id
stringlengths 27
33
| source
stringclasses 1
value | format
stringclasses 1
value | text
stringlengths 13
1.81M
|
|---|---|---|---|
warning/0506/math0506092.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
J. L. Doob was a pioneer in the development of the theory of martingales and its applications to probability theory, potential theory or functional analysis. The fundamental contributions that he made in this field form the cornerstones of one of the richest veins explored in mathematics during the last half-century. In particular, martingales and stochastic calculus provide nowadays key tools for studying the asymptotic behavior of random processes; see the classical books by Ethier and Kurtz and Jacod and Shiryaev . In the present work, we shall apply such techniques to investigate a class of stochastic flows related to certain population dynamics.
The general motivation for the present work is to get a better understanding of the relations between the so-called coalescents with multiple collisions, which were introduced independently by Pitman and Sagitov , and continuous-state branching processes. Note from that coalescents with multiple collisions can be viewed as asymptotic models for the genealogy of a discrete population with a fixed size, and so the existence of connections with branching processes should not come as a surprise. Such connections were already derived in , where the Bolthausen-Sznitman coalescent was shown to describe the genealogical structure of a particular continuous-state branching process introduced by Neveu, and in , where similar relations were obtained between the so-called beta-coalescents and continuous-state branching processes with stable branching mechanism. Here, we do not focus on exact distributional identities, but rather on asymptotics for functionals of coalescent processes, where the limiting objects are given in terms of branching processes. In order to get such asymptotics, we apply the machinery of limit theorems for semimartingales to the so-called generalized Fleming-Viot processes, which where shown in to be duals to the coalescents with multiple collisions.
Generalized Fleming-Viot processes, which model the evolution of a continuous population with fixed size $`1`$, have appeared in articles by Donnelly and Kurtz , and were studied more recently in our work . It is convenient to view a generalized Fleming-Viot process as a stochastic flow $`(F_t,t0)`$ on $`[0,1]`$, such that for each $`t0`$, $`F_t:[0,1][0,1]`$ is a (random) right-continuous increasing map with $`F_t(0)=0`$ and $`F_t(1)=1`$. We should think of the unit interval as a population, and then of $`F_t`$ as the distribution function of a (random) probability measure $`dF_t(x)`$ on $`[0,1]`$. The evolution of the latter is related to the dynamics of the population as follows : For every $`0r_1<r_21`$, the interval $`]F_t(r_1),F_t(r_2)]`$ represents the sub-population at time $`t`$ which consists of descendants of the sub-population $`]r_1,r_2]`$ at the initial time. The transitions of the flow are Markovian, and more precisely, for every $`s,t0`$, we have $`F_{t+s}=\stackrel{~}{F}_sF_t`$, where $`\stackrel{~}{F}_s`$ is a copy of $`F_s`$ independent of $`(F_r,0rt)`$. The distribution of the flow is then characterized by a measure $`\nu `$ on $`]0,1]`$ such that $`_{]0,1]}x^2\nu (dx)<\mathrm{}`$. To explain this, consider the simple case where $`\nu `$ is a finite measure. Let $`((T_i,U_i,\xi _i),iN)`$ denote the sequence of atoms of a Poisson random measure on $`[0,\mathrm{}[\times [0,1]\times [0,1]`$ with intensity $`dtdu\nu (dx)`$, ranked in the increasing order of the first coordinate. The process $`(F_t,t0)`$ starts from $`F_0=\mathrm{Id}`$, remains constant on the intervals $`[T_{i1},T_i[`$ (with the usual convention that $`T_0=0`$), and for every $`iN`$
$$F_{T_i}=\mathrm{\Delta }_iF_{T_{i1}}$$
where
$$\mathrm{\Delta }_i(r)=\xi _i\mathrm{๐}_{\{U_ir\}}+r(1\xi _i),r[0,1].$$
In terms of the population model, this means that at each time $`T_i`$, an individual in the population at time $`T_{i1}`$ is picked uniformly at random and gives birth to a sub-population of size $`\xi _i`$. Simultaneously, the rest of the population shrinks by factor $`1\xi _i`$, so the total size of the population remains $`1`$. The previous description does not apply when $`\nu `$ is infinite, since then the Poisson measure will have infinitely many atoms on a finite time interval. Still, the Fleming-Viot flow can be constructed via a suitable limiting procedure ( Theorem 2).
Our first motivation for studying generalized Fleming-Viot processes came from their remarkable connection with the class of coalescents with multiple collisions considered by Pitman and Sagitov . To describe this connection, fix some time $`T>0`$ viewed as the present date at which the population is observed, and pick a sequence of individuals labelled $`1,2,\mathrm{}`$ independently and uniformly over $`[0,1]`$. For every $`tT`$, we obtain a partition $`\mathrm{\Pi }(t)`$ of $`N`$ by gathering individuals having the same ancestor at time $`Tt`$. The process $`(\mathrm{\Pi }(t),0tT)`$ is then a Markovian coalescent process on the space of partitions of $`N`$. In the terminology of , it is a $`\mathrm{\Lambda }`$-coalescent, with $`\mathrm{\Lambda }(dx)=x^2\nu (dx)`$, started from the partition of $`N`$ into singletons. As a consequence of Kingmanโs theory of exchangeable partitions, for every $`t0`$, each block of $`\mathrm{\Pi }(t)`$ has an asymptotic frequency, also called the size of the block, and the ranked sequence of these frequencies yields a Markov process called the mass-coalescent. As a consequence of the preceding construction, the mass-coalescent at time $`t`$ has the same distribution as the ranked sequence of jump sizes of $`F_t`$.
The first purpose of the present work is to investigate the asymptotic behavior of a rescaled version of the preceding population model. Specifically, we consider a family $`(\stackrel{~}{\nu }^{(a)},a>0)`$ of measures on $`]0,1]`$ such that $`_{]0,1]}x^2\stackrel{~}{\nu }^{(a)}(dx)<\mathrm{}`$ for every $`a>0`$, and the associated generalized Fleming-Viot processes $`\stackrel{~}{F}^{(a)}`$. For each $`a>0`$, we rescale $`\stackrel{~}{F}^{(a)}`$ by a factor $`a`$ in space and time, i.e. we set
$$F_t^{(a)}(r):=a\stackrel{~}{F}_{at}^{(a)}(r/a),r[0,a],t0.$$
So the process $`F^{(a)}`$ describes the evolution of a population with fixed size $`a`$. Roughly speaking, considering $`F^{(a)}`$ in place of $`\stackrel{~}{F}^{(a)}`$ enables us to focus on the dynamics of a sub-population having size of order $`1/a`$. Denote by $`\nu ^{(a)}`$ the image of $`\stackrel{~}{\nu }^{(a)}`$ under the dilation $`xax`$, and assume that the measures $`(x^2x)\nu ^{(a)}(dx)`$ converge weakly as $`a\mathrm{}`$ to a finite measure on $`]0,\mathrm{}[`$, which we may write in the form $`(x^2x)\pi (dx)`$. Then Theorem 1 shows that $`F^{(a)}`$ converges in distribution to the critical continuous-state branching process $`Z`$ with branching mechanism
$$\mathrm{\Psi }(q)=_{]0,\mathrm{}[}(\mathrm{e}^{qx}1+qx)\pi (dx),q0.$$
As a consequence of this limit theorem, we derive a hydrodynamic limit for the associated coalescent processes (Theorem 2). Precisely, we show that under the same assumptions as above, for every $`t0`$, the empirical measure corresponding to the jumps of $`\stackrel{~}{F}_t^{(a)}`$ (or equivalently to the block sizes in the associated coalescent) converges, modulo a suitable rescaling, towards a deterministic measure $`\lambda _t`$. Informally, $`\lambda _t`$ is the distribution ofa cluster at time $`t`$, that is a collection of individuals sharing the same ancestor at the initial time, in the continuous-state branching process with branching mechanism $`\mathrm{\Psi }`$. In a way analogous to the derivation of Smoluchovskiโs coagulation equation from stochastic models (see Aldous , Norris and the references therein for background) we prove that the family $`(\lambda _t,t>0)`$ solves a generalized coagulation equation of the form
$$\frac{d\lambda _t,f}{dt}=\underset{k=2}{\overset{\mathrm{}}{}}\frac{(1)^k\mathrm{\Psi }^{(k)}(\lambda _t,1)}{k!}_{]0,\mathrm{}[^k}(f(x_1+\mathrm{}+x_k)(f(x_1)+\mathrm{}+f(x_k)))\lambda _t(dx_1)\mathrm{}\lambda _t(dx_k)$$
where $`f`$ can be any continuous function with compact support on $`]0,\mathrm{}[`$ (Proposition 3).
In the last part of this work, we study the small time behavior of generalized Fleming-Viot processes and $`\mathrm{\Lambda }`$-coalescents, under a regular variation assumption on the measure $`\nu `$ (recall that $`\mathrm{\Lambda }(dx)=x^2\nu (dx)`$). Precisely, we assume that the tail $`\nu ([\epsilon ,1])`$ is regularly varying with index $`\gamma `$ when $`\epsilon `$ goes to $`0`$. We are interested in the case when the $`\mathrm{\Lambda }`$-coalescent comes down from infinity (i.e. for every $`t>0`$, $`\mathrm{\Pi }_t`$ has finitely many blocks), which forces $`1\gamma 2`$. Leaving aside the boundary cases we suppose that $`1<\gamma <2`$. As a consequence of Theorem 1, we prove that the rescaled Fleming-Viot process
$$F_t^\epsilon (x):=\frac{1}{\epsilon }F_{t/(\epsilon \nu ([\epsilon ,1]))}(\epsilon x)$$
converges in distribution to the continuous-state branching process with stable branching mechanism:
$$\mathrm{\Psi }_\gamma (q)=\frac{\mathrm{\Gamma }(2\gamma )}{\gamma 1}q^\gamma .$$
We then use this result to investigate the small time behavior of the size of blocks in the $`\mathrm{\Lambda }`$-coalescent. Write $`N_t(]0,x[)`$ for the number of blocks with size less than $`x`$ in the $`\mathrm{\Lambda }`$-coalescent at time $`t`$. If $`g(\epsilon )=(\epsilon \nu ([\epsilon ,1]))^1`$, Theorem 4 states that
$$\underset{x]0,\mathrm{}[}{sup}|\epsilon N_{g(\epsilon )}(]0,\epsilon x[)\lambda _1(]0,x[)|\underset{\epsilon 0}{}0,$$
in probability. Furthermore, the measure $`\lambda _1`$ can be characterized by its Laplace transform
$$(1\mathrm{e}^{qr})\lambda _1(dr)=(\mathrm{\Gamma }(2\gamma )+q^{1\gamma })^{1/(1\gamma )}.$$
Theorem 4 is analogous to a classical result for the sizes of blocks in the Kingman coalescent in small time (see Aldous ). The proof uses an intermediate estimate for the total number of blocks in a $`\mathrm{\Lambda }`$-coalescent, which is closely related to the recent paper dealing with beta-coalescents.
The paper is organized as follows. Section 2 gives a few preliminary results about continuous-state branching processes. In particular, the Poisson representation (Proposition 2) may have other applications. Section 3 states our first limit theorem for generalized Fleming-Viot processes. The derivation of the hydrodynamic limit is developed in Section 4, which also discusses the generalized coagulation equation for the family $`(\lambda _t,t0)`$. Finally Section 5 is devoted to the behavior in small time of generalized Fleming-Viot processes and $`\mathrm{\Lambda }`$-coalescents.
Notation. We use the notation $`\mu ,f`$ for the integral of the function $`f`$ with respect to the measure $`\mu `$. We denote by $`_\mathrm{F}`$ the space of all finite measures on $`]0,\mathrm{}[`$, which is equipped with the usual weak topology. We also denote by $`_\mathrm{R}`$ the space of all Radon measures on $`]0,\mathrm{}[`$. The set $`_\mathrm{R}`$ is equipped with the vague topology: A sequence $`(\mu _n,nN)`$ in $`_\mathrm{R}`$ converges to $`\mu _\mathrm{R}`$ if and only if for every continuous function $`f:]0,\mathrm{}[R`$ with compact support, $`lim_n\mathrm{}\mu _n,f=\mu ,f`$.
## 2 Stochastic flows of branching processes
In this section, we give a few properties of continuous-state branching processes that will be needed in the proof of our limit theorems. A critical branching mechanism is a function $`\mathrm{\Psi }:[0,\mathrm{}[[0,\mathrm{}[`$ of the type
$$\mathrm{\Psi }(q)=\beta q^2+_{]0,\mathrm{}[}\left(\mathrm{e}^{rq}1+rq\right)\pi (dr)$$
(1)
where $`\beta 0`$ is the so-called Gaussian coefficient and $`\pi `$ is a measure on $`]0,\mathrm{}[`$ such that $`(rr^2)\pi (dr)<\mathrm{}`$. The continuous-state branching process with branching mechanism $`\mathrm{\Psi }`$ (in short the $`\mathrm{\Psi }`$-CSBP) is the Markov process with values in $`R_+`$, whose transition kernels $`Q_t(x,dy)`$ are determined by the Laplace transform
$$Q_t(x,dy)\mathrm{e}^{qy}=\mathrm{exp}(xu_t(q)),x,t0,q0,$$
(2)
where the function $`u_t(q)`$ solves
$$\frac{u_t(q)}{t}=\mathrm{\Psi }(u_t(q)),u_0(q)=q.$$
(3)
The criticality of $`\mathrm{\Psi }`$ implies that a $`\mathrm{\Psi }`$-CSBP is a nonnegative martingale. If $`Z^1`$ and $`Z^2`$ are two independent $`\mathrm{\Psi }`$-CSBPโs started respectively at $`x_1`$ and $`x_2`$, then $`Z^1+Z^2`$ is also a $`\mathrm{\Psi }`$-CSBP, obviously with initial value $`x_1+x_2`$. From this additivity or branching property, we may construct a two-parameter process $`Z=(Z(t,x),t,x0)`$, such that:
$``$ For each fixed $`x0`$, $`(Z(t,x),t0)`$ is a $`\mathrm{\Psi }`$-CSBP with cร dlร g paths and initial value $`Z(0,x)=x`$.
$``$ If $`x_1,x_20`$, $`Z(,x_1+x_2)Z(,x_1)`$ is independent of the processes $`\left(Z(,x),0xx_1\right)`$ and has the same law as $`Z(,x_2)`$.
These properties entail that for each fixed $`t0`$, $`Z(t,)`$ is an increasing process with independent and stationary increments. Its right-continuous version is a subordinator with Laplace exponent $`u_t`$ determined by (2) and (3). By the Lรฉvy-Khintchin formula, there exists a unique drift coefficient $`d_t0`$ and a unique measure $`\lambda _t`$ on $`]0,\mathrm{}[`$ with $`_{]0,\mathrm{}[}(1x)\lambda _t(dx)<\mathrm{}`$ such that
$$u_t(q)=qd_t+_{]0,\mathrm{}[}(1\mathrm{e}^{qx})\lambda _t(dx),q0.$$
(4)
One refers to $`\lambda _t`$ as the Lรฉvy measure of $`Z(t,)`$. Measures $`\lambda _t`$ play an important role in this work. Informally, we may say that $`\lambda _t`$ is the โdistributionโ of the size of the set of descendants at time $`t`$ of a single individual at time $`0`$. This assertion is informal since $`\lambda _t`$ is not a probability distribution (it may even be an infinite measure). A correct way of stating the above (in the case $`d_t=0`$) is as follows: $`Z(t,x)`$ is the sum of the atoms of a Poisson measure with intensity $`x\lambda _t()`$. Moreover, the study of the genealogical structure of the $`\mathrm{\Psi }`$-CSBP (see e.g. ) allows one to interpret each of these atoms as the size of a family of individuals at time $`t`$ that have the same ancestor at the initial time.
From now on, we assume that $`\beta =0`$ and we exclude the trivial case $`\pi =0`$.
We start by recalling in our special case an important connection between continuous-state branching processes and Lรฉvy processes due to Lamperti . Let $`x>0`$ be fixed, and let $`\xi =(\xi _t,t0)`$ denote a real-valued Lรฉvy process with no negative jumps, started from $`\xi _0=x`$, and whose Laplace exponent is specified by
$$E\left[\mathrm{exp}(q(\xi _t\xi _0))\right]=\mathrm{exp}t\mathrm{\Psi }(q),q0.$$
In particular $`\pi `$ is the Lรฉvy measure of $`\xi `$. The criticality of the branching mechanism $`\mathrm{\Psi }`$ ensures that the Lรฉvy process $`\xi `$ has centered increments and thus oscillates. In particular the first passage time $`\zeta :=inf\{t0:\xi _t=0\}`$ is finite a.s. Next, introduce for every $`t0`$
$$\gamma (t)=_0^{t\zeta }\frac{ds}{\xi _s},C_t=inf\{s0:\gamma (s)>t\}\zeta .$$
Then the time-changed process $`\left(\xi C_t,t0\right)`$ has the same distribution as $`\left(Z(t,x),t0\right)`$.
It follows from this representation that $`(Z(t,x),t0)`$ is a purely discontinuous martingale. We can also use the Lamperti transformation to calculate the compensator of the jump measure of this martingale. By the Lรฉvy-Itรด decomposition, the compensator of the jump measure of $`\xi `$,
$$\underset{\{t:\mathrm{\Delta }\xi _t0\}}{}\delta _{(t,\mathrm{\Delta }\xi _t)},$$
is $`dt\pi (dx)`$. By a time change argument, we can then deduce that the compensator of the measure
$$\underset{\{t:\mathrm{\Delta }Z(t,x)0\}}{}\delta _{(t,\mathrm{\Delta }Z(t,x))}$$
is $`Z(t,x)dt\pi (dr)`$.
Since $`(Z(t,x),t0)`$ is a purely discontinuous martingale, the knowledge of the compensator of its jump measure completely determines the characteristics of this semimartingale, in the sense of Chapter II. We will need the fact that the distribution of $`(Z(t,x),t0)`$, and more generally of the multidimensional process $`((Z(t,x_1),Z(t,x_2),\mathrm{},Z(t,x_p));t0)`$ for any choice of $`p`$ and $`x_1,\mathrm{},x_p`$, is uniquely determined by its characteristics.
Fix an integer $`p1`$ and define
$$๐_p:=\{x=(x_1,\mathrm{},x_p):0x_1x_2\mathrm{}x_p\}.$$
(5)
For every $`(y_1,\mathrm{},y_p)๐_p`$, define a $`\sigma `$-finite measure $`U(y_1,\mathrm{},y_p;dz_1,\mathrm{},dz_p)`$ on $`R_+^p\backslash \{0\}`$ by setting, for any measurable function $`\phi :R_+^pR_+`$ that vanishes at $`0`$,
$$U(y_1,\mathrm{},y_p;dz_1,\mathrm{},dz_p)\phi (z_1,\mathrm{},z_p)=\pi (dr)_0^{\mathrm{}}๐u\phi (r\mathrm{๐}_{\{uy_1\}},\mathrm{},r\mathrm{๐}_{\{uy_p\}}).$$
(6)
###### Proposition 1
Let $`(x_1,\mathrm{},x_p)๐_p`$ and let $`(Z^1,\mathrm{},Z^p)`$ be a $`p`$-dimensional semimartingale taking values in $`๐_p`$, such that $`(Z_0^1,\mathrm{},Z_0^p)=(x_1,\mathrm{},x_p)`$. The following two properties are equivalent:
(i) The processes $`((Z_t^1,\mathrm{},Z_t^p);t0)`$ and $`((Z(t,x_1),\mathrm{},Z(t,x_p));t0)`$ have the same distribution.
(ii) The process $`((Z_t^1,\mathrm{},Z_t^p);t0)`$ is a purely discontinuous local martingale, and the compensator of its jump measure is the measure
$$\theta (dt,dz_1\mathrm{}dz_p)=dtU(Z_t^1,\mathrm{},Z_t^p;dz_1,\mathrm{},dz_p).$$
Proof: The implication (i)$``$(ii) is a straightforward consequence of the remarks preceding the statement and the branching property of continuous-state branching processes. We concentrate on the proof of the converse implication (ii)$``$(i). Let $`q=(q_1,\mathrm{},q_p)]0,\mathrm{}[^p`$, and let $`Y_t=(Y_t^1,\mathrm{},Y_t^p)`$ be defined by $`Y_t^i=Z_t^iZ_t^{i1}`$ if $`i2`$ and $`Y_t^1=Z_t^1`$. Notice that $`Y_t^i0`$. Using property (ii), an application of Itรดโs formula (cf Theorem II.2.42 in ) yields that the process
$`\mathrm{exp}(qY_t)\mathrm{exp}(qY_0)`$
$`{\displaystyle \underset{i=1}{\overset{p}{}}}{\displaystyle _{[0,t]\times [0,\mathrm{}[\times ]0,\mathrm{}[}}\mathrm{exp}(qY_s)\left(\mathrm{e}^{q_ir}1+q_ir\right)\mathrm{๐}_{\{uY_s^i\}}๐s๐u\pi (dr)`$
is a local martingale. This local martingale is bounded over the time interval $`[0,t]`$ for any $`t0`$, hence is a martingale. Taking expectations leads to
$$E[\mathrm{e}^{qY_t}]=E[\mathrm{e}^{qY_0}]+\underset{i=1}{\overset{p}{}}\mathrm{\Psi }(q_i)_0^t๐sE[Y_s^i\mathrm{e}^{qY_s}].$$
(7)
It is immediate to verify from (ii) that each $`Y^i`$ is also a nonnegative local martingale, and so $`E[Y_s^i]E[Y_0^i]=x_ix_{i1}`$ (by convention $`x_0=0`$). If we set $`f_t(q)=E[\mathrm{e}^{qY_t}]`$ we have
$$\frac{f_t(q)}{q_i}=E[Y_t^i\mathrm{e}^{qY_t}]$$
and so we deduce from (7) that
$$\frac{f_t(q)}{t}+\mathrm{\Psi }(q)f_t(q)=\mathrm{\hspace{0.17em}0},$$
(8)
where we write $`\mathrm{\Psi }(q)=(\mathrm{\Psi }(q_1),\mathrm{},\mathrm{\Psi }(q_p))`$. In order to solve (8), fix $`t_1>0`$, and consider the function $`g(t)=(u_{t_1t}(q_1),\mathrm{},u_{t_1t}(q_p))`$ for $`t[0,t_1]`$, where $`u_t(q)`$ is as in (3). Since
$$g^{}(t)=(\mathrm{\Psi }(u_{t_1t}(q_1)),\mathrm{},\mathrm{\Psi }(u_{t_1t}(q_p))),$$
it follows that
$$\frac{f_tg}{t}=\frac{f_t}{t}g+g^{}(t)f_t(g(t))=0$$
by (8). Hence $`f_tg(t)`$ is constant over $`[0,t_1]`$, and
$$f_{t_1}(q)=f_{t_1}(g(t_1))=f_0(g(0))=\mathrm{exp}(\underset{i=1}{\overset{p}{}}(x_ix_{i1})u_{t_1}(q_i)).$$
This shows that
$$(Y_{t_1}^1,\mathrm{},Y_{t_1}^p)\stackrel{(\mathrm{d})}{=}(Z(t_1,x_1),Z(t_1,x_2)Z(t_1,x_1),\mathrm{},Z(t_1,x_p)Z(t_1,x_{p1}))$$
and so
$$(Z_{t_1}^1,\mathrm{},Z_{t_1}^p)\stackrel{(\mathrm{d})}{=}(Z(t_1,x_1),Z(t_1,x_2),\mathrm{},Z(t_1,x_p)).$$
It is easy to iterate this argument to obtain that the processes $`((Z_t^1,\mathrm{},Z_t^p);t0)`$ and $`((Z(t,x_1),\mathrm{},Z(t,x_p));t0)`$ have the same finite-dimensional marginal distributions. The desired result follows since both processes have cร dlร g paths. $`\mathrm{}`$
We now turn our attention to the representation of critical CSBP as stochastic flows on $`[0,\mathrm{}[`$ solving simple stochastic differential equations. On a suitable filtered probability space $`(\mathrm{\Omega },,(_t),P)`$, we consider :
$``$ an $`(_t)`$-Poisson random measure
$$M=\underset{i=1}{\overset{\mathrm{}}{}}\delta _{(t_i,u_i,r_i)},$$
on $`R_+\times [0,\mathrm{}[\times ]0,\mathrm{}[`$, with intensity $`dtdu\pi (dr)`$.
$``$ a collection $`(X_t(x),t0)`$, $`xR_+`$ of cร dlร g $`(_t)`$-martingales with values in $`R_+`$,
$``$ the stochastic differential equation
$$X_t(x)=x+_{[0,t]\times [0,\mathrm{}[\times ]0,\mathrm{}[}M(ds,du,dr)r\mathbf{\hspace{0.17em}1}_{\{uX_s(x)\}}.$$
(9)
The Poissonian stochastic integral in the right-hand side should be understood with respect to the compensated Poisson measure $`M`$ (see e.g. Section II.1 of ). This stochastic integral is well defined according to Definition II.1.37 of , since the increasing process
$$t\left(_{[0,t]\times [0,\mathrm{}[\times [0,\mathrm{}[}M(ds,du,dr)r^2\mathbf{\hspace{0.17em}1}_{\{uX_s(x)\}}\right)^{1/2}$$
is locally integrable under our assumption on $`\pi `$.
A pair $`(M,(X_{}(a),a0))`$ satisfying the above conditions will be called a weak solution of (9).
###### Proposition 2
The equation (9) has a weak solution which satisfies the additional property that $`X_t(x_1)X_t(x_2)`$ for every $`t0`$, a.s. whenever $`0x_1x_2`$. Moreover, for every such solution $`(M,X)`$, for every $`pN`$ and $`0x_1\mathrm{}x_p`$, the process $`((X_t(x_1),\mathrm{},X_t(x_p)),t0)`$ has the same distribution as $`((Z(t,x_1),\mathrm{},Z(t,x_p)),t0)`$.
Proof: The second part of the statement is immediate from the implication (ii)$``$(i) in Proposition 1. The first part can be deduced from Theorem 14.80 in by the same arguments that were used in the proof of Theorem 2 in . We leave details to the reader as this result is not really needed below except for motivation. $`\mathrm{}`$
## 3 Generalized Fleming-Viot flows and their limits
We now recall some results from on generalized Fleming-Viot processes and related stochastic flows. Let $`\nu `$ denote a $`\sigma `$-finite measure on $`]0,1]`$ such that $`_{]0,1]}x^2\nu (dx)<\mathrm{}`$. According to Section 5.1 in , one can associate with $`\nu `$ a Feller process $`(F_t,t0)`$ with values in the space of distribution functions of probability measures on $`]0,1]`$ (i.e. for each $`t0`$, $`F_t`$ is a cร dlag increasing map from $`[0,1]`$ to $`[0,1]`$ with $`F_t(0)=0`$ and $`F_t(1)=1`$), whose evolution is characterized by $`\nu `$ and has been described in Section 1 in the special case when $`\nu `$ is finite.
In , we have shown that such generalized Fleming-Viot processes can be described as the solution to a certain system of Poissonian SDEโs. More precisely, on a suitable filtered probability space $`(\mathrm{\Omega },,(_t),P)`$, one can construct the following processes:
$``$ an $`(_t)`$-Poisson point process $`N`$ on $`R_+\times ]0,1[\times ]0,1]`$ with intensity $`dtdu\nu (dr)`$,
$``$ a collection $`(Y_t(x),t0)`$, $`x[0,1]`$, of adapted cร dlร g processes with values in $`[0,1]`$ with $`Y_t(x_1)Y_t(x_2)`$ for all $`t0`$ a.s. when $`0x_1x_21`$,
in such a way that for every $`r[0,1]`$, a.s.
$$Y_t(x)=x+_{[0,t]\times ]0,1[\times ]0,1]}N(ds,du,dr)r\left(\mathrm{๐}_{\{uY_s(x)\}}Y_s(x)\right).$$
(10)
The Poissonian stochastic integral in the right-hand side should again be understood with respect to the compensated Poisson measure $`N`$.
Weak uniqueness holds for this system of SDEโs (Theorem 2 in ). Furthermore, for every integer $`p1`$ and every $`0x_1\mathrm{}x_p1`$, the processes $`((Y_t(x_1),\mathrm{},Y_t(x_p)),t0)`$ and $`((F_t(x_1),\mathrm{},F_t(x_p)),t0)`$ have the same distribution. Note the similarity with Proposition 2: Compare (9) and (10). This strongly suggests to look for asymptotic results relating the processes $`Z(t,x)`$ and $`F_t(x)`$.
For every integer $`p1`$ and every $`a>0`$, set
$$๐_p^a=๐_p[0,a]^p=\{(x_1,\mathrm{},x_p)R_+^p:0x_1x_2\mathrm{}x_pa\}.$$
From (10) we see that for every $`(x_1,\mathrm{},x_p)๐_p^1`$, the process $`(F_t(x_1),\mathrm{},F_t(x_p))`$ is a purely discontinuous martingale, and the compensator of its jump measure is
$$dtR(F_t(x_1),\mathrm{},F_t(x_p);dz_1,\mathrm{},dz_p),$$
where for every $`(y_1,\mathrm{},y_p)๐_p^1`$, the measure $`R(y_1,\mathrm{},y_p;dz_1,\mathrm{},dz_p)`$ on $`R^p\backslash \{0\}`$ is determined by
$$R(y_1,\mathrm{},y_p;dz_1,\mathrm{},dz_p)\phi (z_1,\mathrm{},z_p)=\nu (dr)_0^1๐u\phi (r(\mathrm{๐}_{\{uy_1\}}y_1),\mathrm{},r(\mathrm{๐}_{\{uy_p\}}y_p)).$$
Consider now a family $`(\stackrel{~}{\nu }^{(a)},a>0)`$ of measures on $`]0,1]`$ with $`_{]0,1]}r^2\stackrel{~}{\nu }^{(a)}(dr)<\mathrm{}`$, and for each $`a>0`$, let $`\stackrel{~}{F}^{(a)}`$ be the associated Fleming-Viot process. We then write
$$F_t^{(a)}(x):=a\stackrel{~}{F}_{at}^{(a)}(x/a),x[0,a],t0$$
for the rescaled version of the Fleming-Viot flow. So, for each $`t0`$, $`F_t^{(a)}`$ is the distribution function of a measure on $`]0,a]`$ with total mass $`a`$. For every fixed real number $`a>0`$, we also denote by $`\nu ^{(a)}`$ the measure on $`]0,\mathrm{}[`$ which is $`0`$ on $`]a,\mathrm{}[`$ and whose restriction to $`]0,a]`$ is given by the image of $`\stackrel{~}{\nu }^{(a)}`$ under the dilation $`rar`$ from $`]0,1]`$ to $`]0,a]`$. In particular $`r^2\nu ^{(a)}(dr)`$ is a finite measure on $`]0,\mathrm{}[`$.
By a scaling argument, we see that, for every $`(x_1,\mathrm{},x_p)๐_p^a`$, $`(F_t^{(a)}(x_1),\mathrm{},F_t^{(a)}(x_p))`$ is a purely discontinuous martingale, with values in $`๐_p^a`$, and the compensator of its jump measure is
$$\mu _{(a)}(dt,dz_1\mathrm{}dz_p)=dtR^{(a)}(F_t^{(a)}(x_1),\mathrm{},F_t^{(a)}(x_p);dz_1,\mathrm{},dz_p)$$
(11)
where
$`{\displaystyle R^{(a)}(y_1,\mathrm{},y_p;dz_1,\mathrm{},dz_p)\phi (z_1,\mathrm{},z_p)}`$
$`={\displaystyle \nu ^{(a)}(dr)_0^a๐u\phi (r(\mathrm{๐}_{\{uy_1\}}a^1y_1),\mathrm{},r(\mathrm{๐}_{\{uy_p\}}a^1y_p))}.`$ (12)
Let $`\pi `$ be as in Section 2 a nontrivial measure on $`]0,\mathrm{}[`$ such that $`(rr^2)\pi (dr)<\mathrm{}`$, and let $`\mathrm{\Psi }`$ be as in (1). Denote by $`(Z(t,x),t0,x0)`$ the associated flow of continuous-state branching processes constructed in Section 2.
Assumption (H) The measures $`(rr^2)\nu ^{(a)}(dr)`$ converge to $`(rr^2)\pi (dr)`$ as $`a\mathrm{}`$, in the sense of weak convergence in $`_\mathrm{F}`$.
###### Theorem 1
Under Assumption (H), for every $`(x_1,\mathrm{},x_p)๐_p`$,
$$((F_t^{(a)}(x_1),\mathrm{},F_t^{(a)}(x_p));t0)\underset{a\mathrm{}}{\overset{(\mathrm{d})}{}}((Z(t,x_1),\mathrm{},Z(t,x_p));t0)$$
in the Skorokhod space $`D(R_+,R^p)`$.
Proof: The proof only uses the facts that $`(F_t^{(a)}(x_1),\mathrm{},F_t^{(a)}(x_p))`$ is a purely discontinuous martingale and that the compensator of its jump measure is given by (11) and (3). The latter properties indeed characterize the law of the process $`(F_t^{(a)}(x_1),\mathrm{},F_t^{(a)}(x_p))`$ (cf Lemma 1 in ), but we do not use this uniqueness property in the proof. We fix a sequence $`(a_n)`$ tending to $`+\mathrm{}`$, and $`(x_1,\mathrm{},x_p)๐_p`$. To simplify notation we write
$$Y_t^n=(Y_t^{n,1},\mathrm{},Y_t^{n,p})=(F_t^{(a_n)}(x_1),\mathrm{},F_t^{(a_n)}(x_p))$$
which makes sense as soon as $`a_nx_p`$, hence for all $`n`$ sufficiently large. We also set
$$Z_t=(Z_t^1,\mathrm{},Z_t^p)=(Z(t,x_1),\mathrm{},Z(t,x_p)).$$
We rely on general limit theorems for semimartingales with jumps which can be found in the book . To this end, we first need to introduce a truncation function $`h:RR`$, that is a bounded continuous function such that $`h(x)=x`$ for every $`x[\delta ,\delta ]`$, for some $`\delta >0`$. We may and will assume that $`h`$ is nondecreasing, $`|h(x)||x|1`$ for every $`xR`$ and that $`h`$ is Lipschitz continuous with Lipschitz constant $`1`$, that is $`|h(x)h(y)||xy|`$ for every $`x,yR`$. We can then consider the associated (modified) triplet of characteristics of the $`p`$-dimensional semimartingale $`Y^n`$:
$$(B^n,\stackrel{~}{C}^n,\mu _{(a_n)}).$$
See Definition II.2.16 in . To be specific, $`\mu _{(a_n)}`$ is defined in (11). Then, since $`Y_t^n`$ is a purely discontinuous martingale, we have $`B_t^n=(B_t^{n,i})_{1ip}`$, with
$$B_t^{n,i}=_{[0,t]\times R^p}\mu _{(a_n)}(dt,dz_1\mathrm{}dz_p)(z_ih(z_i))$$
Similarly, $`\stackrel{~}{C}_t^n=(\stackrel{~}{C}_t^{i,j,n})_{1i,jp}`$, with
$$\stackrel{~}{C}_t^{i,j,n}=_{[0,t]\times R^p}\mu _{(a_n)}(dt,dz_1\mathrm{}dz_p)h(z_i)h(z_j).$$
Write $`C_{}(R^p)`$ for the space of all bounded Lipschitz continuous functions on $`R^p`$ that vanish on a neighborhood of $`0`$. We fix $`gC_{}(R^p)`$ such that $`|g|1`$, and we choose $`\alpha >0`$ such that $`g(z_1,\mathrm{},z_p)=0`$ if $`|z_i|\alpha `$ for every $`i=1,\mathrm{},p`$. Following the notation in , we set
$$(g\mu _{(a_n)})_t=_{[0,t]\times R^p}\mu _{(a_n)}(dt,dz_1\mathrm{}dz_p)g(z_1,\mathrm{},z_p).$$
From formula (11) we have
$`B_t^{n,i}`$ $`=`$ $`{\displaystyle _0^t}๐s\beta ^{n,i}(Y_s^{n,1},\mathrm{},Y_s^{n,p})`$
$`\stackrel{~}{C}_t^{n,i,j}`$ $`=`$ $`{\displaystyle _0^t}๐s\gamma ^{n,i,j}(Y_s^{n,1},\mathrm{},Y_s^{n,p})`$ (13)
$`(g\mu _{(a_n)})_t`$ $`=`$ $`{\displaystyle _0^t}๐s\phi ^n(Y_s^{n,1},\mathrm{},Y_s^{n,p}),`$
where the functions $`\beta ^{n,i},\gamma ^{n,i,j},\phi ^n`$ are defined by
$`\beta ^{n,i}(y_1,\mathrm{},y_p)`$ $`=`$ $`{\displaystyle _{R^p}}R^{(a_n)}(y_1,\mathrm{},y_p;dz_1,\mathrm{},dz_p)(z_ih(z_i))`$
$`\gamma ^{n,i,j}(y_1,\mathrm{},y_p)`$ $`=`$ $`{\displaystyle _{R^p}}R^{(a_n)}(y_1,\mathrm{},y_p;dz_1,\mathrm{},dz_p)h(z_i)h(z_j)`$
$`\phi ^n(y_1,\mathrm{},y_p)`$ $`=`$ $`{\displaystyle _{R^p}}R^{(a_n)}(y_1,\mathrm{},y_p;dz_1,\mathrm{},dz_p)g(z_1,\mathrm{},z_p).`$
Similarly, the (modified) characteristics of the semimartingale $`Z`$ are
$$(B,\stackrel{~}{C},\theta )$$
where $`\theta `$ is as in Proposition 1, and
$`B_t^i`$ $`=`$ $`{\displaystyle _0^t}๐s\beta ^i(Z_s^1,\mathrm{},Z_s^p)`$
$`\stackrel{~}{C}_t^{i,j}`$ $`=`$ $`{\displaystyle _0^t}๐s\gamma ^{i,j}(Z_s^1,\mathrm{},Z_s^p)`$ (14)
$`(g\theta )_t`$ $`=`$ $`{\displaystyle _0^t}๐s\phi (Z_s^1,\mathrm{},Z_s^p),`$
where the functions $`\beta ^i,\gamma ^{i,j},\phi `$ are respectively defined by
$`\beta ^i(y_1,\mathrm{},y_p)`$ $`=`$ $`{\displaystyle _{R^p}}U(y_1,\mathrm{},y_p;dz_1,\mathrm{},dz_p)(z_ih(z_i))`$
$`\gamma ^{i,j}(y_1,\mathrm{},y_p)`$ $`=`$ $`{\displaystyle _{R^p}}U(y_1,\mathrm{},y_p;dz_1,\mathrm{},dz_p)h(z_i)h(z_j)`$
$`\phi (y_1,\mathrm{},y_p)`$ $`=`$ $`{\displaystyle _{R^p}}U(y_1,\mathrm{},y_p;dz_1,\mathrm{},dz_p)g(z_1,\mathrm{},z_p).`$
###### Lemma 1
For every $`(y_1,\mathrm{},y_p)๐_p`$,
$`|\beta ^{n,i}(y_1,\mathrm{},y_p)|`$ $``$ $`2y_i{\displaystyle \nu _{(a_n)}(dr)r\mathbf{\hspace{0.17em}1}_{\{r>\delta \}}}`$
$`|\gamma ^{n,i,j}(y_1,\mathrm{},y_p)|`$ $``$ $`(y_i+y_j){\displaystyle \nu _{(a_n)}(dr)(rr^2)}`$
$`|\phi ^n(y_1,\mathrm{},y_p)`$ $``$ $`{\displaystyle \frac{2}{\alpha }}y_p{\displaystyle \nu _{(a_n)}(dr)r\mathbf{\hspace{0.17em}1}_{\{r>\alpha \}}}.`$
Moreover,
$`\underset{n\mathrm{}}{lim}\beta ^{n,i}(y_1,\mathrm{},y_p)`$ $`=`$ $`\beta ^i(y_1,\mathrm{},y_p)`$
$`\underset{n\mathrm{}}{lim}\gamma ^{n,i,j}(y_1,\mathrm{},y_p)`$ $`=`$ $`\gamma ^{i,j}(y_1,\mathrm{},y_p)`$
$`\underset{n\mathrm{}}{lim}\phi ^n(y_1,\mathrm{},y_p)`$ $`=`$ $`\phi (y_1,\mathrm{},y_p),`$
uniformly on bounded subsets of $`๐_p`$.
Let us postpone the proof of the lemma and complete that of the theorem. The first step is to check the sequence of the laws of the processes $`Y^n`$ is tight in the space of probability measures on $`D(R_+,R^p)`$. This will follow from Theorem VI.4.18 in provided we can check that:
(i) We have for every $`N>0`$ and $`\epsilon >0`$,
$$\underset{b\mathrm{}}{lim}\left(\underset{n\mathrm{}}{lim\; sup}P[\mu _{(a_n)}([0,N]\times \{zR^p:|z|>b\})>\epsilon ]\right)=0.$$
(ii) The laws of the processes $`B^{n,i},\stackrel{~}{C}^{n,i,j},g\mu _{(a_n)}`$ are tight in the space of probability measures on $`C(R_+,R)`$.
To prove (i), set
$$T_A^n=inf\{t0:Y_t^{p,n}>A\}$$
for every $`A>x_p`$. Since $`Y^{n,p}`$ is a (bounded) nonnegative martingale, a classical result states that
$$P[sup\{Y_t^{n,p},t0\}>A]=P[T_A^n<\mathrm{}]\frac{x_p}{A}.$$
(15)
From formulas (11) and (3), we have on the event $`\{sup\{Y_t^{p,n},t0\}A\}`$
$$\mu _{(a_n)}([0,N]\times \{zR^p:|z|>b\})N(A\nu _{(a_n)}(]\frac{b}{p},\mathrm{}[)+a_n\nu _{(a_n)}(]\frac{ba_n}{pA},\mathrm{}[))$$
Under Assumption (H), we have
$$\underset{n\mathrm{}}{lim}a_n\nu _{(a_n)}(]\frac{ba_n}{pA},\mathrm{}[)=0$$
and so, on the event $`\{sup\{Y_t^{p,n},t0\}A\}`$,
$$\underset{n\mathrm{}}{lim\; sup}\mu _{(a_n)}([0,N]\times \{zR^p:|z|>b\})NA\pi ([\frac{b}{p},\mathrm{}[).$$
If we first choose $`A`$ so that $`x_p/A`$ is small, and then $`b`$ large enough so that $`NA\pi ([\frac{b}{p},\mathrm{}[)<\epsilon `$, we see that the statement in (i) follows from (15). Part (ii) is a straightforward consequence of formulas (3), the bounds of the first part of Lemma 1 and (15) again. This completes the proof of the tightness of the sequence of the laws of the processes $`Y^n`$.
Then, we can assume that, at least along a suitable subsequence, $`Y^n`$ converges in distribution towards a limiting process $`Y^{\mathrm{}}=(Y^{\mathrm{},1},\mathrm{},Y^{\mathrm{},p})`$. We claim that $`Y^{\mathrm{}}`$ is a semimartingale whose triplet of (modified) characteristics $`(B^{\mathrm{}},\stackrel{~}{C}^{\mathrm{}},\mu _{\mathrm{}})`$ is such that
$`B_t^{\mathrm{},i}`$ $`=`$ $`{\displaystyle _0^t}๐s\beta ^i(Y_s^{\mathrm{},1},\mathrm{},Y_s^{\mathrm{},p})`$
$`\stackrel{~}{C}_t^{\mathrm{},i,j}`$ $`=`$ $`{\displaystyle _0^t}๐s\gamma ^{i,j}(Y_s^{\mathrm{},1},\mathrm{},Y_s^{\mathrm{},p})`$ (16)
$`(g\mu _{\mathrm{}})_t`$ $`=`$ $`{\displaystyle _0^t}๐s\phi (Y_s^{\mathrm{},1},\mathrm{},Y_s^{\mathrm{},p}),`$
with $`\beta ^i,\gamma ^{i,j},\phi `$ as above. To see this, it is enough to verify that the 4-tuples $`(Y^n,B^n,\stackrel{~}{C}^n,g\mu _{(a_n)})`$ converge in distribution to $`(Y^{\mathrm{}},B^{\mathrm{}},\stackrel{~}{C}^{\mathrm{}},g\mu _{\mathrm{}})`$ (see Theorem IX.2.4 in ). The latter convergence readily follows from the convergence of $`Y^n`$ towards $`Y^{\mathrm{}}`$, formulas (3) and the second part of Lemma 1.
Finally, knowing the triplet of characteristics of $`Y^{\mathrm{}}`$, Theorem II.2.34 in shows that $`Y^{\mathrm{}}`$ is a purely discontinuous martingale, and the compensator of its jump measure is
$$dtU(Y_t^{\mathrm{},1},\mathrm{},Y_t^{\mathrm{},p};dz_1,\mathrm{},dz_p).$$
By Proposition 1, this implies that $`Y^{\mathrm{}}`$ has the same distribution as $`Z`$, and this completes the proof of Theorem 1. $`\mathrm{}`$
Proof of Lemma 1: By definition, for $`(y_1,\mathrm{},y_p)๐_p^{a_n}`$,
$`\beta ^{n,i}(y_1,\mathrm{},y_p)`$ $`=`$ $`{\displaystyle \nu ^{(a_n)}(dr)_0^{a_n}๐u\left(r(\mathrm{๐}_{\{uy_i\}}a_n^1y_i)h(r(\mathrm{๐}_{\{uy_i\}}a_n^1y_i))\right)}`$
$`=`$ $`{\displaystyle \nu ^{(a_n)}(dr)y_i(r(1a_n^1y_i)h(r(1a_n^1y_i)))}`$
$`+{\displaystyle \nu ^{(a_n)}(dr)(a_ny_i)(a_n^1ry_i+h(a_n^1ry_i))}.`$
Recalling that $`h(x)=x`$ if $`|x|\delta `$, we immediately get the bound
$$\beta ^{n,i}(y_1,\mathrm{},y_p)2y_i\nu ^{(a_n)}(dr)r\mathrm{๐}_{\{r>\delta \}}.$$
Furthermore, using the fact that $`h`$ is Lipschitz with Lipschitz constant $`1`$, we have
$`|\beta ^{n,i}(y_1,\mathrm{},y_p)+y_i{\displaystyle \nu ^{(a_n)}(dr)(rh(r))}|`$
$`2a_n^1y_i^2{\displaystyle \nu ^{(a_n)}(dr)r\mathrm{๐}_{\{r>\delta \}}}+y_i{\displaystyle \nu ^{(a_n)}(dr)r\mathrm{๐}_{\{a_n^1ry_i>\delta \}}}`$
and it is easy to verify from Assumption (H) that the right-hand side tends to $`0`$ as $`n\mathrm{}`$, uniformly when $`y_i`$ varies over a bounded subset in $`R_+`$. Since Assumption (H) also implies that
$$\underset{n\mathrm{}}{lim}\nu ^{(a_n)}(dr)(rh(r))=\pi (dr)(rh(r)),$$
we get the first limit of the lemma.
Consider now, for $`(y_1,\mathrm{},y_p)๐_p^{a_n}`$, and $`1ijp`$,
$`\gamma ^{n,i,j}(y_1,\mathrm{},y_p)`$ $`=`$ $`{\displaystyle }\nu ^{(a_n)}(dr){\displaystyle _0^{a_n}}duh(r(\mathrm{๐}_{\{uy_i\}}a_n^1y_i))h(r(\mathrm{๐}_{\{uy_j\}}a_n^1y_j)))`$ (17)
$`=`$ $`{\displaystyle \nu ^{(a_n)}(dr)y_ih(r(1a_n^1y_i))h(r(1a_n^1y_j))}`$
$`+{\displaystyle \nu ^{(a_n)}(dr)(y_jy_i)h(a_n^1ry_i)h(r(1a_n^1y_j))}`$
$`+{\displaystyle \nu ^{(a_n)}(dr)(a_ny_j)h(a_n^1ry_i)h(a_n^1ry_j)}.`$
Using the bounds $`|h|1`$ and $`|h(x)||x|`$, we get
$$|\gamma ^{n,i,j}(y_1,\mathrm{},y_p)|y_j\nu ^{(a_n)}(dr)(r^21)+y_i\nu ^{(a_n)}(dr)r(r1),$$
which gives the second bound of the lemma. Then, using the Lipschitz property of $`h`$,
$`\left|{\displaystyle \nu ^{(a_n)}(dr)h(r(1a_n^1y_i))h(r(1a_n^1y_j))}{\displaystyle \nu ^{(a_n)}(dr)h(r)^2}\right|`$
$`2a_n^1y_j{\displaystyle \nu ^{(a_n)}(dr)rh(r)}0`$
as $`n\mathrm{}`$. Notice that
$$\underset{n\mathrm{}}{lim}y_i\nu ^{(a_n)}(dr)h(r)^2=y_i\pi (dr)h(r)^2=\gamma ^{i,j}(y_1,\mathrm{},y_p),$$
uniformly when $`(y_1,\mathrm{},y_p)`$ varies over a bounded set. To complete the verification of the second limit in the lemma, we need to check that the last two terms in the right-hand side of (17) tend to $`0`$ as $`n\mathrm{}`$. We have first
$$\nu ^{(a_n)}(dr)h(a_n^1ry_i)h(r(1a_n^1y_j))\nu ^{(a_n)}(dr)ra_n^1y_ih(r)0$$
as $`n\mathrm{}`$. It remains to bound
$`\left|a_n{\displaystyle \nu ^{(a_n)}(dr)h(a_n^1ry_i)h(a_n^1ry_j)}\right|`$
$`{\displaystyle \nu ^{(a_n)}(dr)ry_i((a_n^1ry_j)1)}`$
$`y_i{\displaystyle \nu ^{(a_n)}(dr)r\mathrm{๐}_{\{r>A\}}}+y_iy_ja_n^1{\displaystyle \nu ^{(a_n)}(dr)r^2\mathrm{๐}_{\{rA\}}}`$
where $`A>0`$ is arbitrary. If $`\eta >0`$ is given, we can first choose $`A`$ sufficiently large so that
$$\underset{n\mathrm{}}{lim\; sup}\nu ^{(a_n)}(dr)r\mathrm{๐}_{\{r>A\}}\pi (dr)r\mathrm{๐}_{\{rA\}}<\eta .$$
On the other hand, we have also
$$\underset{n\mathrm{}}{lim}a_n^1\nu ^{(a_n)}(dr)r^2\mathrm{๐}_{\{rA\}}=0$$
and together with the preceding estimates, this gives the second limit of the lemma.
Finally, we have
$$\phi ^n(y_1,\mathrm{},y_p)=\nu ^{(a_n)}(dr)_0^{a_n}๐ug(r(\mathrm{๐}_{\{uy_1\}}a_n^1y_1),\mathrm{},r(\mathrm{๐}_{\{uy_p\}}a_n^1y_p)).$$
Since $`|g|1`$ and $`g(z_1,\mathrm{},z_p)=0`$ if $`sup|z_i|\alpha `$, we easily get the bound
$`|\phi ^n(y_1,\mathrm{},y_p)|`$ $``$ $`y_p{\displaystyle \nu ^{(a_n)}(dr)\mathbf{\hspace{0.17em}1}_{\{r>\alpha \}}}+a_n{\displaystyle \nu ^{(a_n)}(dr)\mathbf{\hspace{0.17em}1}_{\{a_n^1ry_p>\alpha \}}}`$
$``$ $`y_p{\displaystyle \nu ^{(a_n)}(dr)\mathbf{\hspace{0.17em}1}_{\{r>\alpha \}}}+{\displaystyle \frac{y_p}{\alpha }}{\displaystyle \nu ^{(a_n)}(dr)r\mathrm{๐}_{\{r>\alpha \}}}`$
which gives the third bound of the lemma. Then, if $`M`$ denotes a Lipschitz constant for $`g`$,
$`\left|\phi ^n(y_1,\mathrm{},y_p){\displaystyle \nu ^{(a_n)}(dr)_0^{a_n}๐ug(r\mathrm{๐}_{\{uy_1\}},\mathrm{},r\mathrm{๐}_{\{uy_p\}})}\right|`$
$`Mp{\displaystyle \nu ^{(a_n)}(dr)_0^{a_n}๐u(\mathrm{๐}_{\{uy_p\}}a_n^1ry_p\mathbf{\hspace{0.17em}1}_{\{r>\alpha \}}+\mathrm{๐}_{\{u>y_p\}}a_n^1ry_p\mathbf{\hspace{0.17em}1}_{\{a_n^1ry_p>\alpha \}})}`$
$`Mp\left({\displaystyle \nu ^{(a_n)}(dr)r\mathrm{๐}_{\{r>\alpha \}}}\right)a_n^1y_p^2+Mpy_p{\displaystyle \nu ^{(a_n)}(dr)r\mathrm{๐}_{\{a_n^1ry_p>\alpha \}}}`$
which tends to $`0`$ as $`n`$ tends to $`\mathrm{}`$, uniformly when $`y_p`$ varies over a compact subset of $`R_+`$. The last convergence of the lemma now follows from Assumption (H). This completes the proof. $`\mathrm{}`$
## 4 Hydrodynamic limits for exchangeable coalescents
The motivation for this section stems from hydrodynamic limit theorems leading from stochastic coalescents to Smoluchowskiโs coagulation equation, which we now summarize.
### 4.1 Stochastic coalescents and Smoluchowskiโs coagulation equation
Consider a symmetric measurable function $`K:]0,\mathrm{}[\times ]0,\mathrm{}[R_+`$ which will be referred to as a coagulation kernel. A stochastic coalescent with coagulation kernel $`K`$ can be viewed as a Markov chain in continuous time $`C=(C_t,t0)`$ with values in the space of finite integer-valued measures on $`]0,\mathrm{}[`$ with the following dynamics. Suppose that the process starts from some state $`_{i=1}^k\delta _{x_i}`$, where $`k2`$ and $`x_i]0,\mathrm{}[`$ for $`i=1,\mathrm{},k`$. For $`1i<jk`$, let $`๐_{i,j}`$ be an exponential variable with parameter $`K(x_i,x_j)`$, such that to different pairs correspond independent variables. The first jump of the process $`C`$ occurs at time $`\mathrm{min}_{1i<jk}๐_{i,j}`$, and if this minimum is reached for the indices $`1\mathrm{}<mk`$ (i.e. $`\mathrm{}`$ and $`k`$ are the indices such that $`\mathrm{min}_{1i<jk}๐_{i,j}=๐_{\mathrm{},m}`$), then the state after the jump is
$$\delta _{x_{\mathrm{}}+x_m}+\underset{i\mathrm{},m}{}\delta _{x_i}.$$
In other words, a stochastic coalescent with coagulation kernel $`K`$ is a finite particle system in $`]0,\mathrm{}[`$ such that each pair of particles $`(x_i,x_j)`$ in the system merges at rate $`K(x_i,x_j)`$, independently of the other pairs.
Now consider a sequence $`(\stackrel{~}{C}_t^{(n)},t0)_{nN}`$ of stochastic coalescents with coagulation kernel $`K`$ and set $`C_t^{(n)}=n^1\stackrel{~}{C}_{t/n}^{(n)}`$ for $`t0`$. Suppose that the sequence of initial states $`C_0^{(n)}`$ converges in probability in $`_\mathrm{R}`$ to a Radon measure $`\mu _0`$. Then under some technical assumptions on the coagulation kernel $`K`$ (see e.g. Norris ), the sequence $`(C_t^{(n)},t0)`$ converges in probability on the space of cร dlร g trajectories with values in $`_\mathrm{R}`$ towards a deterministic limit $`(\mu _t,t0)`$. Moreover this limit is characterized as the solution to Smoluchowskiโs coagulation equation
$$\frac{d\mu _t,f}{dt}=\frac{1}{2}_{]0,\mathrm{}[^2}\left(f(x+y)f(x)f(y)\right)K(x,y)\mu _t(dx)\mu _t(dy),$$
(18)
where $`f:]0,\mathrm{}[R`$ denotes a generic continuous function with compact support.
### 4.2 Hydrodynamic limits
Let $`\nu `$ denote a $`\sigma `$-finite measure on $`]0,1]`$ such that $`_{]0,1]}r^2\nu (dr)<\mathrm{}`$, and write $`\mathrm{\Lambda }(dr)=r^2\nu (dr)`$, which is thus a finite measure on $`]0,1]`$. The so-called $`\mathrm{\Lambda }`$-coalescent (or coalescent with multiple collisions, see ) is a Markov process $`(\mathrm{\Pi }_t,t0)`$ taking values in the set of all partitions of $`N`$. Unless otherwise specified, we assume that $`\mathrm{\Pi }_0`$ is the partition of $`N`$ into singletons. For every $`t0`$, write $`D_t`$ for the sequence of asymptotic frequencies of the blocks of $`\mathrm{\Pi }_t`$, ranked in nonincreasing order (if the number $`k`$ of blocks is finite, then the terms of index greater than $`k`$ in the sequence are all equal to $`0`$). Then (, section 2.2) the process $`(D_t,t0)`$ is a time-homogeneous Markov process with values in the space $`๐ฎ_1^{}`$ of nonincreasing numerical sequences $`๐ฌ=(s_1,\mathrm{})`$ with $`_{i=1}^{\mathrm{}}s_i1`$.
The following connection with generalized Fleming-Viot processes can be found in . Let $`(F_t,t0)`$ be the generalized Fleming-Viot process associated with $`\nu `$, and for every $`t0`$, let $`J_t`$ be the sequence of sizes of jumps of the mapping $`xF_t(x)`$, ranked again in nonincreasing order, and with the same convention if there are finitely many jumps. Then, for each fixed $`t0`$, $`J_t`$ and $`D_t`$ have the same distribution (Theorem 1 in indeed gives a deeper connection, which has been briefly described in Section 1).
For each $`a>0`$, let $`\stackrel{~}{\nu }^{(a)}`$, $`\nu ^{(a)}`$, $`\stackrel{~}{F}^{(a)}`$ and $`F^{(a)}`$ be as in Section 3. Denote by $`\stackrel{~}{\mu }_t^{(a)}`$ the point measure whose atoms are given by the jump sizes of the increasing process $`x\stackrel{~}{F}_t^{(a)}(x)`$:
$$\stackrel{~}{\mu }_t^{(a)}=\underset{\{x]0,1]:\stackrel{~}{F}_t^{(a)}(x)\stackrel{~}{F}_t^{(a)}(x)>0\}}{}\delta _{\stackrel{~}{F}_t^{(a)}(x)\stackrel{~}{F}_t^{(a)}(x)}.$$
Fom the preceding observations, the atoms of $`\stackrel{~}{\mu }_t^{(a)}`$ also correspond to the sizes of the blocks in a $`\mathrm{\Lambda }`$-coalescent at time $`t`$, for $`\mathrm{\Lambda }(dr)=r^2\stackrel{~}{\nu }_t^{(a)}(dr)`$. We then consider the rescaled version $`\mu _t^{(a)}`$, given as the image of $`a^1\stackrel{~}{\mu }_{at}^{(a)}`$ under the dilation $`rar`$. Equivalently, $`\mu ^{(a)}`$ is $`a^1`$ times the sum of the Dirac point masses at the jump sizes of the mapping $`xF_t^{(a)}(x)`$.
###### Theorem 2
Suppose that (H) holds and let $`(Z(t,x);t,x0)`$ be the flow of continuous-state branching processes associated with $`\pi `$. Then for every $`t0`$, $`\mu _t^{(a)}`$ converges to the Lรฉvy measure $`\lambda _t`$ of the subordinator $`Z(t,)`$ as $`a\mathrm{}`$ in probability in $`_\mathrm{R}`$.
Theorem 2 is an immediate consequence of Theorem 1 and the following lemma.
###### Lemma 2
Let $`\sigma =(\sigma _t,t0)`$ be a subordinator with Lรฉvy measure $`\lambda `$. For each $`a>0`$, let $`X^{(a)}=(X_t^{(a)},0ta)`$ be an increasing cร dlร g process with exchangeable increments, with $`X_0^{(a)}=0`$ and $`X_a^{(a)}=a`$ a.s. Suppose that $`X^{(a)}`$ converges to $`\sigma `$ as $`a\mathrm{}`$ in the sense of finite-dimensional distributions. Then the random point measure
$$a^1\underset{0<t<a}{}\delta _{\mathrm{\Delta }X_t^{(a)}}$$
converges to $`\lambda `$ in probability in $`_\mathrm{R}`$ as $`a\mathrm{}`$.
Proof: Pick some nonnegative continuous function $`f:]0,\mathrm{}[R`$ with compact support and write
$$c:=_{]0,\mathrm{}[}f(x)\lambda (dx).$$
By the Lรฉvy-Itรด decomposition for subordinators, the random point measure $`_{\mathrm{\Delta }\sigma _t>0}\delta _{(t,\mathrm{\Delta }\sigma _t)}`$ on $`R_+\times ]0,\mathrm{}[`$ is Poisson with intensity $`dt\lambda (dx)`$. Let $`\rho >0`$. The law of large numbers ensures the existence of a real number $`a_\rho >0`$ such that
$$E\left[\left|a_\rho ^1\underset{0<t<a_\rho }{}f(\mathrm{\Delta }\sigma _t)c\right|\right]<\rho .$$
(19)
Then consider for $`a>a_\rho `$ the bridges with exchangeable increments, bounded variation and no negative jumps on the time interval $`[0,a_\rho ]`$, defined by
$$B_t^{(a)}:=X_t^{(a)}ta_\rho ^1X_{a_\rho }^{(a)},B_t=\sigma _tta_\rho ^1\sigma _{a_\rho },t[0,a_\rho ].$$
Our assumptions entail that $`B^{(a)}`$ converges in the sense of finite dimensional distributions to $`B`$, so according to Kallenberg , the random measure
$$\underset{0<t<a_\rho }{}\delta _{\mathrm{\Delta }B_t^{(a)}}=\underset{0<t<a_\rho }{}\delta _{\mathrm{\Delta }X_t^{(a)}}$$
converges in law on $`_\mathrm{R}`$ towards
$$\underset{0<t<a_\rho }{}\delta _{\mathrm{\Delta }B_t}=\underset{0<t<a_\rho }{}\delta _{\mathrm{\Delta }\sigma _t},$$
and in particular, when $`a\mathrm{}`$,
$$a_\rho ^1\underset{0<t<a_\rho }{}f(\mathrm{\Delta }X_t^{(a)})\stackrel{(\mathrm{d})}{}a_\rho ^1\underset{0<t<a_\rho }{}f(\mathrm{\Delta }\sigma _t).$$
(20)
Let us check that the variables
$$\left|\underset{0<t<a_\rho }{}f(\mathrm{\Delta }X_t^{(a)})\right|,a[a_\rho ,\mathrm{}[$$
are uniformly integrable. Let $`[u,v]`$ be a compact subinterval of $`]0,\mathrm{}[`$ such that the support of $`f`$ is contained in $`[u,v]`$. Denote by $`N_{[u,v]}^a`$ the number of jumps of the process $`X^{(a)}`$ with size in $`[u,v]`$. By classical results about processes with exchangeable increments, conditionally on $`N_{[u,v]}^a=n`$, the number
$$N_{[u,v]}^{(a,a_\rho )}:=\underset{0<t<a_\rho }{}\mathrm{๐}_{[u,v]}(\mathrm{\Delta }X_t^{(a)})$$
has a binomial $`(n,\frac{a_\rho }{a})`$ distribution. Notice that $`N_{[u,v]}^a\frac{a}{u}`$ since $`X_a^{(a)}=a`$. We see that $`N_{[u,v]}^{(a,a_\rho )}`$ is bounded above in distribution by a binomial $`([\frac{a}{u}],\frac{a_\rho }{a})`$ distribution, and the desired uniform integrability readily follows.
It then follows from (19) and (20) that
$$\underset{a\mathrm{}}{lim}E\left[\left|a_\rho ^1\underset{0<t<a_\rho }{}f(\mathrm{\Delta }X_t^{(a)})c\right|\right]=E\left[\left|a_\rho ^1\underset{0<t<a_\rho }{}f(\mathrm{\Delta }\sigma _t)c\right|\right]\rho .$$
Moreover, an easy exchangeability argument shows that we have also
$$\underset{a\mathrm{}}{lim\; sup}E\left[\left|a^1\underset{0<t<a}{}f(\mathrm{\Delta }X_t^{(a)})c\right|\right]\rho .$$
Since $`\rho `$ may be taken arbitrarily small, we have thus shown that
$$\underset{a\mathrm{}}{lim}a^1\underset{ta}{}f(\mathrm{\Delta }X_t^{(a)})=_{]0,\mathrm{}[}f(x)\lambda (dx),$$
in $`L^1`$ for every continuous function $`f`$ with compact support. The conclusion now follows by a standard argument. $`\mathrm{}`$
We will now show that the family $`(\lambda _t,t>0)`$ of Lรฉvy measures, which appears in Theorem 2, solves a certain coagulation equation with multiple collisions. To this end, we introduce the following additional assumption, which also plays a key role in the study of the genealogical structure of continuous-state branching processes (see e.g. ).
Assumption (E) The $`\mathrm{\Psi }`$-CSBP becomes extinct almost surely.
Equivalently, this assumption holds iff $`P[Z(t,x)=0]>0`$ for every $`t>0`$ and $`x0`$. By solving (3), it is easy to verify that Assumption (E) is equivalent to
$$_1^{\mathrm{}}\frac{du}{\mathrm{\Psi }(u)}<\mathrm{}.$$
(21)
In particular, Assumption (E) holds in the so-called stable case $`\mathrm{\Psi }(u)=u^\gamma `$, $`\gamma ]1,2[`$, that will be considered in Section 5 below.
From (4), we see that under Assumption (E) we have $`d_t=0`$, and the total mass $`\lambda _t(]0,\mathrm{}[)=\mathrm{log}P[Z(t,1)=0]`$ is finite for every $`t>0`$. Moreover the function $`t\lambda _t(]0,\mathrm{}[)`$ is nonincreasing.
We denote by $`C_{}(R_+)`$ the space of all bounded continuous functions $`f`$ on $`R_+`$ such that $`f(0)=0`$ and $`f(x)`$ has a limit as $`x+\mathrm{}`$. The space $`C_{}(R_+)`$ is equipped with the uniform norm, which is denoted by $`f`$. For every integer $`k2`$ and $`q>0`$, we denote by $`\mathrm{\Psi }^{(k)}(q)`$ the $`k`$-th derivative of $`\mathrm{\Psi }`$ at $`q`$. It is immediately checked that
$$\mathrm{\Psi }^{(k)}(q)=(1)^k\pi (dr)r^k\mathrm{e}^{qr}.$$
(22)
Obviously, $`(1)^k\mathrm{\Psi }^{(k)}(q)0`$ for every $`k2`$ and $`q>0`$.
###### Proposition 3
Under Assumption (E), for every $`fC_{}(R_+)`$, the function $`t\lambda _t,f`$ solves the equation
$$\frac{d\lambda _t,f}{dt}=\underset{k=2}{\overset{\mathrm{}}{}}\frac{(1)^k\mathrm{\Psi }^{(k)}(\lambda _t,1)}{k!}_{]0,\mathrm{}[^k}(f(x_1+\mathrm{}+x_k)(f(x_1)+\mathrm{}+f(x_k)))\lambda _t(dx_1)\mathrm{}\lambda _t(dx_k)$$
(23)
where the series in the right-hand side converges absolutely.
It is interesting to observe that (23) also holds when $`\mathrm{\Psi }(q)=cq^2`$ for some constant $`c>0`$. Take $`\mathrm{\Psi }(q)=\frac{1}{2}u^2`$ for definiteness (then the $`\mathrm{\Psi }`$-CSBP is the classical Feller diffusion) in such a way that (23) exactly reduces to (18) with $`K1`$. Then $`u_t(q)=2q(2+qt)^1`$, and it follows that
$$\lambda _t(dx)=\frac{4}{t^2}\mathrm{exp}(\frac{2x}{t})dx$$
(24)
so that the density of $`\lambda _t`$ is the classical solution, arising from infinitesimally small initial clusters, of the Smoluchovski equation (18) in the case $`K1`$ (cf Section 2.2 of ).
We can rewrite equation (23) in a somewhat more synthetic way by introducing the following notation. If $`\mu `$ is a measure on $`]0,\mathrm{}[`$ such that $`_{]0,\mathrm{}[}(1x)\mu (dx)<\mathrm{}`$, we write $`\mu ^{}`$ for the distribution on $`[0,\mathrm{}[`$ of the sum of the atoms of a Poisson random measure on $`]0,\mathrm{}[`$ with intensity $`\mu `$. Note that $`\mu ^{}`$ is a probability measure and that, by Campbellโs formula,
$$_{[0,\mathrm{}[}\mathrm{e}^{qx}\mu ^{}(dx)=\mathrm{exp}\left\{_{]0,\mathrm{}[}(1\mathrm{e}^{qx})\mu (dx)\right\},q0.$$
(25)
As we will see in the proof below, (23) follows from the equation
$$\frac{d\lambda _t,f}{dt}=_{]0,\mathrm{}[}\pi (da)\left((a\lambda _t)^{},fa\lambda _t,f\right).$$
(26)
Informally, we may think of $`\lambda _t(dx)`$ as the density at time $`t`$ of particles with size $`x`$ in some infinite system of particles. The right-hand side in (26) can be interpreted by saying at rate $`\pi (da)`$, a โquantityโ $`a`$ of particles coagulates at time $`t`$. More precisely, this โquantityโ is sampled in a Poissonian way, viewing at $`a\lambda _t`$ as an intensity measure for the sampling (so, loosely speaking, the particles involved into the coagulation are sampled uniformly at random amongst the particles present at time $`t`$).
As the proof below will show, (26) still holds without Assumption (E) at least for functions $`f`$ of the type $`f(x)=1\mathrm{exp}(qx)`$, provided that $`d_t=0`$ for every $`t>0`$ (recall from Silverstein that the latter holds whenever $`_{]0,1[}r\pi (dr)=\mathrm{}`$). In that case however, the measures $`\lambda _t`$ may be infinite, and then coagulations involve infinitely many components, so that one cannot write an equation of the form (23).
Proof: We first prove (26). For $`q>0`$, let $`f_{(q)}C_{}(R_+)`$ be defined by $`f_{(q)}(x)=1\mathrm{e}^{qx}`$. By (25) and (4),
$$(a\lambda _t)^{},f_{(q)}=1\mathrm{exp}\left(a\lambda _t(dr)(1\mathrm{e}^{qr})\right)=1\mathrm{exp}(au_t(q)).$$
On the other hand, by (4) again,
$$\lambda _t,f_{(q)}=u_t(q).$$
Thus when $`f=f_{(q)}`$ the right-hand side of (26) makes sense and is equal to
$$_{]0,\mathrm{}[}\pi (da)\left(1\mathrm{exp}(au_t(q))au_t(q)\right)=\mathrm{\Psi }(u_t(q)).$$
Therefore (26) reduces to (3) in that case. Note that we have not used Assumption (E) at this stage (except for the fact that $`d_t=0`$ for every $`t>0`$).
Denote by $``$ the subspace of $`C_{}(R_+)`$ that consists of linear combinations of the functions $`f_{(q)}`$. Then $``$ is dense in $`C_{}(R_+)`$. Obviously, for every $`f`$, (26) holds, and the right-hand side of (26) is a continuous function of $`t]0,\mathrm{}[`$. Fix $`fC_{}(R_+)`$ and a sequence $`(f_n)_{n1}`$ in $``$ that converges to $`f`$. If we also fix $`0<\epsilon <t`$, we have for every $`n1`$,
$$\lambda _t,f_n=\lambda _\epsilon ,f_n+_\epsilon ^t๐s\pi (da)\left((a\lambda _s)^{},f_na\lambda _s,f_n\right).$$
(27)
Plainly, for every $`s>0`$,
$$\lambda _s,f_n\underset{n\mathrm{}}{}\lambda _s,f\text{and}(a\lambda _s)^{},f_n\underset{n\mathrm{}}{}(a\lambda _s)^{},f.$$
We claim that there exists a constant $`C_\epsilon `$ such that, for every $`s\epsilon `$ and $`n1`$, and every $`hC_{}(R_+)`$,
$$|(a\lambda _s)^{},ha\lambda _s,h|C_\epsilon (a^2a)h.$$
(28)
As the quantities $`\lambda _s,1`$, $`s[\epsilon ,\mathrm{}[`$ are bounded above, it is clearly enough to consider $`a1`$. Since $`h(0)=0`$, the definition of $`(a\lambda _s)^{}`$ immediately gives
$$(a\lambda _s)^{},h=\mathrm{e}^{a\lambda _s,1}a\lambda _s,h+O(a^2h)$$
where the remainder $`O(a^2h)`$, which corresponds to the event that a Poisson measure with intensity $`a\lambda _s`$ has at least two atoms, is uniform in $`hC_{}(R_+)`$ and $`s\epsilon `$. The estimate (28) follows.
Using (28) and dominated convergence, we get
$$\underset{n\mathrm{}}{lim}\pi (da)\left((a\lambda _s)^{},f_na\lambda _s,f_n\right)=\pi (da)\left((a\lambda _s)^{},fa\lambda _s,f\right)$$
(29)
uniformly in $`s[\epsilon ,\mathrm{}[`$, and the right-hand side of (29) is a continuous function of $`s`$. Equation (26) in the general case follows by passing to the limit $`n\mathrm{}`$ in (27).
Then, to derive (26) from (23), we write
$`{\displaystyle _{]0,\mathrm{}[}}\pi (da)\left((a\lambda _t)^{},fa\lambda _t,f\right)`$
$`={\displaystyle \pi (da)\left(\left(\underset{k=1}{\overset{\mathrm{}}{}}\frac{a^k}{k!}\mathrm{e}^{a\lambda _t,1}f(x_1+\mathrm{}+x_k)\lambda _t(dx_1)\mathrm{}\lambda _t(dx_k)\right)a\lambda _t,f\right)}`$
$`={\displaystyle \pi (da)\underset{k=1}{\overset{\mathrm{}}{}}\frac{a^k}{k!}\mathrm{e}^{a\lambda _t,1}(f(x_1+\mathrm{}+x_k)(f(x_1)+\mathrm{}+f(x_k)))\lambda _t(dx_1)\mathrm{}\lambda _t(dx_k)}.`$
Notice that the term $`k=1`$ in the last series vanishes. Moreover, bounding the other terms by their absolute value gives a convergent series, whose sum is integrable with respect to $`\pi (da)`$. Hence we may interchange the sum and the integral with respect to $`\pi (da)`$, and we get the statement of the proposition from (26). $`\mathrm{}`$
Remark. To conclude this section, let us observe that Assumption (E) is closely related to the property for a $`\mathrm{\Lambda }`$-coalescent to come down from infinity (cf Pitman and Schweinsberg ). Let $`\nu `$ denote a $`\sigma `$-finite measure on $`]0,1]`$ such that $`_{]0,1]}r^2\nu (dr)<\mathrm{}`$, and let $`\mathrm{\Lambda }(dx)=x^2\nu (dx)`$. Let $`\mathrm{\Psi }`$ be given by (1) with $`\pi =\nu `$ (and $`\beta =0`$). Then the $`\mathrm{\Lambda }`$-coalescent comes down from infinity if and only if the $`\mathrm{\Psi }`$-CSBP becomes extinct almost surely. To see this, recall from Schweinsberg that a necessary and sufficient condition for the $`\mathrm{\Lambda }`$-coalescent to come down from infinity is
$$\underset{b=2}{\overset{\mathrm{}}{}}\left(\underset{k=2}{\overset{b}{}}(k1)\left(\begin{array}{cc}b\hfill & \\ k\hfill & \end{array}\right)r^k(1r)^{bk}\nu (dr)\right)^1<\mathrm{}.$$
(30)
Using the binomial formula, we can rewrite this condition as
$$\underset{b=2}{\overset{\mathrm{}}{}}\left((br1+(1r)^b)\nu (dr)\right)^1<\mathrm{},$$
or equivalently, if we put $`\mathrm{\Phi }(q)=(qr1+(1r)^q)\nu (dr)`$ for every real $`q1`$,
$$_2^{\mathrm{}}\frac{dq}{\mathrm{\Phi }(q)}<\mathrm{}.$$
(31)
(note that the function $`\mathrm{\Phi }`$ is nondecreasing on $`[1,\mathrm{}[`$). Simple estimates give the existence of a constant $`c]0,1[`$ such that, for every $`q2`$,
$$c\mathrm{\Psi }(q)\mathrm{\Phi }(q)\mathrm{\Psi }(q).$$
It follows that (30) and (21) are equivalent. In the spirit of the present work, it would be interesting to give a direct probabilistic proof of the equivalence between the property for a $`\mathrm{\Lambda }`$-coalescent to come down from infinity and Assumption (E) for the associated branching process.
## 5 Small time behavior of flows and coalescents
In this section, we fix a measure $`\nu `$ on $`]0,1]`$ such that $`r^2\nu (dr)<\mathrm{}`$ and we consider the associated generalized Fleming-Viot process $`(F_t,t0)`$.
From now on until the end of the section, we make the following assumption on $`\nu `$.
Assumption (A). The function $`\nu ([\epsilon ,1])`$ is regularly varying with index $`\gamma `$ as $`\epsilon 0`$, for some $`\gamma ]1,2[`$.
As a consequence, there exists a function $`L(\epsilon )`$, $`\epsilon ]0,1]`$ that is slowly varying as $`\epsilon 0`$, such that, for every $`\epsilon ]0,1]`$,
$$\nu ([\epsilon ,1])=\epsilon ^\gamma L(\epsilon ).$$
Fix $`\epsilon _0>0`$ such that $`\nu ([\epsilon _0,1])>0`$. For $`\epsilon ]0,\epsilon _0]`$ we have $`L(\epsilon )>0`$ and so we can set
$$F_t^\epsilon (x)=\frac{1}{\epsilon }F_{L(\epsilon )^1\epsilon ^{\gamma 1}t}(\epsilon x)$$
for $`0x\epsilon ^1`$ and $`t0`$. We also let $`\nu _\epsilon `$ be the measure on $`[0,\epsilon ^1]`$ defined by
$$\nu _\epsilon (dr)\phi (r)=L(\epsilon )^1\epsilon ^\gamma \nu (dr)\phi (\frac{r}{\epsilon }).$$
A simple scaling transformation shows that for every $`(x_1,\mathrm{},x_p)๐_p^{1/\epsilon }`$, $`(F_t^\epsilon (x_1),\mathrm{},F_t^\epsilon (x_p))`$ is a purely discontinuous martingale, with values in $`๐_p^{1/\epsilon }`$, and the compensator of its jump measure is
$$dtR_\epsilon (F_t^\epsilon (x_1),\mathrm{},F_t^\epsilon (x_p);dz_1,\mathrm{},dz_p)$$
where
$`{\displaystyle R_\epsilon (y_1,\mathrm{},y_p;dz_1,\mathrm{},dz_p)\phi (z_1,\mathrm{},z_p)}`$
$`={\displaystyle \nu _\epsilon (dr)_0^{1/\epsilon }๐u\phi (r(\mathrm{๐}_{\{uy_1\}}\epsilon y_1),\mathrm{},r(\mathrm{๐}_{\{uy_p\}}\epsilon y_p))}.`$ (32)
Let $`\pi _\gamma `$ be the measure on $`]0,\mathrm{}[`$ such that $`\pi _\gamma (]a,\mathrm{}[)=a^\gamma `$ for every $`a>0`$, and let
$$\mathrm{\Psi }_\gamma (q)=\pi _\gamma (dr)(\mathrm{e}^{qr}1+qr)=\frac{\mathrm{\Gamma }(2\gamma )}{\gamma 1}q^\gamma .$$
We let $`(Z(t,x),t0,x0)`$ be the flow of continuous-state branching processes constructed in Section 2, with $`\mathrm{\Psi }=\mathrm{\Psi }_\gamma `$.
###### Theorem 3
Under Assumption (A), for every $`(x_1,\mathrm{},x_p)๐_p`$,
$$((F_t^\epsilon (x_1),\mathrm{},F_t^\epsilon (x_p));t0)\underset{\epsilon 0}{\overset{(\mathrm{d})}{}}((Z(t,x_1),\mathrm{},Z(t,x_p));t0)$$
in the Skorokhod space $`D(R_+,R^p)`$.
Proof: This is a simple consequence of Theorem 1, or rather of its proof. Indeed, we immediately see that the kernel $`R_\epsilon (y_1,\mathrm{},y_p;dz_1,\mathrm{},dz_p)`$ coincides with $`R^{(1/\epsilon )}(y_1,\mathrm{},y_p;dz_1,\mathrm{},dz_p)`$ defined in (3), provided we take $`\nu ^{(1/\epsilon )}=\nu _\epsilon `$. From the observation at the beginning of the proof of Theorem 1, we see that Theorem 3 will follow if we can check that Assumption (H) holds in the present setting, that is if
$$\underset{\epsilon 0}{lim}(rr^2)\nu _\epsilon (dr)=(rr^2)\pi _\gamma (dr)$$
(33)
in the sense of weak convergence in $`_\mathrm{F}`$.
In order to prove (33), first note that when $`\epsilon 0+`$,
$$_{[0,\epsilon ]}x^2\nu (dx)=2_0^\epsilon y\nu ([y,1])๐y\epsilon ^2\nu ([\epsilon ,1])\frac{\gamma }{2\gamma }\epsilon ^{2\gamma }L(\epsilon ),$$
where the equivalence follows from Assumption (A) and a classical property of integrals of regularly varying functions. We immediately deduce that
$$\underset{\epsilon 0}{lim}r^2\nu _\epsilon (dr)=r^2\pi _\gamma (dr)$$
(34)
in the sense of vague convergence in the space of Radon measures on $`[0,\mathrm{}[`$.
Next, note that
$$\nu _\epsilon (dr)(rra)=_a^{\mathrm{}}dr\nu _\epsilon ([r,\mathrm{}[)=\epsilon ^\gamma L(\epsilon )^1_a^{\mathrm{}}dr\nu ([r\epsilon ,1])\underset{a\mathrm{}}{}0$$
(35)
uniformly in $`\epsilon ]0,\epsilon _0]`$. From (34) and (35) the family $`((rr^2)\nu _\epsilon (dr),0<\epsilon <1)`$ is tight for the weak topology in $`_\mathrm{F}`$. Together, with (34), this establishes the weak convergence (33). $`\mathrm{}`$
Remark. Suppose that $`(F_t,t0)`$ is the flow of bridges associated with the Kingman coalescent, corresponding to $`\mathrm{\Lambda }=\delta _0`$ in our notation (cf Section 4 in ). If we fix $`(y_1,\mathrm{},y_p)๐_p^1`$, the process $`(F_t(y_1),\mathrm{},F_t(y_p))`$ is a diffusion process in $`๐_p^1`$ with generator
$$๐g(x)=\frac{1}{2}\underset{i,j=1}{\overset{p}{}}x_{ij}(1x_{ij})\frac{^2g}{x_ix_j}(x)$$
(see Theorem 3 in ). Putting $`F_t^\epsilon (x)=\frac{1}{\epsilon }F_{\epsilon t}(\epsilon x)`$, it is a simple matter to verify that our Theorem 3 still holds in that setting, provided we let $`(Z(t,x),t0,x0)`$ be the flow associated with the Feller diffusion ($`\mathrm{\Psi }(q)=\frac{1}{2}q^2`$). Indeed, if we specialize to the case $`p=1`$ and if we let $`(B_t,t0)`$ be a standard linear Brownian motion, this is just saying that, for the Fisher-Wright diffusion $`(X_t(x),t0)`$ solving
$$dX_t=\sqrt{X_t(1X_t)}dB_t,X_0=x,$$
the rescaled processes $`X_t^\epsilon :=\frac{1}{\epsilon }X_{\epsilon t}(\epsilon x)`$ converge in distribution as $`\epsilon 0`$ towards the Feller diffusion $`Y_t(x)`$ solving
$$dY_t=\sqrt{Y_t}dB_t,Y_0=x.$$
We will now use Theorem 3 to derive precise information on the sizes of blocks in a $`\mathrm{\Lambda }`$-coalescent (for $`\mathrm{\Lambda }(dr)=r^2\nu (dr)`$) in small time. As previously, we denote by $`\lambda _1(dr)`$ the Lรฉvy measure of the subordinator $`(Z(1,x),x0)`$. We have for every $`q0`$
$$\mathrm{exp}x(1\mathrm{e}^{qr})\lambda _1(dr)=E(\mathrm{exp}qZ(1,x))=\mathrm{exp}xu_1(q)$$
and the function $`u_1(q)`$ can be calculated from equation (3), with $`\mathrm{\Psi }=\mathrm{\Psi }_\gamma `$. It follows that
$$(1\mathrm{e}^{qr})\lambda _1(dr)=(\mathrm{\Gamma }(2\gamma )+q^{1\gamma })^{1/(1\gamma )}$$
and in particular, the total mass of $`\lambda _1`$ is
$$(\mathrm{\Gamma }(2\gamma ))^{1/(1\gamma )}.$$
We will need the fact that $`\lambda _1`$ has no atoms. An easy way to derive this property is to argue by contradiction as follows. Suppose that $`a>0`$ is an atom of $`\lambda _1`$. From the Lรฉvy-Khintchin decomposition of $`Z(t,x)`$ (see the discussion after (4)), it follows that $`a`$ is also an atom of the distribution of $`Z(1,x)`$, for every $`x>0`$. By a simple scaling argument, for every $`s>0`$, the image of $`\lambda _1(dr)`$ under the mapping $`rs^{1/(\gamma 1)}r`$ is $`s^{1/(\gamma 1)}\lambda _s(dr)`$. Therefore, for every $`s]0,1[`$, $`s^{1/(\gamma 1)}a`$ is also an atom of $`\lambda _s`$, hence of the distribution of $`Z(s,x)`$ for every $`x>0`$. However, applying the Markov property to the process $`(Z(t,1))_{t0}`$ at time $`1s`$, this would imply that for every $`s]0,1[`$, $`s^{1/(\gamma 1)}a`$ is an atom of the distribution of $`Z(1,1)`$, which is absurd.
We set $`g(\epsilon )=L(\epsilon )^1\epsilon ^{\gamma 1}`$ for every $`\epsilon ]0,\epsilon _0]`$.
###### Theorem 4
Assume that (A) holds and let $`\mathrm{\Lambda }(dr)=r^2\nu (dr)`$. For every $`t0`$ and $`r[0,\mathrm{}]`$, denote by $`N_t(]0,r[)`$ the number of blocks at time $`t`$ with frequencies less than $`r`$ in a $`\mathrm{\Lambda }`$-coalescent started from the partition of $`N`$ in singletons. Then,
$$\underset{x]0,\mathrm{}[}{sup}|\epsilon N_{g(\epsilon )}(]0,\epsilon x[)\lambda _1(]0,x[)|\underset{\epsilon 0}{}0$$
in probability.
Again Theorem 4 is a generalization of classical results for the Kingman coalescent. In that case, one has
$$\underset{x]0,\mathrm{}[}{sup}|\epsilon N_\epsilon (]0,\epsilon x[)2(12\mathrm{e}^{2x})|\underset{\epsilon 0}{}0$$
almost surely (cf Section 4.2 of ). This is consistent with Theorem 4 since in the case $`\mathrm{\Psi }(q)=\frac{1}{2}q^2`$, (24) shows that
$$2(12\mathrm{e}^{2x})=_0^x4\mathrm{e}^{2x}dx=\lambda _1(]0,x[).$$
Proof: By the results of recalled at the beginning of subsection 4.2, we know that, for each $`t0`$ the collection $`(N_t(]0,r[),r0)`$ has the same distribution as
$$\left(\underset{0<u<1}{}\mathrm{๐}_{\{0<F_t(u)F_t(u)r\}},r0\right),$$
where $`(F_t,t0)`$ is the generalized Fleming-Viot process associated with $`\nu `$. It then follows from our definitions that
$$(\epsilon N_{g(\epsilon )}(]0,x\epsilon [),x0)\stackrel{(\mathrm{d})}{=}(\epsilon \underset{0<u<1/\epsilon }{}\mathrm{๐}_{\{0<F_1^\epsilon (u)F_1^\epsilon (u)x\}},x0).$$
By combining Theorem 3 and Lemma 2, we get that
$$\epsilon \underset{0<u<1/\epsilon }{}\delta _{F_t^\epsilon (u)F_t^\epsilon (u)}\underset{\epsilon 0}{}\lambda _1$$
(36)
in probability in $`_\mathrm{R}`$. This is indeed the same result as Theorem 2 in our present setting. The preceding convergence is not quite sufficient to conclude: Recalling that $`\lambda _1`$ has no atoms and using Diniโs theorem, we see that the statement of the theorem will follow if we can prove that the convergence (36) holds in the sense of weak convergence in the space $`_\mathrm{F}`$. To get this strengthening of (36), it suffices to prove the convergence of the total masses. Therefore the proof of Theorem 4 will be complete once we have established the following lemma.
###### Lemma 3
We have
$$\underset{\epsilon 0}{lim}\epsilon N_{g(\epsilon )}(]0,\mathrm{}[)=\lambda _1(]0,\mathrm{}[)=(\mathrm{\Gamma }(2\gamma ))^{1/(1\gamma )},$$
in probability.
Remark. The recent paper gives closely related results that were obtained independently of the present work.
Proof: Write $`N_t=N_t(]0,\mathrm{}[)`$ to simplify notation. Then, for every $`t0`$ and $`x]0,1]`$, we have
$$E[x^{N_t}]=P[F_t(x)=1]$$
(cf formula (8) in ). By exchangeability,
$$P[F_t(x)=1]=P[F_t(x)=F_t(1)]=P[F_t(1x)=0].$$
Hence, for $`x]0,1[`$,
$$P[F_t(x)=0]=E[(1x)^{N_t}],$$
and it follows that
$$P[F_1^\epsilon (x)=0]=E[(1\epsilon x)^{N_{g(\epsilon )}}].$$
From the convergence in distribution in Theorem 3, we have for every $`x>0`$,
$$\underset{\epsilon 0}{lim\; sup}P[F_1^\epsilon (x)=0]P[Z(1,x)=0]=\mathrm{exp}x\lambda _1(]0,\mathrm{}[).$$
We have thus obtained that, for every $`x>0`$,
$$\underset{\epsilon 0}{lim\; sup}E[(1\epsilon x)^{N_{g(\epsilon )}}]\mathrm{exp}x\lambda _1(]0,\mathrm{}[).$$
By standard arguments, this implies that for every $`\eta >0`$,
$$\underset{\epsilon 0}{lim}P[\epsilon N_{g(\epsilon )}<\lambda _1(]0,\mathrm{}[)\eta ]=0.$$
(37)
To complete the proof, we need to verify that we have also, for every $`\eta >0`$,
$$\underset{\epsilon 0}{lim}P[\epsilon N_{g(\epsilon )}>\lambda _1(]0,\mathrm{}[)+\eta ]=0.$$
(38)
Fome now on, we fixe $`\eta >0`$ and we prove (38). We will use a different method based on the knowledge of the law of the process of the number of blocks in a $`\mathrm{\Lambda }`$-coalescent. For every integer $`n1`$, write $`N_t^n`$ for the number of blocks at time $`t`$ in a $`\mathrm{\Lambda }`$-coalescent started initially with $`n`$ blocks. Then according to Pitman (Section 3.6), the process $`(N_t^n,t0)`$ is a time-homogeneous Markov chain with values in $`\{1,2,\mathrm{},n\}`$, with only downward jumps, such that for $`2kbn`$, the rate of jumps from $`b`$ to $`bk+1`$ is
$$\alpha _{b,k}=\left(\begin{array}{c}b\hfill \\ k\hfill \end{array}\right)_{]0,1]}x^k(1x)^{bk}\nu (dx).$$
The total rate of jumps from $`b`$ is thus
$$\alpha _b=\underset{k=2}{\overset{b}{}}\alpha _{b,k}=_{]0,1]}(1(1x)^bb(1x)^{b1})\nu (dx).$$
###### Lemma 4
Under Assumption (A), we have
$$\underset{b+\mathrm{}}{lim}(b^\gamma L(1/b))^1\alpha _b=\mathrm{\Gamma }(2\gamma )$$
and, for every integer $`k2`$,
$$\underset{b+\mathrm{}}{lim}(b^\gamma L(1/b))^1\alpha _{b,k}=\frac{\gamma \mathrm{\Gamma }(k\gamma )}{k!}.$$
We leave the easy proof to the reader. Note that
$$\underset{k=2}{\overset{\mathrm{}}{}}\frac{\gamma \mathrm{\Gamma }(k\gamma )}{k!}=\mathrm{\Gamma }(2\gamma ).$$
(39)
This is easily proved by using the definition of the function $`\mathrm{\Gamma }`$ and then an integration by parts. Similarly, we have
$$\underset{k=2}{\overset{\mathrm{}}{}}\frac{\gamma \mathrm{\Gamma }(k\gamma )}{k!\mathrm{\Gamma }(2\gamma )}(k1)=\frac{1}{\gamma 1}.$$
(40)
Let us fix $`\rho ]0,1/8[`$ sufficiently small so that
$$(\mathrm{\Gamma }(2\gamma )^{1/(1\gamma )}+\eta )^{1\gamma }<(16\rho )\mathrm{\Gamma }(2\gamma ).$$
Thanks to (39) and (40), we may choose an integer $`K2\epsilon _0^1`$ sufficiently large so that
$$\frac{1}{\mathrm{\Gamma }(2\gamma )}\underset{k=2}{\overset{K}{}}\frac{\gamma \mathrm{\Gamma }(k\gamma )}{k!}1\rho $$
and
$$\underset{k=2}{\overset{K}{}}\frac{\gamma \mathrm{\Gamma }(k\gamma )}{k!\mathrm{\Gamma }(2\gamma )}(k1)\frac{1}{\gamma 1}\rho .$$
(41)
Then, for every $`k\{2,3,\mathrm{},K\}`$, we may choose $`\rho _k]0,\gamma \mathrm{\Gamma }(k\gamma )/k![`$ sufficiently small so that
$$\frac{1}{\mathrm{\Gamma }(2\gamma )}\underset{k=2}{\overset{K}{}}(k1)\rho _k<\rho .$$
(42)
Now set
$$\beta _{b,k}=\left(\frac{\gamma \mathrm{\Gamma }(k\gamma )}{k!}\rho _k\right)b^\gamma L(\frac{1}{b})$$
for $`bK`$ and $`k\{2,3,\mathrm{},K\}`$. We also put
$$\beta _b=\underset{k=2}{\overset{K}{}}\beta _{b,k}.$$
Notice that
$$\beta _p=\underset{k=2}{\overset{K}{}}\left(\frac{\gamma \mathrm{\Gamma }(k\gamma )}{k!}\rho _k\right)b^\gamma L(\frac{1}{b})(12\rho )\mathrm{\Gamma }(2\gamma )b^\gamma L(\frac{1}{b}).$$
(43)
By Lemma 4, we can choose an integer $`B2K`$ sufficiently large so that, for every $`bBK`$, $`b^{}\{b,b+1,\mathrm{},b+K\}`$ and $`k\{2,\mathrm{},K\}`$, one has
$$\beta _{b^{},k}\alpha _{b,k}.$$
(44)
Denote by $`(U_t^n)_{t0}`$ the continuous-time Markov chain with values in $`N`$, with initial value $`U_0^n=n`$, which is absorbed in the set $`\{1,\mathrm{},B1\}`$ and has jump rate $`\beta _{b,k}`$ from $`b`$ to $`bk+1`$ when $`bB`$ and $`k\{2,3,\mathrm{},K\}`$. Fix $`nB`$. Then thanks to inequality (44), we can couple the Markov chains $`(U_t^n)_{t0}`$ and $`(N_t^n)_{t0}`$ in such a way that
$$U_t^nN_t^n,\text{ for every }tT_B^n:=inf\{s:U_s^n<B\}.$$
Now it is easy to describe the behavior of the Markov chain $`(U_t^n)`$. Note that for $`k\{2,\mathrm{},K\}`$ and $`bK`$ the ratio $`\beta _{b,k}/\beta _b`$ does not depend on $`b`$. Then denote by $`S_i=\xi _1+\mathrm{}+\xi _i`$ ($`i=0,1,2,\mathrm{}`$) a discrete random walk on the nonnegative integers started from the origin and with jump distribution
$$P[\xi _i=k1]=\frac{\beta _{b,k}}{\beta _b}=\frac{(\gamma \mathrm{\Gamma }(k\gamma )/k!)\rho _k}{_{\mathrm{}=2}^K((\gamma \mathrm{\Gamma }(\mathrm{}\gamma )/\mathrm{}!)\rho _{\mathrm{}})},2kK.$$
From (41) and (42) we have
$$E[\xi _i]\frac{1}{\gamma 1}2\rho .$$
(45)
Let $`๐_0,๐_1,\mathrm{}`$ be a sequence of independent exponential variables with mean $`1`$, which are also independent of the random walk $`(S_i)_{i0}`$. We can construct the Markov chain $`(U_t^n)`$ by setting:
$$\begin{array}{cc}U_t^n=n\hfill & \text{if }0t<\frac{๐_0}{\beta _n}\hfill \\ \multicolumn{2}{c}{}\\ U_t^n=nS_1\hfill & \text{if }\frac{๐_0}{\beta _n}t<\frac{๐_0}{\beta _n}+\frac{๐_1}{\beta _{nS_1}}\hfill \end{array}$$
and more generally,
$$U_t^n=nS_p\text{if }\frac{๐_0}{\beta _n}+\frac{๐_1}{\beta _{nS_1}}+\mathrm{}+\frac{๐_{p1}}{\beta _{nS_{p1}}}t<\frac{๐_0}{\beta _n}+\frac{๐_1}{\beta _{nS_1}}+\mathrm{}+\frac{๐_p}{\beta _{nS_p}}$$
provided $`pp_B^n:=inf\{i:nS_i<B\}`$.
Recall that our goal is to prove (38). To this end, note that for $`a>B`$,
$$P[N_{g(\epsilon )}^n>a]P[U_{g(\epsilon )}^n>a]P\left[g(\epsilon )\frac{๐_0}{\beta _n}+\frac{๐_1}{\beta _{nS_1}}+\mathrm{}+\frac{๐_{p_a^n}}{\beta _{nS_{p_a^n}}}\right]$$
(46)
where $`p_a^n:=inf\{i:nS_i<a\}`$.
###### Lemma 5
For $`\epsilon >0`$ set $`a(\epsilon )=(\lambda _1(]0,\mathrm{}[)+\eta )/\epsilon `$. Then,
$$\underset{\epsilon 0}{lim}\left(\underset{na(\epsilon )}{sup}P\left[g(\epsilon )\frac{๐_0}{\beta _n}+\frac{๐_1}{\beta _{nS_1}}+\mathrm{}+\frac{๐_{p_{a(\epsilon )}^n}}{\beta _{nS_{p_{a(\epsilon )}^n}}}\right]\right)=0.$$
The desired bound (38) immediately follows from Lemma 5. Indeed standard properties of $`\mathrm{\Lambda }`$-coalescent give
$$P[\epsilon N_{g(\epsilon )}>\lambda _1(]0,\mathrm{}[)+\eta ]=\underset{n\mathrm{}}{lim}P[\epsilon N_{g(\epsilon )}^n>\lambda _1(]0,\mathrm{}[)+\eta ]=\underset{n\mathrm{}}{lim}P[N_{g(\epsilon )}^n>a(\epsilon )]$$
and by combining (46) and Lemma 5, we see that the latter quantity tends to $`0`$ as $`\epsilon 0`$.
Proof of Lemma 5: By (43), we have for $`a>B`$,
$$\frac{๐_0}{\beta _n}+\frac{๐_1}{\beta _{nS_1}}+\mathrm{}+\frac{๐_{p_a^n}}{\beta _{nS_{p_a^n}}}((12\rho )\mathrm{\Gamma }(2\gamma ))^1\underset{i=0}{\overset{p_a^n}{}}\frac{๐_i}{(nS_i)^\gamma L(\frac{1}{nS_i})}.$$
(47)
Note that
$$E\left[\underset{i=0}{\overset{p_a^n}{}}\frac{๐_i}{(nS_i)^\gamma L(\frac{1}{nS_i})}|S_i,i0\right]=\underset{i=0}{\overset{p_a^n}{}}\frac{1}{(nS_i)^\gamma L(\frac{1}{nS_i})}.$$
Let $`m2`$ be an integer. For $`a>B`$ and $`n>ma`$, a trivial bound shows that
$$a^{\gamma 1}L(\frac{1}{a})\underset{i=0}{\overset{p_{ma}^n}{}}\frac{1}{(nS_i)^\gamma L(\frac{1}{nS_i})}a^{\gamma 1}L(\frac{1}{a})\underset{j=[ma]K}{\overset{\mathrm{}}{}}\frac{1}{j^\gamma L(\frac{1}{j})}$$
and the right-hand side tends to $`0`$ as $`m\mathrm{}`$, uniformly in $`a>B`$. On the other hand, an easy argument using the law of large numbers for the sequence $`(S_i)_{i0}`$ shows that, for each fixed $`m2`$,
$$\underset{a\mathrm{}}{lim}\left(\underset{n>ma}{sup}E\left[\left|a^{\gamma 1}L(\frac{1}{a})\underset{i=p_{ma}^n}{\overset{p_a^n}{}}\frac{1}{(nS_i)^\gamma L(\frac{1}{nS_i})}\frac{1}{E[\xi _1]}_1^m\frac{dx}{x^\gamma }\right|\right]\right)=0.$$
Now recall the bound (45) for $`E[\xi _1]`$. It follows from the preceding considerations that
$$\underset{a\mathrm{}}{lim}\left(\underset{n>a}{sup}P\left[a^{\gamma 1}L(\frac{1}{a})\underset{i=0}{\overset{p_a^n}{}}\frac{1}{(nS_i)^\gamma L(\frac{1}{nS_i})}>\frac{1}{13\rho }\right]\right)=0.$$
(48)
Now we can also get an estimate for the conditional variance
$`\mathrm{var}\left({\displaystyle \underset{i=0}{\overset{p_a^n}{}}}{\displaystyle \frac{๐_i}{(nS_i)^\gamma L(\frac{1}{nS_i})}}|S_i,i0\right)`$ $`=`$ $`{\displaystyle \underset{i=0}{\overset{p_a^n}{}}}{\displaystyle \frac{1}{(nS_i)^{2\gamma }L(\frac{1}{nS_i})^2}}`$
$``$ $`{\displaystyle \underset{j=[aK]}{\overset{n}{}}}{\displaystyle \frac{1}{j^{2\gamma }L(\frac{1}{j})^2}}`$
$``$ $`Ca^{12\gamma }L({\displaystyle \frac{1}{a}})^2`$
for some constant $`C`$ independent of $`a`$ and $`n`$. From this estimate, (48) and an application of the Bienaymรฉ-Cebycev inequality, we get
$$\underset{a\mathrm{}}{lim}\left(\underset{n>a}{sup}P\left[a^{\gamma 1}L(\frac{1}{a})\underset{i=0}{\overset{p_a^n}{}}\frac{๐_i}{(nS_i)^\gamma L(\frac{1}{nS_i})}>\frac{1}{14\rho }\right]\right)=0.$$
(49)
Recalling (47), we arrive at
$$\underset{\epsilon 0}{lim}\left(\underset{na(\epsilon )}{inf}P\left[\frac{๐_0}{\beta _n}+\frac{๐_1}{\beta _{nS_1}}+\mathrm{}+\frac{๐_{p_{a(\epsilon )}^n}}{\beta _{nS_{p_{a(\epsilon )}^n}}}\frac{a(\epsilon )^{1\gamma }L(\frac{1}{a(\epsilon )})^1}{(12\rho )(14\rho )\mathrm{\Gamma }(2\gamma )}\right]\right)=1.$$
However, from our choice of $`\rho `$, we have for $`\epsilon `$ sufficiently small
$$g(\epsilon )>\frac{a(\epsilon )^{1\gamma }L(\frac{1}{a(\epsilon )})^1}{(12\rho )(14\rho )\mathrm{\Gamma }(2\gamma )},$$
and this completes the proof. $`\mathrm{}`$
Remark. It is rather unfortunate that the simple argument we used to derive (37) does not apply to (38). On the other hand, it is interesting to observe that the techniques involved in our proof of (38) would become more complicated if we were trying to use them to get (37).
|
warning/0506/physics0506098.html
|
ar5iv
|
text
|
# Untitled Document
Stock mechanics: predicting recession in S&P500, DJIA, and NASDAQ
รa lar Tuncay
Department of Physics, Middle East Technical University
06531 Ankara, Turkey
caglart@metu.edu.tr
Abstract
An original method, assuming potential and kinetic energy for prices and conservation of their sum is developed for forecasting exchanges. Connections with power law are shown. Semiempirical applications on S&P500, DJIA, and NASDAQ predict a coming recession in them. An emerging market, Istanbul Stock Exchange index ISE-100 is found involving a potential to continue to rise.
Keywords: Potential and kinetic energy; Equations of motion; Power law; Oscillations; Crashes; Portfolio growths
PACS numbers: 89.65.Gh
1. Introduction
As is well known, prices, market indices evolve from one mood to another in time. They may be in calm oscillatory fluctuation mood for some time, then they may increase (or decrease), then again turn into a new oscillatory period, then a crash or a crisis may come, and finally, the current era closes up. Then a new one takes stage; new conditions and new formations take place, so forth ad infinitum. The time terms between long range drastic changes are named as era in the present work, and comparatively medium or short range characteristic time periods in eras are cited as epoch. The whole past and future history of exchanges may be far from obeying analytical functions. Time series may be predicted epoch by epoch, as presently proposed, in terms of some simple analytical functions. These equations of motion for prices are derived in the next section and power law connections are displayed in the third one. In the fourth section three indices S&P500, DJIA, and NASDAQ from New York Stock Exchange (NYSE) and ISE-100 from Istanbul Stock Exchange (ISE) are predicted by the present method. Last section is devoted to conclusion.
2. Potential and kinetic energies and conservation of their sum
A potential energy (taking mass as unity, see <sup>NOTE</sup><sup>1</sup><sup>1</sup>1<sup>NOTE</sup> Some kind of inertia as โresistanceโ against any impact to move the price do exist in time charts of shares.\[1-4, and references therein\]. Yet, not any reasonable โforce equationโ connecting inertia and accelaration of stock prices is proposed. In any case, whether massless potential and kinetic energies are defined, or they are defined per mass, or even mass is taken as unity in them will not be worth much in the present scheme.) in terms of difference in prices ($`\chi =\chi `$(t)) may be defined as
U($`\chi `$$`\chi _{av})`$ = h ($`\chi `$$`\chi _{av})^\alpha `$ , (1)
where h and $`\alpha `$ are some price independent parameters designating the current epoch, and $`\chi _{av}`$ is some time average of prices, which defines the zero-potential-level. Eq. (1) describes a kind of โgravitationalโ potential (as on the earth surface) for $`\alpha `$=1, where h becomes the gravitational (anti-gravitational) constant (g) with 0ยกh=g (0ยฟh=โg). Whereas for $`\alpha `$=2 we have a spring-mass potential energy with a spring constant (Hooke constant) equal to 2h. It is worth to underline that, the parameters of Eq. (1) i.e., h, $`\alpha `$, and $`\chi _{av}`$ may differ from one epoch to the other for any share and also from one share to another in any epoch. Note that, potential energy in Eq. (1) satisfies a power law for any $`\alpha `$, as discussed to some extend in the next section.
Moreover, again by taking mass as unity, one may define kinetic energy for prices as
K = $`1/2`$v<sup>2</sup> , (2)
where v is the usual speed i.e., v(t)=d$`\chi `$(t)/dt. The dimensional unit of K may be taken as (local currency unit/time)<sup>2</sup> for shares, e.g., ($`\mathit{}`$/day)<sup>2</sup> or ($/day)<sup>2</sup>in USA. For indices lcu may be kept in the units or it may be substituted by โvalueโ. Potential energy of Eq. (1) will obviously have the same unit as (lcu/time)<sup>2</sup>, and for $`\alpha `$=1 the factor h will have the unit (lcu/time$`{}_{}{}^{2})`$, where lcu stands for local currency unit. For $`\alpha `$=2, h will have the unit of (time$`{}_{}{}^{\text{}2})`$ i.e., frequency squared.
We may assume conservation of the sum of potential and kinetic energies, as long as friction forces, damping etc. are negligible,
U + K = h($`\chi `$$`\chi _{av})^\alpha `$ \+ $`1/2`$v<sup>2</sup> = E = constant . (3)
Differentiation of Eq. (3) with respect to time, for $`\alpha `$=1 yields
h(d$`\chi `$/dt) + v(dv/dt) = 0 , (4)
from which, after substituting v=d$`\chi `$/dt and a=dv/dt, one may obtain the familiar equation of motion for azimuthal rises and falls as in classical mechanics
$`\chi `$(t$`{}_{m}{}^{})=\chi _0`$ \+ v<sub>0</sub> t<sub>m</sub> \+ $`1/2`$ h t$`{}_{}{}^{2}{}_{m}{}^{}`$ , (5 )
where $`\chi _0`$, and v<sub>0</sub> designate initial price and speed, respectively. Time t<sub>m</sub> runs over exchange process days and may be set to zero at the beginning of any epoch. For $`\alpha `$=2, Eq. (1) yields oscillations. In the expansion of $`\chi _{av}`$(t$`{}_{m}{}^{})=\chi _{av}`$(t<sub>m</sub>=0)+v<sub>av</sub>t<sub>m</sub> sign and magnitude of v<sub>av</sub> indicates the up, down or horizontal character of oscillatory trends;
$`\chi `$(t$`{}_{m}{}^{})=\chi _{av}`$(t<sub>m</sub>=0) + v<sub>av</sub>t<sub>m</sub> \+ Asin(wt<sub>m</sub> \+ $`\mathrm{\Phi })`$ , (6)
where A, w, and $`\mathrm{\Phi }`$ is the usual amplitude, angular frequency (here, (2h)$`{}_{}{}^{1/2})`$, and phase, respectively. They are observed in general to have some medium range time periods about some ten days or so, and fade away after a few (two, or three) full periods.
3. Potential energy, power law, and log-periodic equations of motion
The form of potential energy in Eq. (1) satisfies a power law, which is known in physics for a long time as effective subject, utilized especially to express critical phenomena in statistical mechanics. Some special forms of power law appears outside the physical fields as Paretoโs Law and Zipfโs Law.. It is also utilized in seismic predictions for the rupture times.
Power law states that if the argument x of any observable O(x) is scaled by some $`\lambda `$, (i.e., if x$``$/x=$`\lambda )`$ and if O(x$`)`$/O(x)=$`\zeta `$, then O(x)=x<sup>ฮฑ</sup> is a solution with $`\lambda ^\alpha =\zeta `$ and $`\alpha `$=log$`\zeta `$/log$`\lambda `$. Note that, $`\lambda `$ and $`\zeta `$ are independent of x and the relative value of the observable at two different scales depend only on the ratio of the scaling parameters. This is the way scale invariance is associated to self-similarity and criticality. Note also that there is no condition on $`\alpha `$ to be real. Incorporating $`\lambda ^\alpha `$/$`\zeta `$=1=exp(i2$`\pi `$m), where m is any integer, one may generalize the standard scaling O(x)=x<sup>ฮฑ</sup> to a log-periodic one, O(x)=x<sup>ฮฑ</sup>P(logx/log$`\lambda )`$, where P is a function of period 1. Fourier expansion of P can be performed to obtain the most general form of the relevant function. (For detailed and complete treatment, and for various applications see \[8-31\].) For the sake of simplicity, one may take into account only the first Fourier term;
O($`\tau )`$ (1โ$`\tau )^\alpha `${d<sub>0</sub> \+ d<sub>1</sub>cos\[2$`\pi \mathrm{\Omega }`$ln(1โ$`\tau )+\mathrm{\Psi }`$\]} , (7)
where, $`\tau `$ stands for t/T<sub>c</sub> and t is the general independent time variable, and T<sub>c</sub> is the crtitical time; $`\alpha `$=ln$`\zeta `$/ln$`\lambda `$, $`\mathrm{\Omega }`$= 1/ln$`\lambda `$, and $`\mathrm{\Psi }_n`$ is some general phase term. What is in Eq. (7) relevant to time series of shares and indices is that, near the crash (which is considered as a failure time for the log-periodicity) the frequency of the oscillations and the volatility increases, which can be considered as one of the hall-marks of the coming crash.
Letโs approximate the (1โ$`\tau )^\alpha `$ factor and ln(1-$`\tau )`$ by (1โ$`\alpha \tau )`$ and (-$`\tau )`$ for $`\tau `$0, i.e. much before (and by t $``$ โt, much after) the critical time. After some simple mathematical manipulations Eq. (7) can be written as
O($`\tau )`$ D<sub>0</sub> \+ D$`{}_{1}{}^{}\tau `$ \+ D<sub>2</sub>cos(2$`\pi \mathrm{\Omega }\tau +\mathrm{\Psi })`$ , (8)
where, the constants D<sub>0</sub>, D<sub>1</sub>, D<sub>2</sub> can be calculated out of d<sub>0</sub>, d<sub>1,</sub>and the others of Eq. (7). Close similarity between Eqs. (8) and (6) is a consequence of Eq. (1) obeying power law.
4. Applications
The three NYSE indices S&P500, DJIA, and NASDAQ are extensively studied in literature especially within the formalism of power law. For similar log-periodic predictions performed on ISE-100 see . In all of these indices (as in many other world markets) a severe crash dated about the year of 2000 is common. The present state of the same indices will be investigated utilizing the original method of stock mechanics.
The index of S&P500
A crucial feature in S&P500 (Fig. 1) is the almost symmetric behavior of time series about the critical point near 01.Sept.2000 with the close value of 1520.77. Secondly, starting with the beginning of 1995, oscillatory periods decrease as closes climb to the climax. Afterwards closes start to recede and high frequency oscillations turn back into low frequency ones with increasing amplitudes. Thirdly, with the linear price axis, the time series may be fitted as a first order approximation by partial straight lines. Then simple analytical functions may be utilized for each epoch (j=1, 2, 3, 4) as $`\chi _j`$(t$`{}_{m}{}^{})=\chi _{0j}`$ \+ v<sub>j</sub>t<sub>m</sub>, corresponding to a constant potential energy, i.e. $`\alpha `$=0, and h arbitrary in Eq. (1). In this picture the 1987, 1990, 1998 crashes do seem as normal fluctuations, as well as the others after 2000.
For $`\alpha `$=1 in Eq. (1), the second order expression of Eq. (5) delivers very interesting results for the two epochs; one from the beginning of 1997 to the end of 2002, and the second after 2002 till the present time (May.2005), see Fig. 1. By a simple least square fit (lsf) to daily data, the gravity comes about the same for both of the pronounced epochs as h = โ 0.001101 lcu/day<sup>2</sup>. The initial (shooting) speed is found to be 1.73 lcu/day for the first epoch and 1.11 lcu/day for the next one. Imagining the close values as height of a particle shot up in the given gravitational field; the particle first rises till the climax, then falls down and hits the ground at an elevation of 663 lcu. Afterwards it bounces back with a smaller speed and rises till a lesser height of 1219 lcu. Therefore, during the collision it looses its total energy by 59% and the collision is inelastic. The maximum height in any epoch can be calculated utilizing the relation $`\chi `$$`\chi _0`$ = v$`{}_{}{}^{2}{}_{0}{}^{}`$/(2h). It is worth to forecast that March.2005 values are local maximum for S&P500, and a recessional correction may be expected till the level of 800 back, within the coming 500 days. The pronounced parameters are listed in Table 1 for S&P500 as well as for the other indices.
The index of DJIA
DJIA has similar features as pronounced above for S&P500, see Fig. 2. Focusing on quadratic behavior (Eq. 5), gravity again comes out as common for both of the token epochs, before and after the beginning of 2003 (Fig. 2.), where h = โ 0.00606 lcu/day<sup>2</sup>. Hitting speed is 10.17 lcu/day, and bouncing back speed is 8.43 lcu/day, corresponding to 31% loss in total energy. So the March.2005 height is considerably close the historical top of Jan.2000. Again, in about 500 days, DJIA is forecasted to recede back to 8000โs.
The index of NASDAQ
Nasdaq has the most complicated appearance of all the NYSE indices studied here. Yet, it displays very many similarities with S&P500 and DJIA, Fig. 3. Moreover a common gravity comes out for the two epochs following the beginning of 1997 and separated by the beginning of 2003 (Fig. 3.), where h= โ 0.0041 lcu/day<sup>2</sup>. The hitting and bouncing back speeds are 6.08 lcu/day and 3.10 lcu/day, respectively. Then, the loss in total energy is 75%. So, as expected, the March.2005 heights are quite below the historical maximum. Consecutively, NASDAQ is also forecasted to recede back to 1400โs at least, within the next one and a half year.
The index of ISE-100
ISE is a well known world emerging market, and comparing to the NYSE indices, ISE-100 has many more different aspects than similar ones. A log-linear era (lasting about 20 years from the beginning on) has closed by the 2000 crash. Afterwards a recession with 30% loss in a year has taken place. Between Jan.2000 and Jan.2004, recession epoch has been completed and transition to a new up trend has already taken place.
Within the pronounced epoch, ISE-100 displays a dishlike form, and as can be seen in Fig. 4. a. there exist anti-gravity with h= 0.001811 lcu/day<sup>2</sup>. The work done by this constant anti-gravity results in increasing the total energy, day by day. So, one may expect ISE-100 to continue to rise with some possible decorative up and down fluctuations. It is hard to forecast the time of departure from Eq. (5) and solid curve in Fig. 4. a.; yet, it seems that it lies in the far future.
On the other hand, it can be observed that, at the bottom of the Jan.2000 and Jan.2004 epoch, the trend is horizontal. Meanwhile, many oscillations of type Eqs. (6) and (8) may be expected to exist in ISE-100 and in many ISE shares. In Figs. 4. b. and 4. c., two typical oscillatory epochs with different time domains are exemplified, where time axis is weekly and daily, respectively. The corresponding mechanical parameters are listed in Table 2. For better fits one may take into account many coupled smaller spring-masses.
5. Conclusion
The present analytical method can be applied to shares as well. In general there exists a wide diversity of epochs in world markets, in which the present analytical functions can safely be applied. For more elaborate epoch formations, some more complicated functional forms may be tried in Eq. (1). Or, the solutions of the present form for non-integer fractal powers of $`\alpha `$ may be taken into account. Yet, mismatches between the real and calculated values may always exist, due to unpredictability character of short-range fluctuations about longer-range ones. It is obvious that, such analytical approaches may be used together with the traditional approaches for better prediction of the markets.
Acknowledgement
The author is thankful to Dietrich Stauffer for his encouregement to write the present paper, friendly discussions and corrections, and supplying some references.
References
P. Gopikrishnan et al., Phys. Rev. E 62, 4 (2000).
J-P. Bouchaud, R. Cont, Preprint: cond-mat/9801279.
P. Gopikrishnan et al., Physical Review E 60, 5305 (1999).
R. Cont, J-P Bouchaud, Macroeconomic Dynamics 4, 170 (2000).
D. C. Giancoli, โPhysics for Scientists & Engineersโ, 3<sup>rd</sup> ed. Prentice Hall. pp. 375.
V. Pareto โCours dโeconomie politique reprinted as a volume of Oeuvres Complโetes
(Geneva, Droz, 1965).
G. Zipf, โHuman Behavior and the Principle of Last Effortโ (Cambridge, MA: Addison-
Wesley, 1949).
H. Saleur et al., J.Geophys.Res. 101, 17661 (1996).
D.J. Varnes and C.G. Bufe, Geophys. J. Int. 124, 149 (1996).
J. A. Feigenbaum and P. G.O. Freund, arXiv:cond-mat/9509033.
D. Sornette et al., J. Phys. I Fr. 6, 167 (1996). Preprint: arXiv:cond-mat/9510036.
D. Sornette, Phys. Rep. 297, 239 (1998).
J-P. Bouchaud, Quant. Fin. 1, 105 (2001).
X. Gabaix et al., Nature 423, 267 (2003).
X. Gabaix Quarterly Journal of Economics 114 (3), 739 (1999).
Y. Huang et al., Europhysics Letters 41, 43 (1998). Preprint: arXiv:cond-mat/9612065.
W. I. Newman et al., Phys. Rev. E 52, 4827 (1995).
S. Dro d et al., Eur. Phys. J. B 10, 589 (1999).
S. Dro d et al., Physica A 324, 174 (2003).
J-P. Bouchaud, J. Kockelkoren, M. Potters. Preprint http://xxx.lanl.gov/abs/cond-
mat/0406224.
C. Tannous, and A. Fessant, Preprint arXiv:physics/0101042.
D. Sornette, and A. Johansen, Preprint: arXiv:cond-mat/9704127.
A. Johansen, O. Ledoit, D. Sornette, Int. J. of Theoretical and Applied Finance 3, 219
(2000).
W-X. Zhou, and D. Sornette, Physica A 330, 584, Preprint arXiv:physics/0301023.
D. Sornette and W-X. Zhou, Quant. Fin 2, 468 (2002).
A. Johansen, O. Ledoit, D. Sornette, Int. J. of The. and Appl. Finance 3, 219 (2000).
W-X. Zhou, D. Sornette, Physica A 330, 543, Preprint :arXiv:cond-mat/0212010
J. Laherrรจere and D. Sornette, Eur. Phys. J. B 2, 525 (1998).
A. Johansen, and D. Sornette, Eur. Phys. J. B 18, 163 (2000).
For many other articles of D. Sornette see also several issues of the journal Eur. Phys. J.
B. and search Preprint: http://xxx.lanl.gov/abs/cond-mat/
J-P. Bouchaud, Quant. Fin. 1, 105 (2001).
For detailed information about NYSE shares and indices, URL: http://biz.yahoo.com/i/.
For detailed information about ISE.
URL: http.//.www.imkb.gov.tr/sirket/sirketler\_y\_2003.thm.
Table 1 Physical parameters of S&P500, DJIA, NASDAQ, and ISE-100 for $`\alpha `$=1 in Eq. (1). For the time domains, see text. (lcu) in some units stand for local currency unit.
S&P500 DJIA NASDAQ ISE-100
h (day$`{}_{}{}^{2})`$ -0.001101 -0.00606 -0.004119 +0.001811
v<sub>01</sub> (lcu/day) 1.73 10.17 6.08 -7.521
v<sub>02</sub> (lcu/day) 1.11 8.43 3.10 not present
energy loss (%) 59 31 75 not present
Table 2 Physical parameters of ISE-100 for $`\alpha `$=2 in Eq. (1). (lcu) in some units stand for local currency unit.
2001-2003 2004
h (day$`{}_{}{}^{2})`$ 0.051911 0,093884
v<sub>av</sub> (lcu/day) 14 18,23
A (lcu) 2950 880
Figure captions
Fig. 1. The Sept.2000 crash in S&P500 is described as a second order approximation by an azimuthal rise and fall in a gravity h= โ0.001101 (lcu/day$`{}_{}{}^{2})`$, where the initial (shooting) speed is v<sub>01</sub>=1.73 (lcu/day). The price fall down after the maximum height and inelastically bounces back with v<sub>02</sub>= 1.11 (lcu/day) in the same gravity, and rises up to 1200โs in accordance with the expression $`\chi `$$`\chi _0`$ = v$`{}_{}{}^{2}{}_{0}{}^{}`$/(2h). A recession, back to 800โs is predicted within the coming 500 days.
Fig. 2. The excursion of DJIA about the Apr-Sep.2000 climax is described as a second order approximation by an azimuthal rise and fall of the price in a gravity h= โ0.00606 (lcu/day$`{}_{}{}^{2})`$. The initial (shooting) speed at the beginning of 1995 is v<sub>01</sub>= 10.17 (lcu/day). The price fall down after the maximum height and inelasticly bounces back with v<sub>02</sub>= 8.43 (lcu/day) in the same gravity, and rises up to 11000โs in accordance with the expression $`\chi `$$`\chi _0`$ = v$`{}_{}{}^{2}{}_{0}{}^{}`$/(2h). A recession, back to 8000โs and below is predicted within the coming 500 days.
Fig. 3. NASDAQโs azimuthal motion beginning with the year of 1995 is described by a gravity h = โ0.004119 (lcu/day$`{}_{}{}^{2})`$ and initial (shooting) speed of v<sub>01</sub>=6.08 (lcu/day). The inelastic bouncing speed in the same gravity is v<sub>02</sub>=3.10 (lcu/day). A recession, from the present heights back to 1200โs is predicted within the coming one and a half year.
Fig. 4. a. The epoch begun with the beginning of 2000 has an anti-gravity h=0.001811 (lcu/day$`{}_{}{}^{2})`$. Rise is predicted to last till the departure of the price from the the solid curve.
Fig. 4. b. Long term oscillatory motions of ISE-100 within the dishlike epoch, corresponding to $`\alpha `$=2 in Eq. (1). The horizontal axis is weekly in time and h=0.051911 (week$`{}_{}{}^{2})`$. Oscillation fades away after three full periods (here, about two years and a half), as usual.
Fig. 4. d. Short term oscillatory motions of ISE-100 within the year of 2004 with h= 0,093884 (day$`{}_{}{}^{2})`$. Oscillation fades away after three full periods (here, about two monts or so), as usual. (Notice the relative increase in volume at dips of oscillations.)
Figures
Fig. 1.
Fig. 2.
Fig. 3.
Fig. 4. a.
Fig. 4. b.
Fig. 4. c.
|
warning/0506/astro-ph0506398.html
|
ar5iv
|
text
|
# A Unified, Merger-Driven Model for the Origin of Starbursts, Quasars, the Cosmic X-Ray Background, Supermassive Black Holes and Galaxy Spheroids
## 1. Introduction
The measurement of anisotropies in the cosmic microwave background (e.g. Spergel et al. 2003) combined with observations of high redshift supernovae (e.g. Riess et al. 1998, 2000; Perlmutter et al. 1999) have established a โstandard modelโ for the Universe, in which the energy density is dominated by an unknown form driving accelerated cosmic expansion, and most of the mass is non-baryonic, in a ratio of roughly 5:1 to ordinary matter. On small scales, it is believed that structure formed through gravitational instability. In the currently favored cold dark matter (CDM) paradigm, objects grow hierarchically, with smaller ones forming first and then merging into successively larger bodies. As baryons fall into dark matter potential wells, the gas is shocked and then cools radiatively to form stars and galaxies, in a โbottom-upโ progression (White & Rees 1978).
Even with the many successes of this picture, the processes underlying galaxy formation and evolution are poorly understood. For example, there has yet to be an ab initio calculation, starting from an initial state prescribed by the standard model, resulting in a population of objects that reproduces observed galaxies. However, from the same initial conditions, computer simulations have yielded a new, successful interpretation of the Lyman-alpha forest in which absorption in caused by density fluctuations in the intergalactic medium (e.g. Cen et al. 1994; Zhang et al. 1995; Hernquist et al. 1996), over many orders of magnitude in column density (e.g. Katz et al. 1996a), explicitly related to growth of structure in a CDM universe (e.g. Croft et al. 1998, 1999, 2002; McDonald et al. 2000, 2004; Hui et al. 2001; Viel et al. 2003, 2004). This suggests that the difficulties with understanding galaxy formation and evolution lie not in the initial conditions or with the description of dark matter, but rather with the physics that has been used to model the baryons.
Observations have revealed regularities in the structure of galaxies that point to some of this โmissingโ physics. Supermassive black holes appear to reside at the centers of most galaxies (e.g. Kormendy & Richstone 1995; Richstone et al. 1998; Kormendy & Gebhardt 2001) and the masses of these black holes are correlated with either the mass (Magorrian et al. 1998; McLure & Dunlop 2002; Marconi & Hunt 2003) or the velocity dispersion (i.e. the $`M_{\mathrm{BH}}`$-$`\sigma `$ relation: Ferrarese & Merritt 2000; Gebhardt et al. 2000; Tremaine et al. 2002) of spheroids, demonstrating a direct link between the origin of galaxies and supermassive black holes. Simulations which follow the self-regulated growth of black holes in galaxy mergers (Di Matteo et al. 2005; Springel et al. 2005a) have shown that the energy released through this process has a global impact on the structure of the merger remnant. If this conclusion applies to spheroid formation in general, the simulations demonstrate that models for the origin and evolution of galaxies must account for black hole growth and feedback in a fully self-consistent manner.
Analytical and semi-analytical modeling (Silk & Rees, 1998; Fabian, 1999; Wyithe & Loeb, 2002, 2003; Begelman & Nath, 2005) suggests that, beyond a certain threshold, feedback energy from black holes can expel gas from the centers of galaxies, shutting down accretion onto them and limiting their masses. However, these calculations usually ignore the impact of this process on star formation and therefore do not explain the link between black hole growth and spheroid formation, and furthermore make simplifying assumptions about the dynamics of such accretion. For example, the duration of black hole growth is a free parameter, which is fixed either using observational estimates or assumed to be similar to e.g. the dynamical time of the host galaxy or the $`e`$-folding time for Eddington-limited black hole growth $`t_S=M_{\mathrm{BH}}/\dot{M}=4.5\times 10^7l^1(ฯต_r/0.1)`$yr for accretion with radiative efficiency $`ฯต_r=L/\dot{M}c^20.1`$ and $`l=L/L_{\mathrm{Edd}}1`$ (Salpeter, 1964). Moreover, these studies have adopted idealized models for quasar light curves, usually corresponding to growth at a constant Eddington ratio or on-off, โlight bulb,โ scenarios. As we discuss below, less restrictive modeling suggests that this phase is actually more complex.
Efforts to model quasar accretion and feedback more self-consistently (e.g., Ciotti & Ostriker, 1997, 2001; Granato et al., 2004) by treating the hydrodynamical response of gas to black hole growth have generally been restricted to idealized geometries, such as spherical symmetry, employing simple models for star formation and galaxy-scale quasar fueling. However, these works have made it possible to estimate duty cycles of quasars and shown that the objects left behind have characteristics similar to those observed, with quasar feedback being a critical element in reproducing these features (e.g. Sazonov et al. 2005; Kawata & Gibson 2005; Cirasuolo et al. 2005; for a review, see Ostriker & Ciotti 2005).
Springel et al. (2005b) have incorporated black hole growth and feedback into simulations of galaxy mergers and included a multiphase model for star formation and pressurization of the interstellar gas by supernovae (Springel & Hernquist, 2003) to examine implications of these processes for galaxy formation and evolution. Di Matteo et al. (2005) and Springel et al. (2005a,b) have shown that gas inflows excited by gravitational torques during a merger both trigger starbursts and fuel rapid black hole growth. The growth of the black hole is determined by the gas supply and terminates as gas is expelled by feedback, halting accretion, leaving a dead quasar in an ordinary galaxy. The self-regulated nature of black hole growth in mergers explains observed correlations between black hole mass and properties of normal galaxies (Di Matteo et al., 2005), as well as the color distribution of ellipticals (Springel et al., 2005a). These results lend support to the view that mergers have played an important role in structuring galaxies, as advocated especially by Toomre & Toomre (1972) and Toomre (1977). (For reviews, see, e.g., Barnes & Hernquist 1992; Barnes 1998; Schweizer 1998.)
Subsequent analysis by Hopkins et al. (2005a,b,c,d) has shown that the merger simulations can account for quasar phenomena as a phase of black hole growth. Unlike what has been assumed in e.g. semi-analytical studies of quasars, the simulations predict complicated evolution for quasar lifetimes, fueling rates for black hole accretion, obscuration, and quasar light curves. The light curves were studied by Hopkins et al. (2005a, b), who showed that the self-termination process gives observable lifetimes $`10^7`$yr for bright optical quasars and predicts a large population of obscured sources as a natural stage of quasar evolution, as implied by observations (for a review, see Brandt & Hasinger 2005). Hopkins et al. (2005b) analyzed simulations over a range of galaxy masses and found that the quasar light curves and lifetimes are always qualitatively similar, with both the intrinsic and observed quasar lifetimes being decreasing functions of luminosity, with longer lifetimes at all luminosities for higher-mass (higher peak luminosity) systems. The dependence of the lifetime on luminosity led Hopkins et al. (2005c) to suggest a new interpretation of the quasar luminosity function, in which the steep bright-end consists of quasars radiating near the Eddington limit and is directly related to the distribution of intrinsic peak luminosities (or final black hole masses) as has been assumed previously (e.g., Small & Blandford, 1992; Haiman & Loeb, 1998; Haiman & Menou, 2000; Kauffmann & Haehnelt, 2000; Somerville et al., 2001; Tully et al., 2002; Wyithe & Loeb, 2003; Volonteri et al., 2003; Haiman, Quataert, & Bower, 2004; Croton et al., 2005), but where the shallow, faint-end of the luminosity function describes black holes growing towards or declining from peak phases of quasar activity, with Eddington ratios generally between $`l0.01`$ and 1. The โbreakโ in the luminosity function corresponds directly to the peak in the distribution of intrinsic quasar properties. As argued by Hopkins et al. (2005c, d) this new interpretation of the luminosity function can self-consistently explain various properties of both the quasar and galaxy populations, connecting the origin of galaxy spheroids, supermassive black holes, and quasars.
Motivated by these results, and earlier work by many others which we summarize below, in this paper we consider a picture for galaxy formation and evolution, illustrated schematically as a โcosmic cycleโ in Figure 1, in which starbursts, quasars, and the simultaneous formation of spheroids and supermassive black holes represent connected phases in the lives of galaxies. Mergers are expected to occur regularly in a hierarchical universe, particularly at high redshifts. Those between gas-rich galaxies drive nuclear inflows of gas, triggering starbursts and fueling the growth of supermassive black holes. During most of this phase, quasar activity is obscured, but once a black hole dominates the energetics of the central region, feedback expels gas and dust, making the black hole visible briefly as a bright quasar. Eventually, as the gas is further heated and expelled, quasar activity can no longer be maintained and the merger remnant relaxes to a normal galaxy with a spheroid and a supermassive black hole. In some cases, depending on the gas content of the progenitors, the remnant may also have a disk (Springel & Hernquist 2005; Robertson et al. 2005a). The remnant will then evolve passively and would be available as a seed to repeat the above cycle. As the Universe evolves and more gas is consumed, the mergers involving gas-rich galaxies will shift towards lower masses, explaining the decline in the population of the brightest quasars from $`z2`$ to the present, and the remnants that are gas-poor will redden quickly owing to the termination of star formation by black hole feedback (Springel et al. 2005a), so that they resemble elliptical galaxies, surrounded by hot X-ray emitting halos (e.g. Cox et al. 2005).
There is considerable observational support for this scenario, which has led the development of this picture for the co-evolution of galaxies and quasars over recent decades. Infrared (IR) luminous galaxies are thought to be powered in part by starbursts (e.g. Soifer et al. 1984a,b; Sanders et al. 1986, 1988a,b; for a review, see e.g. Soifer et al. 1987), and the most intense examples locally, ultraluminous infrared galaxies (ULIRGs), are invariably associated with mergers (e.g. Allen et al. 1985; Joseph & Wright 1985; Armus et al. 1987; Kleinmann et al. 1988; Melnick & Mirabel 1990; for reviews, see Sanders & Mirabel 1996 and Jogee 2004). Radio observations show that ULIRGs have large, central concentrations of dense gas (e.g. Scoville et al. 1986; Sargent et al. 1987, 1989), providing a fuel supply to feed black hole growth. Indeed, some ULIRGs have โwarmโ IR spectral energy distributions (SEDs), suggesting that they harbor buried quasars (e.g. Sanders et al. 1988c), an interpretation strengthened by X-ray observations demonstrating the presence of two non-thermal point sources in NGC6240 (Komossa et al., 2003), which are thought to be supermassive black holes that are heavily obscured at visual wavelengths (e.g. Gerssen et al. 2004; Max et al. 2005, Alexander et al. 2005a,b). These lines of evidence, together with the overlap between bolometric luminosities of ULIRGs and quasars, indicate that quasars are the descendents of an infrared luminous phase of galaxy evolution caused by mergers (Sanders et al. 1988a), an interpretation supported by observations of quasar hosts (e.g. Stockton 1978; Heckman et al. 1984; Stockton & MacKenty 1987; Stockton & Ridgway 1991; Hutchings & Neff 1992; Bahcall et al. 1994, 1995, 1997; Canalizo & Stockton 2001).
However, many of the physical processes that connect the phases of evolution in Figure 1 are not well understood. Early simulations showed that mergers produce objects resembling galaxy spheroids (e.g. Barnes 1988, 1992; Hernquist 1992, 1993a) and that if the progenitors are gas-rich, gravitational torques funnel gas to the center of the remnant (e.g. Barnes & Hernquist 1991, 1996), producing a starburst (e.g. Mihos & Hernquist 1996), but these works did not explore the relationship of these events to black hole growth and quasar activity. While a combination of arguments based on time variability and energetics suggests that quasars are produced by the accretion of gas onto supermassive black holes in the centers of galaxies (e.g. Salpeter 1964; Zelโdovich & Novikov 1964; Lynden-Bell 1969), the mechanism that provides the trigger to fuel quasars therefore remains uncertain. Furthermore, there have been no comprehensive models that describe the transition between ULIRGs and quasars that can simultaneously account for observed correlations like the $`M_{\mathrm{BH}}`$-$`\sigma `$ relation.
Here, we study these relationships using numerical simulations of galaxy mergers that account for the consequences of black hole growth. In our simulations, black holes accrete and grow throughout a merger event, producing complex, time-varying quasar activity. Quasars reach a peak luminosity, $`L_{\mathrm{peak}}`$, during the โblowoutโ phase of evolution where feedback energy from black hole growth begins to drive away the gas, eventually slowing accretion. Prior to and following this brief period of peak activity, quasars radiate at instantaneous luminosities, $`L`$, with $`L<L_{\mathrm{peak}}`$. However, we show that even with this complex behavior, the global characteristics that determine the observed properties of quasars, i.e. lifetimes, light curves, and obscuration, can be expressed as functions of $`L`$ and $`L_{\mathrm{peak}}`$, allowing us to make predictions for quasar populations that agree well with observations, supporting the scenario sketched in Figure 1.
In ยง 2, we discuss our methodology and show how the quasar lifetimes and obscuration from our simulations can be expressed as functions of the instantaneous and peak luminosities of quasars. We also define a set of commonly adopted models for the quasar lifetime and obscuration against which we compare our predictions throughout. In ยง 3, we apply our models to the quasar luminosity function, using the observed luminosity function to determine the distribution of quasar peak luminosities, and show that this allows us to simultaneously reproduce the hard X-ray, soft X-ray, and optical quasar luminosity functions at all redshifts $`z3`$, and the distribution of column densities in both optical and X-ray samples. In ยง 4, we determine the time in our simulations when quasars will be observable as broad-line objects, and use this to predict the broad-line luminosity function and fraction of broad-line objects in quasar samples, as a function of luminosity, as well as the mass function of low-redshift, active broad-line quasars. In ยง 5, we estimate the distribution of Eddington ratios in our simulations as a function of luminosity, and infer Eddington ratios in observed samples at different redshifts. In ยง 6, we use our modeling to predict both the mass distribution and total density of present-day relic supermassive black holes, and describe their evolution with redshift. In ยง 7, we similarly apply this model to predict the integrated cosmic X-ray background spectrum, accounting for the observed spectrum from $`1100`$ keV. In ยง 8, we discuss the primary qualitative implications of our results and propose falsifiable tests of our picture. Finally, in ยง 9, we conclude and suggest directions for future work.
Throughout, we adopt a $`\mathrm{\Omega }_\mathrm{M}=0.3`$, $`\mathrm{\Omega }_\mathrm{\Lambda }=0.7`$, $`H_0=70\mathrm{km}\mathrm{s}^1\mathrm{Mpc}^1`$ ($`h=0.7`$) cosmology.
## 2. The Model: Methodology
Our model of quasar evolution has several elements, which we summarize here and describe in greater detail below.
* In what follows, a โquasarโ is taken to mean the course of black hole activity in a single merger event. We use the term โquasar lifetimeโ to refer to the time spent by such a quasar at a given luminosity or fraction of the quasar peak luminosity, integrated over all black hole activity in a single merger event. This is not meant to suggest that this would constitute the entire accretion history of a black hole โ a given black hole may have multiple โlifetimesโ triggered by different mergers, with each merger in principle fueling a distinct โquasarโ with its own lifetime. There is no a priori luminosity threshold for quasar activity โ the time history can include various epochs at low luminosities and accretion rates.
* We model the galaxy mergers using hydrodynamical simulations, varying the orbital parameters of the encounter, the internal properties of the merging galaxies, prescriptions for the gas physics, initial โseedโ black hole masses of the merging systems, and numerical resolution of the simulations. The black hole accretion rate is determined from the surrounding gas (smoothed over the scale of our spatial resolution, reaching $`20`$pc in the best cases), i.e. the density and sound speed of the gas, and its motion relative to the black hole, using Eddington-limited, Bondi-Hoyle-Lyttleton accretion theory. The black hole radiates with a canonical efficiency $`ฯต_r=0.1`$ corresponding to a standard Shakura & Sunyaev (1973) thin disk, and we assume that $`5\%`$ of this radiated luminosity is deposited as thermal energy into the surrounding gas, weighted by the SPH smoothing kernel (which has a $`r^2`$ profile) over the scale of the spatial resolution. This scale is such that we cannot resolve the complex accretion flow immediately around the black hole, but we adopt this prescription because: (1) it reproduces the observed slope and normalization in the $`M_{\mathrm{BH}}\sigma `$ relation (Di Matteo et al. 2005), (2) it follows from observations, based on estimates of the energy contained in highly-absorbed UV portion of the quasar SED (e.g., Elvis et al., 1994; Telfer et al., 2002), (3) it follows from theoretical considerations of momentum coupling to dust grains in the dense gas very near the quasar (Murray et al., 2005) and hydrodynamical simulations of small-scale radiative heating from quasar accretion (Ciotti & Ostriker, 2001), and (4) even if the feedback is initially highly collimated, a driven wind or shock in a dense region such as the center of the merging galaxies will rapidly isotropize, so long as it is decelerated by gravity and the surrounding medium, allowing the high sound speed within the shock to equalize angle-dependent pressure variations (e.g., Koo & McKee, 1990), and furthermore initial local distortions will be washed away in favor of triaxial structure determined by the large-scale density gradients (Bisnovatyi-Kogan & Silich, 1991), as occurs in our simulations.
* For each of our merger simulations, we compute the bolometric black hole luminosity and column density along $`1000`$ lines of sight to the black hole(s) (evenly spaced in solid angle), as a function of time from the beginning of the simulation until the system has relaxed for $`1`$Gyr after the merger.
* We bin different merger simulations by $`L_{\mathrm{peak}}`$, the peak bolometric luminosity of the black hole in the simulation, and the conditional distributions of luminosity, $`P(L|L_{\mathrm{peak}})`$, and column density, $`P(N_\mathrm{H}|L,L_{\mathrm{peak}})`$, are computed using all simulations that fall into a given bin in $`L_{\mathrm{peak}}`$. The final black hole mass (black hole mass at the end of the individual merger โ subsequent mergers and quasar episodes could further increase the black hole mass) is approximately $`M_{\mathrm{BH}}^fM_{\mathrm{Edd}}(L_{\mathrm{peak}})`$ (but not exactly, see ยง 2.4), so we obtain similar results if we bin instead by $`M_{BH}^f`$. Our calculation of $`M_{BH}^f(L_{\mathrm{peak}})`$ allows us to express our conditional distributions of luminosity and column density in terms of either peak luminosity or final black hole mass. Critically, we find that expressed in terms of $`L_{\mathrm{peak}}`$ or $`M_{\mathrm{BH}}^f`$, there is no systematic dependence in the quasar evolution on the varied merger simulation properties โ this allows us to calculate a large number of observables in terms of $`L_{\mathrm{peak}}`$ or $`M_{\mathrm{BH}}^f`$ without the large systematic uncertainties inherent in attempting to directly estimate e.g. quasar light curves in terms of host galaxy mass, gas fraction, multiphase pressurization of the interstellar medium, orbital parameters and merger stage, and other variables.
* The observed quasar luminosity function is the convolution of the time a given quasar spends at some observed luminosity with the rate at which such quasars are created. Knowing the distributions $`P(L|L_{\mathrm{peak}})`$ and $`P(N_\mathrm{H}|L,L_{\mathrm{peak}})`$, we can calculate the time spent by a quasar with some $`L_{\mathrm{peak}}`$ at an observed luminosity in a given waveband. We use this to fit to observational estimates of the bolometric quasar luminosity function $`\varphi (L)`$, de-convolving these quantities to determine the function $`\dot{n}(L_{\mathrm{peak}})`$; i.e. the rate at which quasars of a given peak luminosity must be created or activated (triggered in mergers) in order to reproduce the observed bolometric luminosity function.
* Given these inputs, we determine the joint distribution in instantaneous luminosity and black hole mass, column density distribution, peak luminosity and final black hole mass, as a function of redshift, i.e. $`n(L,L_\nu ,M_{\mathrm{BH}},N_\mathrm{H},L_{\mathrm{peak}},M_{\mathrm{BH}}^f|z)`$, at all redshifts where the observed quasar luminosity function can provide the necessary constraint. From this joint distribution, we can compute, for example, luminosity functions in other wavebands, conditional column density distributions, active black hole mass functions and Eddington ratio distributions, and relic black hole mass functions and cosmic backgrounds. We can compare each of these results to those determined using simpler models for either the quasar lifetime or column density distributions; in ยง 2.5 we describe a canonical set of such models, to which we compare throughout this paper.
### 2.1. The Simulations
The simulations were performed with GADGET-2 (Springel, 2005), a new version of the parallel TreeSPH code GADGET (Springel, Yoshida, & White, 2001). GADGET-2 is based on a fully conservative formulation (Springel & Hernquist, 2002) of smoothed particle hydrodynamics (SPH), which maintains simultaneous energy and entropy conservation when smoothing lengths evolve adaptively (for a discussion, see e.g., Hernquist 1993b, OโShea et al. 2005). Our simulations account for radiative cooling, heating by a UV background (as in Katz et al. 1996b, Davรฉ et al. 1999), and incorporate a sub-resolution model of a multiphase interstellar medium (ISM) to describe star formation and supernova feedback (Springel & Hernquist, 2003). Feedback from supernovae is captured in this sub-resolution model through an effective equation of state for star-forming gas, enabling us to stably evolve disks with arbitrary gas fractions (see, e.g. Springel et al. 2005b; Robertson et al. 2004). In order to investigate the consequences of supernova feedback over a range of conditions, we employ the scheme of Springel et al. (2005b), introducing a parameter $`q_{\mathrm{EOS}}`$ to interpolate between an isothermal equation of state ($`q_{\mathrm{EOS}}=0`$) and the full multiphase equation of state ($`q_{\mathrm{EOS}}=1`$) described above.
Supermassive black holes (BHs) are represented by โsinkโ particles that accrete gas at a rate $`\dot{M}`$ estimated using an Eddington-limited version of Bondi-Hoyle-Lyttleton accretion theory (Bondi 1952; Bondi & Hoyle 1944; Hoyle & Lyttleton 1939). The bolometric luminosity of the black hole is $`L_{\mathrm{bol}}=ฯต_r\dot{M}c^2`$, where $`ฯต_r=0.1`$ is the radiative efficiency. We assume that a small fraction (typically $`5\%`$) of $`L_{\mathrm{bol}}`$ couples dynamically to the surrounding gas, and that this feedback is injected into the gas as thermal energy, as described above.
We have performed several hundred simulations of colliding galaxies, varying the numerical resolution, the orbit of the encounter, the masses and structural properties of the merging galaxies, initial gas fractions, halo concentrations, and the parameters describing star formation and feedback from supernovae and black hole growth. This large set of simulations allows us to investigate merger evolution for a wide range of galaxy properties and to identify any systematic dependence of our modeling. The galaxy models are described in Springel et al. (2005b), and we briefly review their properties here.
The progenitor galaxies in our simulations have virial velocities $`V_{\mathrm{vir}}=80,`$ $`113,`$ $`160,`$ $`226,`$ $`320,`$ $`\mathrm{and}500\mathrm{km}\mathrm{s}^1`$. We consider cases with gas equation of state parameters $`q_{\mathrm{EOS}}=0.25`$ (moderately pressurized, with a mass-weighted temperature of star-forming gas $`10^{4.5}\mathrm{K}`$) and $`q_{\mathrm{EOS}}=1.0`$ (the full, โstiffโ Springel-Hernquist equation of state, with a mass-weighted temperature of star-forming gas $`10^5\mathrm{K}`$), and initial disk gas fractions (by mass) of $`f_{\mathrm{gas}}=0.2,`$ $`0.4,`$ $`0.8,`$ $`\mathrm{and}1.0`$. Finally, we scale these models with redshift, altering the physical sizes of the galaxy components and the dark matter halo concentration in accord with cosmological evolution (Mo, Mao & White, 1998). Details are provided in Robertson et al. (2005b), and here we consider galaxy models scaled appropriately to resemble galaxies of the same $`V_{\mathrm{vir}},f_{\mathrm{gas}},\mathrm{and}q_{\mathrm{EOS}}`$ at redshifts $`z_{\mathrm{gal}}=0,`$ $`2,`$ $`3,`$ $`\mathrm{and}6`$.
For each simulation, we generate two stable, isolated disk galaxies, each with an extended dark matter halo with a Hernquist (1990) profile, motivated by cosmological simulations (e.g. Navarro et al. 1996; Busha et al. 2004) and observations of halo properties (e.g. Rines et al. 2002, 2002, 2003, 2004), an exponential disk of gas and stars, and (optionally) a bulge. The galaxies have masses $`M_{\mathrm{vir}}=V_{\mathrm{vir}}^3/(10GH_0)`$ for $`z_{\mathrm{gal}}=0`$, with the baryonic disk having a mass fraction $`m_\mathrm{d}=0.041`$, the bulge (when present) has a mass fraction $`m_\mathrm{b}=0.0136`$, and the rest of the mass is in dark matter typically with a concentration parameter $`c=9.0`$. The disk scale-length is computed based on an assumed spin parameter $`\lambda =0.033`$, chosen to be near the mode in the observed $`\lambda `$ distribution (Vitvitska et al., 2002), and the scale-length of the bulge is set to $`0.2`$ times the resulting value. In Hopkins et al. (2005a), we describe our analysis of simulation A3, one of our set with $`V_{\mathrm{vir}}=160\mathrm{km}\mathrm{s}^1,f_{\mathrm{gas}}=1.0,q_{\mathrm{EOS}}=1.0,\mathrm{and}z_{\mathrm{gal}}=0`$, a fiducial choice with a rotation curve and mass similar to the Milky Way, and Hopkins et al. (2005b, c, d) used a set of simulations with the same parameters but varying $`V_{\mathrm{vir}}=80,`$ $`113,`$ $`160,`$ $`226,`$ $`\mathrm{and}320\mathrm{km}\mathrm{s}^1`$, which we refer to below as runs A1, A2, A3, A4, and A5, respectively.
Typically, each galaxy is initially composed of 168000 dark matter halo particles, 8000 bulge particles (when present), 24000 gas and 24000 stellar disk particles, and one BH particle. We vary the numerical resolution, with many of our simulations using instead twice as many particles in each galaxy, and a subset of simulations with up to 128 times as many particles. We vary the initial seed mass of the black hole to identify any systematic dependence of our results on this choice. In most cases, we choose the seed mass either in accord with the observed $`M_{\mathrm{BH}}`$-$`\sigma `$ relation or to be sufficiently small that its presence will not have an immediate dynamical effect. Given the particle numbers employed, the dark matter, gas, and star particles are all of roughly equal mass, and central cusps in the dark matter and bulge profiles are reasonably well resolved (see Fig 2. in Springel et al. 2005b). The galaxies are then set to collide from a zero energy orbit, and we vary the inclinations of the disks and the pericenter separation.
A representative example of the behavior of the simulations is provided in Figure 2, which shows the time sequence of a merger involving two bulge-less progenitor galaxies with virial velocities of $`160\mathrm{km}\mathrm{s}^1`$ and initial gas fractions of 20%. During the merger, gas is driven to the galaxy centers by gravitational tides, fueling nuclear starbursts and black hole growth. The quasar activity is short-lived and peaks twice in this merger, both during the first encounter and the final coalescence of the galaxies. To illustrate the bright, optically observable phase(s) of quasar activity which we identify below, we have added nuclear point sources in the center at the position(s) of the black hole(s) at times $`T=1.03`$, $`1.39`$ and $`1.48\mathrm{Gyr}`$, generating a surface density in correspondence to the relative luminosities of stars and quasar at these times. At other times, the accretion activity is either obscured or the black hole accretion rate is negligible. To make the appearance of the quasar visually more apparent, we have put a small part of its luminosity in โraysโ around the quasar. These rays are artificial and are only a visual guide.
### 2.2. Column Densities & Quasar Attenuation
From the simulation outputs, we determine the obscuration of the black hole as a function of time during a merger by calculating the column density to a distant observer along many lines of sight. Typically, we generate $`1000`$ radial lines-of-sight (rays), each with its origin at the black hole location and with directions uniformly spaced in solid angle $`\mathrm{d}\mathrm{cos}\theta \mathrm{d}\varphi `$. For each ray, we begin at the origin and calculate and record the local gas properties using the SPH formalism and move a distance along the ray $`\mathrm{\Delta }r=\eta h_{\mathrm{sml}}`$, where $`\eta 1`$ and $`h_{\mathrm{sml}}`$ is the local SPH smoothing length. The process is repeated until a ray is sufficiently far from the origin ($`100`$ kpc) that the column has converged. We then integrate the gas properties along a particular ray to give the line-of-sight column density and mean metallicity. We have varied $`\eta `$ and find empirically that gas properties along a ray converge rapidly and change smoothly for $`\eta =0.5`$ and smaller. We similarly vary the number of rays and find that the distribution of line-of-sight properties converges for $`100`$ rays.
From the local gas properties, we use the multiphase model of the ISM described in Springel & Hernquist (2003) to determine the mass fraction in โhotโ (diffuse) and โcoldโ (molecular and HI cloud core) phases of dense gas and, assuming pressure equilibrium, we obtain the local density of the hot and cold phases and their corresponding volume filling factors. The resulting values are in rough agreement with those of McKee & Ostriker (1977). Given a temperature for the warm, partially ionized component of the hot-phase $`8000\mathrm{K}`$, determined by pressure equilibrium, we further calculate the neutral fraction of this gas, typically $`0.30.5`$. We denote the neutral and total column densities as $`N_{\mathrm{H}\mathrm{I}}`$ and $`N_\mathrm{H}`$, respectively. Using only the hot-phase density allows us to place an effective lower limit on the column density along a particular line of sight, as it assumes a given ray passes only through the diffuse ISM, with $`90\%`$ of the mass of the dense ISM concentrated in cold-phase โclumps.โ Given the small volume filling factor ($`<0.01`$) and cross section of cold clouds, we expect that the majority of sightlines will pass only through the โhot-phaseโ component.
Using $`L_{\mathrm{bol}}=ฯต_r\dot{M}c^2`$, we model the intrinsic quasar continuum SED following Marconi et al. (2004), based on optical through hard X-ray observations (e.g., Elvis et al., 1994; George et al., 1998; Vanden Berk et al., 2001; Perola et al., 2002; Telfer et al., 2002; Ueda et al., 2003; Vignali et al., 2003), with a reflection component generated by the PEXRAV model (Magdziarz & Zdziarski, 1995). This yields, for example, a B-band luminosity $`\mathrm{log}(L_B/L_{\mathrm{}})=0.800.067+0.017^20.0023^3`$, where $`=\mathrm{log}(L_{\mathrm{bol}}/L_{\mathrm{}})12`$, and we take $`\lambda _B=4400`$ร
, but as we model the entire intrinsic SED we can determine the bolometric correction in any frequency interval.
We then use a gas-to-dust ratio to determine the extinction along a given line of sight at optical frequencies. Observations suggest that the majority of reddened quasars have reddening curves similar to that of the Small Magellanic Cloud (SMC; Hopkins et al. 2004, Ellison et al. 2005), which has a gas-to-dust ratio lower than the Milky Way by approximately the same factor as its metallicity (Bouchet et al., 1985). Hence, we consider both a gas-to-dust ratio equal to that of the Milky Way, $`(A_B/N_{\mathrm{H}\mathrm{I}})_{\mathrm{MW}}=8.47\times 10^{22}\mathrm{cm}^2`$, and a gas-to-dust ratio scaled by metallicity, $`A_B/N_{\mathrm{H}\mathrm{I}}=(Z/0.02)(A_B/N_{\mathrm{H}\mathrm{I}})_{\mathrm{MW}}`$. In both cases we use the SMC-like reddening curve of Pei (1992). The form of the correction for hard X-ray (2-10 keV) and soft X-ray (0.5-2 keV) luminosities is similar to that of the B-band luminosity. We calculate extinction at X-ray frequencies (0.03-10 keV) using the photoelectric absorption cross sections of Morrison & McCammon (1983) and non-relativistic Compton scattering cross sections, similarly scaled by metallicity. In determining the column density for photoelectric X-ray absorption, we ignore the inferred ionized fraction of the gas, as it is expected that the inner-shell electrons which dominate the photoelectric absorption edges will be unaffected in the temperature ranges of interest. We do not perform a full radiative transfer calculation, and therefore do not model scattering or re-processing of radiation by dust in the infrared.
For a full comparison of quasar lifetimes and column densities obtained varying our calculation of $`N_\mathrm{H}`$, we refer to Hopkins et al. (2005b) (see their Figures 1, 5, & 6), and note their conclusion that, after accounting for clumping of most mass in the dense ISM in cold-phase structures, the column density does not depend sensitively on our assumptions for the small-scale physics of the ISM and obscuration โ typically, the uncertainties in the resulting quasar lifetime as a function of luminosity are a factor $`2`$ at low luminosities in the B-band, and smaller in e.g. the hard X-ray. Because our determination of the quasar luminosity functions is similar using the hard X-ray data alone or the hard X-ray, soft X-ray, and optical data simultaneously, the added uncertainties in our calculation of $`\dot{n}(L_{\mathrm{peak}})`$ in ยง 3.2 below owing to the uncertainty in our $`N_\mathrm{H}`$ calculation are small compared to the uncertainties owing to degeneracies in the fitting procedure and uncertain bolometric corrections.
### 2.3. The $`N_\mathrm{H}`$ Distribution as a Function of Luminosity
Next, we consider the distribution of column densities as a function of both the instantaneous and peak quasar luminosities. For each simulation, we consider $`N_\mathrm{H}`$ values at all times with a given bolometric luminosity $`L`$ (in some logarithmic interval in $`L`$), and determine the distribution of column densities at that $`L`$ weighted by the total time along all sightlines with a given $`N_\mathrm{H}`$. At each $`L`$, we approximate the simulated distribution and fit it to a lognormal form,
$$P(N_H)=\frac{1}{\sigma _{N_H}\sqrt{2\pi }}\mathrm{exp}\left[\frac{\mathrm{log}^2(N_H/\overline{N}_H)}{2\sigma _{N_H}^2}\right].$$
(1)
This provides a good fit for all but the brightest luminosities, where quasar feedback becomes important driving the โblowoutโ phase, and the quasar sweeps away surrounding gas and dust to become optically observable.
We show the resulting median column density $`\overline{N}_H`$ at each luminosity $`L`$ in Figure 3. In the upper left panel, simulations with $`z_{\mathrm{gal}}=0`$ are shown in black, those with $`z_{\mathrm{gal}}=2`$ in blue, and those with $`z_{\mathrm{gal}}=3`$ in yellow. In the upper right, simulations with $`f_{\mathrm{gas}}=0.4`$ are shown in black, those with $`f_{\mathrm{gas}}=0.8`$ in red. In the lower left, simulations with $`q_{\mathrm{EOS}}=0.25`$ are shown in black, those with $`q_{\mathrm{EOS}}=1.0`$ in green. And in the lower right, simulations with $`V_{\mathrm{vir}}=80,113,160,226,320,\mathrm{and}500\mathrm{km}\mathrm{s}^1`$ are shown as black asterisks, purple dots, red diamonds, green triangles, yellow squares, and red crosses, respectively. Simulations with other values for these parameters (not shown for clarity, but see e.g. Hopkins et al. \[2005d\]) show similar trends.
While the increase in typical $`N_\mathrm{H}`$ values with luminosity appears to contradict observations suggesting that the obscured fraction decreases with luminosity, this is because the relationship shown above is dominated by quasars in growing, heavily obscured phases. In these stages, the relationship between column density and luminosity is a natural consequence of the fact that both are fueled by strong gas flows into the central regions of the galaxy โ more gas inflow means higher luminosities, but also higher column densities. During these phases, the lognormal fits to column density as a function of instantaneous and peak luminosity presented in this section are reasonable approximations, but they break down in the brightest, short-lived stages of merger activity when the quasar rapidly heats the surrounding gas and drives a powerful wind, lowering the column density, resulting in a bright, optically observable quasar. Including in greater detail the effects of quasar blowout during the final stages of its growth in ยง 4, we find that this modeling actually predicts the observed decrease in obscured fraction with luminosity.
The relationship between $`N_\mathrm{H}`$ and $`L`$ shows no strong systematic dependence on any of the simulation parameters considered. At most, there is weak sensitivity to $`q_{\mathrm{EOS}}`$, in the sense that the simulations with $`q_{\mathrm{EOS}}=1.0`$ have slightly larger column densities at a given luminosity than those with $`q_{\mathrm{EOS}}=0.25`$. We derive an analytical model relating both the observed column density and quasar luminosity to the inflowing mass of gas in Hopkins et al. (2005d), by assuming that while it is growing, the black hole mass is proportional to the inflowing gas mass in the galaxy core (which ultimately produces the Magorrian et al. relation between black hole and bulge mass), and assuming Bondi accretion, with obscuration along a sightline through this (spherically symmetric) gas inflow. Such a model gives the observed correlation between $`N_\mathrm{H}`$ and $`L`$, and explains the weak dependence of the column density-luminosity relation on the ISM gas equation of state. The assumptions above give a relationship of the form
$$N_\mathrm{H}f_0\frac{1}{m_HR_c}\left(\frac{c_s}{c}\right)\left(\frac{cL}{G^2}\right)^{1/3},$$
(2)
where $`f_050`$ is a dimensionless factor depending on the radiative efficiency, mean molecular weight, density profile, and assumed $`M_{\mathrm{BH}}\sigma `$ relation; $`m_H`$ is the mass of hydrogen; $`R_c`$ the radius of the galaxy core ($`100\mathrm{pc}`$); and $`c_s`$ the effective sound speed in the central regions of the galaxy. A $`q_{\mathrm{EOS}}=1.0`$ equation of state, with a higher effective temperature, results in a factor of $`2`$ larger sound speed in the densest regions of the galaxy than a $`q_{\mathrm{EOS}}=0.25`$ equation of state (Springel et al., 2005b), explaining the weak trend seen. In any event, the dependence is small compared to the intrinsic scatter for either equation of state in the value of $`\overline{N}_H`$ at a given luminosity, and further weakens at high luminosity, so it can be neglected. What may appear to be a systematic offset in $`\overline{N}_H`$ with $`V_{\mathrm{vir}}`$ is actually just a tendency for larger $`V_{\mathrm{vir}}`$ systems to be at higher luminosities; there is no significant change in the dependence of $`N_\mathrm{H}`$ on $`L`$.
We use our large set of simulations to improve our fits (relative to those of Hopkins et al. 2005d) to the $`N_\mathrm{H}`$ distribution as a function of instantaneous and peak luminosities. Looking at individual simulations, there appears to be a โbreakโ in the power-law scaling of $`\overline{N}_H`$ with $`L`$ at $`L10^{11}L_{\mathrm{}}`$. We find that the best fit to the median column density $`\overline{N}_H`$ is then
$$\overline{N}_H=\{\begin{array}{cc}10^{22.8}\mathrm{cm}^2\left(\frac{L}{L_{\mathrm{peak}}}\right)^{0.54}\hfill & \mathrm{if}L<10^{11}L_{\mathrm{}}\hfill \\ 10^{21.9}\mathrm{cm}^2\left(\frac{L}{10^{11}L_{\mathrm{}}}\right)^{0.43}\hfill & \mathrm{if}L>10^{11}L_{\mathrm{}}.\hfill \end{array}$$
(3)
Either of these two relations provides an acceptable fit to the plotted $`\overline{N}_H`$ distribution if applied to the entire luminosity range ($`\chi ^2/\nu 2.8,\mathrm{\hspace{0.17em}3.2}`$ for the first and second relations, respectively), but their combination provides a significantly better fit ($`\chi ^2/\nu 1.5`$), although it is clear from the large scatter in $`\overline{N}_H`$ values that any such fit is a rough approximation. Despite the complicated form of this equation, it is, in practice, similar to our $`\overline{N}_HL^{0.35}`$ fit from previous work and $`\overline{N}_HL^{1/3}`$ analytical scaling over the range of relevant luminosities, but is more accurate by a factor $`23`$ at low ($`10^9L_{\mathrm{}}`$) luminosities. For comparison, however, we do consider this simpler form for $`N_\mathrm{H}(L)`$ as well as our more accurate fit above in our subsequent analysis, and find that it makes little difference to most observable quasar properties. At the highest luminosities, near the peak luminosities of the brightest quasars, the scatter about these fitted median $`\overline{N}_H`$ values increases, and as noted above the impact of the quasar in expelling surrounding gas becomes important and column densities vary rapidly. We consider this โblowoutโ phase in more detail in ยง 4.
We find that any dependence of $`\sigma _{N_H}`$ (the fitted lognormal dispersion) on $`L`$ or $`L_{\mathrm{peak}}`$ is not statistically significant, with approximately constant $`\sigma _{N_H}0.4`$ for individual simulations. We similarly find no systematic dependence of $`\sigma _{N_H}`$ on any of our varied simulation parameters. However, it is important to note that while the dispersion in $`N_\mathrm{H}`$ for an individual simulation is $`\sigma _{N_H}0.4`$, the dispersion in $`\overline{N}_H`$ across all simulations at a given luminosity is large, $`1`$ dex. Thus, we fit the effective $`\sigma _{N_H}`$ at a given luminosity for the distribution of quasars and find it is $`\sigma _{N_H}1.2`$. Although we have slightly revised our fits for greater accuracy at low luminosities, we note that this relation is shallower than the relation $`N_\mathrm{H}L`$ roughly expected if $`M_{\mathrm{BH}}`$ is constant ($`L\rho N_\mathrm{H}`$) or $`LM_{\mathrm{BH}}`$ always, and strongly contrasts with unification models which predict static obscuration, or evolutionary models in which $`N_\mathrm{H}`$ is independent of $`L`$ up to some threshold (e.g., Fabian, 1999).
### 2.4. Quasar Lifetimes & Sensitivity to Simulation Parameters
We define the luminosity-dependent quasar lifetime $`t_Q=t_Q(L_{\mathrm{min}})`$ as the time a quasar has a luminosity above a certain reference luminosity $`L_{\mathrm{min}}`$; i.e. the total time the quasar shines at $`LL_{\mathrm{min}}`$. For ease of comparison across frequencies, we measure the lifetime in terms of the bolometric luminosity, $`L`$, rather than e.g. the B-band luminosity. Knowing the distribution of column densities $`N_\mathrm{H}`$ as a function of luminosity and system properties (see ยง 2.3), we can then analytically or numerically calculate the distribution of observed lifetimes at any frequency if we know this intrinsic lifetime. Below $`1`$ Myr, our estimates of $`t_Q`$ become uncertain owing to the effects of quasar variability and our inability to resolve the local small-scale physics of the ISM, but this is significantly shorter than even the most rapid timescales $`10`$ Myr of substantial quasar evolution.
As before, we use our diverse sample of simulations to test for systematic effects in our parameterization of the quasar lifetime. Figure 4 shows the quasar lifetime as a function of reference luminosity $`L_{\mathrm{min}}`$ for both a set of simulations with similar total galaxy mass, $`M_{\mathrm{gal}}10^{12}M_{\mathrm{}}`$, and similar final black hole mass (i.e. similar peak quasar luminosity), $`M_{\mathrm{BH}}^f10^8M_{\mathrm{}}`$. In each case, the simulations cover a range in $`q_{\mathrm{EOS}},`$ $`f_{\mathrm{gas}},`$ $`z_{\mathrm{gal}},`$ $`\mathrm{and}V_{\mathrm{vir}}`$.
As Figure 4 demonstrates, at a given $`M_{\mathrm{gal}}`$, there is a wide range of lifetimes, with a systematic dependence on several quantities. For example, for fixed $`M_{\mathrm{gal}}`$, a lower $`q_{\mathrm{EOS}}`$ means that the gas is less pressurized and more easily collapses to high density, resulting in larger $`M_{\mathrm{BH}}^f`$ and longer lifetimes at higher luminosities. Similarly, higher $`f_{\mathrm{gas}}`$ provides more fuel for black hole growth at fixed $`M_{\mathrm{gal}}`$. However, for a given $`M_{\mathrm{BH}}^f`$, the lifetime $`t_Q`$ as a function of $`L_{\mathrm{min}}`$ is similar across simulations and shows no systematic dependence on any of the varied parameters. We find this for all final black hole masses in our simulations, in the range $`M_{\mathrm{BH}}^f10^610^{10}M_{\mathrm{}}`$. We have further tested this as a function of resolution, comparing with alternate realizations of our fiducial A3 simulation with up to 128 times as many particles, and find similar results as a function of $`M_{\mathrm{BH}}^f`$.
From Figure 4, it is clear that the final black hole mass or peak luminosity is a better variable to use in describing the lifetime than the host galaxy mass. The lack of any systematic dependence of either the quasar lifetime or $`N_\mathrm{H}`$($`L,L_{\mathrm{peak}}`$) on host galaxy properties implies that our earlier results (Hopkins et al. 2005a-d) are reliable and can be applied to a wide range of host galaxy properties, redshifts, and luminosities, although we refine and expand the various fits of these works and their applications herein. Furthermore, the large scatter in $`t_Q`$ at a given galaxy mass has important implications for the quasar correlation function as a function of luminosity, as one cannot associate a single quasar luminosity with hosts of a given mass (see Lidz et al. 2005).
Although the truncated power-laws we have previously fitted to $`t_Q`$ using only the A-series simulations (Hopkins et al., 2005b) provide acceptable fits to all our runs, we use our new, larger set of simulations to improve the accuracy of the fits and average over peculiarities of individual simulations, giving a more robust prediction of the lifetime as a function of instantaneous and peak luminosity. For a given peak luminosity $`L_{\mathrm{peak}}`$, we consider simulations with an $`L_{\mathrm{peak}}`$ within a factor of 2, and take the geometric mean of their lifetimes $`t_Q`$($`L`$) (we ignore any points where $`t_Q<1`$ Myr, as our calculated lifetimes are uncertain below this limit). We can then differentiate this numerically to obtain $`\mathrm{d}t/\mathrm{d}\mathrm{log}L`$ (the time spent in a given logarithmic luminosity interval), and fit some functions to both curves simultaneously. Figure 5 illustrates this and shows the results of our fitting. We find that both the integrated lifetime $`t_Q`$($`L`$) and the differential lifetime $`\mathrm{d}t/\mathrm{d}\mathrm{log}L`$ are well fitted by an exponential,
$$\mathrm{d}t/\mathrm{d}\mathrm{log}L=t_Q^{}\mathrm{exp}[L/L_Q^{}],$$
(4)
where both $`t_Q^{}`$ and $`L_Q^{}`$ are functions of $`M_{\mathrm{BH}}^f`$ or $`L_{\mathrm{peak}}`$. The best-fit such $`\mathrm{d}t/\mathrm{d}\mathrm{log}L`$ is shown in the figure as a solid line for simulations with $`L_{\mathrm{peak}}10^{10}L_{\mathrm{}}`$, and agrees well with both the numerical derivative $`\mathrm{d}t/\mathrm{d}\mathrm{log}L`$ (lower left, black histogram) and the geometric mean $`t_Q(L)`$ (upper left, black histogram). This of course implies
$$t_Q(L)=t_Q^{}_L^{L_{\mathrm{peak}}}e^{L/L_Q^{}}d\mathrm{log}L,$$
(5)
but we are primarily interested in $`\mathrm{d}t/\mathrm{d}\mathrm{log}L`$ in our subsequent analysis.
Although our fitted lifetime involves an exponential, it is in no way similar to the exponential light curve of constant Eddington-ratio black hole growth or the model in, e.g., Haiman & Loeb (1998), which give $`\mathrm{d}t/\mathrm{d}\mathrm{log}L=`$constant$`t_St_Q^{}`$.
Our functional form also has the advantage that, although it should formally be truncated with $`\mathrm{d}t/\mathrm{d}\mathrm{log}L=0`$ for $`L>L_{\mathrm{peak}}`$, the values in this regime fall off so quickly that we can safely use the above fit for all large $`L`$. Similarly, at $`L10^4L_{\mathrm{peak}}`$, $`\mathrm{d}t/\mathrm{d}\mathrm{log}L`$ falls below the constant $`t_Q^{}`$ to which this equation asymptotes. Furthermore, in this regime, the fits above begin to differ significantly from those obtained by fitting e.g. truncated power-laws or Schechter functions. However, these luminosities are well below those we generally consider and well below the luminosities where the contribution of a quasar with some $`L_{\mathrm{peak}}`$ is significant to the observed quantities we predict. Moreover, this turndown (i.e. the lower value predicted by an exponential as opposed to a power-law or Schechter function at low luminosities) is at least in part an artifact of the finite simulation duration. The values here are also significantly more uncertain, as by these low relative accretion rates, the system is likely to be accreting in some low-efficiency, ADAF state (e.g. Narayan & Yi 1995), which we do not implement directly in our simulations. Rather than introduce additional uncertainties into our modeling when they do not affect our predictions, we adopt these exponential fits which are accurate at $`L10^410^3L_{\mathrm{peak}}`$. However, for purposes where the faint-end behavior of the quasar lifetime is important, such as predicting the value and evolution of the faint-end quasar luminosity function slope with redshift, a more detailed examination of the lifetime at low luminosities and relaxation of quasars after the โblowoutโ phase is necessary, and we consider these issues separately in Hopkins et al. (2005f).
We also note that in Hopkins et al. (2005c) we considered several extreme limits to our modeling, neglecting all times before the final merger and applying an ADAF correction at low accretion rates (taken into account a posteriori by rescaling the radiative efficiency $`ฯต_r`$ with accretion rate, given the assumption that such low accretion rates do not have a large dynamical effect on the system regardless of radiative efficiency), and found that this does not change our results โ the lifetime at low luminosities may be slightly altered but the key qualitative point, that the quasar lifetime increases with decreasing luminosity, is robust against a wide range of limits designed to decrease the lifetime at low luminosities.
Figure 5 further shows the fitted $`t_Q^{}`$ (upper right) and $`L_Q^{}`$ (lower right) as a function of peak quasar luminosity for each $`L_{\mathrm{peak}}`$. We find that $`L_Q^{}`$, the luminosity above which the lifetime rapidly decreases, is proportional to $`L_{\mathrm{peak}}`$,
$$L_Q^{}=\alpha _LL_{\mathrm{peak}},$$
(6)
with a best fit coefficient $`\alpha _L=0.20`$ (solid line). The weak dependence of $`t_Q^{}`$ on $`L_{\mathrm{peak}}`$ is well-described by a power-law,
$$t_Q^{}=t_{}^{(10)}\left(\frac{L_{\mathrm{peak}}}{10^{10}L_{\mathrm{}}}\right)^{\alpha _T},$$
(7)
with $`t_{}^{(10)}=1.37\times 10^9\mathrm{yr}`$ and $`\alpha _T=0.11`$
The presence or absence of a stellar bulge in the progenitors can have a significant impact on the quasar light curve (Springel et al. 2005b), primarily affecting the strength of the strong accretion phase associated with initial passage of the merging galaxies (e.g. Mihos & Hernquist 1994). Likewise, the seed mass of the simulation black holes could have an effect, as black holes with smaller initial masses will spend more time growing to large sizes, and more massive black holes may be able to shut down early phases of accretion in mergers in minor โblowoutโ events. In Figure 6, we show various tests to examine the robustness of our fitted quasar lifetimes to these variations. We have re-run our fiducial Milky Way-like A3 simulation both with (right panels) and without (left panels) initial stellar bulges in the merging galaxies and varying the initial black hole seed masses from $`10^410^7M_{\mathrm{}}`$. In each case we compare the lifetime $`t_Q`$ determined directly from the simulations (crosses) to that predicted from our fits above (diamonds), based only on the peak luminosity (final black hole mass) of the simulated quasar. Again, we find that varying these simulation parameters can have a significant effect on the final black hole mass, but that the quasar lifetime as a function of peak luminosity is a robust quantity, independent of initial black hole mass or the presence or absence of a bulge in the quasar host.
We can integrate the total radiative output of our model quasars,
$$E_{\mathrm{rad}}=_{L_{\mathrm{min}}}^{L_{\mathrm{peak}}}L\frac{\mathrm{d}t}{\mathrm{d}\mathrm{log}L}\mathrm{d}\mathrm{log}L,$$
(8)
and using our fitted formulae and $`L_{\mathrm{min}}L_Q^{}`$ we find
$$E_{\mathrm{rad}}=L_Q^{}t_Q^{}\mathrm{log}e(1e^{L_{\mathrm{peak}}/L_Q^{}}).$$
(9)
Knowing $`E_{\mathrm{rad}}=ฯต_rM_{\mathrm{BH}}^fc^2`$, we can compare the final black hole mass as a function of peak luminosity to what we would expect if the peak luminosity were the Eddington luminosity of a black hole with mass $`M_{\mathrm{Edd}}`$, $`L_{\mathrm{Edd}}=ฯต_rM_{\mathrm{Edd}}c^2/t_S`$, where $`t_S`$ is the Salpeter time for $`ฯต_r=0.1`$. Equating $`E_{\mathrm{rad}}=ฯต_rM_{\mathrm{BH}}^fc^2`$ with the value calculated in Equation 9, and using the definition of the Eddington mass at $`L=L_{\mathrm{peak}}`$ and our fitted $`L_Q^{}=\alpha _LL_{\mathrm{peak}}`$, we obtain
$$\frac{M_{\mathrm{BH}}^f(L_{\mathrm{peak}})}{M_{\mathrm{Edd}}(L_{\mathrm{peak}})}=\alpha _L\left(\frac{t_Q^{}}{t_S}\right)\mathrm{log}e1.24f_T,$$
(10)
where $`f_T=(L_{\mathrm{peak}}/10^{13}L_{\mathrm{}})^{0.11}`$ for the power-law fit to $`t_Q^{}`$. For our calculations explicitly involving black hole mass, we adopt this conversion unless otherwise noted, as we have performed our primary calculation (i.e. calculated $`\dot{n}(L_{\mathrm{peak}})`$) in terms of peak luminosity. Moreover, although this agrees well with the black hole masses in our simulations as a function of peak luminosity (as it must if the fitted quasar lifetimes are accurate), this allows us to smoothly interpolate to the highest black hole masses ($`\mathrm{a}\mathrm{few}\times 10^910^{10}M_{\mathrm{}}`$), which are of particular interest in examining the black hole population but for which the number of simulations we have with a given final black hole mass drops rapidly.
This gives explicitly the modifications to the black hole mass compared to that inferred from the โlight bulbโ and โconstant Eddington ratioโ models which we outline below in ยง 2.5, in which quasars shine at constant luminosity or follow exponential light curves, and for which $`M_{\mathrm{BH}}^f=M_{\mathrm{Edd}}(L_{\mathrm{peak}})/l`$, where $`l`$, the (constant) Eddington ratio, is generally adopted. The corrections are small, and therefore most of the black hole mass is accumulated in the bright, near-peak quasar phase, in good agreement with observational estimates (e.g., Soltan, 1982; Yu & Tremaine, 2002); we discuss this in greater detail in ยง 4 and ยง 6. Furthermore, the increase of $`f_T`$ with decreasing $`L_{\mathrm{peak}}`$ implies that lower-mass quasars accumulate a larger fraction of their mass in slower, sub-peak accretion after the final merger, while high-mass objects acquire essentially all their mass in the peak quasar phase. This is seen directly in our simulations, and is qualitatively in good agreement with expectations from simulations and semi-analytical models in which the $`M_{\mathrm{BH}}\sigma `$ relation is set by black hole feedback in a strong quasar phase. Compared to the assumption that $`M_{BH}^f=M_{\mathrm{Edd}}(L_{\mathrm{peak}})`$, this formula introduces a small but non-trivial correction in the relic supermassive black hole mass function implied by the quasar luminosity function and $`\dot{n}(L_{\mathrm{peak}})`$ (see ยง 6).
The predictions of our model for the quasar lifetime and evolution can be applied to observations which attempt to constrain the quasar lifetime from individual quasars, for example using the proximity effect in the Ly$`\alpha `$ forest (Bajtlik, Duncan, & Ostriker, 1988; Haiman & Cen, 2002; Jakobsen et al., 2003; Yu & Lu, 2005) and multi-epoch observations (Martini & Schneider, 2003). However, many observations designed to constrain the quasar lifetime do so not for individual quasars, but using demographic or integral arguments based on the population of quasars in some luminosity interval (e.g., Soltan, 1982; Haehnelt, Natarajan, & Rees, 1998; Yu & Tremaine, 2002; Yu & Lu, 2004; Porciani, Magliocchetti, & Norberg, 2004; Grazian et al., 2004). Our prediction for these observations is similar but slightly more complex, as an observed luminosity function at a given luminosity will consist of sources with different peak luminosities $`L_{\mathrm{peak}}`$, but the same instantaneous luminosity, $`L`$. Furthermore, the lifetime being probed may be either the integrated quasar lifetime above some luminosity threshold or the differential lifetime at a particular luminosity.
For a given determination of the quasar luminosity function using our model for quasar lifetimes and some distribution of peak luminosities, we can predict the distribution of quasar lifetimes as a function of the observed luminosity interval. Figure 7 shows an example of such a result, using the determination of the luminosity function below in ยง 3.2, at redshift $`z=0.5`$. We consider several bolometric luminosities spanning the luminosity function from $`10^910^{14}L_{\mathrm{}}`$, and for each, the distribution of sources (peak luminosities), and the corresponding distribution of quasar lifetimes. We show both the distribution of integrated quasar lifetimes $`t_Q`$ (left panel) and the distribution of differential quasar lifetimes $`\mathrm{d}t/\mathrm{d}\mathrm{log}L`$ (right panel). The evolution with redshift is weak, with the lifetime increasing by $`1.52`$ at a given luminosity at $`z=2`$. There is furthermore an ambiguity of a factor $`2`$, as some of the quasars observed at a given luminosity will only be entering a peak quasar phase, whereas the lifetimes shown are integrated over the whole quasar evolution. This prediction is quite different from that of the optical quasar phase which we describe below in ยง 4 and in Hopkins et al. (2005a), as it considers only the intrinsic bolometric luminosity, but our modeling and the fits provided above for the bolometric lifetime and column density distributions should enable the prediction of these quantities, considering attenuation, in any waveband. In either case, it is clear that the lifetime distribution for lower-luminosity quasars is increasingly more strongly peaked and centered around longer lifetimes, in good agreement with the limited observational evidence from e.g. Adelberger & Steidel (2005). This is a consequence of the fact that in our model quasar lifetimes decrease with increasing luminosity. The range spanned in the figure corresponds well to the range of quasar lifetimes implied by the observations above and others (e.g. Martini, 2004, and references therein).
### 2.5. Alternative Models of Quasar Evolution
Our modeling reproduces at least the observed hard X-ray quasar luminosity function by construction, since we use the observed quasar luminosity functions to determine the birthrate of quasars of a given $`L_{\mathrm{peak}}`$, $`\dot{n}(L_{\mathrm{peak}})`$, in ยง 3.2. It is therefore useful to consider in detail the differences in our subsequent predictions between various models for the quasar lifetime and obscuration, in order to determine to what extent these predictions are implied by any model that successfully reproduces the observed quasar luminosity function, and to what extent they are independent of the observed luminosity functions and instead depend on the model of quasar evolution adopted. To this end, we define two models for the quasar lifetime, and two models for the distribution of quasar column densities, combinations of which have been commonly used in most previous analyses of quasars.
For the quasar lifetime, we consider the following two cases:
โLight-Bulb Modelโ (e.g., Small & Blandford, 1992; Kauffmann & Haehnelt, 2000; Wyithe & Loeb, 2003; Haiman, Quataert, & Bower, 2004). The simplest possible model for the quasar light curve, the โfeast or famineโ or โlight-bulbโ model assumes that quasars have only two states, โonโ and โoff.โ Quasars turn โonโ, shine at a fixed bolometric luminosity $`L=L_{\mathrm{peak}}`$, defined by a โconstantโ Eddington ratio (i.e. $`L_{\mathrm{peak}}=lM_{\mathrm{BH}}^f`$) and constant quasar lifetime $`t_{Q,\mathrm{LB}}`$. Models where quasars live arbitrarily long with slowly evolving mean volume emissivity or mean light curve (e.g. Small & Blandford, 1992; Haiman & Menou, 2000; Kauffmann & Haehnelt, 2000) are equivalent to the โlight bulbโ scenario, as they still assume that quasars observed at a luminosity $`L`$ radiate at that approximately constant luminosity over some universal lifetime $`t_{Q,\mathrm{LB}}`$ at a particular redshift. We adopt $`l=0.3`$ and $`t_{Q,\mathrm{LB}}=10^7`$yr, as is commonly assumed in theoretical work and suggested by observations (given this prior) (e.g. Yu & Tremaine, 2002; Martini, 2004; Soltan, 1982; Yu & Lu, 2004; Porciani, Magliocchetti, & Norberg, 2004; Grazian et al., 2004), and similar to the $`e`$-folding time of a black hole with canonical radiative efficiency $`ฯต_r=0.1`$ (Salpeter, 1964) or the dynamical time in a typical galactic disk or central regions of the merger. These choices control only the normalization of $`\dot{n}(L_{\mathrm{peak}})`$, and therefore do not affect most of our predictions. Where the normalization (i.e. value of the constant $`t_Q`$ or $`l`$) is important, we allow it to vary in order to produce the best possible fit to the observations.
โExponential (Fixed Eddington Ratio) Model.โ A somewhat more physical model of the quasar light curve is obtained by assuming growth at a constant Eddington ratio, as is commonly adopted in e.g. semi-analytical models which attempt to reproduce quasar luminosity functions (e.g. Kauffmann & Haehnelt, 2000; Wyithe & Loeb, 2003; Volonteri et al., 2003). In this model, a black hole accretes at a fixed Eddington ratio $`l`$ from an initial mass $`M_i`$ to a final mass $`M_f`$ (or equivalently, a final luminosity $`L_f=lL_{Edd}(M_f)`$), and then shuts off. This gives exponential mass and luminosity growth, and the time spent in any logarithmic luminosity bin is constant,
$$dt/\mathrm{d}\mathrm{log}(L)=t_S(\mathrm{ln}(10)/l)$$
(11)
for $`L_i<L<L_f`$. This is true for any exponential light curve; i.e. this model includes cases with an exponential decline in quasar luminosity), $`f(t)e^{\pm t/t_{}}`$, such as that of Haiman & Loeb (1998), with only the normalization $`dt/\mathrm{d}\mathrm{log}(L)=t_{}\mathrm{ln}(10)`$ changed, and thus any such model will give identical results with correspondingly different normalizations. As with the โlight-bulbโ model, we are free to choose the characteristic Eddington ratio and corresponding timescale for this lightcurve, and we adopt $`l=0.3`$ (i.e. $`t_{}10^8`$yr) in general. Again, however, we allow the normalization to vary freely where it is important, such that these models have the best chance to reproduce the observations. For our purposes, models in which this timescale is determined by e.g. the galaxy dynamical time and thus are somewhat dependent on host galaxy mass or redshift are nearly identical to this scenario. Further, insofar as the dynamical time increases weakly with increasing host galaxy mass (as, e.g. for a spheroid with $`M_{\mathrm{BH}}M_{\mathrm{vir}}a\sigma ^2/G`$, where $`a`$ is the spheroid scale length and $`M_{\mathrm{BH}}\sigma ^4`$, such that $`t_{\mathrm{dyn}}a/\sigma \sigma M_{\mathrm{vir}}^{1/4}`$), this produces behavior qualitatively opposite to our predictions (of increasing lifetime with decreasing instantaneous luminosity), and yields results which are even more discrepant from our predictions and the observations than the constant (host-galaxy independent) case.
A wide variety of โlight-bulbโ or exponential (constant Eddington ratio) models are possible, allowing for different distributions of typical Eddington ratios and/or quasar lifetimes (see e.g. Steed & Weinberg 2003 for an extensive comparison of several classes of such models), but for our purposes they are essentially identical insofar as they do not capture the essential qualitative features of our quasar lifetimes, namely that the quasar lifetime depends on both instantaneous and peak luminosities, and increases with decreasing instantaneous luminosity.
We fit both of the simple models above to the observed quasar luminosity functions in the same manner described in ยง 3, (i.e. in the same manner as we fit our more complicated models of quasar evolution), to determine $`\dot{n}(L_{\mathrm{peak}})_{\mathrm{LB}}`$ for the โlight-bulbโ model and $`\dot{n}(L_{\mathrm{peak}})_{\mathrm{Edd}}`$ for the โfixed Eddington ratioโ model (see Equations 15 and 16, respectively). Thus all three models of the quasar light curve, the โlight-bulbโ, โfixed Eddington ratioโ, and our luminosity-dependent lifetimes model produce an essentially identical bolometric luminosity function.
We also consider two commonly adopted alternative models for the column density distribution and quasar obscuration:
โStandard (Luminosity-Independent) Torusโ (e.g. Antonucci, 1993). This is the canonical obscuration model, based on observations of local, low-luminosity Seyfert galaxies (e.g., Risaliti et al., 1999). The column density distribution is derived from the torus geometry, where we assume the torus inner radius lies at a distance $`R_\mathrm{T}`$ from the black hole, with a height $`H_\mathrm{T}`$, and a density distribution $`\rho (\theta )\mathrm{exp}(\gamma |\mathrm{cos}\theta |)`$, where $`\theta `$ is the polar angle and the torus lies in the $`\theta =0`$ plane. This results in a column density as a function of viewing angle of
$`N_\mathrm{H}(\theta )`$ $`=`$ $`N_{\mathrm{H},\mathrm{\hspace{0.17em}0}}\mathrm{exp}(\gamma |\mathrm{cos}\theta |)\mathrm{cos}(90\theta )`$ (12)
$`\times \sqrt{\left({\displaystyle \frac{R_\mathrm{T}}{H_\mathrm{T}}}\right)^2\mathrm{sec}^2(90\theta )\left(\left({\displaystyle \frac{R_\mathrm{T}}{H_\mathrm{T}}}\right)^21\right)}`$
(Treister et al., 2004). Here, $`N_{\mathrm{H},\mathrm{\hspace{0.17em}0}}`$ is the column density along a line of sight through the torus in the equatorial plane and $`\gamma `$ parameterizes the exponential decay of density with viewing angle. This is a phenomenological model, and as a result the parameters are essentially all free. We adopt typical values, an equatorial column density $`N_{\mathrm{H},\mathrm{\hspace{0.17em}0}}=10^{24}`$ cm<sup>-2</sup>, radius-to-height ratio $`R_\mathrm{T}/H_\mathrm{T}=1.1`$, and density profile $`\gamma =4`$. This combination of parameters follows Treister et al. (2004), and is designed to fit the observed X-ray column density distribution and give a ratio of obscured to unobscured quasars $`3`$, similar to the mean locally observed value (e.g. Risaliti et al., 1999).
โReceding (Luminosity-Dependent) Torusโ (e.g. Lawrence, 1991). Many observations suggest that the fraction of obscured objects depends on luminosity (Steffen et al., 2003; Ueda et al., 2003; Hasinger, 2004; Grimes, Rawlings, & Willott, 2004; Sazonov & Revnivtsev, 2004; Barger et al., 2005; Simpson, 2005). Therefore, some theoretical works have adopted a โreceding torusโ model, in which the torus radius $`R_\mathrm{T}`$ (i.e. distance from the quasar) is allowed to vary with luminosity, but the height and other parameters remain constant. The torus radius is assumed to increase with luminosity, enlarging the opening angle and thus the fraction of unobscured quasars. In this case, the column densities are identical to those shown above, but now $`R_\mathrm{T}/H_\mathrm{T}=(L/L_0)^{0.5}`$, where $`L_010^{11}L_{\mathrm{}}`$ is the luminosity at which the ratio of obscured to unobscured quasars is $`3:1`$ and the power-law slope is chosen to fit the dependence of obscured fraction on luminosity.
Both of these column density distributions represent phenomenological models with several free parameters, explicitly chosen to reproduce the observed differences in quasar luminosity functions and column density distributions. Despite this, it is not clear that these functional forms represent the best possible fit to the observations they are designed to reproduce. Furthermore, comparison of our results in which column density distributions depend on luminosity and peak luminosity elucidates the importance of proper modeling of the dependence of column density on quasar evolution.
## 3. The Quasar Luminosity Function
### 3.1. The Effect of Luminosity-Dependent Quasar Lifetimes
Given quasar lifetimes as functions of both instantaneous and peak luminosities, the observed quasar luminosity function (in the absence of selection effects) is a convolution of the lifetime with the intrinsic distribution of sources with a given $`L_{\mathrm{peak}}`$. If sources of a given $`L`$ are created at a rate $`\dot{n}(L,t)`$ (per unit comoving volume) at cosmological time $`t_H1/H(z)`$ and live for some lifetime $`\mathrm{\Delta }t_Q(L)`$, the total comoving number density observed will be
$$\mathrm{\Delta }n=_{t_H}^{t_H+\mathrm{\Delta }t_Q(L)}\dot{n}(L,t)dt,$$
(13)
which, for a cosmologically evolving $`\dot{n}(L,t)`$, can be expanded about $`\dot{n}(L,t_H)`$, yielding $`\mathrm{\Delta }n=\dot{n}(L,t_H)\mathrm{\Delta }t_Q(L)`$ to first order in $`\mathrm{\Delta }t_Q(L)/t_H`$. Considering a complete distribution of sources with some $`L_{\mathrm{peak}}`$, we similarly obtain the luminosity function
$$\varphi (L)\frac{\mathrm{d}\mathrm{\Phi }}{\mathrm{d}\mathrm{log}L}(L)=\frac{\mathrm{d}t(L,L_{\mathrm{peak}})}{\mathrm{d}\mathrm{log}(L)}\dot{n}(L_{\mathrm{peak}})\mathrm{d}\mathrm{log}(L_{\mathrm{peak}}).$$
(14)
Throughout, we will denote the differential luminosity function, i.e. the comoving number density of quasars in some logarithmic luminosity interval, as $`\varphi \mathrm{d}\mathrm{\Phi }/\mathrm{d}\mathrm{log}L`$. Here, $`\dot{n}(L_{\mathrm{peak}})`$ is the comoving number density of sources created per unit cosmological time per logarithmic interval in $`L_{\mathrm{peak}}`$, at some redshift, and $`\mathrm{d}t/\mathrm{d}\mathrm{log}L`$ is the differential quasar lifetime, i.e. the total time that a quasar with a given $`L_{\mathrm{peak}}`$ spends in a logarithmic interval in bolometric luminosity $`L`$. This formulation implicitly accounts for the โduty cycleโ (the fraction of active quasars at a given time), which is proportional to the lifetime at a given luminosity. Corrections to this formula owing to finite lifetimes are of order $`(\mathrm{d}t/\mathrm{d}\mathrm{log}L)/t_H`$, which for the luminosities and redshifts considered here (except for Figure 11), are never larger than $`1/5`$ and are generally $`1`$, which is significantly smaller than the uncertainty in the luminosity function itself.
We next consider the implications of our luminosity-dependent quasar lifetimes for the relation between the observed luminosity function and the distribution of peak luminosities (i.e. intrinsic properties of quasar systems). In traditional models of quasar lifetimes and light curves, this relation is trivial. For example, models in which quasars โturn onโ at fixed luminosity for some fixed lifetime (i.e. the โlight-bulbโ model defined in ยง 2.5) imply
$$\dot{n}(L_{\mathrm{peak}})_{\mathrm{LB}}\varphi (L=L_{\mathrm{peak}}),$$
(15)
and models in which quasar light curves are a pure exponential growth or decay with some cutoff(s) (e.g., exponential or fixed Eddington-ratio models) imply
$$\dot{n}(L_{\mathrm{peak}})_{\mathrm{Edd}}\frac{d\varphi }{\mathrm{d}\mathrm{log}(L)}_{L=L_{\mathrm{peak}}}.$$
(16)
These both have essentially identical shape to the observed luminosity function, qualitatively different from our model prediction that $`\dot{n}(L_{\mathrm{peak}})`$ should turn over at luminosities approximately below the break in the observed luminosity function (see, e.g. Fig. 1 of Hopkins et al. 2005e). The luminosity-dependent quasar lifetimes determined from our simulations imply a new interpretation of the luminosity function, with $`\dot{n}(L_{\mathrm{peak}})`$ tracing the bright end of the luminosity function similar to traditional models, but then peaking and turning over below $`L_{\mathrm{peak}}L_{\mathrm{break}}`$, the break luminosity in standard double power-law luminosity functions. In our deconvolution of the luminosity function, the faint end corresponds primarily to sources in sub-Eddington phases transitioning into or out of the phase(s) of peak quasar activity. There is also some contribution to the faint-end lifetime from quasars accreting efficiently (i.e. growing exponentially at high Eddington ratio) early in their activity and on their way to becoming brighter sources, but this becomes an increasingly small fraction of the lifetime at lower luminosities. For example, in Figure 7 of Hopkins et al. (2005b), direct calculation of the quasar lifetime shows that sub-Eddington phases begin to dominate the lifetime for $`L0.1L_{\mathrm{peak}}`$, with $`90\%`$ of the lifetime at $`L10^3L_{\mathrm{peak}}`$ corresponding to sub-Eddington growth. By definition, a โfixed Eddington ratioโ or โlight bulbโ model is dominated at all luminosities by a fixed, usually large, Eddington ratio. Even models which assume an exponential decline in the quasar luminosity from some peak, although they clearly must spend a significant amount of time at low Eddington ratios, have an identical $`\dot{n}(L_{\mathrm{peak}})=\dot{n}(L_{\mathrm{peak}})_{\mathrm{Edd}}`$ (modulo an arbitrary normalization), and predict far less time at most observable ($`10^4L_{\mathrm{peak}}`$) low luminosities and accretion rates (because the accretion rates fall off so rapidly); i.e. the population at any observed luminosity is still dominated by objects near their peak.
From our new, large set of simulations, we test this model of the relationship between the distribution of peak quasar luminosities and observed luminosity functions, namely our assertion that $`\dot{n}(L_{\mathrm{peak}})`$ should peak around the observed break in the luminosity function, and turn over below this peak, with the observed luminosity function faint-end slope dominated by sources with peak luminosities near the break in sub-Eddington (sub-peak luminosity) states. In particular, we wish to ensure that this behavior for $`\dot{n}(L_{\mathrm{peak}})`$ is real, and not some artifact of our fitting functions for the quasar lifetime.
Figure 8 shows the best fit $`\dot{n}(L_{\mathrm{peak}})`$ distribution (solid thick histogram) fitted to the Ueda et al. (2003) hard X-ray quasar luminosity function (solid curve) at redshift $`z=0.5`$, as well as the resulting best-fit luminosity function (solid thin histogram). For ease of comparison with other quasar luminosities, we rescale the luminosity function to the bolometric luminosity using the corrections of Marconi et al. (2004). We determine $`\dot{n}(L_{\mathrm{peak}})`$ by logarithmically binning the range of $`L_{\mathrm{peak}}`$, and considering for each bin all simulations with $`L_{\mathrm{peak}}`$ in the given range. For each bin, then, we take the average binned time the simulations spend in each luminosity interval, and take that to be the quasar lifetime $`\mathrm{d}t/\mathrm{d}\mathrm{log}L`$. We then fit to the observed luminosity function of Ueda et al. (2003), fitting
$$\varphi (L)\underset{i}{}\dot{n}_i(L_{\mathrm{peak},i})\frac{\mathrm{\Delta }t(L,L_{\mathrm{peak},i})}{\mathrm{\Delta }\mathrm{log}L}$$
(17)
and allowing $`\dot{n}_i(L_{\mathrm{peak},i})`$ to be a free coefficient for each binned $`L_{\mathrm{peak}}=L_{\mathrm{peak},i}`$. Despite our large number of simulations, the numerical binning process makes this result noisy, especially at the extreme ends of the luminosity function. However, the relevant result is clear โ the qualitative behavior of $`\dot{n}(L_{\mathrm{peak}})`$ described above is unchanged. For further discussion of the qualitative differences between the $`\dot{n}(L_{\mathrm{peak}})`$ distribution from different quasar models, and the robust nature of our interpretation even under restrictive assumptions (e.g. ignoring the early phases of merger activity or applying various models for radiative efficiency as a function of accretion rate), we refer to Hopkins et al. (2005c).
### 3.2. The Luminosity Function at Different Frequencies and Redshifts
Given a distribution of peak luminosities $`\dot{n}(L_{\mathrm{peak}})`$, we can use our model of quasar lifetimes and the column density distribution as a function of instantaneous and peak luminosities to predict the luminosity function at any frequency. From a distribution of $`N_\mathrm{H}`$ values and some a priori known minimum observed luminosity $`L_\nu ^{\mathrm{min}}`$, the fraction $`f_{\mathrm{obs}}`$ of quasars with a peak luminosity $`L_{\mathrm{peak}}`$ and instantaneous bolometric luminosity $`L`$ which lie above the luminosity threshold is given by the fraction of $`N_\mathrm{H}`$ values below a critical $`N_H^{\mathrm{max}}`$, where $`L_\nu ^{\mathrm{min}}=f_\nu L\mathrm{exp}(\sigma _\nu N_H^{\mathrm{max}})`$. Here, $`f_\nu (L)L_\nu /L`$ is a bolometric correction and $`\sigma _\nu `$ is the cross-section at frequency $`\nu `$. Thus,
$$N_H^{\mathrm{max}}(\nu ,L,L_\nu ^{\mathrm{min}})=\frac{1}{\sigma _\nu }\mathrm{ln}\left(\frac{f_\nu (L)L}{L_\nu ^{\mathrm{min}}}\right),$$
(18)
and for the lognormal distribution above,
$$f_{\mathrm{obs}}(\nu ,L,L_{\mathrm{peak}},L_\nu ^{\mathrm{min}})=\frac{1}{2}\left[1+\mathrm{erf}\left(\frac{\mathrm{log}(N_H^{\mathrm{max}}/\overline{N}_H)}{\sqrt{2}\sigma _{N_H}}\right)\right].$$
(19)
This results in a luminosity function (in terms of the bolometric luminosity)
$`\varphi (\nu ,L,L_\nu ^{\mathrm{min}})`$ $`=`$ $`{\displaystyle f_{\mathrm{obs}}(\nu ,L,L_{\mathrm{peak}},L_\nu ^{\mathrm{min}})}`$ (20)
$`\times {\displaystyle \frac{dt(L,L_{\mathrm{peak}})}{\mathrm{d}\mathrm{log}(L)}}\dot{n}(L_{\mathrm{peak}})d\mathrm{log}(L_{\mathrm{peak}}),`$
where $`\varphi (\nu ,L,L_\nu ^{\mathrm{min}})`$ is the number density of sources with bolometric luminosity $`L`$ per logarithmic interval in $`L`$, with an observed luminosity at frequency $`\nu `$ above $`L_\nu ^{\mathrm{min}}`$.
Based on the direct fit for $`\dot{n}(L_{\mathrm{peak}})`$ in Figure 8, we wish to consider a functional form for $`\dot{n}(L_{\mathrm{peak}})`$ with a well-defined peak and falloff in either direction in $`\mathrm{log}(L_{\mathrm{peak}})`$. Therefore, we take $`\dot{n}(L_{\mathrm{peak}})`$ to be a lognormal distribution, with
$$\dot{n}(L_{\mathrm{peak}})=\dot{n}_{}\frac{1}{\sigma _{}\sqrt{2\pi }}\mathrm{exp}\left[\frac{1}{2}\left(\frac{\mathrm{log}(L_{\mathrm{peak}}/L_{})}{\sigma _{}}\right)^2\right].$$
(21)
Here, $`\dot{n}_{}`$ is the total number of quasars being created or activated per unit comoving volume per unit time; $`L_{}`$ is the center of the lognormal, the characteristic peak luminosity of quasars being born (i.e. the peak luminosity at which $`\dot{n}(L_{\mathrm{peak}})`$ itself peaks), which is directly related to the break luminosity in the observed luminosity function; and $`\sigma _{}`$ is the width of the lognormal in $`\dot{n}(L_{\mathrm{peak}})`$, and determines the slope of the bright end of the luminosity function. Since our model predicts that the bright end of the luminosity function is made up primarily of sources at high Eddington ratio near their peak luminosity, i.e. essentially identical to โlight-bulbโ or โfixed Eddington ratioโ models, the bright-end slope is a fitted quantity, determined by whatever physical processes regulate the bright-end slope of the active black hole mass function (possibly feedback from outflows or threshold cooling processes, e.g. Wyithe & Loeb 2003; Scannapieco & Oh 2004; Dekel & Birnboim 2004), unlike the faint-end slope which is a consequence of the quasar lifetime itself, and is only weakly dependent on the underlying faint-end active black hole mass or $`\dot{n}(L_{\mathrm{peak}})`$ distribution.
We note that although this choice of fitting function has appropriate general qualities, it is ultimately somewhat arbitrary, and we choose it primarily for its simplicity and its capacity to match the data with a minimum of free parameters. We could instead, for example, have chosen a double power-law form with $`\dot{n}(L_{\mathrm{peak}})=\dot{n}_{}/[(L_{\mathrm{peak}}/L_{})^{\gamma _1}+(L_{\mathrm{peak}}/L_{})^{\gamma _2}]`$ and $`\gamma _1<\gamma _2`$, but given that the entire faint end of the luminosity function is dominated by objects with $`L_{\mathrm{peak}}L_{}`$, the observed luminosity function has essentially no power to constrain the faint end slope $`\gamma _1`$, other than setting an upper limit $`\gamma _10`$. The โtrueโ $`\dot{n}(L_{\mathrm{peak}})`$ will, of course, be a complicated function of both halo merger rates at a given redshift and the distribution of host galaxy properties including, but not necessarily limited to, masses, concentrations, and gas fractions.
Having chosen a form for $`\dot{n}(L_{\mathrm{peak}})`$, we can then fit to an observed luminosity function to determine $`(\dot{n}_{},L_{},\sigma _{})`$. We take advantage of the capability of our model to predict the luminosity function at multiple frequencies, and consider both fits to just the Ueda et al. (2003) hard X-ray (2-10 keV) luminosity function, $`\varphi _{HX}`$, and fits to the Ueda et al. (2003), Miyaji et al. (2001) soft X-ray (0.5-2 keV; $`\varphi _{SX}`$), and Croom et al. (2004) optical B-band (4400 ร
; $`\varphi _B`$) luminosity functions simultaneously. These observations agree with other, more recent determinations of $`\varphi _{HX},\varphi _{SX},\varphi _B`$ (e.g. Barger et al., 2005; Hasinger, Miyaji, & Schmidt, 2005; Richards et al., 2005, respectively) at most luminosities, and therefore we do not expect revisions to the observed luminosity functions to dramatically change our results. In order to avoid numerical artifacts from fitting to extrapolated, low-luminosity slopes in the analytical forms of these luminosity functions, we directly fit to the binned luminosity function data. Thus, we fit each luminosity function in all redshift intervals for which we have binned data.
We find good fits ($`\chi ^2/\nu =68.8/1040.66`$) to all luminosity functions at all redshifts with a pure peak-luminosity evolution (PPLE) model, for which
$$L_{}=L_{}^0\mathrm{exp}(k_L\tau ),\dot{n}_{}=constant,\sigma _{}=constant,$$
(22)
where $`\tau `$ is the fractional lookback time ($`\tau H_0_0^z๐t`$) and $`k_L`$ is a dimensionless constant fitted with $`L_{},\dot{n}_{},\sigma _{}`$. It is important to distinguish this from โstandardโ pure luminosity evolution (PLE) models (e.g., Boyle et al., 1988), as with $`\dot{n}(L_{\mathrm{peak}})>0`$ and $`L_{}=L_{}(z)`$ always, the density of sources, especially as a function of observed luminosity at some frequency, evolves in a non-trivial manner.
We do not find significant improvement in the fits if we additionally allow $`\dot{n}_{}`$ or $`\sigma _{}`$ to evolve with redshift ($`\mathrm{\Delta }\chi ^212`$, depending on the adopted form for the evolution), and therefore consider only the simplest parameterization above (Equation 22). We also find acceptable fits for a pure density evolution model, with $`L_{}=`$ constant and $`\dot{n}_{}=\dot{n}_{}^0\mathrm{exp}(k_N\tau )`$ (both keeping $`\sigma _{}`$ fixed and allowing it to evolve as well). However, the fits are somewhat poorer ($`\chi ^2/\nu 1`$), and the resulting parameters over-produce the present-day density of low-mass supermassive black holes and the intensity of the X-ray background by an order of magnitude, so we do not consider them further. In either case, there is a considerable degeneracy between the parameters $`\sigma _{}`$ and $`L_{}`$, where a decrease in $`L_{}`$ can be compensated by a corresponding increase in $`\sigma _{}`$. This degeneracy is present because, as indicated above, the observed luminosity function only weakly constrains the faint-end slope of $`\dot{n}(L_{\mathrm{peak}})`$.
The observations shown are insufficient at high redshift to strongly resolve the โturnoverโ in the total comoving quasar density at $`z23`$, and thus we acknowledge that there must be corrections to this fitted evolution at higher redshift, which we address below. However, as we primarily consider low redshifts, $`z3`$, and show that the supermassive black hole population and X-ray background are dominated by quasars at redshifts for which our $`\dot{n}(L_{\mathrm{peak}})`$ distribution is well determined, this is not a significant source of error in most of our calculations even if we extrapolate our evolution to $`z3`$.
Figure 9 shows the resulting best-fit PPLE luminosity functions from the best-fit $`\dot{n}(L_{\mathrm{peak}})`$ distribution, for redshifts $`z=03`$. This has the best-fit ($`\chi ^2/\nu =0.67`$) values $`(\mathrm{log}L_{},k_L,\mathrm{log}\dot{n}_{},\sigma _{})=(9.94,\mathrm{\hspace{0.17em}5.61},6.29,\mathrm{\hspace{0.17em}0.91})`$ with corresponding errors $`(0.29,0.28,0.13,0.09)`$. Here, $`L_{}`$ is in solar luminosities and $`\dot{n}_{}`$ in comoving $`\mathrm{Mpc}^3\mathrm{Myr}^1`$. Fitting to the hard X-ray data alone gives a similar fit, with the slightly different values $`(\mathrm{log}L_{},k_L,\mathrm{log}\dot{n}_{},\sigma _{})=(9.54,\mathrm{\hspace{0.17em}4.90},5.86,\mathrm{\hspace{0.17em}1.03})\pm (0.66,0.43,0.37,0.13)`$, $`\chi ^2/\nu =0.7`$ (note the degeneracy between $`L_{}`$ and $`\sigma _{}`$ in the two fits). Our best-fit value of $`k_L=5.6`$ compares favorably to the value $`6`$ found by e.g. Boyle et al. (2000) and Croom et al. (2004) for the evolution of the break luminosity in the observed luminosity function, demonstrating that the break luminosity traces the peak in the $`\dot{n}(L_{\mathrm{peak}})`$ distribution at all redshifts. These fits and the errors were obtained by least-squares minimization over all data points (comparing each to the predicted curve at its redshift and luminosity), assuming the functional form we have adopted for $`\dot{n}(L_{\mathrm{peak}})`$.
The agreement we obtain at all redshifts, in each of the hard X-ray (black solid line), soft X-ray (red dashed line), and B-band (dark blue dotted line) is good. This is not at all guaranteed by our procedure, as the fit is highly over-constrained, because we fit three luminosity functions each at five redshifts to only four free parameters. Of course, the choice of the functional form for $`\dot{n}(L_{\mathrm{peak}})`$ ensures that we should be able to reproduce at least one luminosity function and its evolution (e.g. the hard X-ray luminosity function, which is least affected by attenuation), but our modeling of the column density distributions in mergers allows us to simultaneously reproduce the luminosity functions in different wavebands without imposing assumptions about obscured fractions or sources of attenuation. Expressed as bolometric luminosity functions, $`\varphi _B`$, $`\varphi _{SX}`$, and $`\varphi _{HX}`$ would be identical in the absence of obscuration, similar to the predicted $`\varphi _{HX}`$ as obscuration is minimal in the hard X-ray.
For redshifts $`z1`$, we reproduce in our Figure 10, Fig. 2 of Hopkins et al. (2005d), which shows in detail the agreement between hard X-ray (Ueda et al., 2003), soft X-ray (Miyaji et al., 2000), and optical (Boyle et al., 2000) luminosity functions resulting from the time and luminosity dependent column density distributions derived from the simulations. The differential extinction predicted for different frequencies (and magnitude limits) of observed samples based on the column density distributions in our simulations accounts for the different shape of the luminosity function in each band, and the evolution of the luminosity function with redshift is driven by a changing $`L_{}`$, the peak of the $`\dot{n}(L_{\mathrm{peak}})`$ distribution (Equation 22). We emphasize that in our analysis, the key quantity constrained by observations is the fitted $`\dot{n}(L_{\mathrm{peak}})`$ distribution with redshift. All other quantities and distributions are derived from the basic input physics of our simulations, with no further assumptions or adjustable factors in our modeling beyond the prescription for Bondi (Eddington-limited) accretion and $`5\%`$ energy deposition in the ISM, which are themselves constrained by observations and theory as discussed in ยง 2 and in Di Matteo et al. (2005).
We can, of course, fit the previously defined simpler model of quasar lifetimes, either a โlight-bulbโ or exponential light curve/fixed Eddington ratio model, and obtain an identical hard X-ray luminosity function. We determine these fits (see also Equation 15 & 16) and use them throughout when we compare the predictions of such models (described in ยง 2.5) to those of our simulated quasar lifetimes in our subsequent analysis. Applying a standard torus model to any model of the luminosity function reproduces, by design, the mean offset between the B-band and hard X-ray luminosity functions, as the parameters of this model are tuned to reproduce this offset. As many observations show, the fraction of broad-line quasars increases with luminosity (Steffen et al., 2003; Ueda et al., 2003; Hasinger, 2004; Sazonov & Revnivtsev, 2004; Barger et al., 2005; Simpson, 2005), and so reproducing the relationship between B-band and hard X-ray luminosity functions requires adding parameters to the standard torus model which allow luminosity-dependent scalings, i.e. the class of โreceding torusโ models. These, again by construction, reproduce the distinction between hard X-ray and B-band quasar luminosity functions, including the dependence of this difference on luminosity. These are, however, phenomenological models designed to fit these observations. Our simulations, on the other hand, provide a self-consistent description of the column density, which predicts the differences between hard X-ray, soft X-ray, and optical luminosity functions without the addition of tunable parameters or model features designed to reproduce these observations.
Our fits are accurate down to low luminosities, as is clear from our prediction for the X-ray luminosity function at bolometric luminosities $`L10^9L_{\mathrm{}}`$. Furthermore, we have calculated the predicted $`z0.1`$ luminosity function in the B-band as well as in H$`\alpha `$ emission, using the conversion between the two from Hao et al. (2005) and comparing directly to their luminosity functions for Seyfert galaxies and low-luminosity active galactic nuclei (AGN) (both type I and II), and find that our distribution $`\dot{n}(L_{\mathrm{peak}})`$ and model for quasar lifetimes and obscuration reproduces the complete observed luminosity function down to a B-band luminosity $`M_B16`$. Although our prediction falls below the observed Seyfert luminosity function at fainter magnitudes, there is no reason to believe that mergers should be responsible for all nuclear activity at these luminosities (and indeed alternative fueling mechanisms for such faint objects likely exist) - it is surprising, in fact, that this picture reproduces the observed AGN activity to such faint luminosities.
Using the bolometric corrections of Elvis et al. (1994) instead of Marconi et al. (2004) results in a significantly steeper cutoff in the luminosity function at high bolometric luminosities, as the bolometric luminosity inferred for the brightest observed X-ray quasars is almost an order of magnitude smaller using the Elvis et al. (1994) corrections. However, this is because the Elvis et al. (1994) bolometric corrections do not account for any dependence on luminosity, and further the quasars in the sample of Elvis et al. (1994) are X-ray bright (Elvis et al., 2002), whereas it has been well-established that the ratio of bolometric luminosity to hard or soft X-ray luminosity increases with increasing luminosity (e.g., Wilkes et al., 1994; Green et al., 1995; Vignali et al., 2003; Strateva et al., 2005). Recent comparisons between large samples of quasars selected by both optical and X-ray surveys (Risaliti & Elvis, 2005) further suggests that this is an intrinsic correlation, not driven by e.g. the dependence of obscuration on luminosity. For a direct comparison of the bolometric luminosity functions resulting from the two corrections, we refer to Hopkins et al. (2005d). Our analysis uses the form for the UV to X-ray flux ratio, $`\alpha _{\mathrm{OX}}`$, from Vignali et al. (2003), but our results are relatively insensitive to the different values found in the literature. It is important to account for this dependence, as it creates a significant difference in the high-luminosity end of the bolometric quasar luminosity function and implies that a non-negligible fraction of the brightest quasars are not seen in optical surveys (see the discussion in Marconi et al., 2004; Richards et al., 2005).
Finally, our fitted form for the evolution of the break luminosity, with $`L_{}\mathrm{exp}(k_L\tau )`$, cannot continue to arbitrarily high redshift. At redshifts $`z23`$, this asymptotes because $`\tau 1`$, whereas the observed quasar population declines above $`z2`$. This difference is not important for most of our calculated observables, as they are either independent of high-redshift evolution or evolve with cosmic time in some fashion as $`\dot{n}(L_{\mathrm{peak}})dt`$, with little time and thus negligible contributions to integrated totals at high redshifts. However, some quantities, in particular the high-mass end of the black hole mass function (see ยง 6), which is dominated by the small number of the brightest quasars at high redshifts, can receive large relative contributions from these terms. Therefore, it is important in estimating these quantities to be aware of the turnover in the quasar density at high redshifts.
We quantify this in Figure 11, where we show the predicted broad-line luminosity function (where the broad-line phase is determined below in ยง 6) in six luminosity intervals from $`z1.24.8`$. The intervals are those of the COMBO-17 luminosity function from Wolf et al. (2003), but we further compare to the observed luminosity functions of Warren et al. (1994), Schmidt, Schneider, & Gunn (1995), Kennefick, Djorgovski, & De Carvalho (1995), Fan et al. (2001), and Richards et al. (2005) at the appropriate (labeled) redshifts. At each redshift $`z>2`$, we take the fitted $`\dot{n}(L_{\mathrm{peak}})`$ distribution above (Equations 2122) and rescale it according to an exponential cutoff: either pure density evolution (PDE), $`\dot{n}(L_{\mathrm{peak}})\dot{n}(L_{\mathrm{peak}})\times 10^{\alpha _{\mathrm{PDE}}(z2)}`$, or pure peak luminosity evolution (PPLE), $`L_{}L_{}\times 10^{\alpha _{\mathrm{PPLE}}(z2)}`$. Fitting to the data gives $`\alpha _{\mathrm{PDE}}0.65`$ and $`\alpha _{\mathrm{PPLE}}0.55`$, ($`\chi ^2/\nu 1.3`$ for both) in reasonable agreement with the density evolution of e.g. Fan et al. (2001). We note that this evolution, extrapolated as far as $`z6`$, is consistent also with the constraints on $`z6`$ quasars from Fan et al. (2003), especially in the PPLE case.
In each panel, we plot the resulting broad-line luminosity function (see ยง 4), for both the minimum and maximum redshift of the redshift bin, and both the PPLE (solid lines) and PDE (dashed lines) cases. The degeneracy between these possibilities is well-known, as current observations do not resolve the break in the luminosity function. Furthermore, the predicted luminosity function should be considered uncertain especially at low luminosities, as the quasar lifetime at these luminosities and redshifts can become comparable to the age of the Universe, at which point our formalism for the luminosity function as a function of $`\dot{n}(L_{\mathrm{peak}})`$ becomes inaccurate. However, we are able to make testable predictions, based on differences between the two models in integrated galaxy properties (for example, color-magnitude diagrams of red sequence galaxies at low masses or the fraction of recently formed spheroids as a function of mass and redshift), which distinguish the PPLE and PDE models for the evolution of the quasar luminosity function at $`z23`$ (Hopkins et al., 2005e). Owing to these degeneracies and the poor constraints on the observed high-redshift luminosity functions, we have not considered them (those at $`z>3`$) in our fits to $`\dot{n}(L_{\mathrm{peak}})`$, but use them here to roughly constrain the turnover in the quasar density above $`z2`$ (i.e. fitting to $`\alpha _{\mathrm{PDE}}`$ and $`\alpha _{\mathrm{PPLE}}`$). Which form of the turnover we use makes little difference in our subsequent analysis, but, as discussed above, including some turnover is important in calculating select quantities such as the extreme high-mass end of the black hole mass function.
### 3.3. The Observed $`N_\mathrm{H}`$ Distribution
Given the column density distributions and quasar lifetimes calculated from our simulations in ยง 2, and the quantity $`\dot{n}(L_{\mathrm{peak}})`$ determined above (ยง 3.2), we can predict the distribution of column densities observed in a given sample. This will depend not only on the range of observed luminosities and the redshift of the sample, but also on the minimum observed magnitude and frequency (i.e. the selection function) of the sample. For a nearly complete sample or estimate of the luminosity function, for example the hard X-ray luminosity function, at least to $`N_\mathrm{H}10^{25}\mathrm{cm}^2`$, we can integrate the $`N_\mathrm{H}(L,L_{\mathrm{peak}})`$ distribution over the $`\dot{n}(L_{\mathrm{peak}})`$ distribution (weighted by the lifetime at $`L`$).
Figure 12 plots the resulting distribution of column densities for this analysis. The left panel reproduces and expands upon a portion of Fig. 3 of Hopkins et al. (2005b), showing the distribution of column densities (scaled linearly) expected from the characteristic quasars $`L_{\mathrm{peak}}L_{}`$ of the luminosity function observed in optical samples, based on the simulated column density distributions as a function of luminosity and peak luminosity (solid black line). Specifically, we plot the distribution of neutral $`N_{\mathrm{H}\mathrm{I}}`$ values requiring that the observed B-band luminosity be above some reference value $`L_{B,\mathrm{min}}`$. The smooth curve shown is the best-fit to the $`E_{BV}`$ distribution of bright SDSS quasars with $`z<2.2`$, from Hopkins et al. (2004). The curve has been rescaled in terms of the column density (inverting our gas-to-dust prescription) and plotted about a peak (mode) $`N_{\mathrm{H}\mathrm{I}}`$ (undetermined in Hopkins et al. 2004) of $`N_{\mathrm{H}\mathrm{I}}0.5\times 10^{21}\mathrm{cm}^2`$. The observationally implied $`E_{BV}`$ distribution is determined from fitting to the distribution of photometric reddening in all SDSS bands (i.e. using the five-band photometry as a proxy for spectral fitting) in Sloan quasars, relative to the modal quasar colors at each redshift, for quasars with an absolute magnitude limit $`M_i<22`$. The $`i`$-band absolute magnitude limit imposed in the observed sample, $`M_i<22`$, corresponds approximately to our plotted B-band limit $`L_{B,\mathrm{obs}}>10^{11}L_{\mathrm{}}`$. This estimate does not account for bright but strongly reddened quasars having their colors altered to the point where color selection criteria of quasar surveys will not include them. However, this effect would only serve to bring our distribution into better agreement with observations, as it would slightly lower the high-$`N_{\mathrm{H}\mathrm{I}}`$ tail. We also consider the predictions of a standard torus model and receding (luminosity-dependent) torus model in the figure (dashed and dotted lines, respectively). These should not be taken literally in this case โ they reflect that these phenomenological models do not predict the distribution of low/moderate column densities, but rather assume that all lines of sight not intersecting the torus are โunobscured,โ and encounter some constant, small column density (usually chosen to be $`N_\mathrm{H}10^{20}\mathrm{cm}^2`$).
The right panel of 12 shows the integrated distribution (in $`\mathrm{log}N_\mathrm{H}`$) for a complete hard X-ray sample, both as predicted from our simulations based on the joint distribution of column density, luminosity, and peak luminosity (solid), and for both the standard torus model (dashed) and receding torus model (dotted) described in ยง 2.5. The data shown are the results of Treister et al. (2004) (blue squares) and Mainieri et al. (2005) (red circles), with assumed Poisson errors, from multiband Chandra and HST observations of GOODS fields. The solid squares are obtained by assuming an intrinsic photon index for the soft X-ray quasar spectrum of $`\mathrm{\Gamma }=1.9`$, the open squares assuming $`\mathrm{\Gamma }=1.7`$. For the sake of direct comparison with observed distributions, objects with $`N_\mathrm{H}<10^{21}\mathrm{cm}^2`$, for which only an upper limit to the column density would be determined in X-ray observations, are grouped together and plotted as a single bin at $`N_\mathrm{H}=10^{20}\mathrm{cm}^2`$. The actual distribution below $`10^{21}\mathrm{cm}^2`$ is shown as a dot-dashed line. We note that our model of the quasar spectrum assumes a photon index $`\mathrm{\Gamma }=1.9`$ in the soft X-ray, but this has no effect on the column densities calculated from the surrounding gas in our simulations.
The agreement between the observed column density distribution and the result of our simulations once the same selection effect is applied supports our model for quasar evolution, and the good agreement extends to both optical and X-ray samples. Probing to fainter luminosities or frequencies less affected by attenuation broadens the column density distribution, as is seen from the inferred column density distributions in the X-ray. This broadening occurs because, at lower luminosities, observers will see both intrinsically bright periods extinguished by larger column densities (broadening the distribution to larger $`N_\mathrm{H}`$ values) and intrinsically faint periods with small column densities (broadening the distribution to smaller $`N_\mathrm{H}`$ values). The distribution as a function of reference luminosity is a natural consequence of the dynamics of the quasar activity. Throughout much of the duration of bright quasar activity, column densities rise to high levels as a result of the same process that feeds accretion, producing the well-known reddened population of quasars (e.g. Webster et al., 1995; Brotherton et al., 2001; Francis et al., 2001; Richards et al., 2001; Gregg et al., 2002; White et al., 2003; Richards et al., 2003), extending to bright quasars strongly reddened by large $`N_{\mathrm{H}\mathrm{I}}`$. Furthermore, a significant number of quasars are extinguished from optical samples or attenuated to lower luminosities, giving rise to the distinction between luminosity functions in the hard X-ray, soft X-ray, and optical.
The standard torus model described in ยง 2.5, although unable to predict the distribution of column densities seen in optically, relatively unobscured quasars, does a fair job of reproducing the observed distribution of X-ray column densities. The parameters of the model are, of course, chosen to reproduce the data shown (the model parameters are taken from Treister et al., 2004). Nevertheless, our prediction is still a better fit to the observed distribution, with $`\chi ^2/\nu 2`$ as opposed to $`\chi ^2/\nu 7`$ (although the absolute values depend on the estimated systematic errors in the column density estimations). The receding torus model fares even more poorly in reproducing the observed column density distributions, and is ruled out at high significance ($`\chi ^2/\nu 10`$), although this can be alleviated if the observed samples are assumed to be incomplete above $`N_\mathrm{H}10^{23}\mathrm{cm}^2`$. This disagreement results because, in order to match the observed scaling of broad-line fraction with luminosity (see ยง 4 below), this model assumes a larger covering fraction for the torus at lower luminosities, normalized to a similar obscured fraction as the standard torus model near the break in the observed quasar luminosity function. However, since quasars with luminosities below the break dominate the total number counts, this predicts that the cumulative column density distribution must be significantly more dominated by objects with large covering angles, giving a larger Compton-thick population, inconsistent with the actual observed column density distribution.
Although we do not see a significant fraction of extremely Compton-thick column densities $`N_\mathrm{H}10^{26}\mathrm{cm}^2`$ in the distributions from our simulations, our model does not rule out such values. It is possible that bright quasars in unusually massive galaxies or quasars in higher-redshift, compact galaxies which we have not simulated may, during peak accretion periods, reach such values in their typical column densities. Moreover, as our model assumes $`90\%`$ of the mass of the densest gas is clumped into cold-phase molecular clouds, a small fraction of sightlines will pass through such clouds and measure column densities similar to those shown for the โcold phase gasโ in, e.g. Figure 2 of Hopkins et al. (2005a), $`N_\mathrm{H}10^{2526}\mathrm{cm}^2`$.
Furthermore, we have not determined the โshapeโ at any instant of the obscuration (e.g. the dependence of obscuration on radial direction), as in practice, for most of the most strongly obscured phases in peak merger activity, the central regions of the merging galaxies are highly chaotic. Generally, the scale of the obscuration in the peak merger phases is $`100`$pc, quite different than that implied by most traditional molecular torus models, but we note that our resolution limits, $`20`$pc in the dense central regions of the merger, prevent our ruling out collapse of gas in the central regions into a smaller but more dense torus. However, several efforts to model traditional tori through radiative transfer simulations (e.g., Granato & Danese, 1994; Schartmann et al., 2005) suggest significant column densities produced on scales of $`100200`$pc, comparable to our predictions, and we note that only the solid angle covered by a torus, not the absolute torus scale, is constrained in the typical phenomenological torus model (e.g. Antonucci, 1993).
Whether the obscuration of bright quasars originates on larger scales than is generally assumed is observationally testable, either through direct probes of polarized scattered light tracing the obscuring/reflecting structure (e.g., Zakamska et al., 2005), or through correlations between obscuration and e.g. host galaxy morphologies and inclinations (e.g., Donley et al., 2005). These larger scales typical of the central regions of a galaxy are widely accepted as the scales of obscuration in starbursting systems (e.g. Soifer et al. 1984a,b; Sanders et al. 1986, 1988a,b; for a review, see e.g. Soifer et al. 1987), which in our modeling is associated with rapid obscured quasar growth and precedes the quasar phase. Thus, it is natural to associate obscuration with these large scales in any picture which associates starbursts and rapid black hole growth or quasar activity, as opposed to the smaller scales $``$ pc implied by torus models primarily developed to reproduce observations of quiescent, low-luminosity Type II AGN, which are usually not directly associated with merger activity. These low-luminosity AGN are in a relaxed state, suggesting the possibility that the remaining cold gas in the central regions of our merger remnants will collapse once the violent effects of the merger and bright quasar phase have passed, producing a more traditional small torus in a quiescent nucleus. The central point is that regardless of the form of obscuration, the typical magnitude of the obscuration is a strongly evolving function of time, luminosity, and host system properties, and the observed column density distributions reflect this evolution.
## 4. Broad-Line Quasars
### 4.1. Determining the Broad-Line Phase
Optical samples typically identify quasars through their colors, relying on the characteristic non-stellar power-law continua of such objects. However, observations of X-ray selected AGN show a large population of so-called Type 2 AGN, most of which have Seyfert-like luminosities and typical spectra in X-rays and wavelengths longward of $`1\mu `$m (e.g., Elvis et al., 1994), but are optically obscured to the point where no broad lines are visible. Their optical continua, in other words, resemble those of typical galaxies and thus they are not identified by conventional color selection techniques in optical quasar surveys. Traditional unification models (Antonucci, 1993) have postulated a static torus as the explanation for the existence of the Type 2 population, with such objects viewed through the dusty torus and thus optically obscured. Moreover, both synthesis models of the X-ray background (Setti & Woltjer, 1989; Madau et al., 1994; Comastri et al., 1995; Gilli et al., 1999, 2001) and recent direct observations in large surveys (e.g., Zakamska et al., 2004, 2005) indicate the existence of a population of Type 2 quasars, with similar obscuration but intrinsic (unobscured) quasar-like luminosities.
Observations of both radio-loud (Hill, Goodrich, & DePoy, 1996; Simpson, Rawlings, & Lacy, 1999; Willott et al., 2000; Simpson & Rawlings, 2000; Grimes, Rawlings, & Willott, 2004) and radio-quiet (Steffen et al., 2003; Ueda et al., 2003; Hasinger, 2004; Sazonov & Revnivtsev, 2004; Barger et al., 2005; Simpson, 2005) quasars, however, have shown that the broad-line fraction increases with luminosity, with broad-line objects representing a large fraction of all AGN at luminosities above the โbreakโ in the luminosity function and rapidly falling off at luminosities below the break. Modifications to the standard torus unification model explain this via a luminosity-dependent inner torus radius (Lawrence, 1991), but this represents a tunable modification to a purely phenomenological model. Furthermore, as the observations have improved, it has become clear that even these luminosity-dependent torus models cannot produce acceptable fits to the broad line fraction as a function of luminosity (e.g., Simpson, 2005). However, we have shown above that the obscuring column, even at a given luminosity, is an evolutionary effect, dominated by different stages of gas inflow in different merging systems giving rise to varying typical column densities, rather than a single static structure. It is of interest, then, to calculate when quasars will be observed as broad-line objects, and to compare this with observations of broad line quasars and their population as a function of luminosity.
Figure 13 shows the B-band luminosity as a function of time for both the quasars and host galaxies in three representative simulations: the A2, A3, and A5 cases described in detail in ยง 2.1. These simulations each have $`f_{\mathrm{gas}}=1.0,q_{\mathrm{EOS}}=1.0,z_{\mathrm{gal}}=0`$, with virial velocities $`V_{\mathrm{vir}}=113,\mathrm{\hspace{0.17em}160}\mathrm{and}320\mathrm{km}\mathrm{s}^1`$, with resulting final black hole masses $`M_{\mathrm{BH}}^f=3\times 10^7,3\times 10^8,\mathrm{and}2\times 10^9M_{\mathrm{}}`$, respectively. The thick line in each case shows the quasar B-band luminosity, and the thin line shows the integrated B-band luminosity of all stars in the galaxy. New stars are formed self-consistently in the simulations according to the ISM gas properties, equation of state and star formation model described in Springel & Hernquist (2003), with the age and metallicity taken from the local star-forming ISM gas, which is enriched by supernova feedback from previous star formation. We then use the stellar population synthesis model of Bruzual & Charlot (2003) to determine the B-band luminosity (the B-band mass-to-light ratio) of new stars based on the stellar age and metallicity. The dotted line shows the result neglecting bulge particles, which must be initialized at the beginning of the simulation with random or uniform ages and metallicities instead of those quantities being determined self-consistently from the simulation physics. The right panels plot the intrinsic values of these quantities, and the left panels plot the median observed values of these quantities, where we have used our method for determining column densities and dust attenuation (ยง 2.2) to every star and bulge particle for each line of sight.
Unfortunately, the host galaxy luminosity does not scale with instantaneous and peak quasar luminosity as do, for example, the quasar lifetime and obscuration. Rather, there are important systematic dependencies, the largest of which is the dependence on host galaxy gas fraction. If the host galaxies are more massive, more concentrated, or have a weaker ISM equation of state pressurization, then they will more effectively drive gas into the central regions and maintain high gas densities for longer periods of time, as the deeper potential well or lack of gas pressure requires more heat input from the quasar before the gas can be expelled. These conditions will generally produce a quasar with a larger peak luminosity (final black hole mass), but also form more new stars, meaning that the B-band relation between host and quasar luminosity is roughly preserved.
However, the the black hole consumes only a small fraction of the available gas (comparison of e.g. the stellar mass and black hole mass suggests the black hole consumes $`0.1\%`$ of the gas mass), and so, at least above some threshold $`f_{\mathrm{gas}}0.1`$, the quasar peak luminosity does not significantly depend on the galaxy gas fraction (see, e.g. Figure 2 of Robertson et al. 2005b). But, the mass of new stars formed during the merger does strongly depend on the available gas. For example, simulations which are otherwise identical but have initial $`f_{\mathrm{gas}}=0.2,0.4,0.8,1.0`$ (i.e. an increasing fraction of the initial disk mass in gas instead of stars) produce similar peak quasar luminosity and final total stellar mass (within $`30\%`$ of one another), reflecting the conversion of most gas into stars and the fact that the peak quasar luminosity is determined more by the depth of the potential well than the total available gas supply. But, the mass of new stars formed in a merger scales roughly as $`M_{,\mathrm{new}}f_{\mathrm{gas}}`$ (as it must if the initial gas fraction does not change the final total stellar mass), and since young stellar populations dominate the observed B-band luminosity (especially during the peak merger and starburst phases associated with the bright quasar phase of interest), this implies roughly that $`L_Bf_{\mathrm{gas}}`$.
We demonstrate this explicitly in Figure 13, where we show in each panel the host galaxy and stellar B-band light curves for otherwise identical simulations with different gas fractions, $`f_{\mathrm{gas}}=0.2(\mathrm{red}),0.4(\mathrm{blue}),\mathrm{and}1.0(\mathrm{black})`$. In each of these cases, the quasar light curve is nearly identical (we show only the $`f_{\mathrm{gas}}=1.0`$ quasar lightcurve, for clarity, but the others are within $`30\%`$ of the curve shown at most times, with no systematic offset).
In order for a quasar to be classified as a โbroad-lineโ object, the optical spectrum must be visible and identified as such in the observed sample. This is clearly related to the ratio of quasar to host galaxy luminosity, but the threshold for classification is not obvious. In an X-ray or IR-selected sample, optical follow-up should be able to disentangle host galaxy light and identify quasar broad-line spectra with fluxes a factor of several fainter than the host. However, automated optical selection based on color or morphological criteria might well exclude objects unless the quasar luminosity is a factor of several greater than that of the host galaxy. Therefore, there is significant systematic uncertainty in the theoretical definition of a broad-line quasar. To first order, based on the above arguments, we can classify โbroad-line quasarsโ as objects in which the quasar optical luminosity is larger than some multiple $`f_{\mathrm{BL}}`$ of the host galaxy optical luminosity. Because the relevant ratio is different depending on the survey and selection techniques, we consider the range $`f_{\mathrm{BL}}=0.33`$, with a rough median $`f_{\mathrm{BL}}=1`$. Furthermore, because our simulations do not allow us to model the broad-line regions of the quasar or spectral line structures as influenced by e.g. reddening and dust absorption, we adopt the B-band luminosity of the quasar and host galaxy as a proxy for optical luminosity and more complex (but often quite sample-specific) color and morphological selection criteria.
In Figure 13, the B-band host galaxy luminosity is quite flat as a function of time, relative to the quasar B-band luminosity, and is roughly given by $`L_B^{\mathrm{gal}}/L_{\mathrm{}}M_{,\mathrm{new}}/M_{\mathrm{}}`$, where $`M_{,\mathrm{new}}`$ is the mass of new stars formed in the merger. As noted above, this scales approximately linearly with initial gas fraction at fixed final total stellar mass $`M_{}`$, giving $`L_B^{\mathrm{gal}}/L_{\mathrm{}}c_{\mathrm{gal}}(M_{}/M_{\mathrm{}})f_{\mathrm{gas}}`$, where $`c_{\mathrm{gal}}`$ is a correction of order unity which we can fit from the simulations (essentially a mean mass-to-light ratio for the newly formed stars). The bolometric correction of the quasar is usually defined by $`L_{\mathrm{bol}}^{\mathrm{qso}}=c_BL_B^{\mathrm{qso}}`$, and the quasar peak luminosity is $`L_{\mathrm{peak}}=c_LL_{\mathrm{Edd}}(M_{\mathrm{BH}}^f)`$, where again $`c_L`$ is a correction factor of order unity which we can calculate from our form for the quasar lifetime (see Equation 10) or measure in the simulations.
If we require that the quasar B-band luminosity be larger than a factor $`f_{\mathrm{BL}}`$ of the host galaxy B-band luminosity, we obtain
$$L_{\mathrm{bol}}^{\mathrm{qso}}/L_{\mathrm{}}>f_{\mathrm{BL}}c_Bc_{\mathrm{gal}}(M_{}/M_{\mathrm{}})f_{\mathrm{gas}}.$$
(23)
Dividing this through by $`L_{\mathrm{peak}}`$, we have
$$\frac{L_{\mathrm{bol}}^{\mathrm{qso}}}{L_{\mathrm{peak}}}0.4f_{\mathrm{gas}}f_{\mathrm{BL}}\left(\frac{c_{\mathrm{gal}}}{1.0}\right)\left(\frac{c_B}{12.0}\right)\left(\frac{M_{\mathrm{BH}}^f/M_{}}{0.001}\right)^1\left(\frac{c_L}{1.24}\right)^1.$$
(24)
We can test this scaling relation against the results of our simulations, and do so in Figure 14. Rearranging the equations above gives
$`{\displaystyle \frac{L_B^{\mathrm{qso}}}{L_B^{\mathrm{gal}}}}`$ $``$ $`3.4f_{\mathrm{gas}}^1{\displaystyle \frac{L_{\mathrm{bol}}^{\mathrm{qso}}}{L_{\mathrm{peak}}}}`$ (25)
$`\times \left({\displaystyle \frac{c_{\mathrm{gal}}}{1.0}}\right)^1\left({\displaystyle \frac{c_B}{12.0}}\right)^1\left({\displaystyle \frac{M_{\mathrm{BH}}^f/M_{}}{0.001}}\right)\left({\displaystyle \frac{c_L}{1.24}}\right),`$
which we can compare to our direct calculation of $`L_B^{\mathrm{qso}}/L_B^{\mathrm{gal}}`$ and $`L_{\mathrm{bol}}^{\mathrm{qso}}/L_{\mathrm{peak}}`$ for each simulation snapshot.
Ultimately, we are not interested so much in the intrinsic B-band luminosity of the quasar and host galaxy, but rather the observed luminosities; i.e. we are interested in the ratio $`L_{B,\mathrm{obs}}^{\mathrm{qso}}/L_{B,\mathrm{obs}}^{\mathrm{gal}}=(L_B^{\mathrm{qso}}/L_B^{\mathrm{gal}})(\mathrm{exp}\{(\tau _Q\tau _G)\})`$, where $`\tau _Q`$ and $`\tau _G`$ are โeffectiveโ optical depths which we use to denote the mean attenuation of quasar and host galaxy B-band luminosities, respectively. We have considered the distribution of column densities attenuating the quasar as a function of instantaneous and peak quasar luminosity in detail in ยง 2.3 above; the attenuation of the host galaxy as a function of luminosity, observed band, halo mass, and star formation rate are discussed in detail in Jonsson et al. (2005). Combining these fits gives, roughly, $`(\mathrm{exp}\{(\tau _Q\tau _G)\})(M_{\mathrm{BH}}^f/10^8M_{\mathrm{}})^{0.16}`$, but a better approximation can be determined directly from the simulations.
This scaling can be understood roughly using toy models of uniformly mixed luminous sources within the galaxy described by Jonsson et al. (2005), after accounting for the fact that the luminosity (star formation rate) dependent portion of the attenuation scales with luminosity in a similar manner to our quasar attenuation (compare our $`\tau _QN_\mathrm{H}L_{\mathrm{qso}}^{0.430.54}`$ to their $`\tau _GL_{\mathrm{B},\mathrm{gal}}^{0.55}`$). The key consequence of this is that more massive systems (higher bulge and black hole masses) have their host galaxy light proportionally more attenuated in mergers, meaning that (as suggested by the comparison of light curves in Figure 13) the quasar is more likely to be observed with an optical luminosity larger than that of its host.
Figure 14 plots the ratio of the observed (attenuated) B-band quasar luminosity to the observed host galaxy B-band luminosity as a function of the ratio of instantaneous to peak quasar bolometric luminosity. We show the results for four different gas fractions $`f_{\mathrm{gas}}=0.2,0.4,0.8,1.0`$ as labeled. For each gas fraction, we consider our simulations A2 (black diamonds), A3 (blue circles), and A5 (red $`\times `$โs) (the same simulations shown in Figure 13) with $`q_{\mathrm{EOS}}=1.0,z_{\mathrm{gal}}=0`$, and virial velocities $`V_{\mathrm{vir}}=113,\mathrm{\hspace{0.17em}160}\mathrm{and}320\mathrm{km}\mathrm{s}^1`$, using the labeled initial gas fraction. The colored lines in each panel show the predictions of combining the scalings expected for the intrinsic luminosities (Equation 25) and attenuations as above, giving
$$\frac{L_{\mathrm{B},\mathrm{obs}}^{\mathrm{qso}}}{L_{\mathrm{B},\mathrm{obs}}^{\mathrm{gal}}}=7.9\frac{1}{f_{\mathrm{gas}}}\left(\frac{M_{\mathrm{BH}}^f}{10^8M_{\mathrm{}}}\right)^{0.2}\frac{L}{L_{\mathrm{peak}}},$$
(26)
where the colored lines each use the $`M_{\mathrm{BH}}^f`$ and $`f_{\mathrm{gas}}`$ of the simulation of the corresponding color and panel. This scaling provides a good estimate of the observed optical quasar-to-galaxy luminosity ratio, including the complicated effects of attenuation, evolving mass-to-light ratios, metallicities, and host galaxy properties, as a function of gas fraction, final black hole mass, and the ratio of the current to peak quasar luminosity. Although, for clarity, we have not shown a range of simulations varying other parameters, we find that this scaling is robust to the large number of quantities we have considered in our simulations โ there are systematic offsets in e.g. $`L_{\mathrm{peak}}`$ and $`M_{\mathrm{BH}}^f`$ with changes such as e.g. different ISM equations of state, but the scaling in terms of $`L_{\mathrm{peak}}`$ and $`M_{\mathrm{BH}}^f`$ is unchanged.
Because the ratio of observed quasar and host galaxy B-band luminosities in our simulations obeys the scaling of Equation 26, we can use it to predict the properties of โbroad-lineโ quasars, defined by $`L_{\mathrm{B},\mathrm{obs}}^{\mathrm{qso}}>f_{\mathrm{BL}}L_{\mathrm{B},\mathrm{obs}}^{\mathrm{gal}}`$. To do so, however, we must assume a typical host galaxy gas fraction. Unfortunately, because our empirical modeling in terms of the quasar lifetime as a function of $`L`$ and $`L_{\mathrm{peak}}`$ does not have a systematic dependence on host galaxy gas fraction (see ยง 2.4), we have no constraint on this parameter. It is, however, convenient for several reasons to consider $`f_{\mathrm{gas}}=0.3`$ as a typical value for bright quasars.
First, such a gas fraction is capable of yielding the brightest observed quasars; second, scaling a Milky-Way like disk with the observed $`z=0`$ gas fraction $`0.1`$ to the redshifts of peak quasar activity gives a similar gas fraction (e.g., Springel et al., 2005a); third, gas fractions $`30\%`$ in major mergers are needed to explain the observed fundamental plane (Robertson et al. 2005c, in preparation), kinematic properties (Cox et al. 2005c, in preparation), and central phase space densities (Hernquist, Spergel & Heyl 1993) of elliptical galaxies; fourth, this choice implies that the brightest quasars with $`M_{\mathrm{BH}}^f10^{10}M_{\mathrm{}}`$ attain observed B-band luminosities $`1000`$ times that of their hosts at their peaks, as is observed (e.g., McLure & Dunlop, 2004). Finally, and most important, the assumed $`f_{\mathrm{gas}}`$ and $`f_{\mathrm{BL}}`$ are degenerate in our predictions for the broad-line population, as they both enter linearly in the ratio of host galaxy to quasar B-band luminosity. Therefore, the range of $`f_{\mathrm{BL}}=0.33`$ which we consider (for a fixed median $`f_{\mathrm{gas}}=0.3`$) can be equivalently considered, for a fixed median $`f_{\mathrm{BL}}=1`$, to represent a theoretical uncertainty in the host galaxy gas fraction, $`f_{\mathrm{gas}}=0.10.9`$; i.e. spanning the range from present, relatively gas-poor Milky-Way like disks to almost completely gaseous disks. This, then, gives for our โbroad-lineโ criterion,
$$\frac{L}{L_{\mathrm{peak}}}0.2\left(\frac{f_{\mathrm{BL}}}{1.0}\right)\left(\frac{f_{\mathrm{gas}}}{0.3}\right)\left(\frac{M_{\mathrm{BH}}^f}{10^7M_{\mathrm{}}}\right)^{0.2}.$$
(27)
The โbroad-lineโ phase is thus, as is clear from Figure 13 and implicit in our definition of the broad-line phase, closely associated with the final โblowoutโ stages of quasar evolution, when the mass of the quasar reaches that corresponding to its location on the $`M_{\mathrm{BH}}\sigma `$ relation and gas is expelled from the central regions of the galaxy, shutting down accretion (Di Matteo et al., 2005). We note that combining the equation above with our fitted quasar lifetimes gives an integrated time when the quasar would be observable as a broad line object of $`t_{\mathrm{BL}}1020`$ Myr, in good agreement with the optically observable bright quasar lifetimes we calculate directly from our quasar light curves, including the effects of attenuation, and with empirical estimates of the quasar lifetime which are based directly on optically-selected, broad-line quasar samples.
The $`(M_{\mathrm{BH}}^f/10^7M_{\mathrm{}})^{0.2}`$ term in the above equation reflects the fact that, below a certain peak luminosity, quasars are less likely to reach luminosities above that of the host galaxy, as can be seen in the uppermost panels of Figure 13 for a final black hole mass of $`M_{\mathrm{BH}}^f=3\times 10^7`$ โ i.e. the smallest AGN are proportionally less optically luminous than their hosts. This does not imply that such systems are not inherently broad-line objects, but only that the host galaxy light will increasingly dominate at lower luminosities. We also caution against extrapolating this to large or small $`M_{\mathrm{BH}}^f`$, as the attenuation becomes more difficult to predict at these peak luminosities, and the linear formula above is not always accurate (see Figure 14).
We can use this estimate of the broad-line phase and our model of the quasar lifetime to calculate the total energy radiated in this bright, optically observable stage following the calculation of ยง 2.4, but with a minimum luminosity determined by Equation 27. This gives an integrated fraction $`0.30.4`$ ($`\mathrm{exp}\{0.2f_{\mathrm{BL}}(f_{\mathrm{gas}}/0.3)/\alpha _L\}`$) of the total radiant energy emitted during the broad-line phase. Thus, despite the short duration of this optical quasar stage, a large fraction of the total radiated energy is emitted (as it represents the final $`e`$-folding in the growth of the black hole) when most of the final black hole mass (ยง 2.4) is accumulated. Accounting for the luminosity dependence of our bolometric corrections (with the optical fraction of the quasar energy increasing with bolometric luminosity) as well as the small fraction of objects observable at lower luminosities (with larger typical obscuring column densities) increases this fraction to as much as $`0.60.7`$ for bright quasars. Therefore, despite the fact that the duration of the optically observable broad-line quasar phase may be $`1/10`$ that of the obscured quasar growth phase, the changing quasar luminosity over this period and non-trivial quasar lifetime as a function of luminosity implies only small corrections to counting arguments such as that of Soltan (1982), which rely on the total observed optical quasar flux density to estimate the relic supermassive black hole density.
### 4.2. The Broad-Line Fraction as a Function of Luminosity
By estimating the time that a quasar with some $`L_{\mathrm{peak}}`$ will be observable as a broad-line quasar at a given luminosity, we can then calculate the broad-line quasar luminosity function in the same fashion as the complete quasar luminosity function in ยง 3.2. Instead of the full quasar lifetime $`\mathrm{d}t/\mathrm{d}\mathrm{log}L`$, we consider only the time during which broad-lines would be observed (i.e. that the quasar spectrum would be recognized as opposed to the host galaxy spectrum), as identified in our simulations (ยง 4.1).
For a sample selected in hard X-rays (i.e. the selection function only being relevant at column densities $`10^{24}\mathrm{cm}^2`$), we show the resulting โbroad-lineโ luminosity function in Figure 9 (cyan dot-dashed lines), and compare it to the broad-line quasar luminosity function identified in the hard X-ray luminosity function of Barger et al. (2005). The agreement is good at all luminosities, and our model explains both the fact that broad-line quasars dominate the luminosity function at luminosities well above the โbreakโ in the luminosity function, and the downturn in the broad-line quasar population at luminosities below the peak. Essentially, the broad-line quasar population more closely traces the shape of the $`\dot{n}(L_{\mathrm{peak}})`$ distribution, giving rise to the observed behavior as a dual consequence of luminosity-dependent quasar lifetimes and the evolutionary nature of quasar obscuration in our simulations.
Figure 15 compares our theoretical predictions to the 2dF-SDSS (2SLAQ) g-band luminosity function of broad-line quasars from Richards et al. (2005) (black squares), as well as the B-band luminosity function from Croom et al. (2004) (green circles), at several redshifts from $`z0.32`$, over which range the surveys are expected to be relatively complete (for broad-line quasars). The 2dF-SDSS result is the most recent determination of the broad-line luminosity function, but compares well with previous determinations by, e.g., Boyle et al. (1988), Koo & Kron (1988), Marano, Zamorani, & Zitelli (1988), Boyle et al. (1990), Boyle, Jones, & Shanks (1991), Zitelli et al. (1992), Boyle et al. (2000), and Croom et al. (2004). Open squares correspond to bins in luminosity which have been corrected for incompleteness following Page & Carrera (2000), but this correction is uncertain as the bins are not uniformly sampled. We compare this at each redshift to the prediction of our determination of the quasar โbroad-lineโ phase, where we estimate that the quasar is observable as a broad line object when its observed B-band luminosity is greater than a factor $`f_{\mathrm{BL}}=1`$ of that of the host galaxy. We calculate this for both the minimum and maximum observed redshift of each bin to show the range owing to evolution of the luminosity function over each interval in redshift. The systematic uncertainty in our prediction can be estimated from the dotted lines, which show the prediction (at the mean redshift of the bin) if we instead require the observed quasar B-band luminosity to be above a factor of 0.3 (upper lines) or 3 (lower lines) of the observed host galaxy B-band luminosity, which as discussed in ยง 4.1 can alternatively be considered an uncertainty in host galaxy gas fraction, with $`f_{\mathrm{gas}}=0.1`$ and $`f_{\mathrm{gas}}=0.9`$, respectively.
The agreement at all luminosities and redshifts shown is encouraging, given the simplicity of our determination of the broad-line phase from the simulations, but the systematic uncertainties are large, emphasizing the importance of calculating detailed selection effects in contrasting e.g. โbroad-lineโ samples from optical and X-ray surveys, as opposed to assuming a constant obscured fraction at a given luminosity based on the ratio of luminosity functions as has been adopted in previous phenomenological models. The difference between different choices of $`f_{\mathrm{gas}}`$ is suppressed at the high luminosity (and correspondingly high redshift) end of the luminosity function, because the quasar-to-galaxy B-band luminosity ratio scales as $`(M_{\mathrm{BH}}^f)^{0.2}`$; i.e. regardless of the choice of $`f_{\mathrm{BL}}`$, quasars increasingly overwhelm their host galaxy in large systems near their peak luminosity. However, at low luminosity, the predictions rapidly diverge, implying that a measurement of the faint end of the broad-line quasar luminosity function, with a reliable calibration of $`f_{\mathrm{BL}}`$, can constrain the typical gas fractions of quasar host galaxies and the evolution of these gas fractions with redshift.
By dividing out the predicted luminosity function $`\varphi _{HX}`$, we can estimate the fraction of โbroad lineโ objects observed in reasonably complete X-ray samples as a function of luminosity. This is shown in Figure 16, where for ease of comparison we have shown the broad-line fraction as a function of hard X-ray (2-10 keV) luminosity. Our prediction, based on determining the time a quasar with a given luminosity $`L`$ and peak luminosity $`L_{\mathrm{peak}}`$ in our simulations will be observable with a B-band luminosity greater than a fraction $`f_{\mathrm{BL}}=1.0`$ of the host galaxy observed B-band luminosity, is shown as the thick black line. This is compared to the observations of Ueda et al. (2003) (squares), Hasinger (2004) (circles), Grimes, Rawlings, & Willott (2004) (triangles), and Simpson (2005) (diamonds). The data from Hasinger (2004) has been scaled from soft X-ray (0.5-2 keV) using our bolometric corrections, and the data from Grimes, Rawlings, & Willott (2004) and Simpson (2005) have been converted from \[O iii\] luminosity as in Simpson (2005) using the mean correction for Seyfert galaxies (Mulchaey et al., 1994), $`L_{[\mathrm{O}\mathrm{III}]}=0.015\times L_{210\mathrm{keV}}`$.
We also plot as upper and lower dashed lines the results of changing $`f_{\mathrm{BL}}`$, the fraction of the host galaxy B-band luminosity above which the quasar B-band luminosity must be observed for identification as a โbroad-lineโ object, considering $`f_{\mathrm{BL}}=0.3,\mathrm{and}3`$, respectively. We determine this for the low-redshift $`z0.3`$ quasar distribution, from which most of the data are drawn. The red dot-dashed line shows the difference at high redshift, if just $`z1`$ quasars are considered (for $`f_{\mathrm{BL}}=1`$). The broad-line fraction is systematically lower, primarily because the break luminosity in the luminosity function moves to higher luminosity with redshift, meaning that at a fixed luminosity below the break, a smaller fraction of observed objects are at $`LL_{\mathrm{peak}}`$ in the โblowoutโ phase of peak optical quasar luminosity. Finally, the dotted line shows the results assuming a โlight bulbโ model for the broad-line phase (but still using our $`\dot{n}(L_{\mathrm{peak}})`$ distribution, otherwise this translates to a constant obscured fraction with luminosity) lifetimes, with a fixed broad-line lifetime of $`t_Q=20`$Myr.
The prediction of the most basic torus model, with constant broad-line fraction $`0.36`$, is ruled out to high significance ($`\chi ^2/\nu =18.5,17.2`$ if we consider all data points, or if we consider only the most well-constrained data, from Simpson , respectively). Furthermore, the solid cyan line shows the best-fit luminosity-dependent torus model, in which the broad line fraction is given by (e.g., Simpson, 1998; Grimes, Rawlings, & Willott, 2004)
$$f=11/\sqrt{1+3L/L_0},$$
(28)
where $`L_0`$ is the luminosity where the number of broad line objects is equal to the number of non-broad line objects. This fit is at best marginally acceptable over a narrow range in luminosities ($`\chi ^2/\nu =14.0,7.3`$). Modified luminosity-dependent, receding torus models have been proposed which give a better fit to the data by, for example, allowing the torus height to vary with luminosity (e.g., Simpson, 2005), but there is no physical motivation for these changes, and they introduce such variation through additional free parameters that allow a curve of essentially arbitrary slope to be fitted to the data.
However, the prediction of our model agrees reasonably well ($`\chi ^2/\nu =4.0,1.2`$) with the observations over the entire range covered, a span of six orders of magnitude in luminosity. We emphasize that our prediction, which matches the data better than standard torus models that are actually fitted to the data, is not a fit to the observations. Instead, it is derived from the physics of our simulations, including black hole accretion and feedback which are critical in driving the โblowoutโ phase which constitutes most of the time a quasar is visible as a โbroad-lineโ object by our estimation, and from the $`\dot{n}(L_{\mathrm{peak}})`$ distribution implied by our model of quasar lifetimes and the bolometric quasar luminosity function. The agreement suggests that our choice of the parameter combination $`f_{\mathrm{BL}}f_{\mathrm{gas}}=0.3`$ is a good approximation. As noted above, this implies that calibrating $`f_{\mathrm{BL}}`$ for an observed sample, combined with the mean broad-line fraction and our modeling, can provide a constraint (albeit model-dependent) on the host galaxy gas fraction of quasars at a given redshift, which cannot necessarily be directly measured even with difficult, detailed host galaxy probes, as gas is rapidly converted into stars throughout the merger. The uncertainty plotted, while large, actually represents a larger theoretical uncertainty โ as discussed above, if an observational sample were well-defined such that it were complete to broad-line objects with observed optical luminosity above a fraction $`f_{\mathrm{BL}}`$ of the host galaxy luminosity, the range we consider would correspond to a range $`f_{\mathrm{gas}}=0.10.9`$ in the quasar host galaxy gas fraction, which the observations could then constrain.
In our modeling, the broad line fraction as a function of luminosity does not depend sensitively on the observed luminosity function, as evidenced by the relatively similar prediction at high redshift. The evolution we do predict with redshift, in fact, agrees well with that found by Barger et al. (2005) over the redshift range $`z=0.11.2`$ (see also La Franca et al. 2005), an aspect of the observations which is not reproduced in any static or luminosity-dependent torus model but follows from the evolution of the quasar luminosity function in our picture for quasar growth. However, we do caution that gas fractions may systematically evolve with redshift, and as discussed above, a higher gas fraction will give generally shorter โbroad-lineโ lifetimes using our criteria of quasar optical luminosity being higher than some fraction of the host galaxy luminosity, which will also contribute to the evolution in the mean โbroad-lineโ fraction with redshift. Finally, neglecting the role of luminosity-dependent quasar lifetimes gives unacceptable fits to the data ($`\chi ^2/\nu =66.0,77.5`$), as the broad-line fraction as a function of luminosity is a consequence of both the evolution of obscuration and the dependence of lifetime on luminosity.
Our model for quasar evolution provides a direct physical motivation for the change in broad line fraction with luminosity and suggests that it is not a complicated selection effect. As an observational sample considers higher luminosities (i.e. approaches and passes the โbreakโ in the observed luminosity function), a comparison of the luminosity function and the underlying $`\dot{n}(L_{\mathrm{peak}})`$ shows that it is increasingly dominated by sources near their peak luminosity in the final stages of Eddington limited growth. The final stages of this growth expel the large gas densities obscuring the quasar, rendering it a bright, optically observable broad-line object for a short time. Therefore, we expect that the fraction of broad-line objects should increase with luminosity in quasar samples, as indicated by the observations.
Many observational measures do not consider a direct optical analysis of the quasar spectrum in estimating the fraction of broad-line objects as a function of luminosity. For example, Ueda et al. (2003) adopt a proxy, classifying as โobscuredโ any quasars with an X-ray identified column density $`N_\mathrm{H}>10^{22}\mathrm{cm}^2`$, and as โunobscuredโ quasars below this column density. We can compare to their observations, using the column density distributions as a function of luminosity from our simulations, which cover the entire range in luminosity of the observed sample. Specifically, we use a Monte Carlo realization of these distributions, employing our fitted $`\dot{n}(L_{\mathrm{peak}})`$ distribution at each redshift to produce a list of quasar peak luminosities and then generating all other properties based on the probability distribution of a given property in simulations with a similar peak luminosity. We describe this methodology in detail in ยง 8, and provide several such mock quasar distributions at different redshifts.
In Figure 17, we compare our estimated โobscuredโ and โunobscuredโ fractions as a function of hard X-ray luminosity, using the same definitions as well as redshift and luminosity limits as the observed sample. The solid line shows our prediction, with vertical error bars representing Poisson errors, where the number of โcountsโ is proportional to the total time spent by simulations at the plotted luminosity (the absolute value of these errors should not be taken seriously). The โobscuredโ fraction is determined in bins of luminosity $`\mathrm{\Delta }\mathrm{log}L_{210\mathrm{keV}}=0.5`$. Despite our large number of simulations, there is still some artificial โnoiseโ owing to incomplete coverage of the merger parameter space, namely the apparent oscillations in the obscured fraction. However, the mean trend agrees well with that observed (black squares), suggesting that the success of our modeling in reproducing the fraction of โbroad lineโ objects as a function of luminosity is not a consequence of the definitions chosen above. We do not show the predictions of the standard and luminosity-dependent torus models, as (because essentially any line of sight through the torus encounters a column density $`N_\mathrm{H}>10^{22}\mathrm{cm}^2`$) the predictions of these models are identical to those shown and compared to the same observations in Figure 16.
Our prediction that the fraction of broad-line objects should rise with increasing luminosity is counterintuitive, given our fitted column density distributions in which typical (median) column densities increase with increasing luminosity. This primarily owes to the simplicity of our $`N_\mathrm{H}`$ fits; we assume the distribution is lognormal at all times, but a detailed inspection of the cumulative (time-integrated) column density distribution shows that at bright (near-peak) luminosities, the distribution is in fact bimodal (see e.g. Figure 3 of Hopkins et al. 2005b and Figure 2 of Hopkins et al. 2005d), representing both the heavily obscured growth phase and the โblowoutโ phase we have identified here as the โbroad lineโ phase. Over most of a simulation, we find the general trend shown in Figure 3 and discussed above, namely that typical column densities increase with intrinsic (unobscured) luminosity. This is because the total time at moderate to large luminosities is dominated by black holes growing in the obscured/starburst stages; here, the same gas inflows fueling black hole growth also give rise to large column densities and starbursts which obscure the black hole activity. However, when the quasar nears its final, peak luminosity, there is a rapid โblowoutโ phase as feedback from the growing accretion heats the surrounding gas, driving a strong wind and eventually terminating rapid accretion, leaving a remnant with a black hole satisfying the $`M_{\mathrm{BH}}\sigma `$ relation. This can be identified with the traditional bright optical quasar phase, as the final stage of black hole growth with a rapidly declining density (allowing the quasar to be observed in optical samples), giving typical luminosities, column densities, and lifetimes of optical quasars. In these stages, larger luminosities imply more violent โblowoutโ events, i.e. a brighter peak luminosity quasar more effectively expels the nearby gas and dust, rendering a dramatic decrease in column density at these bright stages (see Hopkins et al. 2005f).
We are essentially modeling this bimodality in more detail by directly determining the โbroad-lineโ phase from our simulations. However, the broad line fraction-luminosity relation we predict is also a consequence of the more complicated relationship between column density, peak luminosity, and bolometric and observed luminosity, as opposed to the predictions from a model with correlation between $`N_\mathrm{H}`$ and only observed luminosity. The key point is that we find, near the peak luminosity of the quasar, as feedback drives away gas and slows down accretion, the typical column densities fall rapidly with luminosity in a manner similar to that observed. In our model for the luminosity function, quasars below the observed โbreakโ are either accreting efficiently in early stages of growth or are in sub-Eddington phases coming into or out of their peak quasar activity. Around and above the break, the luminosity function becomes dominated by sources at high Eddington ratio at or near their peak luminosities. Based on the above calculation, we then expect what is observed, that in this range of luminosities, the fraction of objects observed with large column densities will rapidly decrease with luminosity as the observed sample is increasingly dominated by sources at their peak luminosities in this blowout phase. This also further emphasizes that the evolution of quasars dominates over static geometrical effects in determining the observed column density distribution at any given luminosity.
Finally, if host galaxy contamination were not a factor, we would expect from our column density model that, at low luminosities ($`L10^{10}L_{\mathrm{}}`$, well below the range of most observations shown), the broad-line fraction would again increase (i.e. the obscured fraction would decrease), as the lack of gas to power significant accretion would also imply a lack of gas to produce obscuring columns. However, at these luminosities, typical of faint Seyfert galaxies or LINERs, our modeling becomes uncertain; it is quite possible, as discussed previously, that cold gas remaining in relaxed systems could collapse to form a traditional dense molecular torus on scales $``$pc, well below our resolution limits. Furthermore, host galaxy light is likely to overwhelm any AGN broad-line contribution, and selection effects will also become significant at these luminosities.
### 4.3. The Distribution of Active Broad-Line Quasar Masses
Our determination of the โbroad-lineโ or optical phase in quasar evolution allows us to make a further prediction, namely the mass distribution of currently active broad-line quasars. At some redshift, the total number density of observed, currently active broad-line quasars with a given $`L_{\mathrm{peak}}`$ will be (in the absence of selection effects)
$$n_{\mathrm{BL}}(L_{\mathrm{peak}})\dot{n}(L_{\mathrm{peak}})t_{\mathrm{BL}}(L_{\mathrm{peak}}),$$
(29)
where $`t_{\mathrm{BL}}(L_{\mathrm{peak}})`$ is the total integrated time that a quasar with peak luminosity $`L_{\mathrm{peak}}`$ spends as a โbroad-lineโ object (using our criterion for the ratio of the observed quasar B-band luminosity to that of the host galaxy), given by integrating our formulae in ยง 4.1 or directly calculated from the simulations. Since we have determined roughly that a quasar should be observable as a โbroad-lineโ object at times with $`L0.2L_{\mathrm{peak}}`$ primarily just after it reaches its peak luminosity, in the โblowoutโ phase of its evolution, we expect the instantaneous black hole mass at the time of observation as a broad-line quasar to be, on average, $`M_{\mathrm{BH}}^{\mathrm{BL}}M_{\mathrm{BH}}^f(L_{\mathrm{peak}})`$, where $`M_{\mathrm{BH}}^fM_{\mathrm{Edd}}(L_{\mathrm{peak}})`$ modulo the order unity corrections described in ยง 2.4. Using our fitted $`\dot{n}(L_{\mathrm{peak}})`$ distribution from the luminosity function, extrapolated to low redshift ($`z0`$), and combining it with the integrated โbroad-lineโ lifetimes from our simulations as above, we obtain the differential number density of sources in a logarithmic interval in $`L_{\mathrm{peak}}`$. Finally, we use our Equation 10 for $`M_{\mathrm{BH}}^f(L_{\mathrm{peak}})`$ determined from our fitted quasar lifetimes (demanding that $`E_{\mathrm{rad}}=ฯต_rM_{\mathrm{BH}}^fc^2`$) to convert this to a distribution in black hole mass.
Our predicted $`n(M_{\mathrm{BH}})`$, i.e. the number of observed active quasars at low redshift in a logarithmic interval of black hole mass, is shown in Figure 18. We consider the complete distribution of active quasar masses, for both broad-line and non broad-line objects, in the left panel of the figure, and the distribution of broad-line objects only, $`n(M_{\mathrm{BH}}^{\mathrm{BL}})`$, in the right panel. On the left, we show the complete distribution which would be observed without any observational limits (dashed line). We calculate this from the distributions of Eddington ratios in our simulations, as a function of current and peak luminosity, and our fit to $`\dot{n}(L_{\mathrm{peak}})`$ (as, e.g. for our Monte Carlo realizations). We also consider the observed distribution if we apply the luminosity limit for completeness from the SDSS sample of Heckman et al. (2004) (dotted), $`L_{[\mathrm{O}\mathrm{III}]}>10^6L_{\mathrm{}}`$, which using their bolometric corrections yields $`L>3.5\times 10^9L_{\mathrm{}}`$, and then additionally applying their minimum velocity dispersion $`\sigma >70\mathrm{km}\mathrm{s}^1`$ (dot-dashed). Finally, we can weight this distribution by luminosity (solid line) to compare directly to that determined in their Fig. 1. The red points are taken from the luminosity-weighted black hole mass function of Heckman et al. (2004), which serves as a rough estimate of the active black hole mass distribution given their selection effects. Vertical error bars represent the range in parameterizations of the mass function from Heckman et al. (2004), including whether or not star formation is corrected for and limiting the sample to luminosities $`L10^{10}L_{\mathrm{}}`$ or Eddington ratios $`>0.01`$. Horizontal errors represent an uncertainty of $`0.2`$dex in the black hole mass estimation (representative of uncertainties in the $`M_{\mathrm{BH}}\sigma `$ relation used). The agreement is good, especially given the significant effects of the selection criteria and luminosity-weighting.
We also consider the predictions of a โlight-bulbโ or โexponential / fixed Eddington ratioโ model of the quasar lifetime for the active black hole mass distribution (red lines). For purposes of the active black hole mass function, the two predictions are identical and independent of the assumed quasar lifetime (modulo the arbitrary normalization), as both assume that all observed quasars are accreting at a fixed Eddington ratio, giving the distribution of active black hole masses. The dashed line shows the prediction for the complete active black hole mass function, which rises sharply to lower luminosities, as it must given a luminosity function which increases monotonically to lower luminosities. The solid line shows the prediction of such a model with the complete set of selection effects from Heckman et al. (2004) described above applied, as with the solid black line showing the prediction of our modeling. Here, we chose the characteristic Eddington ratio $`1.0`$ by fitting the predicted curve to the Heckman et al. (2004) observations. Note that both the characteristic Eddington ratio and lifetime (normalization) of the curve are fitted, so the relative normalization of this curve and our full model prediction are not the same; for example, the predicted total absolute number of active $`M_{\mathrm{BH}}>10^9`$ quasars is higher in the full model than in the light-bulb or exponential models. Still, it is clear that these models produce too broad a distribution of active black hole masses, in disagreement with the observations. We could, of course, obtain an arbitrarily close agreement with the observations if we fit to the distribution of accretion rates, but such a model would recover a quasar lifetime and accretion rate distribution quite similar to ours, as is evident from the agreement between the predictions of our simulations and the observations. A purely empirical model of this type is considered by e.g. Merloni (2004), who finds that similar qualitative evolution in the quasar lifetime and anti-hierarchical black hole assembly to that predicted by our modeling is implied by the combination of quasar luminosity functions and the black hole mass function.
On the right of the figure, we show our predicted mass distribution for low-redshift, active โbroad-lineโ quasars (solid black lines), where we estimate that an object is a โbroad-lineโ quasar if the observed quasar B-band luminosity is above a factor $`f_{\mathrm{BL}}=1`$ of that of the host galaxy โ dotted and dashed lines show the result if $`f_{\mathrm{BL}}=0.3`$ or 3, respectively, parameterizing the range of different observed samples. As discussed above, the range of $`f_{\mathrm{BL}}`$ shown can be, alternatively, thought of as a parameterization of uncertainty in the host galaxy gas fraction, if (in an observed sample), the sensitivity to seeing quasar broad lines against host galaxy contamination is known. Therefore, the location of the peak in the active broad-line black hole mass function can be used, just as the mean broad line fraction vs. luminosity, as a test of the typical gas fractions of bright quasar host galaxies, and can constrain potential evolution in these gas fractions with redshift.
The prediction shown is testable, but appears to be in good agreement with preliminary results for the distribution of active broad-line black hole masses from the SDSS (e.g., McLure & Dunlop, 2004). The observations may show fewer low-mass black holes than we predict, but this is expected, as observed samples are likely incomplete at the low luminosities of these objects (even at the Eddington limit, a $`10^5M_{\mathrm{}}`$ black hole has magnitude $`M_g16`$). If, in our model, we were to consider instead a standard torus scenario for the definition of the broad-line phase, we would predict the same curve as that shown in the left half of the figure (black dashed; our prediction for the cumulative active black hole mass function). This is because the standard torus model predicts that a constant fraction of objects are broad-line quasars, regardless of mass or luminosity, thus giving identical distributions of Type I and Type II quasar masses. If we consider a luminosity-dependent or receding torus model, the prediction is nearly identical to the black line shown. This is because, as shown in Figure 16, our prediction for the broad line fraction as a function of luminosity is similar to that of the receding torus model. The differences in the model predictions for the broad-line fraction as a function of luminosity do manifest in the prediction for the active broad-line black hole mass function, but the difference in these models is smaller than the $`1\sigma `$ range from different values of $`f_{\mathrm{BL}}`$ shown. However, if we consider different models for the quasar light curve or lifetime, the predicted active broad-line mass function is quite different (as is the cumulative active black hole mass function).
We show the predictions of a light-bulb or exponential light curve model for quasar evolution in the figure, adopting either a standard torus model (red) or receding torus model (blue) to determine the broad-line fraction as a function of luminosity. For the standard torus model, this predicts that the broad line mass function should trace the observed luminosity function, rising monotonically to lower black hole masses in power-law fashion (just as seen in the red dashed line in the left half of the figure for the cumulative black hole mass function). For the receding torus model, the active black hole mass function shows a peak (because, at lower luminosities, there are more observed quasars, but a larger fraction of them are obscured). However, the location of this peak is at roughly an order of magnitude smaller black hole mass than for our prediction. This assumes a typical Eddington ratio $`1`$, which we have fitted to the cumulative black hole mass function โ the peak in the broad-line active black hole mass function in these models could be shifted to larger black hole masses by assuming a smaller typical Eddington ratio, but this would only worsen the agreement with the cumulative black hole mass function of Heckman et al. (2004). Furthermore, a robust difference between the models is that the light bulb or Eddington-limited/exponential models predict, for the standard torus case, no turnover in the active broad-line black hole mass function, and for the receding torus case, a broader distribution in active broad-line quasar black hole masses than is predicted in our modeling. Roughly, the lognormal width of this distribution in our model is $`0.6`$ dex, whereas the light-bulb or exponential light curve models have a distribution with width $`1.0`$ dex. As noted above, we obtain a similar prediction if we adopt our full obscuration model instead of the receding torus model here. A determination of the range of active, broad-line quasar masses can, therefore, constrain quasar lifetimes and light curves.
Our model makes an accurate prediction for the distribution of active black hole masses, even at $`z0`$ where our extrapolation of the luminosity function is uncertain. It is important to distinguish this from the predicted relic black hole mass distribution, derived in ยง 6, which must account for all quasars, i.e. $`\dot{n}(L_{\mathrm{peak}})`$ integrated over redshift. We additionally find for broad-line quasars, as we expect from our prediction of the broad-line phase, that these objects are primarily radiating at large Eddington ratios, $`l0.21`$, but we address this in more detail in ยง 5. The success of this prediction serves not only to support our model, but also implies that we can extrapolate to fairly low luminosities, even bright Seyfert systems at $`z0`$. This suggests that many of these systems, at least at the bright end, may be related to our assumed quasar evolution model, fueled by similar mechanisms and either exhibiting weak interactions among galaxies or relaxing from an earlier, brighter stage in their evolution. As we speculate in ยง 8, our description of self-regulated black hole growth may also be relevant to fainter Seyferts, even those that reside in apparently undisturbed galaxies.
## 5. The Distribution of Eddington Ratios
In traditional models of quasar lifetimes and light curves, the Eddington ratio, $`lL/L_{\mathrm{Edd}}`$ is generally assumed to be constant. Even complex models of the quasar population which allow for a wide range of Eddington ratios according to some probability distribution $`P(l)`$ implicitly associate a fixed Eddington ratio with each individual quasar, and do not allow for $`P(l)`$ to depend on instantaneous luminosity or host system properties. However, this is a misleading assumption in the context of our model, as the Eddington ratio varies in a complicated manner over most of the quasar light curve. Furthermore, the integrated time at a given Eddington ratio is different in different systems, with more massive, higher peak luminosity systems spending more time at large ($`l1`$) Eddington ratios.
The probability of being at a given Eddington ratio should properly be thought of as a conditional joint distribution $`P(l|L,L_{\mathrm{peak}})`$ in both instantaneous and peak luminosity, just as the quasar โlifetimeโ is more properly a conditional distribution $`t_Q(L|L_{\mathrm{peak}})`$. Rather than adopting a uniform Eddington ratio or Eddington ratio distribution, empirical estimates must consider more detailed formulations such as the framework presented in Steed & Weinberg (2003), which allows for a conditional bivariate Eddington ratio distribution and can therefore incorporate these physically motivated dependencies and complications in de-convolving observations of the quasar luminosity function to determine e.g. Eddington ratio distributions, active black hole mass functions, and other physical quantities.
Figure 19 shows the distribution of Eddington ratios as a function of luminosity for the fiducial, Milky Way-like A3 simulation ($`V_{\mathrm{vir}}=160\mathrm{km}\mathrm{s}^1`$). Over the course of the simulation, the system spends a roughly comparable amount of time at a wide range of Eddington ratios from $`l0.0011`$. At high luminosities, $`L>10^{12}L_{\mathrm{}}`$ for a system with $`L_{\mathrm{peak}}10^{13}L_{\mathrm{}}`$, the range of Eddington ratios, is concentrated at high values $`l0.51`$ with some time spent at ratios as low as $`l0.1`$. Note, however, that the y-axis of the plot is scaled logarithmically, so the time spent at $`l0.1`$ in this luminosity interval is a factor $`5`$ smaller than the time spent at $`l0.5`$. Considering lower luminosities $`10^{11}L_{\mathrm{}}<L<10^{12}L_{\mathrm{}}`$, the distribution of Eddington ratios broadens down to $`l0.01`$. Going to lower luminosities still, $`L<10^{11}L_{\mathrm{}}`$, the distribution broadens further, with comparable time spent at ratios as low as $`l0.001`$, and becomes somewhat bimodal. At large luminosities near $`L_{\mathrm{peak}}`$, the system is primarily in Eddington-limited or near-Eddington growth. However, as we consider lower luminosities, we include both early times when the black hole is growing efficiently (high $`l`$) and late or intermediate times when the black hole is more massive but the accretion rate falls (low $`l`$). As we go to lower luminosities, the total time spent in sub-Eddington states increasingly dominates the time spent at $`l1`$, although the time spent at any given value of $`l`$ is fairly flat with $`\mathrm{log}(l)`$.
Roughly, at some luminosity $`L`$, there is a constant probability of being in some logarithmic interval in $`l`$,
$$P(l|L,L_{\mathrm{peak}})\left[\mathrm{log}\left(\frac{L_{\mathrm{peak}}}{L}\right)\right]^1,\frac{L}{L_{\mathrm{peak}}}<l<1,$$
(30)
and $`P(l|L,L_{\mathrm{peak}})=0`$ otherwise. This is especially clear if we compare the distribution of Eddington ratios in each luminosity range obtained if we consider only times after the final merger of the black holes (dotted histograms). At the highest luminosities, the distribution is identical to that obtained previously, since all the time at these luminosities is during the final merger. However, as we move to lower luminosities, the characteristic $`l`$ move systematically lower, as we are seeing only the relaxation after the final โblowoutโ near $`L_{\mathrm{peak}}`$, with characteristic Eddington ratio $`l=L/L_{\mathrm{peak}}`$ at any given luminosity $`L`$. These trends are also clear if we consider the distribution of instantaneous black hole masses in each luminosity interval shown in the figure, which is trivially related to the Eddington ratio distribution at a given luminosity $`L`$ as
$$M_{\mathrm{BH}}=M_0\frac{L}{lL_{\mathrm{Edd}}(M_0)}=\frac{Lt_S}{lฯต_rc^2}.$$
(31)
Of course, it is clear here that $`M_{\mathrm{BH}}M_{\mathrm{BH}}^f=3\times 10^8M_{\mathrm{}}`$ if we consider only times after the final merger.
It has also been argued from observations of stellar black hole binaries that a transition between accretion states occurs at a critical Eddington ratio $`\dot{m}\dot{M}/\dot{M_{\mathrm{Edd}}}`$, from radiatively inefficient accretion flows at low accretion rates (e.g., Esin, McClintock, & Narayan, 1997) to radiatively efficient accretion through a standard Shakura & Sunyaev (1973) disk. Although the critical Eddington ratio for supermassive black holes is uncertain, observations of black hole binaries (Maccarone, 2003) as well as theoretical extensions of accretion models (e.g., Meyer, Liu, & Meyer-Hofmeister, 2000) suggest $`\dot{m}_{\mathrm{crit}}0.01`$. We can examine whether this has a large impact on our predictions for the luminosity function and $`\dot{n}(L_{\mathrm{peak}})`$ distribution, by determining whether the distribution of Eddington ratios is significantly changed by such a correction. Because we assume a constant radiative efficiency $`L=ฯต_r\dot{M}c^2`$ with $`ฯต_r=0.1`$, we account for this effect by multiplying the simulation luminosity at all times by an additional โefficiency factorโ $`f_{\mathrm{eff}}`$ which depends on the Eddington ratio $`l=L/L_{\mathrm{Edd}}`$,
$$f_{\mathrm{eff}}=\{\begin{array}{cc}1\hfill & \mathrm{if}l>0.01\hfill \\ 100l\hfill & \mathrm{if}l0.01.\hfill \end{array}$$
(32)
This choice for the efficiency factor follows from ADAF models (Narayan & Yi, 1995) and ensures that the radiative efficiency is continuous at the critical Eddington ratio $`l_{\mathrm{crit}}=0.01`$. Applying this correction and then examining the distribution of Eddington ratios as a function of luminosity (dashed histograms in Figure 19), we see that the distribution of Eddington ratios is essentially identical, with only a slightly higher probability of observing extremely low Eddington ratios $`l0.001`$. Of course, our modeling of accretion processes does not allow us to accurately describe ADAF-like accretion at these low Eddington ratios, but such low values are not relevant for the observed luminosity functions and quantities with which we make our comparisons. This implies that such a transition in the radiative efficiency with accretion rate should not alter our conclusions regarding the luminosity function and the $`\dot{n}(L_{\mathrm{peak}})`$ distribution (essentially, the corrections are important only at luminosities well below those relevant in constructing the observed luminosity functions; see also Hopkins et al. 2005c for a calculation of the effects of such a correction on the fitted quasar lifetime and $`\dot{n}(L_{\mathrm{peak}})`$ distributions, which leads to the same conclusion).
Despite the broad range of Eddington ratios in the simulations, this entire distribution is unlikely to be observable in many samples. The effect of this can be predicted based on the behavior seen in Figure 19. For example, we consider the distribution of Eddington ratios that would be observed if the B-band luminosity $`L_{B,\mathrm{obs}}10^{11}L_{\mathrm{}}`$, comparable to the selection limits at high redshift of many optical quasar samples. As expected from the change in $`l`$ with luminosity, this restricts the observed range of Eddington ratios to large values $`l0.11`$, in good agreement with the range of Eddington ratios actually observed in such samples. Essentially, it has reduced the observed range to a bolometric luminosity $`L10^{12}L_{\mathrm{}}`$ in the case shown, giving a similar distribution to that seen in the lower panel of the figure.
We compare our predicted distribution of Eddington ratios to observations in Figure 20. Using the distribution of peak luminosities $`\dot{n}(L_{\mathrm{peak}})`$ determined from the luminosity function, we can integrate over all luminosities to infer the observed Eddington ratio distribution,
$`P(l)`$ $``$ $`{\displaystyle \mathrm{d}\mathrm{log}L\mathrm{d}\mathrm{log}L_{\mathrm{peak}}}`$ (33)
$`\times P(l|L,L_{\mathrm{peak}}){\displaystyle \frac{\mathrm{d}t(L,L_{\mathrm{peak}})}{\mathrm{d}\mathrm{log}L}}\dot{n}(L_{\mathrm{peak}}).`$
As our estimate of $`P(l|L,L_{\mathrm{peak}})`$ above is rough, we do this by binning in $`L_{\mathrm{peak}}`$ and averaging the binned $`P(l|L,L_{\mathrm{peak}})\mathrm{d}t/\mathrm{d}\mathrm{log}L`$ for each simulation in the range of $`L_{\mathrm{peak}}`$, then weighting by $`\dot{n}(L_{\mathrm{peak}})`$ and integrating. We consider both the entire distribution that would be observed in the absence of selection effects (red histograms), and the distribution observed demanding a B-band luminosity above some reference value, $`L_{B,\mathrm{obs}}>L_{\mathrm{min}}`$ (black histograms). The results are shown for redshifts $`z<0.5`$ and $`z=1.53.5`$, along with the observed distribution from Vestergaard (2004), with assumed Poisson errors. The observations should be compared to the black histograms, which have luminosity thresholds $`L=10^{10}L_{\mathrm{}}\mathrm{and}10^{11}L_{\mathrm{}}`$ for $`z<0.5`$ and $`z=1.53.5`$, respectively, corresponding approximately to the minimum observable luminosities in the observed samples in each redshift interval.
The agreement is good, given the observational uncertainties, and it suggests that the observed Eddington ratio distribution can be related to the non-trivial nature of quasar lifetimes and light curves we model, rather than some arbitrary distribution of fixed $`l`$ across sources. However, the selection effects in the observed samples are quite significant โ the complete distribution of Eddington ratios is similar in both samples, implying that the difference in the observed Eddington ratio distribution is primarily a consequence of the higher luminosity limit in the observed samples โ and a more detailed test of this prediction requires fainter samples.
Still, there is a systematic offset in the observed samples at $`z<0.5`$ and $`z=1.53.5`$ which does not owe to selection effects. At progressively lower redshifts, more quasars with luminosities further below the โbreakโ in the luminosity function are observed, and therefore the observed Eddington ratio is broadened to lower Eddington ratios $`l0.1`$, whereas at high redshift the distribution is more peaked at slightly higher Eddington ratios. This difference, although not dramatic, is a prediction of our model not captured in โlight bulbโ or โfixed Eddington ratioโ models, even when allowing for a distribution of Eddington ratios, if such a distribution is static. We demonstrate this by fitting the low-redshift Eddington ratio distribution to a Gaussian (blue dashed lines in upper left), and then assuming that this distribution of accretion rates is unchanged with redshift, giving (after applying the same selection effects which yield the black histograms plotted) the blue dashed line in the upper right panel. Although the agreement may appear reasonable, the difference is significant โ such a fit overpredicts the fraction of high redshift objects at Eddington ratios $`0.1`$ and underpredicts the fraction at $`0.3`$, giving a somewhat poor fit overall ($`\chi ^2/\nu =2.7`$, but with typical $`3\sigma `$ overpredictions for Eddington ratios $`0.1`$).
Furthermore, without being modified to allow for a distribution of Eddington ratios, such models are clearly inconsistent with the observations, as they would predict a single, constant Eddington ratio. However, models which fit the observed evolution in the quasar luminosity function with a non-static distribution of accretion rates do recover the broadening of the Eddington ratio distribution at low redshift, so long as strong evolution in the distribution of accretion rates for systems of a given black hole mass is not allowed (Steed & Weinberg, 2003), giving a qualitatively similar picture of the evolution we model. Regardless of the evolution in accretion rates, an advantage of our modeling is that it provides a physically motivated predicted distribution of accretion rates, as opposed to being forced to adopt the distribution of accretion rates from observational input (which can be, as demonstrated in the figure, significantly biased by observational selection effects). The dotted histograms show the distribution if we apply our ADAF correction to the intrinsic distribution, and demonstrate that this does not significantly change the result. We note that our model for black hole accretion employs the Eddington limit as a maximum accretion rate; if we remove this restriction, we find that the simulations spend some small but non-negligible time with $`l12`$, which is also consistent with the observations.
Furthermore, we can make a prediction of this model which can be falsified, namely that the Eddington ratio distribution at luminosities well below the break in the luminosity function should be broader and extend to lower values than the distribution at luminosities above the break luminosity. We quantify this in the lower panels of Figure 20, for the distribution at low redshifts $`z1`$. Here we consider two bins in luminosity, $`L=10^{9.5}10^{10.5}L_{\mathrm{}}`$ and $`L=10^{12.5}10^{13.5}L_{\mathrm{}}`$, for redshifts where the break in the luminosity function is at approximately $`L10^{11}10^{12}L_{\mathrm{}}`$. Clearly, the distribution is broader and extends to lower Eddington ratios in the former luminosity interval, whereas in the latter it is strongly peaked about $`l0.21`$, for both the complete distribution (black) and that with $`L_{B,\mathrm{obs}}10^{11}L_{\mathrm{}}`$ (red). The distribution obtained applying the ADAF correction described above is shown as dotted histograms. Despite the fact that the Eddington ratio distribution at low luminosities will be strongly biased by selection effects, a reasonably complete sample should be able to test this prediction, at least qualitatively.
We illustrate the effects of changing observed waveband, redshift, and luminosity thresholds on the observed Eddington ratio distribution in Figure 21. Here, we plot the predicted distribution of Eddington ratios determined as in Figure 20, from our fitted $`\dot{n}(L_{\mathrm{peak}})`$ distribution at each redshift and the distribution of Eddington ratios as a function of instantaneous and peak luminosity in each of our simulations (specifically, these are drawn from the Monte Carlo realizations of the quasar population described in ยง 8). We show the predictions at three redshifts $`z=0.5`$ (top panels), $`z=1.0`$ (middle), and $`z=2.0`$ (bottom). For each redshift, results are shown in three wavebands, and with three reference luminosities. In B-band, we require $`M_B<19`$ (red), $`M_B<22`$ (blue), and $`M_B<25`$ (black). In soft X-rays, $`\mathrm{log}(L_{SX}[\mathrm{erg}\mathrm{s}^1])>40(\mathrm{red}),42(\mathrm{blue}),44(\mathrm{black})`$. In hard X-rays, $`\mathrm{log}(L_{HX}[\mathrm{erg}\mathrm{s}^1])>41(\mathrm{red}),43(\mathrm{blue}),45(\mathrm{black})`$. The observationally inferred distribution of Eddington ratios at each redshift is loosely estimated by adopting a constant bolometric correction from the observed (attenuated) luminosity in each of three bands shown, i.e. assuming $`L=12L_B^{\mathrm{obs}}`$ ($`4400`$ร
; left), $`L=52L_{SX}^{\mathrm{obs}}`$ (0.5-2 keV; middle), and $`L=35L_{HX}^{\mathrm{obs}}`$ (2-10 keV; right). This follows common practice in many observational estimates of the Eddington ratio distribution and allows for the effects of attenuation, but we caution that it can be misleading.
If we instead use the luminosity-dependent bolometric corrections of Marconi et al. (2004) which we adopt throughout, even given that we are calculating from the observed (attenuated) luminosities, we do not see the large population of highly sub-Eddington (Eddington ratios $`10^3`$) quasars in soft and hard X-ray samples with low luminosity thresholds. This is because these are actually reasonably high-Eddington ratio quasars, but our bolometric corrections imply that a larger fraction of the bolometric luminosity is radiated in the X-ray at low bolometric luminosity, meaning that assuming a constant bolometric correction will underestimate the Eddington ratios of high-bolometric luminosity sources. Regardless, the figure illustrates both the importance of different wavelengths (i.e. the ability to observe more low-Eddington ratio sources in X-ray as compared to optical samples) and luminosity/magnitude limits on the inferred distribution of Eddington ratios. For example, even for relatively deep B-band quasar samples complete to $`M_B<23`$ (i.e. complete to essentially all objects traditionally classified as having โquasar-likeโ luminosities), the expected observed Eddington ratio distribution at $`z0.52`$ is quite sharply peaked about $`0.10.3`$, in good agreement with recent observational results (Kollmeier et al., 2005).
We do not compare to the $`z=0`$ distribution of black hole accretion rates, as this is dominated by objects at extremely low Eddington ratios $`l10^510^4`$ (e.g., Ho, 2002; Marchesini et al., 2004; Jester, 2005), which are well below the range we model, and are not likely to be driven by merger activity (many of these objects are quiescent, low-luminosity Seyferts in normal spiral galaxy hosts); furthermore, many of these objects are not accreting at the Bondi rate (Fabian & Canizares, 1988; Blandford & Begelman, 1999; Di Matteo et al., 2000; Narayan et al., 2000; Quataert & Gruzinov, 2000; Di Matteo et al., 2001; Loewenstein et al., 2001; Bower et al., 2003), clearly showing that our simulations must incorporate more sophisticated models for accretion in quiescent, low-luminosity states (when gravitational torques cannot provide a mechanism to drive large amounts of gas to the central regions of the galaxy) in order to describe such phases.
However, it has been suggested that the rapid โblowoutโ phase and subsequent decay in accretion rates seen in our simulations, coupled with spectral modeling of radiatively inefficient accretion modes, can explain the apparently bimodal distribution of low-redshift accretion rates (Cao & Xu, 2005). Moreover, present-day, relaxed ellipticals are observed to have mass accretion rates $`10^4`$ implying a long relaxation time at moderate and low accretion rates, qualitatively similar to that seen after the โblowoutโ in our modeling (Hopkins et al. 2005f). A pure exponential decay in accretion rate after the peak quasar phase would give $`\dot{m}=\dot{M}/\dot{M}_{\mathrm{Edd}}\mathrm{exp}(t_H/t_Q)`$ at present, where $`t_H`$ is the Hubble time and $`t_Q`$ is the quasar lifetime of order e.g. the Salpeter time $`t_S=4\times 10^7`$ yr, yielding an unreasonably low expected accretion rate $`\dot{m}10^{145}`$. Even assuming an order of magnitude larger quasar lifetime, this gives $`\dot{m}10^{15}`$, far below observed values, implying that regardless of the fueling mechanisms at low luminosities, the basic key point of our modeling must be true to some extent, namely that quasars spend long times relaxing at moderate to low Eddington ratios.
## 6. The Mass Function of Relic Supermassive Black Holes from Quasars
From the $`M_{\mathrm{BH}}`$-$`\sigma `$ relation and other host galaxy-black hole scalings, estimates of bulge and spheroid velocity dispersions have been used to determine the total mass density ($`\rho _{\mathrm{BH}}`$) and mass distribution of local, primarily inactive supermassive black holes (e.g., Salucci et al., 1999; Marconi & Salvati, 2002; Yu & Tremaine, 2002; Ferrarese, 2002; Aller & Richstone, 2002; Marconi et al., 2004; Shankar et al., 2004). These estimates, along with others based on X-ray background synthesis (e.g., Fabian & Iwasawa, 1999; Elvis et al., 2002), have compared these quantities to those expected based on the mass distribution of โrelicโ black holes grown in quasars. It appears that most, and perhaps nearly all of the present-day black hole mass density was accumulated in bright quasar phases, and the $`M_{\mathrm{BH}}\sigma `$ and $`M_{\mathrm{BH}}L_{\mathrm{bulge}}`$ correlations yield estimates of the local mass function in good agreement with those from hard X-ray AGN luminosity functions (Marconi et al., 2004).
However, this modeling is dependent on several assumptions. Namely, the average radiative efficiency $`ฯต_r`$, Eddington ratio $`l`$, and average quasar lifetime $`t_Q`$ are generally taken to be constants and either input into the model or constrained by demanding agreement with the local mass function. In our simulations, we find the quasar lifetime and Eddington ratio to be complex functions of both luminosity and host system properties (as opposed to being constants). We also find that quasars spend a large fraction of their lives in obscured growth phases, suggesting some mass gain outside of the bright quasar phase. It is thus of interest to determine the relic black hole mass function expected from our model for quasar evolution.
Using our estimate for the birthrate of quasars with a given peak luminosity at a particular redshift, $`\dot{n}(L_{\mathrm{peak}})`$, obtained from the luminosity function in ยง 3.2, we can estimate the total number density of relic quasars accumulated by a particular redshift that were born with a given $`L_{\mathrm{peak}}`$ (per logarithmic interval in $`L_{\mathrm{peak}}`$) from
$$n(L_{\mathrm{peak}})=\dot{n}(L_{\mathrm{peak}})dt=\frac{\dot{n}(L_{\mathrm{peak}},z)\mathrm{d}z}{(1+z)H(z)}.$$
(34)
By redshift $`z=0`$, most of these quasars will be โdead,โ with only a small residual fraction having been activated in the recent past.
Using our log-normal form for $`\dot{n}(L_{\mathrm{peak}})`$, with normalization $`\dot{n}_{}`$ and dispersion $`\sigma _{}`$ held constant and only the median $`L_{}=L_{}^0\mathrm{exp}(k_L\tau )`$ evolving with redshift, this integral can be evaluated numerically to give the space density of relic quasars $`n(L_{\mathrm{peak}})`$. Finally, we use $`M_{\mathrm{BH}}^f(L_{\mathrm{peak}})`$, roughly the Eddington mass of the given peak luminosity (but determined more precisely in ยง 2.4) to convert from $`\mathrm{d}n(L_{\mathrm{peak}})/\mathrm{d}\mathrm{log}L_{\mathrm{peak}}`$ to $`\mathrm{d}n(M_{\mathrm{BH}})/\mathrm{d}\mathrm{log}M_{\mathrm{BH}}`$. This formulation implicitly assumes that black holes do not undergo subsequent mergers after the initial quasar-producing event. However, this effect should be small (a factor $`2`$) as subsequent mergers would be dry (gas poor). We explicitly calculate the effects of dry mergers on the spheroid mass function (essentially a rescaling of the black hole mass function calculated here) in Hopkins et al. (2005e), and show that this is a small effect (significantly less than the uncertainties owing to our fit to the quasar luminosity function) even assuming the maximum dry merger rates of e.g. van Dokkum (2005).
This mass function can then be integrated over $`dM_{\mathrm{BH}}`$ to give the total present-day black hole mass density, $`\rho _{\mathrm{BH}}`$. Neglecting temporarily the small corrections to $`M_{\mathrm{BH}}^f(L_{\mathrm{peak}})`$ from ยง 2.4, we expect
$$M_{\mathrm{BH}}^fM_{\mathrm{Edd}}(L_{\mathrm{peak}})=\frac{L_{\mathrm{peak}}t_S}{ฯต_rc^2}$$
(35)
where $`t_S/ฯต_rc^22.95\times 10^5M_{\mathrm{}}/L_{\mathrm{}}`$, so therefore,
$$\rho _{\mathrm{BH}}=\frac{t_S}{ฯต_rc^2}L_{\mathrm{peak}}n(L_{\mathrm{peak}})\mathrm{d}\mathrm{log}L_{\mathrm{peak}}.$$
(36)
This can be combined with the integral over redshift for $`n(L_{\mathrm{peak}})`$, giving, at each $`z`$, a pure Gaussian integral over $`\mathrm{log}(L_{\mathrm{peak}})`$, in the form
$`\rho _{\mathrm{BH}}`$ $`=`$ $`{\displaystyle \frac{L_{}^0t_S}{ฯต_rc^2}}{\displaystyle \frac{\dot{n}_{}}{H_0}}e^{\frac{1}{2}(\sigma _{}\mathrm{ln}10)^2}{\displaystyle \frac{e^{k_L\tau }dz}{(1+z)\widehat{H}(z)}}`$ (37)
$`=`$ $`{\displaystyle \frac{L_{}^0t_S}{k_Lฯต_rc^2}}{\displaystyle \frac{\dot{n}_{}}{H_0}}e^{\frac{1}{2}(\sigma _{}\mathrm{ln}10)^2}\left(e^{k_L\tau _f}e^{k_L\tau }\right),`$
where $`\widehat{H}(z)H(z)/H_0`$ and $`\tau _f`$ is the fractional lookback time at some upper limit. We must modify this integral above $`z2`$ to account for the decreasing space density of bright quasars, applying either our density or peak-luminosity evolution turnover from ยง 3.2, but quasars at these high redshifts contribute only a small fraction to the present-day density. Thus, in this formulation, the evolution of the total supermassive black hole mass density, i.e. $`\rho _{\mathrm{BH}}(z)/\rho _{\mathrm{BH}}(z=0)`$, is given approximately by the dimensionless integral above, and depends only on how $`L_{}`$ evolves, essentially the rate at which the break in the quasar luminosity function shifts. Although this is not strictly true if we include corrections to $`M_{\mathrm{BH}}^f(L_{\mathrm{peak}})`$ based on $`L_{\mathrm{peak}}`$, the difference is small and this behavior is essentially preserved. Note that the total supermassive black hole mass density is independent of corrections from subsequent dry mergers, which (being gas poor) conserve total black hole mass.
Figure 22 shows our prediction for the mass distribution of supermassive black holes, as well as the total density $`\rho _{\mathrm{BH}}`$ and its evolution with redshift. We find a total relic black hole mass density of $`\rho _{\mathrm{BH}}=2.9_{1.2}^{+2.3}\times 10^5M_{\mathrm{}}\mathrm{Mpc}^3`$, in agreement with the observational estimate of $`\rho _{\mathrm{BH}}=2.9\pm 0.5h_{0.7}^2\times 10^5M_{\mathrm{}}\mathrm{Mpc}^3`$, by Yu & Tremaine (2002) ($`h_{0.7}H_0/70\mathrm{km}\mathrm{s}^1\mathrm{Mpc}^1`$; their result is converted from $`h=0.65`$), and within $`1\sigma `$ of the value $`\rho _{\mathrm{BH}}=4.6_{1.4}^{+1.9}h_{0.7}^2\times 10^5M_{\mathrm{}}\mathrm{Mpc}^3`$, of Marconi et al. (2004), based on the observations of Marzke et al. (1994), Kochanek et al. (2001), Nakamura et al. (2003), Bernardi et al. (2003), and Sheth et al. (2003). The fractional evolution of $`\rho _{\mathrm{BH}}`$ with redshift is quite well constrained, and we find, as with previous estimates, that most of the present-day black hole mass density accumulates at moderate to low redshifts $`z0.52.5`$. The $`1\sigma `$ errors are shown as dotted lines in the figure, and are close to our best-fit estimate, as we have demonstrated that this quantity depends only on $`k_L`$, the rate of evolution of the break in the luminosity function with redshift, which is fairly well-constrained by observations (from our fitting to the luminosity functions, $`k_L=5.61\pm 0.28`$). The difference in $`\rho _{\mathrm{BH}}`$ if we include or neglect the small corrections to $`M_{\mathrm{BH}}^f`$ is negligible compared to our errors ($`5\%`$).
Our estimate for the relic black hole mass distribution (thick black line) also agrees well with observational estimates, with all observations within the range allowed by the $`1\sigma `$ errors of our fitting to the luminosity function (dotted lines). The observations shown are again from Marconi et al. (2004), based on the combination of observations by Marzke et al. (1994), Kochanek et al. (2001), Nakamura et al. (2003), Bernardi et al. (2003), and Sheth et al. (2003). The high mass end of the black hole mass function $`M_{\mathrm{BH}}>10^9M_{\mathrm{}}`$ is relatively sensitive to whether or not we apply the $`M_{\mathrm{BH}}^f(L_{\mathrm{peak}})`$ corrections of ยง 2.4, instead of taking $`M_{\mathrm{BH}}^f=M_{\mathrm{Edd}}(L_{\mathrm{peak}})`$ (thin line), as well as to our fitting procedure. However, the agreement is still good, and this is also where the observational estimates of the mass distribution are most uncertain, as they are generally extrapolated to these masses, and are sensitive to the assumed intrinsic dispersions in the $`M_{\mathrm{BH}}\sigma `$ and $`M_{\mathrm{BH}}L_{\mathrm{bulge}}`$ relations (Yu & Tremaine, 2002).
If, instead, we adopt a light-bulb, constant Eddington ratio, or exponential light curve model for quasar evolution, we would have $`M_{\mathrm{BH}}^fL_{\mathrm{peak}}`$, and thus the prediction would be similar to the thin black line shown, a somewhat worse fit at high black hole masses. However, in these models this can be remedied by adjusting the typical Eddington ratios, quasar lifetimes, or radiative efficiencies. We do not show the range of predictions of these models for the relic supermassive black hole mass function, as they have been examined in detail previously (e.g., Salucci et al., 1999; Marconi & Salvati, 2002; Yu & Tremaine, 2002; Ferrarese, 2002; Aller & Richstone, 2002; Marconi et al., 2004; Shankar et al., 2004). These works demonstrate that the observed quasar luminosity functions are consistent with the relic supermassive black hole mass function, given typical radiative efficiencies $`ฯต_r0.1`$ and Eddington ratios $`0.51.0`$, and that most of the mass of black holes is accumulated in bright, observed phases, or else the required radiative efficiency would violate theoretical limits.
That our model of quasar lifetimes and obscuration reproduces the observed $`z=0`$ supermassive black hole mass function explicitly demonstrates that we are consistent with these constraints. By choice, the radiative efficiency in our simulations is $`ฯต_r=0.1`$, and accretion rates are not allowed to exceed Eddington. As noted in ยง 4, most of the black hole mass is accumulated and radiant energy released in the final, โblowoutโ phase of quasar evolution, and here our black hole mass function and cumulative black hole mass density demonstrate that our modeling is consistent with integrated energy and mass arguments such as that of Soltan (1982), despite the fact that quasars spend more time in obscured phases than they do in bright optical quasar phases. In fact, comparison of our predicted total black hole mass density with estimates from the $`z=0`$ black hole mass distribution allows some latitude for significant mass gain in radiatively inefficient growth or black holes in small, disky spheroids, although we emphasize that this is mainly because the uncertainty in our prediction is large, it is not inherent or necessary in our modeling.
The anti-hierarchical nature of black hole formation, where less massive black holes are formed at lower redshift, is reflected in our modeling by the shift of the break in the quasar luminosity function to lower values with decreasing redshift. This can be seen in Figure 22, where the black hole mass distributions are shown at redshifts $`z=1.5,\mathrm{\hspace{0.17em}3.0}\mathrm{and}5.0`$, assuming either pure peak luminosity evolution or pure density evolution for $`z>2`$ (dot-dashed and dashed, respectively). While the choice for the turnover in the $`z>2`$ quasar density matters little for the $`z<2`$ black hole mass functions, the low-$`M_{\mathrm{BH}}`$ distribution at high redshift (where observations do not constrain $`\dot{n}(L_{\mathrm{peak}})`$ well) is quite different between the two models. Figure 23 plots the fractional number density of black holes of a given mass as a function of redshift, i.e. $`n(M,z)/n(M,z=0)`$, where $`n(M)=\mathrm{d}n/\mathrm{d}\mathrm{log}(M)`$ is just the number density at mass $`M`$. This figure demonstrates that higher-mass black holes originated over a larger range of redshifts, and that they mostly formed at higher redshift, compared to lower-mass black holes.
The right panel of Figure 23 compares our prediction to that of a light-bulb or exponential light curve model for quasar lifetimes. In these models, the anti-hierarchical nature of black hole assembly is dramatically suppressed. At the high-mass end, there is no measurable difference in the distribution of formation redshifts (i.e. the $`M_{\mathrm{BH}}=10^9M_{\mathrm{}}`$ and $`M_{\mathrm{BH}}=10^{10}M_{\mathrm{}}`$ curves are indistinguishable), and there is little change in the formation times at $`M_{\mathrm{BH}}=10^8M_{\mathrm{}}`$. The shift in formation redshift at lower masses, although significant, is smaller than that predicted in our model. If spheroids and black holes are produced together, as in our picture, these models of the quasar lifetime would imply that spheroids of masses $`M_{\mathrm{vir}}10^{11}10^{13}M_{\mathrm{}}`$ all formed over nearly identical ranges of redshifts, which is inconsistent with many observations indicating anti-hierarchical growth of the red, elliptical galaxy population (e.g., Treu et al., 2001; van Dokkum et al., 2001; Treu et al., 2002; van Dokkum & Stanford, 2003; Gebhardt et al., 2003; Rusin et al., 2003; van de Ven et al., 2003; Wuyts et al., 2004; Treu et al., 2005; Holden et al., 2005; van der Wel et al., 2005; di Serego Alighieri et al., 2005; Nelan et al., 2005). Implications of our model for the red galaxy sequence are considered in Hopkins et al. (2005e), where we show that this weaker anti-hierarchical black hole (and correspondingly, spheroid) evolution is inconsistent with observed luminosity functions, color-magnitude relations, and mass-to-light ratios of elliptical galaxies.
Our modeling reproduces the observed total density and mass distribution of supermassive black holes at $`z=0`$ with black holes accreting at the canonical efficiency $`ฯต_r=0.1`$ expected for efficient accretion through a Shakura & Sunyaev (1973) disk. Presumably, a large change in $`ฯต_r`$ would give a significantly different relation between peak luminosity and black hole mass (for the same $`L_{\mathrm{peak}}`$, $`M_{\mathrm{BH}}^f1/ฯต_r`$), and thus if the quasar lifetime remained similar as a function of peak luminosity, this would translate to a shift in the black hole mass function. The long obscured stage in black hole evolution does not generate problems in reproducing the black hole mass density, and the final phases of growth are still in bright optical quasar stages. However, a large Compton-thick population of black holes at all luminosities (or even at some range of luminosities at or above the break in the luminosity function) (e.g., Gilli et al., 2001; Ueda et al., 2003), or a large population accreting in a radiatively inefficient ADAF-type solution, as invoked to explain discrepancies in the X-ray background produced by synthesis models (Di Matteo et al., 1999), would result in a significant over-prediction of the present-day supermassive black hole density. As we demonstrate in ยง 7.2, invoking such populations is unnecessary, as our picture for quasar lifetimes and evolutionary obscuration self-consistently reproduces the observed X-ray background.
Finally, we note that we reproduce the $`z=0`$ distribution of black hole masses inferred from the distribution of spheroid velocity dispersions (Sheth et al., 2003) and luminosity functions (Marzke et al., 1994; Kochanek et al., 2001; Nakamura et al., 2003), based on the observed $`M_{\mathrm{BH}}\sigma `$ relation and fundamental plane for galaxy properties (e.g., Bernardi et al., 2003; Gebhardt et al., 2003). Therefore, since our modeling also reproduces the observed $`M_{\mathrm{BH}}\sigma `$ (Di Matteo et al., 2005; Robertson et al., 2005b) and fundamental plane (Robertson et al., in preparation) relations, we implicitly reproduce the $`z=0`$ distribution of spheroid velocity dispersions and spheroid luminosity functions, given our basic assumption that the mergers that produce these spheroids also give rise to luminous quasar activity.
## 7. The Cosmic X-Ray Background
### 7.1. The Integrated Spectra of Individual Quasars
Unresolved extragalactic sources, specifically obscured AGN, have been invoked to explain the cosmic X-ray background (e.g, Setti & Woltjer, 1989). This picture has been confirmed as deep surveys with Chandra and XMM-Newton have resolved most or all of the X-ray background into discrete sources, primarily obscured and unobscured AGN (Brandt et al., 2001; Hasinger et al., 2001; Rosati et al., 2002; Giacconi et al., 2002; Baldi et al., 2002). The X-ray background, however, has a harder X-ray spectrum than typical quasars, with a photon index $`\mathrm{\Gamma }1.4`$ in the $`110`$ keV range (Marshall et al., 1980). Therefore, obscured AGN are important in producing this shape, as absorption in the ultraviolent and soft X-rays hardens the observed spectrum. Indeed, population synthesis models based on observed quasar luminosity functions and involving large numbers of obscured AGN have been successful at matching both the X-ray background intensity and spectral shape (Madau et al., 1994; Comastri et al., 1995; Gilli et al., 1999, 2001). However, these models make arbitrary assumptions about the ratio of obscured to unobscured sources and its evolution with redshift, choosing these quantities to reproduce the X-ray background. Furthermore, as X-ray surveys have been extended to higher redshifts, it has become clear that both the observed redshift distribution of X-ray sources and the ratio of obscured to unobscured sources is inconsistent with that required by these models (Hasinger, 2002; Barger et al., 2003). Even synthesis models based on higher-redshift X-ray surveys and using observationally derived ratios of obscured to unobscured sources (e.g., Ueda et al., 2003) have invoked ad hoc assumptions about additional populations of obscured sources to reproduce the X-ray background shape and intensity.
We can test our model by examining whether the quasar luminosity function, relic AGN mass distribution, and X-ray background can be simultaneously reproduced in a self-consistent manner. Because our formulation describes the birthrate of quasars with a peak luminosity $`L_{\mathrm{peak}}`$, it is most useful to consider the integrated energy spectrum of such a quasar over its lifetime,
$$\nu E_\nu =dt\nu L_\nu (t)=\nu f_\nu (L)L\frac{\mathrm{d}t(L,L_{\mathrm{peak}})}{\mathrm{d}\mathrm{log}L}\mathrm{d}\mathrm{log}L,$$
(38)
where $`f_\nu (L)`$ is the bolometric correction ($`L_\nu f_\nu L`$). As an example, Figure 24 shows the integrated intrinsic spectra (thick solid lines) from the simulations A1, A2, A3, A4, and A5, described in ยง 2.1. The final black hole masses for these simulations are $`M_{\mathrm{BH}}^f=7\times 10^6,\mathrm{\hspace{0.17em}3}\times 10^7,\mathrm{\hspace{0.17em}3}\times 10^8,\mathrm{\hspace{0.17em}7}\times 10^8,\mathrm{and}2\times 10^9M_{\mathrm{}}`$, respectively. The integrated spectral shape in the X-ray, in particular, is ultimately determined by the observationally motivated bolometric corrections of Marconi et al. (2004), with a reflection component in the X-ray determined following Magdziarz & Zdziarski (1995), and, in the case of the observed spectrum, the distribution of column densities calculated from the simulations. Using our fits to the lifetime $`\mathrm{d}t/\mathrm{d}\mathrm{log}L`$ as a function of instantaneous and peak luminosities, we can calculate the expected $`\nu E_\nu `$ from the integral above. These integrated spectra are shown as the dot-dashed lines in the figure, and agree well with the actual integrated spectra of the simulations, demonstrating the self-consistency of our model and applicability of our fitted lifetimes.
This can be compared to idealized models for the quasar lifetime, where we allow the quasar to radiate just at its peak luminosity $`L_{\mathrm{peak}}L_{\mathrm{Edd}}(M_{\mathrm{BH}}^f)`$ for some fixed lifetime $`t_Q^0`$. We determine $`t_Q^0`$ by demanding that the total energetics be correct, $`L_{\mathrm{peak}}t_Q^0=ฯต_rM_{\mathrm{BH}}^fc^2`$. The predicted integrated energy spectra are shown as the dashed lines, and under-predict the soft and hard X-ray energy output by a factor $`1.52`$. This is because higher-luminosity quasars tend to have a larger fraction of their energy radiated in the UV-optical rather than the X-ray (e.g., Wilkes et al., 1994; Green et al., 1995; Vignali et al., 2003; Strateva et al., 2005), reflected in our bolometric corrections. Thus, assuming that the quasar spends all its time at $`L_{\mathrm{peak}}`$ does not account for extended times at lower luminosity, where the ratio of X-ray to total luminosity is higher, which would generate an integrated spectrum with a larger fraction of its energy in the X-ray. Assuming that the quasar undergoes pure Eddington-limited growth to its peak luminosity produces an almost identical integrated spectrum to this light-bulb model, as it is similarly dominated by $`LL_{\mathrm{peak}}`$.
Of course, the intrinsic integrated energy spectrum of the simulations is not what determines the X-ray background, but rather the integrated observed spectrum is the critical quantity. This is shown as the thin lines in the left panel of Figure 24, and in detail for our fiducial A3 simulation in the right panel of the figure (thick solid line). Along a given sightline, the observed integrated spectrum will be
$$\nu \frac{\mathrm{d}E_\nu }{\mathrm{d}\mathrm{\Omega }}=dt\nu \frac{L_\nu (t)}{4\pi }e^{\tau _\nu (\mathrm{\Omega },t)},$$
(39)
where $`\tau _\nu `$ is the optical depth at a given frequency. We can integrate over solid angle and obtain
$$\nu E_{\nu ,\mathrm{obs}}=\nu f_\nu e^{\tau _\nu }L\frac{\mathrm{d}t(L,L_{\mathrm{peak}})}{\mathrm{d}\mathrm{log}L}\mathrm{d}\mathrm{log}L,$$
(40)
where $`e^{\tau _\nu }`$ is the averaged $`e^{\tau _\nu }`$ over the column density distribution $`P(N_\mathrm{H}|L,L_{\mathrm{peak}})`$. Using our fits to the column density distribution and quasar lifetimes and calculating $`\nu E_{\nu ,\mathrm{obs}}`$ as above, we reproduce the integrated observed spectrum quite well (black dashed line). For comparison, we show that it is not a good approximation to redden the spectrum with a constant $`N_\mathrm{H}`$, giving the results for $`N_\mathrm{H}=10^{21},10^{21.5},10^{22},10^{22.5},\mathrm{and}10^{23}\mathrm{cm}^2`$ (thin dot-dashed lines). Even allowing for a distribution of $`N_\mathrm{H}`$ values, the resulting spectrum is a poor match to the observed one if that distribution is taken to be static (i.e. luminosity-independent, as in traditional torus models, for example). We show the results of reddening the intrinsic spectrum by such a (Gaussian) distribution, varying the dispersion $`\sigma _{N_H}=0.4,\mathrm{\hspace{0.17em}0.7},\mathrm{\hspace{0.17em}1.0}`$ (blue, green, and red dashed lines, respectively), for a median column density $`\overline{N}_H=10^{22}\mathrm{cm}^2`$, the median column density expected around $`L_{\mathrm{peak}}`$ in this simulation. Therefore, the luminosity and host system property dependence of both quasar lifetimes and the column density distribution must be accounted for in attempting to properly predict the X-ray background spectrum from observations of the quasar luminosity function. Finally, note that the hard cutoff in the observed UV spectra at 912ร
owes to our calculated cross-sections being incomplete in the extreme UV. Properly modeling the escape fraction and observed emission at these frequencies, while not important for the X-ray background, is critical to calculating the contribution of quasars to reionization, and requires a more detailed modeling of scattering and absorption, especially in the bright optical quasar phase.
### 7.2. The Integrated X-Ray Background
Given the volume emissivity $`j_\nu (z)`$ (per unit comoving volume) of some isotropic process at a given frequency at redshift $`z`$, the resulting background specific intensity at frequency $`\nu _0`$ at $`z=0`$ is (Peacock, 1999)
$$I_{\nu _0}=\frac{c}{4\pi }\frac{j_\nu [(1+z)\nu _0,z]}{(1+z)H(z)}dz.$$
(41)
If we were to consider the emissivity $`j_\nu `$ per unit physical volume, there would be an extra factor of $`(1+z)^3`$ in the integral above. In ยง 7.1, we determined the integrated observed energy $`E_{\nu ,\mathrm{obs}}(L_{\mathrm{peak}})`$ produced by a quasar with peak luminosity $`L_{\mathrm{peak}}`$. We have also inferred $`\dot{n}(L_{\mathrm{peak}})(z)`$ in ยง 3.2, the rate at which quasars of peak luminosity $`L_{\mathrm{peak}}`$ are created per unit comoving volume per unit cosmological time. Therefore, the comoving volume emissivity is just
$$j_\nu (z)=E_{\nu ,\mathrm{obs}}(L_{\mathrm{peak}})\dot{n}(L_{\mathrm{peak}})d\mathrm{log}L_{\mathrm{peak}},$$
(42)
or, expanding $`E_{\nu ,\mathrm{obs}}`$,
$`j_\nu (z)`$ $`=`$ $`{\displaystyle d\mathrm{log}L_{\mathrm{peak}}d\mathrm{log}L}`$ (43)
$`\times f_\nu e^{\tau _\nu }L{\displaystyle \frac{dt(L,L_{\mathrm{peak}})}{\mathrm{d}\mathrm{log}L}}\dot{n}(L_{\mathrm{peak}}).`$
If the column density distribution were independent of $`L_{\mathrm{peak}}`$, as is assumed in even luminosity-dependent torus models or observationally determined $`N_\mathrm{H}`$ functions used for X-ray background synthesis (e.g., Ueda et al., 2003), then we could combine terms in $`L_{\mathrm{peak}}`$ and integrate over them. This simplification, along with the definition of the luminosity function in terms of $`L_{\mathrm{peak}}`$, gives the more traditional formula for the X-ray background in terms of only the observed column density distribution and luminosity function,
$$j_\nu (z)=d\mathrm{log}L\frac{d\mathrm{\Phi }}{d\mathrm{log}L}L_\nu e^{\tau _\nu }.$$
(44)
However, as we showed in ยง 3.3 and ยง 7.1, neglecting the dependence on $`L_{\mathrm{peak}}`$ is not a good approximation at all luminosities and gives an inaccurate estimate of the integrated quasar spectrum; therefore, โpurely observation-basedโ synthesis models of the X-ray background will be inaccurate in a similar manner to synthesis models with an inappropriate model for the quasar lifetime. Essentially, this โaverages outโ the varying distribution of column densities with $`L_{\mathrm{peak}}`$, which changes the shape of the spectrum in a non-linear manner, especially when integrated over varying bolometric corrections as shown above.
Figure 25 (upper panel) shows the predicted X-ray background spectrum from our full modeling of quasar lifetimes and obscuration (solid lines). We use our analytical fits to the quasar lifetime and column density distributions as in ยง 7.1 above, as Figure 24 demonstrates that they accurately reproduce the actual integrated quasar X-ray spectra of the simulations, and the analytical forms are integrated over all luminosities and redshifts. The dotted lines show the deviation resulting from shifting the parameters describing our fitted $`\dot{n}(L_{\mathrm{peak}})`$ distribution by $`1\sigma `$ in either direction, although degeneracies in the parameters suggest that the actual uncertainty in the background prediction is smaller. The dashed line shows the predicted X-ray background if we ignore the broadening of the $`N_\mathrm{H}`$ distribution across simulations ($`\sigma _{N_H}=1.2`$) and instead consider only the dispersion of an individual simulation at a given luminosity ($`\sigma _{N_H}=0.4`$).
These can be compared to the observations of Gruber et al. (1999) (red curve, for $`E3\mathrm{keV}`$) and Barcons et al. (2000) (cyan curve, for $`E10\mathrm{keV}`$). We increase the normalization of the Gruber et al. (1999) spectrum to match that of the best estimate from Barcons et al. (2000) over the range of overlap, determined from combined ASCA, BeppoSAX, and ROSAT data to be $`10.0_{0.9}^{+0.6}\mathrm{keV}\mathrm{cm}^2\mathrm{s}^1\mathrm{sr}^1\mathrm{keV}^1`$ at 1 keV. The uncertainty in the normalization between the two samples, $`20\%`$, is shown as the shaded yellow range (alternatively, this represents the $`2\sigma `$ errors in the ROSAT normalization).
In the middle panel of the figure, we calculate the predicted X-ray background using our full model of the quasar lifetime, but with different models for quasar obscuration. The solid black line shows the prediction using our full model of quasar obscuration, and is identical to the solid black line in the upper panel. The observations are likewise shown in an identical manner to the upper panel. The dashed black line is the prediction adopting the standard torus model for quasar obscuration, and the dotted line adopts the receding (luminosity-dependent) torus model. These models produce the same overall $`30`$keV normalization, as this is relatively unaffected by obscuration, but they predict a slightly ($`20\%`$) higher background at low energies, giving a slightly softer spectrum. This may appear counterintuitive, given that in Figure 12 these models tend to overpredict the number of high-column density sources, but this is because these models predict a strongly bimodal column density distribution, with unobscured sightlines encountering negligible column densities. These unobscured sightlines dominate the soft X-ray integrated spectrum, where the large column densities through the torus attenuate the quasar spectrum heavily. However, this net offset in the predicted background spectrum is generally within the range of the systematic theoretical and observational uncertainties, and can further be alleviated by tuning the parameters of the torus model to fit the X-ray background spectrum (e.g. Treister & Urry 2005, although their fits require a larger fraction of Compton-thick $`N_\mathrm{H}10^{25}\mathrm{cm}^2`$ sources than shown for even the receding torus model in Figure 12). The feature at $`5`$keV in the standard torus model prediction is a consequence of assuming that โunobscuredโ lines of sight encounter negligible column density, and does not appear if such sightlines encounter moderate ($`10^{21}\mathrm{cm}^2`$) columns.
The lower panel of the figure shows the predicted X-ray background spectrum if we instead consider a light-bulb or exponential light curve (fixed Eddington ratio) model for the quasar lifetime, again with various descriptions of quasar obscuration. In such models, the predicted X-ray background spectrum is independent of the quasar lifetime or characteristic Eddington ratio assumed (see Equation 44). However, as shown in Figure 24, these models do imply a different integrated spectrum for quasars; i.e. different effective bolometric corrections for predicting the X-ray background. In particular, in this model, the observed quasar spectrum at a given luminosity (averaged over the quasar population at that luminosity) is the same as the โeffectiveโ quasar spectrum one would use to calculate the total contribution to the X-ray background from quasars of the corresponding observed or peak luminosity, whereas this is not true in our model of quasar lifetimes. The observations are shown in the same manner as the preceding panels. The black solid line shows the prediction with this simplified model for the quasar lifetime, but still adopting our full model for obscuration as a function of instantaneous and peak luminosity, the dashed line assumes instead a standard torus model for obscuration, and the dotted line assumes a receding torus for the obscuration. The variations among different obscuration models are relatively small at most energies, and similar to those discussed above adopting our full model of quasar lifetimes.
In all three cases, however, this model for the quasar lifetime significantly under-predicts the X-ray background, particularly at the $`30`$keV peak. This shortfall is well-known, and earlier attempts (e.g., Madau et al., 1994; Comastri et al., 1995; Gilli et al., 1999, 2001; Pompilio et al., 2000; Ueda et al., 2003) have generally had to invoke additional assumptions about large obscured populations or a strong increase in the obscured fraction with redshift, neither of which is consistent with observations (e.g., Hasinger, 2002; Barger et al., 2003; Ueda et al., 2003; Szokoly et al., 2004; Barger et al., 2005). The difference between the predictions of various quasar lifetime models is, as explained above, attributable to the difference between the integrated quasar spectrum produced in our full model of the quasar lifetime (in which quasars spend long periods of time at low luminosities, with harder X-ray spectra), and the integrated spectrum in these simplified quasar lifetime models, which is proportional to the instantaneous quasar spectrum, and therefore underpredicts the hard X-ray portion of the spectrum by as much as $`50\%`$.
Our prediction of the X-ray background agrees well with the observed spectrum over the range $`1100\mathrm{keV}`$. (At energies above $`100\mathrm{keV}`$ it is likely that processes we have not included, such as those involving magnetic fields, contribute significantly to the background.) Unlike previous synthesis models for the X-ray background, we are able to do so without invoking assumptions about large Compton thick populations or larger obscured populations at different redshifts. In part, this is because our modeling allows us to predict, based on $`\dot{n}(L_{\mathrm{peak}})`$ and our column density formulation, the population of Compton thick sources (see Figure 12). However, as we have demonstrated, it is primarily because the deficit in previous synthesis models can be attributed to their inability to properly account for the dependence of quasar lifetimes and attenuation on both the instantaneous quasar luminosity and the host system properties (peak luminosity). Our picture, on the other hand, yields an estimate for the X-ray background spectrum that is simultaneously consistent with the observed supermassive black hole mass distribution and total density, as well as the โluminosity-dependent density evolutionโ observed in X-ray samples (Hopkins et al., 2005f). The background is primarily built up from $`z2.5`$ to $`z0.5`$, as is evident from the evolution of the black hole mass density in Figure 22, although a harder spectrum at low luminosities will weight this slightly towards lower redshifts (where more low-luminosity quasars are forming). Compton thick and relaxing, low-luminosity sources are accounted for, not as large, independent populations, but as evolutionary phenomena continuously connected to the โnormalโ quasar population.
## 8. Discussion
### 8.1. General Implications of our Model
Our modeling suggests two important paradigm shifts in interpreting quasar populations and evolution:
(1) First, as proposed in Hopkins et al. (2005c), a proper accounting of the luminosity dependence of quasar lifetimes (as opposed to models in which quasars grow in a pure exponential fashion or turn on and off as โlight bulbsโ) implies a novel interpretation of the luminosity function. The steep bright end (luminosities above the โbreakโ in the luminosity function) consists of quasars radiating near their Eddington limits and is directly related to the distribution of intrinsic peak luminosities (or final black hole masses) as has been assumed previously. However, the shallow, faint end of the luminosity function describes black holes either growing in early stages of activity or in extended, quiescent states going into or coming out of a peak bright quasar phase, with Eddington ratios generally between $`l0.01`$ and 1. The โbreakโ luminosity in the luminosity function corresponds directly to the peak in the birthrate of quasars as a function of peak luminosity $`\dot{n}(L_{\mathrm{peak}})`$.
This interpretation resolves inconsistencies in a number of previous theoretical studies. For example, semi-analytical models of the quasar luminosity functions (e.g., Kauffmann & Haehnelt, 2000; Haiman & Menou, 2000; Wyithe & Loeb, 2003) assume, based on simplified models for the quasar lifetime, that quasars at the faint end of the luminosity function correspond to low final-mass black holes (low $`L_{\mathrm{peak}}L`$), presumably in small halos. Consequently, these models overpredict the number of active low-mass black holes (as estimated from radio source counts), especially at high redshift, by orders of magnitude (Haiman, Quataert, & Bower, 2004), and overpredict the number of low-mass spheroids and red galaxies observed (Hopkins et al., 2005e).
Moreover, both observations (McLure & Dunlop, 2004) and comparison of the present-day black hole mass function with radio and X-ray luminosity functions (e.g. Merloni, 2004) suggest anti-hierarchical evolution for the growth of supermassive black holes, where the most massive black holes were produced mainly at high ($`z2`$) redshift, and low-mass black holes mostly formed later, which does not follow from idealized descriptions of quasar lifetimes and the luminosity function (for a review, see e.g. Combes 2005).
A one-to-one correspondence between observed luminosity and black hole mass does produce anti-hierarchical behavior in some sense at the high-mass end, because the most massive black holes are formed at $`z23`$ during the peak of bright quasar activity and the quasar luminosity function evolves to lower luminosities at lower redshifts (as is also the case for our model because the bright end of the luminosity function is dominated by sources near their peak luminosities). However, at black hole masses equal to or below $`10^8M_{\mathrm{}}`$ (i.e. galaxies of stellar mass $`10^{11}M_{\mathrm{}}`$), the evolution in the quasar luminosity function implies a roughly constant production of black holes with these masses at all redshifts, which is inconsistent with observations of galaxy spheroids indicating that typical ages increase with mass, ruling out a large population of low-mass spheroids with ages equal to or older than those of high-mass spheroids (e.g., Treu et al., 2001; van Dokkum et al., 2001; Treu et al., 2002; van Dokkum & Stanford, 2003; Gebhardt et al., 2003; Rusin et al., 2003; van de Ven et al., 2003; Wuyts et al., 2004; Treu et al., 2005; Holden et al., 2005; van der Wel et al., 2005; di Serego Alighieri et al., 2005; Nelan et al., 2005). As demonstrated in Figure 23, such a model does not produce anti-hierarchical growth or any age gradients within the high-mass spheroid population, also inconsistent with observations. Even given observed โluminosity-dependent density evolutionโ (e.g. Page et al., 1997; Miyaji et al., 2000, 2001; La Franca et al., 2002; Cowie et al., 2003; Ueda et al., 2003; Fiore et al., 2003; Hunt et al., 2004; Cirasuolo et al., 2005; Hasinger, Miyaji, & Schmidt, 2005), implying that the densities of lower redshift quasars peak at lower redshift, the inferred anti-hierarchical evolution if observed luminosity directly corresponds to black hole mass (i.e. as in โlight-bulbโ or โfixed Eddington ratioโ models) is not strong enough to account for observed anti-hierarchical growth of the corresponding galaxy spheroids (Hopkins et al., 2005e).
Furthermore, in these earlier models, a โbreakโ in the luminosity function is not necessarily reproduced (Wyithe & Loeb, 2003), and the faint-end slope has no direct physical motivation. The break may be caused by feedback mechanisms which set a characteristic turnover in both the galaxy mass function and quasar luminosity function (e.g., Scannapieco & Oh 2004; Dekel & Birnboim 2004), as in our modeling. The $`\dot{n}(L_{\mathrm{peak}})`$ distributions in our model and โlight bulbโ or โfixed Eddington ratioโ models are comparable at and above the break in the quasar luminosity function, and therefore make similar predictions for some observations at these luminosities. However, the faint-end slope has a different physical motivation in our model. Unlike the bright-end slope, which is determined directly by the active final black hole mass function or peak luminosity distribution (in essentially all models of the quasar lifetime), the faint-end slope in our modeling is a consequence of the quasar lifetime as a function of luminosity, and is a prediction of our simulations and modeling almost independent of the underlying faint-end slope of the active black hole mass function or peak luminosity distribution. In Hopkins et al. (2005f) we examine this in more detail, and demonstrate that it predicts well the evolution in the faint-end quasar luminosity function slope with redshift and the observed โluminosity-dependent density evolutionโ in many samples (Page et al., 1997; Miyaji et al., 2000, 2001; La Franca et al., 2002; Cowie et al., 2003; Ueda et al., 2003; Fiore et al., 2003; Hunt et al., 2004; Cirasuolo et al., 2005; Hasinger, Miyaji, & Schmidt, 2005).
Other observational evidence for our picture exists; for example in the observed distribution of Eddington ratios (see ยง 5), the distribution of low-redshift, active black hole masses (see ยง 4.3), and the turnover in the expected distribution of black hole masses in early-type galaxies at $`10^8M_{\mathrm{}}`$ (e.g., Sheth et al., 2003). Total (integrated) quasar lifetimes estimated from observations are inferred to increase with increasing black hole mass as we predict (Yu & Tremaine, 2002), and furthermore, the Eddington ratios of observed quasar samples are seen to increase systematically with redshift, as the sample becomes increasingly dominated by luminosities above the break in the luminosity function (McLure & Dunlop, 2004).
Moreover, observations show that the evolution of the luminosity function with decreasing redshift is driven by a decrease in the characteristic mass scale of actively accreting black holes (e.g., Heckman et al., 2004), which can be explained in our model by the relation of the observed luminosity function to the peak in the distribution of active black hole masses $`\dot{n}(L_{\mathrm{peak}})`$. This observation, however, has caused considerable confusion, as observations of both radio-quiet (Woo & Urry, 2002) and radio-loud (OโDowd et al., 2002) local (low redshift) AGN indicate that nuclear and host luminosities are uncorrelated, implying that nuclear luminosity does not depend on black hole mass (Heckman et al., 2004), and therefore that the primary variable determining the nuclear luminosity is the Eddington ratio, with the luminosity function spanning a broad range in Eddington ratios (Hao et al., 2005). Furthermore, observations show that this is not true of high redshift quasars, as both direct estimates of accretion rates (e.g., Vestergaard, 2004; McLure & Dunlop, 2004) and the fact that their high luminosities would yield unreasonably large black hole masses rule out substantially sub-Eddington accretion rates for most objects. Many previous empirical and semi-analytical models could not simultaneously account for these observations. To explain just the low-redshift observations, such models adopt tunable distributions of Eddington ratios fitted to the data. However, both these observations are consequences of our interpretation of the luminosity function, as observations of local AGN and the low-redshift luminosity function are dominated by quasars below the break in the luminosity function, which are undergoing sub-Eddington growth and span a wide range of Eddington ratios, while observations at high redshift are dominated by bright objects at or above the break in the luminosity function, which are undergoing Eddington-limited (or near Eddington-limited) growth near their peak luminosity (see ยง 5).
(2) The second paradigm shift indicated by our modeling is that quasar obscuration is not a static or quasi-static geometric effect, but is primarily an evolutionary effect. The physical reasoning for this is simple: the massive gas inflows required to fuel quasar activity produce large obscuring columns, and so column densities are correlated with quasar luminosity. The basic picture of buried quasar activity associated with the early growth of supermassive black holes and starburst activity has been proposed previously and studied for some time (e.g., Sanders & Mirabel, 1996; Fabian, 1999), but our modeling allows us to describe the evolution of obscuration in a self-consistent manner, defining obscured and unobscured phases appropriately and identifying dynamical correlations between the column density distribution and instantaneous and peak luminosities.
There is substantial observational support for this picture. Point-like X-ray sources have been observed in many bright sub-millimeter or infrared and starburst sources, with essentially all very luminous infrared galaxies showing evidence of buried quasar activity (e.g., Sanders & Mirabel, 1996; Komossa et al., 2003; Ptak et al., 2003), indicating simultaneous buried black hole growth and star formation at redshifts corresponding to peak quasar activity ($`z1`$) (Alexander et al., 2005a, b). The buried black holes in high-z starbursting galaxies appear to be active but undermassive compared to the quiescent galaxy black hole-stellar mass relation (Borys et al., 2005), implying that they are rapidly growing in the starburst but have not yet reached their final masses, presumably set in the subsequent โblowoutโ phase. Similarly, observations suggest that obscured AGN are significantly more likely to exhibit strong sub-millimeter emission characteristic of star formation, implying both that obscured black hole growth and star formation are correlated and that obscuration mechanisms (responsible for re-radiation in the submm and IR) may be primarily isotropic in at least some cases (e.g., Page et al., 2004; Stevens et al., 2005). Evidence from quasar emission line structure (e.g., Kuraszkiewicz et al., 2000; Tran, 2003), directly related to the inner broad-line region, suggests that isotropic obscuration of quasars can be important, in contradiction to angle-dependent models. Finally, many observations (e.g., Steffen et al., 2003; Ueda et al., 2003; Hasinger, 2004; Grimes, Rawlings, & Willott, 2004; Sazonov & Revnivtsev, 2004; Barger et al., 2005; Simpson, 2005) indicate that the fraction of broad-line or obscured quasars is a function of luminosity, which cannot be accounted for in traditional static โtorusโ models (e.g., Antonucci, 1993) or reproduced even by modified luminosity-dependent torus models (Lawrence, 1991), an observation that is explained by our model (see ยง 4 for a detailed discussion).
Much of the obscuration in our modeling comes from large scales, arising from the inner regions of the host galaxy on scales $`50`$pc or larger. While our resolution limits prevent our ruling out the possibility of gas collapse to a dense, $``$pc scale torus surrounding the black hole, during the peak obscured phases of the final merger, our simulations indicate that these large scales dominate the contribution to the column density, with quite large columns, which should be observationally testable. Indeed, this is suggested by the typical scales of obscuration in starbursting systems (e.g. Soifer et al. 1984a,b; Sanders et al. 1986, 1988a,b; for a review, see e.g. Soifer et al. 1987), given that, as discussed above, the dominant obscured phase of growth is closely associated with a starburst as implied observationally (Alexander et al., 2005a, b; Borys et al., 2005).
Observations of polarized light in intrinsically bright Type II AGN with unobscured luminosities typical of quasars (as opposed to local, dim Seyfert II objects in relaxed hosts) show scattering on large scales $``$kpc, and in some cases obscuration clearly generated over scales extending beyond the host galaxy in the form of distortions, tidal tails, and streams from interactions and major mergers (Zakamska et al., 2004, 2005). The angular structure seen in these observations is consistent with our modeling. Moreover, in optically faint X-ray quasars (e.g. Donley et al., 2005) it appears that obscuration is generated by the host galaxies, and is directly related to host galaxy morphologies and line-of-sight distance through the host. The critical point is that, regardless of the angular structure of obscuration, typical column densities are strongly evolving functions of time, luminosity, and host system properties, and the observed distribution of column densities is dominated by these effects, not by differences in viewing angle across a uniform population. This is the case in our modeling as the lognormal dispersion (across different lines of sight) in column densities is $`\sigma _{N_H}0.4`$ for a given simulation at some instant, whereas typical column densities across simulations, as a function of instantaneous and peak luminosities, span several orders of magnitude from $`N_\mathrm{H}10^{18}10^{26}\mathrm{cm}^2`$.
### 8.2. Specific Predictions of our Model
Our predictions include:
* Quasar Lifetimes: We find that for a particular source, the quasar lifetime depends sensitively on luminosity, with the observed lifetime in addition depending on the observed waveband. Intrinsic quasar lifetimes vary from $`t_Q10^610^8`$ yrs, with observable lifetimes $`10^7`$ yrs in optical B-band (Hopkins et al., 2005a, b), in good agreement with observational estimates (for a review, see Martini, 2004).
* Luminosity Functions: Using a parameterization of the intrinsic distribution of peak luminosities (final quasar black hole masses) at a given redshift, our model of quasar lifetimes allows us to reproduce the observed luminosity function at all luminosities and redshifts $`z=06`$. Although this is an empirical determination of the peak luminosity distribution, it implies a new interpretation of the luminosity function (Hopkins et al., 2005c), which provides a physical basis for the observed โbreakโ corresponding to the peak in the peak luminosity distribution. Moreover, the faint end slope is not determined by our empirical fitting procedure, but instead by the dependence of the quasar lifetime on luminosity, with its value and redshift evolution predicted by our modeling (Hopkins et al., 2005f). The evolution of typical column densities in different stages of merger activity produces a significant population of obscured quasars, accounting for the difference between hard X-ray (e.g., Ueda et al., 2003), soft X-ray (e.g., Miyaji et al., 2001), and optical B-band (e.g., Croom et al., 2004) luminosity functions (ยง 3.2).
* Column Density Distributions: The evolution of the column densities in our simulations reproduces the observed distribution of columns in optically-selected quasar samples, when the appropriate selection criteria are applied (Hopkins et al., 2005b), as well as complete column distributions in hard X-ray selected samples (ยง 3.3). Column density evolution over the course of a merger yields a wider observed distribution of columns than that produced across different viewing angles at a given point in a merger.
* Broad Line Luminosity Function and Fraction: Using our simulations to estimate when quasars will be observable as broad-line objects (either based on the ratio of quasar to host galaxy optical B-band luminosity or the obscuring column density), we reproduce the luminosity function of broad-line quasars in hard X-ray selected samples as well as optical broad-line quasar surveys, and the fraction of broad-line quasars in a given sample as a function of luminosity, to better precision than traditional or luminosity-dependent (but non-dynamical) torus models which are fitted to the data (ยง 4.2). By providing an a priori prediction of the broad-line fraction as a function of luminosity and redshift which depends systematically on the typical quasar host galaxy gas fraction, we propose that observations of the broad line fraction at different redshifts can be used to constrain the gas fraction of quasar hosts and its evolution with redshift.
* Active Black Hole Mass Functions: Using our prescription for deciding when objects will be visible as โbroad-lineโ quasars, we predict the distribution of low-redshift, broad-line and non-broad line active quasar masses, in good agreement with observations from the SDSS, with expected incompleteness in the observed sample at low $`M_{\mathrm{BH}}10^6M_{\mathrm{}}`$ black hole masses (ยง 4.3). This is a new prediction which can be tested in greater detail by future observations, and our calculations allow us to model the differences in active black hole mass functions of the Type I and Type II populations. The width of the expected broad-line black hole mass function depends significantly on the model of quasar lifetimes, enabling such measurements to probe the statistics of quasar evolution.
* Eddington Ratios: We determine Eddington ratio distributions from our simulations, given the peak luminosity distribution implied by the observed quasar luminosity function. The predicted distribution, once the appropriate observed magnitude limit is imposed, agrees well with observations at both low ($`z<0.5`$) and high ($`1.5<z<3.5`$) redshifts (ยง 5). As noted above, our interpretation of the luminosity function explains seemingly contradictory observations of Eddington ratios at different redshifts. There is even a suggestion (Cao & Xu, 2005) that the evolution of quasars seen in our simulations (with bright phases in mergers and extended relaxation after) can account for observations of bimodal Eddington ratio distributions at $`z0`$ (Marchesini et al., 2004), when coupled with an appropriate description of radiatively inefficient accretion phases, although it is possible that many of these low-redshift black holes are not fueled by mergers, especially in e.g. low-luminosity Seyferts.
* Relic Black Hole Mass Function: With our model for quasar lifetimes, the luminosity function at a given redshift implies a birthrate of sources with given peak luminosities, $`\dot{n}(L_{\mathrm{peak}})`$, which translates to a distribution in final black hole masses. Integrating this over redshift, we predict the present-day mass distribution and total mass density of supermassive black holes. They agree well with observational estimates inferred from local populations of galaxy spheroids. In our picture, these spheroids are produced simultaneously with the supermassive black holes they harbor (ยง 6). We demonstrate that the integrated supermassive black hole density, quasar flux density, and number counts in different wavebands can be reconciled with a radiative efficiency $`ฯต_r=0.1`$, satisfying the constraints of counting arguments such as that of Soltan (1982). Further, we show in ยง 2.4 and ยง 4.1 that the corrections to such observational arguments based on optical quasar samples are small (order unity) when we account for the luminosity dependence of quasar lifetimes, despite an extended obscured phase of quasar growth. In other words, although a quasar spends more time obscured than it does as a bright optical source, the total mass growth and radiated energy are dominated by the final โblowoutโ stage visible as a bright optical quasar.
* X-ray Background: The integrated quasar spectrum from our models of quasar lifetimes and column densities as a function of instantaneous and peak luminosities can be combined with the birthrate of quasars with a given peak luminosity to give the integrated cosmic background in any frequency range. We predict both the normalization and shape of the X-ray background from $`1100`$ keV, with our modeling accounting for quasar obscuration as an evolutionary process (with a corresponding population of Compton-thick objects), avoiding any need for arbitrary assumptions about additional obscured populations (ยง 7.2). For any model in which the quasar spectrum depends on luminosity or accretion rate, we demonstrate that a proper modeling of the quasar lifetime is critical to reproducing observed backgrounds.
* Correlation Functions: In Lidz et al. (2005), we predict the quasar correlation function and bias as a function of redshift and luminosity using our model, and compare it to that expected using โlight bulbโ or exponential light curves. As most quasars in our modeling have a characteristic peak luminosity or final black hole mass corresponding to the peak of the $`\dot{n}(L_{\mathrm{peak}})`$ distribution, they reside in hosts of similar mass, and there is little change in bias with luminosity at a given redshift, in contrast to idealized models for the quasar lifetime and luminosity function. Our predicted bias agrees well with the observations of Croom et al. (2005), who also find no evidence for a dependence of the correlation on quasar luminosity at a given redshift, as we expect. In fact, Porciani, Magliocchetti, & Norberg (2004) and Croom et al. (2005) find that their observations can be explained if quasars lie in hosts with a constant characteristic mass $`2\times 10^{12}M_{\mathrm{}}`$ ($`h=0.7`$). If we consider their redshift range $`z12`$, we predict the quasar population will be dominated by sources with $`L_{\mathrm{peak}}=L_{}(z)`$, which given $`M_{\mathrm{BH}}^f(L_{\mathrm{peak}})`$ and using the $`M_{\mathrm{BH}}M_{\mathrm{halo}}`$ relation of Wyithe & Loeb (2003) yields a nearly constant characteristic host halo mass $`12\times 10^{12}M_{\mathrm{}}`$, in good agreement. Similarly, Adelberger & Steidel (2005) find that the quasar-galaxy cross-correlation function does not vary with luminosity, implying with $`90\%`$ confidence that faint and bright quasars reside in halos with similar masses and that fainter AGN are longer lived, strongly disfavoring traditional โlight bulbโ and exponential light curve models. Furthermore, Hennawi et al. (2005) find an order of magnitude excess in quasar clustering at small scales $`40h^1\mathrm{kpc}`$, with the correlation function becoming progressively steeper at sub-Mpc scales, suggesting that quasar activity is triggered by interactions and mergers.
* Host Galaxy Properties: Because black hole growth and spheroid formation occur together in our picture, our modeling allows us to describe relationships between black hole and galaxy properties. For example, we reproduce both the observed $`M_{\mathrm{BH}}\sigma `$ relation (Di Matteo et al., 2005; Robertson et al., 2005b) and the fundamental plane of elliptical galaxies (Robertson et al., in preparation). Since we also reproduce the distribution of relic black holes inferred from the $`z=0`$ distribution of spheroid velocity dispersions or luminosity functions using the observed versions of these relations, our match to these relations indicates that we also reproduce these distributions of host spheroid properties. We consider this in detail in Hopkins et al. (2005e), and find that we are able to account for a wide range of host galaxy properties, including luminosity and mass functions, color-magnitude relations, mass-to-light ratios, and ages as a function of size, mass, and redshift. With our modeling of the quasar lifetime as motivated by our simulations, the evolution and distribution of properties of red-sequence galaxies and the quasar population are shown to be self-consistent, which is not the case for idealized models of quasar evolution.
Aside from an empirical estimate of the distribution of peak quasar luminosities $`\dot{n}(L_{\mathrm{peak}})`$, we determine all of the quantities summarized above self-consistently from the input physics of our simulations, including a physically motivated dynamic accretion and feedback model in which black holes accrete at the Bondi rate determined from the surrounding gas, and $`5\%`$ of the radiant energy couples thermally to that gas. Beyond this, our simulations enable us to calculate the various predictions above a priori, without the need for additional assumptions or tunable parameters.
We compare each of these predictions to those obtained using idealized descriptions of the quasar lifetime, i.e. โlight-bulbโ and exponential light curve (constant Eddington ratio) models, and the column density distribution, i.e. standard and โrecedingโ (luminosity-dependent) torus models. We fit all these (along with our full model) to the observed luminosity function in the same manner (allowing the same degree of freedom to ensure that they all yield the same observed luminosity function), and we fit the free parameters of these tunable models (e.g. typical Eddington ratios and quasar lifetimes for the โlight-bulbโ or exponential models, typical column densities and torus scalings for the torus models) independently to each observation to maximize their ability to reproduce observations. However, we still find better agreement between our model (with no parameters tuned to match observations) and the observations in nearly every case where the tunable phenomenological model is not guaranteed to reproduce the observation by construction. The one exception is the relic supermassive black hole mass function, for which the predictions of our modeling and idealized lifetime models are essentially identical, reflecting the fact that in both cases black hole growth is dominated by bright, optically observable, high Eddington ratio phases.
Moreover, the best-fit parameters for the idealized models, when fitted independently to each observation, are not self-consistent. For example, calculations of the black hole mass function imply high Eddington ratios $`l0.51`$ (e.g., Yu & Tremaine, 2002), and our fit to the active black hole mass function (Heckman et al., 2004) suggests $`l1`$, but the observed distribution of accretion shows a typical $`l0.3`$ (Vestergaard, 2004), and fitting to the broad-line fraction as a function of luminosity with our full obscuration model but these lifetime models implies a lower $`l0.05`$. Likewise, fitting torus models to the X-ray background suggests typical column densities through the torus of $`N_\mathrm{H}10^{25}\mathrm{cm}^2`$ (e.g., Treister & Urry, 2005), while fitting to the observed column density distributions (Treister et al., 2004; Mainieri et al., 2005) suggests equatorial columns $`N_\mathrm{H}10^{24}\mathrm{cm}^2`$. Clearly then, reproducing the observations listed above, and in particular doing so self-consistently, is not implicit in any model which successfully reproduces the quasar luminosity function, even at multiple frequencies.
### 8.3. Further Testable Predictions of our Model
Our model for quasar evolution makes a number of observationally testable predictions:
* Quasar lifetimes are only weakly constrained by observations (e.g. Martini, 2004), but future studies may be able to measure both the lifetime of individual quasars and the statistical lifetimes of quasar populations as a function of luminosity. We describe in detail our predictions for the evolution of individual quasars and quantify their lifetimes in ยง 2, and further predict the distribution of both integrated and differential lifetimes in an observed sample as a function of luminosity. This should provide a basis for comparison with a wide range of observations, with the most important prediction being that the quasar lifetime should increase with decreasing luminosity.
* For a reasonably complete, optically selected sample above some luminosity, the distribution of observed column densities should broaden to both larger and smaller $`N_\mathrm{H}`$ values as the minimum observed luminosity is decreased, as both intrinsically faint periods with low column density and intrinsically bright periods with high column density become observable.
* Similarly, the Eddington ratio distribution should be a function of observed luminosity, with a broad distribution of Eddington ratios down to $`l0.010.1`$ at luminosities well below the break in the observed luminosity function, and a more strongly peaked distribution about $`l0.21`$ for luminosities above the break (Figure 20).
* In our interpretation, the bright and faint ends of the luminosity function correspond statistically to similar mixes of galaxies, but in various stages of evolution; whereas in all other competing scenarios, the quasar luminosity is directly related to the mass of the host galaxy. Therefore, any observational probe that differentiates quasars based on their host galaxy properties such as, for example, the dependence of the clustering of quasars on luminosity, or the host stellar mass and size as a function of luminosity (although we caution that this is somewhat dependent of the modeling of star formation in mergers), can be used to discriminate our picture from older models. We present a detailed prediction of the quasar correlation function based on our modeling for comparison with observations in Lidz et al. (2005).
* Our distribution $`\dot{n}(L_{\mathrm{peak}})`$ directly translates to a black hole merger rate, as a function of mass, in our modeling, allowing a detailed prediction of the gravitational wave signal from black hole mergers as a function of redshift.
* The broad line fraction as a function of luminosity, defined by requiring that โbroad-lineโ objects have an observed B-band luminosity above a fraction $`f_{\mathrm{BL}}`$ of that of their host galaxy, is a prediction of our model quasar and galaxy light curves. However, the uncertainties are large, primarily because different observational samples have varying sensitivity to quasar vs. host galaxy optical light. Furthermore, the host galaxy gas fraction and $`f_{\mathrm{BL}}`$ are degenerate in these predictions โ a well-defined observational sample complete to some $`f_{\mathrm{BL}}`$ can constrain our modeling of quasar fueling and the relation between quasar and host galaxy light curves. In particular, such observations, either by measuring the faint-end shape of the โbroad-lineโ quasar luminosity function or the mean โbroad-lineโ fraction at a given luminosity as a function of redshift, can constrain the gas fractions of quasar host galaxies and their evolution, essentially a free parameter in our empirical modeling.
* We also predict the distribution of active, low-redshift black hole masses in ยง 4. These predictions can be compared to mass functions for active black holes from numerous quasar surveys, which should include improved mass functions of the entire quasar population complete to lower luminosities as well as future mass functions for the population of active broad-line AGN. We provide predictions for the black hole mass function of all active quasars, and for just the โbroad-lineโ population (as a function of the survey selection).
* Because the evolution of spheroids and supermassive black holes is linked in our modeling, with each affecting the evolution of the other, we can also use the distribution of observed quasar properties to predict galaxy properties such as number counts, spheroid masses and luminosities, and colors as a function of redshift. For the calculation and discussion of these predictions, see Hopkins et al. (2005e).
* In our model, the growth of supermassive black holes is dominated by galaxy mergers. Therefore, at any given redshift, the mass (and as a consequence, luminosity) function of galaxy mergers should have a similar shape to our distribution of quasar birthrates, $`\dot{n}(L_{\mathrm{peak}})`$, distinct from the shapes of either the quasar or total galaxy luminosity functions. Indeed, preliminary observational estimates of both the merger luminosity function (e.g., Xu et al., 2004; Conselice et al., 2003; Wolf et al., 2005) and quasar host galaxy luminosity function (Bahcall et al., 1997; Hamilton et al., 2002), primarily at low redshifts, appear be consistent with this expectation. Theoretically, it may be possible to predict the merger luminosity function using either cosmological simulations or semi-analytical models; we discuss this further in ยง 9.
### 8.4. Mock Quasar Catalogs
In principle, our modeling can be used to predict the distributions of quasar luminosities in various wavebands, column densities, active black hole masses, and peak luminosities for a wide range of observational samples, but it is impractical for us to plot predictions of these quantities for all possible sample selection criteria. To enable comparison with a wider range of observations, we have used our modeling and the conditional probability distributions for these quantities from our simulations to generate Monte Carlo realizations of quasar populations, which we provide publicly via ftp<sup>1</sup><sup>1</sup>1 ftp://cfa-ftp.harvard.edu/pub/phopkins/qso\_catalogs/.
At a particular redshift, we use our fitted $`\dot{n}(L_{\mathrm{peak}})`$ distribution and our suite of simulations to generate a random population of mock โquasars.โ We first generate the peak luminosities of each โquasarโ according to the fitted $`\dot{n}(L_{\mathrm{peak}})`$ at that redshift. For each object, we then use the probability of being at a given instantaneous luminosity in simulations with a similar peak luminosity to generate a current bolometric luminosity. In practice, we calculate the $`P(L|L_{\mathrm{peak}})`$ distribution by summing $`w(L_{\mathrm{peak}},L_{\mathrm{peak},\mathrm{i}})\times P(L|L_{\mathrm{peak},\mathrm{i}})`$, where $`L_{\mathrm{peak}}`$ is the mock quasar peak luminosity, $`L_{\mathrm{peak},\mathrm{i}}`$ is the peak luminosity of each simulation and $`w(L_{\mathrm{peak}},L_{\mathrm{peak},\mathrm{i}})`$ is a Gaussian weighting factor ($`\mathrm{exp}(\mathrm{log}^2(L_{\mathrm{peak}}/L_{\mathrm{peak},\mathrm{i}})/2(0.05)^2)`$). Knowing the instantaneous bolometric luminosity $`L`$ and peak luminosity $`L_{\mathrm{peak}}`$, we then follow an identical procedure to determine the joint distribution $`P(X|L,L_{\mathrm{peak}})`$ of each subsequent quantity $`X`$, from simulations with similar $`L`$ and $`L_{\mathrm{peak}}`$. We have compared this with Monte Carlo realizations based on our fitted probability distributions in this paper, and find that essentially identical results are achieved for e.g. the distribution of $`L`$ and $`L_{\mathrm{peak}}`$, and column densities in phases of growth not near peak luminosity. However, this modeling is not identical for e.g. the distribution of Eddington ratios and column densities around $`LL_{\mathrm{peak}}`$, which reflects the fact that our fits to the Eddington ratio distribution (ยง 5) are rough and that our fits to the column density distribution do not apply to the final โblowoutโ phase of quasar evolution (as discussed in detail in ยง 4).
For each mock quasar, we generate a peak luminosity, final (post-merger) black hole mass, instantaneous bolometric luminosity, intrinsic (un-attenuated) B-band ($`\nu L_\nu `$ at $`\nu =4400`$ร
), soft X-ray (0.5-2 keV), and hard X-ray (2-10 keV) luminosity, observed (attenuated using the generated column density and the reddening/dust extinction modeling described in ยง 2.2, with SMC-like reddening curves and extinction following e.g. Pei 1992, Morrison & McCammon 1983) B-band, soft X-ray, and hard X-ray luminosities, column density of neutral hydrogen, column density of neutral+ionized hydrogen, and instantaneous black hole mass. The intrinsic luminosities in each band are calculated using the bolometric corrections described in Marconi et al. (2004), which account for the luminosity dependence of the optical-to-X-ray luminosity ratio $`\alpha _{\mathrm{OX}}`$ (as discussed in ยง 3.2), and then attenuated to give the observed luminosities. We also provide intrinsic and attenuated luminosities in each waveband using the constant bolometric corrections of Elvis et al. (1994), but we caution that these are not calculated in a completely self-consistent manner, as our assumed bolometric luminosity function to which we fit the $`\dot{n}(L_{\mathrm{peak}})`$ distribution is based on using the luminosity-dependent bolometric corrections. We do not directly calculate Eddington ratios, as these are defined differently in many observed samples, but they should be calculable with the given luminosities and black hole masses.
We calculate these quantities for a mock sample of $`10^9`$ quasars at each redshift $`z=0.2,0.5,1,2,\mathrm{and}3`$. Most of these quasars are at luminosities orders of magnitude below those observed, therefore for space considerations and because our predictions become uncertain at low luminosities, we retain only the $`10^6`$ quasars with brightest bolometric luminosities at each redshift. This introduces some uncertainty in our statistics at the lowest luminosities in any given band, but these luminosities are generally still well below those observed in most samples. At any luminosity, but especially at the brightest luminosities, there is also a significant amount of effective โnoiseโ owing to our incomplete sampling of the enormous parameter space of possible mergers, and decreasing total time across simulations spent at large luminosities, which can be estimated from e.g. Figures 8 and 17. Finally, at each redshift, we generate two distributions, reflecting the $`1\sigma `$ range in $`\dot{n}(L_{\mathrm{peak}})`$, and roughly parameterizing the degeneracies in our fit to the observed luminosity functions and uncertainty in the faint-end of $`\dot{n}(L_{\mathrm{peak}})`$ โ โFit 1โ has a lower $`L_{}`$ (lower peak in $`\dot{n}(L_{\mathrm{peak}})`$), with a larger $`\sigma _{}`$ (broader $`\dot{n}(L_{\mathrm{peak}})`$ distribution), and โFit 2โ has a higher $`L_{}`$ and smaller $`\sigma _{}`$ (more narrowly peaked $`\dot{n}(L_{\mathrm{peak}})`$ distribution). We show a few example โquasarsโ from our $`z=0.2`$ mock distribution in Table 1, to demonstrate the format and units used.
### 8.5. Starburst Galaxies
Although we do not yet model the re-radiation of absorbed light by dust or the contribution of stellar light to quasar host IR luminosities, including these in our picture for quasar evolution will enable us to predict luminosity functions in the IR and sub-mm and their evolution with redshift. We can at this point, however, estimate if our model for quasar lifetimes and merger-driven evolution with $`\dot{n}(L_{\mathrm{peak}})`$ is consistent with the observed distribution of ultraluminous infrared galaxies. Naively, we might expect that since the obscured quasar phase has a duration up to $`10`$ times that of the optically observable quasar phase, there should be $`10`$ times as many ULIRGs as bright optical QSOs. But, this neglects the complicated, luminosity dependent nature of quasar lifetimes.
Given that the bright quasars we simulate attain, during their peak growth phase, an intrinsic luminosity comparable to that of the host starburst, and that this period of peak growth has a similar duration to the starburst phase (see Figure 13 and Di Matteo et al., 2005; Springel et al., 2005b), we can estimate (roughly) the ULIRG bolometric luminosity function from our bolometric quasar luminosity function. Thus, the more accurate comparison to the ULIRG luminosity function is with the hard X-ray quasar luminosity function, as this recovers (and at some luminosities can be dominated by) โburiedโ quasars in starburst phases. This is only applicable above the break in the luminosity function, where quasars are undergoing peak quasar growth. Below the break, quasars are, on average, sub-Eddington and can have luminosities well below that of their star-forming hosts (see Figure 13), so we expect our quasar luminosity function to be significantly shallower than the ULIRG luminosity function at these luminosities. Note also that this does not imply that ULIRGs are all AGN-dominated, as the starburst and peak AGN activity can be (and generally are) somewhat offset, but only says that the lifetime curves at the bright end should be similar.
Considering the luminosity function at $`z=0.15`$, then, we expect ULIRG densities $`d\mathrm{\Phi }/\mathrm{d}M_{\mathrm{bol}}3\times 10^7\mathrm{and}9\times 10^8\mathrm{Mpc}^3\mathrm{mag}^1`$ at $`L1.6\times 10^{12}L_{\mathrm{}}`$ and $`2.5\times 10^{12}L_{\mathrm{}}`$, respectively. These estimates are consistent with the observed density in the IRAS 1 Jy Survey (Kim & Sanders, 1998) at a mean redshift $`z=0.15`$, with $`\mathrm{d}\mathrm{\Phi }/\mathrm{d}M_{\mathrm{bol}}5\times 10^7,7\times 10^8\mathrm{Mpc}^3\mathrm{mag}^1`$ (rescaled to our cosmology), and as expected, our quasar luminosity function slope becomes significantly shallower than the observed 1 Jy survey luminosity function slope below $`L10^{11}10^{12}L_{\mathrm{}}`$, roughly the break luminosity of the luminosity function. We predict these densities to change with redshift according to the evolution of $`\dot{n}(L_{\mathrm{peak}})`$, decreasing by a factor $`1.5`$ at $`z=0.04`$, in good agreement with the evolution of IRAS ULIRG luminosity functions (Kim & Sanders, 1998). Likewise, at $`z13`$, we predict a mean space density $`\mathrm{\Phi }(L>10^{11}L_{\mathrm{}})13\times 10^5\mathrm{Mpc}^3`$, in agreement with the $`5\times 10^5\mathrm{Mpc}^3`$ density of such sources expected to reproduce the observed cumulative source density $`4\times 10^4\mathrm{deg}^2`$ of 1 mJy $`850\mu \mathrm{m}`$ SCUBA sources (Barger et al., 1999). Furthermore, our prediction of the fraction of buried AGN and its evolution with redshift agrees well with determinations from X-ray samples (Barger et al., 2005) and recent Spitzer results in the mid and near-infrared at $`z2`$ (Martinez-Sansigre et al., 2005).
### 8.6. AGN not Triggered by Mergers
Some low redshift quasars (e.g. Bahcall et al. 1996) and many nearby, low-luminosity Seyferts appear to reside in ordinary, relatively undisturbed galaxies. Our picture for quasar evolution does not immediately account for these objects because we suppose that nuclear activity is mainly triggered by tidal torques during a merger.
This work is primarily concerned with the origin of the majority of the mass in spheroids and supermassive black holes, and as a consequence, the relation of this to the abundance and evolution of quasars and the cosmic X-ray background. Based on our present analysis, we believe that a merger-driven picture can account for the main part of each of these, and, as described earlier, that the most relevant phase in the history of the Universe to these phenomena appears to have been at moderate redshifts, $`z2.5`$ to $`z0.5`$.
Our model does not exclude other mechanisms for triggering AGN and it is likely that a variety of stochastic or continuous processes are relevant to nuclear activity in undisturbed disks and residual low-level accretion in relaxed systems. This is not contrary to our picture because most of the total black hole mass density in the Universe is in spheroid-dominated systems. The principal requirement of our model is that AGN activity in undisturbed galaxies should not contribute a large fraction of the black hole mass density in the Universe, to avoid spoiling tight correlations between the black hole and host galaxy properties and producing too large a present-day black hole mass density in violation of the Soltan (1982) constraint.
For example, if a molecular cloud passed through the center of our Galaxy near Sgr A, it is possible that the Milky Way would resemble a Seyfert for some period of time. Alternatively, it has long been recognized that mass loss from normal stellar evolution of bulge stars or stellar clusters near the centers of galaxies can provide a continuous supply of fuel for low-level accretion (e.g., McMillan et al., 1981; MacDonald & Bailey, 1981; Shull, 1983). Typical galactic stellar mass loss rates ($`\dot{M}1M_{\mathrm{}}\mathrm{yr}^1(10^{11}M_{\mathrm{}})^1`$) yield Bondi-Hoyle accretion rates $`10^510^4`$ of Eddington in relaxed, dynamically hot systems; and mass loss rates from O and W-R stars ($`\dot{M}10^6M_{\mathrm{}}\mathrm{yr}^1(10M_{\mathrm{}})^1`$) in young, dense star clusters near the centers of galaxies with sufficient cold gas for continued star formation can yield rates as high as $`10^2`$ of Eddington.
Even though these fueling mechanisms do not involve mergers, the scenario we have discussed might still be relevant to the origin of these black holes. Of course, the black holes and spheroids in disk-dominated systems may have produced in a manner that did not involve mergers. Alternatively, most of the black hole mass in these objects (which is small compared to that in spheroid-dominated galaxies) could have been assembled long ago in mergers with bright quasar phases and then these โdeadโ quasars are resurrected sporadically by other fueling mechanisms.
Independent of how these black holes were formed, elements of our modeling may still account for certain observed properties of Seyferts. The observed Seyfert luminosity function appears to join smoothly onto the quasar luminosity function (Hao et al., 2005). It is not obvious that this would be the case if the two types of objects are produced by entirely distinct mechanisms. In addressing this, it is useful to separate the process by which gas is delivered to the black hole from the subsequent evolution that determines the observed activity. In our picture, gas is delivered to the black hole by gravitational torques during a merger, but other mechanisms, like bar-induced fueling may be important for objects such as Seyferts. Regardless, the induced activity may be generic, if black hole growth is self-regulated in the way we describe it in our simulations.
In Hopkins et al. (2005f) we show that the faint end slope of the quasar luminosity function in our model is partly determined by the time dependence of the โblowoutโ phase of black hole growth. We derive an analytical model for this using a Sedov-Taylor type analysis and show that the impact of this feedback depends on the mass of the host. This analysis does not depend on the fueling mechanism, only on the subsequent evolution. If this self-regulated growth applies to Seyferts as well (for example if Seyfert growth is regulated by a balance between accretion feedback and the spheroid potential, as expected if these objects obey a similar $`M_{\mathrm{BH}}\sigma `$ relation), we would expect the Seyfert luminosity function to smoothly join onto the quasar one, even if the fuel is delivered in a different manner.
## 9. Conclusions
We have studied the evolution of quasars in simulations of galaxy mergers spanning a wide region of parameter space. In agreement with earlier work (Hopkins et al., 2005a), we find that the lifetime of a particular source depends on luminosity and increases at lower luminosities, and that quasar obscuration is time-dependent. Our new, large set of simulations shows that the lifetime and obscuration can be expressed in terms of the instantaneous and peak luminosities of a quasar and that these descriptions are robust, with no systematic dependence on simulation parameters. We have combined these results with a semi-empirical method to describe the cosmological distribution of quasar properties, allowing us to predict a large number of observables as a function of e.g. luminosity and redshift. This approach also makes it possible to compare our picture to simpler models for quasar lifetimes and obscuration.
In the model we examine, quasars are triggered by mergers of gas-rich galaxies, which produce inflows of gas through gravitational torquing, fueling starbursts and rapid black hole growth. The large gas densities obscure the central black hole at optical wavelengths until feedback energy from the growth of the black hole ejects gas and rapidly slows further accretion (โblowoutโ). Quasar lifetimes and light curves are non-trivial, with strong accretion activity during first passage of the merging galaxies and extended quiescent (sub-Eddington) phases leading into and out of the phase of peak quasar activity associated with the final merger. The โblowoutโ phase in which the quasar is visible as a bright, near-Eddington optical source has a lifetime related to the dynamical time in the inner regions of the merging galaxies, which characterizes the timescale over which obscuring gas and dust are expelled, but the quasar spends a longer time at lower luminosities before and after this stage. These evolutionary processes have important consequences which cannot be captured in models of pure exponential or โon/offโ quasar growth.
Our work emphasizes several goals for quasar and galaxy observations and theory. Observationally, it is important to constrain the faint end of the peak luminosity distribution; i.e. the low-mass active black hole distribution. Unfortunately, our modeling of quasar lifetimes implies that the faint-end quasar luminosity function is dominated by quasars with peak luminosities around the break in the luminosity function, and can provide only weak constraints on the faint-end $`L_{\mathrm{peak}}`$ distribution. However, there is still hope, as for example broad-line quasar activity is more closely associated with near-peak luminosities, and thus probing the faint-end of broad-line luminosity functions may in particular improve the estimates. Moreover, studies of the black hole mass distribution (or the distribution of galaxy spheroids) as a function of redshift, extending to small spheroid masses/velocity dispersions probes the faint end of $`\dot{n}(L_{\mathrm{peak}})`$. Other techniques, such as studies of faint radio sources at high redshift (Haiman, Quataert, & Bower, 2004) can similarly constrain these populations. Furthermore, the calculations in this paper can be combined to better determine $`\dot{n}(L_{\mathrm{peak}})`$, as, given a model for the quasar lifetime and obscuration, they all derive from this fundamental quantity. Additional observational tests of the modeling we have presented will provide an important means of constraining models for AGN accretion and feedback; for example, the faint-end slope of the quasar lifetime depends on how the โblowoutโ phase occurs and could provide a sensitive probe of feedback models, enabling the adoption of more realistic and sophisticated feedback prescriptions than we have thus far employed. Of course, improved constraints on the luminosity function at all luminosities at high redshift remains a valuable means of testing theories of quasar evolution.
Our simulations are based on isolated galaxy mergers, and thus do not provide a cosmological prediction for the distribution of peak luminosities $`\dot{n}(L_{\mathrm{peak}})`$, merger rates, or mass functions - we instead have adopted a semi-empirical model, in which we use our modeling of quasar evolution to determine these distributions from the observed luminosity function. While this allows us to predict a large number of observables and to demonstrate that a wide range of quasar and galaxy properties are self-consistent in a model of merger-driven quasar activity with realistic quasar lifetimes, future theoretical work in these areas should predict the distribution of peak luminosities $`\dot{n}(L_{\mathrm{peak}})`$ and its evolution with redshift. These quantities are to be distinguished from the distribution of observed luminosities, as the two are not trivially related in our model or any other with a non-trivial quasar lifetime.
Although the quasar birthrate as a function of peak luminosity will be, in general, a complicated function of galaxy merger rates, gas fractions, morphologies, and other factors, we have parameterized it for comparison with the results of future cosmological simulations and semi-analytical models. This distribution is particularly valuable as a theoretical quantity because it is more directly related to physical galaxy properties than even the complete (intrinsic) luminosity function, and additionally because theoretical modeling which successfully reproduces this $`\dot{n}(L_{\mathrm{peak}})`$ distribution is guaranteed to reproduce the large number of observable quantities we have discussed in detail in this work. We cannot determine the cosmological context in our detailed simulations of the relatively small-scale physics of galaxy mergers, and conversely, cosmological simulations and semi-analytical models cannot resolve the detailed physics driving quasar activity in mergers. However, our determination of quasar evolution as a function of peak luminosity or final black hole mass can be grafted onto these cosmological models to greatly increase the effective dynamic range of such calculations. Combined with our modeling, this would remove the one significant empirical element we have adopted, and allow for a complete prediction of the above quantities from a single theoretical framework.
In these efforts, we emphasize that the mergers relevant to our picture are of a specific type. First, the merging galaxies must contain a supply of cold gas in a rotationally supported disk. Hot, diffuse gas, as in the halos of elliptical galaxies, will not be subject to the gravitational torques which drive gas into galaxy centers and fuel black hole growth. Clearly, gas-poor mergers are also not important for this process. Second, the mergers will likely involve galaxies of comparable, although not necessarily equal, mass, so that the gravitational torques excited are strong enough and penetrate deep enough into galaxy centers to drive substantial inflows of gas. The precise requirement for the mass ratio is somewhat ill-defined because it also depends on the orbit geometry, but mergers with a mass ratio larger than $`10:1`$ are probably not generally important to our model. Simulations of minor mergers involving galaxies with mass ratios $`10:1`$ (e.g. Hernquist 1989; Hernquist & Mihos 1995) have shown that for particular orbital geometries, these events can produce starbursts similar to those in major mergers, leaving behind disturbed remnants with dynamically heated disks (e.g. Quinn et al. 1993; Mihos et al. 1995; Walker et al. 1996). It is of interest to establish whether black hole growth can also be triggered in minor mergers, as these events may be relevant to weak AGN activity like that in some Seyfert galaxies or LINERs.
In summary, the work presented here supports the conjecture that many aspects of galaxy formation and evolution can be understood in terms of the โcosmic cycleโ in Figure 1. To be sure, much of what is summarized in Figure 1 has been proposed elsewhere, either in the context of observations or theoretical models. Our modeling of galaxy formation and evolution emphasizes the possibility that supermassive black holes could be responsible for much of what goes on in shaping galaxies, rather than being bystanders, closing the loop in Figure 1. In this sense, black holes may be the โprime moversโ driving galaxy evolution, as has been proposed earlier for extragalactic radio sources (e.g. Begelman, Blandford & Rees 1984; Rees 1984). It may seem counterintuitive that compact objects with masses much smaller than those of galaxies could have such an impact, but it is precisely the concentrated nature of matter in black holes that makes this idea plausible.
Consider a black hole of mass $`M_{\mathrm{BH}}`$ at the center of a spherical galaxy of mass $`M_{\mathrm{sph}}`$ with a characteristic velocity dispersion $`\sigma `$. The energy available to affect the galaxy through the growth of the black hole will be some small fraction of its rest-mass, $`E_{\mathrm{feed}}ฯต_fM_{\mathrm{BH}}c^2`$. This can be compared with the binding energy of the galaxy, $`E_{\mathrm{bind}}M_{\mathrm{sph}}\sigma ^2`$. Observations indicate that $`M_{\mathrm{BH}}`$ and $`M_{\mathrm{sph}}`$ are correlated and that, roughly $`M_{\mathrm{BH}}(0.0020.005)M_{\mathrm{sph}}`$ (Magorrian et al. 1998; Marconi & Hunt 2003). Therefore, the ratio of the feedback energy to the binding energy of the galaxy is $`E_{\mathrm{feed}}/E_{\mathrm{sph}}>10ฯต_{f,2}\sigma _{300}^2`$, for an assumed efficiency of 1%, $`ฯต_{f,2}ฯต/0.01`$ and scaling the velocity dispersion to $`\sigma _{300}\sigma /300`$ km/sec, as for relatively massive galaxies. This result demonstrates that the supermassive black holes in the centers of spheroidal galaxies are by far the largest supply of potential energy in these objects, exceeding even the galaxy binding energy. When viewed in this way, if even a small fraction of the black hole radiant energy can couple to the surrounding ISM, then black hole growth is not an implausible mechanism for regulating galaxy formation and evolution; in fact, it appears almost inevitable that it should play this role.
We thank our referee, David Weinberg, for many comments and suggestions that greatly improved this paper. We thank Paul Martini, for helpful discussion, and Gordon Richards and Alessandro Marconi, for generously providing data for observational comparisons. This work was supported in part by NSF grants ACI 96-19019, AST 00-71019, AST 02-06299, and AST 03-07690, and NASA ATP grants NAG5-12140, NAG5-13292, and NAG5-13381. The simulations were performed at the Center for Parallel Astrophysical Computing at the Harvard-Smithsonian Center for Astrophysics.
|
warning/0506/nlin0506058.html
|
ar5iv
|
text
|
# Phenomenology of Wall Bounded Newtonian Turbulence
## Introduction
The tremendous amount of work devoted to understanding the apparent experimental deviations from the classical phenomenology of homogeneous and isotropic turbulence MY ; Fri tends to obscure the fact that in many respects this phenomenology is almost right on the mark. Starting with the basic ideas of Richardson and Kolmogorov, and continuing with a large number of ingenious closures, one can offer a reasonable set of predictions regarding the statistical properties of the highly complex phenomenon of homogenous and isotropic turbulence. Thus one predicts the range of scales for which viscous effects are negligible (the so called โinertial rangeโ of turbulence), the cross over scale below which dissipative effects are crucial (also known as the โKolmogorov scaleโ), the exact form of the third order structure function $`S_3(R)`$ (third moment of the longitudinal velocity difference across a scale $`R`$), including numerical pre-factors, and an approximate form of structure function of other orders $`S_n(R)`$ (predicted to scale like $`R^{n/3}`$ but showing deviations in the scaling exponents which grow with the order, giving rise to much of the theoretical work alluded to above). In particular much effort was devoted to calculate the so called โKolmogorov constantโ $`C_2`$ which is the pre-factor of the second order structure function, with closure approximations (see, e.g. Refs. 66Kra ; 86YO ) coming reasonably close to its experimental estimate. Notwithstanding the deviations from the classical phenomenology, one can state that it provides a reasonable first order estimate on many non-trivial aspects of homogeneous and isotropic turbulence. In contrast, the phenomenological theory of wall bounded turbulence is less advanced. In reality most turbulent flows are bounded by one or more solid surfaces, making wall bounded turbulence a problem of paramount importance. Evidently, a huge amount of literature had dealt with problem, with much ingenuity and considerable success 00Pope . In particular one refers to von-Kรกrmรกnโs log-law of the wall which describes the profile of the mean velocity as a function of the distance from the wall. It appears however that the literature lacks an analytically tractable model of wall bounded flows whose predictions can be trusted at a level comparable to the phenomenological theory of homogeneous turbulence.
In this paper we attempt to reduce this gap and offer as simple as possible but still realistic model (a โminimal modelโ) for the viscous-to-turbulent flow in the entire region from the very surface through the logarithmic boundary layer (hereafter, *log-layer*) up to the upper boundary of outer turbulent region. Our final goal is to create clear physical grounds for improved description of the flow-surface interactions in numerical fluid-mechanics models (both engineering and geophysical) where the viscous and the buffer layers cannot be resoled and should be parameterized. To attain these ends we need to obtain analytical solutions (numerical solution would be of no use), which calls for simplification of the governing equations. Accordingly, our strategy is a pragmatic, task-dependent simplification and restrictions. In particular we concentrate on descriptions of the profile of mean flow and the statistics of turbulence on the level of simultaneous, one-point, second-order velocity correlation functions.In other words, the objects that we are after are the entries of the Reynolds-stress tensor as a function of the distance from the wall. The model will be presented for plain geometry; this geometry is relevant for a wide variety of turbulent flows, like channel and plain Couette flows, fluid flows over inclined planes under gravity (modelling river flows), atmospheric turbulent boundary layers over flat planes and, in the limit of large Reynolds numbers, many other turbulent flows, including pipe, circular Couette flows, *etc*.
Suggested in this paper phenomenological theory of wall-bounded flows is based on standard ideas 00Pope ; nevertheless we develop the theory slightly further than anything that exists currently in the literature. In our study we will stress analytical tractability; in other words, we will introduce approximations in order to achieve a model whose properties and predictions can be understood without resort to numerical calculations. Nevertheless we will show that the model appears very dependable in the sense that its predictions check very well in comparison to direct numerical simulations, including some rather non-trivial predictions that are corroborated only by very recent simulations and experiments (which only now reach the sufficient accuracy and high Reynolds numbers).
We should notice, that considering the mean velocity and the second-order statistics in these (neutrally stratified) flows, we neglect some mechanisms and features although present but not essential in the problem under consideration. However, proceeding further, in particular, to account for the density or temperature stratification (out of the scope of this paper), we quite probably will be forced to rule out of some simplifications acceptable in the first task and to account, for example, for a spacial energy flux and even for coherent structures.
In Sec. I we formulate a model which is a version of the โalgebraic Reynolds-stress modelโ 00Pope . In Sec. I.1 we introduce notations and recall the equations describing the mechanical balance; in Sec. I.2 we state the assumptions and detail the approximations used in the context of the balance equations for the components of the Reynolds stress tensor $`W_{ij}`$. The result of these considerations is a set of 5 equations for the mean shear $`S`$ and $`W_{ij}`$ which is described in Sec. I.3. For actual calculations this set of equations is still too rich since it contains 12 adjustable parameter. Eight of these parameters control the nonlinear behavior of the system in the outer layer and four additional parameters govern the energy dissipation in the viscous sub-layer. Clearly, further reduction of the model is called for. This is accomplished in Sec. II. First, in Sec. II.1 we consider the full 12 parametrical solution of the model, and present a comparison with experimental observations in Sec. II.1.4. This comparison indicates that an adequate description of the entire turbulent boundary layer phenomenology can be achieved with only four parameters instead of twelve. We refer to the four-parameter model as the โminimal modelโ. In Sec. II.2 we reap the benefit of the minimal model: we find simple and physically transparent Eqs. (37) for the profiles of the Reynolds stress tensor $`W_{ij}(y)`$ and the mean shear $`S(y)`$ ($`y`$ is the distance from the wall) in terms of the root-mean-square turbulent velocity $`v\sqrt{W_{ii}}`$. Unfortunately, the equation for the $`v(y)`$ profile is quite cumbersome and cannot be solved analytically. Nevertheless we employ an effective iteration procedure that allows reaching highly accurate solutions with one or at most two iteration steps.
Section III is devoted to a comparison of the predictions of the minimal model with results of experiments and direct numerical simulations . In particular, in Sec. III.2 we show that the model describes the mean velocity profile in a channel flow with $`1\%`$-accuracy almost everywhere. Only in the core the model fails to describe so-called โvelocity defectโ (the upward deviation from the log-law) which is observed near the mid-channel (of about 5-6 units in $`V^+`$, independent on Reynolds number). For our purposes this mismatch in not essential. In Sec. III.3 we show that the minimal model provides a good qualitative description of kinetic energy profile, including position, amplitude and width of the peak of the kinetic energy in the buffer layer. In Sec. III.4 we show that with the same set of four parameters the model offers also a good qualitative description of the Reynolds stress profiles and the profiles of โpartialโ kinetic energies (in the streamwise, wall-normal and cross-stream directions) almost in the entire channel. The final Sec. IV presents a short summary of our results, including a discussion of the limitations of the minimal model. Possible improvements of the suggested model will have to start by addressing these limitations.
## I Formulation of the model
Our starting point is the standard Reynolds decomposion 00Pope of the fluid velocity $`๐ผ(๐,t)`$ into its average (over time) $`๐ฝ`$ and the fluctuating components $`๐`$. In wall-bounded planar geometry the mean velocity is oriented in the (stream-wise) $`\widehat{๐ฑ}`$ direction, depending on the vertical (wall-normal) coordinate $`y`$ only:
$$๐ผ(๐,t)=๐ฝ(y)+๐(๐,t),๐ฝ(y)๐ผ(๐,t)=\widehat{๐ฑ}V(y).$$
(1)
The mean velocity and the fluctuating parts are used to construct the objects of the theory which are the components of the Reynolds stress tensor $`๐พ(y)`$ and the mean shear:
$$W_{ij}(y)u_iu_j,S(y)\frac{dV(y)}{dy}.$$
(2)
We note that in previous applications 04LPPT ; 04DCLPPT ; 04BLPT ; 05LPPT ; 05LPPTa ; 04BDLPT ; 04LPT ; 05BCLLP ; 05BDLP we have employed a model in which only the trace of $`๐พ(y)`$ and its $`xy`$ component were kept in a simplified description. For the present purposes we consider all the component of this tensor, paying a price of having more equations to balance, but reaping the benefit of a significantly improved phenomenology. We discuss now the equations relating these variables to each other.
### I.1 Equation for the mechanical balance
The first equation relates the Reynolds stress $`W_{xy}`$ to the mean shear; it describes the balance of the flux of mechanical momentum, it follows as an exact result from Navier-Stokes equations and has the familiar form:
$$W_{xy}(y)+\nu _0S(y)=P(y).$$
(3a)
$`W_{xy}`$ on the left hand side (LHS) is the turbulent (reversible) contribution to the momentum flux whereas $`\nu _0S(y)`$ is the viscous (dissipative) contribution to the momentum flux. The RHS is the momentum flux, which may have different origin. For example, in a channel or pipe flow $`P(y)`$ is generated by the pressure head. In the channel flow with the pressure gradient $`p^{}=dp/dx`$,
$$P(y)=p^{}(Ly),$$
(3b)
where $`L`$ is the half width of the channel. In the pipe flow $`P(y)`$ is given by the same Eq. (3b) with $`L`$ being a half of the pipe radius. In a water flow over incline in the gravity field $`p^{}`$ should be replaced by $`g\mathrm{sin}a`$, where $`g`$ is the gravity acceleration and $`\alpha `$ is the inclination angle. For Re$`{}_{\lambda }{}^{}1`$, near the wall one can neglect the $`y`$ dependence of $`P(y)`$, replacing $`P(y)`$ by its value at the wall: $`P(y)P_0P(0)`$. Here the so-called โwall-basedโ Reynolds number Re<sub>ฮป</sub> is defined by:
$$\text{Re}_\lambda \frac{L\sqrt{P(0)}}{\nu _0}.$$
(4)
### I.2 Balance of the Reynolds tensor
The next set of equations relates the various components of the Reynolds tensor, $`W_{ij}(y)`$ defined by Eq. (2). In contrast to Eq. (3a) this set of equations is only partially exact. We need to model some of the terms, as explained below. We start from the Navier-Stokes equations and write the following set of equations:
$$\frac{dW_{ij}}{dt}+ฯต_{ij}+I_{ij}=S(W_{iy}\delta _{jx}+W_{jy}\delta _{ix}).$$
(5a)
The RHS of these equations is exact, describing the production term in the equations for $`W_{ix}=W_{xi}`$ which is caused by the existence of a mean shear. On the LHS of Eq. (5a)
$$ฯต_{ij}=2\nu _0\frac{u_i}{x_k}\frac{u_j}{x_k}$$
(5b)
is the exact term presenting the viscous energy dissipation. The problem is that $`ฯต_{ij}`$ involve new object, which requires evaluation via $`W_{ij}`$. This can be easily done in regions where the velocity field is rather smooth, and in particular in the viscous sub-layer, the velocity gradient exists and thus the spatial derivatives in Eq. (5b) are estimated using a characteristic length which is the distance from the wall $`y`$. In order to write equalities we employ the dimensionless constants $`a_{ij}1`$:
$$ฯต_{ij}ฯต_{ij}^{\text{vis}}=\gamma _{ij}^{\text{vis}}W_{ij},\gamma _{ij}^{\text{vis}}\nu _0\left(\frac{a_{ij}}{y}\right)^2.$$
(5c)
In general the constants $`a_{ij}`$ are different for every $`i,j`$.
In the buffer sublayer and in the log-layer the energy cascades down the scales until it dissipates at the Kolmogorov (inner) scale that is much smaller than the distance $`y`$ from the wall. Therefore the main contribution to the dissipation $`ฯต_{ij}`$ from all scales smaller than $`y`$ is due to the energy flux, i.e. has a *nonlinear character*. Due to the asymptotical isotropy of fine-scale turbulence, the nonlinear contribution should be diagonal in $`i`$, $`j`$ (see, e.g. 00Pope ):
$$ฯต_{ij}ฯต_{ij}^{\text{nl}}=\gamma \frac{W}{3}\delta _{ij},W\text{Tr}\{๐พ\},$$
(6a)
where prefactor $`\frac{1}{3}`$ is introduced to simplify equations below. The characteristic โnonlinear flux frequencyโ $`\gamma ^{\text{nl}}`$, can be estimated using a standard Kolmogorov-41 dimensional analysis:
$$\gamma (y)=\frac{b}{y}\sqrt{W(y)},,$$
(6b)
again with some constants $`b1`$. The โouter scaleโ of turbulence is estimated in Eq. (6b) by the only available characteristic length, $`y`$, the distance to the wall.
As one sees from Eq. (6), the dissipation of particular component of the Reynolds-stress tensor, say $`W_{xx}`$, depends not only on $`W_{xx}`$ itself, but also on other components, $`W_{yy}`$ and $`W_{zz}`$ in our case. It means that $`ฯต_{ij}`$, given by Eq. (6a) leads, in the framework of Eq. (5) not only to the dissipation of total energy, but also to its redistribution between different components of $`W_{ii}`$. In order to separate these effects we divide $`ฯต_{ij}`$ into two parts as follows:
$`ฯต_{ij}^{\text{nl}}`$ $`=`$ $`ฯต_{ij}^{\text{nl,1}}+ฯต_{ij}^{\text{nl,2}},`$ (7a)
$`ฯต_{ij}^{\text{nl,1}}`$ $`=`$ $`\gamma W_{ij}\delta _{ij},`$ (7b)
$`ฯต_{ij}^{\text{nl,2}}`$ $`=`$ $`\gamma \left(W_{ij}\delta _{ij}{\displaystyle \frac{W}{3}}\right)\delta _{ij}.`$ (7c)
Clearly, $`ฯต_{ij}^{\text{nl,1}}`$ describes the damping of each component $`W_{ii}`$ separately, without changing of their ratios, while the traceless part, $`ฯต_{ij}^{\text{nl,1}}`$, does not contribute to the dissipation of total energy and leads only to redistribution of energy between components of the Reynolds-stress tensor. This contribution we will include into the โreturn to isotropyโ term (14b), that will be discussed below.
Actually, we presented $`ฯต_{ij}`$ as the sum
$$ฯต_{ij}=ฯต_{ij}^{\text{dis}}+ฯต_{ij}^{\text{nl,2}},$$
(8)
in which for the total energy dissipation is responsible only first term in the RHS. In the buffer layer both contributions to $`ฯต_{ij}^{\text{dis}}`$, the viscous dissipation, $`ฯต_{ij}^{\text{vis}}`$, and the nonlinear one, $`ฯต_{ij}^{\text{nl,1}}`$ are important. Their relative role depending on the turbulent statistics. We will employ two simple interpolation formulas which lead to two versions of the minimal model:
$`ฯต_{ij}^{\text{dis}}`$ $`=`$ $`\mathrm{\Gamma }_{ij}W_{ij},`$ (9)
$`\mathrm{\Gamma }_{ij}(y)`$ $`=`$ $`\gamma _{ij}^{\text{vis}}(y)+\gamma (y)\delta _{ij},\text{โsumโ},`$ (10)
$`\mathrm{\Gamma }_{ij}(y)`$ $`=`$ $`\sqrt{\gamma _{ij,\mathrm{vis}}^2(y)+\gamma ^2(y)\delta _{ij}},\text{โrootโ}.`$ (11)
The versions of the resulting model will be referred to as the โsumโ and the โrootโ versions correspondingly. A-priori there is no reason to prefer one or the other, and we leave the choice for later, after the comparisons with the data.
The term $`I_{ij}`$ in Eq. (5a) is caused by the pressure-strain correlations:
$$I_{ij}=\frac{1}{\rho }_0p\left(\frac{u_i}{x_j}+\frac{u_j}{x_i}\right),$$
(12)
and is known in the literature as the โReturn to Isotropyโ 00Pope . Due to incompressibility constraint $`I_{ij}`$ is traceless tensor and therefore does not contribute to the total energy balance, leading only to redistribution of partial kinetic energy between different vectorial components. Also, this term does not exist in isotropic turbulence where $`W_{ij}=\frac{1}{3}W\delta _{ij}`$. We adopt the simplest linear Rota approximation for the โReturn to Isotropyโ term 00Pope , using yet another different characteristic frequencies $`\overline{\gamma }_{ij}`$, estimated as follows:
$`I_{ij}`$ $`=`$ $`\overline{\gamma }_{ij}\left(3W_{ij}\delta _{ij}W\right),`$ (13a)
$`\overline{\gamma }_{ij}(y)`$ $``$ $`\overline{b}_{ij}{\displaystyle \frac{\sqrt{W(y)}}{y}},\overline{b}_{ij}1.`$ (13b)
One sees that $`I_{ij}`$ has precisely the same structure as $`ฯต_{ij}^{\text{nl\hspace{0.17em},2}}`$, introduced by Eq. (7c). Therefore it is convenient to treat these contributions together, introducing
$`\stackrel{~}{I}_{ij}`$ $``$ $`I_{ij}+ฯต_{ij}^{\text{nl\hspace{0.17em},2}},`$ (14a)
$`\stackrel{~}{I}_{ij}`$ $``$ $`\stackrel{~}{\gamma }_{ij}\left(3W_{ij}\delta _{ij}W\right),`$ (14b)
$`\stackrel{~}{\gamma }_{ij}(y)`$ $``$ $`\stackrel{~}{b}_{ij}{\displaystyle \frac{\sqrt{W(y)}}{y}},\stackrel{~}{b}_{ij}=\overline{b}_{ij}{\displaystyle \frac{b}{3}}\delta _{ij}.`$ (14c)
Recall that tensor $`\stackrel{~}{I}_{ij}`$ must have zero trace for any values of $`W_{ij}`$. This is possible only if $`\stackrel{~}{b}_{xx}=\stackrel{~}{b}_{yy}=\stackrel{~}{b}_{zz}\stackrel{~}{b}_\mathrm{d},\stackrel{~}{b}_{xx}\stackrel{~}{b}`$, and consequently
$$\stackrel{~}{\gamma }_{xx}=\stackrel{~}{\gamma }_{yy}=\stackrel{~}{\gamma }_{zz}\stackrel{~}{\gamma }_\mathrm{d},\stackrel{~}{\gamma }_{xy}\stackrel{~}{\gamma }.$$
(15)
Thus, representation (14b) involves only two free parameters $`\stackrel{~}{b}_\mathrm{d}`$ and $`\stackrel{~}{b}`$.
Equations (5) for $`W_{ij}`$ with $`ij=xx,yy,zz`$ and $`xy`$ involve 7 constants $`a_{ij}`$, $`b`$, $`\stackrel{~}{b}_\mathrm{d}`$ and $`\stackrel{~}{b}`$. Our goal is to formulate the simplest possible model, with a minimal number of adjustable constants. The strategy will be now to use experimental and simulational data, coupled with reasonable physical considerations, to reduce the number of parameters to four, each of which being responsible for a separate fragment of the underlying physics.
We should stress that we neglect in Eq. (5a) the spatial energy transport term $`ฯต_{\mathrm{tr}}`$, caused by the tripple-velocity correlations, pressure-velocity correlations and by the viscosity 00Pope . In the high Re<sub>ฮป</sub> limit the density of turbulent kinetic energy becomes space independent in the log-law region. Accordingly, the spatial transport term is very small in that log-law region. More detailed analysis, see, e.g. Fig. 3 in Ref. DNS , shows that even for a relatively small Re<sub>ฮป</sub> in the log-law turbulent region this term is small with respect to the energy transfer term from scale to scale which is represented by $`\gamma W_{ij}`$ in the equations above. On the other hand, in the viscous sub-layer the mean velocity is determined by the viscous term and thus the influence of the spatial energy transfer term can be again neglected. To keep the model simple we will neglect $`ฯต_{\mathrm{tr}}`$ term also in the buffer layer where it is of the same order as the other terms of the model. The reason for this simplification, which evidently will cause some trouble in the buffer layer, is that the energy balance equations used below become local in space. This is a great advantage of the model, allowing us to advance analytically to obtain a very transparent phenomenology of wall-bounded turbulence. It was already demonstrated in Ref. 04LPT that the simple description (10) gives a uniformly reasonable description of the rate of the energy dissipation in the entire boundary layer. Here we improve this description further, effectively accounting for the energy transfer term in the balance equation by an appropriate decrease in the viscous layer parameters $`a_{ij}`$.
### I.3 Summary of the two versions of the model
For the sake of further analysis we present the model with the final notation:
$`W_{xy}(y)+\nu _0S(y)`$ $`=`$ $`P(y),`$ (16a)
$`[\mathrm{\Gamma }_{xx}+3\stackrel{~}{\gamma }_\mathrm{d}]W_{xx}`$ $`=`$ $`\stackrel{~}{\gamma }_\mathrm{d}W2SW_{xy},`$ (16b)
$`[\mathrm{\Gamma }_{yy}+3\stackrel{~}{\gamma }_\mathrm{d}]W_{yy}`$ $`=`$ $`\stackrel{~}{\gamma }_\mathrm{d}W,`$ (16c)
$`[\mathrm{\Gamma }_{zz}+3\stackrel{~}{\gamma }_\mathrm{d}]W_{zz}`$ $`=`$ $`\stackrel{~}{\gamma }_\mathrm{d}W,`$ (16d)
$`[\mathrm{\Gamma }_{xy}+3\stackrel{~}{\gamma }]W_{xy}`$ $`=`$ $`SW_{yy},`$ (16e)
In the traditional theory of wall-bounded turbulence one employs the โwall unitsโ $`u_\tau ,\tau `$ and $`\mathrm{}_\tau `$ for the velocity, time and length 00Pope which for a fluid of density $`\rho `$ are:
$$u_\tau \sqrt{\frac{P_0}{\rho }},\tau \frac{\nu _0}{P_0},\mathrm{}_\tau \frac{\nu _0}{\sqrt{\rho P_0}}.$$
(17)
Using these scales one defines the wall-normalized dimensionless objects
$`y^+`$ $``$ $`{\displaystyle \frac{y}{\mathrm{}_\tau }},V^+(y){\displaystyle \frac{V_x(y)}{u_\tau }},v_\mathrm{T}^+(y){\displaystyle \frac{v_\mathrm{T}(y)}{u_\tau }},etc.,`$
$`S^+`$ $``$ $`S\tau ,W_{ij}^+(y){\displaystyle \frac{W_{ij}(y)}{u_\tau ^2}},etc.`$ (18)
In our model we can use the property of locality in space to introduce โlocal unitsโ:
$$\stackrel{~}{u}_\tau (y)\sqrt{\frac{P(y)}{\rho }},\stackrel{~}{\tau }(y)\frac{\nu _0}{P(y)},\stackrel{~}{\mathrm{}}_\tau (y)\frac{\nu _0}{\sqrt{\rho P(y)}},$$
(19)
similar to traditional wall units Eq. (17), but with the replacement $`P_0P(y)`$, and โlocally normalizedโ dimensionless objects, analogous to Eq. (18):
$`y^{}`$ $``$ $`{\displaystyle \frac{y}{\stackrel{~}{\mathrm{}}_\tau (y)}},v_\mathrm{T}^{}(y^{}){\displaystyle \frac{v_\mathrm{T}(y)}{\stackrel{~}{u}_\tau (y)}},`$ (20)
$`S^{}(y^{})`$ $``$ $`S\stackrel{~}{\tau }(y),W_{ij}^{}(y^{}){\displaystyle \frac{W_{ij}(y)}{\stackrel{~}{u}_\tau ^2(y)}},etc..`$
Then the dimensionless version of Eq. (16a) reduces to
$`W_{xy}^{}+S^{}`$ $`=`$ $`1,`$ (21a)
$`\left(\mathrm{\Gamma }_{xx}^{}+3\stackrel{~}{\gamma }_\mathrm{d}^{}\right)W_{xx}^{}`$ $`=`$ $`\stackrel{~}{\gamma }_\mathrm{d}^{}W^{}2S^{}W_{xy}^{},`$ (21b)
$`\left(\mathrm{\Gamma }_{yy}^{}+3\stackrel{~}{\gamma }_\mathrm{d}\right)W_{yy}^{}`$ $`=`$ $`\stackrel{~}{\gamma }_\mathrm{d}W^{},`$ (21c)
$`\left(\mathrm{\Gamma }_{zz}^{}+3\stackrel{~}{\gamma }_\mathrm{d}^{}\right)W_{zz}`$ $`=`$ $`\stackrel{~}{\gamma }_\mathrm{d}^{}W^{},`$ (21d)
$`\left(\mathrm{\Gamma }_{xy}^{}+3\stackrel{~}{\gamma }^{}\right)W_{xy}^{}`$ $`=`$ $`S^{}W_{yy}^{}.`$ (21e)
Introducing $`v^{}\sqrt{W^{}}`$ we can write:
$`\mathrm{\Gamma }_{ij}^{}`$ $`=`$ $`\left({\displaystyle \frac{a_{ij}}{y^{}}}\right)^2+{\displaystyle \frac{bv^{}}{y^{}}}\delta _{ij},\text{for the sum model},`$ (22a)
$`\mathrm{\Gamma }_{ij}^{}`$ $`=`$ $`\sqrt{{\displaystyle \frac{a_{ij}^4}{(y^{})^4}}+{\displaystyle \frac{b^2v_{}^{}{}_{}{}^{2}}{(y^{})^2}}\delta _{ij}},\text{for the root model},`$ (22b)
$`\stackrel{~}{\gamma }_\mathrm{d}^{}`$ $`=`$ $`{\displaystyle \frac{\stackrel{~}{b}_\mathrm{d}v^{}}{y^{}}},\stackrel{~}{\gamma }^{}={\displaystyle \frac{\stackrel{~}{b}v^{}}{y^{}}},\text{for both versions}.`$ (22c)
## II Analysis of the model
### II.1 Solution of the 7-parameters version of the model
#### II.1.1 Solutions in the viscous sub-layer
The four Eqs. (16b) โ (16e) can be considered as a homogeneous โlinearโ set of equations for $`W_{xx}`$, $`W_{yy}`$, $`W_{zz}`$ and $`W_{xy}`$ (with coefficients that are functions of $`W`$). They can have a trivial solution $`W=0`$ for which Eq. (16a) gives
$$S=P_0/\nu _0,V=yP_0/\nu _0W_{ij}=0,\text{Laminal layer}.$$
(23)
The complete absence of turbulent activity in the viscous layer in our model is a consequence of leaving out the energy transport in physical space.
#### II.1.2 Analysis of turbulent solution
Equations (21b21e) have non-trivial โturbulent solutionโ with $`W^{}0`$:
$`W_{xx}^{}`$ $`=`$ $`{\displaystyle \frac{W^{}}{2}}\left[{\displaystyle \frac{\mathrm{\Gamma }_{yy}^{}+\stackrel{~}{\gamma }_\mathrm{d}^{}}{\mathrm{\Gamma }_{yy}^{}+3\stackrel{~}{\gamma }_\mathrm{d}^{}}}+{\displaystyle \frac{\mathrm{\Gamma }_{zz}^{}+\stackrel{~}{\gamma }_\mathrm{d}^{}}{\mathrm{\Gamma }_{zz}^{}+3\stackrel{~}{\gamma }_\mathrm{d}^{}}}\right],`$ (24a)
$`W_{yy}^{}`$ $`=`$ $`{\displaystyle \frac{W^{}\stackrel{~}{\gamma }_\mathrm{d}^{}}{\mathrm{\Gamma }_{yy}^{}+3\stackrel{~}{\gamma }_\mathrm{d}^{}}},W_{zz}^{}={\displaystyle \frac{W^{}\stackrel{~}{\gamma }_\mathrm{d}^{}}{\mathrm{\Gamma }_{zz}^{}+3\stackrel{~}{\gamma }_\mathrm{d}^{}}},`$ (24b)
$`W_{xy}^{}`$ $`=`$ $`{\displaystyle \frac{W^{}S^{}\stackrel{~}{\gamma }_\mathrm{d}^{}}{\left(\mathrm{\Gamma }_{xy}^{}+3\stackrel{~}{\gamma }^{}\right)\left(\mathrm{\Gamma }_{yy}^{}+3\stackrel{~}{\gamma }_\mathrm{d}^{}\right)}},`$ (24c)
if its determinant $`\mathrm{\Delta }`$ vanishes. The solvability condition $`\mathrm{\Delta }=0`$ gives:
$`\left(S^{}\right)^2`$ $`=`$ $`{\displaystyle \frac{\mathrm{\Gamma }_{xy}^{}+3\stackrel{~}{\gamma }_{xy}^{}}{2\stackrel{~}{\gamma }_{yy}^{}\left(\mathrm{\Gamma }_{zz}^{}+3\stackrel{~}{\gamma }_{zz}^{}\right)}}[\mathrm{\Gamma }_{xx}^{}\mathrm{\Gamma }_{yy}^{}\mathrm{\Gamma }_{zz}^{}`$
$`+2\stackrel{~}{\gamma }_\mathrm{d}^{}\left(\mathrm{\Gamma }_{xx}^{}\mathrm{\Gamma }_{yy}^{}+\mathrm{\Gamma }_{xx}^{}\mathrm{\Gamma }_{zz}^{}+\mathrm{\Gamma }_{yy}^{}\mathrm{\Gamma }_{zz}^{}\right)`$
$`+3(\stackrel{~}{\gamma }_\mathrm{d}^{})^2(\mathrm{\Gamma }_{xx}^{}+\mathrm{\Gamma }_{yy}^{}+\mathrm{\Gamma }_{zz}^{})].`$
Substitution $`W_{xy}^{}`$ and $`S^{}`$ in Eq. (21a) gives a closed equation for the function $`W^{}(y^{})`$, (or for $`v^{}\sqrt{W^{}}`$). To present the resulting Eq. in explicit form, introduce
$`A(v^{})`$ $``$ $`S^{}/v^{},B(v^{})W_{xy}^{}/S^{}v^{},`$ (25)
$`R_{ij}`$ $``$ $`\mathrm{\Gamma }_{ij}^{}/v^{},\stackrel{~}{r}_\mathrm{d}\stackrel{~}{\gamma }_\mathrm{d}^{}/v^{},\stackrel{~}{r}\stackrel{~}{\gamma }^{}/v^{}.`$
Using Eqs. (24) and (24c) we find:
$`A^2(v^{})`$ $`=`$ $`{\displaystyle \frac{R_{xy}+3\stackrel{~}{r}}{2\stackrel{~}{r}_\mathrm{d}\left(R_{zz}+3\stackrel{~}{r}_\mathrm{d}\right)}}[R_{xx}R_{yy}R_{zz}`$
$`+2\stackrel{~}{r}_\mathrm{d}\left(R_{xx}R_{yy}+R_{xx}R_{zz}+R_{yy}R_{zz}\right)`$
$`+3\stackrel{~}{r}_\mathrm{d}^2(R_{xx}+R_{yy}+R_{zz})],`$
$`B(v^{})`$ $`=`$ $`{\displaystyle \frac{\stackrel{~}{r}_\mathrm{d}}{\left(R_{xy}+3\stackrel{~}{r}\right)\left(R_{yy}+3\stackrel{~}{r}_\mathrm{d}\right)}}.`$
Now Eq. (21a) can be presented as
$$A(v^{})v^{}\left[1+B(v^{})v^{}\right]=1.$$
(27)
Together with Eqs. (24) this provides the full solution of Eqs. (21).
#### II.1.3 Outer layer, $`y^{}>50`$
In the outer layer, far away from the wall, all the viscous terms in Eqs. (21) can be neglected. In this case Eqs. (22) for both the sum and the root minimal models give:
$$\mathrm{\Gamma }_{ij}^{}\stackrel{~}{\gamma }^{}\delta _{ij},$$
(28)
and Eqs. (24b24b) and the solution (37) simplifies drastically:
$$W_{yy}^{}=W_{zz}^{}=\frac{W^{}\stackrel{~}{b}_\mathrm{d}}{b+3\stackrel{~}{b}_\mathrm{d}}.$$
(29)
By analyzing results of experiments and numerical simulations (as discussed in Sec. III) we found that in the outer layer $`W_{xx}^{}=\frac{1}{2}W^{}`$, and $`W_{yy}^{}=W_{zz}=\frac{1}{4}W^{}`$. The model reproduces these findings if we choose:
$$b=\stackrel{~}{b}_\mathrm{d}.$$
(30)
Using this relation and solution of (21a): $`W_{xy}^{}=1`$, in the rest of Eqs. (21), one finds
$$W^{}=\sqrt{\frac{24\stackrel{~}{b}}{b}},\frac{1}{\kappa }=\left(\frac{b}{2}\right)^{1/4}(3\stackrel{~}{b})^{3/4}.$$
(31)
Here $`\kappa `$ is nothing but the von-Kรกrmรกn constant, that determines the slope of the logarithmic mean velocity profile in the log-law turbulent region:
$`V^+(y^+)`$ $`=`$ $`\kappa ^1\mathrm{ln}y^++C,\mathrm{for}z^+30,`$ (32)
$`\kappa `$ $``$ $`0.436,C6.13.`$
The experimental value of $`\kappa `$ and the intercept $`C`$, were taken from 97ZS . Using the simulations result $`W^+6.85`$ of Ref. DNS which is reproduced in Fig. 6, we find
$$b0.256,\stackrel{~}{b}0.500.$$
(33)
#### II.1.4 Reduction of the number parameters: the minimal model
The parameters $`a_{ij}`$ are responsible for the difference between the energy dissipation and the energy transfer in the viscous sub-layer. To further simplify the model we reduce the number of independent parameters $`a_{ij}`$ from 4 ($`a_{xx}`$, $`a_{yy}`$, $`a_{zz}`$ and $`a_{xy}`$) to two, denoted as $`a`$ and $`\stackrel{~}{a}`$. Among various possibilities (including $`a_{xx}=a_{zz}=a`$, $`a_{yy}=\stackrel{~}{a}`$, $`a_{xy}=(a+\stackrel{~}{a})/2`$) we choose a parametrization similar to the situation with the outer layer parameters:
$$a_{ii}=a,a_{xy}=\stackrel{~}{a}.$$
(34)
The analytical solution given in next Sec. II.2 simplifies considerably with the 4-parameter version of the model. A further simplification $`a=\stackrel{~}{a}`$ could be considered, but we rule it out since it yields a monotonic dependence of the turbulent kinetic energy $`W^+(y^+)`$ with $`y^+`$, while experimentally there is a pronounced peak of $`W^+(y^+)`$ in the buffer sub-layer, see Sec. III.3. We thus consider the four-parameter model as the โminimal modelโ (MM).
Below we will use mostly the following set of constants:
$`a`$ $`=`$ $`1.0,\stackrel{~}{a}=10.67,\text{ sum-MM},`$ (35a)
$`a`$ $`=`$ $`1.0,\stackrel{~}{a}=12.95,\text{root-MM},`$ (35b)
$`b`$ $`=`$ $`0.256,\stackrel{~}{b}=0.500,\text{both MMs}.`$ (35c)
This choice is based on the analysis of the simulational and experimental data presented in Sec. III.
Notice that eliminating $`b`$ from Eqs. (31) (valid for both sum- and root models) one gets:
$$v^{}=12\kappa \stackrel{~}{b}6\kappa .$$
(36)
With the simulational values $`\kappa =0.436`$ and $`W_{}^{}{}_{\mathrm{}}{}^{}=6.85`$ this relationship is valid with a precision that is better than 1%.
### II.2 Analysis of the Minimal Models
#### II.2.1 $`a,\stackrel{~}{a}`$-parametrization of the general solution
With the minimal parametrization, given by Eq. (34), the solution (24) takes on a simpler form:
$`W_{yy}^{}`$ $`=`$ $`W_{zz}^{}={\displaystyle \frac{v^{}}{4v_4}}W^{},W_{xx}^{}={\displaystyle \frac{v_2}{2v_4}}W^{},`$ (37a)
$`W_{xy}^{}`$ $`=`$ $`{\displaystyle \frac{W^{}}{2}}\sqrt{{\displaystyle \frac{bv^{}v_1}{6\stackrel{~}{b}v_3v_4}}},S^{}={\displaystyle \frac{1}{y^{}}}\sqrt{{\displaystyle \frac{6b\stackrel{~}{b}v_1v_3v_4}{v^{}}}}.`$ (37b)
Here we introduced the following short-hand notations $`v_j`$ for the sum-MM:
$`v_1`$ $``$ $`v^{}+{\displaystyle \frac{a^2}{by^{}}},v_2v^{}+{\displaystyle \frac{a^2}{2by^{}}},`$ (38a)
$`v_3`$ $``$ $`v^{}+{\displaystyle \frac{\stackrel{~}{a}^2}{3\stackrel{~}{b}y^{}}},v_4v^{}+{\displaystyle \frac{a^2}{4by^{}}}.`$
For the root-MM instead of Eq. (38a) we take:
$`v_1`$ $``$ $`\sqrt{v_{}^{}{}_{}{}^{2}+{\displaystyle \frac{a^4}{\left(by^{}\right)^2}}},v_2{\displaystyle \frac{v_1+v^{}}{2}},v_4{\displaystyle \frac{v_1+3v^{}}{4}},`$
$`v_3`$ $``$ $`{\displaystyle \frac{v^{}b}{\stackrel{~}{b}}}+\sqrt{(\stackrel{~}{b}b)^2v_{}^{}{}_{}{}^{2}+{\displaystyle \frac{\stackrel{~}{a}^4}{(3\stackrel{~}{b}y^{})^2}}}.`$ (38b)
With the minimal parametrization Eq. (27) takes a very simple explicit form:
$$v_{}^{}{}_{}{}^{2}+\frac{12\stackrel{~}{b}v^{}}{y^{}}r_3r_4=\sqrt{\frac{24\stackrel{~}{b}r_3r_4}{br_1}},r_jv_j/v^{}.$$
(39a)
This form of equation for $`v_\mathrm{T}^{}`$ serves below as a starting point for an approximate (iterative) analytical solution. One can also seek an exact solution by numerical methods; to this aim it is better to use the following form of the same equation:
$`F(v^{},y^{})`$ $``$ $`{\displaystyle \frac{b}{24\stackrel{~}{b}}}v_{}^{}{}_{}{}^{6}v_1+v^{}v_3v_4\left[{\displaystyle \frac{b}{y}}v_{}^{}{}_{}{}^{2}v_11\right]`$ (39b)
$`+{\displaystyle \frac{6b\stackrel{~}{b}}{y^2}}v_1v_3^2v_4^2=0.`$
Equation (39b) has seven roots for the sum-MM, (and 27 for the root-MM) but only two of them, denoted as $`v_\pm ^{}`$ are real and positive for large enough $`y^{}`$. These two roots approach each other upon decreasing the distance from the wall. At some value of $`y^{}`$ these roots merge:
$$v_+^{}(y_{\mathrm{vs}})=v_{}^{}(y_{\mathrm{vs}})v_{}<v_{}^{}{}_{\mathrm{}}{}^{}.$$
(40)
The values $`y_{\mathrm{vs}}`$ and $`v_{}`$ as functions of the problem parameters follow from the polynomial (39b):
$$F(v^{},y^{})=0,\frac{F(v^{},y^{})}{v^{}}=0.$$
(41)
For $`y^{}<y_{\mathrm{vs}}`$ there are no physical (positive definite) solutions of Eq. (39b). This is a laminar region that was discussed before as the viscous sub-layer. In Tab. 2 we present the corresponding values of $`y_{\mathrm{vs}}`$ and $`v_{}`$ for $`b=0.256,\stackrel{~}{b}=0.5`$ and various pairs of $`a,\stackrel{~}{a}`$.
#### II.2.2 Iterative solution of Eq. (39a) for rms turbulent velocity $`v^{}(y^{})`$
To develop further analytic insight we employ an iterative procedure to find an approximate solution for $`v_+^{}(y)`$ for all $`y^{}>y_{\mathrm{vs}}`$. For this goal we forget for a moment that $`r_j`$ depends on $`v^{}`$, and consider Eq. (39a) as a quadratic equation with a positive solution:
$$v^{}=\sqrt{\sqrt{\frac{24\stackrel{~}{b}r_3r_4}{br_1}}+\left(\frac{6\stackrel{~}{b}r_3r_4}{y^{}}\right)^2}\frac{6\stackrel{~}{b}r_3r_4}{y^{}}.$$
(42)
However $`r_j`$ does depend on $`v^{}`$. For example, for the sum-MM:
$`r_1(v^{})`$ $`=`$ $`1+{\displaystyle \frac{a^2}{by^{}v_\mathrm{T}^{}}},r_2(v^{})=1+{\displaystyle \frac{a^2}{2by^{}v^{}}},`$ (43)
$`r_3(v^{})`$ $``$ $`1+{\displaystyle \frac{\stackrel{~}{a}^2}{3\stackrel{~}{b}y^{}v^{}}},r_4(v^{})=1+{\displaystyle \frac{a^2}{4by^{}v^{}}}.`$
Nevertheless, for very large $`y^{}`$ all $`r_j1`$ and an asymptotic solution of (42) reproduces the asymptotic value of $`v^{}=v_{\mathrm{}}^{}=(24\stackrel{~}{b}/b)^{1/4}`$, given by Eq. (31).
A much better approximation for $`v^{}(y^{})`$ (denoted as $`v_1^{}`$) is obtained using in Eq. (42) a $`v^{}`$-independent $`r_{j,0}r_j(v_{\mathrm{}}^{})`$ instead of $`r_j=1`$:
$$v_1^{}=\sqrt{\sqrt{\frac{24\stackrel{~}{b}r_{3,0}r_{4,0}}{br_{1,0}}}+\left(\frac{6\stackrel{~}{b}r_{3,0}r_{4,0}}{y^{}}\right)^2}\frac{6\stackrel{~}{b}r_{3,0}r_{4,0}}{y^{}}.$$
(44a)
Clearly, this iterative procedure can be prolonged further and one can find the velocity $`v^{}`$ at the $`n+1`$ iteration step, $`v_{n+1}^{}(y^{})`$, using the relations $`r_{j,n}r_j(v_n^{})`$, found with the velocity $`v_n^{}`$ of the previous, $`n`$-th step:
$$v_{n+1}^{}=\sqrt{\sqrt{\frac{24\stackrel{~}{b}r_{3,n}r_{4,n}}{br_{1,n}}}+\left(\frac{6\stackrel{~}{b}r_{3,n}r_{4,n}}{y^{}}\right)^2}\frac{6\stackrel{~}{b}r_{3,n}r_{4,n}}{y^{}}.$$
(44b)
The numerical verification of the iteration procedure is given in Appendix A. The conclusion is that already the first few iterations are sufficiently accurate for all practical purposes: often one can use the first iteration and occasionally the second one.
Remarkably, the first iteration can be formulated directly in terms of the basic Eqs. (21) by replacing the turbulent velocity profile $`v^{}(y^{})`$ in Eqs. (22) for $`\mathrm{\Gamma }_{ij}^{}`$ and $`\stackrel{~}{\gamma }_{ij}^{}`$ by its asymptotic value in log-law region $`v_{\mathrm{}}^{}=\left(24\stackrel{~}{b}/b\right)^{1/4}`$.
#### II.2.3 Iterative solution for the mean velocity and Reynolds tensor profiles
Consider first the resulting plots for the mean velocity profile, $`V_n^{}(y^{})`$, computed with the help of turbulent velocity $`v_n^{}`$ at the $`n`$-th iteration step:
$$V_n^{}(y^{})=y_{\mathrm{vs}}+_{y_{\mathrm{vs}}}^y^{}S_n^{}(\xi )๐\xi ,y^{}>y_{\mathrm{vs}}.$$
(45)
Here $`S_n^{}(x)`$ denotes $`S^{}(x)`$, given by Eqs. (37), with $`v^{}=v_n^{}`$. Figure 1 displays plots of $`V_n^{}(y^{})`$ for $`n=1,\mathrm{\hspace{0.17em}2},\mathrm{\hspace{0.17em}3},\mathrm{\hspace{0.17em}4}`$ and the โexactโ (numerical) result $`V^{}(y^{})`$. All the plots almost coincide within the linewidth. This means that for the purpose of computing $`V^{}(y^{})`$ one can use the first approximation $`v_1^{}`$ given by Eq. (44a) instead of the exact solution $`v^{}`$.
Next we present in Fig. 2 log-plots for the trace of the Reynolds-stress tensor $`W_n^{}(y^{})`$, (computed with the $`n`$-th iteration step for $`n=1,2,3,4`$) together with the โexactโ numerical solution $`W^{}(y^{})`$ for the sum-MM. Evidently, the iterative procedure for the kinetic energy does not converge as rapidly as for the mean velocity profile: one can distinguish the plots of $`W_1^{}(y^{})`$, $`W_2^{}(y^{})`$ and $`W_3^{}(y^{})`$; the plots of $`W_3^{}(y^{})`$ and $`W_4^{}(y^{})`$ coincide within the line width. Nevertheless, for $`y^{}>5`$ (i.e. in the buffer layer and in the outer region) alreadly the first iterative solution provides a very reasonable approximation to the exact solution for the kinetic energy profile.
## III Analysis of the numerical and experimental data and comparison with the model prediction
In this Section we analyze and compare the predictions of the minimal models to results of experiments and comprehensive direct numerical simulations of high Re channel flows. We refer to results that were made available in the public domain by R. G. Moser, J. Kim, and N. N. Mansour DNS , to Large-Eddy-Simulation performed by C. Casciola LES , and to recent laboratory experiments in a vertical water tunnel by A. Angrawal, L. Djenidi and R.A. Antonia Exp . The choice of the outer layer parameters $`b_{ij}`$ and $`\stackrel{~}{b}_{ij}`$ is based on our analysis of the anisotropy in the log-low region, presented in Sec. III.1. The relation between the viscous layer parameters $`a`$ vs. $`\stackrel{~}{a}`$ is based on the comparison between the DNS and the model mean velocity profiles, presented in Sec. III.2. The final choice of $`a`$ and $`\stackrel{~}{a}`$ is motivated by the DNS data for the kinetic profile which is compared with the model prediction in Sec. III.3. Section III.4 is devoted to the comparison of the model results with the DNS profiles of the Reynolds stress $`W_{xy}^{}`$ and partial kinetic energies $`W_{xx}^{}`$, $`W_{yy}^{}`$, $`W_{zz}^{}`$.
### III.1 Anisotropy of the log-layer: Relative partial kinetic energies $`R_{xx}`$, $`R_{yy}`$, $`R_{zz}`$ in the outer layer
The anisotropy of turbulent boundary layer can be characterized by the dimensionless ratios
$$R_{ii}(y^+)\frac{W_{ii}(y^+)}{W(y^+)}=\frac{W_{ii}^+(y^+)}{W^+(y^+)}.$$
(46)
This anisotropy plays an important role in various phenomena and was a subject of experimental and theoretical concern for many decades, see, e.g. MY ; 00Pope . Nevertheless, up to now the dispersion of results on the subject appears quite large. There is a widely spread opinion, based on atmospheric measurements, that the wall-normal turbulent fluctuations $`W_{yy}`$ are much smaller than the other ones. For example, Monin and Yaglom MY reported that for a neutrally stratified log-boundary layer $`R_{xx}=54\%`$, $`R_{yy}=6\%`$ and $`R_{zz}=40\%`$. This contradicts recent DNS results for Re<sub>ฮป</sub>=590 which are available in Ref. DNS , as shown in Fig. 3. Note that there is a region about $`100<y^+<\frac{2}{3}\text{Re}_\lambda `$ where the plots of $`R_{ii}(y^+)`$ are nearly horizontal, as expected in the log-law region. From these plots we can conclude that is this region $`R_{xx}53\%`$ which is close to the value $`54\%`$, stated in MY . Nevertheless, the DNS data for $`R_{yy}`$ and $`_{zz}`$ are completely different. From Fig. 3 one gets $`_{yy}22\%`$ and $`_{zz}27\%`$. Thus $`_{yy}`$ can be considered roughly equal to $`_{zz}`$. We should mention here that various models of turbulent boundary layers give $`_{yy}=_{zz}`$ in the asymptotic log-law region. We propose that the difference between $`_{yy}`$ and $`_{zz}`$ which is observed in Fig. 3 is due to the effect of the energy transfer. This effect practically vanishes in the asymptotic limit Re$`{}_{\lambda }{}^{}\mathrm{}`$, but is still present at values of Re<sub>ฮป</sub> which are available in DNS DNS . Indeed, for both values of Re<sub>ฮป</sub> shown in Fig. 3, $`W_{yy}=W_{zz}`$ in the center of the channel, where the energy flux vanishes by symmetry. Clearly, there is no energy flux also in space homogeneous cases, for example for a constant shear flow, in which, according to the model, one should expect $`W_{yy}=W_{zz}`$ in the entire space.
Our expectation that $`W_{yy}=W_{zz}`$, which is based on symmetry considerations, is confirmed by the Large Eddy Simulation (LES) of the constant shear flow LES . As one sees in Fig. 4 in this flow $`_{xx}0.46`$, while $`W_{yy}W_{zz}0.27`$.
As stated, for sufficiently large values of Re<sub>ฮป</sub> the energy transfer terms should almost vanish in the log-law region and, according to our model, one can expect in that region $`W_{yy}=W_{zz}`$ also in the channel flow. This viewpoint was confirmed in the aforementioned laboratory experiment Exp in a vertical water channel with Re$`{}_{\lambda }{}^{}=1000`$, reproduced in Fig. 5. The experimental values of $`_{xx}`$, $`_{yy}`$ and $`_{zz}`$ in the log-law turbulent region are in the excellent quantitative agreement with the values $`_{xx}=0.5`$ and $`_{yy}=_{zz}=0.25`$ shown in Figs. 3 and 5 by horizontal dashed lines.
Table 1 summaries the DNS, LES and experimental values of the relative kinetic energies in comparison with the model expectation. The conclusion is that, in contradiction with the old and still wide spread viewpoint MY that the wall-normal turbulent activity is strongly suppressed, $`_{yy}<0.1`$, the turbulent kinetic energy in the log-law region is distributed in a very simple manner: the stream-wise component contains a half of total energy, $`_{xx}=\frac{1}{2}`$ and the rest is equally distributed between the wall-normal and span-wise components: $`_{yy}=_{zz}=\frac{1}{4}`$. As shown in Sec. II.1.4, this very simple energy distribution is predicted by the minimal model if one assumes that the characteristic nonlinear times scales in the energy transfer term and in the return-to-isotropy term are identical.
Thus anisotropy predicted by our minimal model agree reasonably accurately with those obtained from DNS, LES and vertical water channel. However, we must admit that all these are not yet the nature. Indeed DNS, LES and lab experiments, done at relatively modest Re<sub>ฮป</sub> impose limits on the low-frequency intervals in the spectra of the streamwise and the transverse velocity components, because of side walls or periodic boundary conditions. In other words, it must not be ruled out that DNS, as well as lab experiments, cat off the largest-scale ejections, observed in the atmosphere in the form of coherent structures, which pump additional streamwise- and transverse-velocity energies into the log-layer. Be it as it may, we leave a detailed discussion of the above problem and geophysical applications of our theory for further work. At the present stage, following our strategy of โpragmatic, task-dependent simplificationโ, we consider our minimal model as definitely relevant to flows in channels. Its extension to geophysical (atmospheric and oceanic) boundary layers needs further efforts.
### III.2 Mean velocity profile in channel flows
To compute the mean velocity profile $`V^+(y^+)`$ in our approach we need to connect first $`S^+(y^+)`$ with $`S^{}(y^{})`$. According to the definitions (18) โ (20):
$$S^+(y^+)=\left(1\frac{y^+}{\text{Re}_\lambda }\right)S^{}\left(y^+\sqrt{1\frac{y^+}{\text{Re}_\lambda }}\right),$$
(47)
where
$$\text{Re}_\lambda L/\mathrm{}_\tau ,$$
(48)
and $`y^+>y_{\mathrm{vs}}`$. For $`y^+<\widehat{๐ฒ}_{\mathrm{vs}}`$ we can take $`S^+(y^+)=1`$ and integrate the resulting shear over the distance to the wall with no-slip boundary condition. The resulting profiles $`V^+(y^+)`$ for Re$`{}_{\lambda }{}^{}=590`$ and the parameters (33) are shown in Fig. 6 as a dashed line for the sum-MM and as a dot-dashed line for the root-MM. The DNS profile of DNS for the same Re<sub>ฮป</sub> is shown as a solid line. There is no significant difference (less than $`1\%`$) between these plots in the viscous sublayer, buffer and outer layers, where $`y^+300`$ i.e. in about 50% of the channel half-width $`L^+=\text{Re}_\lambda =590`$. This robustness of the mean velocity profile $`V^+(y^+)`$ is a consequence of the fact that $`V^+(y^+)`$ is an integral of the mean shear $`S^+`$ which is described very well both in the viscous and the outer layers.
Notice that our model does not describe the upward deviation from the log-low which is observed near the mid-channel (of about 5-6 units in $`V^+`$, independent on Reynolds number). We consider this minor disagreement as an acceptable price for the simplicity of the minimal model which neglects the energy transport term toward the centerline of the channel. This transport is the only reason for some turbulent activity near the centerline where both the Reynolds stress $`W_{xy}`$ and the mean shear $`S`$ vanish due to symmetry. Just at the center line the source term in our energy equation, $`2SW_{xy}`$, is zero, and the missing energy transport term is felt.
The plots in Fig. 6 have a reasonably straight logarithmic region from $`y^+20`$ to $`y^+200`$. On the other hand, the Reynolds stress profile at the same Re$`{}_{\lambda }{}^{}=590`$ shown in Fig. 8, has no flat region at all. Such a flat region is expected in the true asymptotic regime of Re$`{}_{\lambda }{}^{}\mathrm{}`$, where $`W^+=1`$. Therefore if one plots the model profiles $`V^+`$ at different Re<sub>ฮป</sub> and fits them by log-linear profiles (32) one can get a Re<sub>ฮป</sub>-dependence of the โeffectiveโ intercept in the von-Kรกrmรกn log-law. We think that this explains, why measured value of the log-low intercept can depend on the Reynolds number and on the flow geometry (channel vs. pipe): both in DNS and in physical experiments one usually does not reach high enough values of Re<sub>ฮป</sub>.
### III.3 Profiles of the total kinetic energy density and the choice of the pair $`a,\stackrel{~}{a}`$
The quality of the profiles $`V^+(y^+)`$ calls for a bit more thinking. In fact, one find that the minimal model produces practically the same profiles $`V^+(y^+)`$ not only for the parameters (35) but for a wide choice of the pairs $`a,\stackrel{~}{a}`$, for example for $`a=2`$ and $`\stackrel{~}{a}=8.6`$. Actually, for any $`0a4`$ one can find a value of $`\stackrel{~}{a}`$ that gives a mean velocity profile in good agreement with Fig. 6. In other words, in the $`(a,\stackrel{~}{a})`$-plane there exist a long narrow corridor that produces a good quantitative description of $`V^+(y^+)`$. Within this corridor there exists a line that provides a โbest fitโ of $`V^+(y^+)`$, minimizing the mean square deviation $`\delta V^+`$
$$\delta V^+\sqrt{[V_{\mathrm{MM}}^+(y^+)V_{\mathrm{DNS}}^+(y^+)]^2}$$
(49)
of the model prediction $`V_{\mathrm{MM}}^+(y^+)`$ from the DNS profile $`V_{\mathrm{DNS}}^+(y^+)`$ in the inner region $`y^+<140`$. Some of the best pairs are given in Table 2 together with the corresponding values of $`\delta V^+`$. Table 2 also presents values of $`y_{\mathrm{vs}}`$ and $`v_{}`$; recall: for $`y^{}<y_{\mathrm{vs}},v^{}=0`$, for $`y^{}=y_{\mathrm{vs}}+0`$ there is a jump of $`v^{}`$ from zero to $`v^{}=v_{}`$. The most striking difference for different $`(a,\stackrel{~}{a})`$ pairs is in the behavior of the Reynolds stress profiles $`W^{}(y^{})`$ that can be used to select the best values of these parameters.
Clearly, the minimal model with only 4 fit parameters cannot fit perfectly the profiles of all the physical quantities that can be measured. Therefore the actual values of $`a`$ and $`\stackrel{~}{a}`$ should be determined with a choice of the characteristics of turbulent boundary layers that we desire to describe best. Foremost in any modeling should be the mean velocity profile which is of crucial importance in a wide variety of transport phenomena. Next we opt to fit well the profile of the kinetic energy density (\[or, equivalently, the profile of the Reynolds stress tensor trace $`W^{}(y^{})`$\] . Figure 7, upper panel, shows the DNS profiles of the trace of the Reynold-stress tensor $`W^+(y^+)`$ for Re$`{}_{\lambda }{}^{}=590`$ (solid lower line) and Re$`{}_{\lambda }{}^{}=395`$ (dashed lower line). There are no plateau in these plots, meaning that these values of Re<sub>ฮป</sub> are not large enough to have a true scale-invariant log-law region. Nevertheless, the plots of
$$W^{}(y^+)=(1\frac{y^+}{\mathrm{Re}_\lambda })^1W^+(y^+)$$
(50)
(shown in the same upper panel of Fig. 7) display clear plateaus, according to the theoretical prediction for Re$`{}_{\lambda }{}^{}\mathrm{}`$. This means that the decay of $`W^+(y^+)`$ is related to the decrease of the momentum flux $`P(y)`$ and that the dimensionless โ โ variables, (20), that use the $`y`$-dependent value of the momentum flux $`P(y)`$ represent the asymptotic physics of the wall bounded turbulent flow, at lower values of Re<sub>ฮป</sub> than the traditional โwall unitsโ (18), which are based on the wall value of the momentum flux $`P_0`$.
To compare the model prediction with simulational results we have to relate $`W_{ij}^{}(y^{})`$ with $`W_{ij}^+(y^+)`$ in channel flows. According to Eqs. (18) โ (20):
$$W_{ij}^+(y^+)=\left(1\frac{y^+}{\text{Re}_\lambda }\right)W_{ij}^{}\left(y^+\sqrt{1\frac{y^+}{\text{Re}_\lambda }}\right),$$
(51)
and similar Eqs. for the its trace $`W^+(y^+)`$. Figure 7 shows a peak of $`W^{}(y^+)`$, $`W_{\mathrm{max}}^{}W^{}(y_{\mathrm{max}})9.8`$ at $`y^+=y_{\mathrm{max}}18`$. As one sees from the Tabl. 2, the minimal model reproduces the peak in $`W^{}(y^{})`$ with an amplitude of about $`8รท8.6`$ for $`a2`$. To be specific we choose $`a=1`$ in both versions of the minimal model, sum-MM and root-MM. With this choice we plot in Fig. 7, lower panel, both theoretical profiles, $`W_{_{{\scriptscriptstyle }}}^{}(y^+)`$ and $`W_{_{\sqrt{}}}^{}(y^+)`$, in comparison with the simulational profile $`W_{\mathrm{DNS}}^{}(y^+)`$. It appears that the root-MM is in better correspondence with the simulation than the sum-MM. However, for the sake of analytic calculations, the sum-MM is simpler. Therefore, again, the choice of the version of MM depends on what is more important for a particular application: calculational simplicity or accuracy of fit.
### III.4 Profiles of the Reynolds stress tensor
In Fig. 8 we present (by solid lines) simulational profiles of the Reynolds stress $`W_{xy}^+(y^+)`$ for Re$`{}_{\lambda }{}^{}=395`$ and Re$`{}_{\lambda }{}^{}=590`$ in comparison with the model predictions (dashed lines) for the root-MM. The upper panel shows the comparison in linear coordinates, the lower panel in linear-log coordinates, stressing the buffer layer region. In the model profiles we used the values of parameters (35), chosen to fit the simulational profiles for the mean velocity and the kinetic energy. In other words, in comparing the profiles of $`W_{xy}^+(y^+)`$ in Fig. 8 *no further fitting was exercised*. Having this in mind, we consider the agreement as very encouraging. The only difference between the model predictions and the simulational profiles of $`W_{xy}^+(y^+)`$ is in a steeper front of the model profiles for $`y^+<20`$. This is again because the model does not account for the energy transfer that can only flatten the front. As already mentioned, even for Re=590 the maximum value of the Reynolds stress does not reach it asymptotical value $`|W_{xy}^+|=1`$, as it should in the true log-law region. The corresponding comparison for the sum-MM looks very similar and is therefore not shown.
Next, we present in Fig. 9 the simulational and root-MM profiles of the diagonal components of the Reynolds-stress tensor, $`W_{ii}^+(y^+)`$ for the channel flow with Re$`{}_{\lambda }{}^{}=590`$. Solid lines present simulational profiles, dashed lines โ the model profiles. The stream-wise and span-wise profiles, $`W_{xx}^+(y^+)`$ and $`W_{zz}^+(y^+)`$, are in good agreement in the most of the channel, $`70<y^+<470`$, while for the wall-normal component, the model profile $`W_{yy}^+(y^+)=W_{zz}^+(y^+)`$ and differs from the simulational one. The model also predicts semi-quantitatively increase in the streamwise part of the kinetic energy and the decrease in the span-wise and wall-normal components in the buffer layer which is observed in simulations. The physical reason of this is simple: as is well known, the energy from the mean flow is transferred only to the stream-wise component of the turbulent fluctuations. Accordingly, in the model one sees the energy production term ($`2SW_{xy}`$) only in the RHS of equation for $`W_{xx}`$. The energy redistributes between other components due to โreturn-to-isotropyโ term $`I_{ij}`$, Eq. (14b) with the isotropisation frequency $`1/y`$. The relative importance of $`I_{ij}`$ (in comparison with the energy relaxation term) decreases toward the wall due to the viscous contribution $`1/y^2`$. Accordingly, near the wall only a small part of the kinetic energy can be transferred from the streamwise to the wall-normal and the span-wise components of the velocity during the relaxation time (that $`1/\mathrm{\Gamma }`$). Also, the model describes well the part (about $`50\%`$ in the outer layer) of the total kinetic energy that contains the streamwise components.
In the core of the flow ($`y^+>450`$) the model gives smaller values of all the components $`W_{ii}^+`$, as compared to simulations and experiments. This is again because the model neglects the energy transfer toward the centerline of the channel, where the energy input into turbulence, $`2SW_{xy}`$, disappears due to the symmetry reason.
Also, there is a quantitative disagreement between the model and the simulations in the buffer layer. One can relate this with the fact that the model neglects the energy flux toward the wall, which plays a considerable role in the energy balance. The minimal models are local in space, but this effect can be effectively accounted for by an appropriate choice of the dissipation constants, taking $`a_{yy}>a_{xx}=a_{zz}`$. We do not propose to take this route; in the buffer layer the turbulent flow is strongly affected by highly intermittent events (coherent structures) connected with the near-wall instabilities of the laminal sub-layer. This is confirmed by the very large values of the flatness (above 30), as shown in Fig. 10. Only for $`y^+>50`$ the flatness reaches the Gaussian value of 3 and one can successfully utilize various lower-order closure model for describing wall bounded flows.
## IV Summary: strength and limitations of the minimal model
The minimal model as formulated in this paper is a version of the algebraic Reynolds stress models. Its aim is to describe, for wall bounded turbulent flows, the profile of mean flow and the statistics of turbulence on the level of simultaneous, one-point, second-order velocity correlation functions, i.e. the entries of the Reynolds-stress tensor $`W_{ij}`$. The model was developed explicitly for plain geometry, including a wide variety of turbulent flows, like channel and plain Couette flows, to some extent fluid flows over inclined planes under gravity (modelling river flows), atmospheric turbulent boundary layers over flat planes and, in the limit of large Reynolds numbers, many other turbulent flows, including pipe, circular Couette flows, *etc*.
In developing a simple model one needs to decide what are the physically important aspects of the flow statistics, those which determine the mean-flow and the turbulent transport phenomena. The choice of the Reynolds-stress approach was dictated by the decision to emphasize the accurate description of $`V(y)`$\- and $`W_{ij}(y)`$\- profiles. The main criteria in constructing the model were simplicity, physical transparency, and maximal analytical tractability of the resulting model. That is why we took liberty to ignore the spatial energy flux, and, thanks to the plain geometry, to estimate the spatial derivatives and the outer scale of turbulence using the distance to the wall $`y`$. The same motivations led to choosing the simplest linear Rotta approximation of the โReturn to isotropy termโ 51Rot and the simplest dimensional form of the nonlinear term for energy flux down the scales, also in agreement with 51Rot .
By proper parametrization the number of fit parameters was reduced from twelve to four. Two of these, $`a`$ and $`\stackrel{~}{a}`$ are responsible for the viscous dissipation of the diagonal, $`W_{ii}`$ , and the off-diagonal, $`W_{xy}`$, components of the Reynolds-stress tensor. The other two parameters - $`b`$ and $`\stackrel{~}{b}`$ โ control the nonlinear relaxation of $`W_{ii}`$ and $`W_{xy}`$. It appears that one cannot decrease the number of fit parameters further with impunity. The outer layer parameters $`b=0.256`$ and $`\stackrel{~}{b}=0.500`$ where chosen to describe the observed constant values of von-Kรกrmรกn in the log-law (32) and the asymptotic level of the density of kinetic energy. The viscous layer parameters were chosen to describe the observed values of the intersection $`C`$ in the von-Kรกrmรกn log-law (32) and the peak of the kinetic energy in the buffer sub-layer. The resulting set of 5 equations for the mean shear $`S(y)`$, Reynolds stress $`W_{xy}`$ and $`W_{xx}`$, $`W_{yy}`$, $`W_{zz}`$ with just four fit parameters is referred to as the minimal model.
As demonstrated in Sec. III the minimal model with the given set (35) of four parameters describes five functions:
* the mean velocity profile $`V(y)`$ is describe with accuracy of $`1\%`$ โ almost throughout the channel (except of small velocity defect in the core of the flow), cf. Fig. 6;
* the Reynolds stress profile $`W_{xy}(y)`$ is also described with accuracy of few percents (except in the viscous layer $`y^+<5`$ in which $`W_{xy}`$ does not contribute to the mechanical balance), cf. Fig. 8;
* the total kinetic energy profile $`\frac{1}{2}W(y)`$ is reproduced with reasonable (semi-quantitative) accuracy, including the position and width of its peak in the buffer sub-layer, cf. Fig. 7;
* The profiles of the partial kinetic energies, $`\frac{1}{2}W_{xx}(y)`$, $`\frac{1}{2}W_{yy}(y)`$ and $`\frac{1}{2}W_{zz}(y)`$, are reproduced, see Fig. 9, including the simple $`\frac{1}{2}`$-$`\frac{1}{4}`$-$`\frac{1}{4}`$ distribution in the asymptotic outer region. This distribution is supported by recent experimental, DNS and LES data, as shown in Figs. 3, 4, and 5.
We consider all this as a good support of the minimal model; too much data is being reproduced to be an accident. It appears that the minimal model takes into account the essential physics almost throughout the channel flow.
On the other hand, one should accept that such a simple model cannot pretend to describe all the aspects of the turbulent statistics in wall bounded flows. For example, the minimal model ignores the quasi-two dimensional character of turbulence and the existence of coherent structures in the very vicinity of the wall. The minimal model does not attempt to take into account many-point and high-order turbulent statistics, including three-point velocity correlation functions and pressure-velocity correlations, responsible for the spatial energy flux and for the isotropization of turbulence. Finally, our choice of dissipation term definitely contradicts to the near-wall expansion, (and see Sec. 11.7.4 of 00Pope ), in disagreement with various known improvements of 51Rot . We propose that all this is a reasonable price for the simplicity and transparency of the minimal model, which is constructed with emphasis on the fundamental characteristics $`V(y)`$ and $`W_{ij}(y)`$ which are crucial for most applications.
We trust that a proper generalization of the minimal model will be found useful in the futures in studies of more complicated turbulent flows, laden with heavy particles, bubbles, **etc**.
###### Acknowledgements.
We thank T.S. Lo for his critical reading of the manuscript and many insightful remarks. We are grateful to Carlo Casciola for sharing with us his LES data. We express our appreciation to R. G.Moser, J. Kim, and N. N.Mansour, for making their comprehensive DNS data of high Re channel flow available to all in Ref. DNS . This work was supported in part by the US-Israel Binational Scientific Foundation and the European Commission under a TMR research grant. SSZ acknowledges supports form the EU Marie Curie Chair Project MEXC-CT-2003-509742; ARO Project โAdvanced parameterization and modelling of turbulent atmospheric boundary layersโ - contract number W911NF-05-1-0055; and EU Project FUMAPEX EVK4-2001-00281.
## Appendix A Validation of the iterative procedure
To see, how the iterative procedure described in Sec. II.2.2 works, we plotted in Fig. 11 iterative profiles of the turbulent velocity $`v_n^{}(y^{})`$ for $`n=1,\mathrm{}5`$ together with the (numerical) solutions of Eq. (39a), $`v_+^{}(y^{})`$ (the thick solid line) and $`v_{}^{}(y^{})`$ (dot-dashed curve). The horizontal straight line presents the asymptotic value $`v_{\mathrm{}}^{}`$. The critical point $`\{v_{}^{},y_{\mathrm{vs}}^{}\}`$ is shown by a black circle. Our analysis shows (and see also Fig. 11), that already the simple Eq. (44a) gives the relative accuracy (with respect to $`v_{\mathrm{}}^{}`$) better than 1% for $`y^{}>30`$. The second iteration works with this accuracy in wider region $`y^{}>10`$, the third iteration gives 1% accuracy for $`y^{}5`$, which is about the critical value $`y_{\mathrm{vs}}^{}4.8`$. Unexpectedly, the approximate solutions work even below the $`y_{\mathrm{vs}}^{}`$, where exact solution is $`v^{}=0`$. One observes with increasing $`n`$ the widening of the region, in which $`v_n^{}`$ practically indistinguishable from zero. The overall conclusion from these observations is that already the fist few iterations give a very good accuracy for all practical purposes, and very often one can use only the first or the second iteration.
|
warning/0506/hep-th0506070.html
|
ar5iv
|
text
|
# I Introduction
## I Introduction
The computation of correlation functions is one of the challenging problem in the theory of quantum integrable lattice models . In this paper, we compute the correlation functions of the free fermion model by means of the algebraic Bethe ansatz method . Our computation are based on the recent progress on the Drinfeld twists. Working in the $`F`$-bases provided by the $`F`$-matrices (Drinfeld twists), the authors in managed to derive the determinant representations of the form factors and correlation functions of the XXX and XXZ models in the framework of algebraic Bethe ansatz.
Recently we have constructed the Drinfeld twists for both the rational $`gl(m|n)`$ and the quantum $`U_q(gl(m|n))`$ supersymmetric models and resolved the hierarchy of their nested Bethe vectors in the $`F`$-basis . These results serve as the basis of our computation in this paper of the correlation functions of the $`U_q(gl(1|1))`$ free fermion model.
Correlation functions of the free fermion model based on the XX0 spin chain (XY model ) with periodic boundary condition were studied in -. As is seen in section VI, by using the Jordan-Wigner transform, our $`U_q(gl(1|1))`$ free fermion model is equivalent to a twisted XX0 model, and the one-point functions we obtained (see (V.5) and (V.7) below) give the $`m`$-point correlation functions of the twisted XX0 model (see e.g. (VI.6)).
The present paper is organized as follows. In section II, we review the background of the $`U_q(gl(1|1))`$ model and its algebraic Bethe ansatz. In section III, we construct the Drinfeld twists for the model. In section IV, we obtain the determinant representation of the scalar products of the $`U_q(gl(1|1))`$ Bethe states. Then in section V, we compute correlation functions of the local fermionic operators of the model. We conclude the paper by offering some discussions in section VI.
## II $`U_q(gl(1|1))`$ free fermion model
### II.1 The background of the model
Let $`V`$ be the 2-dimensional $`U_q(gl(1|1))`$-module and $`REnd(VV)`$ the $`R`$-matrix associated with this module. $`V`$ is $`Z_2`$-graded, and in the following we choose the FB grading for $`V`$, i.e. $`[1]=1,[2]=0`$. The $`R`$-matrix depends on the difference of two spectral parameters $`u_1`$ and $`u_2`$ associated with the two copies of $`V`$, and is, in the FB grading, given by
$`R_{12}(u_1,u_2)=R_{12}(u_1u_2)`$ $`=`$ $`\left(\begin{array}{ccccccccc}c_{12}& 0& 0& 0& & & & & \\ 0& a_{12}& b_{12}^+& 0& & & & & \\ 0& b_{12}^{}& a_{12}& 0& & & & & \\ 0& 0& 0& 1& & & & & \end{array}\right),`$ (II.5)
where
$`a_{12}=a(u_1,u_2){\displaystyle \frac{\mathrm{sinh}(u_1u_2)}{\mathrm{sinh}(u_1u_2+\eta )}},b_{12}^\pm =b^\pm (u_1,u_2){\displaystyle \frac{e^{\pm (u_1u_2)}\mathrm{sinh}\eta }{\mathrm{sinh}(u_1u_2+\eta )}},`$
$`c_{12}=c(u_1,u_2){\displaystyle \frac{\mathrm{sinh}(u_1u_2\eta )}{\mathrm{sinh}(u_1u_2+\eta )}}`$ (II.7)
with $`\eta C`$ being the crossing parameter. One can easily check that the $`R`$-matrix satisfies the unitary relation
$$R_{21}R_{12}=1.$$
(II.8)
Here and throughout $`R_{ij}R_{ij}(u_i,u_j)`$. The $`R`$-matrix satisfies the graded Yang-Baxter equation (GYBE)
$$R_{12}R_{13}R_{23}=R_{23}R_{13}R_{12}.$$
(II.9)
In terms of the matrix elements defined by
$$R(u)(v^i^{}v^j^{})=\underset{i,j}{}R(u)_{ij}^{i^{}j^{}}(v^iv^j),$$
(II.10)
the GYBE reads
$`{\displaystyle \underset{i^{},j^{},k^{}}{}}R(u_1u_2)_{ij}^{i^{}j^{}}R(u_1u_3)_{i^{}k}^{i^{\prime \prime }k^{}}R(u_2u_3)_{j^{}k^{}}^{j^{\prime \prime }k^{\prime \prime }}(1)^{[j^{}]([i^{}]+[i^{\prime \prime }])}`$ (II.11)
$`=`$ $`{\displaystyle \underset{i^{},j^{},k^{}}{}}R(u_2u_3)_{jk}^{j^{}k^{}}R(u_1u_3)_{ik^{}}^{i^{}k^{\prime \prime }}R(u_1u_2)_{i^{}j^{}}^{i^{\prime \prime }j^{\prime \prime }}(1)^{[j^{}]([i]+[i^{}])}.`$
The quantum monodromy matrix $`T(u)`$ of the free fermion model on a lattice of length $`N`$ is defined as
$`T_0(u)=R_{0N}(u,z_N)R_{0N1}(u,z_{N1})_{\mathrm{}}R_{01}(u,z_1),`$ (II.12)
where the index 0 refers to the auxiliary space and $`\{z_i\}`$ are arbitrary inhomogeneous parameters depending on site $`i`$. $`T(u)`$ can be represented in the auxiliary space as the $`2\times 2`$ matrix whose elements are operators acting on the quantum space $`V^N`$:
$`T_0(u)=\left(\begin{array}{ccc}A(u)& B(u)& \\ C(u)& D(u)& \end{array}\right)_{(0)}.`$ (II.15)
By using the GYBE, one may prove that the monodromy matrix satisfies the GYBE
$`R_{12}(uv)T_1(u)T_2(v)=T_2(v)T_1(u)R_{12}(uv).`$ (II.16)
or in matrix form,
$`{\displaystyle \underset{i^{},j^{}}{}}R(uv)_{ij}^{i^{}j^{}}T(u)_i^{}^{i^{\prime \prime }}T(v)_j^{}^{j^{\prime \prime }}(1)^{[j^{}]([i^{}]+[i^{\prime \prime }])}`$
$`={\displaystyle \underset{i^{},j^{}}{}}T(v)_j^j^{}T(u)_i^i^{}R(uv)_{i^{}j^{}}^{i^{\prime \prime }j^{\prime \prime }}(1)^{[j^{}]([i]+[i^{}])}.`$ (II.17)
Define the transfer matrix $`t(u)`$
$`t(u)=str_0T_0(u),`$ (II.18)
where $`str_0`$ denotes the supertrace over the auxiliary space. With the help of the GYBE, one may check that the transfer matrix satisfies the commutation relation $`[t(u),t(v)]=0,`$ ensuring the integrability of the system. The transfer matrix gives the Hamiltonian of the system:
$`H`$ $`=`$ $`{\displaystyle \frac{d\mathrm{ln}t(u)}{du}}|_{u=0}`$ (II.19)
$`=`$ $`{\displaystyle \frac{1}{\mathrm{sinh}\eta }}{\displaystyle \underset{j=1}{\overset{N}{}}}(E_{(j)}^{12}E_{(j+1)}^{21}+E_{(j)}^{21}E_{(j+1)}^{12}2\mathrm{cosh}\eta E_{(j)}^{11}E_{(j+1)}^{11}`$
$`(e^\eta E_{(j)}^{11}E_{(j+1)}^{22}+e^\eta E_{(j)}^{22}E_{(j+1)}^{11})),`$
where $`E_{(k)}^{ij}`$ are generators, which act on the $`k`$th space, of the superalgebra $`U_q(gl(1|1))`$.
Using the standard fermionic representation
$`E_{(k)}^{12}=c_k,E_{(k)}^{21}=c_k^{},E_{(k)}^{11}=1n_k,E_{(k)}^{22}=n_k,n_k=c_k^{}c_k,`$ (II.20)
the Hamiltonian can be rewritten as
$`H={\displaystyle \frac{1}{\mathrm{sinh}\eta }}{\displaystyle \underset{j=1}{\overset{N}{}}}\left(c_jc_{j+1}^{}+c_j^{}c_{j+1}2\mathrm{cosh}\eta (1n_j)\right).`$ (II.21)
### II.2 Algebraic Bethe ansatz
The transfer matrix (II.18) can be diagonalized by using the algebra Bethe ansatz. Define the Bethe state of the system
$`\mathrm{\Phi }_N(v_1,v_2,\mathrm{},v_n)={\displaystyle \underset{i=1}{\overset{n}{}}}C(v_i)|0>,`$ (II.22)
where $`|0>`$ is the pseudo-vacuum,
$`|0>={\displaystyle \underset{k=1}{\overset{N}{}}}\left(\begin{array}{c}0\\ 1\end{array}\right)_{(k)}`$ (II.25)
and the index $`(k)`$ indicates the $`k`$th space.
Applying the elements of the monodromy matrix (II.15) to the pseudo-vacuum $`|0>`$ and its dual, we easily obtain
$`B(u)|0>=0,<0|C(u)=0,D|0>=|0>,<0|D(u)=<0|,`$
$`A(u)|0>={\displaystyle \underset{i=1}{\overset{N}{}}}a(u,z_i)|0>,<0|A(u)={\displaystyle \underset{i=1}{\overset{N}{}}}a(u,z_i)<0|.`$ (II.26)
With the help of the GYBE (II.16), we obtain the commutation relations between the elements of the monodromy matrix
$`C(u)C(v)`$ $`=`$ $`c(u,v)C(v)C(u),`$ (II.27)
$`D(u)D(v)`$ $`=`$ $`D(v)D(u),`$ (II.28)
$`A(u)C(v)`$ $`=`$ $`{\displaystyle \frac{c(u,v)}{a(u,v)}}C(v)A(u)+{\displaystyle \frac{b^+(u,v)}{a(u,v)}}C(u)A(v),`$ (II.29)
$`D(u)C(v)`$ $`=`$ $`{\displaystyle \frac{1}{a(v,u)}}C(v)D(u){\displaystyle \frac{b^{}(v,u)}{a(v,u)}}C(u)D(v),`$ (II.30)
$`B(u)C(v)`$ $`=`$ $`C(v)B(u)+{\displaystyle \frac{b^+(u,v)}{a(u,v)}}[D(v)A(u)D(u)A(v)]`$ (II.31)
$`=`$ $`C(v)B(u)+{\displaystyle \frac{b^+(u,v)}{a(u,v)}}[D(u)t(v)D(v)t(u)].`$
Thus applying the transfer matrix $`t(u)=D(u)A(u)`$ to the Bethe state and using the commutation relations repeatedly, we obtain the eigenvalues of $`t(u)`$ as
$`t(u)\mathrm{\Phi }_N=\mathrm{\Lambda }(u,\{v_k\})\mathrm{\Phi }_N=\left[{\displaystyle \underset{k=1}{\overset{n}{}}}{\displaystyle \frac{1}{a(v_k,u)}}{\displaystyle \underset{j=1}{\overset{N}{}}}a(u,z_j){\displaystyle \underset{k=1}{\overset{n}{}}}{\displaystyle \frac{c(u,v_k)}{a(u,v_k)}}\right]\mathrm{\Phi }_N`$ (II.32)
providing $`v_k(k=1,2,\mathrm{},n)`$ satisfying the Bethe ansatz equations (BAE)
$`{\displaystyle \underset{j=1}{\overset{N}{}}}a(v_k,z_j)=1.`$ (II.33)
For late use, we define the state of the free fermion chain of length $`N`$
$`|a_1a_2\mathrm{}a_N>=|a_1>_{(1)}|a_2>_{(2)}\mathrm{}|a_N>_{(N)}`$ (II.34)
and its dual
$`|a_1a_2\mathrm{}a_N>^{}=<a_N|_{(N)}<a_{N1}|_{(N1)}\mathrm{}<a_1|_{(1)}<a_Na_{N1}\mathrm{}a_1|.`$ (II.35)
## III Drinfeld twists of the model
### III.1 Factorizing $`F`$-matrix and its inverse
Following , we now introduce the notation $`R_{1\mathrm{}N}^\sigma `$, where $`\sigma `$ is any element of the permutation group $`๐ฎ_N`$. We note that we may rewrite the GYBE as
$`R_{23}^{\sigma _{23}}T_{0,23}=T_{0,32}R_{23}^{\sigma _{23}},`$ (III.1)
where $`T_{0,23}R_{03}R_{02}`$ and $`\sigma _{23}`$ is the transposition of space labels (2,3). It follows that $`R_{1\mathrm{}N}^\sigma `$ is a product of elementary $`R`$-matrices , corresponding to a decomposition of $`\sigma `$ into elementary transpositions. With the help of the GYBE, one may generalize (III.1) to a $`N`$-fold tensor product of spaces
$`R_{1\mathrm{}N}^\sigma T_{0,1\mathrm{}N}=T_{0,\sigma (1\mathrm{}N)}R_{1\mathrm{}N}^\sigma ,`$ (III.2)
where $`T_{0,1\mathrm{}N}R_{0N}\mathrm{}R_{01}.`$ This implies the โdecompositionโ law
$`R_{1\mathrm{}N}^{\sigma ^{}\sigma }=R_{\sigma ^{}(1\mathrm{}N)}^\sigma R_{1\mathrm{}N}^\sigma ^{},`$ (III.3)
for a product of two elements in $`๐ฎ_N`$. Note that $`R_{\sigma ^{}(1\mathrm{}N)}^\sigma `$ satisfies the relation
$`R_{\sigma ^{}(1\mathrm{}N)}^\sigma T_{0,\sigma ^{}(1\mathrm{}N)}=T_{0,\sigma ^{}\sigma (1\mathrm{}N)}R_{\sigma ^{}(1\mathrm{}N)}^\sigma .`$ (III.4)
Let us write the elements of $`R_{1\mathrm{}N}^\sigma `$ as
$`\left(R_{1\mathrm{}N}^\sigma \right)_{\beta _N\mathrm{}\beta _1}^{\alpha _{\sigma (N)}\mathrm{}\alpha _{\sigma (1)}},`$ (III.5)
where the labels in the upper indices are permuted relative to the lower indices according to $`\sigma `$.
We proved in that for the $`R`$-matrix $`R_{1\mathrm{}N}^\sigma `$, there exists a non-degenerate lower-diagonal $`F`$-matrix (the Drinfeld twist) satisfying the relation
$`F_{\sigma (1\mathrm{}N)}(z_{\sigma (1)},\mathrm{},z_{\sigma (N)})R_{1\mathrm{}N}^\sigma (z_1,\mathrm{},z_N)=F_{1\mathrm{}N}(z_1,\mathrm{},z_N).`$ (III.6)
Explicitly,
$`F_{1,\mathrm{}N}={\displaystyle \underset{\sigma ๐ฎ_N}{}}{\displaystyle \underset{\alpha _{\sigma (1)}\mathrm{}\alpha _{\sigma (N)}}{\overset{}{}}}{\displaystyle \underset{j=1}{\overset{N}{}}}P_{\sigma (j)}^{\alpha _{\sigma (j)}}S(c,\sigma ,\alpha _\sigma )R_{1\mathrm{}N}^\sigma ,`$ (III.7)
where the sum $`^{}`$ is over all non-decreasing sequences of the labels $`\alpha _{\sigma (i)}`$:
$`\alpha _{\sigma (i+1)}\alpha _{\sigma (i)},\text{if}\sigma (i+1)>\sigma (i),`$
$`\alpha _{\sigma (i+1)}>\alpha _{\sigma (i)},\text{if}\sigma (i+1)<\sigma (i)`$ (III.8)
and the c-number function $`S(c,\sigma ,\alpha _\sigma )`$ is given by
$`S(c,\sigma ,\alpha _\sigma )\mathrm{exp}\left\{{\displaystyle \frac{1}{2}}{\displaystyle \underset{l>k=1}{\overset{N}{}}}\left(1(1)^{[\alpha _{\sigma (k)}]}\right)\delta _{\alpha _{\sigma (k)},\alpha _{\sigma (l)}}\mathrm{ln}(1+c_{\sigma (k)\sigma (l)})\right\}.`$ (III.9)
The inverse of the $`F`$-matrix is given by
$$F_{1\mathrm{}N}^1=F_{1\mathrm{}N}^{}\underset{i<j}{}\mathrm{\Delta }_{ij}^1$$
(III.10)
with
$`\mathrm{\Delta }_{ij}=\text{diag}((1+c_{ij})(1+c_{ji}),a_{ji},a_{ij},1)`$ (III.11)
and
$`F_{1\mathrm{}N}^{}`$ $`=`$ $`{\displaystyle \underset{\sigma ๐ฎ_N}{}}{\displaystyle \underset{\alpha _{\sigma (1)}\mathrm{}\alpha _{\sigma (N)}}{\overset{}{}}}S(c,\sigma ,\alpha _\sigma )R_{\sigma (1\mathrm{}N)}^{\sigma ^1}{\displaystyle \underset{j=1}{\overset{N}{}}}P_{\sigma (j)}^{\alpha _{\sigma (j)}},`$
where the sum $`^{}`$ is taken over all possible $`\alpha _i`$ which satisfies the following non-increasing constraints:
$`\alpha _{\sigma (i+1)}\alpha _{\sigma (i)},\text{if}\sigma (i+1)<\sigma (i),`$
$`\alpha _{\sigma (i+1)}<\alpha _{\sigma (i)},\text{if}\sigma (i+1)>\sigma (i).`$ (III.13)
### III.2 Symmetric representation of the Bethe state
The non-degeneracy of the $`F`$-matrix means that its column vectors form a complete basis, which is called the $`F`$-basis. By the procedure in , we find that in the $`F`$-basis, the simple generators of the superalgebra $`U_q(gl(1|1))`$ have the symmetric form:
$`\stackrel{~}{E}^{12}`$ $`=`$ $`F_{12\mathrm{}N}E^{12}F_{12\mathrm{}N}^1`$ (III.14)
$`=`$ $`{\displaystyle \underset{i=1}{\overset{N}{}}}E_{(i)}^{12}_{ji}\text{diag}(2e^\eta \mathrm{cosh}\eta ,e^\eta )_{(j)},`$
$`\stackrel{~}{E}^{21}`$ $`=`$ $`F_{12\mathrm{}N}E^{21}F_{12\mathrm{}N}^1`$ (III.15)
$`=`$ $`{\displaystyle \underset{i=1}{\overset{N}{}}}E_{(i)}^{12}_{ji}\text{diag}(e^\eta (2a_{ji}\mathrm{cosh}\eta )^1,e^\eta a_{ji}^1)_{(j)}.`$
Similarly, the diagonal element $`D(u)`$ of the monodromy matrix in the $`F`$-basis is given by
$`\stackrel{~}{D}(u)`$ $`=`$ $`F_{12\mathrm{}N}D(u)F_{12\mathrm{}N}^1=_{j=1}^N\text{diag}(a_{oj},1),`$ (III.16)
where $`a_{0j}a(u,z_j)`$.
Then, the creation and annihilation operators $`C(u)`$ and $`B(u)`$ read, in the $`F`$-basis,
$`\stackrel{~}{C}(u)`$ $`=`$ $`F_{12\mathrm{}N}C(u)F_{12\mathrm{}N}^1=(q^1\stackrel{~}{E}_{(i)}^{12}\stackrel{~}{D}(u)\stackrel{~}{D}(u)\stackrel{~}{E}_{(i)}^{12})q^{_{i=1}^Nh_{(i)}}`$ (III.17)
$`=`$ $`{\displaystyle \underset{i=1}{\overset{N}{}}}b_{0i}^{}E_{(i)}^{12}_{ji}\text{diag}(2a_{0j}\mathrm{cosh}\eta ,1)_{(j)},`$
$`\stackrel{~}{B}(u)`$ $`=`$ $`F_{12\mathrm{}N}B(u)F_{12\mathrm{}N}^1=q^{_{i=1}^Nh_{(i)}}(\stackrel{~}{E}^{21}\stackrel{~}{D}q\stackrel{~}{D}\stackrel{~}{E}^{21})`$ (III.18)
$`=`$ $`{\displaystyle \underset{i=1}{\overset{N}{}}}b_{0i}^+E_{(i)}^{21}_{ji}\text{diag}(a_{0j}(2a_{ji}\mathrm{cosh}\eta )^1,a_{ji}^1)_{(j)},`$
where $`b_{0j}^\pm b^\pm (u,z_j)`$, $`q=e^\eta `$ and $`hE^{11}E^{22}`$.
Acting the $`F`$-matrix (III.7) on the state (II.25), one sees that the pseudo-vacuum is invariant. Therefore in the $`F`$-basis, the Bethe state (II.22) becomes,
$`\stackrel{~}{\mathrm{\Phi }}_N(v_1,v_2,\mathrm{},v_n)F_{1\mathrm{}N}\mathrm{\Phi }_N(v_1,\mathrm{},v_n)={\displaystyle \underset{i=1}{\overset{n}{}}}\stackrel{~}{C}(v_n)|0>.`$ (III.19)
Substituting (III.17) into (III.19), we obtain
$`\stackrel{~}{\mathrm{\Phi }}_N(v_1,\mathrm{},v_n)=(2\mathrm{cosh}\eta )^{\frac{n(n1)}{2}}{\displaystyle \underset{i_1<\mathrm{}<i_n}{}}B_n^{}(v_1,\mathrm{},v_n|z_{i_1},\mathrm{},z_{i_n})E_{(i_1)}^{12}\mathrm{}E_{(i_n)}^{12}|0>,`$
(III.20)
where
$`B_n^\pm (v_1,\mathrm{},v_n|z_{i_1},\mathrm{},z_{i_n})`$ $`=`$ $`{\displaystyle \underset{\sigma ๐ฎ_n}{}}\text{sign}(\sigma ){\displaystyle \underset{k=1}{\overset{n}{}}}b^\pm (v_k,z_{i_{\sigma (k)}}){\displaystyle \underset{l=k+1}{\overset{n}{}}}a(v_k,z_{i_{\sigma (l)}})`$ (III.21)
$`=`$ $`\text{det}^\pm (\{v_k\},\{z_j\})`$
with $`^\pm (\{v_i\},\{z_j\})`$ being a $`n\times n`$ matrix with matrix elements
$$_{\alpha \beta }^\pm =b^\pm (v_\alpha ,z_\beta )\underset{\gamma =1}{\overset{\alpha 1}{}}a(v_\gamma ,z_\beta ).$$
(III.22)
Similarly, acting $`\stackrel{~}{B}(u_n)\mathrm{}\stackrel{~}{B}(u_1)`$ on the dual pseudo-vacuum state, we have,
$`<0|\stackrel{~}{B}(u_n)\mathrm{}\stackrel{~}{B}(u_1)`$ $`=`$ $`(1)^n(2\mathrm{cosh}\eta )^{\frac{n(n1)}{2}}{\displaystyle \underset{i_1<\mathrm{}<i_n}{}}{\displaystyle \underset{l=1}{\overset{n}{}}}{\displaystyle \underset{k=1,i_l}{\overset{N}{}}}a^1(z_k,z_{i_l})`$ (III.23)
$`\times \text{det}^+(\{v_k\},\{z_{i_j}\})<0|E_{(i_n)}^{21}\mathrm{}E_{(i_1)}^{21}.`$
## IV Determinant representation of the scalar product of the Bethe states
In , the authors gave the determinant representation of the scalar product of the Bethe state for the spin-1/2 XXZ model. In this section, we derive the determinant representation of the scalar product of the $`U_q(gl(1|1))`$ Bethe states defined by
$`S_n(\{u_j\},\{v_k\})=<0|B(u_n)\mathrm{}B(u_1)C(v_1)\mathrm{}C(v_n)|0>.`$ (IV.1)
The $`F`$-invariance of the pseudo-vacuum state $`|0>`$ and its dual state $`<0|`$ leads to
$`S_n(\{u_j\},\{v_k\})=<0|\stackrel{~}{B}(u_n)\mathrm{}\stackrel{~}{B}(u_1)\stackrel{~}{C}(v_1)\mathrm{}\stackrel{~}{C}(v_n)|0>.`$ (IV.2)
Following , we define
$`G^{(m)}(\{v_k\},u_1,\mathrm{},u_m,i_{m+1},\mathrm{},i_n)`$
$`=<i_n,\mathrm{},i_{m+1}|\stackrel{~}{B}(u_m)\mathrm{}\stackrel{~}{B}(u_1)\stackrel{~}{C}(v_1)\mathrm{}\stackrel{~}{C}(v_n)|0>,`$ (IV.3)
where $`i_k`$ $`(k=m+1,\mathrm{},n)`$, ordered as $`i_{m+1}<\mathrm{}<i_n`$, indicate the positions having state $`\left(\begin{array}{c}1\\ 0\end{array}\right)`$, and other positions have state $`\left(\begin{array}{c}0\\ 1\end{array}\right)`$ . One sees that when $`m=n`$, $`G^{(n)}=S_n`$. Inserting a complete set and noticing (III.18), (IV.3) becomes
$`G^{(m)}(\{v_k\},u_1,\mathrm{},u_m,i_{m+1},\mathrm{},i_n)`$
$`={\displaystyle \underset{ji_{m+1},\mathrm{},i_n}{\overset{N}{}}}<i_n,\mathrm{},i_{m+1}|\stackrel{~}{B}(u_m)|i_{m+1},\mathrm{},i_{m+p},j,i_{m+p+1},\mathrm{},i_n>`$
$`\times G^{(m1)}(\{v_k\},u_1,\mathrm{},u_{m1},i_{m+1},\mathrm{},i_{m+p},j,i_{m+p+1},\mathrm{},i_n).`$ (IV.4)
In view of (III.18), we have
$`<i_n,\mathrm{},i_{m+1}|\stackrel{~}{B}(u_m)|i_{m+1},\mathrm{},i_{m+p},j,i_{m+p+1},\mathrm{},i_n>`$
$`=(2\mathrm{cosh}\eta )^{(nm)}(1)^pb^+(u_m,z_j){\displaystyle \underset{kj}{\overset{N}{}}}a^1(z_k,z_j){\displaystyle \underset{l=m+1}{\overset{n}{}}}a(u_m,z_{i_l}).`$ (IV.5)
With the help of (III.20), we obtain $`G^{(0)}`$:
$`G^{(0)}(\{v_k\},i_1,\mathrm{},i_n)=<i_n,\mathrm{},i_1|{\displaystyle \underset{k=1}{\overset{n}{}}}\stackrel{~}{C}(v_k)|0>`$ (IV.6)
$`=`$ $`(2\mathrm{cosh}\eta )^{\frac{n(n1)}{2}}\text{det}^{}(\{v_k\},\{z_{i_l}\}).`$
We now compute $`G^{(1)}`$ by using the recursion relation (IV.4). Substituting (IV.5) and (IV.6) into (IV.4), we obtain
$`G^{(1)}(\{v_k\},u_1,i_2,\mathrm{},i_n)`$
$`={\displaystyle \underset{ji_2,\mathrm{},i_n}{\overset{N}{}}}<i_n,\mathrm{},i_2|\stackrel{~}{B}(u_1)|i_2,\mathrm{},i_{p+1},j,i_{p+2},\mathrm{},i_n>`$
$`\times G^{(0)}(\{v_k\},i_2,\mathrm{},i_{p+1},j,i_{p+2},\mathrm{},i_n)`$
$`=(2\mathrm{cosh}\eta )^{\frac{(n1)(n2)}{2}}{\displaystyle \underset{ji_2,\mathrm{},i_n}{\overset{N}{}}}(1)^pb^+(u_1,z_j){\displaystyle \underset{kj}{\overset{N}{}}}a^1(z_k,z_j){\displaystyle \underset{l=2}{\overset{n}{}}}a(u_1,z_{i_l})`$
$`\times \text{det}^{}(\{v_k\},z_{i_2},\mathrm{},z_{i_{p+1}},z_j,z_{i_{p+2}},\mathrm{},z_{i_n}),(k=1,\mathrm{},n).`$ (IV.7)
Let $`v_k(k=1,\mathrm{},n)`$ label the row and $`z_l(l=i_2,\mathrm{},j,\mathrm{},i_n)`$ label the column of the matrix $`^{}`$. From (IV.6), one sees that the column indices in (IV.7) satisfy the sequence $`i_2<\mathrm{}<j<\mathrm{}<i_n`$. Therefore, moving the column $`j`$ in the matrix $`^{}`$ to the first column, we have
$`G^{(1)}(\{v_k\},u_1,i_2,\mathrm{},i_n)`$
$`=(2\mathrm{cosh}\eta )^{\frac{(n1)(n2)}{2}}{\displaystyle \underset{ji_1,\mathrm{},i_n}{\overset{N}{}}}b^+(u_1,z_j){\displaystyle \underset{kj}{\overset{N}{}}}a^1(z_k,z_j){\displaystyle \underset{l=2}{\overset{n}{}}}a(u_1,z_{i_l})`$
$`\times \text{det}(\{v_k\},z_j,z_{i_2},\mathrm{},z_{i_n})`$
$`=(2\mathrm{cosh}\eta )^{\frac{(n1)(n2)}{2}}\text{det}(^{})^{(1)}(\{v_k\},u_1,z_{i_2},\mathrm{},z_{i_n}),`$ (IV.8)
where the matrix $`(^{})^{(1)}(\{v_k\},u_1,z_{i_2},\mathrm{},z_{i_n})`$ is given by
$`(_{\alpha \beta }^{})^{(1)}`$ $`=`$ $`a(u_1,z_{i_\beta })_{\alpha \beta }^{}\text{for}\beta 2,`$ (IV.9)
$`(_{\alpha 1}^{})^{(1)}`$ $`=`$ $`{\displaystyle \underset{ji_2,\mathrm{},i_n}{\overset{N}{}}}b^+(u_1,z_j)b^{}(v_\alpha ,z_j){\displaystyle \underset{\gamma =1}{\overset{\alpha 1}{}}}a(v_\gamma ,z_j){\displaystyle \underset{k=1,j}{\overset{N}{}}}a^1(z_k,z_j).`$ (IV.10)
Using the properties of determinant, one finds that if $`j=i_2,\mathrm{},i_n`$, the corresponding terms in (IV.10) contribute zero to the determinant. Thus, we may rewrite (IV.10) as
$`(_{\alpha 1}^{})^{(1)}`$ $`=`$ $`{\displaystyle \underset{j=1}{\overset{N}{}}}{\displaystyle \frac{e^{u_1v_\alpha }\mathrm{sinh}^2\eta }{\mathrm{sinh}(u_1z_j+\eta )\mathrm{sinh}(v_\alpha z_j+\eta )}}{\displaystyle \underset{\gamma =1}{\overset{\alpha 1}{}}}{\displaystyle \frac{\mathrm{sinh}(v_\gamma z_j)}{\mathrm{sinh}(v_\gamma z_j+\eta )}}`$ (IV.11)
$`\times {\displaystyle \underset{k=1,j}{\overset{N}{}}}{\displaystyle \frac{\mathrm{sinh}(z_kz_j+\eta )}{\mathrm{sinh}(z_kz_j)}}`$
$``$ $`e^{u_1}f(u_1).`$
Thanks to the Bethe ansatz equation (II.33), we may construct the function
$`_{\alpha \beta }^\pm `$ $`=`$ $`e^{u_\beta }g(u_\beta )`$ (IV.12)
$`=`$ $`{\displaystyle \frac{e^{\pm (v_\alpha u_\beta )}\mathrm{sinh}\eta }{\mathrm{sinh}(v_\alpha u_\beta )}}{\displaystyle \underset{\gamma =1}{\overset{\alpha 1}{}}}{\displaystyle \frac{\mathrm{sinh}(v_\gamma u_\beta \eta )}{\mathrm{sinh}(v_\gamma u_\beta )}}\left\{1{\displaystyle \underset{k=1}{\overset{N}{}}}{\displaystyle \frac{\mathrm{sinh}(u_\beta z_k)}{\mathrm{sinh}(u_\beta z_k+\eta )}}\right\}.`$
Comparing $`f(u_1)`$ in (IV.11) with $`g(u_1)`$ in (IV.12), one finds that as functions of $`u_1`$, they have the same residues at the simple pole $`u_1=z_j\eta `$ mod($`i\pi `$), and that when $`u_1\mathrm{}`$, they are bounded. Moreover, one may prove that the residues of $`g(u_1)`$ at $`u_1=v_\nu (\nu =1,\mathrm{},\alpha )`$ are zero because $`v_\nu `$ are solutions of the Bethe ansatz equation (II.33). Therefore, we have
$`(_{\alpha 1}^{})^{(1)}=_{\alpha 1}^{}={\displaystyle \frac{b^{}(v_\alpha ,u_1)}{a(v_\alpha ,u_1)}}{\displaystyle \underset{\gamma =1}{\overset{\alpha 1}{}}}a^1(u_1,v_\gamma )\left(1{\displaystyle \underset{k=1}{\overset{N}{}}}a(u_1,z_k)\right).`$ (IV.13)
Then, by using the function $`G^{(0)},G^{(1)}`$ and the intermediate function (IV.4) repeatedly, we obtain $`G^{(m)}`$ as
$`G^{(m)}(\{v_k\},u_1,\mathrm{},u_m,i_{m+1},\mathrm{},i_n)`$
$`=(1)^m(2\mathrm{cosh}\eta )^{\frac{n(n1)m(2nm1)}{2}}{\displaystyle \underset{1j<km}{}}a^1(u_k,u_j)`$
$`\times \text{det}(^{})^{(m)}(\{v_k\},u_1,\mathrm{},u_m,i_{m+1},\mathrm{},i_n)`$ (IV.14)
with the matrix elements
$`(_{\alpha \beta }^{})^{(m)}`$ $`=`$ $`{\displaystyle \underset{k=1}{\overset{m}{}}}a(u_k,z_{i_\beta })_{\alpha \beta }^{},\text{for }\beta >m,`$
$`(_{\alpha \beta }^{})^{(m)}`$ $`=`$ $`_{\alpha \beta }^{},\text{for }\beta m.`$ (IV.15)
(IV.14) can be proved by induction. Firstly from (IV.8), (IV.9) and (IV.13), (IV.14) is true for $`m=1`$. Assume (IV.14) for $`G^{(m1)}`$. Let us show (IV.14) for general $`m`$. Substituting $`G^{(m1)}`$ and (IV.5) into intermediate function (IV.4), we have
$`G^{(m)}(\{v_k\},u_1,\mathrm{},u_m,i_{m+1},\mathrm{},i_n)`$
$`={\displaystyle \underset{ji_{m+1},\mathrm{},i_n}{\overset{N}{}}}<i_n,\mathrm{},i_{m+1}|\stackrel{~}{B}(u_m)|i_{m+1},\mathrm{},i_{m+p},j,i_{m+p+1},\mathrm{},i_n>`$
$`\times G^{(m1)}(\{v_k\},u_1,\mathrm{},u_{m1},i_{m+1},\mathrm{},i_{m+p},j,i_{m+p+1},\mathrm{},i_n)`$
$`=(2\mathrm{cosh}\eta )^{(nm)}{\displaystyle \underset{ji_{m+1},\mathrm{},i_n}{\overset{N}{}}}b^+(u_m,z_j){\displaystyle \underset{kj}{\overset{N}{}}}a^1(z_k,z_j){\displaystyle \underset{l=m+1}{\overset{n}{}}}a(u_m,z_{i_l})`$
$`\times G^{(m1)}(\{v_k\},u_1,\mathrm{},u_{m1},j,i_{m+1},\mathrm{},i_n)`$
$`=(1)^m(2\mathrm{cosh}\eta )^{\frac{n(n1)m(2nm1)}{2}}{\displaystyle \underset{1j<km1}{}}a^1(u_k,u_j)`$
$`\times \text{det}_{}^{}{}_{}{}^{(m)}(\{v_k\},u_1\mathrm{},u_m,i_{m+1},\mathrm{},i_n),`$ (IV.16)
where the matrix elements $`_{}^{}{}_{\alpha \beta }{}^{(m)}`$ are given by
$`_{}^{}{}_{\alpha \beta }{}^{(m)}`$ $`=`$ $`{\displaystyle \underset{k=1}{\overset{m}{}}}a(u_k,z_{i_\beta })_{\alpha \beta }^{}\text{for }\beta >m,`$
$`_{}^{}{}_{\alpha \beta }{}^{(m)}`$ $`=`$ $`_{\alpha \beta }^{}\text{for }\beta <m,`$
$`_{}^{}{}_{\alpha m}{}^{(m)}`$ $`=`$ $`{\displaystyle \underset{i=1}{\overset{m1}{}}}a(u_i,z_j){\displaystyle \underset{ji_{m+1},\mathrm{},i_n}{}}b^+(u_m,z_j)b^{}(v_\alpha ,z_j){\displaystyle \underset{\gamma =1}{\overset{\alpha 1}{}}}a(v_\gamma ,z_j)`$ (IV.17)
$`\times {\displaystyle \underset{k=1,j}{\overset{N}{}}}a^1(z_k,z_j).`$
Thus, by the procedure leading to $`(_{\alpha \beta }^{})^{(1)}`$, we can prove
$`_{}^{}{}_{\alpha m}{}^{(m)}={\displaystyle \underset{i=1}{\overset{m1}{}}}a^1(u_m,u_i)_{\alpha m}^{}.`$ (IV.18)
Then one sees that $`_{}^{}{}_{\alpha \beta }{}^{(m)}=_{\alpha \beta }^{(m)}`$. Therefore we have proved that (IV.14) holds for all $`m`$.
When $`m=n`$, we obtain the scalar product $`S_n(\{u_i\},\{v_j\})`$,
$`S_n(\{u_i\},\{v_j\})=(1)^n{\displaystyle \underset{k>l}{\overset{n}{}}}a^1(u_k,u_l)\text{det}^{}(\{v_j\},\{u_i\}),`$ (IV.19)
where the matrix elements of $`^{}`$ are given by
$`_{\alpha \beta }^\pm ={\displaystyle \frac{b^\pm (v_\alpha ,u_\beta )}{a(v_\alpha ,u_\beta )}}{\displaystyle \underset{\gamma =1}{\overset{\alpha 1}{}}}a^1(u_\beta ,v_\gamma )\left(1{\displaystyle \underset{k=1}{\overset{N}{}}}a(u_\beta ,z_k)\right).`$ (IV.20)
By using the expression of the eigenvalues of the system (II.32), the scalar product (IV.19) can be rewritten as
$`S_n(\{u_i\},\{v_j\})=(1)^n{\displaystyle \underset{k>l}{\overset{n}{}}}a^1(u_k,u_l)\text{det}\widehat{}^{}(\{v_j\},\{u_i\})`$ (IV.21)
with the matrix $`\widehat{}^\pm `$ being
$`\widehat{}_{\alpha \beta }^\pm =e^{\pm (v_\alpha u_\beta )}\mathrm{sinh}(u_\beta v_\alpha ){\displaystyle \underset{\mu \alpha }{}}a(v_\mu ,u_\beta ){\displaystyle \underset{\gamma =1}{\overset{\alpha 1}{}}}a^1(u_\alpha ,v_\gamma ){\displaystyle \frac{\mathrm{\Lambda }(u_\beta ,\{v_\alpha \})}{v_\alpha }}.`$ (IV.22)
Remark: In the derivation of (IV.19), the parameters $`v_i`$ in the state $`\stackrel{~}{C}(v_1)\mathrm{}\stackrel{~}{C}(v_n)|0>`$ are required to satisfy the BAE (II.33). However, the parameters $`u_j`$ $`(j=1,\mathrm{},n)`$ in the dual state $`<0|\stackrel{~}{B}(u_n)\mathrm{}\stackrel{~}{B}(u_1)`$ do not need to satisfy the BAE.
On the other hand, if we compute the scalar product by starting from the dual state $`<0|B(v_n)\mathrm{}B(v_1)`$, then by using the same procedure, we have
$`S_n(\{v_i\},\{u_j\})`$ $`=`$ $`<0|\stackrel{~}{B}(v_n)\mathrm{}\stackrel{~}{B}(v_1)\stackrel{~}{C}(u_1)\mathrm{}\stackrel{~}{C}(u_n)|0>`$ (IV.23)
$`=`$ $`(1)^n{\displaystyle \underset{k>l}{\overset{n}{}}}a^1(u_k,u_l)\text{det}^+(\{v_i\},\{u_j\}).`$
In the above equation, we have assumed that $`\{v_i\}`$ satisfy the BAE.
Noticing the BAE (II.33), one sees that the scalar product $`S_n(\{u_i\},\{v_j\})=0`$ if both parameter sets $`\{u_i\}`$ and $`\{v_j\}(\{v_j\}\{u_i\}`$ $`i,j=1,\mathrm{},n)`$ in (IV.19) and (IV.23) satisfy the BAE.
Let $`u_\alpha v_\alpha `$ $`(\alpha =1,\mathrm{},n)`$ in (IV.19), we obtain the Gaudin formula for the norm of the $`U_q(gl(1|1))`$ Bethe state.
$`๐ฎ_n`$ $`=`$ $`S_n(\{v_j\},\{v_k\})=<0|B(v_n)\mathrm{}B(v_1)C(v_1)\mathrm{}C(v_n)|0>`$ (IV.24)
$`=`$ $`(1)^n\mathrm{sinh}^n\eta {\displaystyle \underset{k>j}{\overset{n}{}}}{\displaystyle \frac{\mathrm{sinh}^2(v_kv_j+\eta )}{\mathrm{sinh}^2(v_kv_j)}}\left[{\displaystyle \underset{\alpha =1}{\overset{n}{}}}{\displaystyle \frac{1}{v_\alpha u_\alpha }}\left(1{\displaystyle \underset{l=1}{\overset{N}{}}}{\displaystyle \frac{\mathrm{sinh}(u_\alpha z_l)}{\mathrm{sinh}(u_\alpha z_l+\eta )}}\right)\right]_{u_\alpha v_\alpha }`$
$`=`$ $`(1)^n\mathrm{sinh}^n\eta {\displaystyle \underset{k>j}{\overset{n}{}}}{\displaystyle \frac{\mathrm{sinh}^2(v_kv_j+\eta )}{\mathrm{sinh}^2(v_kv_j)}}\left[{\displaystyle \underset{\alpha =1}{\overset{n}{}}}{\displaystyle \frac{}{u_\alpha }}\mathrm{ln}\left({\displaystyle \underset{l=1}{\overset{N}{}}}{\displaystyle \frac{\mathrm{sinh}(u_\alpha z_l)}{\mathrm{sinh}(u_\alpha z_l+\eta )}}\right)\right]_{u_\alpha v_\alpha }`$
$`=`$ $`(1)^n\mathrm{sinh}^{2n}\eta {\displaystyle \underset{k>j}{\overset{n}{}}}{\displaystyle \frac{\mathrm{sinh}^2(v_kv_j+\eta )}{\mathrm{sinh}^2(v_kv_j)}}{\displaystyle \underset{\alpha =1}{\overset{n}{}}}{\displaystyle \underset{l=1}{\overset{N}{}}}{\displaystyle \frac{1}{\mathrm{sinh}(v_\alpha z_l)\mathrm{sinh}(v_\alpha z_l+\eta )}},`$
where we have used the BAE (II.33).
## V Correlation functions
Having obtained the scalar product and the norm, we are now in the position to compute the k-point correlation functions of the model. In general, a k-point correlation function is defined by
$`F_n^{ฯต^1,\mathrm{},ฯต^k}=<0|B(u_n)\mathrm{}B(u_1)ฯต_{i_1}^1\mathrm{}ฯต_{i_k}^kC(v_1)\mathrm{}C(v_n)|0>,`$ (V.1)
where $`ฯต_{i_j}^j`$ stand for the local fermion operators $`c_{i_j},c_{i_j}^{}`$ or $`n_{i_j}`$, and the lower indices $`i_j`$ indicate the positions of the fermion operators.
The authors in proved that the local spin and field operators of the fundamental graded models can be represented in terms of monodromy matrix. Specializing to the current system, we obtain
$`c_j^{}`$ $`=`$ $`{\displaystyle \underset{k=1}{\overset{j1}{}}}(A(z_k)+D(z_k))B(z_j){\displaystyle \underset{k=j+1}{\overset{N}{}}}(A(z_k)+D(z_k)),`$ (V.2)
$`c_j`$ $`=`$ $`{\displaystyle \underset{k=1}{\overset{j1}{}}}(A(z_k)+D(z_k))C(z_j){\displaystyle \underset{k=j+1}{\overset{N}{}}}(A(z_k)+D(z_k)),`$ (V.3)
$`n_j`$ $`=`$ $`{\displaystyle \underset{k=1}{\overset{j1}{}}}(A(z_k)+D(z_k))D(z_j){\displaystyle \underset{k=j+1}{\overset{N}{}}}(A(z_k)+D(z_k)).`$ (V.4)
### V.1 One point functions
In this subsection, we compute the one point functions for the local operators $`c_m^{},c_m`$ and $`n_m`$, respectively.
We first calculate $`c_m^{}`$ . Noticing that the Bethe state and its dual are eigenstates of the transfer matrix under the constraint of the BAE, we have, from (V.2),
$`F_n^{}(\{u_j\},z_m,\{v_k\})`$ (V.5)
$`=`$ $`<0|B(u_n)\mathrm{}B(u_1)c_m^{}C(v_1)\mathrm{}C(v_{n+1})|0>`$
$`=`$ $`\varphi _{m1}(\{u_j\})\varphi _m^1(\{v_k\})<0|B(u_n)\mathrm{}B(u_1)B(z_m)C(v_1)\mathrm{}C(v_{n+1})|0>`$
$`=`$ $`\varphi _{m1}(\{u_j\})\varphi _m^1(\{v_k\})<0|\stackrel{~}{B}(u_n)\mathrm{}\stackrel{~}{B}(u_1)\stackrel{~}{B}(z_m)\stackrel{~}{C}(v_1)\mathrm{}\stackrel{~}{C}(v_{n+1})|0>`$
$`=`$ $`\varphi _{m1}(\{u_j\})\varphi _m^1(\{v_k\})S_{n+1}(u_n,\mathrm{},u_1,z_m,\{v_j\})`$
$`=`$ $`(1)^{n+1}\varphi _{m1}(\{u_j\})\varphi _m^1(\{v_k\}){\displaystyle \underset{k>j}{\overset{n}{}}}a^1(u_k,u_j){\displaystyle \underset{l=1}{\overset{n}{}}}a^1(u_l,z_m)`$
$`\times \text{det}^{}(\{v_j\},z_m,u_1,\mathrm{},v_n),`$
where $`\varphi _i(\{u_j\})=_{k=1}^i_{l=1}^na(u_l,u_k)`$. As mentioned in the remark of the previous section, $`F_n^{}=0`$ if the parameter set $`\{u_i\}`$ $`(i=1,\mathrm{},n)`$ is not a subset of $`\{v_j\}`$ $`(j=1,\mathrm{},n+1)`$. When $`\{u_i\}\{v_j\}`$, (V.5) can be simplified to a simple function. For example, if $`u_i=v_{i+1}`$ $`(i=1,\mathrm{},n)`$, the one point function $`F^{}`$ becomes
$`F_n^{}(v_{n+1},\mathrm{},v_2,z_m,v_1,\mathrm{},v_{n+1})`$ (V.6)
$`=`$ $`(1)^{n+1}{\displaystyle \frac{\varphi _{m1}(\{u_j\})}{\varphi _m(\{v_k\})}}{\displaystyle \frac{e^{(v_1z_m)}\mathrm{sinh}^{2n+1}\eta }{\mathrm{sinh}(v_1z_m)}}{\displaystyle \underset{k>j=2}{\overset{n+1}{}}}{\displaystyle \frac{\mathrm{sinh}^2(v_kv_j+\eta )}{\mathrm{sinh}^2(v_kv_j)}}{\displaystyle \underset{j=2}{\overset{n+1}{}}}{\displaystyle \frac{\mathrm{sinh}(v_jz_m+\eta )}{\mathrm{sinh}(v_jz_m)}}`$
$`\times {\displaystyle \underset{j=2}{\overset{n+1}{}}}{\displaystyle \frac{\mathrm{sinh}(v_jv_1+\eta )}{\mathrm{sinh}(v_jv_1)}}{\displaystyle \underset{\alpha =2}{\overset{n+1}{}}}{\displaystyle \underset{l=1}{\overset{N}{}}}{\displaystyle \frac{1}{\mathrm{sinh}(v_\alpha z_l)\mathrm{sinh}(v_\alpha z_l+\eta )}}.`$
Similarly, when $`\{u_i\}\{v_j\}`$, we obtain the one-point function involving the operator $`c_m`$:
$`F_n^+(\{v_k\},z_m,\{u_j\})`$ (V.7)
$`=`$ $`<0|B(v_{n+1})\mathrm{}B(v_1)c_mC(u_1)\mathrm{}C(u_n)|0>`$
$`=`$ $`\varphi _{m1}(\{v_j\})\varphi _m^1(\{u_k\})S_{n+1}(\{v_j\},z_m,u_1,\mathrm{},u_n)`$
$`=`$ $`(1)^{n+1}\varphi _{m1}(\{v_j\})\varphi _m^1(\{u_k\}){\displaystyle \underset{k>j}{\overset{n}{}}}a^1(u_k,u_j){\displaystyle \underset{l=1}{\overset{n}{}}}a^1(u_l,z_m)`$
$`\times \text{det}^+(\{v_j\},z_m,u_1,\mathrm{},v_n).`$
$`F_n^+`$ is non-vanishing if $`\{u_i\}\{v_j\}`$. When $`\{u_i\}\{v_j\}`$, (V.7) can also be simplified to a simple function. In the case $`u_i=v_{i+1}`$ $`(i=1,\mathrm{},n)`$, the one point function $`F^+`$ becomes
$`F_n^+(v_{n+1},\mathrm{},v_2,z_m,v_1,\mathrm{},v_{n+1})`$ (V.8)
$`=`$ $`(1)^{n+1}{\displaystyle \frac{\varphi _{m1}(\{v_j\})}{\varphi _m(\{u_k\})}}{\displaystyle \frac{e^{(v_1z_m)}\mathrm{sinh}^{2n+1}\eta }{\mathrm{sinh}(v_1z_m)}}{\displaystyle \underset{k>j=2}{\overset{n+1}{}}}{\displaystyle \frac{\mathrm{sinh}^2(v_kv_j+\eta )}{\mathrm{sinh}^2(v_kv_j)}}{\displaystyle \underset{j=2}{\overset{n+1}{}}}{\displaystyle \frac{\mathrm{sinh}(v_jz_m+\eta )}{\mathrm{sinh}(v_jz_m)}}`$
$`\times {\displaystyle \underset{j=2}{\overset{n+1}{}}}{\displaystyle \frac{\mathrm{sinh}(v_jv_1+\eta )}{\mathrm{sinh}(v_jv_1)}}{\displaystyle \underset{\alpha =2}{\overset{n+1}{}}}{\displaystyle \underset{l=1}{\overset{N}{}}}{\displaystyle \frac{1}{\mathrm{sinh}(v_\alpha z_l)\mathrm{sinh}(v_\alpha z_l+\eta )}}.`$
The one-point function involving the operator $`n_m`$ is defined by
$`F_n^{n_m}(\{u_j\},z_m,\{v_k\})=<0|B(u_n)\mathrm{}B(u_1)n_mC(v_1)\mathrm{}C(v_{n+1})|0>.`$ (V.9)
Substituting (V.4) into the above equation and considering the BAE, we have
$`F_n^{n_m}(\{u_j\},z_m,\{v_k\})=<0|B(u_n)\mathrm{}B(u_1)n_mC(v_1)\mathrm{}C(v_n)|0>`$ (V.10)
$`=`$ $`{\displaystyle \frac{\varphi _{m1}(\{u_j\})}{\varphi _{m1}(\{v_k\})}}<0|\stackrel{~}{B}(u_n)\mathrm{}\stackrel{~}{B}(u_1)\stackrel{~}{D}(z_m)\stackrel{~}{C}(v_1)\mathrm{}\stackrel{~}{C}(v_n)|0>.`$
With the help of (II.30), we see
$`D(z_m)C(v_1)\mathrm{}C(v_n)|0>`$
$`=`$ $`{\displaystyle \underset{k=1}{\overset{n}{}}}a^1(v_k,z_m)C(v_1)\mathrm{}C(v_n)|0>`$
$`{\displaystyle \underset{j=1}{\overset{n}{}}}{\displaystyle \frac{b^{}(v_j,z_m)}{a(v_j,z_m)}}{\displaystyle \underset{l=1}{\overset{j1}{}}}{\displaystyle \frac{c(v_l,v_j)}{c(v_l,z_m)}}{\displaystyle \underset{k=1,j}{\overset{n}{}}}a^1(v_k,v_j)`$
$`\times C(v_1)\mathrm{}C(v_{j1})C(z_m)C(v_{j+1})\mathrm{}C(v_n)|0>.`$
Therefore, substituting (LABEL:eq:commu-DCn) into (V.10), we obtain
$`F_n^{n_m}(\{u_j\},z_m,\{v_k\})`$ (V.12)
$`=`$ $`{\displaystyle \frac{\varphi _{m1}(\{u_j\})}{\varphi _{m1}(\{v_k\})}}{\displaystyle \underset{k=1}{\overset{n}{}}}a^1(v_k,z_m)S_n(\{u_i\},\{v_j\})`$
$`{\displaystyle \underset{j=1}{\overset{n}{}}}{\displaystyle \frac{b^{}(v_j,z_m)}{a(v_j,z_m)}}{\displaystyle \underset{l=1}{\overset{j1}{}}}{\displaystyle \frac{c(v_l,v_j)}{c(v_l,z_m)}}{\displaystyle \underset{k=1,j}{\overset{n}{}}}a^1(v_k,v_j)S_n(\{u_i\},v_1,\mathrm{},v_{j1},z_m,v_{j+1},\mathrm{},v_n)`$
$`=`$ $`(1)^n{\displaystyle \frac{\varphi _{m1}(\{u_j\})}{\varphi _{m1}(\{v_k\})}}{\displaystyle \underset{k=1}{\overset{n}{}}}a^1(v_k,z_m){\displaystyle \underset{k>j}{}}a^1(v_k,v_j)`$
$`\times \text{det}\left[M^+(\{u_i\},\{v_j\})๐ฉ(\{u_i\},\{v_j\},z_m)\right],`$
where $`๐ฉ`$ is a rank-one matrix with the following matrix elements
$`๐ฉ_{\alpha \beta }(\{u_i\},\{v_j\},z_m)`$ $`=`$ $`{\displaystyle \frac{e^{u_\alpha v_\beta }\mathrm{sinh}^2\eta }{\mathrm{sinh}(u_\alpha z_m)\mathrm{sinh}(v_\beta z_m+\eta )}}{\displaystyle \underset{i=1}{\overset{\alpha 1}{}}}{\displaystyle \frac{\mathrm{sinh}(z_mu_i+\eta )}{\mathrm{sinh}(z_mu_i)}}.`$ (V.13)
In the above derivation, we have used the following property of determinant: If $`๐`$ is an arbitrary $`n\times n`$ matrix and $``$ is a rank-one $`n\times n`$ matrix, then the determinant of $`๐+`$ is given by
$`\text{det}(๐+)=\text{det}๐+{\displaystyle \underset{i=1}{\overset{n}{}}}\text{det}๐^{(i)},`$ (V.14)
where
$`๐_{\alpha \beta }^{(i)}=๐_{\alpha \beta }\text{ for }\beta i,`$
$`๐_{\alpha i}^{(i)}=_{\alpha i}.`$
### V.2 Correlation function of two adjacent operators
In the subsection, we compute the correlation function of two adjacent operators $`c_m`$ and $`c_{m+1}`$ defined by
$`F_n^+(\{u_i\},z_m,z_{m+1},\{v_j\})=<0|B(u_n)\mathrm{}B(u_1)c_mc_{m+1}^{}C(v_1)\mathrm{}C(v_n)|0>.`$ (V.15)
Substituting (V.3) and (V.2) into the above definition and considering the fact $`_{k=1}^Nt(z_k)=1`$, we have
$`F_n^+(\{u_i\},\{v_j\},z_m,z_{m+1})`$
$`={\displaystyle \frac{\varphi _{m1}(\{u_i\})}{\varphi _{m+1}(\{v_j\})}}<0|\stackrel{~}{B}(u_n)\mathrm{}\stackrel{~}{B}(u_1)\stackrel{~}{C}(z_m)\stackrel{~}{B}(z_{m+1})\stackrel{~}{C}(v_1)\mathrm{}\stackrel{~}{C}(v_n)|0>.`$ (V.16)
By using the commutation relation (II.31), we obtain
$`B(z_{m+1})C(v_1)\mathrm{}C(v_n)|0>=(1)^nC(v_1)\mathrm{}C(v_n)B(z_{m+1})|0>`$
$`+{\displaystyle \underset{j=1}{\overset{n}{}}}(1)^{j+1}{\displaystyle \frac{b^+(z_{m+1},v_j)}{a(z_{m+1},v_j)}}C(v_1)\mathrm{}C(v_{j1})D(z_{m+1})t(v_j)C(v_{j+1})C(v_n)|0>`$
$`+{\displaystyle \underset{j=1}{\overset{n}{}}}(1)^j{\displaystyle \frac{b^+(z_{m+1},v_j)}{a(z_{m+1},v_j)}}C(v_1)\mathrm{}C(v_{j1})t(z_{m+1})D(v_j)C(v_{j+1})C(v_n)|0>,`$ (V.17)
where $`\stackrel{~}{t}(u)F_{1\mathrm{}N}t(u)F_{1\mathrm{}N}^1`$. On the rhs of the above equation, one easily finds that the first term is zero. Using the BAE, one may check that the second term also equals to zero. Therefore, only the third term survives on the rhs of the above equation and we have
$`B(z_{m+1})C(v_1)\mathrm{}C(v_n)|0>`$
$`={\displaystyle \underset{j=1}{\overset{n}{}}}(1)^j{\displaystyle \frac{b^+(z_{m+1},v_j)}{a(z_{m+1},v_j)}}{\displaystyle \underset{k=j+1}{\overset{n}{}}}a^1(v_k,z_{m+1}){\displaystyle \underset{l=j+1}{\overset{n}{}}}a^1(v_l,v_j)`$
$`\times C(v_1)\mathrm{}C(v_{j1})C(v_{j+1})C(v_n)|0>`$
$`+{\displaystyle \underset{j=1}{\overset{n}{}}}(1)^{j+1}{\displaystyle \frac{b^+(z_{m+1},v_j)}{a(z_{m+1},v_j)}}{\displaystyle \underset{k=j+1}{\overset{n}{}}}a^1(v_k,z_{m+1})`$
$`\times {\displaystyle \underset{l=j+1}{\overset{n}{}}}{\displaystyle \frac{b^{}(v_l,v_j)}{a(v_l,v_j)}}{\displaystyle \underset{m=j+1}{\overset{l1}{}}}{\displaystyle \frac{c(v_m,v_l)}{c(v_m,v_j)}}{\displaystyle \underset{i=j+1,l}{\overset{n}{}}}a^1(v_i,v_l)`$
$`\times C(v_1)\mathrm{}C(v_{j1})C(v_{j+1})\mathrm{}C(v_{l1})C(v_j)C(v_{l+1})\mathrm{}C(v_n)|0>`$
$`{\displaystyle \underset{j=1}{\overset{n}{}}}M_jC(v_1)\mathrm{}C(v_{j1})C(v_{j+1})C(v_n)|0>`$
$`+{\displaystyle \underset{j=1}{\overset{n}{}}}{\displaystyle \underset{l=j+1}{\overset{n}{}}}M_{j,l}C(v_1)\mathrm{}C(v_{j1})C(v_{j+1})\mathrm{}C(v_{l1})C(v_j)C(v_{l+1})\mathrm{}C(v_n)|0>.`$
(V.18)
Substituting (V.18) into (V.16), we obtain two-point correlation function $`F_n^+`$
$`F_n^+(\{u_i\},z_m,z_{m+1},\{v_j\})`$
$`={\displaystyle \frac{\varphi _{m1}(\{u_i\})}{\varphi _{m+1}(\{v_j\})}}[{\displaystyle \underset{j=1}{\overset{n}{}}}M_jS_n(\{u_i\},z_m,v_1,\mathrm{},v_{j1},v_{j+1},v_n)`$
$`+{\displaystyle \underset{j=1}{\overset{n}{}}}{\displaystyle \underset{l=j+1}{\overset{n}{}}}M_{j,l}S_n(\{u_i\},z_m,v_1,\mathrm{},v_{j1},v_{j+1},\mathrm{},v_{l1},v_j,v_{l+1},\mathrm{},v_n)].`$ (V.19)
## VI Discussion
In this paper, with the help of the factorizing $`F`$-matrix ($`F`$-basis), we have obtained the determinant representations of the scalar products and correlation functions of the $`U_q(gl(1|1))`$ free fermion model.
In -, the authors studied the correlation functions of the free fermion model based on the finite XX0 spin chain (XY model ) with periodic boundary condition
$`H_{XX0}={\displaystyle \underset{j=1}{\overset{N}{}}}\left(\sigma _j^x\sigma _{j+1}^x+\sigma _j^y\sigma _{j+1}^y+h\sigma _j^z\right),`$ (VI.1)
where $`\sigma ^ฯต(ฯต=x,y,z)`$ are the Pauli matrices and $`h`$ is an external classical magnetic field. The equivalence between the free fermion model and the XX0 model can be proved by using the Jordan-Wigner transform
$`c_k=\mathrm{exp}[i\pi Q_{k1}]\sigma _k^+,`$ (VI.2)
$`c_k^{}=\sigma _k^{}\mathrm{exp}[i\pi Q_{k1}],`$ (VI.3)
where $`\sigma ^\pm =\frac{1}{2}(\sigma ^x\pm \sigma ^y)`$, $`Q_k=_{j=1}^k\frac{1}{2}(1\sigma _k^z)`$. Because of the periodic boundary condition of the finite XX0 spin chain, we have
$$\sigma _{N+1}^\pm =\sigma _1^\pm .$$
(VI.4)
Substituting the Jordan-Wigner transforms into the above relation, we obtain
$`c_{N+1}=\mathrm{exp}[i\pi Q_N]c_1,c_{N+1}^{}=c_1^{}\mathrm{exp}[i\pi Q_N].`$ (VI.5)
Thus, comparing the above boundary condition with that of the $`U_q(gl(1|1))`$ free fermion model (II.21), we find that the free fermion model arising from the XX0 model has a twisted boundary condition which depends on the operator $`\sigma ^z=_{i=1}^N\sigma _i^z`$.
On the other hand, by means of the Jordan-Wigner transform, the $`U_q(gl(1|1))`$ free fermion model is equivalent to a twisted XX0 model, and the one-point correlation functions (V.5) and (V.7) give rise to the $`m`$-point correlation functions of the twisted XX0 model. For example: substituting (VI.3) into (V.5), we obtain
$`F_n^{}(\{u_j\},z_m,\{v_k\})`$ (VI.6)
$`=`$ $`<0|B(u_n)\mathrm{}B(u_1)c_m^{}C(v_1)\mathrm{}C(v_{n+1})|0>`$
$`=`$ $`<0|B(u_n)\mathrm{}B(u_1)\sigma _1^z\mathrm{}\sigma _{m1}^z\sigma _m^{}C(v_1)\mathrm{}C(v_{n+1})|0>.`$
Acknowledgements: This work was financially supported by the Australian Research Council. S.Y. Zhao was supported by the UQ Postdoctoral Research Fellowship.
|
warning/0506/cond-mat0506601.html
|
ar5iv
|
text
|
# Quantum transitions induced by the third cumulant of current fluctuations
## Abstract
We investigate the transitions induced by non-Gaussian external fluctuations on a small quantum system. The rates for the transitions between the energy states are calculated using the real-time Keldysh formalism for the density matrix evolution. We detail the effects of the third cumulant of current fluctuations coupled to a quantum system with a discrete level spectrum and propose a setup for detecting the frequency-dependent third cumulant through the transitions it induces. We especially discuss a scheme where the fluctuations are coupled to a Josephson flux qubit.
The study of fluctuations has been in the center of interest in physics for decades. The relevance of noise and fluctuations is underlined by the fundamental relation between fluctuations and dissipation in physical systems. One very concrete example of fluctuations is the current noise in electric circuits. At equilibrium, it obeys the fluctuation-dissipation theorem which relates the magnitude of fluctuations to the temperature and the impedance of the circuit. For a quantum system with a finite number of levels interacting with an environment, the magnitude of these fluctuations in the environment then determines the steady state of the system, along with the rate with which this steady state is approached.
During the past decade, the theory of electric fluctuations in mesoscopic systems has been significantly developed to characterize them also out of equilibrium, q ; SZ where a finite average current leads to shot noise. The study yields information about the microscopic physical phenomena inside electric conductors and the effects of the electromagnetic environment on mesoscopic circuits. In large wires the current statistics is Gaussian and fully characterized by the average current and the noise power. The experimental development in manufacturing smaller circuits has enabled the study of the non-Gaussian character of fluctuations in mesoscopic samples. ReuletBomze ; lindell In principle, the knowledge of these fluctuations allows for an improved characterization of the conductors, q or the study of the effect their non-Gaussian character causes on other mesoscopic systems. tobiska ; sonin ; heikkila ; ankerhold
With a nonvanishing average current, the probability distribution of current fluctuations no longer needs to be symmetric around the average current. In particular, the third cumulant of fluctuations describing the skewness of the current distribution may be finite. It is also the lowest cumulant indicating a non-Gaussian distribution. Despite the strong theoretical effort describing the nature of the higher-order cumulants,q measuring even the third cumulant with conventional techniques has turned out to be difficult and so far its only measurements exist for the case of a tunnel junction. ReuletBomze The attention is thus turning towards using other mesoscopic systems as fluctuation detectors. deblock ; tobiska ; heikkila ; pekola ; schoelkopf
In this Paper we analyze the transitions caused by external fluctuations on a probe quantum system. First, we present a formula correcting the Golden Rule transition rates by taking into account the next order effects that are dependent on the third cumulant. This is essential in developing generic methods for detecting non-Gaussian fluctuations. We can establish conditions imposed to suitable probes of third-cumulant induced excitations. Although we concentrate on current fluctuations, our general analysis is independent of the physical system as long as the fluctuations are linearly coupled to the probe system. To demonstrate the results, we consider a quantum two-state system (qubit) as a probe candidate and propose a setup for measuring the effects of the frequency-dependent third cumulant of current fluctuations by a Josephson flux qubit.mooij99 ; makhlin This can be viewed as a generalization of using qubits as spectrometers of the quantum noise power, schoelkopf a method which has already been experimentally demonstrated.astafiev
Our starting point is the Hamiltonian
$`H=H_{\mathrm{ext}}+H_s+H_{\mathrm{int}},`$ (1)
where $`H_{\mathrm{ext}}`$ and $`H_s`$ describe the environment where the current fluctuates and the quantum system we use as a probe for the fluctuations, respectively, and $`H_{\mathrm{int}}`$ is the interaction Hamiltonian between the environment and the probe. Motivated by the case of a current-biased Josephson junction and the magnetic interaction between two circuits considered below, we study the bilinear coupling of the form $`H_{\mathrm{int}}=g\delta I\varphi `$. Here $`\delta I`$ is the current fluctuation operator acting on the environment, $`\varphi `$ is an operator acting on the probe system and $`g`$ is the coupling constant of the interaction. We assume that the quantum system is described by a set of energy eigenstates $`\{|n\}`$ and the average current effect $`gI\varphi `$ is included in $`H_s`$. Treating $`H_{\mathrm{int}}`$ as a perturbation, the Fermi Golden Rule predicts the transition rate $`\mathrm{\Gamma }_{nn^{}}^{(2)}=\frac{2\pi g^2}{\mathrm{}^2}|\varphi _{nn^{}}|^2S_{\delta I}(\frac{E_nE_n^{}}{\mathrm{}})`$ between the eigenstates of the probe system.schoelkopf The matrix element is defined as $`\varphi _{nn^{}}=n|\varphi |n^{}`$ and the noise power $`S_{\delta I}(\omega )=\frac{1}{2\pi }_{\mathrm{}}^{\mathrm{}}e^{i\omega t}\delta I(t)\delta I(0)๐t`$. The lowest-order estimate $`\mathrm{\Gamma }^{(2)}`$ is thus proportional to the second cumulant of current fluctuations. The correlator in the above expression is calculated with respect to the environment Hamiltonian $`H_{\mathrm{ext}}`$ as if the probe system did not exist. Below, we correct the transition rate $`\mathrm{\Gamma }^{(2)}`$ by calculating the next order contribution $`\mathrm{\Gamma }^{(3)}`$, depending on the third cumulant.
We solve the density matrix evolution using the real-time Keldysh method, as outlined in Refs. schoeller, ; feynman, , which is a natural formalism for studying a small subsystem in a larger environment. We are interested in the dynamics of the probe system in particular, so we study the reduced density operator $`\rho (t)=\mathrm{Tr}_{\mathrm{ext}}\rho _{\mathrm{tot}}(t)`$ where $`\rho _{\mathrm{tot}}`$ is the density operator for the system and the environment. The trace goes over a complete set of environment states. The idea is to solve the temporal evolution of a diagonal element of the reduced density matrix $`\rho _{n^{}n^{}}(t)=n^{}|\rho (t)|n^{}`$ with the initial condition $`\rho _{nn}(t_0)=1`$ $`(nn^{})`$. In the long-time limit $`\rho _{n^{}n^{}}(t)`$ is proportional to the total evolution time $`tt_0`$, the coefficient being the transition rate $`\mathrm{\Gamma }_{nn^{}}`$. We calculate the rates between well-specified states of the reduced system. Therefore, without loss of generality, we use an initial state of the form $`\rho _{\mathrm{tot}}(t_0)=\rho _{\mathrm{ext}}(t_0)\rho (t_0)`$, where $`\rho _{\mathrm{ext}}(t_0)`$ describes the initial state of the environment.
As the lowest-order contribution to $`\mathrm{\Gamma }_{nn^{}}`$ is the well-known Golden-Rule result $`\mathrm{\Gamma }_{nn^{}}^{(2)}`$, we concentrate on the next order contribution $`\mathrm{\Gamma }_{nn^{}}^{(3)}`$. The total rate is then given by $`\mathrm{\Gamma }_{nn^{}}=\mathrm{\Gamma }_{nn^{}}^{(2)}+\mathrm{\Gamma }_{nn^{}}^{(3)}`$. Treating $`H_{\mathrm{int}}=g\delta I\varphi `$ as a perturbation and using the graphical rules derived in Ref. schoeller, , we find six different diagrams contributing to $`\mathrm{\Gamma }_{nn^{}}^{(3)}`$, see Fig. 1.
To present the result in a compact way we define a correlator
$`\delta ^3I(\omega _1,\omega _2)={\displaystyle \frac{1}{(2\pi )^2}}{\displaystyle _{\mathrm{}}^{\mathrm{}}}d(t_3t_1){\displaystyle _{\mathrm{}}^{\mathrm{}}}d(t_2t_1)\times `$
$`\times e^{i\omega _1(t_2t_1)+i\omega _2(t_3t_1)}\stackrel{~}{T}[\delta I(t_1)\delta I(t_2)]\delta I(t_3),`$ (2)
where $`\stackrel{~}{T}`$ denotes the anti-time-ordering operator. The time-dependent correlator is calculated with respect to the free external Hamiltonian $`H_{\mathrm{ext}}`$ with the density operator $`\rho _{\mathrm{ext}}(t_0)`$. We assume $`H_{\mathrm{ext}}`$ to be independent of time and $`\rho _{\mathrm{ext}}(t_0)`$ to describe a stationary state with respect to $`H_{\mathrm{ext}}`$. Our results can be also stated with the help of the Fourier transform of $`\delta I(t_3)T[\delta I(t_2)\delta I(t_1)]`$, which is the complex conjugate of the previous correlator, so it is a matter of choice which one to use. Our definition of the frequency-dependent third cumulant (Quantum transitions induced by the third cumulant of current fluctuations) differs from the one studied in Ref. galaktionov, , which consists of the sum of all possible Keldysh orderings. Whereas that definition is relevant in studying the evolution of the off-diagonal density matrix dynamics, transition rates cannot be obtained from that form. If the state of the environment is invariant under time reversal as usually in equilibrium at low magnetic fields, the correlator (Quantum transitions induced by the third cumulant of current fluctuations) vanishes.
Evaluating and summing the different contributions shown in Fig. 1, we obtain the result
$`\mathrm{\Gamma }_{nn^{}}^{(3)}={\displaystyle \frac{4\pi g^3}{\mathrm{}^3}}\mathrm{Re}{\displaystyle \underset{n_1}{}}[`$ $`{\displaystyle _{\mathrm{}}^{\mathrm{}}}{\displaystyle \frac{\delta ^3I(\frac{E}{\mathrm{}},\frac{E_n^{}E_n}{\mathrm{}})}{E(E_{n_1}E_n^{})i\eta }}๐E`$
$`\times \varphi _{n^{},n}\varphi _{n_1,n^{}}\varphi _{n,n_1}].`$ (3)
The summation is extended over all the eigenstates of $`H_s`$ and $`\eta `$ denotes a positive infinitesimal quantity. With the help of the identity $`\frac{1}{xx_0\pm i\eta }=P\frac{1}{xx_0}i\pi \delta (xx_0)`$, where $`P`$ stands for a principle value integral, we can write (Quantum transitions induced by the third cumulant of current fluctuations) in the form
$`\mathrm{\Gamma }_{nn^{}}^{(3)}={\displaystyle \frac{4\pi g^3}{\mathrm{}^3}}\times \mathrm{Im}{\displaystyle \underset{n_1}{}}[iP{\displaystyle \frac{\delta ^3I(\frac{E}{\mathrm{}},\frac{E_n^{}E_n}{\mathrm{}})}{E(E_{n_1}E_n^{})}}`$
$`+\pi \delta ^3I({\displaystyle \frac{E_{n_1}E_n^{}}{\mathrm{}}},{\displaystyle \frac{E_n^{}E_n}{\mathrm{}}})]\varphi _{n^{},n}\varphi _{n_1,n^{}}\varphi _{n,n_1}.`$ (4)
Even without the knowledge of $`\delta ^3I(\omega _1,\omega _2)`$, the general results (Quantum transitions induced by the third cumulant of current fluctuations,Quantum transitions induced by the third cumulant of current fluctuations) contain some information about the requirements made for the meter designed to detect the third-cumulant effects. The structure of the product of the matrix elements $`\varphi _{n_in_j}`$ restricts the possible physical realizations used in detecting the transitions induced by the third cumulant. Generally the operator $`\varphi `$ should either couple several states of the system, or both matrix elements $`\varphi _{n,n}`$ and $`\varphi _{n,n+1}`$ should be finite.
Next we turn to study the case where the probe system is a qubit. The system Hamiltonian can be written as $`H_s=\frac{1}{2}B_z\sigma _z\frac{1}{2}B_x\sigma _x`$ and the interaction term as $`H_{\mathrm{int}}=g\delta I\sigma _z`$. The system Hamiltonian has the eigenstates $`|E_1=\alpha |+\beta |`$, $`|E_0=\beta |+\alpha |`$ and the eigenenergies $`E_1=\frac{1}{2}\sqrt{B_x^2+B_z^2}`$, $`E_0=\frac{1}{2}\sqrt{B_x^2+B_z^2}`$. The coefficients can be parametrized as $`\alpha =\mathrm{cos}\frac{\varphi }{2}`$ and $`\beta =\mathrm{sin}\frac{\varphi }{2}`$, where $`\varphi =\mathrm{arctan}(\frac{B_x}{B_z})`$. We denote the energy difference between the two eigenstates as $`\mathrm{\Delta }E=\sqrt{B_x^2+B_z^2}`$. Using the above conventions and the general result (Quantum transitions induced by the third cumulant of current fluctuations), we can express the corrections to the transition rates as
$`\mathrm{\Gamma }_{E_1E_0}^{(3)}={\displaystyle \frac{16\pi g^3}{\mathrm{}^3}}F({\displaystyle \frac{\mathrm{\Delta }E}{\mathrm{}}})(\alpha \beta )^2(\alpha ^2\beta ^2)`$
$`\mathrm{\Gamma }_{E_0E_1}^{(3)}={\displaystyle \frac{16\pi g^3}{\mathrm{}^3}}F({\displaystyle \frac{\mathrm{\Delta }E}{\mathrm{}}})(\alpha \beta )^2(\alpha ^2\beta ^2).`$ (5)
The function $`F(\omega )`$ contains the information about the third cumulant and is defined as
$`F(\omega )=\mathrm{Im}[iP{\displaystyle \frac{\delta ^3I(\frac{E}{\mathrm{}},\omega )}{E\mathrm{}\omega }}+iP{\displaystyle \frac{\delta ^3I(\frac{E}{\mathrm{}},\omega )}{E}}+`$
$`+\pi \delta ^3I(\omega ,\omega )\pi \delta ^3I(0,\omega )].`$ (6)
Comparing the result (Quantum transitions induced by the third cumulant of current fluctuations) with the Golden Rule rates
$`\mathrm{\Gamma }_{E_1E_0}^{(2)}={\displaystyle \frac{8\pi g^2}{\mathrm{}^2}}S_{\delta I}({\displaystyle \frac{\mathrm{\Delta }E}{\mathrm{}}})(\alpha \beta )^2`$
$`\mathrm{\Gamma }_{E_0E_1}^{(2)}={\displaystyle \frac{8\pi g^2}{\mathrm{}^2}}S_{\delta I}({\displaystyle \frac{\mathrm{\Delta }E}{\mathrm{}}})(\alpha \beta )^2,`$ (7)
one notices that the function $`F(\omega )`$ plays a similar role in $`\mathrm{\Gamma }^{(3)}`$ as the noise power in $`\mathrm{\Gamma }^{(2)}`$.
Supposing we can control the effective magnetic fields $`B_x`$ and $`B_z`$, we can optimize the parameters $`\alpha `$ and $`\beta `$ to produce the maximum effect from $`\mathrm{\Gamma }^{(3)}`$. The absolute value of the expression $`(\alpha \beta )^2(\alpha ^2\beta ^2)`$ is maximized by choosing $`\alpha =0.89`$ and $`\beta =0.46`$ or vice versa, i.e., $`B_x=1.4B_z`$ or $`B_z=1.4B_x`$. By changing the magnitude of $`\mathrm{\Delta }E=\sqrt{B_x^2+B_z^2}`$ but keeping $`B_x/B_z`$ fixed, one can probe $`F(\omega )`$ as a function of frequency.
A physical qubit always has some intrinsic noise mechanism, in solid-state realizations produced by the electromagnetic environment, which cannot be neglected (we consider the external fluctuation circuit as an additional environment). To be measurable, the external current fluctuation effects have to be significant compared to transitions due to the intrinsic noise.
A possible physical realization for the system considered above is a Josephson flux qubit mooij99 ; makhlin coupled inductively to the external circuit, see Fig. 2. The interaction Hamiltonian is of the form $`H_{\mathrm{int}}=\frac{M\mathrm{\Delta }\varphi }{2L_{qb}}\delta I\sigma _z`$, where $`M`$ is the mutual inductance between the qubit and the external circuit, $`\mathrm{\Delta }\varphi `$ is the flux difference between the two states of the flux qubit and $`L_{qb}`$ is the inductance of the qubit. We choose $`\delta I=0`$, since the effects of the finite average external current can be included in redefining $`B_z`$ or eliminated by a flux control. The effective magnetic fields can be controlled by external fluxes through the loops so the qubit can be biased to the optimal point for detecting $`\mathrm{\Gamma }^{(3)}`$. The transition rates follow from Eqs. (Quantum transitions induced by the third cumulant of current fluctuations) and (Quantum transitions induced by the third cumulant of current fluctuations) after the identification $`g=\frac{M\mathrm{\Delta }\varphi }{2L_{qb}}`$. We assume that the energy gap to the higher states is large compared to any other energy scales in the system, allowing us to make the two-state approximation and to neglect the effective interaction terms nonlinear in $`\delta I`$.
Let us estimate $`\mathrm{\Gamma }^{(3)}`$ in a flux qubit for a specific setup. Suppose that the external circuit consists of a scatterer with resistance $`R`$ and loop inductance $`L`$. We assume that the third cumulant of current fluctuations in the scatterer is frequency independent in the frequency scale of the circuit, $`\omega _LR/L`$. This is generally the case provided that the voltage $`eV`$ over and the Thouless energy $`E_T`$ of the scatterer, defined as the inverse time of flight through it, satisfy $`eV,E_T\mathrm{}\omega _L`$.galaktionov ; pilgram Then the frequency dependence of the correlator (Quantum transitions induced by the third cumulant of current fluctuations) arises solely from the classical effect of the inductance $`L`$ modifying the noise. In this limit Eq. (Quantum transitions induced by the third cumulant of current fluctuations) can be approximated by
$$\delta ^3I(\omega _1,\omega _2)=\frac{F_3e^2I(2\pi )^1}{(1+\frac{i\omega _1L}{R})(1+\frac{i\omega _2L}{R})(1\frac{i(\omega _1+\omega _2)L}{R})},$$
(8)
where $`I`$ is the average current in the circuit and $`F_3`$ is a scatterer-specific proportionality constant (โFano factorโ) between the third cumulant and the current.
In deriving (8), we assumed $`L_{qb}I_{qb}^2LI^2`$ where $`I_{qb}`$ is the current in the qubit, allowing us to neglect the back-action of the qubit on these fluctuations. This leads to the rates $`\mathrm{\Gamma }_{E_0E_1}^{(3)}=\mathrm{\Gamma }_{E_1E_0}^{(3)}\mathrm{\Gamma }^{(3)}`$ given by
$$\mathrm{\Gamma }^{(3)}=A\frac{\mathrm{\Delta }E\omega _L^3}{(\mathrm{\Delta }E^2+\mathrm{}^2\omega _L^2)(\mathrm{\Delta }E^2+4\mathrm{}^2\omega _L^2)},$$
(9)
where $`A32\pi F_3e^2Ig^3(\alpha \beta )^2(\beta ^2\alpha ^2)`$. The noise power for the setup can be written as
$$S_{\delta I}(\omega )=\frac{F_2(eI\frac{\mathrm{}|\omega |}{R})\theta (eV\mathrm{}|\omega |)+\frac{\mathrm{}\omega \theta (\omega )}{R}}{1+\omega ^2/\omega _L^2}$$
(10)
where $`F_2`$ is the Fano factor for the second cumulant and $`\theta (x)`$ is the Heaviside step function. This formula includes the quantum fluctuations (last term) and is valid for our case provided that the temperature $`T`$ is low, $`k_BT\mathrm{\Delta }E`$. In the limit $`eV\mathrm{\Delta }E`$ we get from Eqs. (Quantum transitions induced by the third cumulant of current fluctuations), (9) and (10) that $`\mathrm{\Gamma }_{E_1E_0}^{(2)}=\mathrm{\Gamma }_{E_0E_1}^{(2)}\mathrm{\Gamma }^{(2)}`$ and
$`\gamma _3{\displaystyle \frac{\mathrm{\Gamma }^{(3)}}{\mathrm{\Gamma }^{(2)}}}=2(\beta ^2\alpha ^2)\stackrel{~}{g}{\displaystyle \frac{F_3}{F_2}}{\displaystyle \frac{\mathrm{\Delta }E\mathrm{}\omega _L}{\mathrm{\Delta }E^2+4\mathrm{}^2\omega _L^2}},`$ (11)
with $`\stackrel{~}{g}=\left(\frac{M\mathrm{\Delta }\varphi e}{\mathrm{}L_{qb}}\right)`$. For the optimal parameters $`\alpha `$ and $`\beta `$ mentioned above, $`\alpha ^2\beta ^2=0.58`$. The phase difference of the two flux states can be of order $`\mathrm{\Phi }_0/4=h/8e`$, so we may estimate $`\stackrel{~}{g}\left(\frac{2\pi M}{8L_{qb}}\right)`$. Consequently, it can be made of order unity or greater by an efficient inductive coupling and a large external inductance $`L`$. The factor $`F_3/F_2`$ depends solely on the nature of noise produced by the scatterer.f3note For realistic parameters $`\omega _L/(2\pi )=10\mathrm{GHz}`$ and $`\mathrm{\Delta }E/h=1\mathrm{GHz}`$ the last factor is about 2.5%. Optimizing the setup one could expect a relative effect $`\left|\frac{\mathrm{\Gamma }^{(3)}}{\mathrm{\Gamma }^{(2)}}\right|`$ up to roughly 10 %, which shows that the third cumulant effect can be significant.
Now suppose that the intrinsic relaxation of the qubit is caused by an independent zero-averaged fluctuating Gaussian field. Then the second-order rate should be replaced by the sum of rates caused by the field and the external circuit, the third-order rate remaining unchanged. In the case of a zero-temperature environment, this intrinsic relaxation rate $`\mathrm{\Gamma }_{\mathrm{int}}`$ can be quantified by the $`Q`$-factor, $`\mathrm{\Gamma }_{\mathrm{int}}=\mathrm{\Delta }E/\mathrm{}Q`$. In this case, its ratio to the rate $`\mathrm{\Gamma }^{(2)}`$ is
$$\gamma _Q^{\mathrm{int}}\frac{\mathrm{\Gamma }_{\mathrm{int}}}{\mathrm{\Gamma }_{E_0E_1}^{(2)}}=\frac{R}{4R_Q}\frac{1}{Q\stackrel{~}{g}^2}\frac{1+\mathrm{\Delta }E^2/\mathrm{}\omega _L^2}{F_2(\frac{eV}{\mathrm{\Delta }E}1)\theta (eV\mathrm{\Delta }E)(\alpha \beta )^2}.$$
(12)
Here $`R_Q=h/e^2`$.
One possibility to detect $`\mathrm{\Gamma }^{(3)}`$ is to let the qubit reach the stationary state and then determine the probabilities $`P_{E_0}`$ and $`P_{E_1}=1P_{E_0}`$ of the states $`|E_0`$ and $`|E_1`$. This can be achieved by repeated measurements of the qubit. From detailed balance we get
$$p\frac{P_{E_1}}{P_{E_0}}=\frac{\mathrm{\Gamma }_{E_0E_1}^{(2)}+\mathrm{\Gamma }_{E_0E_1}^{(3)}}{\mathrm{\Gamma }_{E_1E_0}^{(2)}+\mathrm{\Gamma }_{E_1E_0}^{(3)}}=\frac{1+\gamma _3}{1+\gamma _Q+\gamma _3},$$
(13)
where $`\gamma _Q=\gamma _Q^{\mathrm{int}}+\gamma _Q^\mathrm{c}`$ and $`\gamma _Q^\mathrm{c}=(\mathrm{\Gamma }_{E_1E_0}^{(2)}\mathrm{\Gamma }_{E_0E_1}^{(2)})/\mathrm{\Gamma }_{E_0E_1}^{(2)}`$. Now inverting the external current $`I`$, $`\mathrm{\Gamma }^{(3)}`$ changes sign and we get $`p^{}P_{E_1}^{}/P_{E_0}^{}=(1\gamma _3)/(1+\gamma _Q\gamma _3)`$. From the above relations one can solve $`\mathrm{\Gamma }_{E_1E_0}^{(3)}`$ and $`\mathrm{\Gamma }_{E_0E_1}^{(3)}`$ provided that the probabilities, $`\mathrm{\Gamma }_{\mathrm{int}}`$ and $`\mathrm{\Gamma }_i^{(2)}`$ are known. One can evaluate $`\mathrm{\Gamma }_i^{(2)}`$ by applying Eq. (Quantum transitions induced by the third cumulant of current fluctuations) or the rates can be determined experimentally. From Eqs.โ (Quantum transitions induced by the third cumulant of current fluctuations) and (Quantum transitions induced by the third cumulant of current fluctuations) we see that when $`\alpha =\beta `$, $`\mathrm{\Gamma }^{(3)}`$ vanishes but $`\mathrm{\Gamma }_i^{(2)}`$ remains finite. By keeping $`\mathrm{\Delta }E`$ fixed but setting $`B_z=0`$ it is possible to measure $`\mathrm{\Gamma }_i^{(2)}`$ independently. Figure 3 shows the asymmetry $`pp^{}`$ in the change of polarization with respect to the current in the source as a function of the magnitude of the current (bias voltage $`V=RI`$).
In conclusion, we have studied the transitions induced by the third cumulant of current fluctuations on a probe quantum system. We have calculated a general formula for the transition rates and propose a scheme to measure the predicted results using a Josephson flux qubit. We have shown that the third-order transition rates are governed by the variant of the third cumulant which to our knowledge has not been studied before.
We thank Valentina Brosco, Frank Hekking and Jukka Pekola for fruitful discussions. TTH acknowledges the funding from the Academy of Finland.
|
warning/0506/math0506163.html
|
ar5iv
|
text
|
# Monotone invariants and embeddings of statistical manifolds
## 1. Introduction
A statistical model is a family $`M`$ of probability measures on a measurable space $`\mathrm{\Omega }`$. There are two natural geometrical structures on any statistical model equipped with a differentiable manifold structure. They are the Fisher tensor and the Chentsov-Amari tensor.
The Fisher tensor was given by Fisher in 1925 as an information characterization of a statistical model. Rao \[Rao(1945)\] proposed to consider this tensor as a Riemannian metric on the manifold of probability distributions. This Fisher metric has been systematically studied in \[Chentsov1972\], \[M-C 1990\], \[A-N2000\] and others \[Lauritzen1987\], \[Rao1987\], \[Ay2002\], \[Jost2005\], ect. in the field of geometric aspects of statistics and information theory.
Chentsov \[Chentsov1972\] and Amari \[Amari1997\] independently also discovered a natural structure on statistical models, namely a 1-parameter family of invariant connections, which includes the Levi-Civita connection of the Fisher metric. This family of invariant connections is defined by a 3-symmetric tensor $`T`$ together with the Levi-Civita connection of the Fisher metric.
Motivated by the question how much we can describe a statistical model via their Fisher metric and Chentsov-Amari tensor $`T`$, in 1987 Lauritzen proposed to call a Riemannian manifold $`(M,g)`$ with a 3-symmetric tensor $`T`$ a statistical manifold. Since two 3-symmetric tensors $`T`$ and $`kT`$, $`k0`$, define the same family of Chentsov-Amari connections, we shall say that two statistical manifolds $`(M,g,T)`$ and $`(M,g,kT)`$ are conformal equivalent.
A natural and important question in the mathematical statistics is to understand, if a given family $`M`$ of probability distributions can be considered as a subfamily of another given one $`N`$. In the language of statistical manifolds, this question can be formulated as a problem of isostatistical embedding of a statistical manifold $`(M,g,T)`$ into another one $`(N,g^{},T^{})`$. Here we say that an immersion $`f:(M,g,T)(\overline{M},\overline{g},\overline{T})`$ is called isostatistical, if $`f^{}(\overline{g})=g`$ and $`f^{}(\overline{T})=T`$.
We shall see in section 2 that the problem of the existence of an isostatistical embedding includes also the Lauritzen question in 1987, if any statistical manifold is a statistical model. It also concerns the following important problem posed by Amari in 1997, if any finite dimensional statistical model can be embedded into the space $`Cap^N`$ of probability distributions of the sample space $`\mathrm{\Omega }^N`$ of $`N`$ elementary events for some finite $`N`$.
We shall construct a class of $`C^0`$ (and $`C^1`$) monotone invariants of statistical manifolds, which present obstructions to embedding of a given $`C^k`$ statistical manifold $`M`$ into another one $`N^n`$. Here a $`C^k`$ statistical manifold $`(M,g,T)`$ is a smooth differentiable manifold with $`C^k`$ sections $`gS^2T^{}M`$ and $`TS^3T^{}M`$. These invariants measure certain relations between the metric tensor $`g`$ and the 3-symmetric tensor $`T`$. In particular, using these invariants we show that no statistical manifold which is conformal equivalent to the space $`Cap^N`$ can be embedded into the product of $`m`$ copies of the normal Gaussian manifolds for any $`N>3`$ and any finite $`m`$. In the Main Theorem (section 5) we prove that any compact smooth ($`C^1`$ resp.) statistical manifold $`M^m`$ can be isostatistically embedded to a the space $`Cap^N`$ for some $`N`$ big enough.
As a consequence we also get a new proof of Matumoto theorem on the existence of the contrast function for a compact statistical manifold (see 2.8).
Acknowledgement. I am thankful to Jรผrgen Jost and Nihat Ay for their introduction to the field of information geometry and helpful discussions.
## 2. Statistical models and statistical manifolds.
In this section we recall the definitions of the Fisher metric and the Chentsov-Amari connections on statistical models. We introduce the notion of a weak Fisher metric and a weak potential function. At the end of the section we discuss the problem, if a given statistical manifold is a statistical model. Most of the facts in this section can be found in \[A-N2000\].
Suppose that $`M`$ is a statistical model - a family of probability measures on a space $`\mathrm{\Omega }`$. We assume throughout this note that $`M`$ and $`\mathrm{\Omega }`$ are differentiable manifolds, and $`\mathrm{\Omega }`$ is equipped with a fixed Borel measure $`d\omega `$. We also write
$`(2.1)`$
$$p(x,\omega )=p(x,\omega )d\omega ,$$
where $`p(x,\omega )`$ in LHS of (2.1) is a Borel measure in $`M`$ and $`p(x,\omega )`$ in the RHS of (2.1) is a non negative (density) function on $`M\times \mathrm{\Omega }`$ which satisfies
$`(2.1.a)`$
$$_\mathrm{\Omega }p(x,\omega )๐\omega =1xM.$$
The Fisher metric $`g^F(x)`$ is defined on $`M`$ as follows. For any $`V,WT_xM`$ we put
$`(2.2)`$
$$g^F(V,W)_x=_\mathrm{\Omega }(_V\mathrm{ln}p(x,\omega ))(_W\mathrm{ln}p(x,\omega ))p(x,\omega ).$$
The function under integral in (2.2) is well defined, if
$`(2.1.b)`$
$$p(x,\omega )>0,$$
Denote by $`Cap(\mathrm{\Omega })`$ the space of all probability measures on $`\mathrm{\Omega }`$. Clearly we can consider the density function $`p(x,\omega )`$ as a mapping $`MCap(\mathrm{\Omega })`$. Thus we shall call a function $`p(x,\omega )`$ a probability potential of the metric $`g_F`$, if $`p(x,\omega )`$ satisfies (2.1.a), (2.1.b), (2.2). (It is known that for a given Riemannian metric $`g_F`$ on a smooth manifold $`M`$ there exist many probability potentials $`f(x,\omega )`$ for $`g_F`$, even if we fix the space $`(\mathrm{\Omega },d\omega )`$.)
Some time it is useful to consider functions $`p(x,\omega )`$ which satisfy (2.2) and (2.1.b) but not necessary (2.1.a). In this case, the Riemannian metric $`g^F`$ will be called weak Fisher metric, and the function $`p(x,\omega )`$ will be called a weak probability potential of $`g^F`$.
2.3. Example of a weak Fisher metric: the standard Euclidean metric $`g^0`$ on the positive quadrant $`_+^N(x_i>0)`$. It is straightforward to check that $`g_0`$ admits a weak probability potential $`\{p_i(x)=\frac{1}{4}x_i^2,i=\overline{1,N}.\}`$ Here $`\mathrm{\Omega }=\mathrm{\Omega }^N`$ \- the sample space of $`N`$ elementary events.
2.4. The Fisher metric on the space $`(Cap^N)_+`$ of all positive probability distributions on $`\mathrm{\Omega }^N`$ (see also \[A-N2000\], \[Jost2005\], \[Chentsov1972\]). By definition we have
$$Cap_+^N:=\{(p_1,\mathrm{},p_N)|p_i>0\text{ for }i=\overline{1,N}\&p_i=1\}.$$
We define the embedding map
$$f:Cap_+^NS^{N1}(2),$$
$$(p_1,\mathrm{},p_N)(q_1=2\sqrt{p_1},\mathrm{},q_N=2\sqrt{p_N}).$$
It is easy to see that the Fisher metric in the new coordinates $`(q_i)`$ is the standard metric of constant positive curvature on the sphere $`S^{N1}(2)`$.
2.5. Divergence potential (see \[A-N2000\], \[Rao(1987)\].) A function $`\rho `$ on $`M\times M`$ with the following property
$`(\mathrm{2.5.1})`$
$$\rho (x,y)g^Fe0\text{ with equality iff }x=y$$
is called a divergence function. A divergence function $`\rho `$ is called a divergence potential for a metric $`g`$ on $`M`$, if
$`(\mathrm{2.5.2})`$
$$g(X,Y)_x=Hess(\rho )(i_1(X),i_1(Y)).$$
where
$$T_{(x,x)}(M,M)=(T_xM,0)(0,T_xM)=(i_1(T_xM))(i_2(T_xM)).$$
An example of a divergence potential for a Fisher metric is the Jensen function $`J_H^{\lambda ,\mu }(x,y)`$ of the entropy function $`H(x)`$ on $`M`$, or a Kullback relative entropy function $`K(x,y)`$ on $`M\times M`$.
2.6. Chentsov-Amari connections. Let $`p(x,\omega )`$ be a probability potential for a Riemannian metric $`g`$. We define a symmetric 3-tensor $`T`$ on $`M`$ as follows
$`(\mathrm{2.6.1})`$
$$T(X,Y,Z)=(_X\mathrm{ln}p(x,\omega ))(_Y\mathrm{ln}p(x,\omega ))(_Z\mathrm{ln}p(x,\omega ))p(x,\omega ).$$
We denote by $`^F`$ the Levi-Civita connection of the (weak) Fisher metric $`g^F`$. We define
$`(\mathrm{2.6.2})`$
$$<_X^tY,Z>:=<_X^FY,Z>+tT(X,Y,Z).$$
The connections $`^t`$ are called the Chentsov-Amari connections.
2.6.3. Remark. (\[A-N2000\], \[Matsumoto1993\]) Any divergence function $`\rho (x,y)`$ on $`M\times M`$ defines a tensor $`T`$ on $`M`$ via the following formula
$$T(X,Y,Z)_x=_{i_2(Z)}Hess(\rho )(i_1(X),i_1(Y))_{(x,x)}+_{i_1(Z)}Hess(\rho )(i_2(X),i_2(Y))_{(x,x)}.$$
If $`g`$ and $`T`$ are defined by the same divergence function $`\rho (x,y)`$, we shall call $`\rho (x,y)`$ a divergence potential for the statistical manifold $`(M,g,T)`$. It is a known fact that the Kullback relative entropy function is a divergence potential for the associated statistical model.
2.7. Statistical submanifolds.
A submanifold $`N`$ in a statistical manifold $`(M,g,T)`$ with the induced Riemannian metric $`g_{|N}`$ and induced tensor $`T_{|N}`$ is called statistical submanifold of $`(M,g,T)`$. Clearly, if $`f(x,\omega )`$ is a (weak) probability potential for $`(M,g,T)`$, then its restriction to any submanifold $`NM`$ is a (weak) probability potential of the induced statistical structure.
2.8. Statistical models and statistical manifolds. Since any probability function $`p(x,\omega )`$ defines a map $`MCap(\mathrm{\Omega })`$, we shall say that a statistical manifold $`(M,g,T)`$ is a statistical model, if there probability potential $`p(x,\omega )`$ for $`g`$ and $`T`$. By the remark in 2.7, we get that a statistical submanifold of a statistical model is also a statistical model. Furthermore, if a statistical manifold $`(M,g,T)`$ is a statistical model, then it must admit a divergence potential. Hence we obtain the following
2.8.1 Theorem. (cf. \[Matumoto1993\] ) For any compact statistical manifold $`(M,g,T)`$ there exists a divergence potential $`\rho `$ for $`g`$ and for $`T`$.
Note that Matumotoโs theorem does not requires the compactness of $`(M,g,T)`$.
## 3. Embeddings of linear statistical spaces.
An Euclidean space $`(^n,g^0)`$ equipped with a 3 -symmetric tensor $`T`$ will be called a linear statistical spaces. We observe that the equivalence class of linear statistical spaces coincides with the orbit space of 3-symmetric tensors $`T`$ under the action of the orthogonal group $`O(n)`$. In this section we discuss certain invariants of these orbits and we show several necessary and sufficient conditions for the existence of embedding of one linear statistical space into another linear statistical space by studying these invariants. A class of our necessary conditions consists of monotone invariants $`\lambda `$, i.e. we assign to any linear statistical space $`(^n,g^0,T)`$ a number $`\lambda (^n,g^0,T)`$ such that, if $`(^n,g^0,T)`$ is a statistical submanifold of $`(^m,g^0,T^{})`$, then we have
$$\lambda (^n,g^0,T)\lambda (^m,g^0,T^{}).$$
Since a tangent space of a statistical manifold is a linear statistical manifold, these invariants play important role in the problem of isostatistical immersion.
3.1. Trace type of a symmetric 3-tensor. Let us denote by $`^n`$ the subspace in $`S^3(^n)`$ consisting of the following 3-symmetric tensors
$$T^v(x,y,z)=<v,x><y,z>+<v,y><x,z>+<v,z><x,y>,$$
where $`v^n`$. Using the standard representation theory (see e.g. \[O-N1988\]) we have the decomposition
$`(3.2)`$
$$S^3(^n)=(3\pi _1)^n.$$
The component $`^n`$ is defined by taking the trace of $`T`$
$$Tr:S^3(^n)^{}(^n)^{},Tr(T)(v):=Tr(vT).$$
Clearly $`Tr`$ is an $`SO(n)`$-equivariant map with nonzero image. Using the identity $`Tr(T^v)=(n+2)v^{}`$, we get
3.3. Lemma. We have
$`(3.4)`$
$$\pi _2(T)=\frac{1}{n+2}T^{Tr(T)}.$$
In view of Lemma 3.3 we shall call any tensor $`T^n`$ of trace type.
We note that
$$dimS^3(^n)=C_n^3+2C_n^2+n=\frac{n(n+1)(n+2)}{6}.$$
Thus the dimension of the quotient $`S^3(^n)/SO(n)`$ is at least $`C_n^3+C_n^2+n`$. A direct computation shows that the dimension of the orbit $`SO(n)([_{i=1}^na_iv_i^3])`$ is $`C_n^2=dimSO(n)`$, if $`\mathrm{\Pi }a_i0`$. Here $`\{v_i\}`$ is an orthonormal basis in $`^n`$. Hence the dimension of $`S^3(^n)/O(n)=C_n^3+C_n^2+n`$. This dimension is exactly the number of all complete invariants of pairs consisting of a positive definite bilinear form $`g`$ and a 3-symmetric tensor $`T`$.
Since the dimension of $`G_k(^n)=k(nk)`$, it follows that generically it is impossible to embed a linear statistical space $`(R^k,g^0,T)`$ into a given statistical linear space $`(R^n,g^0,T)`$, unless $`k(nk)g^FeC_k^3+C_k^2+k`$. Clearly the dimension condition is not sufficient as the following proposition shows.
3.5. Proposition. A linear statistical space $`(^k,g^0,T)`$ can be embedded into a linear statistical space $`(^N,g^0,T^v)`$, if and only if $`Ng^Fek`$ and $`T`$ is also a trace type: $`T=T^w`$ with $`|w||v|`$.
Proof. The necessary condition follows from the fact that the restriction of $`T^v`$ to $`^k`$ equals $`T^{\overline{v}}`$, where $`\overline{v}`$ is the orthogonal projection of $`v`$ to $`^k`$. Conversely, if $`|w||v|`$ we can find an orthogonal transformation, such that $`w`$ equals the orthogonal projection of $`v`$ on $`^k`$. $`\mathrm{}`$
3.6. Commasses as monotone invariants. Since the metric $`g`$ extends canonically on the space $`S^3(^n)`$, we can define the absolute norm
$$T:=\sqrt{<T,T>}.$$
Now we define comasses of a 3-symmetric tensor $`T`$ as follows
$$^3(T):=\underset{|x|=1,|y|=1,|z|=1}{\mathrm{max}}T(x,y,z),$$
$$^2(T):=\underset{|x|=1,|y|=1}{\mathrm{max}}T(x,y,y),$$
$$^1(T):=\underset{|x|=1}{\mathrm{max}}T(x,x,x).$$
Clearly we have
$$0^1(T)^2(T)^3(T)T.$$
3.7. Proposition. The comasses $`^i`$, $`i[1,3]`$, are nonnegative linear monotone invariants, which vanish if and only if $`T=0`$.
Proof. Clearly $`^i(T)0`$ for $`i=1,2,3`$. Now we are going to show that $`^1`$ vanishes at $`T`$ only if $`T=0`$. Observe that $`^1=0`$ if and only if $`T(x,x,x)=0`$ for all $`x^n`$. Writing $`T`$ in coordinate expression $`T(x,y,z)=a_{ijk}x^iy^jz^k`$, we note that $`T(x,x,x)=0`$ if and only if $`T=0`$, since $`T`$ is symmetric.
Next we shall show that $`^i(T)`$ is a linear monotone invariant for $`i=1,2,3`$. Assume that $`e`$ is a linear embedding $`(^n,g,T)`$ into $`(^m,\overline{g},\overline{T})`$. Then $`T`$ is a restriction of the 3-symmetric tensor $`\overline{T}`$. Hence we have
$$^i(T)^i(\overline{T}),\text{ for }i=1,2,3.$$
This implies that $`^i`$ are linear monotone invariants. $`\mathrm{}`$
Now for a space $`(^n,g^0,T)`$ and for $`1kn`$ we put
$$\lambda _k(T):=\underset{^k^n}{\mathrm{min}}^1(T_{|^k}).$$
We can easily check that if $`\overline{T}`$ is a restriction of $`T`$ to a subspace $`^m^n`$, then
$$\lambda _k(\overline{T})\lambda _k(T)0\text{ for all }km.$$
Thus $`\lambda _k(T)`$ is a monotone invariant of linear statistical manifolds. These invariants are related by the following inequalities
$$^1(T)=\lambda _n(T)\lambda _{n1}(T)\mathrm{}\lambda _2(T)\lambda _1(T)=0.$$
The last equality follows from the fact, that the function $`T(x,x,x)`$ is anti-symmetric on $`S^{n1}(|x|=1)^n`$ and $`S^{n1}`$ is connected. We observe that if $`T`$ is of trace type, then $`\lambda _{n1}(T)=\mathrm{}=\lambda _1(T)=0`$.
We are going to give a lower bound of the monotone invariant $`\lambda _{n1}`$ of a linear statistical space of certain type. The equality $`\lambda _{n1}(^n,g^0,T)g^FeA`$ means that no hyperplane with the norm $`^1`$ strictly less than $`A`$ can be embedded in $`(^{n1},g^0,T)`$.
3.8. Lemma. a) Let $`T=_{i=1}^n(N\epsilon _i)(x^i)^3`$ be a 3-symmetric tensor on $`^n`$ with $`n4`$, $`N4`$ and $`|\epsilon _i|1/4`$. Then we have
$$\lambda _{n1}(T)\frac{N}{\sqrt{1}0}1/4.$$
b ) Let $`T=N_{i=1}^n(x^i)^3`$, and $`H`$ be a hyperplane in $`^n`$ which is orthogonal to $`(kn,1,1,\mathrm{},1)`$, and let $`n5,k3`$. Then we have
$$\lambda _{n2}(T_{|H})\frac{N}{5}1.$$
c) Let $`x=((1\epsilon ),\frac{1}{kn},\mathrm{},\frac{1}{kn})S^n(1)^{n+1}`$, where $`n4,kn`$. We denote by $`H`$ the tangential plane $`T_xS^n`$, and by $`T^0`$ the following 3-symmetric tensor on $`_+^{n+1}`$:
$`(\mathrm{3.8.1})`$
$$T_{ijk}^0(x_1,\mathrm{}x_N)=\delta _{ijk}\frac{2}{x_i}.$$
Then we have
$$\lambda _{n1}(T_{|H}^0)\frac{kn}{5}1.$$
3.8.2. Remark. The tensor $`T^0`$ in (3.8.1) defines on $`(^n,g^0)`$ a statistical structure with a weak probability potential $`\{\frac{1}{4}x_i^2,i=1,n\}`$, see also 2.3.
Proof of Lemma 3.8. The reader shall see that a proof of Lemma 3.8 can be done in the same scheme of the proof of Sublemma 5.10. Therefore we do not repeat this argument here.
3.8.3. Remark. Lemma 3.8.a holds also for $`n=3`$ but not for $`n=2`$, Lemma 3.8.b holds also for $`n=4`$, but not for $`n=3`$, and Lemma 3.8.c holds also for $`n=3`$ but not for $`n=2`$.
There are also several obvious monotone invariants of $`T`$.
$$A^1(T):=\underset{|x|=|y|=|z|=1,<x,y>=<y,z>=<z,x>=0}{\mathrm{max}}T(x,y,z)$$
is well-defined for $`n3`$.
$$A^2(T):=\underset{|x|=|y|=1,<x,y>=0}{\mathrm{max}}T(x,y,y),$$
is well-defined for $`n2`$. We can check that
$$\mathrm{ker}A^1=^n.$$
On the other hand we have
$$\mathrm{ker}A^2(3\pi _1).$$
Thus $`A^1`$ and $`A^2`$ are different invariants.
3.9. Lemma. Let $`\pi _1`$ be the first component of $`T`$ in decomposition (3.2). Then $`T_1:=\pi _1(T)`$ is a monotone invariant of $`T`$.
Proof. Let $`^k`$ be a subspace of $`^n`$. We denote by $`\pi _k^nT`$ the restriction of $`T`$ to $`^k`$. Clearly
$$\pi _k^n(T)=\pi _k^n(\pi _1T)+\pi _k^n(\pi _2T).$$
We have noticed in Proposition 3.5 that the restriction of the trace form $`\pi _2T`$ to any subspace is also a trace form. Thus $`\pi _k^n(\pi _2)`$ is an element in $`^kS^3(^k).`$ Hence we have
$`(\mathrm{3.9.1})`$
$$\pi _1(\pi _k^nT)=\pi _1(\pi _k^n(\pi _1T)).$$
Since all the projections $`\pi _1`$, $`\pi _k^n`$ decrease the norm $`||.||`$, we get
$$\pi _k^nT_1=\pi _1(\pi _k^nT)=\pi _1(\pi _k^n(\pi _1T))\pi _1(T)=T_1.$$
$`\mathrm{}`$
3.10. Proposition. A statistical line $`(,g^0,T)`$ can be embedded into $`(^N,g^0,T^{})`$, if and only if $`^1(T)^1(T^{})`$.
Proof. It suffices to show that we can embed $`(,g^0,T)`$ into $`(^N,g^0,T^{})`$, if we have $`^1(T)^1(T^{})`$. We note that $`T^{}(v,v,v)`$ defines an anti-symmetric function on the sphere $`S^{N1}(|v|=1)^N`$. Thus there is a point $`vS^{N1}`$ such that $`T^{}(v,v,v)=^1(T)`$. Clearly the line $`v`$ defines the required embedding. $`\mathrm{}`$
Let us consider the embedding problem for 2-dimensional linear statistical spaces. It is easy to see that
$$S^3(^2)=^2^2.$$
Thus the quotient $`S^3(^2)/SO(2)`$ equals $`(^2^2)/S^1`$. Geometrically there are several ways to see this. In the first way we denote components of $`TS^3(^2)`$ via $`T_{111},T_{112},T_{122},T_{222}.`$
3.11. Lemma. There exists an oriented orthonormal basic in $`^2`$ such that $`T_{111}=^1(T)>0,T_{112}=0`$ for all non-vanishing $`T`$. These numbers $`(T_{111},T_{122},T_{222})`$ are called canonical coordinates of $`T`$. Two tensors $`T`$ and $`T^{}`$ are equivalent, if and only if they have the same canonical coordinates.
Proof. We choose an oriented orthonormal basis $`(v_1,v_2)`$ by taking as $`v_1`$ a point on $`S^1(|x|=1)`$, where the function $`T(x,x,x)`$ reaches the maximum. The first variation formula shows that in this case $`T_{112}=0`$. This shows the existence of the canonical coordinates. Clearly, if two tensors have the same canonical coordinates, then they are equivalent. Next, if two tensors $`T`$ and $`T^{}`$ are equivalent, then their norms $`^1`$ are the same. We need to take care the case, when there are several points $`x`$ at which $`T(x,x,x)`$ reaches the maximum. In any case, they have the same first coordinates. Next we note that
$$<Tr(T),Tr(T)>=(T_{111}+T_{122})^2+T_{222}^2,$$
$$T^2=T_{111}^2+T_{122}^2+T_{222}^2.$$
Thus if two tensors are equivalent and have the same first coordinates, they must have the same third coordinate $`T_{122}`$, and this third coordinate is uniquely defined up to sign. The condition on the orientation tells us that the sign must be $`+`$. This proves the second statement. $`\mathrm{}`$
3.12. Proposition. We can always embed the 2-dimensional statistical space $`(^2,g^0,0)`$ into any linear statistical space $`(^n,g^0,T)`$, if $`n7`$.
Proof. It suffices to prove for $`n=7`$. We denote by $`๐ช(T)`$ the set of of all unit vectors $`vS^6`$ such that $`T(v,v,v)=0`$. Clearly $`๐ช(T)`$ is a set of dimension 5 in $`S^6`$. Since $`T`$ is anti-symmetric, there exists a connected component $`๐ช^0(T)`$ of $`๐ช(T)`$ which is invariant under the anti-symmetry involution. Now we consider the following function $`f`$ on $`๐ช^0(T)`$. For each $`v๐ช^0(T)`$ we denote by $`A^v`$ the bilinear symmetric 2-form on the space $`T_x๐ช^0(T)`$ considered as a subspace in $`^n`$:
$$A^v(y,z)=T(v,y,z).$$
Then we define $`f(v)`$ equal to $`det(A^v)`$. Since $`๐ช(T)`$ has dimension 5, the function $`f(v)`$ is anti-symmetric on $`๐ช^0(T)`$. Hence the set $`๐ช_0^0(T)`$ of all $`v๐ช^0(T)`$ with $`f(v)=0`$ has dimension 4 and it contains a connected component which is also invariant under the anti-symmetric involution. For the simplicity we denote this connected component also by $`๐ช_0^0(T)`$. Now we consider the following two possible cases.
Case 1. We assume that there is a point $`v๐ช_0^0(T)`$ such that the nullity of $`A^v`$ is at least 2. Then there are two linear independent vectors $`y,zT_v`$ such that the restriction of $`A^v`$ on the plane $`^2(y,z)`$ vanishes. Since the set $`๐ช^0(T)`$ is connected and anti-symmetric and of co dimension 1 in $`S^{n1}`$, the plane $`(y,z)`$ has a non-empty intersection with $`๐ช^0(T)`$ at a point $`w`$. Then the restriction of $`T`$ on the plane $`^2(v,w)`$ is vanished, because
$$T(v,v,v)=T(w,w,w)=0$$
$$T(v,w,w)=0(\text{since }A^v(w,w)=0),$$
$$T(v,v,w)=0(\text{since }wT_v๐ช^0(T)).$$
Case 2. We assume that the nullity of $`A^v`$ on $`๐ช_0^0(T)`$ is constantly 1. Using the anti-symmetric property of $`A^v`$ we conclude that the restriction of $`A^v`$ to the plane $`^4(v)`$ which is orthogonal to the kernel of $`A^v`$ has index constantly 2. Thus there exists a vector $`z`$ which is orthogonal to the kernel $`y`$ of $`A^v`$ such that $`A^v(z,z)=0`$. Clearly the restriction of $`A^v`$ to the plane $`^2(y,z)`$ vanishes. Now we can repeat the argument in the case 1 to get a vector $`w`$ such that the restriction of $`T`$ to $`^2(v,w)`$ vanishes. $`\mathrm{}`$
3.13. Theorem. a) Any statistical space $`(^n,g^0,T)`$ can be embedded in the statistical space $`(^{n(n+1)},g^0,T^{}=2T_{i=1}^{N(n)}x_i^3)`$, where $`x_i`$ are the canonical Euclidean coordinates on $`^{n(n+1)}`$.
b) The trivial space $`(^n,g^0,0)`$ can be embedded into $`(^{2n},g^0,_{i=1}^{2n}(dx^i)^3)`$ for all $`n`$.
Proof. a) We prove by induction. The statement for $`n=1`$ follows from Proposition 3.8. Suppose that the statement is valid for all $`nk`$.
3.14. Lemma. Suppose that $`TS^3(^{k+1})`$. Then there are orthonormal coordinates $`x_1,\mathrm{},x_k`$ such that
$`(\mathrm{3.14.1})`$
$$T=x_1\underset{i=1}{\overset{k+1}{}}a_ix_i^2+\underset{1<i,j,k}{}a_{ijk}x_ix_jx_k.$$
Proof of Lemma 3.14. We choose $`v_1`$ as the unit vector in $`S^k^{k+1}`$, on which the function $`T(v,v,v)`$ reaches the maximum on the unit sphere $`S^k`$. The first variation formula shows that $`T(v_1,v_1,w)=0`$ for all $`w`$ which is orthogonal to $`v_1`$. We denote by $`^k`$ the orthogonal complement to $`v_1`$. Now we consider a bilinear symmetric form $`A`$ on $`^k`$ defined as follows
$$A(x,y)=S(v_1,x,y).$$
There is an orthonormal basis on $`^k`$, where we can write $`A(x,y)=_{i=2}^{k+1}a_ix_i^2`$. Clearly in this orthonormal basis we can write $`T`$ in the form in (3.14.1). $`\mathrm{}`$
Continuation of the proof of Theorem 3.13.a We shall show explicitly that that any statistical space $`(^2,g^0,T=a_2x_1(x_2)^2)`$ can be embedded in $`(^4,g^0,_{i=1}^4(y_i)^3)`$, if $`0|a_2|1/2`$. We put
$`(\mathrm{3.15.1})`$
$$L(v_1):=\pm (\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2})$$
$`(\mathrm{3.15.2})`$
$$L(v_2):=(\sqrt{\frac{1+2a_2}{2}},\sqrt{\frac{1+2a_2}{2}},\sqrt{\frac{12a_2}{2}},\sqrt{\frac{12a_2}{2}}).$$
Here we take the sign $`+`$ in (3.16.1), if $`a_2>0`$, and we take the sign $``$, if $`a_2<0`$. Clearly, $`L`$ defines the required embedding $`^2^4`$.
This together with Proposition 3.8 and the induction assumption completes the proof of Theorem 3.13. a.
Proof of Theorem 3.13. b. We decompose the embedding $`f:(^n,g^0,0)`$ to $`(^{2n},g^0,_{i=1}^{2n}(x^i)^3)`$ as follows
$$f(x_1,\mathrm{},x_n)=(f^1(x_1),\mathrm{},f^n(x_n))$$
where $`f^i`$ embeds the line $`(,(dx^i)^2,0)`$ into $`(^2,(dx^{2i1})^2+(dx^{2i})^2,(dx^{2i1})^3+(dx^{2i})^3)`$. Clearly, $`f`$ is the required embedding. $`\mathrm{}`$
## 4. Monotone invariants and obstructions to embeddings of statistical manifolds
Let $`K(M,e)`$ denote the category of statistical manifolds $`M`$ with morphisms being embeddings. Functors of this category are called monotone invariants of statistical manifolds. Clearly any monotone invariant is an invariant of statistical manifolds.
4.1. Examples. There are many monotone invariants which arise from our analysis in section 3.
a) Trace type of a statistical manifold. A statistical manifold $`(M,g,T)`$ will be called of trace type, if for all $`xM`$ the form $`T(x)`$ is of trace type (see 3.1.) It follows from Proposition 3.5 that any statistical submanifold of a statistical manifold of trace type is also of trace type. Thus the trace type is a monotone invariant. In particular we cannot embed the statistical space $`Cap^N`$ and the normal Gaussian space into any statistical space of trace type. On the other hand, unlike the linear case, we cannot embed a statistical manifold of trace type into another one of trace type, even if the norm condition is satisfied. For example, if the trace form is closed (or exact), then the trace form of its submanifolds is also closed (resp. exact). Hence within a class of statistical manifolds of trace type we get a new monotone invariants which can be expressed via the closedness and the cohomology class of the corresponding trace form.
b) Decomposability of a statistical manifold. We note that the class of 3-symmetric tensors of trace form is a subclass of all decomposable tensors $`T^3`$ which are a symmetric product of 1-forms and symmetric 2-forms. Any statistical submanifold of a statistical manifold with a decomposable tensor $`T`$ has also the (induced) decomposable tensor. Thus the decomposability is also a monotone invariant. The Gaussian normal 2-dimensional manifold is an example of decomposable type but not of trace type.
c) Rank and comass. We define for any statistical manifold $`(M,g,T)`$ the following number
$$rank(T)=suprank(T(x))$$
$$T_0=\underset{xM}{sup}T(x).$$
$$^1(T)_0=\underset{xM}{sup}^1(T(x)).$$
$$T_{1,0}=\underset{xM}{sup}T(x)_1.$$
Clearly these four numbers are monotone invariants of statistical manifolds.
We recall that the normal Gaussian statistical manifold is the two dimensional statistical model which is upper half of the plane $`^2(\mu ,\sigma )`$ with the potential
$$p(\mu ,\sigma )(x)=\frac{1}{\sqrt{2\pi }\sigma }\mathrm{exp}(\frac{(x\mu )^2}{2\sigma ^2}),$$
here $`x`$.
4.2. Proposition. Any statistical manifold which is conformal equivalent to the space $`Cap^N`$ cannot be embedded into the direct product of $`m`$ copies of the normal Gaussian statistical manifold 2.3.3.a for any $`N3`$ and finite $`m`$.
Proof. It is easy to check that $`^1(Cap^N)=\mathrm{}`$. Thus any statistical manifold which is conformal equivalent to $`Cap^N`$ has also the infinite invariant $`^1`$. On the other hand, we compute easily that the norm $`^1`$ of the Gaussian normal manifold, as well as the norm $`^1`$ of a direct product of its finite copies, is finite. Namely the norm $`^1(\mu ,\sigma )`$ is $`\sqrt{2}`$ for all $`(\mu ,\sigma )`$. $`\mathrm{}`$
4.3. Diameters of statistical manifolds. For a positive number $`\rho >0`$ and a statistical manifold $`(M,g,T)`$ we set
$$d_\rho (M,g,T):=sup\{lR^+\mathrm{}|\text{ an immersion of }([0,l],dx^2,\rho (dx)^3)\text{ to }(M,g,T).\}$$
We shall call $`d_\rho (M,g,T)`$ the diameter with weight $`\rho `$ of $`(M,g,T)`$. Clearly $`d_\rho `$ are monotone invariants for all $`\rho `$.
To estimate the diameter with weight $`\rho `$ of a given statistical manifold $`(M,g,T)`$ we can proceed as follows. For each point $`xM`$ we denote by $`D_\rho (x)`$ the set of all unit tangential vector $`vT_xM`$ such that $`T(v,v,v)=\rho `$. We denote by $`D_\rho ^i(x)`$ the connected components of $`D_\rho ^i(x)`$. We say that a unite vector $`v`$ in $`T_xM`$ is $`\rho `$-characteristic with weight $`c(x)`$, if there exists $`i`$ such that we have
$$c(x)=\underset{wD_\rho ^i(x)}{\mathrm{min}}<v,w>>0.$$
We shall say that a point $`xM`$ is $`\rho `$-regular, if there is an open neighborhood $`U_\epsilon (x)M`$ such that $`D_\rho (U_\epsilon )=U_\epsilon \times D_\rho (x)`$. It is easy to see that the set of all $`\rho `$-regular points is open and dense in $`M`$ for any given $`\rho `$.
4.4. Proposition. The diameter $`d_\rho `$ of $`(M^m,g,T)`$ is infinite, if $`m3`$ and there exists a number $`\epsilon >0`$ such that one of the following 2 conditions holds:
a)There exists a $`(\rho +\epsilon )`$-regular point $`xM`$ such that the convex hull $`Cov(D_{\rho +\epsilon }^i(x))`$ of one of connected components $`D_{\rho +\epsilon }^i(x)`$ contains the origin point $`0T_xM^m`$ as it interior point.
b) $`(M^m,g,T)`$ has a complete Riemannian submanifold $`(N,\overline{g})`$ such that there exists a smooth section $`x(D_{\rho +\epsilon }(x)TN)`$ over $`N`$.
Proof. The statement under the first condition a) is based on the fundamental Lemma of the convex integration technique of Gromov. Namely Gromov proved that \[2.4.1.A, Gromov(1986)\], if the convex hull of some path connected subset $`A_0^q`$ contains a small neighborhood of the origin, then there exists a map $`f:S^1^q`$ whose derivative sends $`S^1`$ into $`A_0`$.
4.5. Lemma. Under the condition in Proposition 4.4.1 there exists a small neighborhood $`U_\delta (x)`$ in $`M`$ and an embedded oriented curve $`S^1U_\delta (x)`$ such that for all point $`s(t)S^1`$ we have $`^1(T_{s(t)}S^1)\rho +(\epsilon /2)`$.
Proof of Lemma 4.5. We denote by $`Exp`$ the exponential map $`T_xM^mM^m`$ and by $`DExp`$ the differential of this exponential map restricted to $`S^{m1}\times T_xM^mT(T_xM^m)`$. Here $`S^{m1}`$ is the unit sphere in $`T_xM^m`$. The space $`T_xM^m`$ is a linear statistical space, so we denote by $`_x^1`$ the induced norm-function on $`S^{m1}\times T_xM^m`$ as follows:
$$_x^1(l)=T_x(l,l,l).$$
Since $`DExp`$ is a continuous function, whose restriction to $`S^{m1}\times 0`$ is the identity, there exists a ball $`B(0,\delta )`$ with center in $`0T_xM`$ such that
$`(\mathrm{4.5.1})`$
$$^1(DExp(l))_x^1(l))<\epsilon /4$$
for all $`lS^{m1}\times B(\delta )T(T_xM^m)`$. We can assume that $`\delta `$ is so small such that $`DExp`$ is a homeomorphism on $`S^{m1}\times B(0,\delta )`$.
Now we apply the above mentioned Gromov Lemma \[2.4.1.A, Gr1986\] to get a oriented curve $`S^1(t)`$ in the linear space $`T_xM`$ such that
$`(\mathrm{4.5.2})`$
$$T(\frac{(/t)S^1(t)}{|(/t)S^1(t)|})=\rho +\epsilon $$
for all $`t`$. Next we observe that for all $`\alpha >0`$ the curve $`\alpha S^1(t)`$ has the same norm as $`S^1(t)`$, i.e.
$$_x^1(T_{|(\alpha S^1)}(t))=_x^1(T_{|(S^1)}(t))=\rho +\epsilon .$$
Thus we can assume that our curve $`S^1(t)`$, which satisfies (4.5.2), lies in the ball $`B(0,\delta )`$. By our choice of $`\delta `$ ( see (4.5.1)), we get from (4.5.2)
$`(\mathrm{4.5.3})`$
$$\rho +\frac{3}{4}\epsilon ^1(Exp(S^1(t)))\rho +\frac{5}{4}\epsilon ,$$
for all $`t`$. This curve $`Exp(S^1(t))`$ is an immersed curve. $`\mathrm{}`$
Now let us to continue the proof of Proposition 4.4.a. We denote by $`S^1(t)`$ the embedded curve in Lemma 4.5. Next by choosing a tubular neighborhood of $`S^1(t)`$ we can get a (small, thin) oriented embedded solid torus $`T^3(t,s,r)=S^1(t)\times S^1(s)\times [0,R]`$ in $`M^m`$ such that our embedded curve is exactly the mean curve $`S^1(t)\times \{0\}\times \{0\}`$ on the solid torus. We can choose this torus $`T^3`$ so thin, such that for all $`s,t,r`$ we have
$`(\mathrm{4.5.4})`$
$$^1(T_r^2(t,s))\rho +\frac{\epsilon }{4}.$$
Using (4.5.4) we choose a smooth unit vector field $`V(t,s)`$ on the torus $`T^3(t,s,r)`$ which is tangential to each torus $`T_r^2(t,s)`$ such that $`T(V,V,V)=\rho `$. The integral curve of this vector field is either a circle or an curve of infinite length. If there exists an integral curve of infinite length, then this curve is our desired curve for the Proposition 4.5. Assume now that all the integral curves are circles. Then there exist an embedding $`S^1(t)\times [0,\mu ]\times [0,\mu ]`$ such that for all $`(s,r)[0,\mu ]\times [0,\mu ]`$ the circle $`S^1(t)\times \{s\}\times \{r\}`$ is an integral curve of $`V`$. Now we perturb $`V`$ in a neighborhood $`[0,\alpha ]\times [0,\mu ]\times [0,\alpha ]`$ with a very small $`\alpha `$ such that the perturbed unit vector field $`V^{}`$ satisfies $`T(V^{},V^{},V^{})=\rho `$ and the integral curve of vector field $`V^{}`$ is not any more periodic. This completes the proof of the first part in Proposition 4.4.
Using the same argument we can prove the second part b) of Proposition 4.4. First we get the existence of an embedded curve $`S^1(t)`$ of arbitrary length on $`M`$ such that $`^1(T_{|S^1}(t))\rho +(1/4)\epsilon `$. Now we consider a torus tubular neighborhood of this curve in $`M`$ and apply the same argument in the first part, namely we get on each torus $`T^2(t,s)`$ an integral curve whose unit tangential vector $`V=(/t)S^1(t;s,r)`$ satisfies the condition:
$$T(V,V,V)=\rho .$$
If there exists an infinite integral curve, then we are done. If not, that means all integral curve are circles, then we apply the perturbation method in the proof of the first part and get our desired curve. $`\mathrm{}`$
## 5. Existence of isostatistical embeddings into $`Cap^N`$.
Main Theorem. Any compact smooth ($`C^1`$ resp.) statistical manifold $`(M,g,T)`$ can be immersed into the statistical manifold $`(Cap_+^N,g^F,T^{AC})`$ for some finite number $`N`$. Hence any statistical manifold is a statistical model.
We first deduce our Main Theorem from Theorem 5.1 and Theorem 5.5.
5.1. Theorem. Let $`(M^m,g,T)`$ be a compact smooth ($`C^1`$ resp.) statistical manifold. Then there exist numbers $`N^+`$ and $`A0`$ as well as a smooth ($`C^1`$ resp.) embedding $`f:(M^m,g,T)(^N,g_0,AT_0)`$ such that $`f^{}(g_0)=g`$ and $`f^{}(AT_0)=T`$.
Our proof of Theorem 5.1 uses the Nash embedding theorem, the Gromov embedding theorem and an algebraic trick. The existence of monotone invariants prevents us extend Theorem 5.1 for non-compact case (in contrast to the Riemannian case.)
5.2. The Nash embedding theorem. \[Nash1954, Nash1956\] Any smooth ($`C^1`$ resp.) -Riemannian manifold $`(M^n,g)`$ can be isometrically embedded into $`(^N,g_0)`$ for some $`N`$ depending on $`M^n`$.
We denote by $`T_0`$ the โstandardโ 3-tensor on $`^n`$:
$$T_0=\underset{i=1}{\overset{n}{}}dx_i^3.$$
5.3. The Gromov immersion theorem. \[Gromov1986, 2.4.9.3โ and 3.1.4\] Suppose that $`M^m`$ is given with a smooth ($`C^1`$ resp.) symmetric 3-form $`T`$. Then there exists an embedding $`f:M^m^{N_1(m)}`$ with $`N_1(m)=3(n+(_2^{n+1})+(_3^{n+2}))`$ such that $`f^{}(T_0)=T`$.
Proof of Theorem 5.1. First we shall take an immersion $`f_1:(M^m,g,T)(^{N_1(m)},g_0,T_0)`$ such that
$$f_1^{}(T_0)=T.$$
The existence of $`f_1`$ follows from the Gromov immersion theorem.
Then we choose a positive number $`A^1`$ such that
$$gA^1(f_1^{}(g_0))=g_1$$
is a Riemannian metric on $`M`$, i.e. $`g_1`$ is a positive symmetric bi-linear form. Such a number $`A`$ exists, since $`M`$ is compact.
Now we shall choose an isometric immersion $`f_2:(M^m,g_1)(^N,g_0)`$. The existence of $`f_2`$ follows from the Nash isometric immersion theorem.
5.4. Lemma. There is a linear isometric embedding $`L_{m+1}:^{m+1}^{2m+2}`$ such that $`L_{n+1}(T_0)=0`$.
Proof. We put
$$L_{m+1}(x_1,\mathrm{},x_{m+1})=(f^1(x_1),\mathrm{},f^{m+1}(x_{m+1}))$$
where $`f^i`$ embeds the line $`(,(dx^i)^2,0)`$ into $`(^2,(dx^{2i1})^2+(dx^{2i})^2,(dx^{2i1})^3+(dx^{2i})^3)`$:
$$f^i(x_i)=\frac{1}{\sqrt{2}}(x_{2i1}x_{2i}).$$
Clearly, $`L_{m+1}`$ is the required embedding. $`\mathrm{}`$
Completion of the proof of Theorem 5.1. Finally we take an embedding
$$f_3:M^m^{(m+1)(m+2)+m}^{2m+2}$$
as follows.
$$f_3(x)=A^1f_1(x)(L_{n+1}f_2).$$
Since $`f_2`$ is an embedding, $`f_3`$ is the required embedding map for Theorem 5.1. $`\mathrm{}`$
5.5. Theorem. Suppose that $`C`$ is a compact subset in $`Cap_+^{4n}`$. Then any bounded domain $`D`$ in a linear statistical manifold $`(^n,g_0,AT_0)`$ can be realized as an immersed statistical submanifold of $`(Cap_+^{4n},g^F,T^{AC})`$.
Set
$$T^{}:=\underset{i=1}{\overset{n}{}}\frac{2dx_i^3}{x_i}.$$
We denote by $`S_{r,+}^n`$ the positive sector of the sphere of radius $`r`$ centered at the origin in $`^{n+1}`$.
Proof of Theorem 5.5. We choose a very large positive number
(5.1)
$$\overline{A}=\overline{A}(n,A)$$
to be specified in Lemma 5.1 later. First, $`\overline{A}`$ in (5.1) is required to be so large such that there exists a number $`1<\lambda <2`$ satisfying the following equation
(5.2)
$$\lambda ^2+\frac{3n}{(2\overline{A})^2}=4.$$
Equation (5.2) implies that $`(\lambda ,(2\overline{A})^1,(2\overline{A})^1,(2\overline{A})^1)^4`$ is a point in $`S_{2/\sqrt{n},+}^3`$. Hence there exists a positive number $`r(\overline{A})`$ such that for all $`0<rr(\overline{A})`$ the ball $`U(\overline{A},r)`$ of radius $`r`$ in the sphere $`S_{2/\sqrt{n}}^3`$ that is centered at the point $`(\lambda ,(2\overline{A})^1,(2\overline{A})^1,(2\overline{A})^1)`$ belongs also to the positive quadrant $`S_{2/\sqrt{n},+}^3`$. Hence $`U(\overline{A},r)\times _{\text{ n times }}U(\overline{A},r)`$ is a subset in $`S_{2,+}^{4n1}^{4n}`$.
Next, we note that Theorem 5.5 is a consequence of the following
###### Lemma 5.1.
For given positive numbers $`R>0`$ and $`A0`$ there exist a positive number $`\overline{A}`$, satisfying (5.2) and depending only on $`n`$ and $`A`$, a positive number $`r<r(\overline{A})`$ and an isostatistical immersion $`h`$ from the bounded domain $`[0,R]\times _{\text{ n times}}[0,R](^n,g_0,AT_0)`$ into $`(Cap_+^{4N},g^F,T^{AC})`$ such that $`h:[0,R]\times _{\text{ n times}}[0,R]U(\overline{A},r)\times _{\text{ n times }}U(\overline{A},r)`$.
###### Proof.
Set
$$T^{}:=\underset{i=1}{\overset{n}{}}\frac{2dx_i^3}{x_i}.$$
Since $`(U(\overline{A},r),(g_0)|_{U(\overline{A},r)},T^{}|_{U(\overline{A},r)})`$ is a statistical submanifold of $`(_+^4,g_0,T^{})`$, the direct product
$$(U(\overline{A},r)\times _{\text{n times}}U(\overline{A},r),_{i=1}^n(g_0)|_{U(\overline{A},r)},_{i=1}^nT^{}|_{U(\overline{A},r)})$$
is a statistical submanifold of $`(_+^{4n},g_0,T^{})`$. Since $`(Cap_+^N,g^F,T^{AC})`$ is a statistical submanifold of $`(R_+^N,g_0,T^{})`$ , we conclude that
$$(U(\overline{A},r)\times _{\text{ n times}}U(\overline{A},r),_{i=1}^n(g_0)|_{U(\overline{A},r)},_{i=1}^nT^{}|_{U(\overline{A},r)})$$
is a statistical submanifold of $`(Cap_+^{4N},g^F,T^{AC})`$. Hence, to prove Lemma 5.1, it suffices to show that there are positive numbers $`\overline{A}=\overline{A}(n,A)`$, $`r<r(\overline{A})`$ and an isostatistical immersion $`f:([0,R],dx^2,Adx^3)(U(\overline{A},r),(g_0)|_{U(\overline{A},r)},T^{}|_{U(\overline{A},r)})`$. On $`U(\overline{A},r)`$ we consider the distribution $`D(\rho )`$ defined by
$$D_x(\rho ):=\{vT_xU(\overline{A},r):|v|_{g_0}=1,T^{}(v,v,v)=\rho \}$$
for any given $`\rho >0`$. Clearly the existence of an isostatistical immersion $`f:([0,],dx^2,Adx^3)(U(\overline{A},r),(g_0)|_{U(\overline{A},r)},T^{}|_{U(A,r)})`$ is equivalent to the existence of an integral curve with the length $`R`$ of the distribution $`D(A)`$ on $`U(\overline{A},r)`$.
Now we are going to prove the following
###### Lemma 5.2.
There exist a positive number $`\overline{A}=\overline{A}(n,A)`$ and an embedded torus $`T^2`$ in $`U(\overline{A},r)`$ which is provided with a unit vector field $`V`$ on $`T^2`$ such that $`T^{}(V,V,V)=A`$.
###### Proof of Lemma 5.2.
Let us denote
$$x_0:=(\lambda ,(2\overline{A})^1,(2\overline{A})^1,(2\overline{A})^1)S^3(2/\sqrt{n})$$
with $`\lambda `$ defined by (5.2). We shall need the following
###### Lemma 5.3.
There exists a positive number $`\overline{A}=\overline{A}(n,A)`$ such that the following assertion holds. Let $`H`$ be any 2-dimensional subspace in $`T_{x_0}U(\overline{A},r)^4`$. Then there exists a unit vector $`wH`$ such that $`T^{}(w,w,w)\sqrt{2}A`$.
###### Proof of Lemma 5.3.
Denote by $`\stackrel{}{x}_0`$ the vector in $`^4`$ with the same coordinates as those of the point $`x_0`$. For any given $`H`$ as in Lemma 5.3 there exists a unit vector $`\stackrel{}{h}`$ in $`^4`$, which is not co-linear with $`\stackrel{}{x}_0`$ and which is orthogonal to $`H`$, such that a vector $`w^4`$ belongs to $`H`$ if and only if $`w`$ is a solution to following two linear equations:
(5.3)
$$w,\stackrel{}{x}_0=0,$$
(5.4)
$$w,\stackrel{}{h}=0.$$
Adding a multiple of $`\stackrel{}{x}_0`$ to $`\stackrel{}{h}`$ if necessary, and taking the normalization, we can assume that
$$\stackrel{}{h}=(0=h_1,h_2,h_3,h_4)\text{ and }\underset{i}{}h_i^2=1.$$
Case 1. Suppose that not all the coordinates $`h_i`$ of $`\stackrel{}{h}`$ are of the same sign, so w.l.o.g. we assume that $`h_1=0,h_20,h_3>0`$. We put
$`k_2:={\displaystyle \frac{h_2}{\sqrt{(h_2)^2+(h_3)^2}}},k_3:={\displaystyle \frac{h_3}{\sqrt{(h_2)^2+(h_3)^2}}},`$
(5.5) $`w:=(w_1,w_2=(1\epsilon _2)k_3,w_3=(1\epsilon _2)k_2,0=w_4)^4.`$
Obviously, for any choice of $`w_1`$ and $`\epsilon _2`$ the equation (5.4) for $`w`$ is satisfied. Now we choose $`w_1,\epsilon _2`$ to be solutions of the following equations
(5.6)
$$\lambda w_1+(1\epsilon _2)(2\overline{A})^1(k_2+k_3)=0,$$
(5.7)
$$w_1^2=(2\epsilon _2\epsilon _2^2).$$
Note that (5.6) is equivalent to (5.3) and (5.7) normalizes $`w`$. From (5.6) we get
(5.8)
$$w_1=\frac{(1\epsilon _2)(k_2+k_3)}{\lambda 2\overline{A}}.$$
Substituting the value of $`w_1`$ into (5.7), we get
$$(\frac{(k_2+k_3)^2}{(\lambda 2\overline{A})^2}+1)\epsilon _2^2(2+\frac{2(k_2+k_3)^2}{(\lambda 2\overline{A})^2})\epsilon _2+(\frac{k_2+k_3}{\lambda 2\overline{A}})^2=0,$$
which we simplify as follows:
(5.9)
$$\epsilon _2^22\epsilon _2+\frac{(k_2+k_3)^2}{(k_2+k_3)^2+4\lambda ^2\overline{A}^2}=0.$$
Clearly, the following choice of $`\epsilon _2`$ is a solution to (5.9)
(5.10)
$$\epsilon _2=1\frac{2\lambda \overline{A}}{\sqrt{(k_2+k_3)^2+4\lambda ^2\overline{A}^2}}.$$
By our assumption on $`h_2`$ and $`h_2`$, we have $`0k_2,k_31`$. Since $`1<\lambda <2`$ by (5.2), we conclude that when $`\overline{A}`$ goes to infinity, the value $`\epsilon _2`$ goes to zero. Hence there exists a number $`N_1>0`$ such that if $`\overline{A}>N_1`$ then
(5.11)
$$\epsilon _2>0\text{ and }(1\epsilon _2)^2\frac{3}{4}.$$
We shall show that for $`\epsilon _2`$ in (5.10) that also satisfies (5.11) if $`\overline{A}`$ is sufficiently large, and for $`w_1`$ defined by (5.8), the vector $`w`$ defined by (5.5) satisfies the required condition of Lemma 5.3). Since $`x_0=(\lambda ,(2\overline{A})^1,(2\overline{A})^1,(2\overline{A})^1)`$ we have
(5.12)
$$T_{x_0}^{}(w,w,w)=\frac{2w_1^3}{\lambda }+(4\overline{A})(w_2^3+w_3^3).$$
Now assume that $`\overline{A}>N_1`$. Noting that $`\epsilon _2`$ is positive and close to zero, and using $`k_20`$, $`k_30`$, we obtain from (5.5)
(5.13)
$$w_20,w_30.$$
Since $`0<\epsilon <1`$, $`0<k_2+k_3<2`$, and $`\lambda ,\overline{A}`$ are positive, we obtain from (5.8)
(5.14)
$$w_1<0\text{ and }|w_1|<\frac{1}{\lambda \overline{A}}.$$
Taking into account (5.5) and (5.11), we obtain
(5.15)
$$w_2^2+w_3^2=(1\epsilon _2)^2\frac{3}{4}.$$
Using (5.14), we obtain from (5.12)
(5.16)
$$T_{x_0}^{}(w,w,w)\frac{2}{\lambda ^4\overline{A}^3}+(4\overline{A})(w_2^3+w_3^3).$$
Observing that the function $`x^{3/2}+(cx)^{3/2}`$ is convex on interval $`[0,c]`$ for any $`c>0`$, using (5.13) and (5.15), we obtain from (5.16)
(5.17)
$$T_{x_0}^{}(w,w,w)\frac{2}{\lambda ^4\overline{A}^3}+(4\overline{A})2(\frac{\sqrt{3}}{\sqrt{2}})^3)=\frac{2}{\lambda ^4\overline{A}^3}+8(\sqrt{\frac{3}{2}})^3\overline{A}.$$
Increasing $`\overline{A}`$ if necessary, noting that $`1<\lambda =\lambda (A)`$, equation (5.17) implies that there exists a large positive number $`\overline{A}(n,A)`$ depending only on $`n`$ and $`A`$ such that any subspace $`H`$ defined by the equations (5.3) and (5.4), where $`h`$ is in Case 1, contains a unit vector $`w`$ that satisfies the condition in Lemma 5.3, i.e. the RHS of (5.17) is larger than $`\sqrt{2}A`$.
Case 2. W.l.o.g. we assume that $`h_2h_3h_4>0`$ and therefore we have
(5.18)
$$\alpha :=\frac{h_2+h_3}{h_4}2.$$
We shall search the required vector $`w`$ for Lemma 5.3 in the following form
(5.19)
$$w:=(w_1,w_2=(1\epsilon _2),w_3=(1\epsilon _2),w_4=\alpha (1\epsilon _2)).$$
The equations (5.19) and (5.18) ensure that $`w,\stackrel{}{h}=0`$ for any choice of parameters $`(w_1,\epsilon _2)`$ of $`w`$ in (5.19). Next we require that the parameters $`(w_1,\epsilon _2)`$ of $`w`$ satisfy the following two equations
(5.20)
$$\lambda w_1+\frac{(1\epsilon _2)(\alpha 2)}{2\overline{A}}=0,$$
(5.21)
$$w_1^2+(1\epsilon _2)^2(2+\alpha ^2)=1.$$
Note that (5.20) is equivalent to (5.3) and (5.21) normalizes $`w`$. From (5.20) we express $`w_1`$ in terms of $`\epsilon _2`$ as follows
(5.22)
$$w_1=\frac{(1\epsilon _2)(\alpha 2)}{\lambda 2\overline{A}}.$$
Set
(5.23)
$$B:=(2+\alpha ^2)+\frac{(\alpha 2)^2}{4\lambda ^2\overline{A}^2}.$$
Plugging (5.22) into (5.21) and using (5.23), we obtain the following equation for $`\epsilon _2`$
$$(1\epsilon _2)^2B1=0,$$
which is equivalent to the following equation
(5.24)
$$(1\epsilon _2)^2=\frac{1}{B}.$$
Since $`\alpha 2`$ by (5.18), from (5.23) we have $`B>0`$. Clearly
(5.25)
$$\epsilon _2:=1\frac{1}{\sqrt{B}}$$
is a solution to (5.24).
Since $`\alpha 2`$ and $`\epsilon _21`$ by (5.25), we obtain from (5.22) that $`w_10`$. Taking into account $`1<\lambda `$, $`\overline{A}>0`$, we derive from (5.22) and (5.25) the following estimates
$$T_{x_0}^{}(w,w,w)=\frac{2w_1^3}{\lambda }+(4\overline{A})(1\epsilon _2)^3(\alpha ^32)$$
$$>2w_1^3+(4\overline{A})(\alpha ^32)(1\epsilon _2)^3$$
$$=\frac{(\alpha 2)^3}{4\overline{A}^3(\sqrt{B})^3}+4\overline{A}\frac{(\alpha ^32)}{(\sqrt{B})^3}$$
$$\frac{\alpha ^32}{4\overline{A}^3(\sqrt{B})^3}+4\overline{A}\frac{(\alpha ^32)}{(\sqrt{B})^3}(\text{ since }\alpha 2)$$
(5.26)
$$=\frac{\alpha ^32}{(\sqrt{B})^3}(\frac{1}{4\overline{A}^3}+4\overline{A}).$$
###### Lemma 5.4.
There exists a large number $`\overline{A}=\overline{A}(n,A)`$ depending only on $`n`$ such that for all choice of $`\alpha 2`$ we have
$$\frac{(\alpha ^32)}{(\sqrt{B})^3}\frac{1}{10^2}.$$
###### Proof.
To prove Lemma 5.4 it suffices to show that for $`\alpha 2`$ we have
(5.27)
$$10^4(\alpha ^32)^2B^3.$$
Clearly there exists a positive number $`N_2`$ such that if $`\overline{A}>N_2`$, then by (5.23), we have
(5.28)
$$B<\frac{3}{2}(2+\alpha ^2)$$
for any $`\alpha 2`$. Hence (5.27) is a consequence of the following relation
(5.29)
$$10^4(\alpha ^32)^2[\frac{3}{2}(2+\alpha ^2)]^3,$$
which we shall establish now. To prove (5.29) it suffices to show that
(5.30)
$$10^3(\alpha ^32)^2(2+\alpha ^2)^3.$$
The inequality (5.30) is equivalent to the following
(5.31)
$$999\alpha ^66\alpha ^44000\alpha ^312\alpha ^2+39920.$$
Since $`\alpha 2`$ it follows that $`\alpha ^38`$ and hence
(5.32)
$$999\alpha ^64000\alpha ^3=499\alpha ^6+500\alpha ^3(\alpha ^38)499\alpha ^6.$$
Using $`2\alpha ^66\alpha ^4`$, we obtain
(5.33)
$$499\alpha ^66\alpha ^4497\alpha ^6.$$
Using $`a^416`$, we obtain
(5.34)
$$497\alpha ^612\alpha ^2=496\alpha ^6+\alpha ^2(\alpha ^412)>496\alpha ^6.$$
From (5.32), (5.33), (5.34) we obtain
(5.35)
$$999\alpha ^66\alpha ^44000\alpha ^312\alpha ^2+3992496\alpha ^6+3992>0.$$
This proves (5.30) and hence completes the proof of Lemma 5.4. โ
Lemma 5.4 implies that when $`\overline{A}=\overline{A}(A,n)`$ is sufficiently large, the RHS of (5.26) is larger than $`\sqrt{2}A`$. This proves the existence of $`\overline{A}`$, which depends only on $`n`$ and $`A`$, for Case 2.
This completes the proof of Lemma 5.3.
From Lemma 5.3 we obtain immediately the following.
###### Corollary 5.5.
The exists a small neighborhood $`U_1x_0`$ in $`\overline{U}(\overline{A},r)`$ such that the following statement holds. For any $`xU_1`$ and any two-dimensional subspace $`HT_xU_1`$ we have
$$\mathrm{max}\{T^{}(v,v,v)|vH\text{ and }|v|_{g_0}=1\}\frac{5}{4}A.$$
Completion of the proof of Lemma 5.2. Let $`\overline{A}=\overline{A}(n,A)`$ satisfy the condition of Lemma 5.3. Now we choose a small embedded torus $`T^2`$ in $`U_1U(\overline{A},r)`$. By Corollary 5.5, for all $`xT^2`$ we have
(5.36)
$$\mathrm{max}\{T^{}(v,v,v)|vT_xT^2\text{ and }|v|_{g_0}=1\}\frac{5}{4}A.$$
Denote by $`T_1T^2`$ the bundle of the unit tangent vectors of $`T^2`$. Since $`T^2=^2/^2`$ is parallelizable, we have $`T_1T^2=T^2\times S^1`$. Thus the existence of a vector field $`V`$ required in Lemma 5.2 is equivalent to the existence of a function $`T^2S^1`$ satisfying the condition of Lemma 5.2. Next we claim that there exists a unit vector field $`W`$ on $`T^2`$ such that $`T^{}(W,W,W)=0`$. First we choose some orientation for $`T^2`$, that induces an orientation on $`T_1T^2`$ and hence on the circle $`S^1`$. Take an arbitrary unit vector field $`W^{}`$ on $`T^2`$, equivalently we pick a function $`W^{}:T^2S^1`$. Now we consider the fiber bundle $`F`$ over $`T^2`$ whose fiber over $`xT^2`$ consists of the interval $`[W^{},W^{}]`$ defined by the chosen orientation on the circle of unit vectors in $`T_xS^2`$. Since $`T^{}(W^{},W^{},W^{})=T^{}(W,W,W)`$, for each $`xT^2`$ there exists a value $`W`$ on $`F(x)`$ such that $`T^{}(W,W,W)=0`$ and $`W`$ is closest to $`W^{}`$. Using $`W`$ we identify the circle $`S^1`$ with the interval $`[0,1)`$. The existence of $`W`$ implies that the existence of a function $`V:T^2[0,1)`$, regarded as a unit vector field $`V`$ on $`T^2`$, that satisfies the condition of Lemma 5.2 is equivalent to the existence of a function $`f:T^2[0,1)`$ satisfying the same condition. Now let $`V(x)`$ be the smallest value of unit vector $`V(x)[0,1)S^1(T_xT^2)`$ such that
$$T^{}(V(x),V(x),V(x))=A$$
for each $`xT^2`$. The existence of $`V(x)`$ follows from (5.36). This completes the proof of Lemma 5.2. โ
As we have noted, Lemma 5.2 implies Lemma 5.1. โ
This finishes the proof of Theorem 5.5. $`\mathrm{}`$
###### Proof of Main Theorem.
The existence of an isostatistical immersion of a compact statistical manifold $`(M,g,T)`$ into $`(Cap_+^N,g^F,T^{AC})`$ for some finite $`N`$ follows from Theorem 5.1 and Theorem 5.5.
###### Theorem 5.6.
Any smooth ($`C^1`$ resp.) compact statistical manifold $`(M^n,g,T)`$ admits an isostatistical embedding into the statistical manifold $`(๐ซ_+([N]),g^F,T^{AC})`$ for some finite number $`N`$.
###### Proof.
To prove Theorem 5.6 we repeat the proof of Main Theorem, replacing the Nash immersion theorem by the Nash embedding theorem. First we observe that our immersion $`f_3`$ constructed in the proof of Lemma 5.4 is an embedding, if $`f_2`$ is an isometric embedding. The existence of an isometric embedding $`f_2`$ is ensured by the Nash theorem. Hence, if $`M^n`$ is compact, to prove the existence of an isostatistical embedding of $`(M^n,g,T)`$ into $`(๐ซ_+([N]),g^F,T^{AC})`$ it suffices to prove the strengthened version of Theorem 5.5, where the existence of an isostatistical immersion is replaced by the existence of an isostatistical embedding.
Recall that the proof of Theorem 5.5 is reduced to the proof of the existence of an isostatistical immersion of a bounded statistical interval $`([0,R],dt^2,Adt^3)`$ into a torus $`T^2`$ of a small domain in $`(S_{2/\sqrt{n},+}^3,g_0,T^{})`$, see the proof of Lemma 5.1. Here for simplicity of notation, we abbreviate the restriction of $`T^{}`$ to the sphere in consideration as $`T^{}`$.
The statistical immersion produced with the help of Lemma 5.2 will be an embedding if not all the integral curves of the distribution $`D(A)`$ on the torus $`T^2`$ are closed curves. Now we shall search for an isostatistical embedding of $`([0,R],dt^2,Adt^3)`$ into a torus $`T^2\times T^2`$ of a small domain in $`(S_{1/\sqrt{n},+}^3,g_0,T^{})\times (S_{1/\sqrt{n},+}^3,g_0,T^{})(^8,g_0,T^{})`$. Since $`T^4`$ is parallelizable, repeating the argument at the end of the proof of Lemma 5.1, we choose a distribution $`D(A)TT^4`$ such that $`D(A)=T^4\times S^2`$ and
$$D_xA=\{vT_xT^4||v|_{g_0}=1,\text{ and }T^{}(v,v,v)=A\}.$$
Now assume that the integral curves of $`D(A)`$ that lie on the first factor $`T^2\times y`$ for all $`yS_{1/\sqrt{n},+}^3`$ are closed. Since $`T^2`$ is compact, there is a positive number $`p_1`$ such that the periods of these integral curves are at least $`p_1`$.
Now let us consider the following integral curve $`\gamma (t)`$ of $`D(A)`$ on $`T^4`$. The curve $`\gamma (t)`$ begins at a point $`(0,0,0,0)T^4`$. Here we identify $`T^1`$ with $`[0,1]/(0=1)`$. The integral curve lies on $`T^2\times (0,0)`$ until it approaches $`(0,0,0,0)`$ again. Since $`D_x(A)=S^2`$, we can slightly modify the direction of $`\gamma (t)`$ and let it leave the torus $`T^2\times (0,0)`$ and after a very short time $`\gamma (t)`$ must stay on the torus $`T^2\times (\epsilon ,\epsilon )`$ where $`\epsilon `$ is sufficiently small. W.l.o.g. we assume that the period of any closed curve of the distribution $`D(A)T(T^2\times (\epsilon ,\epsilon ))`$ is at least $`p_1`$. Repeating this procedure, since $`R`$ and $`p_1`$ are finite, we produce an embedding of $`([0,R],dt^2,Adt^3)`$ into $`T^4(S_{1/\sqrt{n},+}^3,g_0,T^{})\times (S_{1/\sqrt{n},+}^3,g_0,T^{})`$.
This completes the proof of Theorem 5.6.
|
warning/0506/quant-ph0506133.html
|
ar5iv
|
text
|
# The cryptographic power of misaligned reference frames
## An Asymptotically Secure Scheme
For our scenario we first assume that Bob can measure the coordinates of a three-dimensional vector that he receives with infinite precision. We then argue that the security still holds in the case of finite precision.
The idea behind the scheme is two-fold. We show that a single use of a channel can give rise to an (asymptotically) perfectly secure commitment when the channel acts on vectors in an arbitrarily high-dimensional space. However, spatial reference frames are only three-dimensional objects which seems to suggest that only a partially secure bit commitment may be achievable. Our scheme overcomes this apparent problem by parametrizing three-dimensional vectors by coordinates of a $`d`$-dimensional lattice.
Alice sends a vector $`\stackrel{}{v}^3`$ to Bob. In Bobโs spatial frame of reference this vector looks like $`R\stackrel{}{v}`$ for some random rotation $`R`$ taken from the probability distribution $`\mu `$. The distribution $`\mu `$ of $`R`$ is the following. Let there be a set of angles $`\theta \{\theta _1,\theta _2,\mathrm{},\theta _d\}`$ which are not linearly related, meaning that for all integers (positive or negative) $`n_1,n_2,\mathrm{},n_d`$, $`_{i=1}^dn_i\theta _i=0mod\pi `$ if and only if $`n_1=n_2=\mathrm{}=n_d=0`$. One of these angles $`\theta `$ is picked uniformly at random. Then $`R`$ is given by
$$R=\{\begin{array}{cc}\left(\begin{array}{ccc}\mathrm{cos}\theta & \mathrm{sin}\theta & 0\\ \mathrm{sin}\theta & \mathrm{cos}\theta & 0\\ 0& 0& 1\end{array}\right)\hfill & \text{with prob. }1/2,\hfill \\ & \\ \left(\begin{array}{ccc}\mathrm{cos}2\theta & \mathrm{sin}2\theta & 0\\ \mathrm{sin}2\theta & \mathrm{cos}2\theta & 0\\ 0& 0& 1\end{array}\right)\hfill & \text{with prob. }1/2.\hfill \end{array}$$
(1)
Let us describe our bit commitment protocol with this distribution $`\mu `$, known to Alice and Bob. Let $`L`$ be a positive integer.
* Commit. To commit to $`b\{0,1\}`$, Alice chooses $`๐=(a_1,\mathrm{},a_d)\{0,1,\mathrm{},L1\}^d`$ uniformly at random, but conditioned on $`_{i=1}^da_i=bmod2`$. Let $`\alpha (๐)=_{i=1}^da_i\theta _i`$. Alice sends the vector $`\stackrel{}{v}=\stackrel{}{v}(๐)=(\mathrm{cos}\alpha (๐),\mathrm{sin}\alpha (๐),0)`$. Bob receives this vector as $`R\stackrel{}{v}(๐)=\stackrel{}{v}(๐^{})`$ with rotated angle $`\alpha ^{}=_ia_i^{}\theta _i`$. He determines the $`d`$-dimensional lattice vector $`๐^{}`$. If he cannot find such a vector, he aborts the protocol.
* Reveal. In the reveal phase, Alice simply sends the classical bits $`(b,๐)`$ to Bob. Bob accepts when $`b=_ia_imod2`$ and all coordinates of $`๐๐^{}`$ are 0 except one for which $`a_i^{}a_i`$ is either 1 or 2. Otherwise he aborts.
Let us now show that this protocol has the desired security properties.
Soundness. We assume that both parties are honest. To understand a given realization of this protocol, let $`j`$ be such that the randomly chosen $`\theta `$ equals $`\theta _j`$. In that case the coordinates $`a_i^{}`$ of $`๐^{}`$ are
$$a_i^{}=\{\begin{array}{cc}a_i+\delta _{ij}\hfill & \text{with prob. }1/2,\hfill \\ a_i+2\delta _{ij}\hfill & \text{with prob. }1/2.\hfill \end{array}$$
(2)
Due to the linear independence of the $`\theta `$-angles Bob can perfectly compute $`๐^{}\{0,1,\mathrm{},L+1\}^d`$ from $`\stackrel{}{v}(๐^{})`$. Bob will accept Aliceโs message in the reveal phase since this vector $`๐^{}`$ is within distance 2 of the original $`๐`$ and the noise acts on only one of the coordinates.
Concealing. A bit commitment protocol is called $`ฯต`$-concealing when for two different messages $`b=0`$ and $`b=1`$ the distributions over random variables as viewed by Bob are $`ฯต`$-close with respect to their variation distance. In our case Bob learns $`๐^{}`$, thus we consider the distance $`ฯต=_๐^{}|(๐^{}b=0)(๐^{}b=1)|`$. For an $`๐^{}`$ with all coordinates at least 2 and at most $`L`$ we have $`\left(๐^{}b=0\right)=\left(๐^{}b=1\right)`$. All other $`๐^{}`$s are โboundaryโ cases, which we denote as $`๐^{}`$, for which these two conditional probabilities can be different. We upper bound this boundary term as
$`{\displaystyle \underset{๐^{}}{}}|(๐^{}|b=0)(๐^{}|b=1)|{\displaystyle \underset{๐^{}}{}}\left(๐^{}\right)`$
$`1\left({\displaystyle \frac{L1}{L+2}}\right)^d,`$ (3)
which can be made arbitrarily small for large enough $`L`$ for any fixed $`d`$.
Binding. Consider Aliceโs cheating strategies. She could have sent a different vector, say $`\stackrel{}{w}(\beta )`$. In case $`\stackrel{}{w}`$ is not in the x-y plane or if $`\beta _ib_i\theta _i`$ for some integers $`b_i\{0,\mathrm{},L+1\}`$ Bob simply aborts. Alice could try to cheat by revealing an $`๐^{}`$ and $`b^{}b`$ that pass Bobโs test. Aliceโs best option is to choose $`๐^{}`$ that is the same as $`๐`$ except, say, the $`k`$th coordinate, which is $`a_k^{}=a_k+1`$. This implies that that the parity $`b^{}`$ of $`๐^{}`$ is opposite to $`b`$. With probability $`1/d`$ the noise acts on the $`k`$th coordinate and so $`(๐^{},b^{})`$ passes Bobโs test. The protocol is $`1/d`$-binding.
In reality we should assume that Bob can only determine $`R\stackrel{}{v}`$ with finite precision, which means that Bob finds some vector $`\stackrel{}{w}`$ at Euclidean distance $`\epsilon `$ from $`R\stackrel{}{v}`$. If $`\epsilon `$ is small enough, we can ensure that for all $`๐ฑ,๐ฒ\{0,\mathrm{},L+1\}^d`$, $`\stackrel{}{v}(๐ฑ)\stackrel{}{v}(๐ฒ)>2\epsilon `$, so Bob can still determine $`๐^{}`$ from $`\stackrel{}{w}R\stackrel{}{v}`$, if Alice behaves honestly.
But what if Alice cheats and sends some arbitrary vector $`\stackrel{}{w}`$ to Bob? Notice, however, that this strategy could only work if $`R\stackrel{}{w}\stackrel{}{v}(๐^{})\epsilon `$, which happens if and only if $`\stackrel{}{w}\stackrel{}{v}(๐)\epsilon `$. In particular, $`๐`$ is the only $`d`$-dimensional vector such that $`\stackrel{}{v}(๐)`$ is $`\epsilon `$-close to $`\stackrel{}{w}`$. Thus if Alice later reveals $`๐^{}๐`$, her cheating still succeeds only with probability $`1/d`$.
Remark: By increasing $`d`$ and running the above protocol in parallel several times, it is also possible to commit more than one bit.
It remains an open problem to get a more complete overview of the (im)possibility of bit commitment for general distributions $`\mu `$. In particular it would be interesting to find realistic noise models for, say, polarized photons, that would allow for secure or approximately secure bit commitment.
RO and BMT would like to acknowledge support by the NSA and the ARDA through ARO contract number W911NF-04-C-0098. AWH thanks the IBM Watson quantum information group for their hospitality while doing this work.
|
warning/0506/physics0506028.html
|
ar5iv
|
text
|
# Influence of saving propensity on the power law tail of wealth distribution
## I Introduction
Multi-agent models of closed economy systems have received considerable attention in recent years due to the fact that they seem to predict realistic shapes of wealth distribution from very simple underlying dynamics, basically equivalent to kinetic theory of ideal gases in classical statistical mechanics Bennati (1988a, b, 1993); Ispolatov et al. (1998); Dragulescu and Yakovenko (2000); Chakraborti and Chakrabarti (2000); Chakraborti (2002); Chatterjee et al. (2003); Iglesias et al. (2004); Chatterjee et al. (2004); Repetowicz et al. ; Patriarca et al. (2004a, b); Chatterjee et al. ; Patriarca et al. (2005). A notable plus-point of these simple models is represented by the ability to reproduce the main features of the empirical wealth distributions: a Boltzmann distribution at intermediate values of wealth, and a power law at the highest values (see e.g. Dragulescu and Yakovenko (2001a, b); Fujiwara et al. (2003); Levy and Solomon (1997); Sinha ). The power law form in the tail of the distribution was observed more than a century ago by the economist Vilfredo Pareto Pareto (1897), who found that the wealth of individuals in a stable economy has a cumulative distribution $`F(x)x^\alpha `$, where $`\alpha `$, the Pareto exponent, has usually a value between $`1`$ and $`2`$.
In these models $`N`$ agents interact exchanging a quantity $`x`$, which can be interpreted as representing any economic entity contributing to the agent wealth, expressed in the same unit of measure, e.g. in monetary units. Depending on the parameters of the kinetic model, in particular on the values of the saving propensities $`\{\lambda _i\}`$ ($`i=1,\mathrm{},N`$) of the $`N`$ agents, the equilibrium wealth distribution can be a simple Boltzmann distribution for $`\lambda _i0`$ Bennati (1988a, b); Dragulescu and Yakovenko (2000), a Gamma distribution with a similar exponential tail but a well defined mode $`x_\mathrm{m}>0`$ for $`\lambda _i\lambda _0>0`$ Chakraborti and Chakrabarti (2000); Chakraborti (2002), or a distribution with a power law tail for randomly distributed $`\lambda _i`$ Chatterjee et al. (2003, 2004). It has been recently recognized Chatterjee et al. (2004); Patriarca et al. (2005) that the observed power law arises from the overlap of Gamma distributions, resulting from (subsets of) agents with similar values of $`\lambda `$. That is, in systems where saving propensity is distributed according to an arbitrary distribution function $`f(\lambda `$), agents relax individually toward Maxwell-Boltzmann distributions, similarly to systems with a global saving propensity $`\lambda _0`$, but with the important difference that in this case the various Gamma distributions with different $`\lambda _0`$ parameters will overlap and provide the final (power law) equilibrium distribution.
The aim of the present paper is to further investigate the relation between the saving propensity distribution and the shape of the final equilibrium wealth distribution, with particular attention to reproduce a realistic distribution. In Sec. II we recall the main features of kinetic multi-agent models, while in Sec. III we consider how the equilibrium distribution is affected by a particular choice of the parameters of the saving propensity distribution and provide some examples. Results are summarized in Sec. IV.
## II Kinetic multi-agent models
In kinetic multi-agent models $`N`$ agents interact with each other through a pair interaction โ but this is only one of the many possibilities โ exchanging a quantity $`x`$, generally referred to as โwealthโ in the following. Agents are characterized by their current wealths $`\{x_i\},i=1,2,\mathrm{},N`$ and, possibly, by some parameters, such as the saving propensity $`\lambda _i`$. The evolution of the system is then carried out in the following way. At every time step two agents $`i`$ and $`j`$ are extracted randomly and an amount of wealth $`\mathrm{\Delta }x`$ is exchanged between them,
$`x_i^{}`$ $`=`$ $`x_i\mathrm{\Delta }x,`$
$`x_j^{}`$ $`=`$ $`x_j+\mathrm{\Delta }x.`$ (1)
It can be noticed that in this way the quantity $`x`$ is conserved during the single transactions, $`x_i^{}+x_j^{}=x_i+x_j`$, where $`x_i^{}`$ and $`x_j^{}`$ are the agent wealths after the transaction has taken place.
#### II.0.1 The basic model
In the basic versions of the model the quantity $`x`$ represents money and $`\mathrm{\Delta }x`$ the money exchanged, assumed to have a constant value Bennati (1988a, b, 1993),
$$\mathrm{\Delta }x=\mathrm{\Delta }x_0,$$
(2)
or to be proportional to the initial values Dragulescu and Yakovenko (2000),
$$\mathrm{\Delta }x=\overline{ฯต}x_iฯตx_j,$$
(3)
where $`ฯต`$ is a random number uniformly distributed between 0 and 1 and $`\overline{ฯต}=1ฯต`$. The form of $`\mathrm{\Delta }x`$ in Eq. (3) represents a random reshuffling of the wealths of the two agents Dragulescu and Yakovenko (2000), since Eq. (1) can in this case be rewritten as
$`x_i^{}`$ $`=`$ $`ฯต(x_i+x_j),`$
$`x_j^{}`$ $`=`$ $`\overline{ฯต}(x_i+x_j).`$ (4)
These dynamics rules, together with the constraint that transactions can take place only if $`x_i^{}>0`$ and $`x_j^{}>0`$, lead to an equilibrium state characterized by an exponential Boltzmann distribution,
$$f(x)=x^1\mathrm{exp}(x/x),$$
(5)
where the effective temperature $`T_\lambda `$ of the system is just the average wealth (see curve $`\lambda =0`$ in Fig. 1).
Despite its intrinsic simplicity, the basic model has the merit of having shown that economic interactions can be modeled in terms of simple statistical mechanisms leading to corresponding universal statistical laws.
#### II.0.2 Models with a global saving propensity
A first generalization toward a more realistic model is based on the introduction of a saving criterion. Agents save a fraction $`\lambda `$ (the saving propensity, with $`0<\lambda <1`$) before entering a trade and only exchange the remaining fraction $`(1\lambda )`$ of their wealth, Chakraborti and Chakrabarti (2000); Chakraborti (2002):
$`x_i^{}`$ $`=`$ $`\lambda x_i+ฯต(1\lambda )(x_i+x_j),`$
$`x_j^{}`$ $`=`$ $`\lambda x_j+\overline{ฯต}(1\lambda )(x_i+x_j),`$ (6)
corresponding to a $`\mathrm{\Delta }x`$ in Eq. (1) given by
$$\mathrm{\Delta }x=(1\lambda )[\overline{ฯต}x_iฯตx_j].$$
(7)
The corresponding equilibrium distribution is well fitted by the gamma distribution Patriarca et al. (2004a, b)
$$f(\xi )=\frac{1}{\mathrm{\Gamma }(D_\lambda /2)}\xi ^{D_\lambda /21}\mathrm{exp}(\xi )\gamma _{D_\lambda /2}(\xi ),$$
(8)
as shown in Fig. 1. Here the dimensionless variable
$$\xi =\frac{x}{T_\lambda },$$
(9)
is just the variable $`x`$ rescaled with respect to the effective temperature $`T_\lambda `$ and
$`{\displaystyle \frac{D_\lambda }{2}}`$ $`=`$ $`1+{\displaystyle \frac{3\lambda }{1\lambda }}={\displaystyle \frac{1+2\lambda }{1\lambda }},`$
$`T_\lambda `$ $`=`$ $`{\displaystyle \frac{1\lambda }{1+2\lambda }}x.`$ (10)
The parameter $`D_\lambda `$ plays the role of an effective dimension, since the Gamma distribution $`\gamma _n(\xi )`$ given by Eq. (8) is identical to the Maxwell-Boltzmann distribution of kinetic energy for a system of molecules at temperature $`T_\lambda `$ in $`D_\lambda `$ dimensions (of course only for integer or half-integer values of $`n=D_\lambda /2`$) Patriarca et al. (2004b); Patriarca et al. (2005). In further support of this analogy, it is worth noting that $`T_\lambda `$ and $`D_\lambda `$ are related to each other through an โequipartition theoremโ,
$$x=\frac{D_\lambda T_\lambda }{2}.$$
(11)
The equivalence between kinetic theory and closed economy models, suggested by the basic version of the kinetic multi-agent models Bennati (1988a, b, 1993); Dragulescu and Yakovenko (2000), can thus be extended to values $`\lambda 0`$ Patriarca et al. (2005), as summarized in Table 1.
While $`\lambda `$ varies between 0 and 1, the effective dimension $`D_\lambda `$ increases monotonically between 2 and $`\mathrm{}`$. In fact in a higher number of dimensions the fraction of kinetic energy exchanged between particles during a collision is smaller. At the same time, the market temperature $`T_\lambda `$ decreases with increasing $`\lambda `$, signaling smaller fluctuations of $`x`$ during trades, consistently with the presence of a saving criterion, i.e. $`\lambda >0`$. One can notice that $`T_\lambda =(1\lambda )x/(1+2\lambda )(1\lambda )x`$ is on average the amount of wealth exchanged during an interaction between agents, see Eqs. (6).
#### II.0.3 Models with a continuous distributions of saving propensity
As a further generalization, various investigations concerned models in which agents have realistically been diversified from each other by assigning them different saving propensities $`\lambda _i`$ Chatterjee et al. (2003); Das and Yarlagadda ; Chatterjee et al. (2004); Repetowicz et al. ; Chatterjee et al. ; Patriarca et al. (2005). In particular, uniformly distributed $`\lambda _i`$ in the interval (0,1) have been studied numerically in Refs. Chatterjee et al. (2003, 2004). This model is described by the trading rule
$`x_i^{}`$ $`=`$ $`\lambda _ix_i+ฯต[(1\lambda _i)x_i+(1\lambda _j)x_j],`$
$`x_j^{}`$ $`=`$ $`\lambda x_j+\overline{ฯต}[(1\lambda _i)x_i+(1\lambda _j)x_j],`$ (12)
or, equivalently, by a $`\mathrm{\Delta }x`$ โ as defined in Eq. (1) โ given by
$$\mathrm{\Delta }x=\overline{ฯต}(1\lambda _i)x_iฯต(1\lambda _j)x_j.$$
(13)
One of the main features of this model, which is supported by theoretical considerations Das and Yarlagadda ; Repetowicz et al. ; Chatterjee et al. , is that the wealth distribution exhibits a robust power law at large values of $`x`$,
$$f(x)=x^{\alpha 1},$$
(14)
with a Pareto exponent $`\alpha =1`$ largely independent of the details of the $`\lambda _i`$-distribution. As remarked in Ref. Chatterjee et al. (2004), the wealth distribution of the single agents are not of a power law type but have a well defined mode and an exponential tail, similarly to the case a global saving propensity $`\lambda _0`$. The power law actually results from the overlap of these partial distributions corresponding to the various $`\lambda `$โs, which are Gamma distributions, whose average value is proportional to $`1/(1\lambda )`$ and thus extend to very large values of $`x`$ Patriarca et al. (2005). These results are also in agreement with theoretical approaches to kinetic multi-agent models Repetowicz et al. ; Das and Yarlagadda ; Chatterjee et al. . This phenomenon is illustrated in Fig. 2.
Arbitrarily small (random) irregularities in the distance between two consecutive values of $`\lambda `$ close enough to 1, in a uniform distribution of saving propensity, are amplified in the wealth distribution as a consequence of the correlation between average wealth and saving propensity, resulting in isolated peaks in the wealth distribution Patriarca et al. (2005). This is shown by the simple example in Fig. 3, where two distributions, in principle equivalent to each other since associated to a uniform $`\lambda `$-distribution in (0,1), look actually quite different: The first distribution (left) has been obtained by randomly extracting the values of $`\lambda `$ with a random number generator in the interval (0,1). It can be noticed that the corresponding equilibrium wealth distribution is more irregular than that obtained from the second distribution (right), in which the values $`\lambda _i`$ were set to be equidistant from each other in the interval (0,1) by defining them as $`\lambda _i=i/N`$. The deterministic and random uniform distributions are equivalent to each other in principle but not in practice (within numerical simulations), where a finite number of agents is necessarily used. The reason is that the $`\lambda `$ values extracted randomly present fluctuations and therefore wider intervals between neighbor values which are amplified in the final wealth distribution. Since in these numerical simulations one tries to mimic continuous distributions by use of the smallest possible number of variables, it is convenient to avoid the irregular fluctuations present in randomly extracted sets of numbers and use a deterministically extracted sets $`\{\lambda _i\}`$ of saving propensities. This can be achieved easily with the method prescribed in Appendix A.
## III Construction of a realistic model
Here we study how some aspects of the $`\lambda `$-distribution $`f(\lambda )`$ influence the equilibrium form of the wealth distribution $`f(x)`$, i.e. its shape at smaller and the tail at larger values of $`x`$, and in particular under which conditions an exponential and a power law can appear in different ranges of the same distribution.
### III.1 Wealth distribution at small and intermediate values of wealth
The equilibrium distribution of the basic model is the simple exponential function in Eq. (5). Such a form of distribution decreases monotonously with $`x`$ and does not have rich agents nor a power law tail, a point dealt with in greater detail in the next section. In the small $`x`$ limit, the exponential distribution is $`f(x0)>0`$ which implies that many agents have a wealth $`x0`$. In fact the mode of the distribution is $`x_\mathrm{m}=0`$ and the fraction of agents outside a given interval $`(0,x)`$ โ which is just the upper cumulative distribution function โ has a pure exponential form, $`F(x)=\mathrm{exp}(x/x)`$. Real data about wealth and income distributions, on the other hand, show that wealth distribution functions have a mode $`x_\mathrm{m}>0`$ Dragulescu and Yakovenko (2001b); SalaiMartin and Mohapatra (2002); SalaiMartin (2002); Aoyama et al. (2003); Ferrero (2004); Silva and Yakovenko (2005). The introduction of a (global) saving propensity $`\lambda >0`$ solves this problem Chakraborti and Chakrabarti (2000) since it leads to an equilibrium Gamma distribution Patriarca et al. (2004a, b), which has a mode $`x_m>0`$ and a zero limit for $`x0`$, see Fig. 1. This functional form has been shown to interpolate well real data about income distributions at small and intermediate values Dragulescu and Yakovenko (2001a, b); Ferrero (2004); Silva and Yakovenko (2005).
### III.2 The tail of the wealth distribution
The tail of wealth distributions is known to follow a power law with Pareto exponent between 1 and 2, depending on the sample analyzed. The model under consideration, when saving propensities are continuously distributed, predicts a power law tail in $`f(x)`$, despite with a lower Pareto exponent $`\alpha =1`$, a feature which has been shown to be very robust and independent of the details of $`f(\lambda )`$.
Both numerical and theoretical analyzes of kinetic multi-agent models show that agents with large values of $`\lambda `$โs (i.e. $`\lambda `$ close to 1) give a major contribution to the power law tail. This is illustrated e.g. by the fact that when the power law is decomposed into partial distributions of agents within a given interval of the saving propensity, the partial distribution corresponding to the interval with the highest $`\lambda `$ is in turn a power law, while the distributions corresponding to lower values of $`\lambda `$ are localized and have an exponential tail, as shown in Fig. 2. However, they sum up to give a power law at lower values of $`x`$. All this suggests that the crucial factor for having a power law extending beyond a certain value $`x`$ is the highest $`\lambda `$ present in the sample, that is the cutoff of the $`\lambda `$-distribution.
Thus, rather than varying the functional form of $`f(\lambda )`$, the influence of the cut-off $`\lambda _\mathrm{M}`$ of the $`\lambda `$-distribution โ which is a parameter characterizing numerical simulations as well as real systems โ has been analyzed. A uniform deterministic distribution of saving propensity for the $`N`$ agents in the interval ($`0,\lambda _\mathrm{M}`$), has been generated through the formula
$$\lambda _i=\left(\frac{i}{N}\right)\lambda _\mathrm{M},i=1,\mathrm{},N,\lambda _\mathrm{M}<1,$$
(15)
as described in greater detail in Appendix A.
In fact we found that varying the cutoff $`\lambda _\mathrm{M}`$ influences in turn the cut-off of the wealth distribution $`f(x)`$ and the shape of the distribution at small $`x`$ โ but not the shape of the tail which remains a power law with exponent $`\alpha =1`$. Decreasing $`\lambda _\mathrm{M}`$ has the effect to decrease the interval of wealth $`x`$ in which the power law appears, until it eventually disappears for $`\lambda _\mathrm{M}0.92`$. Results are shown in Fig. 4, where the various curves represent the distribution functions obtained for some values of the cutoff $`\lambda _\mathrm{M}`$ chosen in the interval $`\lambda _\mathrm{M}=(0.9,0.9999)`$ in a system of $`10^5`$ agents. Curves from left to right correspond to increasing values of cutoff.
The transition from an exponential to a power law form of the wealth distribution, as the cut-off $`\lambda _\mathrm{M}`$ decreases, takes place by a shrinking of the power law interval, rather than as a change of the functional form of the tail.
As a final remark it is to be noted that the cutoff of the $`\lambda `$-distribution is naturally linked to that of the $`x`$-distribution, as a consequence of the correlation existing between average wealth $`x`$ and saving propensity Patriarca et al. (2005) in this model,
$$x_i(1\lambda _i)=\mathrm{const}.$$
(16)
Here the constant on the right hand side of the equation is the same number for all the agents in the system. This relation clearly shows that the highest average wealth is determined in turn by the highest $`\lambda _i`$.
### III.3 Superposing an exponential form at intermediate values and a power law tail
In real wealth distributions, an exponential form at intermediate values of wealth is known to coexist with a power law tail at larger values Silva and Yakovenko (2005). The power law is mainly due to a small percentage of population, of the order of a few per cent, while the majority of the population with smaller average wealth give rise to the exponential part of the distribution. In this section we try to construct a realistic example of such a type of wealth distribution by collecting some of the results obtained so far:
* A global saving propensity $`\lambda _0`$ is associated to an equilibrium Gamma distribution, which always has an exponential tail.
* A set of agents with a continuous $`\lambda `$-distribution produces a power law in the equilibrium wealth distribution.
* The average wealth $`x_i`$ of an agent and the corresponding saving propensity $`\lambda _i`$ are linked to each other through Eq. (16), which implies that agents with high $`\lambda 1`$ contribute to the large-$`x`$ part of the distribution.
It is then natural to ask if a suitable $`\lambda `$-distribution may lead to the desired equilibrium wealth distribution. To answer this question we have constructed a hybrid $`\lambda `$-distribution โ on the base of the results listed above and a very similar prescription mentioned in Ref. Chatterjee et al. (2003)โ in the following way:
* A small fraction of agents $`p_0`$ with saving propensities $`\lambda _i`$ uniformly distributed in the interval (0,1) according to Eq. (19).
* The remaining fraction $`1p_0`$ with a constant value of the saving propensity $`\lambda _0`$.
The corresponding distribution for $`p_0=0.01`$ (1 per cent) and $`\lambda _0=0`$ is shown in Fig. 5, both in the small $`x`$-scale, where the distribution has an exponential shape, and in the long $`x`$-range, where the power law with exponent $`2`$, which characterizes this type of multi-agent model, is observed.
It is noteworthy that the coexistence of an exponential and a power law tail is possible only for small values of $`p_0`$, in agreement with the fact that it is a small percentage of the population in real systems that is responsible of the power law form of wealth distribution at large values of wealths. For larger values of $`p_0`$ the exponential part shrinks and the power law dominates. This effect is in a sense contrary to that considered in Sec. II.0.3, where decreasing the cutoff of the $`\lambda `$-distribution induced a shrinking of the power law range.
It may be noticed that, due to the choice $`\lambda _0=0`$ for that part of agents with a constant saving propensity, the distribution in Fig. 5 still has a mode $`x_\mathrm{m}=0`$. However, one recovers a distribution with a well defined mode $`x_\mathrm{m}>0`$ as soon as one chooses a $`\lambda _00`$. The distribution in Fig. 6 corresponds to a $`\lambda _0=0.2`$ for 99% of the agents and a uniform $`\lambda `$-distribution for the remaining agents.
### III.4 Meaning of the saving propensity
The central role of the saving propensity $`\lambda `$ โ or risk aversion as referred to in Ref. Iglesias et al. (2004) โ for the considerations made above is evident. However, it is to be remarked that the relation between saving propensity $`\lambda _i`$ of an agent and the corresponding average wealth $`x_i`$ should not considered to be of a cause-effect type. It is true that in the present model the $`\lambda _i`$โs are fixed parameters, so that the natural dynamical interpretation is that the saving propensity $`\lambda `$ determines the final average wealth. However, in a real situation the value of $`\lambda `$ itself may vary according to various factors, e.g. the wealth itself: a high average wealth probably puts the agent in a situation which allows to carry on trades investing the same amount of wealth while saving more respect to agents with smaller wealths. Therefore, the model contains in its very dynamics a positive correlation between $`\lambda `$ and $`x`$ supported by real data Dynan et al. (2004) but leaves the question of the actual dynamical relation between them to a more detailed microscopic analysis. Multi-agents models like that considered here describe flux of wealth on a mesoscopic level, i.e. on a coarse grained scale in time or wealth, rather than reflecting the single agent strategy to save or reduce risks.
## IV Conclusions
We have shown that within the framework of kinetic multi-agent models it is possible to obtain realistic wealth distributions $`f(x)`$ characterized by a zero limit for small $`x`$, and the coexistence of an exponential form at intermediate and power law tail at larger values of $`x`$. In agreement with observations on real systems, this is possible only if the percentage of rich agents does not exceed a critical threshold of the order of 1 per cent. Also, the model naturally produces a positive correlation between average wealth $`x`$ and saving propensity $`\lambda `$ exhibited in real data samples.
###### Acknowledgements.
Numerical computations were partially carried out on the facilities of the Laboratory of Computational Engineering, Helsinki University of Technology, under support by the Academy of Finland, Research Centre for Computational Science and Engineering, project no. 44897 (Finnish Centre for Excellence Program 2000-2005). The work at Brookhaven National Laboratory was carried out under Contract No. DE-AC02-98CH10886, Division of Material Science, U.S. Department of Energy.
## Appendix A Extraction of a variable $`\lambda `$ with cumulative distribution $`F(\lambda )`$
It is possible to define a sequence of $`N`$ numbers $`\lambda _i`$, $`i=1,\mathrm{},N`$, which becomes distributed according to an arbitrary distribution function $`f(\lambda )=dF(\lambda )/d\lambda `$ in the continuous limit ($`N\mathrm{}`$), in at least two ways, randomly or deterministically. The two methods are equivalent to each other only in the continuous limit, while in numerical simulations a finite $`N`$ is necessarily employed and they may provide different results. As discussed in Sec. II.0.3, in some cases it may be preferable to have a regular, rather than a randomly extracted sequence.
* Random extraction. A generator of random numbers $`\varphi `$, $`0<\varphi <1`$, uniformly distributed between 0 and 1, can be employed to extract a set of numbers $`\lambda _i`$ distributed in the continuous limit according to an arbitrary cumulative distribution function $`F(\lambda )`$, with $`F(0)=0`$ and $`F(1)=1`$. The cumulative distribution function for the random variable $`\varphi `$ is simply $`F(\varphi )=\varphi `$ and $`dFd\varphi `$ is the (constant) probability to extract the next random number between $`\varphi `$ and $`\varphi +d\varphi `$. The algorithm is based on the identity $`d\varphi =dF=f(\lambda )d\lambda `$, which shows that if values $`F_i`$ are extracted randomly and uniformly in the interval (0,1), then the corresponding values $`\lambda _i`$ obtained by inverting $`F=F(\lambda _i)`$ will be distributed with probability density $`f(\lambda )`$.
* Deterministic extraction The same result can be obtained by a deterministic assignment of the values $`\lambda _i`$ which does not make use of random number generators. If the sequence $`\{\lambda _i\}`$ is assumed to be labeled in increasing order, i.e. $`0\lambda _1<\lambda _2<\mathrm{}<\lambda _N1`$, then the function of $`i`$
$$\lambda (i)=\lambda _i,$$
(17)
increases monotonously with $`i`$ and it is possible to invert it to express $`i`$ as a function of $`\lambda _i`$ to define the function
$$F(\lambda _i)=\frac{i}{N},$$
(18)
which represents the fraction of agents with saving propensity less or equal to $`\lambda _i`$: $`F(\lambda )`$ is just the (lower) cumulative distribution function and as such $`0<F(\lambda )<1`$ for every $`\lambda `$, $`F(\lambda 0)0`$, and $`F(\lambda 1)1`$. For instance the cumulative distribution function of a uniformly distributed variable $`\lambda `$ in the interval $`\lambda (0,1)`$ is just the linear function $`F(\lambda )=\lambda `$, with $`0<\lambda <1`$. Then Eq. (18) provides the corresponding deterministic sequence as
$$\lambda _i=\frac{i}{N},i=1,\mathrm{},N.$$
(19)
If there is an upper cutoff $`\lambda _\mathrm{M}`$ in the distribution, the equation is modified as in (15). In the general case of a given cumulative function $`F(\lambda )`$, it is sufficient to invert Eq. (18) to obtain the sequence in the form $`\lambda _i=\lambda (i/N)`$, $`i=1,\mathrm{},N`$, where $`\lambda (\mathrm{})`$ is the inverse function of $`F(\mathrm{})`$. The values $`\lambda _i`$ thus obtained will be distributed in the continuous limit with a probability distribution $`f(\lambda )=dF(\lambda )/d\lambda `$.
|
warning/0506/gr-qc0506072.html
|
ar5iv
|
text
|
# On the area expectation values in area tensor Regge calculus in the Lorentzian domain
## Abstract
Wick rotation in area tensor Regge calculus is considered. The heuristical expectation is confirmed that the Lorentzian quantum measure on a spacelike area should coincide with the Euclidean measure at the same argument. The consequence is validity of probabilistic interpretation of the Lorentzian measure as well (on the real, i.e. spacelike areas).
The problem of the relation between the Lorentzian and Euclidean versions of a field theory is especially nontrivial in gravity, namely, in general relativity theory where the Wick rotation means not only formal rescaling time and timelike vector components by $`\sqrt{1}`$ but also real change of topology . The discrete counterpart of general relativity, Regge calculus (RC), provides a simplification connected with coordinatelessness of this formulation of general relativity: there is no time coordinate to be scaled by $`\sqrt{1}`$; instead, some edge lengths should be made complex (the notion of โtimeโ can arise at the intermediate steps simply as a parameter which labels the 3D layers by โฆ, 1, 2, 3, โฆ) .
In the present paper we consider Wick rotation in the framework of the approach to quantisation of RC developed in a number of our papers. To begin with, we shortly discuss that quantization problem. Up to now the canonical (Dirac) quantization prescription is considered as a fundamental principle not contradicting to experiment, so let us adhere to it. However, there is a problem to implement it in the discrete theory such as RC because of the lack of a regular continuous coordinate playing the role of time. Although quantum theory still can be formulated with the help of the functional integral in this case, the absence of a strict framework leaves the quantum measure practically arbitrary. In fact, various measures are used in applications (e.g., in phase analysis of simplicial quantum gravity ).
Despite of the lack of a continuous coordinate in the completely discrete RC, such the coordinate is present in the limiting so-called (3+1) RC when one of the coordinates is made continuous by shrinking sizes of all the simplices along this coordinate to those infinitely close to zero. Therefore one can try to develop Hamiltonian formalism and canonical quantization and represent the result in the form of the Feynman path integral measure. Next it is natural to ask whether a measure in the original completely discrete RC exists such that one could define the limit of this measure when one of the coordinates is made continuous and the limiting measure would coincide with the above Feynman path (canonical quantization) measure with this continuous coordinate playing the role of time. Equivalence of the different coordinates means that this situation should take place irrespectively of what coordinate is made continuous.
A difficulty with RC in the continuous time limit is that the description of the infinitely flattened in some direction simplex purely in terms of the lengths is singular. The idea is to use description in terms of the variables of the types of both lengths and angles. This might be achieved in the Regge analog of the Hilbert-Palatini form of the Einstein action. The discrete analogs of the tetrad and connection, link vectors and finite rotation matrices, were first considered by Bander . We have found exact representations of the Regge action in terms of link vectors and finite rotation matrices as independent variables . This representation results in the exact Regge action if rotation matrices are excluded via equations of motion, that is, on classical level. Rotation matrices are just the desired angle type variables which allow us to formulate continuous time (3+1) RC in a nonsingular way. After that the above strategy can be implemented. In we write out canonical form of RC.
Next we can try to solve the problem of finding measure in the full discrete RC which has the desired continuous time limit corresponding to the canonical quantization irrespectively of what coordinate is taken as a time and made continuous. Although this last condition is rather restrictive, the problem has solution in 3 dimensions . In 4 dimensions solution can be found for a certain version of the so-called โarea RCโ where areas are treated as independent variables . Since the number of areas is larger than the number of lengths, this means that the lengths of the same edge defined in the different simplices are in general different, i.e. ambiguous. The configuration superspace of the area RC contains the hypersurface corresponding to the ordinary RC; at the same time it is exactly soluble just as 3D model. Appropriate version in our case is โarea tensor RCโ with independent area tensors (i.e., in particular, in general there are no link vectors corresponding to them). Just the corresponding superspace (extended as compared to that of ordinary RC) is that space on which the measure under above restrictive conditions can be found .
Remarkable feature simplifying solution of the above problem of finding full discrete measure for the 3D RC and for the 4D area tensor RC in the tetrad-connection variables is commutativity of the constraints (arising in the Hamiltonian formalism, i.e. in the continuous time limit). These constraints as well as canonical quantization itself are analogous to their completely continuum counterparts. The commuting constraints for the 3D discrete gravity were first deduced by Waelbroeck for general system (not aโpriori restricted to be RC). Analogous first class system of constraints arises in the area tensor RC in the above mentioned derivation of the measure .
Finally, we need to reduce the quantum measure in the extended superspace to the RC hypersurface. The idea is to consider area tensor RC system and ordinary RC system as particular case of the simplicial complex with discontinuous metrics. The point is that the piecewise flat manifold possesses metric whose normal component undergoes discontinuity when passing across any 3D face, but the tangential components remain unchanged. Now we go further and consider system where tangential components of metric are also discontinuous. It is the system with independent simplices which do not necessarily fit each other on their common faces. In the superspace of all the simplicial discontinuous metrics RC corresponds to the hypersurface singled out by conditions of the tangential metric continuity on the faces. The problem is to reduce the above constructed quantum measure in โarea tensor RCโ to this hypersurface. For that some $`\delta `$-function-like factor is introduced in the measure which fixes equality of tangential metric across any face. This factor is found in our paper by using the principle of โminimum of the lattice artefactsโ. Namely, we require that the factor should not depend on the form and size of any face across which metrics are compared, only on the hyperplane in which the given face is placed. We show that such the factor preserving equivalence of the different simplices exists and is unique. Consequences of the viewpoint on area RC (now not tensor one) as a system with discontinuous metrics were also discussed in .
It is quite natural that at the intermediate stage of our construction in area tensor RC the measure is factorizable (on certain conditions) since area tensors are here independent just as edge vectors in the (exactly soluble) 3D model, the typical factor as applied to averaging a function $`f(\pi )`$ of a given area tensor $`\pi `$ being of the form (in the case of the Euclidean spacetime signature)
$`<f(\pi )>`$ $`=`$ $`{\displaystyle f(i\pi )\mathrm{d}^6\pi e^{i\pi R}๐R}.`$ (1)
Here rotation of the integration contours used to define integral is performed via substitution of the integration (dummy) variables $`\pi `$ $``$ $`i\pi `$. The integration variable $`R`$ is SO(4) matrix, $`AB`$ $``$ $`A^{ab}B_{ab}/2`$. The formula (1) looks reasonable from path integral and symmetry viewpoint: $`\pi R`$ reminds Regge action in the sense that it (for independent tensors $`\pi ^{ab}`$) gives the same constraints $`R`$ = 0 arising in the canonical formalism; $`\mathrm{d}^6\pi `$ and $`๐R`$ are the invariant Lebesgue and Haar measures. Upon splitting antisymmetric matrices into self- and antiselfdual ones like
$$\pi _{ab}\frac{1}{2}^+\pi _k^+\mathrm{\Sigma }_{ab}^k+\frac{1}{2}^{}\pi _k^{}\mathrm{\Sigma }_{ab}^k$$
(2)
(the basis of self- and antiselfdual matrices $`i^\pm \mathrm{\Sigma }_{ab}^k`$ obeys the Pauli matrix algebra) the formula (1) reads
$`<f(\pi )>`$ $`=`$ $`{\displaystyle f(\pi )\frac{\nu (|^+๐
|)}{|^+๐
|^2}\frac{\nu (|^{}๐
|)}{|^{}๐
|^2}\frac{\mathrm{d}^3{}_{}{}^{+}๐
}{4\pi }\frac{\mathrm{d}^3{}_{}{}^{}๐
}{4\pi }},`$
$`\nu (l)={\displaystyle \frac{l}{\pi }}{\displaystyle \underset{0}{\overset{\pi }{}}}{\displaystyle \frac{\mathrm{d}\phi }{\mathrm{sin}^2\phi }}e^{l/\mathrm{sin}\phi }.`$
This is positive measure which gives finite (due to exponential cut-off) nonzero expectation values of positive powers of area (and even of negative powers $`|\pi |^k`$ at $`k`$ $`>`$ $`2`$, $`|\pi |`$ $``$ $`(\pi \pi )^{1/2}`$) which thus are certain numbers in the Plank units. Further, โglueingโ together neighbouring 4-simplices on their common faces we can go over from the considered area tensor RC to the genuine RC and make qualitative scaling estimate showing that the length expectation values are finite nonzero . The properties of the measure such as positivity and finite nonzero area (length) expectation values remain true.
Also note the following. If the matrices $`R`$ in the formula (1) were substituted by their continuum analogs, antisymmetric matrices (elements of the Lie group so(4)), we would have $`\delta `$-function-like measure and zero expectation values for nonconstant monomials in $`\pi ^{ab}`$. Thus, nonzero area expectation values arise due to nonlinearity of the finite rotations, i. e. eventually are connected with the discreteness of the theory.
Nonvanishing and finiteness of the area (length) expectation values may speak well for the internal consistency on dynamics level of using the discrete type variables to describe gravity. If these expectations were equal to zero, this would mean that the functional integral is saturated by smooth manifolds which are just the limiting case of the piecewise flat manifolds if edge lengths tend to zero; that is, we would return to the continuum theory.
More detailed and still concise discussion of the above points is given in our work .
When passing from Euclidean to the Lorentzian domain the Euclidean area becomes spacelike (real) Lorentzian area, and one would expect that positivity of the measure in the Euclidean domain would correspond to positivity in the Lorentzian domain on real areas. The technical complication is that now self- and antiselfdual parts of area tensors and of the rotation matrices are, first, complex, second, are related to each other by complex conjugation. Therefore a more careful analysis is required.
Note that integration over the invariant measure is representable as
$$f(R)๐R=f(m)\delta ^{10}(\eta ^{ab}m_{af}m_{bg}\eta _{fg})\mathrm{d}^{16}m,$$
(4)
where $`\eta _{ab}`$ = $`\mathrm{diag}(\pm 1,1,1,1)`$ in the Euclidean/Lorentzian case. Invariance of this measure is evident<sup>1</sup><sup>1</sup>1In fact, originally in our three-dimensional analysis the invariant measure just arises in the form analogous to (4) upon introducing, as in , the variables $`P_{ab}`$ = $`l^c\mathrm{\Omega }_a^fฯต_{cfb}`$ and $`\mathrm{\Omega }^{ab}`$ for each edge which are canonically conjugate in the usual sense. Then we get $`\delta `$-functions in the functional integral which take into account the II class constraints on $`P`$, $`\mathrm{\Omega }`$, namely, $`\delta ^6(\overline{\mathrm{\Omega }}\mathrm{\Omega }1)\delta ^6(\overline{\mathrm{\Omega }}P+\overline{P}\mathrm{\Omega })`$. The invariant measure $`D๐๐\mathrm{\Omega }`$ follows upon integrating these $`\delta `$-functions. Representing $`\delta `$-function in (4) as the Fourier transform over a 10-component variable symmetrical matrix $`\lambda ^{ab}`$ we can easily perform Gaussian integration over $`m_{ab}`$. As a result, we get
$`{\displaystyle e^{i\pi R}๐R}`$ $`=`$ $`{\displaystyle \mathrm{exp}\left[\frac{i}{2}\pi ^{ab}m_{ab}+i\lambda ^{ab}(\eta ^{fg}m_{af}m_{bg}\eta _{ab})\right]\mathrm{d}^{16}m_{ab}\mathrm{d}^{10}\lambda ^{ab}}`$ (5)
$`=`$ $`{\displaystyle \frac{\mathrm{d}^{10}\lambda ^{ab}}{(det\lambda ^{ab})^2}\mathrm{exp}\left[\frac{i}{16}\pi ^{fa}\pi ^{gb}\eta _{fg}(\lambda ^1)_{ab}i\eta _{ab}\lambda ^{ab}\right]}.`$
This is in some sense a matrix analog of a Bessel function. Let us make the formal substitution of the variables $`\lambda ^{ab}`$ = $`\underset{f,g}{}(\eta ^{1/2})^{af}(\eta ^{1/2})^{bg}\mathrm{\Lambda }^{fg}`$ where for definiteness $`(\eta ^{\pm 1/2})^{ab}`$ = $`\mathrm{diag}(\pm i,1,1,1)`$ for the Lorentzian case. The result reads
$`{\displaystyle e^{i\pi R}๐R}=`$ (6)
$`{\displaystyle \frac{\mathrm{d}^{10}\mathrm{\Lambda }^{ab}}{(det\mathrm{\Lambda }^{ab})^2}\mathrm{exp}\left[\frac{i}{16}\underset{c,d}{}\pi ^{fc}\pi ^{gd}(\eta ^{1/2})^{ca}(\eta ^{1/2})^{db}\eta _{fg}(\mathrm{\Lambda }^1)_{ab}i\delta _{ab}\mathrm{\Lambda }^{ab}\right]}.`$
Real $`\lambda ^{ab}`$ imply purely imaginary $`\mathrm{\Lambda }^{0i}`$ ($`i`$ = 1, 2, 3). Let us make rotations in the complex plane to the purely real $`\mathrm{\Lambda }^{0i}`$. The possibility to proceed in such the way is provided by the sufficiently rapidly decreasing expression under the integral sign at $`|\mathrm{\Lambda }^{0i}|`$ $``$ $`\mathrm{}`$ both separately for each $`i`$ and simultaneously. Note that the dangerous in this respect exponential factor $`\mathrm{exp}(i\delta _{ab}\mathrm{\Lambda }^{ab})`$ does not contain $`\mathrm{\Lambda }^{0i}`$ while $`det\mathrm{\Lambda }^{ab}`$ is bilinear in each $`\mathrm{\Lambda }^{0i}`$ at almost all the values of the rest of variables. As a result, integrations over the circles of a large radius $`|\mathrm{\Lambda }^{0i}|`$ = $`const`$ $``$ $`\mathrm{}`$ result in zero at almost all the values of the rest of variables $`\mathrm{\Lambda }^{ab}`$. Therefore $`\mathrm{\Lambda }^{ab}`$ in (6) can be considered real.
An advantage of the formula (6) is that the same expression can be found also for the Euclidean case, the transition between these two consisting in substitution of Lorentzian $`\underset{c,d}{}\pi ^{fc}\pi ^{gd}(\eta ^{1/2})^{ca}(\eta ^{1/2})^{db}\eta _{fg}`$ by the Euclidean $`\pi ^{fa}\pi _f^b`$ and vice versa (remind that the exponent is monotonic at genuine real area tensor $`\pi `$ in the latter case, not oscillating as at earlier used analytical continuation $`\pi `$ $``$ $`i\pi `$). One of the conditions imposed when passing from the area tensor Regge calculus (where $`\pi `$ are independent tensors) to the ordinary Regge calculus (that is to say, on physical surface) is $`\pi \pi `$ $``$ $`\pi ^{ab}`$ $`\pi ^{cd}`$ $`ฯต_{abcd}/4`$ = 0, therefore $`\pi ^{ab}\pi _{ab}`$ is the only scalar which could be constructed of $`\pi ^{ab}`$. Thereby this is what should be replaced by $`\pi ^{ab}\pi _{ab}`$ when passing from the Euclidean metric to the Lorentzian one. Thus we find the same exponential cut-off factor in the Lorentzian measure as $`\nu (|๐
|)`$ in the Euclidean one,
$$e^{i\pi R}๐R\frac{\nu (|๐
|)^2}{|๐
|^4},\nu (l)=\frac{l}{\pi }\underset{0}{\overset{\pi }{}}\frac{\mathrm{d}\phi }{\mathrm{sin}^2\phi }e^{l/\mathrm{sin}\phi },$$
(7)
but with the argument
$$|๐
|=\sqrt{\underset{i}{}(\pi ^{0i})^2\underset{i<j}{}(\pi ^{ij})^2}.$$
(8)
If $`\pi ^{ab}`$ is timelike, i.e. the components $`\pi ^{0i}`$ dominate, we find still positive measure. According to the definition of $`\pi ^{ab}`$ as dual tensor, this corresponds to the spacelike (real) area. If the area is imaginary (timelike), the measure oscillates and is even complex and does not admit the usual probabilistic interpretation. Physically, this does not seem surprising since not all the areas (or lengths) must fluctuate, some of them should be fixed by hand to specify triangulation (in analogy with gauge lapse-shift vectors in the ordinary continuum general relativity). At the same time, the set of functions of $`\pi ^{ab}\pi _{ab}`$ on which the measure (as functional) is positive seems to be even larger than in the Euclidean case. This set includes, first, positive functions with support on the (negative) semiaxis of $`\pi ^{ab}\pi _{ab}`$, such as $`\pi ^{ab}\pi _{ab}\theta (\pi ^{ab}\pi _{ab})`$ ($`\theta `$ is Heaviside function) as it follows from the considered explicit positivity of the measure on this semiaxis. Second, rotation of contours in the complex plane of $`\pi ^{ab}`$ immediately in the Lorentzian expectation value $`<f(\pi ^{ab}\pi _{ab})>`$ reduces this value to the Euclidean $`<f(\pi ^{ab}\pi _{ab})>`$; therefore the measure is positive also on the analytical functions positive on the negative semiaxis of $`\pi ^{ab}\pi _{ab}`$, e. g. $`\pi ^{ab}\pi _{ab}`$.
Thus, the quantum measure is positive on the area tensors $`\pi _{\sigma ^2}^{ab}`$ spacelike w.r.t. the local frame indices. On the other hand, the notation for the triangle $`\sigma ^2`$ in the index of $`\pi _{\sigma ^2}^{ab}`$ can be considered as an analog of the world index, or, more accurately, a pair of such indices. An analog of the spacelike components of area tensor w.r.t. the world index are $`\pi _{\sigma ^2}^{ab}`$ for the leaf/diagonal triangles $`\sigma ^2`$. Now we observe consistency between the properties of the quantum measure on area tensor w.r.t. the local indices and w.r.t. the world indices of this tensor. Indeed, earlier we have found that integrations over $`\mathrm{d}^6\pi _{\sigma ^2}`$ are present in the measure not for all $`\sigma ^2`$ (otherwise we could not, e.g., normalize the measure: integrations of $`\mathrm{exp}(i\pi _{\sigma ^2}R_{\sigma ^2})`$ would give the product of $`\delta `$-functions of (the antisymmetric part of) $`R_{\sigma ^2}`$ which are not independent due to the Bianchi identities , now for the matrices $`R_{\sigma ^2}(\mathrm{\Omega })`$). Some regular choice is that $`\mathrm{d}^6\pi _{\sigma ^2}`$ are present in the measure only for the leaf and diagonal triangles $`\sigma ^2`$ (or, rather, this can serve a definition of what do we call the leaf/diagonal triangles). Correspondingly, only $`\pi _{\sigma ^2}^{ab}`$ for the leaf and diagonal triangles do fluctuate. It is important that the quantum measure be positive on the sufficiently large subspace of possible fluctuations of those objects which do fluctuate. This just takes place: the areas fluctuate which are spacelike in the world index, that is, from 3D subset, and the quantum measure is positive also on the spacelike areas, now w.r.t. the local frame index, that is, again on 3D subspace (or, taking into account also the Euclidean case, not smaller than 3D subspace). It looks surprising that consistent are the things which seem to have perfectly different origins. Namely, properties of the quantum measure w.r.t. the world index of area tensor are consequence of the structure of the Bianchi identities on the curvature matrices $`R_{\sigma ^2}`$, whereas considered in the present paper properties of the quantum measure w.r.t. the local frame index refer to the dependence of the measure on a single area. The analogous situation had been encountered also in the three-dimensional model .
The present work was supported in part by the Russian Foundation for Basic Research through Grant No. 05-02-16627-a.
|
warning/0506/cond-mat0506086.html
|
ar5iv
|
text
|
# Current-driven switching of magnetisation- theory and experiment
## I Introduction
Recently there has been a lot of interest in magnetic nanopillars of 10-100 nm in diameter. The pillar is a metallic layered structure with two ferromagnetic layers, usually of cobalt, separated by a non-magnetic spacer layer, normally of copper. Non-magnetic leads are attached to the magnetic layers so that an electric current may be passed through the structure. In the simplest case the pillar may exist in two states, with the magnetisation of the two magnetic layers parallel or anti-parallel. The state of a pillar can be read by measuring its resistance, this being smaller in the parallel state than in the anti-parallel one. This dependence of the resistance on magnetic configuration is the giant magnetoresistance (GMR) effect 1 . A dense array of these nanopillars could form a magnetic memory for a computer. Normally one of the magnetic layers in a pillar is relatively thick and its magnetisation direction is fixed. In order to write into the memory the magnetisation direction of the second thinner layer must be switched. This might be achieved by a local magnetic field of suitable strength and methods have been proposed 2 for providing such a local field by currents in a criss-cross array of conducting strips. However an alternative, and potentially more efficient method, proposed by Slonczewski 3 makes use of a current passing up the pillar itself. Slonczewskiโs effect relies on โspin transferโ and not on the magnetic field produced by the current which in the nanopillar geometry is ineffective. The idea of spin-transfer is as follows. In a ferromagnet there are more electrons of one spin orientation than of the other so that current passing through the thick magnetic layer (the polarising magnet) becomes spin-polarised. In general its state of spin-polarisation changes as it passes through the second (switching) magnet so that spin angular momentum is transferred to the switching magnet. This transfer of spin angular momentum is called spin-transfer torque and, if the current exceeds a critical value, it may be sufficient to switch the direction of magnetisation of the switching magnet. This is called current-induced switching.
In the next section we show how to calculate the spin-transfer torque for a simple model.
## II Spin-transfer torque in a simple model
For simplicity we consider a structure of the type shown in Fig. 1, where p and m are unit vectors in the direction of the magnetisations. This models the layered structure of the pillars used in experiments but the atomic planes shown are considered to be unbounded instead of having the finite cross-section of the pillar. This means that there is translational symmetry in the $`x`$ and $`z`$ directions. The structure consists of a thick (semi-infinite) left magnetic layer (polarising magnet), a non-magnetic metallic spacer layer, a thin second magnet (switching magnet) and a semi-infinite non-magnetic lead. In the simplest model we assume the atoms form a simple cubic lattice, with lattice constant $`a`$, and we adopt a one-band tight-binding model with hopping Hamiltonian
$$H_0=t\underset{๐ค_{}\sigma }{}\underset{n}{}c_{k_{}n\sigma }^{}c_{k_{}n1\sigma }+\mathrm{h}.\mathrm{c}..$$
(1)
Here $`c_{k_{}n\sigma }^{}`$ creates an electron on plane $`n`$ with two-dimensional wave-vector $`๐ค_{}`$ and spin $`\sigma `$, and $`t`$ is the nearest-neighbour hopping integral.
In the tight-binding description the operator for spin angular momentum current between planes $`n1`$ and $`n`$, which we require to calculate spin-transfer torque, is given by
$$๐ฃ_{n1}=\frac{\mathrm{i}t}{2}\underset{๐ค_{}}{}(c_{k_{},n,\sigma }^{},c_{k_{},n,}^{})๐(c_{k_{},n1,},c_{k_{},n1,})^{}+\mathrm{h}.\mathrm{c}..$$
(2)
Here $`๐=(\sigma _x,\sigma _y,\sigma _z)`$ where the components are Pauli matrices. Eq. (2) yields the charge current $`j_{n1}^\mathrm{c}`$ if $`\frac{1}{2}๐`$ is replaced by a unit matrix multiplied by the number $`e/\mathrm{}`$, where $`e`$ is the electronic charge (negative). All currents flow in the $`y`$ direction, perpendicular to the layers, and the components of the vector $`๐ฃ`$ correspond to transport of $`x`$, $`y`$ and $`z`$ components of spin. The justification of Eq. (2) for $`๐ฃ_{n1}`$ relies on an equation of continuity, as pointed out in Section IV.
To define the present model completely we must supplement the hopping Hamiltonian $`H_0`$ by specifying the on-site potentials in the various layers. For simplicity we assume the on-site potential for both spins in non-magnetic layers, and for majority spin in ferromagnetic layers, is zero. We assume an infinite exchange splitting in the ferromagnets so that the minority spin potential in these layers is infinite. Thus minority spin electrons are completely excluded from the ferromagnets. Clearly the definition of majority and minority spin relate to spin quantisation in the direction of the local magnetisation. We take $`\alpha =0`$, so that the magnetisation of the switching magnet is in the z direction and take $`\theta =\psi `$, where $`\psi `$ is the angle between the magnetisations.
To describe spin transport in the structure we adopt the generalised Landauer approach of Waintal et. al. 4 . Thus the structure is placed between two reservoirs, one on the left and one on the right, with electron distributions characterised by Fermi functions $`f(\omega \mu _\mathrm{L})`$, $`f(\omega \mu _\mathrm{R})`$ respectively. The system is then subject to a bias voltage $`V_\mathrm{b}`$ given by $`eV_\mathrm{b}=\mu _\mathrm{L}\mu _\mathrm{R}`$, the difference between the chemical potentials. We discuss the ballistic limit where scattering occurs only at interfaces, the effect of impurities being negligible. We label the atomic planes so that $`n=0`$ corresponds to the last atomic plane of the polarising magnet. the planes of the spacer layer correspond to $`n=1,2N`$ and $`n=N+1`$ is the first plane of the switching magnet.
Consider first an electron incident from the left with wave-function $`|k,maj`$, where $`k>0`$, which corresponds to a Bloch wave $`|k=_n\mathrm{e}^{\mathrm{i}kna}|๐ค_{}n`$ with majority spin in the polarising magnet. In this notation the label $`๐ค_{}`$ is suppressed. The particle is partially reflected by the structure and finally emerges as a partially transmitted wave in the lead, with spin $``$ corresponding to majority spin in the switching magnet. Thus the wave-function is of the form
$$|P_k=|k,maj+B|k,maj$$
(3)
in the polariser and
$$|L_k=F|k,$$
(4)
in the lead. A majority spin in either ferromagnet enters or leaves the spacer without scattering, since in our simple model there is no potential step. Also the minority spin wave-function entering a ferromagnet is zero. The spacer wave-function may therefore be written in two ways:
$$|S_k=F|k,+E\left(\mathrm{e}^{\mathrm{i}k(N+1)a}|k,\mathrm{e}^{\mathrm{i}k(N+1)a}|k,\right)$$
(5)
or
$`|S_k`$ $`=`$ $`|k,maj+B|k,maj+D\left(|k,min|k,min\right)`$
$`=`$ $`\mathrm{cos}\left(\psi /2\right)|k,+\mathrm{sin}\left(\psi /2\right)|k,+B\left[\mathrm{cos}\left(\psi /2\right)|k,+\mathrm{sin}\left(\psi /2\right)|k,\right]`$
$`+D\left[\mathrm{sin}\left(\psi /2\right)|k,+\mathrm{cos}\left(\psi /2\right)|k,+\mathrm{sin}\left(\psi /2\right)|k,\mathrm{cos}\left(\psi /2\right)|k,\right].`$
On equating coefficients of $`|k,`$, $`|k,`$, $`|k,`$, $`|k,`$ in expressions (5) and (II) we have four equations which may be solved for $`B`$, $`D`$, $`E`$, $`F`$. In particular the transmission coefficient $`T`$ is given by
$$T=\left|F\right|^2=\frac{4\mathrm{cos}^2(\psi /2)\mathrm{sin}^2k(N+1)a}{\mathrm{sin}^4(\psi /2)+4\mathrm{cos}^2(\psi /2)\mathrm{sin}^2k(N+1)a}.$$
(7)
Similarly an electron incident from the right with wave-function $`|k`$ in the lead is partially reflected and finally emerges as a partially transmitted wave $`F^{}|k,maj`$ in the polarising magnet. It is found that $`F^{}=F`$ so that the transmission coefficient is the same for particles from left or right.
The spin angular momentum current in a particular layer, which we shall denote by S although it need not be the spacer layer, is the sum of currents carried by left and right moving electrons. Thus we have a Landauer-type formula 5
$$๐ฃ_\mathrm{s}=\frac{a}{2\pi }\underset{๐ค_{}}{}\left\{_{k>0}dk\left[S_k|๐ฃ_{n1}|S_kf(\omega \mu _L)+S_k|๐ฃ_{n1}|S_kf(\omega \mu _R)\right]\right\}$$
(8)
where $`|S_k`$, $`|S_k`$ are wave-functions in the layer considered corresponding to electrons incident from left and right, respectively. Here $`\omega `$, the energy of the Bloch wave $`k`$, is given by the tight-binding formula
$$\omega =u_๐ค_{}+2t\mathrm{cos}ka$$
(9)
where $`u_๐ค_{}=2t(\mathrm{cos}k_xa+\mathrm{cos}k_za)`$. We take $`t<0`$ so that positive $`k`$ corresponds to positive velocity $`\mathrm{}^1\omega /k`$ as we have assumed. The current $`๐ฃ_\mathrm{s}`$ in layer $`S`$ calculated by Eq. (8) does not depend on the particular planes $`n1`$, $`n`$ between which it is calculated. On changing the integration variable in Eq. (8) we find
$$๐ฃ_\mathrm{s}=\frac{1}{2\pi }\underset{๐ค_{}}{}d\omega \left[๐_+f(\omega \mu _\mathrm{L})+๐_{}f(\omega \mu _\mathrm{R})\right]$$
(10)
where
$$๐_\pm =\frac{S_{\pm k}|๐ฃ_{n1}|S_{\pm k}}{2t\mathrm{sin}ka}.$$
(11)
Here $`k=k(\omega ,๐ค_{})`$ is the positive root of Eq. (9). Eq. (10) may be written as
$$๐ฃ_\mathrm{s}=\frac{1}{4\pi }\underset{๐ค_{}}{}d\omega \left\{\left(๐_++๐_{}\right)\left[f(\omega \mu _\mathrm{L})+f(\omega \mu _\mathrm{R})\right]+\left(๐_+๐_{}\right)\left[f(\omega \mu _\mathrm{L})f(\omega \mu _\mathrm{R})\right]\right\}.$$
(12)
Before discussing this spin current we briefly consider the charge current $`j^\mathrm{c}`$, and we denote the analogues of $`๐_\pm `$ by $`J_\pm ^\mathrm{c}`$. Since the charge current is conserved throughout the structure $`J_+^\mathrm{c}`$ and $`J_{}^\mathrm{c}`$ can be calculated in different ways, e.g. in the lead for $`J_+^\mathrm{c}`$ and in the polariser for $`J_{}^\mathrm{c}`$. Since $`T=\left|F\right|^2=\left|F^{}\right|^2`$ we find $`J_+^\mathrm{c}+J_{}^\mathrm{c}=0`$ and for small bias $`eV_b=\mu _\mathrm{L}\mu _\mathrm{R}`$ the charge current is given by
$$j^\mathrm{c}=\frac{2e^2V_\mathrm{b}}{h}\underset{๐ค_{}}{}T$$
(13)
where the transmission coefficient $`T`$ is given by Eq. (7) with $`k=k(\mu ,๐ค_{})`$, $`\mu `$ being the common chemical potential as $`V_\mathrm{B}0`$. This is the well-known Landauer formula 5 .
The spin transfer torque on the switching magnet is given by
$$๐^{\mathrm{s}\mathrm{t}}=๐ฃ_{\mathrm{spacer}}๐ฃ_{\mathrm{lead}},$$
(14)
where $`๐ฃ_{\mathrm{spacer}}`$ and $`๐ฃ_{\mathrm{lead}}`$ are spin currents in the spacer and lead respectively. For zero bias ($`\mu _\mathrm{L}=\mu _\mathrm{R}`$) there is clearly no charge current in the structure and straight-forward calculation shows that all components of spin current in the spacer and the lead vanish, except for a non-zero $`y`$-spin current in the spacer. There is therefore a non-zero $`y`$ component of spin-transfer torque acting on the switching magnet for zero bias, and its dependence on the angle $`\psi `$ between the magnetisations is found to be approximately $`\mathrm{sin}\psi `$. This torque is due to exchange coupling, analogous to an RKKY coupling, between the two magnetic layers. This coupling oscillates as a function of spacer thickness and tends to zero as the thickness tends to infinity. For finite bias $`V_\mathrm{B}`$ the second term in the integrand of Eq. (12) comes into play. In general this leads to finite $`x`$ and $`y`$ components of $`๐^{\mathrm{s}\mathrm{t}}`$ proportional to $`V_\mathrm{b}`$ (for small $`V_\mathrm{b}`$) whereas $`T_z^{\mathrm{s}\mathrm{t}}=0`$. However for the special model considered here with infinite exchange splitting in both ferromagnets it turns out that $`T_y^{\mathrm{s}\mathrm{t}}=0`$. For this model the only non-zero component of $`๐^{\mathrm{s}\mathrm{t}}`$ proportional to $`V_\mathrm{b}`$ is found to be
$$T_x^{\mathrm{s}\mathrm{t}}=\frac{\mathrm{}j^\mathrm{c}}{2|e|}\mathrm{tan}\frac{\psi }{2}$$
(15)
where $`j^\mathrm{c}`$ is the charge current given by Eq. (13).
Slonczewski 3 originally obtained this result for the analogous parabolic band model. From Eqs. (15), (13) and (7) it follows that $`T_x^{\mathrm{s}\mathrm{t}}`$ contains an important factor $`\mathrm{sin}\psi `$ although this does not represent the whole angle dependence. Clearly, from Eq. (15), the torque proportional to bias remains finite for arbitrarily large spacer thickness, in the ballistic limit. For this model, with infinite exchange splitting, the torque is independent of the thickness of the switching magnet.
From the results of this simple model we can infer a general form of the spin-transfer torque $`๐^{\mathrm{s}\mathrm{t}}`$ which is independent of the choice of coordinate axes. Thus we write
$$๐^{\mathrm{s}\mathrm{t}}=๐_{}+๐_{}$$
(16)
where
$`๐_{}`$ $`=`$ $`\left(g^{\mathrm{ex}}+g_{}eV_\mathrm{b}\right)(๐ฆ\times ๐ฉ)`$
$`๐_{}`$ $`=`$ $`g_{}eV_\mathrm{b}๐ฆ\times (๐ฉ\times ๐ฆ).`$ (17)
With the choice of axes in Fig. 1 $`๐_{}`$ corresponds to the $`x`$ component of torque, that is the component parallel to the plane containing the magnetisation directions $`๐ฆ`$ and $`๐ฉ`$. Similarly $`๐_{}`$ corresponds to the $`y`$ component of torque, this being perpendicular to the plane of $`๐ฆ`$ and $`๐ฉ`$. The modulus of both the vectors $`๐ฆ\times ๐ฉ`$ and $`๐ฆ\times (๐ฉ\times ๐ฆ)`$ is $`\mathrm{sin}\psi `$, so that the factors $`g^{\mathrm{ex}}`$, $`g_{}`$ and $`g_{}`$ are functions of $`\psi `$ which contain deviations from the simple $`\mathrm{sin}\psi `$ behaviour. The bias-independent term $`g^{\mathrm{ex}}`$ corresponds to the interlayer exchange coupling, as discussed above, and henceforth we assume that the spacer is thick enough for this term to be negligible. Sometimes the $`\mathrm{sin}\psi `$ factor accounts for most of the angular dependence of $`T_{}`$ and $`T_{}`$ so that $`g_{}`$ and $`g_{}`$ may be regarded as constant parameters for the given structure. In the next section we use Eqs. (II) for the spin-transfer torque in a phenomenological theory of current-induced switching of magnetisation. This phenomenological treatment enables us to understand most of the available experimental data. It is more usual in experimental works to relate spin-transfer torque to current rather than bias. However in theoretical work, based on the Landauer or Keldysh approach, bias is more natural. In practice the resistance of the system considered is rather constant (the GMR ratio is only a few percent) so that bias and current are in a constant ratio.
## III Phenomenological treatment of current-induced switching of magnetisation
In this section we explore the consequences of the spin-transfer torque acting on a switching magnet using a phenomenological Landau Lifshitz equation with Gilbert damping (LLG equation). This is essentially a generalisation of the approach used originally by Slonczewski 3 and Sun sun . We assume that there is a polarising magnet whose magnetisation is pinned in the $`xz`$-plane in the direction of a unit vector $`๐ฉ`$, which is at general fixed angle $`\theta `$ to the $`z`$-axis as shown in Fig. 1.
The pinning of the magnetisation of the polarising magnet can be due to its large coercivity (thick magnet) or a strong uniaxial anisotropy. The role of the polarising magnet is to produce a stream of spin-polarised electrons, i.e. spin current, that is going to exert a torque on the magnetisation of the switching magnet whose magnetisation lies in the general direction of a unit vector $`๐ฆ`$. The orientation of the vector $`๐ฆ`$ is defined by the polar angles $`\alpha `$, $`\varphi `$ shown in Fig. 1. There is a non-magnetic metallic layer inserted between the two magnets whose role is merely to separate magnetically the two magnetic layers and allow a strong charge current to pass. The total thickness of the whole trilayer sandwiched between two non-magnetic leads must be smaller than the spin diffusion length $`l_{\mathrm{sf}}`$ so that there are no spin flips due to impurities or spin-orbit coupling. A typical junction in which current-induced switching is studied experimentally albert is shown schematically in Fig. 2. The thickness of the polarising magnet is 40nm, that of the switching magnet 2.5nm and the non-magnetic spacer is 6nm thick. The materials for the two magnets and the spacer are cobalt and copper, respectively, which are those most commonly used. The junction cross section is oval-shaped with dimensions 60nm$`\times `$130nm. A small diameter is necessary so that the torque due to the Oersted field generated by a charge current of $`10^7`$-$`10^8`$ A/cm<sup>2</sup>, required for current-induced switching, is much smaller than the spin-transfer torque we are interested in.
The aim of most experiments is to determine the orientation of the switching magnet moment as a function of the current (applied bias) in the junction. Sudden jumps of the magnetisation direction, i.e. current-induced switching, are of particular interest. The orientation of the switching magnet moment $`๐ฆ`$ relative to that of the polarising magnet $`๐ฉ`$, which is fixed, is determined by measuring the resistance of the junction. Because of the GMR effect, the resistance of the junction is higher when the magnetisations of the two magnets are anti-parallel than when they are parallel, In other words, what is observed are hysteresis loops of resistance versus current. A typical experimental hysteresis loop of this type 16 is reproduced in Fig. 3.
It can be seen from Fig. 3 that, for any given current, the switching magnet moment is stationary (the junction resistance has a well defined value), i.e. the system is in a steady state. This holds everywhere on the hysteresis loop except for the two discontinuities where current-induced switching occurs. As indicated by the arrows jumps from the parallel (P) to anti-parallel (AP) configurations of the magnetisation, and from AP to P configurations, occur at different currents. It follows that in order to interpret experiments which exhibit such hysteresis behaviour, the first task of the theory is to determine from the LLG equation all the possible states and then investigate their dynamical stability. At the point of instability the system seeks out a new steady state, i.e. a discontinuous transition to a new steady state with the switched magnetisation occurs. We have tacitly assumed that there is always a steady state available for the system to jump to. There is now experimental evidence that this is not always the case. In the absence of any stable steady state the switching magnet moment remains permanently in the time-dependent state. This interesting case is implicit in the phenomenological LLG treatment and we shall discuss it in detail later.
In describing the switching magnet by a unique unit vector $`๐ฆ`$, we assume that it remains uniformly magnetised during the switching process. This is only strictly true when the exchange stiffness of the switching magnet is infinitely large. It is generally a good approximation as long as the switching magnet is small enough to remain single domain, so that the switching occurs purely by rotation of the magnetisation as in the Stoner-Wohlfarth theory 17 of field switching. This seems to be the case in many experiments albert ; 16 ; 18 ; 19
Before we can apply the LLG equation to study the time evolution of the unit vector $`๐ฆ`$ in the direction of the magnetisation of the switching magnet, we need to determine all the contributions to the torque acting on the switching magnet. Firstly, there is the spin-transfer torque $`๐^{\mathrm{s}\mathrm{t}}`$ which we discussed in Section II. Secondly, there is a torque due to the uniaxial in-plane and easy plane (shape) anisotropies. The easy-plane shape anisotropy torque arises because the switching magnet is a thin layer typically only a few nanometers thick. The in-plane uniaxial anisotropy is usually also a shape anisotropy arising from an elongated cross section of the switching magnet albert . We take the uniaxial anisotropy axis of the switching magnet to be parallel to the $`z`$-axis of the coordinate system shown in Fig. 1. Since the switching magnet lies in the $`xz`$-plane, we can write the total anisotropy field as
$$๐_\mathrm{A}=๐_\mathrm{u}+๐_\mathrm{p}$$
(18)
where $`๐_\mathrm{u}`$ and $`๐_\mathrm{p}`$ are given by
$$๐_\mathrm{u}=H_{\mathrm{u0}}(๐ฆ๐_z)๐_z,$$
(19)
$$๐_\mathrm{p}=H_{\mathrm{p0}}(๐ฆ๐_y)๐_y.$$
(20)
Here $`๐_x`$, $`๐_y`$ and $`๐_z`$ are unit vectors in the directions of the axes shown in Fig. 1. If we write the energy of the switching magnet in the anisotropy field as $`๐_\mathrm{A}๐_{\mathrm{tot}}`$, where $`๐_{\mathrm{tot}}`$ is the total spin angular momentum of the switching magnet, them $`H_{\mathrm{u0}}`$,$`H_{\mathrm{p0}}`$ which measure the strengths of the uniaxial and easy-plane anisotropies have dimensions of frequency. These quantities may be converted to a field in tesla by multiplying by $`\mathrm{}/2\mu _\mathrm{B}=5.69\times 10^{12}`$.
We are now ready to study the time evolution of the unit vector $`๐ฆ`$ in the direction of the switching magnet moment. The LLG equation takes the usual form
$$\frac{\mathrm{d}๐ฆ}{\mathrm{d}t}+\gamma ๐ฆ\times \frac{\mathrm{d}๐ฆ}{\mathrm{d}t}=๐ช$$
(21)
where the reduced total torque $`๐ช`$ acting on the switching magnet is given by
$$๐ช=\left[\left(๐_\mathrm{A}+๐_{\mathrm{ext}}\right)\times ๐_{\mathrm{tot}}+๐_{}+๐_{}\right]/\left|๐_{\mathrm{tot}}\right|.$$
(22)
Here $`๐_{\mathrm{ext}}`$ is an external field, in the same frequency units as $`๐_\mathrm{A}`$, and $`\gamma `$ is the Gilbert damping parameter. Following Sun sun , Eq. (21) may be written more conveniently as
$$\left(1+\gamma ^2\right)\frac{\mathrm{d}๐ฆ}{\mathrm{d}t}=๐ช\gamma ๐ฆ\times ๐ช.$$
(23)
It is also useful to measure the strengths of all the torques in units of the strength of the uniaxial anisotropy sun . We shall, therefore, write the total reduced torque $`๐ช`$ in the form
$$๐ช=H_{uo}\left\{(๐ฆ๐_z)๐ฆ\times ๐_zh_\mathrm{p}(๐ฆ๐_y)๐ฆ\times ๐_y+v_{}(\psi )๐ฆ\times (๐ฉ\times ๐ฆ)+\left[v_{}(\psi )+h_{\mathrm{ext}}\right]๐ฆ\times ๐ฉ\right\}$$
(24)
where the relative strength of the easy plane anisotropy $`h_\mathrm{p}=H_{\mathrm{p0}}/H_{\mathrm{u0}}`$ and $`v_{}(\psi )=vg_{}(\psi )`$, $`v_{}(\psi )=vg_{}(\psi )`$ measure the strengths of the torques $`๐_{}`$ and $`๐_{}`$. The reduced bias is defined by $`v=eV_\mathrm{b}/(|๐_{\mathrm{tot}}|H_{\mathrm{u0}})`$ and has the opposite sign from the bias voltage since $`e`$ is negative. Thus positive $`v`$ implies a flow of electrons from the polarising to the switching magnet. The last contribution to the torque in Eq. (24) is due to the external field $`H_{\mathrm{ext}}`$ with $`h_{\mathrm{ext}}=H_{\mathrm{ext}}/H_{\mathrm{u0}}`$. The external field is taken in the direction of the magnetisation of the polarising magnet, as is the case in most experimental situations.
It follows from Eq. (21) that in a steady state $`๐ช=0`$. We shall first consider some cases of experimental importance where the steady state solutions are trivial and the important physics is concerned entirely with their stability. To discuss stability, we linearise Eq. (23), using Eq. (24), about a steady state solution $`๐ฆ=๐ฆ_0`$. Thus
$$๐ฆ=๐ฆ_0+\xi ๐_\alpha +\eta ๐_\varphi ,$$
(25)
where $`๐_\alpha `$, $`๐_\varphi `$ are unit vectors in the direction $`๐ฆ`$ moves when $`\alpha `$ and $`\varphi `$ are increased independently. The linearised equation may be written in the form
$$\frac{\mathrm{d}\xi }{\mathrm{d}\tau }=A\xi +B\eta ;\frac{\mathrm{d}\eta }{\mathrm{d}\tau }=C\xi +D\eta .$$
(26)
Following Sun sun , we have introduced the natural dimensionless time variable $`\tau =tH_{\mathrm{u0}}/(1+\gamma ^2)`$. The conditions for the steady state to be stable are
$$F=A+D0;G=ADBC0$$
(27)
excluding $`F=G=0`$ 20 . For simplicity we give these conditions explicitly only for the case where either $`v_{}^{}(\psi _0)=v_{}^{}(\psi _0)=0`$, with $`\psi _0=\mathrm{cos}^1(๐ฉ๐ฆ_0)`$, or $`๐ฆ_0=\pm ๐ฉ`$. The case $`๐ฆ_0=\pm ๐ฉ`$ is very common experimentally as is discussed below. The stability condition $`G0`$ may be written
$`Q^2v_{}^2+(Qh+\mathrm{cos}2\alpha _0)(Qh+\mathrm{cos}^2\alpha _0)+h_\mathrm{p}\left\{Qh(13\mathrm{sin}^2\varphi _0\mathrm{sin}^2\alpha _0)+\mathrm{cos}2\alpha _0(12\mathrm{sin}^2\alpha _0\mathrm{sin}^2\varphi _0)\right\}`$
$`h_\mathrm{p}^2\mathrm{sin}^2\alpha _0\mathrm{sin}^2\varphi _0(12\mathrm{sin}^2\varphi _0\mathrm{sin}^2\alpha _0)0,`$ (28)
where $`v_{}=v_{}(\psi _0)`$, $`h=v_{}(\psi _0)+h_{\mathrm{ext}}`$ and $`Q=\mathrm{cos}\psi _0`$. The condition $`F0`$ takes the form
$$2(v_{}+\gamma h)Q\gamma (\mathrm{cos}2\alpha _0+\mathrm{cos}^2\alpha _0)\gamma h_\mathrm{p}(13\mathrm{sin}^2\varphi _0\mathrm{sin}^2\alpha _0)0.$$
(29)
We now discuss several interesting examples, the first of these relating to experiments of Grollier et al. 18 and others. In these experiments the magnetisation of the polarising magnet, the uniaxial anisotropy axis and the external field are all collinear (along the in-plane $`z`$-axis in our convention). In this case the equation $`๐ช=0`$, with $`๐ช`$ given by Eq. (24), shows immediately that possible steady states are given by $`๐ฆ_0=\pm ๐ฉ(\alpha _0=0,\pi )`$, corresponding to the switching magnet moment along the $`z`$-axis. These are the only solutions when $`h_\mathrm{p}=0`$. For $`h_\mathrm{p}0`$ other steady-state solutions may exist but in the parameter regime which has been investigated they are always unstable 13 . We shall assume this is always the case and concentrate on the solutions $`๐ฆ_0=\pm ๐ฉ`$. In the state of parallel magnetisation (P) $`๐ฆ_0=๐ฉ`$ we have $`v_{}=vg_{}(0)`$, $`h=vg_{}(0)+h_{\mathrm{ext}}`$, $`\alpha _0=0`$ and $`Q=1`$. The stability conditions (III) and (29) become
$$\left[g_{}(0)\right]^2v^2+(vg_{}(0)+h_{\mathrm{ext}}+1)^2+h_\mathrm{p}\left[vg_{}(0)+h_{\mathrm{ext}}+1\right]0$$
(30)
$$g_{}(0)v+\gamma \left[vg_{}(0)+h_{\mathrm{ext}}+1+\frac{1}{2}h_\mathrm{p}\right]0.$$
(31)
In the state of anti-parallel magnetisation (AP) $`๐ฆ_0=๐ฉ`$ we have $`v_{}=vg_{}(\pi )`$, $`h=vg_{}(\pi )+h_{\mathrm{ext}}`$, $`\alpha _0=\pi `$ and $`Q=1`$. The stability conditions for the AP state are thus
$$\left[g_{}(\pi )\right]^2v^2+(vg_{}(\pi )h_{\mathrm{ext}}+1)^2+h_\mathrm{p}\left[vg_{}(\pi )h_{\mathrm{ext}}+1\right]0$$
(32)
$$g_{}(\pi )v+\gamma \left[vg_{}(\pi )+h_{\mathrm{ext}}1\frac{1}{2}h_\mathrm{p}\right]0.$$
(33)
In the regime of low external field ($`h_{\mathrm{ext}}1`$, i.e. $`H_{\mathrm{ext}}H_{u0}`$) we have $`H_\mathrm{p}>>H_{\mathrm{ext}}`$ ($`h_\mathrm{p}100`$). Eqs. (30) and (32) may be then approximated by
$$vg_{}(0)+h_{\mathrm{ext}}+1>0$$
(34)
$$vg_{}(\pi )+h_{\mathrm{ext}}1<0.$$
(35)
Equation (34) corresponds to P stability and (35) to AP stability. It is convenient to define scalar quantities $`T_{}`$, $`T_{}`$ by $`T_{}=g_{}(\psi )\mathrm{sin}\psi `$, $`T_{}=g_{}(\psi )\mathrm{sin}\psi `$, these being scalar components of spin-transfer torque in units of $`eV_\mathrm{b}`$ (cf. Eq. (II)). Then $`g_i(0)=[\mathrm{d}T_i/\mathrm{d}\psi ]_{\psi =0}`$ and $`g_i(\pi )=[\mathrm{d}T_i/\mathrm{d}\psi ]_{\psi =\pi }`$ with $`i=,`$. Model calculations 13 show that both $`g_{}`$ and $`g_{}`$ can be of either sign, although positive values are more common. Also there is no general rule about the relative magnitude of $`g_i(0)`$ and $`g_i(\pi )`$.
We now illustrate the consequences of the above stability conditions by considering two limiting cases. We first consider the case $`g_{}(\psi )=0`$, $`g_{}>0`$, as assumed by Grollier et. al 18 in the analysis of their data. In Fig. 4 we plot the regions of P and AP stability deduced from Eqs. (31),(33)-(35), in the ($`v`$,$`h_{\mathrm{ext}}`$)-plane. Grollier et al. plot current instead of bias but this should not change the form of the figure. Theirs is rather more complicated, owing to a less transparent stability analysis with unnecessary approximation. The only approximations made above, to obtain Eqs. (34) and(35), can easily be removed, which results in the critical field lines $`h_{\mathrm{ext}}=\pm 1`$ acquiring a very slight curvature given by $`h_{\mathrm{ext}}1+[vg_{}(\pi )]^2/h_\mathrm{p}`$ and $`h_{\mathrm{ext}}1[vg_{}(0)]^2/h_\mathrm{p}`$. The critical biases in the figure are give by
$`v_{\mathrm{AP}\mathrm{P}}`$ $`=`$ $`\gamma \left[1+{\displaystyle \frac{1}{2}}h_\mathrm{p}h_{\mathrm{ext}}\right]/g_{}(\pi )`$
$`v_{\mathrm{P}\mathrm{AP}}`$ $`=`$ $`\gamma \left[1+{\displaystyle \frac{1}{2}}h_\mathrm{p}+h_{\mathrm{ext}}\right]/g_{}(0).`$ (36)
A downward slope from left to right of the corresponding lines in Fig. 4 is not shown there. Since the damping parameter $`\gamma `$ is small ($`\gamma 0.01`$) this downward slope of the critical bias lines is also small. From Fig. 4 we can deduce the behaviour of resistance versus bias in the external field regimes $`|h_{\mathrm{ext}}|<1`$ and $`|h_{\mathrm{ext}}|>1`$.
Consider first the case $`|h_{\mathrm{ext}}|<1`$. Suppose we start in the AP state with a bias $`v=0`$ which is gradually increased to $`v_{\mathrm{AP}P}`$. At this point the AP state becomes unstable and the system switches to the P state as $`v`$ increases further. On reducing $`v`$ the hysteresis loop is completed via a switch back to the AP state at the negative bias $`v_{\mathrm{P}AP}`$. The hysteresis loop is shown in Fig. 5(a). The increase in resistance R between the P and AP states is the same as would be produced by varying the applied field in a GMR experiment.
Now consider the case $`h_{\mathrm{ext}}<1`$. Starting again in the AP state at $`v=0`$ we see from Fig. 4 that, on increasing $`v`$ to $`v_{\mathrm{AP}\mathrm{P}}`$, the AP state becomes unstable but there is no stable P state to switch to. This point is marked by an asterisk in Fig. 5(b). For $`v>v_{\mathrm{AP}\mathrm{P}}`$, the moment of the switching magnet is in a persistently time-dependent state. However, if $`v`$ is now decreased below $`v_{\mathrm{P}\mathrm{AP}}`$ the system homes in on the stable AP state and the overall behaviour is reversible, i.e. no switching and no hysteresis occur. When $`h_{\mathrm{ext}}>1`$ similar behaviour, now involving the P state, occurs at negative bias, as shown in Fig. 5(b). The dashed curves in Fig. 5(b) show a hypothetical time-averaged resistance in the regions of time-dependent magnetisation. As discussed later time-resolved measurements of resistance suggest that several different types of dynamics can occur in these regions.
It is clear from Fig. 5(a) that the jump AP$``$P always occurs for positive bias $`v`$, which corresponds to flow of electrons from the polarising to the switching magnet. This result depends on the assumption that $`g_{}>0`$; if $`g_{}<0`$ it is easy to see that the sense of the hysteresis loop is reversed and the jump P$``$AP occurs for positive $`v`$. To our knowledge this reverse jump has never been observed, although $`g_{}<0`$ can occur in principle and is predicted theoretically 13 for the Co/Cu/Co(111) system with a switching magnet consisting of a single atomic plane of Co. It follows from Eq. (III) that $`|v_{\mathrm{P}\mathrm{AP}}/v_{\mathrm{AP}\mathrm{P}}|=|g_{}(\pi )/g_{}(0)|`$ in zero external field. Experimentally this ratio, essentially the same as the ratio of critical currents, may be considerably less than 1 (e.g. $`<0.5`$ albert ), greater than 1 (e.g. $`2`$ 19 ) or close to 1 16 . Usually the field dependence of the critical current is found to be stronger than that predicted by Eq. (III) albert ; 16 .
We now discuss the reversible behaviour shown in Fig. 5(b) which occurs for $`|h_{\mathrm{ext}}|>1`$. The transition from hysteretic to reversible behaviour at a critical external field seems to have been first seen in pillar structures by Katine et al. 21 . Curves similar to the lower one in Fig. 5(b) are reported with $`|v_{\mathrm{P}\mathrm{AP}}|`$ increasing with increasing $`h_{\mathrm{ext}}`$, as expected from Eq. (III). Plots of the differential resistance $`\mathrm{d}V/\mathrm{d}I`$ show a peak near the point of maximum gradient of the dashed curve. Similar behaviour has been reported by several groups 22 ; 23 ; 24 . It is particularly clear in the work of Kiselev at al. 22 that the transition from hysteretic behaviour (as in Fig. 5(a)) to reversible behaviour with peaks in $`\mathrm{d}V/\mathrm{d}I`$ occurs at the coercive field 600 Oe of the switching layer ($`h_{\mathrm{ext}}=1`$). The important point about the peaks in $`\mathrm{d}V/\mathrm{d}I`$ is that for a given sign of $`h_{\mathrm{ext}}`$ they only occur for one sign of the bias. This clearly shows that this effect is due to spin-transfer and not to Oersted fields. Myers et al. 25 show a current-field stability diagram similar to the bias-field one of Fig. 4 with a critical field of 1500 Oe. They examine the time dependence of the resistance at room temperature with the field and current adjusted so that the system is in the โboth unstableโ region in the fourth quadrant of Fig. 4 but very close to its top left-hand corner. They observe telegraph-noise-type switching between approximately P and AP states with slow switching times in the range 0.1-10 s. Similar telegraph noise with faster switching times was observed by Urazhdin et al. 23 at current and field close to a peak in $`\mathrm{d}V/\mathrm{d}I`$. In the region of P and AP instability Kiselev et al. 22 and Pufall et al. 24 report various types of dynamics of precessional type and random telegraph switching type in the microwave Ghz regime. Kiselev et al. 22 propose that systems of the sort considered here might serve as nanoscale microwave sources or oscillators, tunable by current and field over a wide frequency range.
We now return to the stability conditions (31),(33)-(35) and consider the case of $`g_{}(\psi )0`$ but $`h_{\mathrm{ext}}=0`$. These conditions of stability of the P state may be written approximately, remembering that $`\gamma <<1`$, $`h_\mathrm{p}>>1`$, as
$$vg_{}(0)>1,vg_{}(0)>\frac{1}{2}\gamma h_\mathrm{p}.$$
(37)
The conditions for stability of the AP state are
$$vg_{}(\pi )<1,vg_{}(\pi )<\frac{1}{2}\gamma h_\mathrm{p}.$$
(38)
In Fig. 6 we plot the regions of P and AP stability, assuming $`g_{}(0)=g_{}(\pi )=g_{}`$ and $`g_{}(0)=g_{}(\pi )=g_{}`$ for simplicity. We also put $`r=g_{}/g_{}`$.
For $`r>0`$ we find the normal hysteresis loop as in Fig. 5(a) if we plot $`R`$ against $`vg_{}`$ (valid for either sign of $`g_{}`$). In Fig. 7 we plot the hysteresis loops for the cases $`r_\mathrm{c}<r<0`$ and $`r<r_\mathrm{c}`$, where $`r_\mathrm{c}=2/(\gamma h_\mathrm{p})`$ is the value of $`r`$ at the point $`X`$ in Fig. 6.
The points labelled by asterisks have the same significance as in Fig. 5(b). If in Fig. 7(a) we increase $`vg_{}`$ beyond its value indicated by the right-hand asterisk we move into the โboth-unstableโ region where the magnetisation direction of the switching magnet is perpetually in a time-dependent state. Thus negative $`r`$ introduces behaviour in zero applied field which is similar to that found when the applied field exceeds the coercive field of the switching magnet for $`r=0`$. This behaviour was predicted by Edwards et al. 13 , in particular for a Co/Cu/Co(111) system with the switching magnet consisting of a Co monolayer. Zimmler et al. 26 use methods similar to the ones described here to analyse their data on a Co/Cu/Co nanopillar and deduce that $`g_{}>0`$, $`r=g_{}/g_{}0.2`$. It would be interesting to carry out time-resolved resistance measurements on this system at large current density (corresponding to $`vg_{}<1`$) and zero external field.
So far we have considered the low-field regime ($`H_{\mathrm{ext}}`$ coercive field of switching magnet) with both magnetisations and the external field in-plane. There is another class of experiments in which a high field, greater than the demagnetising field ($`>2T`$), is applied perpendicular to the plane of the layers. The magnetisation of the polarising magnet is then also perpendicular to the plane. This is the situation in the early experiments where a point contact was employed to inject high current densities into magnetic multilayers 27 ; 28 ; 29 . In this high-field regime a peak in the differential resistance $`\mathrm{d}V/\mathrm{d}I`$ at a critical current was interpreted as the onset of current-induced excitation of spin waves in which the spin-transfer torque leads to uniform precession of the magnetisation 6 ; 27 ; 28 . No hysteretic magnetisation reversal was observed and it seemed that the effect of spin-polarised current on the magnetisation is quite different in the low- and high-field regimes. Recently, however, รzyilmaz et al. 30 have studied Co/Cu/Co nanopillars ($`100`$nm in diameter) at $`T=4.2`$K for large applied fields perpendicular to the layers. They observe hysteretic magnetisation reversal and interpret their results using the Landau-Lifshitz equation. We now give a similar discussion within the framework of this section.
Following รzyilmaz et al., we neglect the uniaxial anisotropy term in Eq. (24) for the reduced torque $`๐ช`$ while retaining $`H_{u0}`$ as a scalar factor. Hence
$$๐ช=H_{u0}\left\{\left[h_{\mathrm{ext}}+v_{}(\psi )h_\mathrm{p}\mathrm{cos}\psi \right]๐ฆ\times ๐ฉ+v_{}(\psi )๐ฆ\times (๐ฉ\times ๐ฆ)\right\}$$
(39)
where $`๐ฉ`$ is the unit vector perpendicular to the plane. When $`v_{}(\psi )0`$ the only possible steady-state solutions of $`๐ช=0`$ are $`๐ฆ_0=\pm ๐ฉ`$. On linearizing Eq. 23 about $`๐ฆ_0`$ as before we find that the condition $`G0`$ is always satisfied. The second stability condition $`F<0`$ becomes
$$\left[v_{}(\psi _0)+\gamma (v_{}(\psi _0)+h_{\mathrm{ext}}h_\mathrm{p})\right]\mathrm{cos}\psi _0>0$$
(40)
where $`\psi _0=\mathrm{cos}^1(๐ฆ_0๐ฉ)`$. Applying this to the P state ($`\psi _0=0`$) and the AP state ($`\psi _0=\pi `$) we obtain the conditions
$$v>\gamma (h_\mathrm{p}h_{\mathrm{ext}})/g(0)$$
(41)
$$v<\gamma (h_\mathrm{p}+h_{\mathrm{ext}})/g(\pi ),$$
(42)
where the first condition applies to the P stability and the second to the AP stability. Here $`g(\psi )=g_{}(\psi )+\gamma g_{}(\psi )`$. The corresponding stability diagram is shown in Fig. 8, where we have assumed $`g(\pi )>g(0)>0`$ for definiteness.
The boundary lines cross at $`h_{\mathrm{ext}}=h_\mathrm{c}`$, where $`h_\mathrm{c}=h_\mathrm{p}[g(\pi )+g(0)]/[g(\pi )g(0)]`$. This analysis is only valid for fields larger than the demagnetising field ($`h_{\mathrm{ext}}>h_\mathrm{p}`$) and we see from the figure that for $`h_{\mathrm{ext}}>h_\mathrm{c}`$ hysteretic switching occurs. This takes place for only one sign of the bias (current) and the critical biases (currents) increase linearly with $`h_{\mathrm{ext}}`$ as does the width of the hysteresis loop $`|v_{\mathrm{P}\mathrm{AP}}v_{\mathrm{AP}\mathrm{P}}|`$. This accords with the observations of รzyilmaz et al. The critical currents are not larger than those in the low-field or zero-field regimes (cf. Eqs. (41), (42) with Eq. (III)) and yet the magnetisation of the switching magnet can be switched against a very large external field. However, in this case the AP state is only stabilised by maintaining the current.
The experiments on spin transfer discussed above have mainly been carried out at constant temperature, typically $`4.2`$K or room temperature. The effect on current-driven switching of varying the temperature has recently been studied by several groups 23 ; 25 ; 31 . The standard Nรฉel-Brown theory of thermal switching 32 does not apply because the Slonczewski in-plane torque is not derivable from an energy function. Li and Zhang 33 have generalised the standard stochastic Landau-Lifschitz equation, which includes white noise in the effective applied field, to include spin transfer torque. In this way they have successfully interpreted some of the experimental data. A full discussion of this work is outside the scope of the present review. However it should be pointed out that in addition to the classical effect of white noise there is an intrinsic temperature dependence of quantum origin. This arises from the Fermi distribution functions which appear in expressions for the spin-transfer torque (see Eqs. (14) and (12)).
So far we have discussed steady-state solutions of the LLG equation (23). It is important to study the magnetisation dynamics of the switching layer in the situation during the jumps AP$``$P and P$``$AP of the hysteresis curve in zero external field, and secondly under conditions where only time-dependent solutions are possible, for example in the regions of sufficiently strong current and external field marked โboth unstableโ in Fig. 4. The first situation has been studied by Sun sun , assuming single-domain behaviour of the switching magnet, and by Miltat et al. 34 with more general micromagnetic configurations. Both situations have been considered by Li and Zhang 35 . In the second case they find precessional states, and the possibility of โtelegraph noiseโ at room temperature, as seen experimentally in Refs. 22 ; 24 . Switching times (AP$``$P and P$``$AP) are estimated to be of the order 1ns. Micromagnetic simulations 34 indicate that the Oersted field cannot be completely ignored for typical pillars with diameter of the order of 100nm.
Finally, in this section, we briefly discuss some practical considerations which may ultimately decide whether current-induced switching is useful in spintronics. Sharp switching, with nearly rectangular hysteresis loops, is obviously desirable and this demands single-domain behaviour. In experiments on nanopillars of circular cross section 21 multidomain behaviour was observed with the switching transition spread over a range of current. Subsequently the same group albert found sharp switching in pillars whose cross-section was an elongated hexagon, which introduces strong uniaxial in-plane shape anisotropy. It was known from earlier magnetisation studies of nanomagnet arrays 36 that such a shape anisotropy can result in single domain behaviour. A complex switching transition need not necessarily indicate multidomain behaviour. It could also arise from a marked departure of $`T_{}(\psi )`$ and/or $`T_{}(\psi )`$ from sinusoidal behaviour, such as occurs near $`\psi =\pi `$ in calculations for Co/Cu/Co(111) with two atomic planes of Co in the switching magnet (see Fig. 9(b)). In the calculations of the corresponding hysteresis loops (Fig. 11) the torques were approximated by sine curves but an accurate treatment would certainly complicate the AP$``$P transition which occurs at negative bias in Fig. 11(b). Studies of this effect are planned.
The critical current density for switching is clearly an important parameter. From Eq. (III) the critical reduced bias for the P$``$AP transition is to a good approximation given by $`\gamma h_\mathrm{p}/[2g_{}(0)]`$. Using the definitions of reduced quantities given after Eq. (24), we may write the actual critical bias in volts as
$$V_{PAP}=M\gamma M_\mathrm{s}H_\mathrm{d}/[2g_{}(0)|e|],$$
(43)
where $`M`$ is the number of atomic planes in the switching magnet, $`M_\mathrm{s}`$ is the average moment ($`J/T`$) of the switching magnet per atomic plane per unit area, and $`H_\mathrm{d}=\mathrm{}H_{\mathrm{p0}}/(2\mu _\mathrm{B})`$ is the easy-plane anisotropy field in tesla. As expressed earlier $`g_{}(0)=(\mathrm{d}T_{}/\mathrm{d}\psi )_{\psi =0}`$ where the torque $`T_{}`$ is per unit area in units of $`eV_\mathrm{B}`$. (The calculated torques in Figs. 9 and 10 of Sec. V are per surface atom so that if these are used to determine $`g_{}(0)`$ in Eq. (43) $`M_\mathrm{s}`$ must be taken per surface atom.)
An obvious way to reduce the critical bias, and hence the critical current, is to reduce $`M`$, the thickness of the switching magnet. Calculations show 13 (see also Fig. 10) that $`g_{}`$ does not decrease with $`M`$ and may, in fact, increase for small values such as $`M=2`$. Careful design of the device might also increase $`g_{}(0)`$ beyond the values ($`<0.01`$ per surface atom) which seem to be obtainable in simple trilayers 13 . Jiang et al. 37 ; 38 , have studied various structures in which the polarising magnet is pinned by an adjacent antiferromagnet (exchange biasing) and in which a thin Ru layer is incorporated between the switching layer and the lead. Critical current densities of $`2\times 10^6`$Acm<sup>-2</sup> have been obtained which are substantially lower than those in Co/Cu/Co trilayers. Such structures can quite easily be investigated theoretically by the methods of Section V.
Decreasing the magnetisation $`M_\mathrm{s}`$, and hence the demagnetising field ($`H_\mathrm{d}`$), would be favourable but $`g_{}`$ then tends to decrease also 13 . A possible way of decreasing $`H_\mathrm{d}`$ without decreasing local magnetic moments in the system is to use a synthetic ferrimagnet as the switching magnet 39 . The Gilbert damping factor $`\gamma `$ is another crucial parameter but it is uncertain whether this can be decreased significantly. However, the work of Capelle and Gyorffy 40 is an interesting theoretical development. The search for structures with critical current densities low enough for use in spintronic devices ($`10^5`$Acm<sup>-2</sup> perhaps) 41 is an enterprise where experiment and quantitative calculations 13 should complement each other fruitfully.
## IV Quantitative theory of spin-transfer torque
### IV.1 General principles
To put the phenomenological treatment of Sec. III on a first-principle quantitative basis we must calculate the spin-transfer torques (Eqs. (II) in a steady state for real systems. For this purpose it is convenient to describe the magnetic and nonmagnetic layers of Fig. 1 by tight-binding models, in general multiorbital with s, p, and d orbitals whose one-electron parameters are fitted to first-principle bulk band structure 42 . The hamiltonian is therefore of the form
$$H=H_0+H_{\mathrm{int}}+H_{\mathrm{anis}}$$
(44)
where the one-electron hopping term $`H_0`$ is given by
$$H_0=\underset{k_{}\sigma }{}\underset{m\mu ,n\nu }{}t_{m\mu ,n\nu }(๐ค_{})c_{๐ค_{}m\mu \sigma }^{}c_{๐ค_{}n\nu \sigma },$$
(45)
where $`c_{k_{}m\mu \sigma }^{}`$ creates an electron in a Bloch state, with in-plane wave vector $`๐ค_{}`$ and spin $`\sigma `$, formed from a given atomic orbital $`\mu `$ in plane $`m`$. Eq. 45 generalises the single orbital eq. (1). $`H_{\mathrm{int}}`$ is an on-site interaction between electrons in d orbitals which leads to an exchange splitting of the bands in the ferromagnets and is neglected in the spacer and lead. Finally, $`H_{\mathrm{anis}}`$ contains anisotropy fields in the switching magnet and is given by
$$H_{\mathrm{anis}}=\underset{n}{}๐_n๐_\mathrm{A},$$
(46)
where $`๐_n`$ is the operator of the total spin angular momentum of plane $`n`$ and $`๐_\mathrm{A}`$ is given by Eqs. (18)-(20) with the unit vector $`๐ฆ`$ in the direction of $`_n๐_n`$, where $`๐_n`$ is the thermal average of $`๐_n`$. We assume here that the anisotropy fields $`H_{\mathrm{u0}}`$,$`H_\mathrm{p}`$ are uniform throughout the switching magnet but we could generalise to include, for example, a surface anisotropy.
In the tight-binding description, the spin angular momentum operator $`๐_n`$ is given by
$$๐_n=\frac{1}{2}\mathrm{}\underset{k_\mu }{}(c_{k_{}n\mu }^{},c_{k_{}n\mu }^{})๐(c_{k_{}n\mu },c_{k_{}n\mu })^\mathrm{T}$$
(47)
and the corresponding operator for the spin angular momentum current between planes $`n1`$ and $`n`$ is
$$๐ฃ_{n1}=\frac{1}{2}\mathrm{}\underset{๐ค_{}\mu \nu }{}t(๐ค_{})_{n\nu ,n1\mu }(c_{k_{}n\nu }^{},c_{k_{}n\nu }^{})๐(c_{k_{}n1\mu },c_{k_{}n1\mu })^\mathrm{T}+\mathrm{h}.\mathrm{c}.,$$
(48)
which generalises the single orbital expression (2). The rate of change of $`๐_n`$ in the switching magnet is given by
$$\mathrm{i}\mathrm{}\frac{\mathrm{d}๐_n}{\mathrm{d}t}=[๐_n,H_0]+[๐_n,H_{\mathrm{anis}}].$$
(49)
This results holds since the spin operator commutes with the interaction hamiltonian $`H_{\mathrm{int}}`$.
It is straightforward to show that
$$[๐_n,H_0]=\mathrm{i}\mathrm{}(๐ฃ_{n1}๐ฃ_n),$$
(50)
and
$$[๐_n,H_{\mathrm{anis}}]=\mathrm{i}\mathrm{}(๐_\mathrm{A}\times ๐_n).$$
(51)
On taking the thermal average, Eq. (49) becomes
$$\frac{\mathrm{d}๐_n}{\mathrm{d}t}=๐ฃ_{n1}๐ฃ_n๐_\mathrm{A}\times ๐_{\mathrm{tot}},$$
(52)
This corresponds to an equation of continuity, stating that the rate of change of spin angular momentum on plane $`n`$ is equal to the difference between the rate of flow of this quantity onto and off the plane, plus the rate of change due to precession around the field $`๐_\mathrm{A}`$. When Eq. (52) is summed over all planes in the switching magnet we have
$$\frac{\mathrm{d}}{\mathrm{d}t}๐_{\mathrm{tot}}=๐^{\mathrm{s}\mathrm{t}}๐_\mathrm{A}\times ๐_{\mathrm{tot}},$$
(53)
where the total spin-transfer torque $`๐^{\mathrm{s}\mathrm{t}}`$ is given by Eq. (14) and $`๐_{\mathrm{tot}}`$ is the total spin angular momentum of the switching magnet. Equation (53) is equivalent to Eq. (21), for zero external field, in the absence of damping. Equation (14) shows how $`๐^{\mathrm{s}\mathrm{t}}`$ required for the phenomenological treatment of Sec. III is to be determined from the calculated spin currents in the spacer and lead. As discussed in Sec. III, the magnetization of a single-domain sample is essentially uniform and the spin-transfer torque $`๐^{\mathrm{s}\mathrm{t}}`$ depends on the angle $`\psi `$ between the magnetisations of the polarising and switching magnets.
To consider time-dependent solutions of Eq. (21) it is necessary to calculate $`๐^{\mathrm{s}\mathrm{t}}`$ for arbitrary angle $`\psi `$ and for this purpose $`๐_\mathrm{A}`$ can be neglected. To reduce the calculation of the spin-transfer torque to effectively a one-electron problem, we replace $`H_{\mathrm{int}}`$ by a selfconsistent exchange field term $`_n๐_n๐ซ_n`$, where the exchange field $`๐ซ_n`$ should be determined selfconsistently in the spirit of an unrestricted Hartree-Fock (HF) or local spin density (LSD), approximation. The essential selfconsistency condition in any HF or LSD calculation is that the local moment $`๐_n`$ in a steady state is in the same direction as $`๐ซ_n`$. Thus we require
$$๐ซ_n\times ๐_n=0$$
(54)
for each atomic plane of the switching magnet. It is useful to consider first the situation when there is no applied bias and the polarising and switching magnets are separated by a spacer which is so thick that the zero-bias oscillatory exchange coupling 44 is negligible. In that case we have two independent magnets and the selfconsistent exchange field in every atomic plane of the switching magnet is parallel to its total magnetisation which is uniform and assumed to be along the $`z`$-axis. Referring to Fig. 1 the selfconsistent solution therefore corresponds to uniform exchange fields in the polarising and switching magnets which are at an assumed angle $`\psi =\theta `$ with respect to one another.
When a bias $`V_\mathrm{b}`$ is applied , with a uniform exchange field $`๐ซ=\mathrm{\Delta }๐_z`$ in the switching magnet imposed, the calculated local moments $`๐_n`$ will deviate from the $`z`$-direction so that the solution is not selfconsistent. To prepare a selfconsistent state with $`๐ซ`$ and all $`๐_n=๐`$ in the $`z`$-direction it is necessary to apply fictitious constraining fields $`๐_n`$ of magnitude proportional to $`V_\mathrm{b}`$. The local field for plane $`n`$ is thus $`๐ซ+๐_n`$ but to calculate the spin currents in the spacer and lead, and hence $`๐^{\mathrm{s}\mathrm{t}}`$ from Eq. (14), the fields $`๐_n`$, of the order of $`V_\mathrm{b}`$, may be neglected compared with $`๐ซ`$. Although the fictitious constraining fields $`๐_n`$ need therefore never be calculated, it is interesting to see that they are in fact related to $`๐^{\mathrm{s}\mathrm{t}}`$. For the constrained self-consistent steady state ($`๐_n=๐`$, $`\dot{๐}_n=0`$) in the presence of the constraining fields, with $`๐_\mathrm{A}`$ neglected as discussed above, it follows from Eq. (52) that
$$๐ฃ_{๐ง\mathrm{๐}}๐ฃ_๐ง=(\mathrm{\Delta }+๐_n)\times S=๐_n\times ๐,$$
(55)
where the local field $`๐ซ+๐_n`$ replaces $`๐_\mathrm{A}`$. On summing over all atomic planes $`n`$ in the switching magnet we have
$$๐^{\mathrm{s}\mathrm{t}}=๐ฃ_{\mathrm{spacer}}๐ฃ_{\mathrm{lead}}=\underset{n}{}๐_n\times ๐.$$
(56)
Thus, as expected, in the prepared state with a given angle $`\psi `$ between the magnetisations of the magnetic layers the spin-transfer torque is balanced by the total torque due to the constraining fields.
In the simple model of Section II, with infinite exchange splitting in the magnets, the local moment is constrained to be in the direction of the exchange field so the question of selfconsistency is not raised.
The main conclusion of this Section is that the spin-transfer torque for a given angle $`\psi `$ between magnetisations may be calculated using uniform exchange fields making the same angle with one another. Such calculations are described in Sec. II and V. The use of this spin-transfer torque in the LLG equation of Section III completes what we shall call the โstandard modelโ (SM). It underlies the original work of Slonczewski 3 and most subsequent work. The spin-transfer torque calculated in this way should be appropriate even for time-dependent solutions of the LLG equation. This is based on the reasonable assumption that the time for the electronic system to attain a โconstrained steady stateโ with given $`\psi `$ is short compared with the time-scale ($``$1ns) of the macroscopic motion of the switching magnet moment.
Although the SM is a satisfactory way of calculating the spin-transfer torque its lack of selfconsistency leads to some non-physical concepts. The first of these is the โtransverse spin accumulationโ in the switching magnet 46 ; 47 , This refers to the deviations of local moments $`๐_n`$ from the direction of the exchange field, assumed uniform in the SM. In a self-consistent treatment such deviations do not occur because the exchange field is always in the direction of the local moment. A related non-physical concept is the โspin decoherence lengthโ over which the spin accumulation is supposed to decay 46 ; 47 , More detailed critiques of these concepts are given elsewhere 13 ; em .
### IV.2 Keldysh formalism for fully realistic calculations of the spin-transfer torque
The wave-function approach to spin-transfer torque described in Section II is difficult to apply to realistic multiorbital systems. For this purpose Green functions are much more convenient and Keldysh 11 developed a Green function approach to the non-equilibrium problem of electron transport. In this section we apply this method to calculate spin currents in a magnetic layer structure, following Edwards et al. 13 .
The structure we consider is shown schematically in Fig. 1. It consists of a thick (semi-infinite) left magnetic layer (polarising magnet), a nonmagnetic metallic spacer layer of $`N`$ atomic planes, a thin switching magnet of $`M`$ atomic planes, and a semi-infinite lead. The broken line between the atomic planes $`n1`$ and $`n`$ represents a cleavage plane separating the system into two independent parts so that charge carriers cannot move between the two surface planes $`n1`$ and $`n`$. It will be seen that our ability to cleave the whole system in this way is essential for the implementation of the Keldysh formalism. This can be easily done with a tight-binding parametrisation of the band structure by simply switching off the matrix of hopping integrals $`t_{n\nu ,n1\mu }`$ between atomic orbitals $`\nu `$, $`\mu `$ localised in planes $`n1`$ and $`n`$. We therefore adopt the tight-binding description with the Hamiltonian defined by Eqs. (44-47).
To use the Keldysh formalism 11 ; 12 ; 53 to calculate the charge or spin currents flowing between the planes $`n1`$ and $`n`$, we consider an initial state at time $`\tau =\mathrm{}`$ in which the hopping integral $`t_{n\nu ,n1\mu }`$ between planes $`n1`$ and $`n`$ is switched off. Then both sides of the system are in equilibrium but with different chemical potentials $`\mu _\mathrm{L}`$ on the left and $`\mu _\mathrm{R}`$ on the right, where $`\mu _\mathrm{L}\mu _\mathrm{R}=eV_\mathrm{b}`$. The interplane hopping is then turned on adiabatically and the system evolves to a steady state. The cleavage plane, across which the hopping is initially switched off, may be taken in either the spacer or in one of the magnets or in the lead. In principle, the Keldysh method is valid for arbitrary bias $`V_\mathrm{b}`$ but here we restrict ourselves to small bias corresponding to linear response. This is always reasonable for a metallic system. For larger bias, which might occur with a semiconductor or insulator as spacer, electrons would be injected into the right part of the system far above the Fermi level and many-body processes neglected here would be important. Following Keldysh 11 ; 12 , we define a two-time matrix
$$G_{\mathrm{RL}}^+(\tau ,\tau ^{})=\mathrm{i}c_\mathrm{L}^{}(\tau ^{})c_\mathrm{R}(\tau ),$$
(57)
where $`R(n,\nu ,\sigma ^{})`$ and $`L(n1,\mu ,\sigma )`$, and we suppress the $`k_{}`$ label. The thermal average in Eq. (57) is calculated for the steady state of the coupled system. The matrix $`G_{\mathrm{RL}}^{}`$ has dimensions $`2m\times 2m`$ where $`m`$ is the number of orbitals on each atomic site, and is written so that the $`m\times m`$ upper diagonal block contains matrix elements between $``$ spin orbitals and the $`m\times m`$ lower diagonal block relates to $``$ spin. $`2m\times 2m`$ hopping matrices $`t_{\mathrm{LR}}`$ and $`t_{\mathrm{RL}}`$ are written similarly and in this case only the diagonal blocks are nonzero. If we denote $`t_{\mathrm{LR}}`$ by $`t`$, then $`t_{\mathrm{RL}}=t^{}`$. We also generalise the definition of $`๐`$ so that its components are now direct products of the $`2\times 2`$ Pauli matrices $`\sigma _x`$, $`\sigma _y`$, $`\sigma _z`$, and the $`m\times m`$ unit matrix. The thermal average of the spin current operator, given by Eq. (49), may now be expressed as
$$๐ฃ_{n1}=\frac{1}{2}\underset{๐ค_{}}{}\mathrm{Tr}\left\{\left[G_{\mathrm{RL}}^+(\tau ,\tau )tG_{\mathrm{LR}}^+(\tau ,\tau )t^{}\right]๐\right\}.$$
(58)
Introducing the Fourier transform $`G^+(\omega )`$ of $`G^+(\tau ,\tau ^{})`$ , which is a function of $`\tau \tau ^{}`$, we have
$$๐ฃ_{n1}=\frac{1}{2}\underset{๐ค_{}}{}\frac{\mathrm{d}\omega }{2\pi }\mathrm{Tr}\left\{\left[G_{\mathrm{RL}}^+\left(\omega \right)tG_{\mathrm{LR}}^+(\omega )t^{}\right]๐\right\}.$$
(59)
The charge current is given by Eq. (59) with $`\frac{1}{2}๐`$ replaced by the unit matrix multiplied by $`e/\mathrm{}`$.
Following Keldysh 11 ; 12 we now write
$$G_{\mathrm{AB}}^+(\omega )=\frac{1}{2}\left(F_{\mathrm{AB}}+G_{\mathrm{AB}}^\mathrm{a}G_{\mathrm{AB}}^\mathrm{r}\right),$$
(60)
where the suffices $`A`$ and $`B`$ are either $`R`$ or $`L`$. $`F_{\mathrm{AB}}(\omega )`$ is the Fourier transform of
$$F_{\mathrm{AB}}(\tau ,\tau ^{})=\mathrm{i}[c_\mathrm{A}(\tau ),c_\mathrm{B}^{}(\tau ^{})]_{}$$
(61)
and $`G^\mathrm{a}`$, $`G^\mathrm{r}`$ are the usual advanced and retarded Green functions 54 . Note that in 11 and 12 the definitions of $`G^\mathrm{a}`$ and $`G^\mathrm{r}`$ are interchanged and that in the Green function matrix defined by these authors $`G^+`$ and $`G^{}`$ should be interchanged.
Charge and spin current are related by Eqs. (59) and (60) to the quantities $`G^\mathrm{a}`$, $`G^\mathrm{r}`$ and $`F_{\mathrm{AB}}`$. The latter are calculated for the coupled system by starting with decoupled left and right systems, each in equilibrium, and turning on the hopping between planes L and R as a perturbation. Hence, we express $`G^\mathrm{a}`$, $`G^\mathrm{r}`$ and $`F_{\mathrm{AB}}`$ in terms of retarded surface Green functions $`g_Lg_{\mathrm{LL}}`$, $`g_\mathrm{R}g_{\mathrm{RR}}`$ for the decoupled equilibrium system. It is then found 13 that the spin current between the planes $`n1`$ and $`n`$ can be written as the sum $`๐ฃ_{n1}=๐ฃ_{n1}_1+๐ฃ_{n1}_2`$, where the two contributions to the spin current $`๐ฃ_n_1`$, $`๐ฃ_n_2`$ are given by
$$\mathrm{j}_{n1}_1=\frac{1}{4\pi }\underset{๐ค_{}}{}d\omega \mathrm{}\mathrm{Tr}[(BA)๐][f(\omega \mu _\mathrm{L})+f(\omega \mu _\mathrm{R})].$$
(62)
$$\mathrm{j}_{n1}_2=\frac{1}{2\pi }\underset{๐ค_{}}{}d\omega \mathrm{}\mathrm{Tr}\left\{[g_\mathrm{L}tABg_\mathrm{R}^{}t^{}AB+\frac{1}{2}(A+B)]๐\right\}[f(\omega \mu _\mathrm{L})f(\omega \mu _\mathrm{R})].$$
(63)
Here, $`A=[1g_\mathrm{R}t^{}g_\mathrm{L}t]^1`$, $`B=[1g_\mathrm{R}^{}t^{}g_\mathrm{L}^{}t]^1`$, and as in Section II $`f(\omega \mu )`$ is the Fermi function with chemical potential $`\mu `$ and $`\mu _\mathrm{L}\mu _\mathrm{R}=eV_\mathrm{b}`$. In the linear-response case of small bias which we are considering, the Fermi functions in Eq. (63) are expanded to first order in $`V_\mathrm{b}`$. Hence the energy integral is avoided, being equivalent to multiplying the integrand by $`eV_\mathrm{b}`$ and evaluating it at the common zero-bias chemical potential $`\mu _0`$.
It can be seen that Eqs. (62) and (63), which determine the spin and the charge currents, depend on just two quantities, i.e. the surface retarded one-electron Green functions for a system cleaved between two neighbouring atomic planes. The surface Green functions can be determined without any approximations by the standard adlayer method (see e.g. 42 ; 44 ) for a fully realistic band structure.
We first note that there is a close correspondence between Eqs. (62), (63) and the generalised Landauer formula (12). The first term in Eq. (12) corresponds to the zero-bias spin current $`๐ฃ_{n1}_1`$ given by Eq. (62). When the cleavage plane is taken in the spacer, the spin current $`๐ฃ_{n1}_1`$ determines the oscillatory exchange coupling between the two magnets and it is easy to verify that the formula for the exchange coupling obtained from Eq. (62) is equivalent to the formula used in previous total energy calculations of this effect 42 ; 44 . The contribution to the transport spin current given by Eq. (63) clearly corresponds to the second term in the Landauer formula (12) which is proportional to the bias in the linear response limit. Placing the cleavage plane first between any two neighbouring atomic planes in the spacer and then between any two neighbouring planes in the lead, we obtain from Eq. (63) the total spin-transfer torque $`๐^{\mathrm{s}\mathrm{t}}`$ of Eq. (14) in Section II.
The equivalence of the Keldysh and Landauer methods has been demonstrated by calculating the currents (62) and (63) analytically for the simple single orbital model of Section II. The results of that section, such as Eq. (15) are reproduced 13 .
## V Quantitative results for Co/Cu/Co(111)
We now discuss the application of the Keldysh formalism to a real system. In particular we consider a realistic multiorbital model of fcc Co/Cu/Co(111) with tight-binding parameters fitted to the results of the first-principles band structure calculations, as described previously 42 ; 44 .
Referring to Fig. 1, the system considered by Edwards et al. 13 consists of a semi-infinite slab of Co (polarising magnet), the spacer of 20 atomic planes of Cu, the switching magnet containing $`M`$ atomic planes of Co, and the lead which is semi-infinite Cu. The spacer thickness of 20 atomic planes of Cu was chosen so that the contribution of the oscillatory exchange coupling term is so small that it can be neglected. The spin currents in the right lead and in the spacer were determined from Eq. (63). Figure 9(a),(b) shows the angular dependences of $`T_{}`$, $`T_{}`$ for the cases $`M=1`$ and $`M=2`$. respectively.
For the monolayer switching magnet, the torques $`T_{}`$ and $`T_{}`$ are equal in magnitude and they have opposite sign. However, for $`M=2`$, the torques have the same sign and $`T_{}`$ is somewhat smaller than $`T_{}`$. A negative sign of the ratio of the two torque components has important and unexpected consequences for hysteresis loops as already discussed in Section III. It can be seen that the angular dependence of both torque components is dominated by a $`\mathrm{sin}\psi `$ factor but distortions from this dependence are clearly visible. In particular, the slopes at $`\psi =0`$ and $`\psi =\pi `$ are quite different. As pointed out in Section III, this is important in the discussion of the stability of steady states and leads to quite different magnitudes of the critical biases $`V_\mathrm{P}V_{\mathrm{AP}}`$ and $`V_{\mathrm{AP}}V_\mathrm{P}`$.
In Fig. 10 we reproduce the dependence of $`T_{}`$ and $`T_{}`$ on the thickness of the Co switching magnet. It can be seen that the out-of-plane torque $`T_{}`$ becomes smaller than $`T_{}`$ for thicker switching magnets.
However, $`T_{}`$ is by no means negligible (27$`\%`$ of $`T_{}`$) even for a typical experimental thickness of the switching Co layer of ten atomic planes. It is also interesting that beyond the monolayer thickness, the ratio of the two torques is positive with the exception of $`M=4`$.
The microscopically calculated spin-transfer torques for Co/Cu/Co(111) were used by Edwards et al. 13 as an input into the phenomenological LLG equation. For simplicity the torques as functions of $`\psi `$ were approximated by sine curves but this is not essential. The LLG equation was first solved numerically to determine all the steady states and then the stability discussion outlined in the phenomenological section was applied to determine the critical bias for which instabilities occur. Finally, the ballistic resistance of the structure was evaluated from the real-space Kubo formula at every point of the steady state path. Such a calculation for the realistic Co/Cu system then gives hysteresis loops of the resistance versus bias which can be compared with the observed hysteresis loops. The LLG equation was solved including a strong easy-plane anisotropy with $`h_\mathrm{p}=100`$. If we take $`H_{\mathrm{u0}}=1.86\times 10^9`$sec<sup>-1</sup>, corresponding to a uniaxial anisotropy field of about 0.01T, this value of $`h_\mathrm{p}`$ corresponds to the shape anisotropy for a magnetisation of $`1.6\times 10^6`$A/m, similar to that of Co sun . Also a realistic value sun of the Gilbert damping parameter $`\gamma =0.01`$ was used. Finally, referring to the geometry of Fig. 1, two different values of the angle $`\theta `$ were employed in these calculations: $`\theta =2`$rad and $`\theta =3`$rad, the latter value being close to the value of $`\pi `$ which is realised in most experiments.
We first reproduce in Fig. 11 the hysteresis loops for the case of Co switching magnet consisting of two atomic planes. We note that the ratio $`r=T_{}/T_{}0.65`$ deduced from Fig. 9 is positive in this case. Fig. 11(a) shows the hysteresis loop for $`\theta =2`$ and Fig. 11(b) that for $`\theta =3`$.
The hysteresis loop for $`\theta =3`$ shown in Fig. 11(b) is an illustration of the stability scenario in zero applied field with $`r>0`$ discussed in Section III. As pointed out there the hysteresis curve is that of Fig. 5(a) which agrees with Fig. 11(b) when we remember that the reduced bias used in Fig. 5 has the opposite sign from the bias in volts used in Fig. 11. It is rather interesting that the critical bias for switching is $`0.2`$mV both for $`\theta =2`$ and $`\theta =3`$. When this bias is converted to the current density using the calculated ballistic resistance of the junction, it is found 13 that the critical current for switching is $`10^7`$A/cm<sup>2</sup>, which is in very good agreement with experiments albert .
The hysteresis loops for the case of the Co switching magnet consisting of a single atomic plane are reproduced in Fig. 12. The values of $`h_\mathrm{p}`$, $`\gamma `$, $`H_{\mathrm{u0}}`$, and $`\theta `$ are the same as in the previous example.
However the ratio $`r1`$ is now negative and the hysteresis loops in Fig. 12 illustrate the interesting behaviour discussed in Section III when the system subjected to a bias higher than a critical bias moves to the โboth unstableโ region shown in Fig. 6. As in Fig. 7 the points on the hysteresis loop in Fig. 12 corresponding to the critical bias are labelled by asterisks. Fig. 12(b) and Fig. 7(a) are in close correspondence because Fig. 7(a) is for $`r_\mathrm{c}<r<0`$ and in the present case $`r=1`$, $`r_\mathrm{c}=2/(\gamma h_\mathrm{p})=2`$. Also, from Fig. 9(a), $`g_{}<0`$ so that $`vg_{}`$ in Fig. 7(a) has the same sign as the voltage $`V`$ in Fig. 12(b).
## VI Summary
Spin-transfer torque is responsible for current-driven switching of magnetisation in magnetic layered structures. The simplest theoretical scheme for calculating spin-transfer torque is a generalised Landauer method and this is used in Section II to obtain analytical results for a simple model. The general phenomenological form of spin-transfer torque is deduced in Section III and this is introduced into the Landau-Lifshitz-Gilbert equation, together with torques due to anisotropy fields. This describes the motion of the magnetisation of the switching magnet and the stability of the steady states (constant current and stationary magnetisation direction) is studied under different experimental conditions, with and without external field. This leads to hysteretic and reversible behaviour in resistance versus bias (or current) plots in agreement with a wide range of experimental observations. In Section IV the general principles of a self-consistent treatment of spin-transfer torque are discussed and the Keldysh formalism for quantitative calculations is introduced. This approach to the non-equilibrium problem of electron transport uses Green functions which are very convenient to calculate for a realistic multiorbital tight-binding model of the layered-structure. In Section V quantitative calculations for Co/Cu/Co(111) systems are presented which yield switching currents of the observed magnitude.
This study of current-driven switching of magnetisation was carried out in collaboration with J. Mathon and A. Umerski and financial support was provided by the UK Engineering and Physical Research Council (EPSRC).
|
warning/0506/cond-mat0506437.html
|
ar5iv
|
text
|
# Superfluid stability in BEC-BCS crossover
(June 13, 2005)
## Abstract
We consider a dilute atomic gas of two species of fermions with unequal concentrations under a Feshbach resonance. We find that the system can have distinct properties due to the unbound fermions. The uniform state is stable only when either (a) beyond a critical coupling strength, where it is a gapless superfluid, or (b) when the coupling strength is sufficiently weak, where it is a normal Fermi gas mixture. Phase transition(s) must therefore occur when the resonance is crossed.
Feshbach resonance Feshbach has opened up a new playground for the field of cold trapped atoms. Using this resonance, the effective interaction between the atoms can be varied over a wide range. In particular, for two fermion species with a Feshbach resonance between them, the ground state can be tuned from a weak-coupling Bardeen Cooper Schrieffer (BCS) superfluid to a strong coupling regime where the Fermions pair-up to form Bosons which in turn undergo Bose Einstein Condensation (BEC) EL ; SRE93 .
Though this problem has been under intense theoretical theo and experimental expr investigations, almost all works thus far are restricted to the case where the concentrations of the two fermionic species are equal. We here generalize this study to the case of unequal populations of the two species, and investigate in detail the thermodynamic stability of this system, in particular the question when the uniform state can be stable.
Studies of fermions with unequal populations or mismatched Fermi surfaces and a pairing interaction have a long history. It was studied by Fulde and Ferrell, Larkin and Ovchinnikov (FFLO) FFLO in the 1960โs with relation to superconductivity in materials with ferromagnetically coupled paramagnetic impurities. It was found that in this case the system is likely to have an inhomogeneous gapless superconducting phase. Advances in techniques of manipulating dilute ultracold atoms have revived interests in the related problems FFLOn . These studies, in our present language, are still restricted to the weak-coupling regime. We, however, would extend our analysis to all coupling strengths.
In the โcanonicalโ problem of two species of fermions with equal mass (say, spin up and spin down electrons) and equal concentrations (thus a single Fermi surface), if the cross-species interaction is varied from weak to strong coupling, at low temperatures the system would undergo a smooth crossover from a superfluid with loosely bound Cooper pairs (the โBCSโ limit) to one with condensation of tightly bound bosonic molecules (the โBECโ limit) EL . The situation, however, can be very different if one considers two species of fermions with unequal concentrations (i.e. mismatched Fermi surfaces), even if they have identical mass. This can be anticipated because, on the one hand, far into the BCS side, the system is basically in the FFLO regime FFLOn and therefore must go into a spatially inhomogeneous phase. On the other hand, in the far end of the BEC side, the system is expected to behave like an ordinary (weakly interacting) Bose-Fermi mixture and thus has a stable homogeneous phase. Here the bosons are the fermion pairs and the fermions are the โleftoverโ unpaired atoms of the majority species. Deep into the BEC regime, the size of the Fermion pairs is small and the interaction between the bosons and the leftover unpaired fermions are expected to be weak. It is therefore a very interesting question as to what happens in between. This is the question we want to address in this paper.
For simplicity, we shall assume that the resonance is sufficiently wide that the physics reduces effectively to a single channel regime. This is probably valid wide for many Feshbach resonances under current experimental investigations. Thus, in our calculations, we would not invoke explicitly the presence of the โclosed channelโ which leads to this Feshbach resonance. We simply model the fermions as interacting through a short-range, s-wave effective interaction (dependent on the external magnetic field) characterized by the corresponding scattering length $`a`$. $`1/a`$ varies from $`\mathrm{}`$ for large negative detuning (closed channel bound state energy much below continuum threshold) to $`\mathrm{}`$ for large positive detuning.
Now we proceed to the details of our calculation and results. We consider two fermion species, denoted as โspinโ $``$ and $``$, of equal mass $`m`$. Because of the unequal concentrations of the two species and the possible existence of pairing, it is useful to introduce three fields: the chemical potentials $`\mu _\sigma `$ ($`\sigma =`$ or $``$) and the pairing field $`\mathrm{\Delta }`$. We shall confine ourselves to zero temperature and generalize the BCS mean field approach of SRE93 . The excitation spectrum for each spin is (see e.g. WY03 for details)
$$E_\sigma (๐ค)=\frac{\xi _\sigma (๐ค)\xi _\sigma (๐ค)}{2}+\sqrt{\left(\frac{\xi _\sigma (๐ค)+\xi _\sigma (๐ค)}{2}\right)^2+\mathrm{\Delta }^2},$$
(1)
where $`\xi _\sigma (๐ค)=\mathrm{}^2k^2/2m\mu _\sigma `$ are the quasi-particle excitation energies for normal fermions, and $``$. The density of each spin species is then
$$n_\sigma =\frac{d^3k}{(2\pi )^3}\left[u_๐ค^2f(E_\sigma )+v_๐ค^2f(E_\sigma )\right],$$
(2)
with the coherence factors $`u_๐ค^2=1v_๐ค^2={\displaystyle \frac{E_{}(k)+\xi _{}(k)}{E_{}(k)+E_{}(k)}}`$. Here $`f`$ is the Fermi function. The equation for the order parameter $`\mathrm{\Delta }`$ reads:
$$\frac{m}{4\pi a}\mathrm{\Delta }=\mathrm{\Delta }\frac{d^3k}{(2\pi )^3}\left[\frac{1f(E_{})f(E_{})}{E_{}+E_{}}\frac{m}{\mathrm{}^2k^2}\right].$$
(3)
We solve equations (2) and (3) self-consistently for fixed total density $`nn_{}+n_{}`$ and density difference $`n_dn_{}n_{}`$. We shall always take $``$ to be the majority species so that $`n_d0`$.
It is convenient to introduce the average chemical potential $`\mu (\mu _{}+\mu _{})/2`$ and the difference $`h(\mu _{}\mu _{})/2\mathrm{\hspace{0.17em}0}`$. Then we have
$$E_,(๐ค)=\sqrt{\xi (๐ค)^2+\mathrm{\Delta }^2}h,$$
(4)
where $`\xi (๐ค)\mathrm{}^2k^2/2m\mu `$. Hence $`E_{}(k)>0`$ always. From Eq. (2) we get
$$n_d=\frac{d^3k}{(2\pi )^3}f(E_{}(๐ค))$$
(5)
and so the integration is only over the region where $`E_{}(๐ค)<0`$. In the following, it is useful to note that the smallest (or most negative) $`E_{}(๐ค)`$ occurs at $`\xi (๐ค)=0`$ for $`\mu >0`$, where it is $`\mathrm{\Delta }h`$, and at $`k=0`$ for $`\mu <0`$, where it is $`\sqrt{\mu ^2+\mathrm{\Delta }^2}h`$.
As in the case of equal concentrations, it is convenient to express our results in dimensionless variables. We shall define an inverse length scale $`k_F`$ through the total density $`n`$ via $`k_F(3\pi ^2n)^{1/3}`$, and an energy scale $`ฯต_F\mathrm{}^2k_F^2/2m`$. We thus write $`\stackrel{~}{\mu }\mu /ฯต_F`$, $`\stackrel{~}{h}h/ฯต_F`$, $`\stackrel{~}{\mathrm{\Delta }}\mathrm{\Delta }/ฯต_F`$, $`\stackrel{~}{n}_dn_d/n`$, and define the dimensionless coupling constant $`g1/(\pi k_Fa)`$, which varies from $`\mathrm{}`$ for large negative detuning to $`\mathrm{}`$ for large positive detuning.
We now describe the results of our calculations. We first make contact with the BCS-BEC cross-over for equal concentrations. The inset of Fig. 1 shows the typical behaviors of $`\stackrel{~}{\mu }`$, $`\stackrel{~}{\mathrm{\Delta }}`$ and $`\stackrel{~}{h}`$ as a function of $`g`$ for a given density difference $`\stackrel{~}{n}_d`$. The behavior of $`\stackrel{~}{\mu }`$ or $`\stackrel{~}{\mathrm{\Delta }}`$ is similar to that in the case of equal concentrations SRE93 . For example, $`\stackrel{~}{\mu }`$ is large and negative in the BEC limit whereas it is of order $`1`$ in the BCS regime. Unlike that case however, $`g`$ has to be larger than a minimum coupling $`g_c`$ in order for a finite order parameter $`\mathrm{\Delta }`$ to exist. For $`g<g_c`$, Eq. (3) requires that $`\mathrm{\Delta }=0`$ and the system is in the normal state. (For clarity of this inset, we plot only the $`\mathrm{\Delta }0`$ solutions.) The main Fig. 1 shows $`\stackrel{~}{h}`$ as function of $`g`$ in the intermediate regime ($`|g|1`$) for three different $`\stackrel{~}{n}_d`$ (0.2, 0.5, and 0.8). The horizontal dotted lines indicate the normal state in which the gap function $`\mathrm{\Delta }`$ is zero (described above).
The behavior for $`g<g_c`$ is easy to understand. For sufficently large and negative $`g`$, the interaction is too weak to produce pairing since the concentrations are unequal. The system reduces to a Fermi gas. In this case, the chemical potentials are given by $`\mu _\sigma =(6\pi ^2n_\sigma )^{2/3}/(2m)`$ which implies $`\mu =[(6\pi ^2)^{2/3}/4m](n_{}^{2/3}+n_{}^{2/3})`$ and $`h=[(6\pi ^2)^{2/3}/4m)](n_{}^{2/3}n_{}^{2/3})`$ (both independent of $`g`$). On the other hand, for large and positive $`g`$ (the strong-coupling BEC limit), one can show Yip02 from Eqs. (2), (3) and (5) that both $`h`$ and $`|\mu |`$ are large and to leading order given by $`\mathrm{}^2/(2ma^2)`$. However, $`(\mu +h)/2`$ $`=\mu _{}`$ $`=(6\pi ^2n_d)^{2/3}/(2m)`$ $`|\mu |`$ or $`h`$. These expressions simply reflect that the system becomes a Bose-Fermi mixture with boson concentration $`n_{}`$ and free fermion concentration $`n_d`$.
Notice that the lines (for $`\mathrm{\Delta }0`$) of $`\stackrel{~}{h}`$ versus $`g`$ cross each other from small to large $`\stackrel{~}{n}_d`$ near $`g0.15`$. For $`g0.17`$, $`\stackrel{~}{h}`$ increases with $`\stackrel{~}{n}_d`$ for fixed $`g`$. For $`g0.15`$, $`\stackrel{~}{h}`$ decreases as $`\stackrel{~}{n}_d`$ increases when the coupling strength is fixed. We shall return to these features again below.
Now we make contact with the superconductivity literature. It is helpful here to note that $`h`$ plays the role of an effective external Zeeman field. We plot in Fig. 2 $`\stackrel{~}{\mathrm{\Delta }}`$ as a function of $`\stackrel{~}{h}`$ for various coupling strengths $`g`$. The horizontal portion of each curve corresponds to $`n_d=0`$. In this region, $`h<\mathrm{\Delta }`$ and so that $`E_{}(๐ค)>0`$ for all $`๐ค`$ (see Eq. (5)). The other part of the curve corresponds to $`n_d>0`$, and exists only in the region $`h>\mathrm{\Delta }`$ (More precisely, for larger $`g`$ where $`\mu `$ becomes negative, this condition should read $`h>\sqrt{\mu ^2+\mathrm{\Delta }^2}`$). For small $`g`$ ($`0.1`$), $`\stackrel{~}{\mathrm{\Delta }}`$ decreases with decreasing $`\stackrel{~}{h}`$. This solution is the generalization of that first discovered by Sarma Sarma . We find that this superfluid state corresponds to one where $`\stackrel{~}{n}_d`$ increases with decreasing $`\stackrel{~}{h}`$, and hence unstable (to be discussed again below). For sufficiently large coupling ($`g0.17`$), $`\stackrel{~}{\mathrm{\Delta }}`$ decreases with increasing $`\stackrel{~}{h}`$. This state has $`\stackrel{~}{n}_d`$ increases with $`\stackrel{~}{h}`$, and satisfies one of the stability conditions to be discussed below.
In Fig. 3 the chemical potential difference $`\stackrel{~}{h}`$ is plotted as a function of $`\stackrel{~}{n}_d`$. These results correspond to those of Fig. 1 presented in a different manner. Let us explain how this graph should be read, with $`g=0.1`$ as an example. For $`\stackrel{~}{n}_d=0`$, $`\stackrel{~}{h}`$ can take any value up to $`\stackrel{~}{h}_10.5`$ given by the intersection of the line labelled by $`g=0.1`$ with the $`\stackrel{~}{n}_d=0`$ axis. This portion corresponds to the line with $`\mathrm{\Delta }`$ being a constant in Fig 2. For $`0<\stackrel{~}{n}_d<0.46`$ the dependence of $`\stackrel{~}{h}`$ on $`\stackrel{~}{n}_d`$ is given by the solid line labelled by $`g=0.1`$. This line corresponds to the state with $`\mathrm{\Delta }0`$ but $`\mathrm{\Delta }<h`$ in Fig. 2 discussed above. For $`\stackrel{~}{n}_d>0.46`$, the system enters into the normal state with the $`(\stackrel{~}{h},\stackrel{~}{n}_d)`$ relation represented by the dotted lines, given by $`\stackrel{~}{h}=\mathrm{\hspace{0.17em}0.5}[(1+\stackrel{~}{n}_d)^{2/3}(1\stackrel{~}{n}_d)^{2/3}]`$ (see the discussion on Fig. 1 above). Lastly, for $`\stackrel{~}{n}_d=1`$, $`\stackrel{~}{h}`$ can take any value larger than $`\stackrel{~}{h}_20.5\times 2^{2/3}0.79`$. This is because this line corresponds simply to a Fermi gas with only $``$ particles, and $`h`$ can take any value larger than $`\mu `$ so that $`\mu _{}=(\mu h)/2<0`$. For $`g0.1`$, the graph can be read in a similar manner except that the dotted line representing the normal state is not involved.
For the uniform superfluid to be stable, two criterions must be fulfilled WY03 ; Bedaque . First, the susceptibilities matrix $`n_\sigma /\mu _\sigma ^{}`$ can have only positive eigenvalues. One can show that this requires that $`\stackrel{~}{n}_d/\stackrel{~}{h}`$, evaluated at constant $`g`$, must be positive matrix . That is, the plot of $`\stackrel{~}{h}`$ versus $`\stackrel{~}{n}_d`$ must have positive slope. From Fig. 3, we see that for small $`g`$ ($`0.1`$), the slope of each curve is always negative which indicates the superfluid state is unstable. For $`g`$ greater than about 0.1, the slope of these curves change sign at some $`\stackrel{~}{n}_d`$ between $`0`$ and $`1`$. In this case a stable superfluid state may occur for sufficiently large $`\stackrel{~}{n}_d`$. After $`g0.17`$, the slopes of these curves are positive for all $`n_d0`$ and the above stability criterion is satisfied for all $`n_d`$.
The second stability criterion is that the superfluid density $`\rho _s`$ must be positive WY03 . $`\rho _s`$ can be evaluated as discussed in WY03 . The analytic result can be expressed as
$$\frac{\rho _s}{n}=\left[1\frac{\theta (\sqrt{\mu ^2+\mathrm{\Delta }^2}h)\stackrel{~}{k}_1^3+\stackrel{~}{k}_2^3}{2\sqrt{1(\mathrm{\Delta }/h)^2}}\right],$$
(6)
with $`\stackrel{~}{k}_{1,2}=[(\stackrel{~}{\mu }\sqrt{\stackrel{~}{h}^2\stackrel{~}{\mathrm{\Delta }}^2})]^{1/2}`$ and $`\theta (x)`$ is the step function. The line $`\rho _s=0`$ is plotted as the dashed lines in Fig. 3, with $`\rho _s<0`$ below and $`\rho _s>0`$ above. Thus the states correspond to $`\mathrm{\Delta }0`$ below this dashed line are all unstable. $`\rho _s<0`$ indicates that the system is unstable towards a state with spatially varying $`\mathrm{\Delta }`$ and hence a state such as FFLO can be more preferable. From our results, it turns out that the condition for $`\rho _s>0`$ is actually a slightly weaker requirement than the positive susceptibility discussed in the last paragraph (though we are not aware of any reason why it must be so).
We here note that, for $`\stackrel{~}{n}_d0`$, the location where $`\rho _s`$ changes sign is exactly at $`\mu =0`$. Though this can be seen from Eq. (6), it is more instructive to return to the more basic equation for the superfluid density: $`\rho _s=n+\rho _p`$. Here $`\rho _p`$, the paramagnetic response, is related to the backflow of quasiparticles and is given by WY03 $`\rho _p=\frac{1}{6\pi ^2m}_0^{\mathrm{}}๐kk^4\delta (E_{}(k))`$. For $`\mu <0`$, the solution to $`E_{}(k)=0`$ exists only when $`h>\sqrt{\mu ^2+\mathrm{\Delta }^2}`$ and takes place at $`k=k_2`$ with $`k_2^2/2m=\sqrt{h^2\mathrm{\Delta }^2}+\mu `$. For $`n_d0^+`$, $`h`$ is just slightly larger than $`\sqrt{\mu ^2+\mathrm{\Delta }^2}`$ \[see Eq. (5)\]. $`k_2`$ is small and hence $`\rho _p0^+`$ and $`\rho _sn>0`$. For $`\mu >0`$, $`E_{}(k)=0`$ happens when $`h>\mathrm{\Delta }`$ and take place at two $`k`$ values: $`k=k_2`$ as already given in the $`\mu <0`$ case above and $`k=k_1`$ with $`k_1^2/2m=\sqrt{h^2\mathrm{\Delta }^2}+\mu `$. For $`n_d0^+`$, $`h`$ is just slightly larger than $`\mathrm{\Delta }`$ and the $`E_{}(k)=0`$ points occur near $`\xi (k)0`$ hence $`E_{}(k)/k0`$. Since $`k_1`$ and $`k_2`$ are finite, $`\rho _p\mathrm{}`$ and $`\rho _s<0`$.
Finally we show our phase diagram in Fig. 4 covering the entire BEC to BCS regimes. On the BEC side, with $`g0.17`$, the superfluid state is stable in which both the slope of $`\stackrel{~}{h}(\stackrel{~}{n}_d)`$ and the superfluid density are positive (see Fig. 3). On the upper right of Fig. 4, the pairing order parameter $`\mathrm{\Delta }`$ is zero and the system is in the normal state. In constructing the boundary of this phase, we have simply taken the intersection of the full lines in Fig. 3 with the dotted lines. A more correct approach would involve the Maxwell construction. However, to do this we also need to know the equation of state for the non-uniform FFLO superfluid state beyond the weak-coupling regime. Since this information is not yet available, we shall leave this investigation to the future.
Lastly we discuss the โbreached gapโ state of Liu and Wilczek LW . For this state to exist, one need $`E_{}(๐ค)<0`$ for a region of $`k`$ that lies between $`k_1<k<k_2`$ where $`k_1>0`$. ($`k_1`$, $`k_2`$ already given in the paragraph before last). This is possible only if $`E_{}(๐ค)`$ is not monotonic with $`k`$ and hence $`\mu >0`$. Moreover, since $`E_{}(0)>0`$, we have $`\sqrt{\mu ^2+\mathrm{\Delta }^2}>h`$ yet $`h>\mathrm{\Delta }`$. Though there is a region of stability with $`\mu >0`$ (the region near the upper right of Fig. 3), we find that it rather corresponds to $`\sqrt{\mu ^2+\mathrm{\Delta }^2}<h`$. (That this is the case near $`\stackrel{~}{n}_d1`$ is obvious, since the line $`\stackrel{~}{n}_d=1`$ corresponds to $`h>\mu `$ and $`\mathrm{\Delta }0`$. Note also on the dash-dotted lines where $`\mu =0`$, we also have $`\sqrt{\mu ^2+\mathrm{\Delta }^2}<h`$ since $`h>\mathrm{\Delta }`$). Therefore gapless excitations exist only near $`k=k_2`$. Moreover this region has $`n_{}(k)=1`$ for $`0kk_2`$ (and $`n_{}(k)=n_{}(k)=v^2(k)<1`$ for $`k>k_2`$), thus the leftover unpaired majority spin-up particles form a rather normal Fermi sphere with radius $`k_2`$, although the energy required to create a hole $`E_{}(k)`$ can actually be non-monotonic as a function of $`k`$, being maximum at an intermediate value $`k=\sqrt{2m\mu }`$ (where it is $`h\mathrm{\Delta }`$) but not $`k=0`$ (where it is $`h\sqrt{\mu ^2+\mathrm{\Delta }^2}`$).
In conclusion, we have investigated the stability of a fermion mixture with unequal concentrations under a Feshbach resonance. We show that, in contrast to the case of equal concentrations, there is no smooth BCS-BEC crossover. The system is a uniform stable superfluid Bose-Fermi mixture only for sufficiently large coupling. For weak interactions the normal state is the only stable uniform state. The uniform state is unstable for intermediate coupling strengths. Phase transitions must occur when the Feshbach resonance is varied between large positive detuning and large negative detuning.
This research was supported by the National Science Council of Taiwan under grant numbers NSC93-2112-M-194-002 (CHP), NSC93-2112-M-194-019 (STW) and NSC93-2112-M-001-016 (SKY), with additional support from National Center for Theoretical Sciences, Hsinchu, Taiwan.
|
warning/0506/math0506283.html
|
ar5iv
|
text
|
# (1.5) Proposition
The Diagonal Distribution for the Invariant Measure
of a Unitary Type Symmetric Space
Doug Pickrell
Mathematics Department
University of Arizona
Tucson, Arizona 85721
pickrellmath.arizona.edu
Abstract. Let $`\mathrm{\Theta }`$ denote an involution for a simply connected compact Lie group $`U`$, let $`K`$ denote the fixed point set, and let $`\mu `$ denote the $`U`$-invariant probability measure on $`U/K`$. Consider the geodesic embedding $`\varphi :U/KU:uuu^\mathrm{\Theta }`$ of Cartan. In this paper we compute the Fourier transform of the diagonal distribution for $`\varphi _{}\mu `$, relative to a compatible triangular decomposition of $`G`$, the complexification of $`U`$. This boils down to a Duistermaat-Heckman exact stationary phase calculation, involving a Poisson structure on the dual symmetric space $`G_0/K`$ discovered by Evens and Lu.
ยง0. Introduction.
Suppose that $`K`$ is a simply connected compact Lie group, and let $`G`$ denote the complexification. Given a triangular decomposition $`๐ค=๐ซ^{}๐ฅ๐ซ^+`$, a generic $`gK`$ has a unique โLDU decompositionโ, $`g=lmau`$, where $`lN^{}`$ (lower triangular), $`uN^+`$ (upper triangular), and $`maH`$, with $`mHK`$ (unitary) and $`aexp(๐ฅ_{}=๐ฅi๐จ)`$ (positive). A formula of Harish-Chandra (essentially equivalent to the Weyl dimension formula) asserts that for $`\lambda ๐ฅ_{}^{}`$
$$_Ka^{i\lambda }=๐(2\delta i\lambda )=\underset{\alpha >0}{}\frac{2\delta ,\alpha }{2\delta i\lambda ,\alpha },$$
$`0.1`$
where the integral is with respect to normalized Haar measure, the product is over positive complex roots, and $`2\delta `$ is the sum of the positive complex roots (there are various ways in which the $`๐`$-function arises, and this formula has many extensions and interpretations; see e.g. \[H2\], especially ยง5-6 of chapter IV).
The purpose of this paper is to present a generalization of this formula, and some of the related geometry, in which $`K`$ is replaced by a compact symmetric space.
Suppose that $`X`$ is a simply connected compact symmetric space with a fixed basepoint. From this we obtain (1) a diagram of groups,
$$\begin{array}{ccc}& & G\\ & & & \\ G_0& & & & U\\ & & & \\ & & K\end{array},$$
$`0.2`$
where $`U`$ is the universal covering of the identity component of $`Aut(X)`$, $`XU/K`$, $`G`$ is the complexification of $`U`$, and $`G_0/K`$ is the noncompact type symmetric space dual to $`X`$; and (2) a diagram of equivariant totally geodesic (Cartan) embeddings of symmetric spaces:
$$\begin{array}{ccc}U/K& \mathrm{@}>\varphi >>& U\\ & & \\ G/G_0& \mathrm{@}>\varphi >>& G& \genfrac{}{}{0pt}{}{\psi }{}& G/U\\ & & & & \\ & & G_0& \genfrac{}{}{0pt}{}{\psi }{}& G_0/K\end{array}.$$
$`0.3`$
We also consider one additional ingredient: a triangular decomposition of $`๐ค`$, $`๐ค=๐ซ^{}๐ฅ๐ซ^+`$, which is $`\mathrm{\Theta }`$-stable and for which $`๐ฑ_0=๐ฅ๐จ`$ is maximal abelian in $`๐จ`$, where $`\mathrm{\Theta }`$ is the involution corresponding to the pair $`(U,K)`$.
Given this triangular structure, a generic element of $`\varphi (U/K)`$ can be written as $`g=l๐จa_\varphi ml^\mathrm{\Theta }`$, where $`l`$, $`a_\varphi `$, $`m`$ are roughly as before, and $`๐จT_0^{(2)}`$ (elements of order two); the possible $`๐จ`$ index the connected components of the set of generic elements. In the special case in which $`\mathrm{\Theta }`$ is an inner automorphism, the generalization of $`(0.1)`$ which we consider is of the form
$$_{\varphi (U/K)}a_\varphi ^{i\lambda }=\frac{|W(K)|}{|W(U)|}\underset{๐จ}{}\stackrel{๐จ}{}\frac{\delta ,\alpha }{\delta i\lambda ,\alpha }$$
$`0.4`$
where, given $`๐จ`$, the product is over positive roots $`\alpha `$ which are of noncompact type for the involution $`Ad(๐จ)\mathrm{\Theta }`$, and $`|W()|`$ denotes the order of the Weyl group.
The plan of the paper is the following. In ยง1 we compute the intersections of the $`\varphi `$-images in $`(0.3)`$ with the triangular decomposition of $`G`$. A notable qualitative fact is that just as the map $`U/KG/G_0`$ in $`(0.3)`$ is a homotopy equivalence, so also are the intersections with the triangular decomposition of $`G`$. Possibly everything in this section is known; it can certainly be generalized and packaged in various ways (the canonical source is \[Wolf\]).
In ยง2 the general formulation and a proof of $`(0.4)`$ is presented. It turns out that, in the inner case for example, the $`๐จ=1`$ term in $`(0.4)`$ equals
$$_{G_0/K}๐(g_0)^{2\delta 2(\delta i\lambda )}๐V(g_0K)=_{\psi (G_0/K)}a_\psi ^{\delta (\delta i\lambda )}$$
$`0.5`$
where $`g_0=๐๐๐ฆ`$ is an Iwasawa decomposition in $`G`$, $`\psi (g_0K)=g_0g_0^{}=๐a_\psi ๐^{}`$ is an LDU decomposition, and the integrals are with respect to a $`G_0`$-invariant measure. It is remarkable, although not a surprise, that $`๐(g_0)^{2\delta }dV(g_0K)`$ is the volume element for a symplectic form having a momentum map $`log(๐(g_0K))`$. Hence $`(0.5)`$ can be evaluated using (a noncompact version of) the Duistermaat-Heckman exact stationary phase method. The symplectic structure was discovered by Evens and Lu, in a general setting (\[EL\]); the relevance of this structure was pointed out to me by Foth and Otto (\[FO\]), to whom I am grateful.
It is natural to consider the more general integral
$$\mathrm{\Psi }(g)=_{G_0}๐(g_0g)^{2\delta 2(\delta i\lambda )}๐g_0$$
$`0.6`$
for $`gG_0\backslash G/U`$. This is an eigenfunction for $`G`$-invariant differential operators on $`G/U`$. This can also be evaluated exactly, by the same method.
In \[Pi1,2\] I have discussed conjectural generalizations of $`(0.1)`$ and $`(0.4)`$ to loop spaces, and other kinds of infinite symmetric spaces. The localization argument applies in a heuristic way. In appendix A there is a proof of $`(0.1)`$, involving an explicit factorization of the integral, which has elements that seem useful in the loop space context.
Notation. $`,`$ will denote the Killing form for $`๐ค`$. For an automorphism $`\theta `$ of $`๐ค`$, we will often write $`\theta (x)=x^\theta `$, and more briefly, $`Ad(g)(x)=x^g`$. We will write $`x=x_{}+x_๐ฅ+x_+`$ for the triangular decomposition of $`x๐ค`$, and $`x=x_๐จ+x_๐ญ`$ for the Cartan decomposition of $`x๐ค_0`$.
ยง1. Symmetric Spaces and Triangular Decomposition.
Throughout the remainder of this paper, $`U`$ will denote a simply connected compact Lie group, $`\mathrm{\Theta }`$ will denote an involution of $`U`$, with fixed point set $`K`$, and $`X`$ will denote the quotient, $`U/K`$. This implies that $`K`$ is connected and $`X`$ is simply connected (Theorem 8.2 of \[H1\]).
Corresponding to the diagram of groups in $`(0.1)`$, there is a Lie algebra diagram
$$\begin{array}{ccccc}& & ๐ค=๐ฒi๐ฒ& & \\ & & & & & \\ ๐ค_0=๐จ๐ญ& & & & ๐ฒ=๐จi๐ญ\\ & & & \\ & & ๐จ\end{array}$$
$`1.1`$
where $`\mathrm{\Theta }`$, acting on the Lie algebra level and extended complex linearly to $`๐ค`$, is $`+1`$ on $`๐จ`$ and $`1`$ on $`๐ญ`$. We let $`()^{}`$ denote the Cartan involution for the pair $`(G,U)`$. The Cartan involution for the pair $`(G,G_0)`$ is given by $`\sigma (g)=g^\mathrm{\Theta }`$. Since $``$, $`\mathrm{\Theta }`$, $`\sigma `$, and $`()^1`$ commute, our practice of writing $`g^\mathrm{\Theta }`$ for $`\mathrm{\Theta }(g)`$, etc, should not cause any confusion.
We have natural maps
$$\begin{array}{ccccc}K& & U& & U/K\\ & & & & \\ G_0& & G& & G/G_0\end{array}.$$
$`1.2`$
The vertical arrows (given by inclusion) are homotopy equivalences; more precisely, there are diffeomorphisms (polar or Cartan decompositions)
$$K\times ๐ญG_0,U\times i๐ฒG,U\times _Ki๐จG/G_0,$$
$`1.3`$
in each case given by the formula $`(g,X)gexp(X)`$ (mod $`G_0`$ in the last case). In turn there are totally geodesic embeddings of symmetric spaces
$$\begin{array}{ccccccc}U/K& \mathrm{@}>\varphi >>& U& :& gK& & gg^\mathrm{\Theta }\\ & & \\ G/G_0& \mathrm{@}>\varphi >>& G& :& gG_0& & gg^\mathrm{\Theta }=gg^\sigma \end{array},$$
$`1.4`$
where the symmetric space structures are derived from the Killing form.
A group element of the form $`g=g_1g_1^\sigma `$ satisfies the equation $`g^{}=g^\mathrm{\Theta }`$ (i.e. $`gg^\sigma =1`$); $`g^{}=g^\mathrm{\Theta }`$ implies that $`Ad(g)\sigma `$ is an antilinear involution; and if $`g=g_1g_1^\sigma `$, then $`Ad(g)\sigma =Ad(g_1)\sigma Ad(g_1^1)`$, hence $`\sigma `$ and $`Ad(g)\sigma `$ are inner conjugate. These considerations lead to the following well-known
###### (1.5) Proposition
(a) In terms of $`gG`$,
$$\begin{array}{ccc}\varphi (U/K)=\{g^1=g^{}=g^\mathrm{\Theta }\}_0& & U=\{g^1=g^{}\}\\ & & \\ \varphi (G/G_0)=\{g^{}=g^\mathrm{\Theta }\}_0& & G\end{array}$$
where $`\{\}_0`$ denotes the connected component containing the identity.
(b) The connected components of $`\{g^1=g^{}=g^\mathrm{\Theta }\}`$ are determined by the map which sends $`g`$ to the inner conjugacy class of the involution $`\eta =Ad(g)\mathrm{\Theta }`$, subject to the constraint that $`\eta `$ equals $`\mathrm{\Theta }`$ in $`Out(U)=Ad(U)\backslash Aut(U)`$. A similar statement applies to $`\{g^{}=g^\mathrm{\Theta }\}`$, with $`\sigma `$ and antilinear automorphisms of $`G`$ in place of $`\mathrm{\Theta }`$ and involutions of $`K`$.
###### Demonstration Proof of (1.5)
We first recall why $`\{gg^\sigma =1\}`$ is smooth.
Consider the map $`\psi :GG:ggg^\sigma `$. If we use right translation to identify the tangent space at any point of $`G`$ with $`๐ค`$, the derivative at $`g`$ is given by $`xx+Ad(g)[\sigma (x)]`$. Thus $`ker(d\psi |_g)`$ is identified with the $`1`$ eigenspace of $`Ad(g)\sigma `$ acting on $`๐ค`$.
Now suppose $`gg^\sigma =1`$. Since $`Ad(g)\sigma `$ is an involution, the spectrum of $`Ad(g)\sigma `$ is fixed. Thus the dimension of the $`1`$ eigenspace of $`Ad(g)\sigma `$ is constant on $`\{gg^\sigma =1\}`$. It follows that $`\psi `$ has constant rank on the connected components of $`\psi ^1(1)`$. Since $`\psi `$ is an algebraic map, this implies that $`\{g^{}=g^\mathrm{\Theta }\}`$ is an embedded submanifold. A similar argument applies to the intersection with $`U`$.
The action
$$G\times \{gg^\sigma =1\}\{gg^\sigma =1\}:g,g_1gg_1g^\mathrm{\Theta }$$
$`1.6`$
is isometric (for the symmetric space structure). The constancy of the rank of $`\psi `$ on connected components is equivalent to the statement that the dimension of the isotropy subgroup for the action of $`G`$ is constant on connected components of $`\{gg^\sigma =1\}`$ (in fact this dimension is the same on all components). Hence the action of $`G`$ must be transitive on connected components. The same applies to the same action of $`U`$ on $`\{gU:gg^\mathrm{\Theta }=1\}`$. This implies $`(a)`$.
For the first part of $`(b)`$, note that in fact the map
$$\{gU:g^1=g^\mathrm{\Theta }\}\{\eta Aut(U)^{(2)}:Out(\eta )=Out(\mathrm{\Theta })\}:gAd(g)\mathrm{\Theta }$$
is a universal covering for each connected component (For the identity component this covering is understood more intellibly by identifying the total space with $`U/K`$:
$$C_U(K)/C(U)U/K\mathrm{@}>q>>Ad(U)\mathrm{\Theta }$$
$`1.7`$
where $`q(g_1K)=Ad(g_1)\mathrm{\Theta }Ad(g_1)^1`$; to obtain a similar picture for another component, we replace $`\mathrm{\Theta }`$ by $`Ad(g)\mathrm{\Theta }`$, for some $`g`$ in the component).
The second part of $`(b)`$ is similar (We could also note that the inclusion $`\{g^1=g^{}=g^\mathrm{\Theta }\}\{g^{}=g^\mathrm{\Theta }\}`$ is a homotopy equivalence, since we know this is true for the identity component, and we are free to change $`\mathrm{\Theta }`$ to $`Ad(g)\mathrm{\Theta }`$; the fact that the $`\pi _0`$โs are the same is a reflection of the fact that classifying $`\mathrm{\Theta }`$โs and classifying $`\sigma `$โs are canonically isomorphic problems (see e.g. 2. of ยง6, chapter 10 of \[H\])). โ
Remarks (1.8). (a) I do not know of a uniform way to define an invariant for the class of an involution $`\eta `$ as in $`(b)`$. However it is a simple matter to produce an invariant in a case by case manner from the classification of symmetric spaces (see Table V of \[H1\]).
(b) The groups and maps in $`(1.4)`$ exist for any automorphism $`\mathrm{\Theta }`$ of $`K`$. However it seems that there is a linear characterization of the $`\varphi `$-images (up to connectedness issues), and $`G_0`$ is a real form, only in the symmetric case, $`\mathrm{\Theta }^2=1`$.
Fix a maximal abelian subalgebra $`๐ฑ_0๐จ`$. We then obtain $`\mathrm{\Theta }`$-stable Cartan subalgebras
$$๐ฅ_0=๐ต_{๐ค_0}(๐ฑ_0)=๐ฑ_0๐_0,๐ฑ=๐ฑ_0i๐_0,and๐ฅ=๐ฅ_0^{}$$
$`1.9`$
for $`๐ค_0`$, $`๐ฒ`$, and $`๐ค`$, respectively, where $`๐_0๐ญ`$ (see $`(6.60)`$ of \[Kn\]). We let $`T_0`$ and $`T`$ denote the maximal tori in $`K`$ and $`U`$ corresponding to $`๐ฑ_0`$ and $`๐ฑ`$, respectively.
Let $`\mathrm{\Delta }`$ denote the roots for $`๐ฅ`$ acting on $`๐ค`$; $`\mathrm{\Delta }๐ฅ_{}^{}`$, where $`๐ฅ_{}=๐_0i๐ฑ_0`$. We choose a Weyl chamber $`C^+`$ which is $`\mathrm{\Theta }`$-stable (to prove that $`C^+`$ exists, we must show that $`i๐ฑ_0`$, the $`+1`$ eigenspace of $`\mathrm{\Theta }`$ acting on $`๐ฅ_{}`$, contains a regular element of $`๐ค`$; this is equivalent to the fact that $`๐ฅ_0`$ in $`(1.8)`$ is a Cartan subalgebra). Since $`\sigma =()^\mathrm{\Theta }`$ and $`()^{}`$ is the identity on $`๐ฅ_{}`$, $`\sigma (C^+)=C^+`$.
Given our choice of $`C^+`$, we obtain a $`\mathrm{\Theta }`$-stable triangular decomposition $`๐ค=๐ซ^{}๐ฅ๐ซ^+`$, so that $`\sigma (๐ซ^\pm )=๐ซ^{}`$. Let $`N^\pm =exp(๐ซ^\pm )`$, $`H=exp(๐ฅ)`$, and $`B^\pm =HN^\pm `$. We also let $`W=W(G,T)`$ denote the Weyl group, $`W=N_U(T)/TN_G(H)/H`$.
At the group level we have the Birkhoff or triangular or LDU decomposition for $`G`$,
$$G=\underset{W}{}\mathrm{\Sigma }_w^G,\mathrm{\Sigma }_w^G=N^{}wHN^+,$$
$`1.10`$
where $`\mathrm{\Sigma }_w^G`$ is diffeomorphic to $`(N^{}wN^{}w^1)\times H\times N^+`$. When we intersect this decomposition with $`\varphi (G/G_0)`$, and the other spaces in (b) of $`(1.5)`$, we obtain various decompositions. We will first determine the structure of the pieces in the $`\{g^{}=g^\mathrm{\Theta }\}`$ case (thus we initially ignore connectedness issues).
###### (1.11) Proposition
Fix $`wW`$.
(a) The intersection $`\{g^{}=g^\mathrm{\Theta }\}\mathrm{\Sigma }_w^G`$ is nonempty if and only if there exists $`๐จwN_U(T)`$, such that $`๐จ^\mathrm{\Theta }=๐จ`$; $`๐จ`$ is unique modulo the action
$$T\times \{๐จN_U(T):๐จ^\mathrm{\Theta }=๐จ\}\{๐จ^\mathrm{\Theta }=๐จ\}:\lambda ,๐จ\lambda ๐จ\lambda ^\mathrm{\Theta }.$$
(b) For the action $`B^{}\times \{g^{}=g^\mathrm{\Theta }\}\{g^{}=g^\mathrm{\Theta }\}`$:$`b,gbgb^\mathrm{\Theta }`$, the stability subgroup is given by
$$B_๐จ^{}=\{b:๐จ^1b๐จ=\sigma (b)\}\{lN^{}:๐จ^1l๐จ=\sigma (l)N^+\}\times \{hH:h^{w^1}=\sigma (h)\}.$$
(c) The orbits of $`B^{}`$ in $`\{g^{}=g^\mathrm{\Theta }\}\mathrm{\Sigma }_w^G`$ are open and indexed by
$$\pi _0(\{๐จw:๐จ^\mathrm{\Theta }=๐จ\})\{๐จw:๐จ^\mathrm{\Theta }=๐จ\}/T,$$
where $`T`$ acts as in part (a).
(d) The map
$$N^{}N^w\times \{hH,LN^{}N^{+w}:h^{w^1\mathrm{\Theta }}=h^{},\sigma (L)^{๐จh}=L^1\}\{g^{}=g^\mathrm{\Theta }\}\mathrm{\Sigma }_w^G$$
given by $`l,h,LlL^1๐จh(lL^1)^\mathrm{\Theta }`$ is a diffeomorphism onto the connected component containing $`๐จ`$. This component is homotopic to the torus $`exp(\{Ad(w^1)\mathrm{\Theta }|_๐ฑ=1\})`$.
(e) In particular for $`w=1`$, the map
$$N^{}\times (T_0^{(2)}\times _{exp(i๐_0)^{(2)}}exp(i๐_0))\times exp(i๐ฑ_0)\{g^{}=g^\mathrm{\Theta }\}\mathrm{\Sigma }_1^G$$
$$l,[๐จ,m],a_\varphi g=l๐จma_\varphi l^\mathrm{\Theta }$$
is a diffeomorphism, so that the connected components for $`\{g^{}=g^\mathrm{\Theta }\}\mathrm{\Sigma }_1^G`$ are indexed by $`T_0^{(2)}/exp(i๐_0)^{(2)}`$.
###### Demonstration Proof of (1.11)
Suppose that $`g\mathrm{\Sigma }_w^G`$. We write $`g=l๐จhu`$, for some $`lN^{}`$, $`๐จwN_U(T)`$, $`hexp(๐ฅ_{})`$, $`u`$$`N^+`$. If we additionally require that $`lN^{}(N^{})^w`$, then this decomposition is unique, but we will not require this at the outset.
We have $`g=g^\mathrm{\Theta }`$ if and only if
$$l๐จhu=u^\mathrm{\Theta }(๐จh)^\mathrm{\Theta }l^\mathrm{\Theta }$$
$`1.12`$
if and only if
$$(๐จh)^\mathrm{\Theta }=(u^\sigma l)(๐จh)(ul^\sigma )$$
$$=\{(u^\sigma l)_{}(u^\sigma l)_+\}(๐จh)(ul^\sigma )$$
$$=(u^\sigma l)_{}(๐จh)\{(u^\sigma l)_+^{(๐จh)^1}ul^\sigma \}$$
$`1.13`$
where $`L=L_{}L_+`$ denotes the decomposition induced by the diffeomorphism
$$N^{}(N^{})^w\times N^{}(N^+)^wN^{}:L_{},L_+L=L_{}L_+$$
$`1.14`$
Thus $`(1.13)`$ holds if and only if
$$(u^\sigma l)_{}=1=(u^\sigma l)_+^{(๐จh)^1}ul^\sigma $$
$`1.15`$
and $`(๐จh)^\mathrm{\Theta }=๐จh`$, or, using the fact that $`h`$ is real,
$$h^\mathrm{\Theta }=h^wand๐จ๐จ^\mathrm{\Theta }=1.$$
$`1.16`$
Consider part (a). If $`g`$ is in the intersection, then we have just seen that $`๐จ`$ must satisfy $`๐จ^\mathrm{\Theta }=๐จ`$. Conversely, given a unitary representative $`๐จ`$ for $`w`$ satisfying $`๐จ^\mathrm{\Theta }=๐จ`$, the intersection contains $`๐จ`$ and hence is nonempty. This proves (a).
Part (b) is straightforward.
Now consider (c). We first write $`g\{g^{}=g^\mathrm{\Theta }\}\mathrm{\Sigma }_w^G`$ uniquely as $`l\omega u`$, where $`lN^{}wN^{}w^1`$, $`\omega =๐จh`$, and $`uN^+`$. For the first part of (c) we must prove that we can relax the constraint on $`l`$ to arrange for $`u=l^\mathrm{\Theta }`$. We can write
$$g=lL^1\omega \{(\omega L\omega ^1)u\},$$
$`1.17`$
where $`LN^{}wN^+w^1`$ is arbitrary. We must prove the existence of $`L`$ such that $`lL^1=u^\mathrm{\Theta }\omega \sigma (L^1)\omega ^1`$, or
$$u^\sigma l=\omega L^\sigma \omega ^1L.$$
$`1.18`$
The basic fact is that this equation has a unique solution $`LN^{}N^{+w}`$ satisfying $`\omega L^\sigma \omega ^1=L^1`$, namely $`L=(u^\sigma l)^{1/2}`$ (square root has an unambiguous meaning in a simply connected nilpotent Lie group). To see this simply plug such an $`L`$ into $`(1.18)`$. We obtain the equation $`L^2=u^\sigma l`$. The fact that $`u^\sigma l`$, and its square root, satisfy $`\omega L^\sigma \omega ^1=L^1`$ follows from $`(1.15)`$, and uniqueness of the square root.
As we remarked previously, the existence of a solution $`L`$ proves that $`B^{}`$ has open orbits. The rest of part (c) is relatively straightforward, using $`(1.16)`$.
The uniqueness of the solution $`L`$, subject to the constraint we imposed, implies the first part of (d). The second statement in $`(d)`$ follows routinely from the first part.
For part (e), to clarify the statement, observe that
$$exp(i๐_0)^{(2)}=Kexp(i๐_0)=T_0^{(2)}exp(i๐_0).$$
$`1.19`$
Now suppose that $`w=1`$. In this case $`๐จ`$ is in the kernel of the homomorphism $`TT:๐จ๐จ๐จ^\mathrm{\Theta }`$, and this equals the subgroup generated by $`T_0^{(2)}`$ and $`exp(i๐_0)`$. We can modify $`๐จ`$ by multiplying by something in the image of the homomorphism $`TT:\lambda \lambda \lambda ^\mathrm{\Theta }`$. This image is $`exp(i๐_0)`$. Therefore we can choose $`๐จT_0^{(2)}`$, but this choice is unique only modulo the intersection of $`T_0^{(2)}`$ and $`exp(i๐_0)`$. This proves (e). โ
Example (1.20). In the group case, $`U=K\times K`$, where $`K`$ embeds diagonally. The image of $`๐ฑ_0`$ inside $`๐ฒ`$ is $`\{(x,x):x๐ฑ_0\}`$, while $`i๐_0=\{(x,x):x๐ฑ_0\}`$. This implies that the quotient $`T_0^{(2)}/exp(i๐_0)^{(2)}`$ is trivial. Thus in this group case, the set of generic elements (considered in part (e)) is connected, as we already know.
###### (1.22) Notation
Given $`๐จ`$ as in $`(c)`$ of $`(1.11)`$, we let $`\mathrm{\Sigma }_๐จ^{\{g^{}=g^\mathrm{\Theta }\}}`$ denote the corresponding connected component of $`\{g^{}=g^\mathrm{\Theta }\}\mathrm{\Sigma }_w^G`$ (the $`B^{}`$-orbit of $`๐จ`$, in the sense of (c) of $`(1.11)`$). If $`๐จ\varphi (G/G_0)`$, then we will write $`\mathrm{\Sigma }_๐จ^{\varphi (G/G_0)}`$ for this component. We also set $`\mathrm{\Sigma }_๐จ^{\varphi (U/K)}=\varphi (U/K)\mathrm{\Sigma }_๐จ^{\varphi (G/G_0)}`$.
Having understood the intersection of $`\{g^{}=g^\mathrm{\Theta }\}`$ with the triangular decomposition, we now want to specialize this to the identity component. In an abstract way this is answered by $`(1.5b)`$ (and Remark $`(1.8a)`$). Concerning open orbits, we have the following
###### (1.23) Proposition
Suppose that $`๐จN_U(T)`$ satisfies $`๐จ^\mathrm{\Theta }=๐จ`$. The following are equivalent:
(a) $`\mathrm{\Sigma }_๐จ^{\{g^{}=g^\mathrm{\Theta }\}}`$ is an open $`B^{}`$-orbit in the identity component, $`\varphi (G/G_0)`$.
(b) There exists $`๐จ_1N_U(T_0)`$ such that $`\varphi (๐จ_1K)=๐จ`$.
Hence the open orbits can be parameterized by either $`N_U(T_0)/N_K(T_0)`$ (the intrinsic point of view), or the set of $`๐จT_0^{(2)}/exp(i๐_0)^{(2)}`$ such that $`Ad(๐จ)\mathrm{\Theta }`$ is equivalent to $`\mathrm{\Theta }`$ in the sense of $`(1.5b)`$ (the nonintrinsic point of view, as in $`(1.11e)`$).
In addition, the $`๐จ_1K`$ are exactly the $`T_0`$ fixed points in $`U/K`$.
###### Demonstration Proof of (1.23)
Determining the possible (open) $`B^{}`$ orbits in $`G/G_0`$ is equivalent to determining the possible (open) $`G_0`$ orbits in $`B^{}\backslash G`$. Thus the equivalence of $`(a)`$ and $`(b)`$ follows from Theorem 4.6 and its Corollaries in \[Wolf\]. The other statements are obvious. โ
In general it apparently remains an open question to systematically obtain representatives for all $`B^{}`$ orbits in $`G/G_0`$, from the intrinsic point of view (see \[WZ\] for the Hermitian symmetric case). In this regard the nonintrinsic point of view of $`(1.11)`$ seems to have some utility.
Let $`q:GG/B^+`$ denote the quotient map. The map $`q`$ applied to the decomposition $`(1.9)`$ induces the (more conventional) triangular stratification for the flag space,
$$U/TG/B^+=\underset{W}{}\mathrm{\Sigma }_w,\mathrm{\Sigma }_w=N^{}wB^+,$$
$`1.24`$
where each $`\mathrm{\Sigma }_w`$ is a cell ($`N^{}wN^{}w^1`$). As a consequence, for the pieces of the induced decomposition for $`U`$, there are diffeomorphisms
$$\mathrm{\Sigma }_w^U=U\mathrm{\Sigma }_w^G\mathrm{\Sigma }_w\times T.$$
$`1.25`$
The inclusions $`\mathrm{\Sigma }_w^U\mathrm{\Sigma }_w^G`$ are homotopy equivalences, because $`T`$ is homotopy equivalent to $`B^+`$:
$$\begin{array}{ccccc}T& & \mathrm{\Sigma }_w^U& \mathrm{@}>q>>& \mathrm{\Sigma }_w\\ & & & & \\ B^+& & \mathrm{\Sigma }_w^G& \mathrm{@}>q>>& \mathrm{\Sigma }_w\end{array}$$
$`1.26`$
The main point of this section is now to describe the generalization of this to $`U/KG/G_0`$. We consider the Iwasawa decomposition for $`G`$, which we write as
$$GN^{}\times A\times U:g=๐(g)๐(g)๐ฆ(g),$$
$`1.27`$
where $`A=exp(๐ฅ_{})`$. There is an induced right action
$$U\times T\times G_0U:(u,t,g_0)t^1๐ฆ(ug_0)$$
$`1.28`$
arising from the identification of $`U`$ with $`N^{}A\backslash G`$.
###### (1.29) Proposition
Suppose that $`๐จN_U(T)`$ satisfies $`๐จ^\mathrm{\Theta }=๐จ`$ and $`\mathrm{\Sigma }_๐จ^{\{g^{}=g^\mathrm{\Theta }\}}\varphi (G/G_0)`$. Fix a choice of $`๐จ_1U`$ such that $`๐จ_1๐จ_1^\mathrm{\Theta }=๐จ`$.
(a) The map
$$T\times G_0\mathrm{\Sigma }_๐จ^{\varphi (U/K)}:(t,g_0)\varphi (t^1๐ฆ(๐จ_1g_0))$$
$`1.30`$
is surjective, and induces a diffeomorphism
$$T\times _{exp(\{Ad(๐จ)\mathrm{\Theta }|_๐ฑ=1\})}R\backslash G_0/K\mathrm{\Sigma }_๐จ^{\varphi (U/K)}$$
$`1.31`$
where $`R=(N^{}A)^{๐จ_1^1}G_0`$ is a contractible subgroup of $`G_0`$, and $`\lambda exp(\{Ad(๐จ)\mathrm{\Theta }|_๐ฑ=1\})`$ is identified with a pair $`(\lambda ,\lambda ^{๐จ_1^1})`$.
(b) The inclusion
$$\mathrm{\Sigma }_๐จ^{\varphi (U/K)}\mathrm{\Sigma }_๐จ^{\varphi (G/G_0)}.$$
$`1.32`$
is a homotopy equivalence; each is homotopic to the torus $`exp(\{Ad(w^1)\mathrm{\Theta }|_๐ฑ=1\})`$.
(c) The connected components of $`\varphi (U/K)`$ intersected with $`\mathrm{\Sigma }_1^U`$ are indexed by $`๐จT_0^{(2)}/exp(i๐_0)^{(2)}`$ such that $`๐จ=๐จ_1๐จ_1^\mathrm{\Theta }`$, for some $`๐จ_1N_U(T_0)`$; for such a $`๐จ`$, and choice of $`๐จ_1`$, the diffeomorphism in (a) simplifies to
$$exp(i๐_0)/exp(i๐_0)^{(2)}\times A_0\backslash G_0/K\mathrm{\Sigma }_๐จ^{\varphi (U/K)},$$
$`1.33`$
where $`A_0=exp(๐_0)`$.
###### Demonstration Proof of $`(1.29)`$
In proving the first part of (a), it is convenient to work with $`U/K`$ instead of $`\varphi (U/K)`$. Thus we consider a point in the intersection of $`U/K`$ and the $`B^{}`$-orbit of $`๐จ_1G_0G/G_0`$. This point can be represented by a $`uU`$ such that $`u=b^{}๐จ_1g_0`$, for some $`b^{}B^{}`$ and $`g_0G_0`$. This immediately implies that $`u`$ is in the $`T\times G_0`$-orbit of $`๐จ_1`$, and this proves surjectivity of the first map in (a).
For the second part of (a), we first calculate the stabilizer for the action $`(1.28)`$ at the point $`๐จ_1`$. Suppose that $`tT`$ and $`g_0G_0`$ satisfy $`t๐ฆ(๐จ_1g_0)=๐จ_1`$. This is equivalent to
$$๐จ_1g_0๐จ_1^1=๐(๐จ_1g_0)๐(๐จ_1g_0)t.$$
$`1.34`$
This implies that $`g_0G_0(B^{})^{๐จ_1^1}`$, and $`t=T(g_0^{๐จ_1})`$. Conversely if $`g_0G_0(B^{})^{๐จ_1^1}`$, then $`(1.34)`$ holds with $`t=T(g_0^{๐จ_1})`$. Thus the stabilizer is isomorphic to
$$\{(T(g_0^{๐จ_1}),g_0):g_0G_0(B^{})^{๐จ_1^1}\}T\times G_0.$$
$`1.35`$
The group $`G_0(B^{})^{๐จ_1^1}`$ is connected and solvable. The torus part is isomorphic to $`\{\lambda T:\lambda ^{๐จ_1^1}G_0\}`$. This condition on $`\lambda `$ is equivalent to $`(\lambda ^{๐จ_1^1})^\mathrm{\Theta }=\lambda ^{๐จ_1^1}`$, or $`\lambda exp(\{Ad(๐จ)\mathrm{\Theta }|_๐ฑ=1\})`$. This implies $`(1.31)`$.
From (a), since $`R\backslash G_0/K`$ is contractible, it follows that the double coset space in $`(1.31)`$ is homotopic to $`exp(\{Ad(๐จ)\mathrm{\Theta }|_๐ฑ=1\})`$, modulo elements of order $`2`$. A $`t`$ in this torus is mapped in $`(1.30)`$ to $`t^1๐จt^\mathrm{\Theta }=๐จ(t^{๐จ^1}t^\mathrm{\Theta })`$. It is straightforward to check that $`t^{๐จ^1}t^\mathrm{\Theta }`$ belongs to $`exp(\{Ad(\mathrm{w}^1)\mathrm{\Theta }|_๐ฑ=1\})`$. Together with $`(d)`$ of $`(1.11)`$, this implies $`(b)`$.
Part $`(c)`$ follows from $`(a)`$. โ
We want to explain how this proposition is related to familiar facts in special cases. First consider the group case $`X=K`$. We have already explained why the generic set is connected; see Example $`(1.20)`$. In the group case, $`(1.1)`$ has the form
$$\begin{array}{ccc}& & \{(x,y)๐จ^{}๐จ^{}\}\\ & & & \\ \{(x,x^{}):x๐จ^{}\}& & & & \{(x,y)๐จ๐จ\}\\ & & & \\ & & \{(x,x):x๐จ\}\end{array}$$
$`1.36`$
Given $`g_0G_0=K^{}`$, $`g_0`$ maps to $`(g_0,g_0^{})G=K^{}\times K^{}`$. Given an arbitrary triangular decomposition $`๐จ^{}=\stackrel{~}{๐ซ}_{}\stackrel{~}{๐ฅ}\stackrel{~}{๐ซ}_+`$, we obtain a $`\mathrm{\Theta }`$-stable triangular decomposition for $`๐ค`$ by defining $`๐ซ_\pm =\stackrel{~}{๐ซ}_\pm \times \stackrel{~}{๐ซ}_\pm `$, and $`๐ฅ=\stackrel{~}{๐ฅ}\times \stackrel{~}{๐ฅ}`$. The Iwasawa factorization $`(1.27)`$ is equivalent to the two Iwasawa factorizations
$$g_0=l_1a_1k_1,g_0^{}=l_2a_2k_2,$$
$`1.37`$
where $`a_iexp(\stackrel{~}{๐ฅ}_{})`$, $`l_i\stackrel{~}{N}^{}`$, $`k_iK`$. We have an equivariant isomorphism $`U/KK:(g,h)gh^1`$. The map in (c) of $`(1.29)`$ (using $`()^{}`$ in place of inverse) is given by
$$T/T^{(2)}\times A\backslash K^{}/K\mathrm{\Sigma }_1^K:t,g_0Kta_1^1l_1^1l_2^{}a_2^{}t.$$
$`1.38`$
So $`a_\varphi =a_1^1a_2^{}=(a_1a_2)^1`$. Thus in $`(1.29)`$ we are using $`exp(\stackrel{~}{๐ฅ}_{})g_0Kexp(\stackrel{~}{๐ฅ}_{})\backslash K^{}/K`$ as coordinate, which is completely equivalent to using $`l_1`$ or $`l_2N^{}`$. From this point of view, $`l_1`$ is a horocycle coordinate.
Now suppose that $`U/K`$ is Hermitian symmetric. In this case $`\mathrm{\Theta }`$ is inner, $`P=K^{}B^+`$ is a parabolic subgroup of $`G`$, and the natural map $`๐ฆ:U/KG/P`$ is a $`U`$-equivariant isomorphism. The natural map $`\eta :G_0/KG/P`$ is an open holomorphic embedding, and the image is contained $`๐ฆ((U/K)_r)`$, the regular set (see ch VIII of \[H1\], especially Prop 7.14).
There is a commutative diagram
$$\begin{array}{ccccc}G_0/K& \mathrm{@}>๐ฆ>>& \mathrm{\Sigma }_1^{U/K}:& g_0K๐ฆ(g_0)K& \\ I& & ๐ฆ\\ G_0/K& \mathrm{@}>\eta >>& G/P\end{array}$$
$`1.39`$
where the top arrow $`๐ฆ`$ is a diffeomorphism, by $`(b)`$ of $`(1.29)`$, and the map $`I`$ is defined in the following way: given $`g_0KG_0/K`$, we can write $`๐ฆ(g_0)=exp(ix)k`$ uniquely, where $`x๐ญ`$, $`exp(itx)K`$ is a geodesic of minimal length joining the basepoint to $`๐ฆ(g_0)K`$, and $`kK`$; we set $`I(g_0K)=(g_0k^1)^1K`$. To see that the diagram is commutative, note that because $`g_0k^1G_0`$, $`(g_0k^1)^1=(g_0k^1)^\mathrm{\Theta }=exp(ix)a^\mathrm{\Theta }l^\mathrm{\Theta }`$, and $`l^\mathrm{\Theta }N^+`$; thus $`I(g_0K)`$ equals $`๐ฆ(g_0)`$ mod $`P`$.
Thus in the Hermitian symmetric case, $`\mathrm{\Sigma }_1^{U/K}`$ is the usual model of $`G_0/K`$ inside $`U/K`$, but the parameterization in $`(b)`$ of $`(1.29)`$ is related to the natural holomorphic map $`\eta `$ in a clumsy way.
ยง2. Diagonal Distribution
Suppose that $`๐จT_0^{(2)}=T\{g^{}=g^\mathrm{\Theta }\}`$. Then $`Ad(๐จ)\mathrm{\Theta }`$ is an involution, and $`๐ซ^{}๐ฅ๐ซ^+`$ is $`Ad(๐จ)\mathrm{\Theta }`$-stable. We also suppose that $`๐จ\varphi (U/K)`$.
Given $`g\mathrm{\Sigma }_๐จ^{\varphi (U/K)}`$, we factor $`g`$ as in (e) of $`(1.11)`$,
$$g=l๐จma_\varphi l^\mathrm{\Theta },$$
$`2.1`$
where $`lN^{}`$, $`a_\varphi exp(i๐ฑ_0)`$, and $`[๐จ,m]T_0^{(2)}\times _{exp(i๐_0)^{(2)}}exp(i๐_0)`$.
###### (2.2) Theorem
For $`\lambda (i๐ฑ_0)^{}`$,
$$_{\mathrm{\Sigma }_๐จ^{\varphi (U/K)}}a_\varphi (g)^{i\lambda }=\frac{1}{M}\stackrel{๐จ}{}\frac{\delta ,\alpha }{\delta i\lambda ,\alpha }$$
$`2.3`$
where the product is over pairs $`(\alpha ,Ad(๐จ)\mathrm{\Theta }(\alpha ))`$ of positive complex roots which are $`\underset{ยฏ}{not}`$ of compact type for $`Ad(๐จ)\mathrm{\Theta }`$, and $`M=|N_U(T_0)/N_K(T_0)|`$. Hence
$$_{\varphi (U/K)}a_\varphi (g)^{i\lambda }=\frac{1}{M}(\stackrel{๐จ}{}\frac{\delta ,\alpha }{\delta i\lambda ,\alpha })$$
$`2.4`$
where the sum is over representatives $`๐จ`$ for the connected components of $`\varphi (U/K)\mathrm{\Sigma }_1^U`$.
Note that it does not matter whether we take $`\alpha `$ or $`Ad(๐จ)\mathrm{\Theta }(\alpha )`$ in the product $`(2.3)`$, because $`\delta `$ and $`\lambda `$ are fixed by $`Ad(๐จ)\mathrm{\Theta }`$. In the case in which $`\mathrm{\Theta }`$ is inner, i.e. $`๐_0=0`$, all roots are either of compact or noncompact type. Hence in this case the product in $`(2.3)`$ is over the noncompact type roots.
There is a more intrinsic way to write $`(2.4)`$. The right hand side can be expressed as a sum over $`๐จ_1N_U(T_0)/N_K(T_0)`$, using $`Ad(๐จ)\mathrm{\Theta }=Ad(๐จ_1)\mathrm{\Theta }Ad(๐จ_1^1)`$ (see $`(1.23)`$).
To prove $`(2.2)`$ we will first note that we can reduce to the case $`๐จ=1`$ in $`(2.3)`$. We will then need several Lemmas.
To see that it suffices to prove $`(2.3)`$ in the case $`๐จ=1`$, observe that $`\underset{ยฏ}{left}`$ translation by $`๐จT_0^{(2)}`$, which is an isometric map for the Riemannian structure of $`U`$, maps the $`๐จ=1`$ component to the $`๐จ`$-component:
$$L_๐จ:\mathrm{\Sigma }_1^{\varphi (U/K)}\mathrm{\Sigma }_๐จ^{\varphi (U/K)}.$$
$`2.5`$
This map interchanges the canonical factorization from that relative to $`\mathrm{\Theta }`$ to the one relative to $`Ad(๐จ)\mathrm{\Theta }`$: if $`g`$ has the unique decomposition $`g=lma_\varphi l^\mathrm{\Theta }`$, then $`๐จg`$ has the unique decomposition $`๐จg=l^๐จ๐จma_\varphi (l^๐จ)^{Ad(๐จ)\mathrm{\Theta }}`$. Since $`a_\varphi `$ is unchanged, the integral is evaluated in the same way, except the meaning of the roots changes.
We henceforth suppose $`๐จ=1`$. Consider the parameterization
$$\mathrm{\Phi }:exp(i๐_0)\times A_0\backslash G_0/K\mathrm{\Sigma }_1^{\varphi (U/K)}:(t,A_0g_0K)\varphi (t^1๐ฆ(g_0)).$$
$`2.6`$
from $`(1.33)`$. Note that $`a_\varphi =๐(g_0)^1๐(g_0)^\tau exp(i๐ฑ_0)`$.
###### (2.7) Lemma
We have
$$\mathrm{\Phi }^{}(dV_{U/K})=a_\varphi ^{2\delta }(dV_{exp(i๐_0)}\times dV_{A_0\backslash G_0/K})$$
where $`dV_{A_0\backslash G_0/K}`$ is obtained by integrating a $`G_0`$-invariant measure on $`A_0\backslash G_0`$ over $`K`$. Thus $`(2.3)`$ equals
$$a_\varphi (๐ฆ(g_0))^{2\delta i\lambda }๐V_{A_0\backslash G_0/K}.$$
$`2.8`$
###### Demonstration Proof of (2.7)
We will consider a slight reformulation of the problem. We identify $`U/K`$ with $`\varphi (U/K)`$. Let $`S`$ denote the inverse image of $`\mathrm{\Sigma }_1^{U/K}`$ in $`U`$, with respect to the projection $`UU/K`$. Consider the lift
$$\mathrm{\Psi }:exp(i๐_0)\times A_0\backslash G_0S:(t,A_0g_0)t^1๐ฆ(g_0)$$
$`2.9`$
of $`\mathrm{\Phi }`$ in $`(2.6)`$. We must show that the Jacobian for the mapping $`\mathrm{\Psi }`$, with respect to the Riemannian structures induced by the Killing form, is equal to a constant times $`a_\varphi ^{2\delta }`$. To do this we identify $`i๐_0`$, $`๐ค_0๐_0`$, and $`๐ฒ`$ with the tangent spaces to $`exp(i๐_0)`$, $`A_0\backslash G_0`$ and $`U`$, respectively, using the exponential map and right translation (we use right translation because $`A_0`$ appears on the left). Let $`P:๐ค๐ฒ`$ denote the projection with kernel $`๐ซ^{}๐ฅ_{}`$. We compute
$$d\mathrm{\Psi }|_{(t,A_0g_0)}:i๐_0(๐ค_0๐_0)๐ฒ:(\chi ,x)\frac{d}{dฯต}|_{ฯต=0}(te^{ฯต\chi })^1๐ฆ(e^{ฯตx}g_0)๐ฆ(g_0)^1t$$
$$=Ad(t^1)\{\chi +P(Ad(๐^1๐^1)(x))\}.$$
$`2.10`$
The operator $`Ad(t^1)`$ preserves $`๐ฒ`$-volume, so it can be ignored.
Write $`๐=๐_1๐_0`$, relative to the decomposition $`A=exp(i๐ฑ_0)A_0`$. Since $`๐_0G_0`$, $`Ad(๐_0)`$ will preserve $`๐ค_0`$-volume. Thus the determinant of $`(2.10)`$ equals the determinant of the map
$$๐ค_0๐_0๐ฒi๐_0:xP_1(Ad(๐^{})Ad(๐_1^1)(x)),$$
$`2.11`$
where $`๐^{}=๐_1^1๐a_1N^{}`$ and $`P_1`$ is $`P`$ followed by the projection to $`๐ฒi๐_0`$.
Given $`x๐ค`$, if $`x=x_{}+x_๐ฅ+x_+`$ is its triangular decomposition, then
$$P(x)=x_+^{}+\frac{1}{2}(x_๐ฅx_๐ฅ^{})+x_+.$$
$`2.12`$
If $`x๐ค_0`$, then $`x_{}=x_+^\sigma `$, and $`x_๐ฅ=x_{๐ฑ_0}๐ฑ_0`$. Because $`๐^{}๐_1^1`$ maps $`๐ซ_{}`$ into itself, $`(2.11)`$ is given by
$$P_1(Ad(๐^{}๐_1^1)(x))=[(x_+^{๐^{}๐_1^1})_+]^{}+(x_{๐ฑ_0}+(x_+^{๐^{}๐_1^1})_{๐ฑ_0})+(x_+^{๐^{}๐_1^1})_+.$$
$`2.13`$
Thus the determinant of $`(2.10)`$ is the same as the (real) determinant of the map $`x_+(x_+^{๐^{}๐_1^1})_+`$. Because of the unipotence of $`Ad(๐^{})`$, this is equal to
$$\underset{\alpha >0}{}|๐_1^\alpha |^2=๐_1^{4\delta }=a_\varphi ^{2\delta }.$$
$`2.14`$
We will now show that the integral $`(2.8)`$ can be computed using a Duistermaat-Heckman exact stationary phase calculation. The relevant Poisson structure was discovered by Evens and Lu in a very general setting (\[EL\]). We will introduce this structure directly, but to understand why it is natural the reader will need to consult the original paper.
To do calculations we will use the isomorphism of vector bundles
$$G_0\times _K๐ญT(G_0/K):[g_0,x]\frac{d}{dt}|_{t=0}(g_0e^{tx}K),$$
and we will use the Killing form to identify $`๐ญ^{}`$ with $`๐ญ`$.
Consider the $`Ad(T_0)`$ and $`Ad(A_0)`$-stable decomposition of $`๐ค`$ as a direct sum of subalgebras:
$$๐ค=๐ค_0(๐ซ^{}i๐ฅ_0).$$
$`2.15`$
Let $`pr_{๐ค_0}`$ denote the projection $`๐ค๐ค_0`$ along this decomposition. Given $`x๐ค`$, with triangular decomposition $`x=x_{}+x_0+x_+`$,
$$pr_{๐ค_0}(x)=(x_+^\sigma +(x_0)_{๐ฅ_0}+x_+).$$
$`2.16`$
The Evens-Lu Poisson bivector is given by
$$\mathrm{\Pi }([g_0,x][g_0,y])=\mathrm{\Omega }(g_0)(x),y,$$
$`2.17`$
where $`\mathrm{\Omega }(g_0):๐ญ๐ญ`$ is given by
$$\mathrm{\Omega }(g_0)(x)=\{(pr_{๐ค_0}(ix^{g_0}))^{g_0^1}\}_๐ญ.$$
$`2.18`$
The operator $`\mathrm{\Omega }`$ satisfies the equivariance condition
$$\mathrm{\Omega }(a_0g_0k)=Ad(k)^1\mathrm{\Omega }(g_0)Ad(k).$$
$`2.19`$
To understand $`\mathrm{\Omega }`$, it is useful to consider the augmented operator $`\stackrel{~}{\mathrm{\Omega }}:๐ค_0๐ค_0`$ given by
$$\stackrel{~}{\mathrm{\Omega }}(g_0)(x_๐จ+x_๐ญ)=\{(pr_{๐ค_0}((x_๐จ+ix_๐ญ)^{g_0}))^{g_0^1}\}.$$
$`2.20`$
Relative to the decomposition $`๐ค_0=๐จ๐ญ`$,
$$\stackrel{~}{\mathrm{\Omega }}=\left(\begin{array}{cc}1& \\ 0& \mathrm{\Omega }\end{array}\right).$$
$`2.21`$
This augmented operator can be factored as the composition of four operators
$$๐ค_0\mathrm{@}>I>>๐ฒ\mathrm{@}>Ad(๐ฆ(g_0))>>๐ฒ\mathrm{@}>T>>๐ค_0\mathrm{@}>Ad(๐_0^1g_0)^1>>๐ค_0$$
$`2.22`$
where the first operator is given by $`I(x_๐จ+x_๐ญ)=x_๐จ+ix_๐ญ`$, and $`T(g_0)`$ maps $`x=x_+^{}+(x_{t_0}+x_{i๐_0})+x_+`$ to
$$T(g_0)(x)=pr_{๐ค_0}(x^{๐^{}๐_1(g_0)})=$$
$$[(x_+^{๐^{}๐_1})_+]^\sigma +(x_{t_0}+(x_+^{๐^{}๐_1})_{๐ฑ_0+๐_0})+(x_+^{๐^{}๐_1})_+,$$
$`2.23`$
where $`๐^{}=๐_0๐๐_0^1`$, and the last equality depends upon the fact that conjugation by $`๐^{}๐_1(g_0)`$ maps $`๐ซ^{}`$ into itself, and that $`๐ซ^{}`$ terms disappear when we use $`(2.16)`$.
###### (2.24) Lemma
(a) $`\mathrm{\Omega }so(๐ญ)`$; the Schouten bracket $`[\mathrm{\Pi },\mathrm{\Pi }]`$ vanishes, so that $`(G_0/K,\mathrm{\Pi })`$ is a Poisson manifold.
(b) $`ker(\mathrm{\Omega }(g_0))=\{[g_0,(๐_0^{๐ฆ(g_0)^1})_๐ญ]\}`$
(c) $`Pfaffian(\mathrm{\Omega }(g_0)|_{ker(\mathrm{\Omega })^{}})=๐_1(g_0)^{2\delta }`$.
###### Demonstration Proof of (2.24)
For (a) let $`X=x^{g_0}`$, $`Y=y^{g_0}`$, $`x,y๐ญ`$. Then
$$\mathrm{\Omega }(g_0)x,y=pr_{๐ค_0}(iX),Y$$
$$=iX_+^\sigma +iX_+,Y_+^\sigma +Y_{๐ฅ_0}+Y_+=2iX_+,Y_+^\sigma .$$
$`2.25`$
This is clearly skew-symmetric in $`X`$ and $`Y`$, because $`\sigma `$ preserves the Killing form and it is complex antilinear. For the second part of $`(a)`$ we refer to \[EL\] (or see ยง3 of \[FO\] for an exposition specific to this case).
For (b), note that $`(2.23)`$ implies the kernel of $`T`$ is $`i๐_0`$. Thus $`(2.22)`$ implies
$$ker(\stackrel{~}{\mathrm{\Omega }}(g_0))=\{[g_0,x]:(x_๐จ+ix_๐ญ)i๐_0^{๐ฆ(g_0)^1}\}$$
$`2.26`$
This, together with $`(2.21)`$, implies $`(b)`$.
For (c), note that in $`(2.22)`$ the first, second and fourth operators preserve volume determined by the Killing form. The determinant of $`T`$ (relative to the Killing form volumes) is the same as the determinant of the operator on $`๐ซ^+`$ which maps $`x_+`$ to $`(x_+^{๐^{}๐_1})_+`$. This determinant equals
$$\underset{\alpha >0}{}๐_1^{2\alpha }=๐_1^{4\delta }$$
$`2.27`$
Thus the Pfaffian is $`๐_1^{2\delta }`$. โ
By $`(b)`$ the tangent directions in $`G_0/K`$ determining the symplectic leaves are given by $`[g_0,x]`$ such that $`x^๐ฆ๐_0`$. This is clearly $`A_0`$-invariant, because $`๐ฆ(a_0g_0)=๐ฆ(g_0)`$. Thus the left action of $`A_0`$ permutes the symplectic leaves. The symplectic form is given by the formula
$$\omega ([g_0,x],[g_0,y])=\mathrm{\Omega }(g_0)|_{ker(\mathrm{\Omega })^{}})^1(x),y.$$
$`2.28`$
This form does not in general descend to a form on the quotient $`A_0\backslash G_0/K`$. However $`(c)`$ of the preceding Lemma does imply that the volume form descends.
###### (2.29) Proposition
(a) The action of $`T_0`$ is Hamiltonian with momentum map
$$\mu :G_0/K(๐ฑ_0)^{}:g_0Kilog(๐_1(g_0)),,$$
This momentum map is proper, and it is semibounded.
(b) The symplectic measure is
$$\omega ^d/d!=๐_1(g_0K)^{2\delta }dV_{A_0\backslash G_0/K}(A_0g_0K)$$
(where the invariant measure is suitably normalized).
###### Demonstration Proof of (2.29)
Part (a) is proven in \[FO\] (Lemma $`3.3`$, which in turn refers to a result of Van Den Ban). Part (b) follows from (c) of $`(2.24)`$. โ
We can now apply the Duistermaat-Heckman exact stationary phase method, as generalized to noncompact manifolds in \[PW\]. For definiteness we will consider the symplectic leaf through the basepoint of $`G_0/K`$.
We must first find the fixed points of the $`T_0`$ action. Suppose that $`g_0K`$ is fixed by $`T_0`$. If we choose $`\lambda T_0`$ which generates a dense subgroup of $`T_0`$, this is equivalent to $`g_0^1\lambda g_0K`$. Since $`T_0`$ is maximal abelian in $`K`$, we can assume (by multiplying $`g_0`$ on the right by $`kK`$ if necessary) that $`g_0^1\lambda g_0T_0`$. Since $`N_{G_0}(T_0)=N_K(T_0)exp(๐ฅ_0)`$, $`g_0K=a_0K`$ for some $`a_0A_0`$. Thus each symplectic leaf has exactly one $`T_0`$ fixed point. Since we are considering the leaf through the basepoint, there is just one $`T_0`$ fixed point, the basepoint.
If $`X`$ denotes the element of $`๐ฑ_0`$ corresponding to $`\delta +\mathrm{\Lambda }`$ ($`\mathrm{\Lambda }=i\lambda `$), then the Pfaffian of the infinitesimal action of $`X`$ at the basepoint equals
$$Pf(ad(X)|_๐ญ)=\delta +\mathrm{\Lambda },\alpha ,$$
$`2.30`$
where the product is over pairs of positive roots $`(\alpha ,\mathrm{\Theta }(\alpha ))`$ which are not of compact type. The Duistermaat-Heckman formula now implies $`(2.3)`$ in the case $`๐จ=1`$. This concludes the proof of $`(2.2)`$.
We end this section with two brief remarks. First, it is interesting to consider the integral
$$\psi _\mathrm{\Lambda }(g)=_{A_0\backslash G_0}๐_1(g_0g)^{2\delta 2(\mathrm{\Lambda }+\delta )}๐g_0,$$
$`2.31`$
for $`gG`$, which generalizes $`(2.8)`$. When this is well-defined, (1) this is a function of $`gG_0\backslash G/U`$, (2) this is a $`G_0`$-invariant eigenfunction for $`G`$-invariant differential operators on $`G/U`$; see Lemma 5.15 of ch2 section 5 of \[H2\] (one is usually interested in $`U`$-invariant eigenfunctions, i.e. spherical functions).
To explicitly evaluate $`(2.31)`$, first note that $`G_0exp(i๐ฑ_0)U=G`$ (this is existence of polar decomposition for the nonRiemannian symmetric space $`G_0\backslash G`$). Thus we can suppose that $`g=aexp(i๐ฑ_0)`$. In this case $`(2.31)`$ is an integral over $`A_0\backslash G_0/C_K(a)`$. One can define a Poisson structure on $`G_0/C_K(a)`$, using the Evens-Lu method, as above (see ยง3 of \[FO\]). As in $`(2.29)`$, the momentum map can be identified with $`log(๐_1(ga))`$, and the symplectic volume of a symplectic leaf can be identifed with the form $`๐_1(g_0a)^{2\delta }dV_{A_0\backslash G_0/C_K(a)}`$, via the projection to the double coset space. The fixed points for the $`T_0`$ action are of the form $`a_0wC_K(a)`$, where $`a_0A_0`$ and $`wW(K,T_0)/W(C_K(a),T_0)`$ (see Prop 4.3 of \[FO\]). Thus $`(2.31)`$ equals
$$\underset{\{w\}}{}\frac{๐_1(wa)^{2(\delta +\mathrm{\Lambda })}}{^w(\delta +\mathrm{\Lambda })^{w^1},\alpha }$$
where given $`wW(K,T_0)/W(C_K(a),T_0)`$, the product is over (1) pairs of positive roots $`(\alpha ,\mathrm{\Theta }(\alpha ))`$ which are not of compact type (relative to $`\mathrm{\Theta }`$), and (2) positive compact type roots which vanish on $`C_๐จ(a)`$.
The second remark is that there is a kind of โdualโ Poisson structure, on all of $`U/K`$, which can be used so that the sum $`(2.4)`$ has the structure of an exact stationary phase calculation. In the terminology of the paper \[EL\], in this section we used the Lagrangian splitting $`(2.15)`$, to obtain a Poisson structure on $`G_0/K`$; the โdualโ is the (Iwasawa) Lagrangian splitting $`๐ค=๐ฒ(๐ฅ_{}+๐ซ^{})`$, which induces a Poisson structure on $`U/K`$. This will hopefully be taken up elsewhere.
Appendix. Special Features of the Group Case.
In this appendix we will present a proof of $`(0.1)`$, using facts about Bott-Samelson resolutions of Schubert varieties. One rationale for including this appendix is that many of the arguments are valid in the more general context of Kac-Moody Lie algebras and groups. Throughout this appendix, we will use the notation and basic results in \[Kac\].
We start with the following data: $`A`$ is an irreducible symmetrizable generalized Cartan matrix; $`๐ค=๐ค(A)`$ is the corresponding Kac-Moody Lie algebra, realized via its standard (Chevalley-Serre) presentation; $`๐ค=๐ซ^{}๐ฅ๐ซ^+`$ is the triangular decomposition; $`๐=๐ฅ๐ซ^+`$ the upper Borel subalgebra; $`G=G(A)`$ is the algebraic group associated to $`A`$ by Kac-Peterson; $`H,N^\pm `$ and $`B`$ are the subgroups of $`G`$ corresponding to $`๐ฅ`$, $`๐ซ^\pm `$, and $`๐`$, respectively; $`K`$ is the โunitary formโ of $`G`$; $`T=KH`$ the maximal torus; and $`W=N_K(T)/TN_G(H)/H`$ is the Weyl group.
A basic fact is that $`(G,B,N_G(H))`$ with Weyl group $`W`$ is an abstract Tits system. This yields a complete determination of all the (parabolic) subgroups between $`B`$ and $`G`$. They are described as follows.
Let $`\mathrm{\Phi }`$ be a fixed subset of the simple roots. The subgroup of $`W`$ generated by the simple reflections corresponding to roots in $`\mathrm{\Phi }`$ will be denoted by $`W(\mathrm{\Phi })`$. The parabolic subgroup corresponding to $`\mathrm{\Phi }`$, $`P=P(\mathrm{\Phi })`$, is given by $`P=BW(\mathrm{\Phi })B`$.
Given $`wN_K(T)`$, we will denote its image in $`W/W(\mathrm{\Phi })`$ by $`\overline{w}`$.
The basic structural features of $`G/P`$ which we will need are the Birkhoff and Bruhat decompositions
$$G/P=\mathrm{\Sigma }_{\overline{w}},\mathrm{\Sigma }_{\overline{w}}=N^{}wP$$
$`A.1`$
$$G/P=C_{\overline{w}},C_{\overline{w}}=BwP,$$
$`A.2`$
respectively, where the indexing set is $`W/W(\mathrm{\Phi })`$ in both cases. The strata $`\mathrm{\Sigma }_{\overline{w}}`$ are infinite dimensional if $`๐ค`$ is infinite dimensional, while the cells $`C_{\overline{w}}`$ are always finite dimensional. Our main interest lies in the Schubert variety $`\overline{C}_{\overline{w}}`$, the closure of the cell.
Fix $`\overline{w}W/W(\mathrm{\Phi })`$. We choose a representative $`wN(T)`$ of minimal length $`n`$; for definiteness we will always take $`w`$ of the form
$$w=r_n\mathrm{}r_1$$
$`A.3`$
where $`r_i=i_{\alpha _i}(\left(\begin{array}{cc}0& 1\\ 1& 0\end{array}\right))`$, and $`i_{\alpha _i}:SL_2G`$ is the canonical homomorphism of $`SL_2`$ onto the root subgroup corresponding to the simple root $`\alpha _i`$.
###### (A.4) Proposition
For $`w`$ as in (A.3), the map
$$r_nexp(๐ค_{\alpha _n})\times ..\times r_1exp(๐ค_{\alpha _1})G/P:(p_j)p_n..p_1P$$
is a complex analytic isomorphism onto $`C_{\overline{w}}`$.
This result is essentially (5) of \[Kac\] together with Titsโs theory. We will sketch a proof for completeness.
###### Demonstration Proof of (A.4)
Let $`\mathrm{\Delta }^+`$ denote the positive roots, $`\mathrm{\Delta }^+(\mathrm{\Phi })`$ the positive roots which are combinations of elements from $`\mathrm{\Phi }`$. The โLie algebra of $`P`$โ is $`๐ญ=\mathrm{\Sigma }๐ค_\beta ๐`$ where the sum is over $`\beta \mathrm{\Delta }^+(\mathrm{\Phi })`$; this is the Lie algebra of $`P`$ in the sense that it is the subalgebra generated by the root spaces $`๐ค_\gamma `$ for which $`exp:๐ค_\gamma G`$ is defined and have image contained in $`P`$. The subgroups $`exp(๐ค_\gamma )`$ generate $`P`$. We also let $`๐ญ^{}`$ denote the subalgebra opposite $`๐ญ`$: $`๐ญ=๐ค_\gamma `$, where the sum is over $`\gamma \mathrm{\Delta }^+\mathrm{\Delta }^+(\mathrm{\Phi })`$. The corresponding group will be denoted by $`P^{}`$.
The cell $`C_w`$ is the image of the map $`N^+G/P:uuwP`$. The stability subgroup at $`wP`$ is $`N^+wPw^1`$.
At the Lie algebra level we have the splitting
$$๐ซ^+=๐ซ^+Ad(w)(๐ญ)๐ซ^+Ad(w)(๐ญ^{}).$$
$`A.5`$
The second summand equals
$$๐ซ_w^+=๐ค_\beta $$
$`A.6`$
where the sum is over roots $`\beta >0`$ with $`w^1\beta (\mathrm{\Delta }^+\mathrm{\Delta }^+(\mathrm{\Phi }))`$. These roots $`\beta `$ are necessarily real, so that $`exp:๐ซ_w^+N_w^+N^+`$ is well-defined.
For $`q^+`$ let $`N_q^+`$ denote the subgroup corresponding to $`๐ซ_q^+=span\{๐ค_\beta :height(\beta )q\}`$. Then $`N^+/N_q^+`$ is a finite dimensional nilpotent Lie group, and it is also simply connected. By taking $`q`$ sufficiently large and considering the splitting (A.5) modulo $`๐ซ_q^+`$, we conclude by finite dimensional considerations that each element in $`N^+`$ has a unique factorization $`n=n_1n_2`$, where $`n_1N_w^+`$ and $`n_2N^+wPw^1`$:
$$N^+N_w^+\times (n^+wPw^1).$$
$`A.7`$
The important point here is that modulo $`N_q^+`$ we can control $`N^+wPw^1`$ by the exponential map.
We now recall the following standard
###### (A.8) Lemma
In terms of the minimal factorization $`w=r_n\mathrm{}r_1`$, the roots $`\beta >0`$ with $`w^1\beta <0`$ are given by
$$\beta _j=r_n\mathrm{}r_{j+1}(\alpha _j)=r_n\mathrm{}r_j(\alpha _j),1jn.$$
Because $`w`$ is a representative of $`\overline{w}W/W(\mathrm{\Phi })`$ of minimal length, all of these $`\beta _j`$ satisfy $`w^1\beta _j(\mathrm{\Delta }^+\mathrm{\Delta }^+(\mathrm{\Phi }))`$. Otherwise, if say $`w^1\beta _j\mathrm{\Delta }^+(\mathrm{\Phi })`$, then
$$w^1r_{\beta _j}w=r_1\mathrm{}r_{j1}r_jr_{j1}\mathrm{}r_1N(T)P$$
$`A.9`$
and $`w^{}=w(w^1r_{\beta _j}w)=r_n\mathrm{}\widehat{r}_j\mathrm{}r_1`$ would be a representative of $`\overline{w}`$ of length $`<n`$ (here we have used the fact that $`W(\mathrm{\Phi })=N(T)P/T`$, which follows from the Bruhat decomposition). For future reference we note this proves that
$$N_w^+=N^+(N^{})^w=N^+(P^{})^w$$
$`A.10`$
and (2.4) shows that
$$N_w^+\times wC_{\overline{w}}.$$
$`A.11`$
Now for any $`1pqn`$, $`_{pjp}g_{\beta _j}`$ is a subalgebra of $`n_w^+`$. Thus by (2.7)
$$exp(๐ค_{\beta _n})\times \mathrm{}\times exp(๐ค_{\beta _1})\times wC_{\overline{w}}.$$
$`A.12`$
This yields (A.4) when we write
$$exp(๐ค_{\beta _j})=r_n\mathrm{}r_jexp(๐ค_{\alpha _j})r_j\mathrm{}r_n.$$
$`A.13`$
We now note several important corollaries of (A.4).
For each $`i`$, let $`P_i`$ denote the parabolic subgroup $`i_{\alpha _i}(SL_2)B`$. Let
$$\mathrm{\Gamma }_w=P_n\times _B\mathrm{}\times _BP_1/B$$
$`A.14`$
where
$$P_n\times \mathrm{}\times P_1\times B\times \mathrm{}\times BP_n\times \mathrm{}\times P_1$$
$`A.15`$
is given by
$$(p_j)\times (b_j)(p_nb_n,b_n^1p_{n1}b_{n1},\mathrm{},b_2^1p_1b_1).$$
$`A.16`$
We have written โ$`\mathrm{\Gamma }_w`$โ instead of โ$`\mathrm{\Gamma }_{\overline{w}}`$โ to indicate that this compact complex manifold depends upon the factorization.
###### (A.17) Corollary
The map
$$\mathrm{\Gamma }_w\overline{C}_{\overline{w}}:(p_j)p_n\mathrm{}p_1P$$
is a desingularization of $`\overline{C}_{\overline{w}}`$.
Let
$$SL_2^{}=\{g=\left(\begin{array}{cc}a& b\\ c& d\end{array}\right)SL(2,):a0\}.$$
$`A.17`$
###### (A.18) Corollary
Let $`\varphi `$ denote the surjective map
$$SL_2\times \mathrm{}\times SL_2\overline{C}_{\overline{w}}:(g_j)r_ni_{\alpha _n}(g_n)\mathrm{}r_1i_{\alpha _1}(g_1)P.$$
The inverse image of $`C_{\overline{w}}`$ under $`\varphi `$ is $`SL_2^{}\times \mathrm{}\times SL_2^{}`$.
###### Demonstration Proof of (A.18)
Let $`\sigma =r_{n1}\mathrm{}r_1`$. It suffices to show that for the natural actions
$$r_ni_{\alpha _n}(SL_2^{})\times C_{\overline{\sigma }}C_{\overline{w}},$$
$`A.19`$
$$r_ni_{\alpha _n}(SL_2SL_2^{})\times C_{\overline{\sigma }}\overline{C}_{\overline{w}},$$
$`A.20`$
and
$$r_ni_{\alpha _n}(SL_2)\times (\overline{C}_{\overline{\sigma }}C_{\overline{\sigma }})\overline{C}_{\overline{w}}C_{\overline{w}}.$$
$`A.21`$
The first line, $`(A.19)`$, follows from (A.4) since $`i_{\alpha _n}(SL_2^{})exp(๐ค_{\alpha _n})B`$ and $`B\times C_{\overline{\sigma }}C_{\overline{\sigma }}`$. The second line follows from
$$r_ni_{\alpha _n}\left(\begin{array}{cc}0& b\\ c& d\end{array}\right)C_{\overline{\sigma }}=i_{\alpha _n}\left(\begin{array}{cc}c& b\\ 0& d\end{array}\right)C_{\overline{\sigma }}C_{\overline{\sigma }}.$$
$`A.22`$
For the third line itโs clear that the image of the left hand side is a union of cells, since we can replace $`r_ni_{\alpha _n}(SL_2)`$ by $`P_n`$. This image is at most $`n1`$ dimensional. Therefore it must have null intersection with $`C_{\overline{w}}`$. โ
Fix an integral functional $`\lambda ๐ฅ^{}`$ which is antidominant. Denote the (algebraic) lowest weight module corresponding to $`\lambda `$ by $`L(\lambda )`$, and a lowest weight vector by $`\sigma _\lambda `$. Let $`\mathrm{\Phi }`$ denote the simple roots $`\alpha `$ for which $`\lambda (h_\alpha )=0`$, where $`h_\alpha `$ is the coroot, $`P=P(\mathrm{\Phi })`$ the corresponding parabolic subgroup. The Borel-Weil theorem in this context realizes $`L(\lambda )`$ as the space of strongly regular functions on $`G`$ satisfying
$$f(gp)=f(g)\lambda (p)^1$$
$`A.23`$
for all $`gG`$ and $`pP`$, where we have implicitly identified $`\lambda `$ with the character of $`P`$ given by
$$\lambda (u_1w\mathrm{exp}(x)u_2)=\mathrm{exp}\lambda (x)$$
$`A.24`$
for $`x๐ฅ,u_1,u_2N^+,wW(\mathrm{\Phi })`$. Thus we can view $`L(\lambda )`$ as a space of sections of the line bundle
$$_\lambda =G\times _\lambda G/P.$$
$`A.25`$
If $`๐ค`$ is of finite type, then $`L(\lambda )=H^0(_\lambda )`$; if $`๐ค`$ is affine (and untwisted), then $`L(\lambda )`$ consists of the holomorphic sections of finite energy, as in \[PS\].
Normalize $`\sigma _\lambda `$ by $`\sigma _\lambda (1)=1`$.
###### (A.26) Proposition
Let $`\overline{w}W/W(\mathrm{\Phi })`$, and let $`w=r_n\mathrm{}r_1`$ be a representative of minimal length $`n`$. The positive roots mapped to negative roots by $`w`$ are given by
$$\tau _j=r_1\mathrm{}r_{j1}(\alpha _j),1jn;$$
let $`\lambda _j=\lambda (h_{\tau _j})`$, where $`h_\tau `$ is the coroot corresponding to $`\tau `$. Then
$$\sigma _\lambda ^w(r_ni_{\alpha _n}(g_n)\mathrm{}r_1i_{\alpha _1}(g_1))=\underset{1}{\overset{n}{}}a_j^{\lambda _j}$$
where $`g=\left(\begin{array}{cc}a& b\\ c& d\end{array}\right)SL_2`$.
###### Demonstration Proof of (A.26)
The claim about the $`\tau _j`$ is easily derived from (2.5). None of these roots lie in $`\mathrm{\Delta }^+(\mathrm{\Phi })`$, by the same argument as follows (2.5). Thus each $`\lambda _j>0`$. It follows that $`\mathrm{\Pi }a_j^{\lambda _j}`$ is nonzero precisely on the set $`SL_2^{}\times \mathrm{}\times SL_2^{}`$.
Now $`\sigma _\lambda ^w`$, viewed as a section of $`_\lambda G/P`$, is nonzero precisely on the $`w`$-translate of the largest strata,
$$w\mathrm{\Sigma }_0=wP^{}P=(P^{})^wwP.$$
$`A.27`$
We claim the intersection of this with $`\overline{C}_{\overline{w}}`$ is $`C_{\overline{w}}`$. In one direction
$$C_{\overline{w}}=\left(N^+(P^{})^w\right)wP(P^{})^wwP$$
$`A.28`$
by (2.6). On the other hand $`(N^+(P^{})^w)`$ is a closed finite dimensional subgroup of $`(P^{})^w`$. Since $`(P^{})^w`$ is topologically equivalent to $`w\mathrm{\Sigma }_0`$, the limit points of $`C_{\overline{w}}`$ must be in the complement of $`w\mathrm{\Sigma }_0`$. This establishes the other direction.
It now follows from (A.4) that $`\sigma _\lambda ^w`$ is also nonzero precisely on $`SL_2^{}\times \mathrm{}\times SL_2^{}`$, viewed as a function of $`(g_n,\mathrm{},g_1)`$.
We now calculate that
$$\sigma _\lambda ^w(r_ni_{\alpha _n}(g_n)\mathrm{}r_1i_{\alpha _1}(g_1))=\sigma _\lambda (w^1r_ni_{\alpha _n}(g_n)\mathrm{}r_1i_{\alpha _1}(g_1))$$
$$=\sigma _\lambda \left(\omega _{n1}i_{\alpha _n}(g_n)\omega _{n1}^1\omega _{n2}i_{\alpha _{n1}}(g_{n1})\omega _{n2}^1\mathrm{}\omega _0i_{\alpha _1}(g_1)\omega _0^1\right)$$
$$=\sigma _\lambda \left(i_{\tau _n}(g_n)i_{\tau _{n1}}(g_{n1})\mathrm{}i_{\tau _1}(g_1)\right)$$
$`A.29`$
where we have set $`\omega _i=r_1..r_i,\mathrm{\hspace{0.33em}0}i<n`$, and we have used
$$\omega _{i1}(\alpha _i)=r_1\mathrm{}r_{i1}(\alpha _i)>0$$
$`A.30`$
to conclude that $`\omega _{i1}i_{\alpha _i}(g)\omega _{i1}^1=i_{\tau _i}(g)`$.
The map
$$SL_2\times \mathrm{}\times SL_2w^1\overline{C}_{\overline{w}}:(g_j)i_{\tau _n}(g_n)\mathrm{}i_{\tau _1}(g_1)P$$
$`A.31`$
is surjective and the inverse image of $`\mathrm{\Sigma }_0w^1\overline{C}_{\overline{w}}`$ is precisely $`SL_2^{}\times \mathrm{}\times SL_2^{}`$.
For $`g=\left(\begin{array}{cc}a& b\\ c& d\end{array}\right)SL_2^{}`$, write $`g=LDU`$, where
$$L=\left(\begin{array}{cc}1& 0\\ ca^1& 1\end{array}\right),D=\left(\begin{array}{cc}a& 0\\ 0& a^1\end{array}\right),U=\left(\begin{array}{cc}1& a^1b\\ 0& 1\end{array}\right).$$
$`A.32`$
Then for $`(g_j)SL_2^{}\times \mathrm{}\times SL_2^{}`$, (3.2) equals
$$\sigma _\lambda (i_{r_n}(L_nD_nU_n)\mathrm{}i_{\tau _1}(L_1D_1U_1))$$
$$=\sigma _\lambda (i_{\tau _n}(L_nU_n^{})i_{\tau _{n1}}(L_{n1}^{}U_{n1}^{})\mathrm{}i_{\tau _1}(L_1^{}U_1^{})i_{\tau _n}(D_n)\mathrm{}i_{\tau _1}(D_1))$$
$$=\sigma _\lambda (i_{r_n}(L_nU_n^{})\mathrm{}i_{\tau _1}(L_1^{}U_1^{}))\mathrm{\Pi }a_j^{\lambda _j}$$
$`A.33`$
where each $`L_j^{}`$ ($`U_j^{}`$) has the same form as $`L_j`$ ($`U_j`$, respectively). This follows from the fact that $`H`$ normalizes each $`exp(๐ค_{\pm r})`$.
Now each $`L_j^{}U_j^{}SL_2^{}`$, so that $`i_{\tau _n}(L_nU_n^{})\mathrm{}i_{\tau _1}(L_1^{}U_1^{})`$ is in $`\mathrm{\Sigma }_0`$. We now conclude that
$$\sigma _\lambda (i_{\tau _n}(L_nU_n^{})\mathrm{}i_{\tau _1}(L_1^{}U_1^{}))=1,$$
$`A.34`$
by the fundamental theorem of algebra, since this is polynomial and never vanishes. โ
###### (A.35) Proposition
Suppose that $`๐ค`$ is finite dimensional. Given $`gK`$ such that $`gT\mathrm{\Sigma }_1`$, we can write $`g`$ uniquely as $`g=lmau`$, where $`lN^{}`$, $`mT`$, $`aexp(๐ฅ_{})`$, and $`uN^+`$. Then
$$_Ka(g)^{i\lambda }=\underset{\alpha >0}{}\frac{2\delta ,\alpha }{2\delta i\lambda ,\alpha },$$
where the integral is with respect to the normalized Haar measure of $`K`$, and $`2\delta `$ denotes the sum of the positive complex roots.
###### Demonstration Proof of $`(A.35)`$
Let $`\{\mathrm{\Lambda }_j\}`$ denote the set of basic dominant integral functionals.
We apply $`(A.26)`$ to $`w=w_0`$. We write $`w_0=r_n..r_1`$ as in $`(A.3)`$. Then
$$gT=i_{\tau _n}(g_n)i_{\tau _{n1}}(g_{n1})\mathrm{}i_{\tau _1}(g_1)T$$
$`A.36`$
$$a(g)=\underset{j=1}{\overset{l}{}}|\sigma _{\mathrm{\Lambda }_j}(g)|^{h_j}=\underset{j=1}{\overset{l}{}}(\underset{k=1}{\overset{n}{}}|a_k|^{\mathrm{\Lambda }_j(h_{\tau _k})})^{h_j}=\underset{k=1}{\overset{n}{}}|a_k|^{h_{\tau _k}},$$
$`A.37`$
since the $`\mathrm{\Lambda }_j`$ are dual to the $`h_j`$. Therefore
$$a(g)^{i\lambda }=\underset{k=1}{\overset{n}{}}|a_k|^{i\lambda (h_{\tau _k})}.$$
$`A.38`$
Also, in terms of the coordinates $`a_k`$, the invariant measure is given by
$$a(g)^{2\delta }\underset{k}{}|a_k|^2dm(a_k),$$
$`A.39`$
up to a normalization factor.
The roots $`\tau _k`$ range over all the positive complex roots. Thus by $`(A.26)`$,
$$_Ka(g)^{i\lambda }=_{\mathrm{\Sigma }_1}a(gT)^{i\lambda }$$
$$๐ต^1\underset{\alpha >0}{}_{SU(2)}|a|^{(2\delta i\lambda )(h_\alpha )}|a|^2=๐ต^1\underset{\alpha >0}{}_0^1r^{(2\delta i\lambda )(h_\alpha )1}๐r$$
$$=๐ต^1\underset{\alpha >0}{}\frac{1}{(2\delta i\lambda )(h_\alpha )}=\underset{\alpha >0}{}\frac{2\delta ,\alpha }{2\delta i\lambda ,\alpha }.$$
This proof was given a Poisson-theoretic interpretation in \[Lu\].
References
\[EL\] S Evens and J-H Lu, On the variety of Lagrangian subalgebras, Ann Scient Ec Norm Sup 4 (34) (2001) 631-668.
\[FO\] P Foth and M Otto, A symplectic realization of Van Den Banโs convexity theorem, math.SG/0505063.
\[H1\] S Helgason, Differential Geometry, Lie Groups, and Symmetric Spaces, Academic Press (1978)
\[H2\] โโโ, Groups and Geometric Analysis, Academic Press (1984).
\[Kac\] V Kac, Constructing groups from infinite dimensional Lie algebras, Infinite Dimensional Groups with Applications, edited by V Kac, MSRI publication, Springer-Verlag (1985) 167-216.
\[Kn\] A Knapp, Lie groups beyond an introduction, 2nd edition, Birkhauser (2002).
\[Lu\] J Lu, Coordinates on Schubert cells, Kostantโs harmonic forms, and the Bruhat-Poisson structure on $`G/B`$, Transf. Groups 4, No. 4 (1999) 355-374.
\[Pi1\] D Pickrell, Invariant measures for unitary forms of Kac-Moody Lie groups, Memoirs of the AMS, Vol 146, No 693 (2000).
\[Pi2\] โโโ, An invariant measure for the loop space of a simply connected compact symmetric space, submitted to J Funct Anal.
\[PW\] E Prato and S Wu, Duistermaat-Heckman measures in a noncompact setting, Compositio Math. 94 no. 2 (1994) 113-128.
\[Wolf\] J Wolf, The action of a real semisimple group on a complex flag manifold. I: Orbit structure and holomorphic arc components, Bull A.M.S. 75 (1969) 1121-1237.
\[WZ\] J Wolf and R Zierau, Cayley transforms and orbit structure in complex flag manifolds, Transf. Groups 2 (1997) no. 4, 391-405.
|
warning/0506/math0506610.html
|
ar5iv
|
text
|
# Untitled Document
THE ALTERNATING GROUPS AND K3 SURFACES
D. -Q. Zhang
Abstract. In this note, we consider all possible extensions $`G`$ of a non-trivial perfect group $`H`$ acting faithfully on a $`K3`$ surface $`X`$. The pair $`(X,G)`$ is proved to be uniquely determined by $`G`$ if the transcendental value of $`G`$ is maximum. In particular, we have $`G/H(/(2))^2`$, if $`H`$ is the alternating group $`A_5`$ and normal in $`G`$.
Introduction
We work over the complex numbers field $``$. A K3 surface $`X`$ is a simply connected projective surface with a nowhere vanishing holomorphic 2-form $`\omega _X`$. In this note, we will consider finite groups in $`\text{Aut}(X)`$. An element $`h\text{Aut}(X)`$ is symplectic if $`h`$ acts trivially on the 2-form $`\omega _X`$. A group $`G_N\text{Aut}(X)`$ is symplectic if every element of $`G_N`$ is symplectic.
According to Nikulin \[Ni1\], Mukai \[Mu1\] and Xiao \[Xi\], there are exactly 80 abstract finite groups which can act symplectically on $`K3`$ surfaces. Among these 80, there are exactly four perfect groups ($`G`$ is perfect if the commutator subgroup $`[G,G]=G`$): $`A_5,L_2(7),A_6,M_{20}=C_2^4:A_5`$ (the Mathieu group of degree 20), where the first three are also the only non-abelian simple groups which can act on a $`K3`$ surface symplectically, and the last is the the symplectic finite group with the largest order $`960`$.
The common thing shared by the three bigger perfect groups $`G_N=L_2(7),A_6`$ and $`M_{20}`$, is that they all can be extended to a bigger group $`G=G_N.\mu _4`$ acting faithfully on a $`K3`$ surface $`X`$. Moreover, the pair $`(X,G)`$ turns out to be unique in each case, \[Ko2\], \[OZ3\], \[KOZ1\].
So one would expect that $`A_5`$, being a smaller one, should be extendable to a bigger group $`G=A_5.\mu _I`$ for some $`I3`$. However, our result below shows that this is not the case. Indeed, only $`I=1`$, or $`2`$ is possible.
Theorem A. Suppose that a finite group $`G`$ acts faithfully on a $`K3`$ surface. Suppose further that $`G`$ contains $`A_5`$ as a normal subgroup. Then $`G`$ equals one of the following four groups, each realizable (see Example 1.10):
$$A_5,S_5,A_5\times \mu _2,S_5\times \mu _2.$$
To be precise, as in (1.0) below, for every finite group $`G`$ acting on a $`K3`$ surface $`X`$, the symplectic elements of $`G`$ (i.e., those $`h`$ acting trivially on the non-zero 2-form $`\omega _X`$) form a normal subgroup $`G_N`$ such that $`G/G_N\mu _I`$ (the cyclic group of order $`I`$ in $`^{}`$). Namely, we have $`G=G_N.\mu _I`$ (see Notation below). The natural number $`I=I(G)`$ is determined by the action of $`G`$ on $`X`$ and called the transcendental value of (the action of) $`G`$.
It is proved in \[Ko2\], \[OZ3\] and \[KOZ1\] that for the three bigger perfect groups $`G_N`$ above, there is an extension $`G=G_N.\mu _I`$ such that the transcendental value $`I=I(G)`$ equals $`4`$. However, for the smaller perfect (and also simple) group $`A_5`$, we have:
Theorem B. Suppose that a finite group $`G`$ acts faithfully on a $`K3`$ surface. Suppose further that $`G`$ contains $`A_5`$ as a normal subgroup. Then the transcendental value $`I(G)`$ equals 1 or 2 (both attainable as shown in Example 1.10).
A bit more surprise comes from the next result: the existence of action by a perfect group (together with the transcendental value being $`4`$) will guarantee the existence of action by a quite large group $`G`$ as well as the uniqueness of the pair $`(X,G)`$.
Theorem C. Suppose that a finite group $`G`$ acts faithfully on a $`K3`$ surface $`X`$. Suppose further that $`G`$ contains a non-trivial perfect group $`H`$ as a subgroup (not necessarily normal). Then we have:
(1) The transcendental value $`I(G)4`$.
(2) If $`I(G)=4`$, then $`G=L_2(7).\mu _4`$, $`A_6.\mu _4`$ or $`M_{20}.\mu _4`$, and the pair $`(X,G)`$ is unique, up to isomorphisms, in all three cases.
Remark D. (1) The three subgroups $`L_2(7)`$, $`A_6`$ and $`M_{20}`$ of $`G`$ in Theorem C are all equal to $`G_N`$ in notation of (1.0), and are the only perfect groups among the 11 maximum symplectic $`K3`$ groups \[Mu1\]. So the maximality of the transcendental value $`I(G)`$ in the situation of Theorem C guarantees the maximality of the symplectic part $`G_N`$ of $`G`$. This also shows the importance of studying non-symplectic K3 groups.
(2) Regarding Theorems B and C, the readers may wonder whether the action of $`\stackrel{~}{A}_6=A_6:\mu _4`$ on a $`K3`$ surface $`X`$ induces an action of $`H.\mu _4`$ on $`X`$ with $`H=A_5`$ a smaller perfect (indeed simple) group. To elaborate, the unique group structure of $`\stackrel{~}{A}_6`$ (and also the unique pair $`(X,\stackrel{~}{A}_6)`$) is described in \[KOZ1, 2\]. In particular, the natural conjugation map $`\stackrel{~}{A}_6\text{Aut}(A_6)`$ ($`xc_x`$; see Notation below) has the Mathieu group $`M_{10}`$ as its image; therefore, the conjugation $`\mu _4`$ action switches the two different conjugacy classes of order 3 in $`A_6`$ \[CS, Ch 10, ยง1.5\]. On the other hand, for $`\stackrel{~}{A}_6`$ to contain an $`A_5.\mu _4`$, the conjugation $`\mu _4`$ action should stabilize at least one $`A_5`$ in $`A_6`$ and also preserve the unique conjugacy class of order 3 in this $`A_5`$, which is impossible.
(3) The same construction in \[OZ3, Appendix\] shows that there is a smooth non-isotrivial family of $`K3`$ surfaces $`f:๐ณ^1`$ such that all fibres admit $`A_6`$ actions and infinitely many of them are algebraic K3 surfaces. So, the symplectic part alone can not determine the surface uniquely, and the study of transcendental value is needed.
The main tools of the paper are the Lefschetz fixed point formula (both the topological version and vector bundle version due to Atiyah-Segal-Singer \[AS2, 3\]), the representation theory on the $`K3`$ lattice and the study in \[Z2\] on automorphism groups of Niemeier lattices (in connection with Golay binary or ternary codes) where the latter is much inspired by Conway-Sloane \[CS\], Kondo \[Ko1\] and Mukai \[Mu2\].
The reduction to the study of automorphisms of Niemeier lattices was pioneered by Nikulin (see e.g., \[Ni3, end of section 1.14\]) and further developed by Kondo (see e.g. \[Ko1\]).
We believe that the way of combining different very powerful machinaries to deduce results as done in the paper should be applicable to the study of other problems. Our humble paper also demonstrates the powerfulness and depth of these algebraic results in the study of geometry. The information we compute in Proposition 1.4 (and its generalization in the future) should be of independent interest and use in understanding the geometry of $`K3`$ surfaces.
Note.Mapleโ was used in solving the linear equations in the crucial Proposition 1.4. We refer to Shimada \[Sh1, Sh2, Sh3\] for more computations in Algebraic Geometry.
Notation.
1. When we write $`G=G_N.\mu _I`$ we mean that $`G`$ acts on a $`K3`$ surface $`X`$ satisfying the situation in (1.0) below.
2. $`S_n`$ is the symmetric group in $`n`$ letters, $`A_n`$ ($`n3`$) the alternating group in $`n`$ letters and $`\mu _I=\text{exp}(2\pi \sqrt{1})/I`$ the multiplicative group of order $`I`$ in $`^{}`$.
3. For a group $`G`$, we write $`G=A.B`$ if $`A`$ is normal in $`G`$ so that $`G/A=B`$. We write $`G=A:B`$ if assume further that $`A`$ is normal in $`G`$ and $`B`$ is a subgroup of $`G`$ such that the composition $`BGG/A=B`$ is the identity (we say then that $`G`$ is a semi-direct product of $`A`$ and $`B`$).
4. For groups $`HG`$ and $`xG`$ we denote by $`c_x:HG`$ ($`hc_x(h)=x^1hx`$) the conjugation map.
5. For a $`K3`$ surface $`X`$, we let $`S_X`$ and $`T_X`$ be the Neron-Severi and transcendental lattices. So the $`K3`$ lattice $`H^2(X,)`$ contains $`S_XT_X`$ as a sublattice of finite index.
Acknowledgement. This work was done during the authorโs visits to Hokkaido University, University of Tokyo and Korea Institute for Advanced Study in the summer of 2004. The author would like to thank the institutes and Professors I. Shimada, K. Oguiso and J. Keum for the support and warm hospitality.
ยง1. Preparations and Examples
(1.0). In this section, we will prepare some basic results to be used late. Let $`X`$ be a $`K3`$ surface with a non-zero 2-form $`\omega _X`$ and let $`G\text{Aut}(X)`$ be a finite group of automorphisms. For every $`hG`$, we have $`h^{}\omega _X=\alpha (h)\omega _X`$ for some scalar $`\alpha (h)^{}`$. Clearly, $`\alpha :G^{}`$ is a homomorphism. A fact in basic group theory says that $`\alpha (G)`$ is a finite cyclic group, so $`\alpha (G)=\mu _I=\text{exp}(2\pi \sqrt{1}/I)`$ for some $`I1`$. This natural number $`I=I(G)`$ is called the transcendental value of $`G`$. It is known that $`I=I(G)`$ for some $`G`$ if and only if that the Euler function $`\phi (I)21`$ and $`I60`$ \[MO\].
Set $`G_N=\text{Ker}(\alpha )`$. Then we have the basic exact sequence below:
$$1G_NG\stackrel{๐ผ}{}\mu _I1.$$
For the $`G`$ in the basic exact sequence, we write $`G=G_N.\mu _I`$, though there is no guarantee that $`G=G_N:\mu _I`$ (a semi-direct product).
Fact 1.0A. If $`G`$ is a finite perfect group, i.e., the commutator group $`[G,G]=G`$ (especially, if $`G`$ is a non-abelian simple group like $`A_5`$), then $`G=G_N`$.
1.0B. $`G_N`$ acts trivially on the transcendental lattice $`T_X`$ (Lefschetz theorem on $`(1,1)`$-classes).
1.0C. If a subgroup $`HG_N`$ fixes a point $`P`$, then $`H<SL(T_{X,P})SL_2()`$ \[Mu1, (1.5)\]. The finite subgroups of $`SL_2()`$ are listed up in \[Mu1, (1.6)\]. These are cyclic $`C_n`$, binary dihedral (or quaternion) $`Q_{4n}`$ ($`n2`$), binary tetrahedral $`T_{24}`$, binary octahedral $`O_{48}`$ and binary icosahedral $`I_{120}`$.
Lemma 1.1. Suppose that $`G:=A_5.\mu _I`$ (with $`G_N=A_5`$) acts faithfully on a $`K3`$ surface $`X`$.
(1) The Picard number $`\rho (X)19`$, and $`I=1,2,3,4,6`$. Moreover, $`\rho (X)=20`$ if $`I3`$.
(2) We have $`G=A_5:\mu _I`$, i.e., a semi-prodcut of a normal subgroup $`A_5`$ and a subgroup $`\mu _I`$ of $`G`$. Moreover, $`G=A_5\times \mu _I`$ if $`I=3`$.
Proof. (1) In notation of \[Xi, the list\], $`\rho (X)=\text{rank}S_Xc+1=19`$. Also the Euler function $`\phi (I)`$ divides $`\text{rank}T_X=22\rho (X)`$ by \[Ni1, Theorem 0.1\]. So (1) follows.
(2) Let $`gG`$ such that $`\alpha (g)`$ is a generator of $`\mu _I`$. Since $`\text{Aut}(A_5)=S_5>A_5`$ and the conjugation homomorphism $`A_5\text{Aut}(A_5)`$ ($`xc_x`$) is an isomorphism onto $`A_5`$, the conjugation map $`c_g`$ equals $`c_{(12)a}`$ or $`c_a`$ on $`A_5`$ for some $`aA`$. Replacing $`g`$ by $`ga^1`$, we may assume that $`c_g=c_{(12)}`$ or $`c_{\text{id}}`$. Thus $`g^2`$ commutes with every element in $`A_5`$. If $`2|I`$, then $`g^I\text{Ker}(\alpha )=A_5`$ is in the centre of $`A_5`$ (which is trivial) and hence $`\text{ord}(g)=I`$; thus $`G=A_5:\mu _I`$. If $`I=3`$, then $`\text{gcd}(3,\text{ord}(g)/3)=1`$ as proved in \[IOZ\] or \[Og, Proposition 5.1\]; so replacing $`g`$ by $`g^{\mathrm{}}`$ with $`\mathrm{}=\text{ord}(g)/3`$ (or $`2\text{ord}(g)/3`$), we have $`G=A_5\times g=A_5\times \mu _3`$.
The third result below \[Ni1, ยง5\] is crucial in classifying symplectic groups in \[Mu1\]. The second uses the fact $`A_5\text{Aut}(X)`$ in an essential way.
Lemma 1.2. (1) Let $`h`$ be a non-symplectic involution on a $`K3`$ surface $`X`$. Then $`X^h`$ is a disjoint union of $`s`$ smooth curves $`C_i`$ with $`0s10`$. To be precise, $`X^h`$ (if not empty) is either a disjoint union of a genus $`2`$ curve $`C`$ and a few $`^1`$โs, or a disjoint union of a few elliptic curves and $`^1`$โs, or a disjoint union of a few $`^1`$โs.
(2) For $`h`$ in (1), suppose further that $`A_5\text{Aut}(X)`$. Then $`\chi _{\text{top}}(X^h)18`$.
(3) If $`\delta `$ is a non-trivial symplectic automorphism of finite order on a $`K3`$ surface $`X`$, then $`\text{ord}(\delta )8`$ and $`X^\delta `$ is a finite set. To be precise, if $`\text{ord}(\delta )=2,3,4,5,6,7,8`$, then $`|X^\delta |=8,6,4,4,2,3,2`$, respectively; see \[Ni1, ยง5\] for the proof. In particular, if $`A_5\text{Aut}(X)`$ then $`_{\delta A_5}\chi _{\text{top}}(X^\delta )=360`$ (see (1.0A)).
Proof. (1) Locally, at a point $`PX^h`$, we have $`h|P:(x,y)(x,y)`$ for some coordinates around $`P`$, because $`h`$ is non-symplectic. Thus around $`P`$, our $`X^h=\{y=0\}`$ which is smooth. For the range of $`s`$, see \[Ni2\] or \[Z1\]. If $`X^h`$ contains a genus $`2`$ curve $`C`$, then the big and nefness of $`C`$ and the Hodge index theorem show that the other $`s1`$ curves are negative definite, whence are $`^1`$โs. So (1) is true.
(2) Let $`X^h=_{i=1}^sC_i`$ be as in (1). Then $`\chi _{\text{top}}(X^h)=_{i=1}^s(22g(C_i))2s20`$. If (2) is false, then $`s=10`$ and $`C_i^1`$. Thus, by \[OZ1, Theorem 4\], $`X`$ equals $`X_4`$: the unique $`K3`$ surface of Picard number $`\rho (X)=20`$ and $`|\text{Pic}X|=4`$. Now $`A_5\text{Aut}(X_4)`$, where the latter is given in \[Vi\]. This is impossible by the simplicity of $`A_5`$ and the precise description of $`\text{Aut}(X_4)`$ there (see the proof of \[KOZ1, Prop 4.1 (3)\]).
For an automorphism $`h`$ on a smooth algebraic surface $`Y`$, we split the pointwise fixed locus as the disjoint union of 1-dimensional part and the isolated part: $`Y^h=Y_{1\text{dim}}^hY_{\text{isol}}^h`$. The proof of (1) below is similar to that for (1) in (1.2).
Fact 1.3. (1) $`Y_{1\text{dim}}^h`$ (if not empty) is a disjoint union of smooth curves.
(2) The Euler number $`\chi _{\text{top}}(Y_{1\text{dim}}^h)=_C(22g(C))=2n_h`$ for some integer $`n_h`$, where $`C`$ runs in the set $`Y_{1\text{dim}}^h`$ of curves.
(3) The Euler number $`\chi _{\text{top}}(Y^h)=m_h+2n_h`$, where $`m_h=|Y_{\text{isol}}^h|`$.
The results of \[IOZ\] below follow from the application of Lefschetz fixed point formula to the trivial vector bundle in Atiyah-Segal-Singer \[AS2, AS3, pages 542 and 567\]. The results themselves should be very useful and informative for other studies in the future.
Important Proposition 1.4. Let $`X`$ be a $`K3`$ surface and let $`h\text{Aut}(X)`$ be of order $`I`$ such that $`h^{}\omega _X=\eta _I\omega _X`$ for some primitive $`I`$-th root $`\eta _I`$ of 1.
(1) Suppose that $`I=3`$. Then $`m_h=3+n_h`$ and hence $`\chi _{\text{top}}(X^h)=3(1+n_h)`$. Moreover, $`3n_h6`$.
(2) Suppose that $`I=4`$. Then $`m_h=4+2n_h`$ and hence $`\chi _{\text{top}}(X^h)=4(1+n_h)`$. Moreover, $`2n_h4`$.
(3) Suppose that $`I=3`$, or $`4`$. If $`\delta \text{Aut}(X)`$ is symplectic of order 5 and commutes with $`h`$. Then $`|X^{h\delta }|=4`$.
(4) Suppose that $`I=4`$. If $`\delta \text{Aut}(X)`$ is symplectic of order 3 and commutes with $`h`$ then $`6|X^{h^2\delta }||X^{h\delta }|\{2,4,6\}`$.
Proof. (1) The first part is proved in \[OZ1, Lemma 2.3\]. Note that $`h^{}|T_X`$ can be diagonalized as $`\text{diag}[\eta _3,\eta _3^2]^s`$ ($`s1`$) by \[Ni1, Theorem 0.1\]. So as in (1.7) below, $`\chi _{\text{top}}(X^h)=2+\text{Tr}(h^{}|T_X)+\text{Tr}(h^{}|S_X)2s+\text{rank}S_X21`$, whence $`n_h6`$. Also $`m_h0`$ implies that $`n_h3`$.
(2) As in \[OZ1, Lemma 2.3\], we calculate the holomorphic Lefschetz number $`L(h)`$ in two ways as in \[AS2, 3, pages 542 and 567\], where $`X_{\text{isol}}^h=\{P_j|1jm_h\}`$ (so $`h^{}|T_{P_j}=(\eta _4^1,\eta _4^2)`$ up to switching the coordinates of the tangent plane at $`P_j`$), $`X_{\text{1-dim}}^h=\{C_k\}`$, $`gC_k=g(C_k)`$ the genus, and $`\eta _4^1`$ the eigenvalue of the action $`h_{}`$ on the normal bundle of $`C_k`$ (in the first equation below we used Serre duality, while the last is from the first two with $`x=\eta _4`$):
$$\begin{array}{c}L(h)=\underset{i=0}{\overset{2}{}}(1)^i\text{Tr}(h^{}|H^i(X,๐ช_X))=1+\eta _4^1,\\ L(h)=\underset{j=1}{\overset{m_h}{}}a(P_j)+\underset{k}{}b(C_k),\\ a(P_j)=1/\text{det}(1h^{}|T_{P_j})=1/(1\eta _4^1)(1\eta _4^2),\\ b(C_k)=(1gC_k)/(1\eta _4)\eta _4C_k^2/(1\eta _4)^2=(1gC_k)(1+\eta _4)/(1\eta _4)^2,\\ 0=(1+x^1)+m_h/(1x^1)(1x^2)+n_h(1+x)/(1x)^2.\end{array}$$
Noting that $`x=\eta _4`$ satisfies $`x^2=1`$ and solving the last equation, we get $`m_h=4+2n_h`$. The second part of (2) is similar to (1), noting that $`h^{}|T_X`$ can be diagonalized as $`\text{diag}[\eta _4,\eta _4]^s`$ ($`s1`$).
(3) $`\&`$ (4). In (4), note that $`X^{h^i\delta }=X^{h^i}X^\delta `$ ($`i=1,2`$). So the inequalities there hold and we have only to calculate $`|X^{h\delta }|`$; see (1.2).
Let $`g\text{Aut}(X)`$ such that $`\text{ord}(g)=kI`$ and $`g^{}\omega _X=\eta ^k\omega _X`$ where $`\eta =\eta _{kI}`$ is a primitive $`kI`$-th root of 1. (We set $`g=h\delta `$ in (3) and (4).) If $`k2`$ and $`\text{gcd}(k,I)=1`$ (these are true in (3) and (4)), then $`g^I`$ is of order $`k`$ and symplectic, so $`X^gX^{g^I}`$ is a finite set by (1.2). Namely, $`X^g=X_{\text{isol}}^g=\{P_j|1jm_g\}`$ say. Let $`M_g(i)`$ be the set of points $`P`$ in $`X^g`$ satisfying $`g^{}|T_P=(\eta ^i,\eta ^{k+i})`$ (up to switching the coordinates of the tangent plane at $`P`$; so $`a(P)=1/(1\eta ^i)(1\eta ^{k+i})`$ in the notation for the formula of $`L(g)`$). Put $`m_g(i)=|M_g(i)|`$. Then for $`(I,k)=(3,5)`$ (the first case in (3)), we have $`X^g=M_g(i)`$ and $`m_g=_im_g(i)`$, where $`i\{1,\mathrm{},4,11,12\}`$; for $`(I,k)=(4,5)`$ (the second case in (3)), we have $`m_g=_im_g(i)`$, where $`i\{1,\mathrm{},4,6,7,16,17\}`$; for $`(I,k)=(4,3)`$ (the case in (4)), we have $`m_g=_im_g(i)`$, where $`i\{1,2,4,10\}`$.
As in (2), we have the following, where $`x=\eta =\eta _{kI}`$ and $`i`$ runs in the set specified above:
$$()0=(1+x^k)+\underset{i}{}\underset{PM_g(i)}{}a(P)=(1+x^k)+\underset{i}{}m_g(i)/(1x^i)(1x^{k+i}).$$
For $`(I,k)=(3,5)`$, $`x`$ satisfies the minimal polynomial $`\mathrm{\Phi }_g(x)=1x+x^3x^4+x^5x^7+x^8`$ and also $`x^{15}=1`$, $`x^{10}=1x^5`$. Substituting these into the equation (\*) multiplied by the common denomenator (which is not zero), we will get an equation of degree $`7`$ in $`x`$ with coefficients linear in $`m_g(i)`$. The minimality of $`\mathrm{\Phi }_g(x)`$ implies that all 8 coefficients are zero. Solving these 8 linear equations, we obtain, where $`m_i=m_g(i)`$:
$$()m_1=m_4,m_2=1+m_3,m_{11}=1+m_4,m_{12}=m_3.$$
By (1.2), we have $`4=m_{g^3}m_g=_{i=1}^4m_i+_{i=11}^{12}m_i=2+3(m_3+m_4)`$. So $`m_3+m_42`$. This together with the condition $`m_i0`$ and the relations in (\**), imply that $`[m_1,m_2,m_3,m_4,m_{11},m_{12}]=[1,0,1,1,0,1]`$. In particular, $`m_g=4`$.
For $`(I,k)=(4,5)`$, $`x`$ satisfies the minimal polynomial $`\mathrm{\Phi }_g(x)=1x^2+x^4x^6+x^8`$ and also $`x^{20}=1`$, $`x^{10}=1`$. As above, solving $`()`$, we obtain, where $`m_i=m_g(i)`$:
$$\begin{array}{c}()m_1=3+2m_33m_4+4m_62m_7,m_2=1+m_32m_4+2m_6,\\ m_{16}=5+2m_34m_4+5m_62m_7,m_{17}=3+2m_42m_6+m_7.\end{array}$$
One can check that the following is the only possibility of $`m_i`$ satisfying the relations in $`()`$ and that $`0m_im_gm_{g^4}=4`$ by (1.2); in particular, $`m_g=4`$:
$$[m_1,m_2,m_3,m_4,m_6,m_7,m_{16},m_{17}]=[1,1,0,0,1,0,0,1].$$
For $`(I,k)=(4,3)`$, $`x`$ satisfies the minimal polynomial $`\mathrm{\Phi }_g(x)=1x^2+x^4`$ and also $`x^{12}=1`$, $`x^6=1`$. As above, solving $`()`$, we obtain, where $`m_i=m_g(i)`$:
$$()m_1=3+3m_22m_4,m_{10}=1+2m_2m_4.$$
One can check that the following are the only possibilities of $`m_i`$ satisfying the relations in $`()`$ and $`0m_im_gm_{g^4}=6`$, (1.2); in particular, $`m_g=2,4,6`$ (so (1.4) is done):
$$[m_1,m_2,m_4,m_{10}]=[3,0,0,1],[1,0,1,0],[2,1,2,1],[0,1,3,0].$$
The following two results can be found in \[Ni1, Theorem 0.1\], \[MO, Lemma (1.1)\], or \[OZ3, Lemma (2.8)\].
Lemma 1.5. Suppose that $`X`$ is a $`K3`$ surface of Picard number $`\rho (X)=20`$ and $`g`$ an order-4 automorphism such that $`g^{}\omega _X=\eta _4\omega _X`$ with a primitive 4-th root $`\eta _4`$ of 1. Then we can express the transcendental lattice $`T_X`$ as $`T_X=[t_1,t_2]`$ so that $`t_2=g^{}(t_1)`$, $`g^{}(t_2)=t_1`$. In particular, the intersection form $`(t_i.t_j)=\text{diag}[2m,2m]`$ for some $`m1`$.
Now we assume that $`G=G_N.\mu _I`$ (with $`I=I(G)`$) acts on a $`K3`$ surface $`X`$. When $`G_N=A_5`$, we will determine the action of $`G_N`$ on the Neron Severi lattice $`S_X`$ of $`X`$:
Lemma 1.6. (1) Suppose that $`A_5`$ acts on a $`K3`$ surface $`X`$, and $`\text{rank}S_X=20`$ (this is true if $`I3`$ by (1.1)). Then we have the irreducible decomposition below (in the notation of Atlas for the characters of $`A_5`$), where $`\chi _1`$ (the trivial character), $`\chi _4`$ and $`\chi _5`$ have dimensions 1, 4 and 5, respectively, where $`\chi _i^{}`$ is a copy of $`\chi _i`$:
$$S_X=\chi _1\chi _1^{}\chi _4\chi _4^{}\chi _5\chi _5^{}.$$
(2) For conjugacy class $`nA`$ (and $`nB`$) of order $`n`$ in $`A_5`$ and the characters $`\chi _i`$ of $`A_5`$, we have the following Table 1 from \[Atlas\], where $`Z`$ is respectively $`1A`$, $`2A`$, $`3A`$, $`5A`$ or $`5B`$:
$$\begin{array}{c}[\chi _1,\chi _2,\chi _3,\chi _4,\chi _5](Z)=[1,3,3,4,5],[1,1,1,0,1],[1,0,0,1,1],\\ [1,(1\sqrt{5})/2,(1+\sqrt{5})/2,1,0],[1,(1+\sqrt{5})/2,(1\sqrt{5})/2,1,0].\end{array}$$
Proof. Applying the Lefschetz fixed point formula to the action of $`A_5`$ on $`H^{}(X,)=_{i=0}^4H^i(X,)`$ and noting that $`H^2(X,)`$ contains $`S_XT_X`$ as a finite index sublattice, we obtain (see also (1.0A-B) and (1.2)):
$$2+\text{rank}T_X+\text{rank}(S_X)^{A_5}=\text{rank}H^{}(X,)^{A_5}=\frac{1}{|A_5|}\underset{aA_5}{}\chi _{\text{top}}(X^a)=360/60=6.$$
Thus $`\text{rank}S_X^{A_5}=2`$. So the irreducible decomposition is of the following form, where $`a_i`$ are non-negative integers:
$$S(X)=2\chi _1a_2\chi _2a_3\chi _3a_4\chi _4a_5\chi _5.$$
As in (1.7) below, using the topological Lefschetz fixed point formula, the fact that $`\text{rank}T(X)`$ $`=2`$ and (1.0B), we have, for $`aA_5`$, that:
$$\chi _{\text{top}}(X^a)=4+\text{Tr}(a^{}|S(X))$$
Running $`a`$ through the five conjugacy classes and calculating both sides, using (1.2) and the character Table 1 in (2), we obtain the following system of equations:
$$\begin{array}{c}20=2+3(a_2+a_3)+4a_4+5a_5,\\ 4=2(a_2+a_3)+a_5,\\ 2=2+a_4a_5,\\ 0=2+\frac{1\sqrt{5}}{2}a_2+\frac{1+\sqrt{5}}{2}a_3a_4,\\ 0=2+\frac{1+\sqrt{5}}{2}a_2+\frac{1\sqrt{5}}{2}a_3a_4.\end{array}$$
Now, we get the result by solving this system of Diophantine equations.
(1.7). Note that $`\text{Aut}(A_5)=S_5`$. For a group $`G=A_5.\mu _I`$ (and the map $`\alpha `$) in (1.0), we have the natural homomorphism below, which is injective (since its restriction on $`A_5`$ is an injection onto $`A_5\times (1)`$), where $`c_x:ac_x(a)=x^1ax`$ is the conjugation map:
$$\begin{array}{c}G\text{Aut}(A_5)\times \mu _I=S_5\times \mu _I,\\ x(c_x,\alpha (x)).\end{array}$$
Lemma. Suppose that $`G=A_5.\mu _4`$ acts on a $`K3`$ surface $`X`$ (i.e., $`G_N=A_5`$ and $`I(G)=4`$). Then $`G=A_5:\mu _4`$, but $`GA_5\times \mu _4`$. Our $`GS_5\times \mu _4`$ ($`(x(c_x,\alpha (x))`$) is an injective homomorphism and the group structure of such $`G`$ is unique up to isomorphisms.
Proof. By (1.1), we have $`G=A_5:\mu _4`$. Suppose the contrary $`G=A_5\times \mu _4`$. Write $`\mu _4=`$ $`g`$. In notation of (1.6), the $`g`$ either stabilizes $`\chi _i`$ or swtiches $`\chi _i`$ with $`\chi _i^{}`$ ($`i=4`$ or $`5`$; then denoted as $`\chi _i\stackrel{๐}{}\chi _i`$, and $`\text{Tr}(g^{}|(\chi _i\chi _i^{}))=0`$)). Since $`G`$ stabilizes an ample line bundle (the pull back of an ample line bundle on $`X/G`$) and since $`G`$ acts on $`S_X^{A_5}`$ (whose $``$-extension is $`\chi _1\chi _1^{}`$), we may assume that $`g^{}|(\chi _1\chi _1^{})=\text{diag}[1,\pm 1]`$ w.r.t. to a suitable basis. If $`\chi _i`$ is $`g`$-stable then $`g^{}|\chi _i`$ is a scalar $`\zeta _4^c`$ with $`\zeta _4=\text{exp}(2\pi \sqrt{1}/4)`$, by Schurโs lemma.
Let $`aA_5`$. Then $`(ga)^{}|T_X=g^{}|T_X`$ (see (1.0B)) and the latter can be diagonalized as $`\text{diag}[\zeta _4,\zeta _4^1]`$ by \[Ni1, Theorem 0.1\] and (1.1). Hence $`\text{Tr}(ga)^{}|T_X=0`$. By the topological Lefschetz fixed point formula and noting that $`H^2(X,)`$ contains $`S_XT_X`$ as a sublattice of finite index, we have $`\chi _{\text{top}}(X^{ga})=_{i=0}^4\text{Tr}(ga)^{}|H^i(X,)`$ $`=2+\text{Tr}(ga)^{}|S_X+\text{Tr}(ga)^{}|T_X=2+\text{Tr}(ga)^{}|S_X`$. For $`a=5A`$ (an order-5 element) in $`A_5`$, by (1.4) and Table 1 in (1.6)) (and Schurโs lemma), we have: $`4=\chi _{\text{top}}(X^{g5A})=2+\text{Tr}(g^{}|\chi _1\chi _1^{})+\text{Tr}(g5A)^{}|(\chi _4\chi _4)+0`$, so one of the following cases occurs (using Schurโs lemma):
Case(i). $`g^{}|S_X=\text{diag}[1,1,I_4,I_4,\mathrm{?},\mathrm{?}]`$,
Case(ii). $`g^{}|S_X=\text{diag}[1,1,\chi _4\stackrel{๐}{}\chi _4,\mathrm{?},\mathrm{?}]`$,
Case(iii). $`g^{}|S_X=\text{diag}[1,1,I_4,I_4,\mathrm{?},\mathrm{?}]`$,
Case(iv). $`g^{}|S_X=\text{diag}[1,1,\zeta _4I_4,\zeta _4^1I_4,\mathrm{?},\mathrm{?}]`$.
By (1.4), we have $`():4\chi _{\text{top}}(X^g)=4(1+n_g)=0`$ (mod 4) with $`2n_g4`$. So $`\chi _{\text{top}}(X^g)=4`$ in Cases (ii), (iii) and (iv) (using Schurโs lemma). Thus $`n_g=0`$ and $`m_g=4+2n_g=4`$ by (1.4). Now $`A_5`$ (commuting with $`g`$) acts on the four isolated points $`P_i`$ in $`X^g`$, whence fixing these four points (see (1.8) below). So $`A_5<SL(T_{X,P_1})`$, contradicting (1.0C). In Case(i), by the fact (\*) above and Schurโs lemma, we have $`\chi _{\text{top}}(X^g)=2+(1144+5+5)=4`$, which will lead to the same contradiction.
By the proof of (1.1) and the result in the above paragraph, we may assume that there is an order-4 element $`gG`$ such that $`\alpha (g)`$ is the generator of $`\mu _4`$, so that $`G=A_5:g=A_5:\mu _4`$ and the conjugation map $`c_g=c_{(12)}`$ on $`A_5`$. Clearly, the group structure of $`G`$ is unique. The lemma is proved.
The two results below are either easy or well known and will be frequently used in the arguments of the subsequent sections.
Lemma 1.8. Let $`f:A_5S_r`$ ($`r2`$) be a homomorphism.
(1) If $`r=2`$, $`3`$, or $`4`$, then $`f`$ is trivial.
(2) If $`\text{Im}(f)`$ is a transitive subgroup of the full symmetry group $`S_r`$ in $`r`$ letters $`\{1,2,\mathrm{},r\}`$ (whence $`r5`$ by (1)), then $`r||A_5|`$ with $`|A_5|/r`$ equal to the order of the subgroup of $`A_5`$ stabilizing the letter 1, so $`r\{5,6,10,12,15,20,30\}`$.
Lemma 1.9. (1) $`\text{Aut}(^1)`$ includes $`A_5`$ but not $`S_5`$ \[Su, Theorem 6.17\].
(2) If $`\text{id}f\text{Aut}(^1)`$ is an automorphism of finite order, then $`f`$ fixes exactly two point of $`^1`$ (by the diagonalization of a lifting of $`f`$ to $`GL_2()`$).
(3) If $`f_r`$ ($`r=2`$ or $`3`$) is an order$`r`$ automorphism of an elliptic curve $`E`$, then either $`f_r`$ acts freely on $`E`$, or the fix locus satisfies $`|X^{f_r}|=4`$ (resp. $`=3`$) if $`r=2`$ (resp. $`r=3`$) (by the Hurwitz formula).
The examples below are to show the existence of the groups in Theorems A and B.
Example 1.10. (1) $`G=G_N.\mu _I=S_5\times \mu _2`$ (with $`G_N=S_5`$, $`I=2`$) acts on a $`K3`$.
Let $`X=\{_{i=1}^5X_i=_{i=1}^6X_i^2=_{i=1}^5X_i^3=0\}^5`$. We define the symplectic action of $`\sigma S_5`$ on $`X`$ (the same as in \[Mu1, $`n^{}3`$\]) and a non-symplectic involution $`g`$ on $`X`$ as follows (see \[Mu1, Lemma 2.1\]):
$$\begin{array}{c}\sigma :[X_1:\mathrm{}:X_6][X_{\sigma (1)}:\mathrm{}:X_{\sigma (5)}:(\text{sign}\sigma )X_6],\\ g:[X_1:\mathrm{}:X_6][X_1:\mathrm{}:X_5:X_6].\end{array}$$
Let $`G=S_5,g`$. Then $`G=S_5\times g`$ is the required one.
(2) $`G=G_N.\mu _I=A_5:\mu _2=S_5`$ (with $`G_N=A_5`$, $`I=2`$) acts on a $`K3`$ surface.
Let $`X=\{_{i=1}^6X_i=_{i=1}^6X_i^2=_{i=1}^6X_i^3=0\}^5`$. We define the action of $`\sigma S_6`$ on $`X`$ (the same as in \[Mu1, $`n^{}2`$\]):
$$\sigma :[X_1:\mathrm{}:X_6][X_{\sigma (1)}:\mathrm{}:X_{\sigma (6)}].$$
Since $`A_6`$ is perfect, its action on $`X`$ is symplectic (1.0A). If we let $`\stackrel{~}{G}=S_6`$, then $`\stackrel{~}{G}=\stackrel{~}{G}_N.\mu _2`$ with $`\stackrel{~}{G}_N=A_6`$ and $`I=2`$ (see \[Mu1, Lemma 2.1\]). Now a subgroup $`G=S_5`$ of $`\stackrel{~}{G}`$ is the required one.
ยง2. The determination of some topological invariants
Let $`X`$ be a $`K3`$ surface with a faithful action by a group of the form $`G:=A_5.\mu _4`$ as in (1.0). Then $`G=A_5:\mu _4`$ and the unique group structure of such $`G`$ is given in (1.7).
We will use the notation in (1.6). Let $`g`$ be a generator of $`\mu _4<G`$. We may also assume the following is true (after a change of $`g`$):
Lemma 2.1. (1) The conjugation action $`c_g(.)=c_{(12)}(.)`$ on $`A_5`$. So $`g^2`$ is in the centre of $`G`$ and $`G\text{Aut}(A_5)=S_5`$ ($`xc_x`$) induces an isomorphism $`G/g^2S_5`$.
(2) $`g^{}\omega _X=\zeta _4\omega _X`$ with $`\zeta _4=\text{exp}(2\pi \sqrt{1}/4)`$.
(3) $`g^2`$ is a non-symplectic involution on $`X`$ and commutes with every element in $`A_5`$.
(4) Set $`\sigma =(12)(34)`$ and $`\tau =(345)`$. Then $`g`$ commutes with every element in $`\sigma ,\tau =S_3`$. So $`G=A_5:\mu _4>S_3\times \mu _4`$.
(5) Set $`\sigma =(12)(34)`$, $`\gamma =(123)`$. Then $`g`$ normalizes $`\sigma ,\gamma =A_4`$. So $`G=A_5:\mu _4>A_4:\mu _4`$. Set $`\sigma _1=\sigma `$ and $`\sigma _2=(13)(24)`$ (all in $`A_4`$).
(6) $`g`$ stabilizes both $`\chi _1`$ and $`\chi _1^{}`$; the restrictions $`g^{}|\chi _1=\text{id}`$ and $`g^{}|\chi _1^{}=\pm \text{id}`$ (after a change of basis).
(7) $`g`$ either stabilizes both $`\chi _4`$ and $`\chi _4^{}`$ (so the restrictions of $`g^{}`$ on $`\chi _4`$ and $`\chi _4^{}`$ are scalar multiplications), or switches $`\chi _4`$ with $`\chi _4^{}`$.
(8) $`g`$ either stabilizes both $`\chi _5`$ and $`\chi _5^{}`$ (so the restrictions of $`g^{}`$ on $`\chi _5`$ and $`\chi _5^{}`$ are scalar multiplications), or switches $`\chi _5`$ with $`\chi _5^{}`$.
(9) Both $`g^2|\chi _i`$ and $`g^2|\chi _i^{}`$ ($`i=4,5`$) are scalar multiplications.
Proof. (1) is from the last part of the proof of (1.7). The (2) is true because $`g`$ is a generator of $`\mu _4<G=A_5:\mu _4`$. The (3), (4) and (5) follow from (1). The (6) is true because $`G=A_5:g`$ stabilizes one ample line bundle (the pull back of an ample line bundle on $`X/G`$) and $`g`$ acts on $`S_X^{A_5}`$ (defined over $``$) whose $``$-extension is $`\chi _1\chi _1^{}`$. (7), (8) and (9) are from the form of the decomposition in (1.6) and Schurโs lemma.
In the rest of the section, we will prove the Key result (2.2) below which will be used in the proof of Theorems A, B and C in ยง5 and is the consequence of (2.6)-(2.9) below. The representation theory (mainly on $`A_5`$) is fully applied. We divide into cases according to whether $`g`$ stablilizes or switches $`\chi _i`$ ($`i=4,5`$).
Key Proposition 2.2. Suppose that $`G=A_5:\mu _4`$ acts on a $`K3`$ surface $`X`$. Then with the notation in (2.1) and (1.4), $`(n_g,m_g;\chi _{\text{top}}(X^g),\chi _{\text{top}}(X^{g\tau }),\chi _{\text{top}}(X^{g^2\tau })`$, $`\chi _{\text{top}}(X^{g^2}))`$ is one of the following:
$$(1,6;8,2,6,0),(0,4;4,4,6,0),(1,2;0,6,6,0).$$
The result below is used in (2.4) to determine the representation of $`S_3\times \mu _4<G`$ there.
Lemma 2.3. (1) Suppose that $`g`$ stabilizes $`\chi _4`$. Then w.r.t. to one and the same basis $`\{v_1,\mathrm{},v_4\}`$, we have the following matrix representation of $`A_4:\mu _4`$ on $`\chi _4`$:
$$\sigma _1^{}=\text{diag}[1,1,1,1],\sigma _2^{}=[1,1,1,1],$$
$$\gamma ^{}=\left(\begin{array}{cccc}1& 0& 0& 0\\ 0& 0& 0& \beta _4\\ 0& \beta _2& 0& 0\\ 0& 0& \beta _3& 0\end{array}\right),g^{}=\left(\begin{array}{cccc}\alpha _1& 0& 0& 0\\ 0& \alpha _2& 0& 0\\ 0& 0& 0& \alpha _5\\ 0& 0& \alpha _4& 0\end{array}\right).$$
We have exactly the same kind of matrix representation of $`A_4:\mu _4`$ w.r.t. one and the same basis $`\{v_1^{},\mathrm{},v_4^{}\}`$ of $`\chi _4^{}`$. But we use $`\beta _i^{}`$ and $`\alpha _i^{}`$ for $`\gamma ^{}|\chi _4^{}`$ and $`g^{}|\chi _4^{}`$ instead.
(2) Suppose $`g`$ stabilizes $`\chi _5`$. Then w.r.t. to one and the same basis $`\{y_1,\mathrm{},y_5\}`$, we have the following matrix representation of $`A_4:\mu _4`$ on $`\chi _5`$, where $`\eta _3`$ is a primitive 3rd root of 1:
$$\sigma _1^{}=\text{diag}[1,1,1,1,1],\sigma _2^{}=[1,1,1,1,1],$$
$$\gamma ^{}=\left(\begin{array}{ccccc}\eta _3& 0& 0& 0& 0\\ 0& \eta _3^2& 0& 0& 0\\ 0& 0& 0& 0& b_5\\ 0& 0& b_3& 0& 0\\ 0& 0& 0& b_4& 0\end{array}\right),g^{}=\left(\begin{array}{ccccc}0& a_2& 0& 0& 0\\ a_1& 0& 0& 0& 0\\ 0& 0& a_3& 0& 0\\ 0& 0& 0& 0& a_5\\ 0& 0& 0& a_4& 0\end{array}\right).$$
We have exactly the same kind of matrix representation of $`A_4:\mu _4`$ w.r.t. one and the same basis $`\{y_1^{},\mathrm{},y_5^{}\}`$ of $`\chi _5^{}`$. But we use $`b_i^{}`$ and $`a_i^{}`$ for $`\gamma ^{}|\chi _5^{}`$ and $`g^{}|\chi _5^{}`$ instead.
Proof. This follows from the character Table 1 in (1.6) and the fact that the conjugation $`c_g`$ fixes $`\sigma _1`$, and exchanges $`\sigma _2`$ with $`\sigma _1\sigma _2`$ and $`\gamma `$ with $`\gamma ^1`$.
Lemma 2.4. (1) Suppose that $`g`$ stabilizes $`\chi _4`$. Then w.r.t. to one and the same basis $`\{u_1,\mathrm{},u_4\}`$, we have the following matrix representation of $`S_3\times \mu _4`$ on $`\chi _4`$, where $`\eta _3`$ is a primitive 3rd root of 1. Moreover, $`d_3=\pm d_1`$ and $`(g^2)^{}|\chi _4=d_1^2\text{id}`$:
$$\begin{array}{c}\tau ^{}=[1,1,\eta _3,\eta _3^2],g^{}=\text{diag}[d_1,d_3,d_3,d_3],\\ \sigma ^{}=\text{diag}[1,1,\left(\begin{array}{cc}0& 1\\ 1& 0\end{array}\right)].\end{array}$$
We have exactly the same kind of matrix representation of $`S_3\times \mu _4`$ w.r.t. one and the same basis $`\{u_1^{},\mathrm{},u_4^{}\}`$ of $`\chi _4^{}`$. But we use $`d_i^{}`$ for $`g^{}|\chi _4^{}`$ instead.
(2) Suppose that $`g`$ stabilizes $`\chi _5`$. Then w.r.t. to one and the same basis $`\{x_1,\mathrm{},x_5\}`$, we have the following matrix representation of $`S_3\times \mu _4`$ on $`\chi _5`$, where $`\eta _3`$ is a primitive 3rd root of 1. Moreover, $`e_2=\pm e_1`$, $`(g^2)^{}|\chi _5=e_1^2\text{id}`$ (and $`e_1`$ equals $`a_3`$ in (2.3)):
$$\begin{array}{c}\tau ^{}=\text{diag}[1,\eta _3,\eta _3^2,\eta _3,\eta _3^2],g^{}=[e_1,e_2,e_2,e_2,e_2],\\ \sigma ^{}=\text{diag}[1,\left(\begin{array}{cc}0& 1\\ 1& 0\end{array}\right),\left(\begin{array}{cc}0& 1\\ 1& 0\end{array}\right)].\end{array}$$
We have exactly the same kind of matrix representation of $`S_3\times \mu _4`$ w.r.t. one and the same basis $`\{x_1^{},\mathrm{},x_5^{}\}`$ of $`\chi _5^{}`$. But we use $`e_i^{}`$ for $`g^{}|\chi _5^{}`$, instead.
Proof. (1) follows from the character Table 1 in (1.6) and the fact that $`g`$ commutes with both $`\sigma ,\tau `$, if we claim only $`g^{}|\chi _4=\text{diag}[d_1,d_2,d_3,d_3]`$ instead. It suffices to show that $`d_2=d_3`$. On the one hand, over the eigenspace $`V_4(\sigma =1)\chi _4`$ of $`\sigma `$ corresponding to the eigenvalue $`1`$, we have $`g^{}|V_4(\sigma =1)=\text{diag}[d_2,d_3]`$. On the other hand, by (2.3), $`g^{}|V_4(\sigma =1)=\text{diag}[\sqrt{\alpha _4\alpha _5},\sqrt{\alpha _4\alpha _5}]`$. Thus $`d_2=d_3`$. Now $`d_1=\pm d_3`$ follows from the fact that $`(g^2)^{}|\chi _i`$ is a scalar.
(2) is similar, except the determination of $`e_i`$ in $`g^{}=\text{diag}[e_1,e_2,e_2,e_4,e_4]`$. Indeed, comparing the diagonalization in (2.3) and here we see also that $`\text{diag}[e_2,e_4]=g^{}|V_5(\sigma =1)=\text{diag}[\sqrt{a_4a_5},\sqrt{a_4a_5}]`$, whence $`e_4=e_2`$. Taking trace in (2.3) and here, we obtain $`a_3=\text{Tr}(g^{}|\chi _5)=e_1`$.
Lemma 2.5. (1) Suppose that $`g`$ switches $`\chi _4`$ with $`\chi _4^{}`$. Then w.r.t. to one and the same basis $`\{u_1,\mathrm{},u_8\}`$, we have the following matrix representation of $`S_3\times \mu _4`$ on $`\chi _4\chi _4^{}`$, where $`\eta _3`$ is a primitive 3rd root of 1. Moreover, $`(g^2)^{}|\chi _4=(d_1d_5)\text{id}=(g^2)^{}|\chi _4^{}`$:
$$\tau ^{}=[1,1,\eta _3,\eta _3^2,1,1,\eta _3,\eta _3^2],$$
$$\sigma ^{}=\text{diag}[\left(\begin{array}{cc}1& 0\\ 0& 1\end{array}\right),\left(\begin{array}{cc}0& 1\\ 1& 0\end{array}\right),\left(\begin{array}{cc}1& 0\\ 0& 1\end{array}\right),\left(\begin{array}{cc}0& 1\\ 1& 0\end{array}\right)],$$
$$g^{}=\left(\begin{array}{cccccccc}0& 0& 0& 0& d_5& 0& 0& 0\\ 0& 0& 0& 0& 0& d_6& 0& 0\\ 0& 0& 0& 0& 0& 0& d_7& 0\\ 0& 0& 0& 0& 0& 0& 0& d_8\\ d_1& 0& 0& 0& 0& 0& 0& 0\\ 0& d_2& 0& 0& 0& 0& 0& 0\\ 0& 0& d_3& 0& 0& 0& 0& 0\\ 0& 0& 0& d_4& 0& 0& 0& 0\end{array}\right).$$
(2) Suppose that $`g`$ switches $`\chi _5`$ with $`\chi _5^{}`$. Then w.r.t. to one and the same basis $`\{x_1,\mathrm{},x_{10}`$ $`\}`$, we have the following matrix representation of $`S_3\times \mu _4`$ on $`\chi _5\chi _5^{}`$, where $`\eta _3`$ is a primitive 3rd root of 1. Moreover, $`(g^2)^{}|\chi _5=(e_1e_6)\text{id}=(g^2)^{}|\chi _5^{}`$:
$$\tau ^{}=[1,\eta _3,\eta _3^2,\eta _3,\eta _3^2,1,\eta _3,\eta _3^2,\eta _3,\eta _3^2],$$
$$\sigma ^{}=\text{diag}[1,\left(\begin{array}{cc}0& 1\\ 1& 0\end{array}\right),\left(\begin{array}{cc}0& 1\\ 1& 0\end{array}\right),1,\left(\begin{array}{cc}0& 1\\ 1& 0\end{array}\right),\left(\begin{array}{cc}0& 1\\ 1& 0\end{array}\right)],$$
$$g^{}=\left(\begin{array}{cccccccccc}0& 0& 0& 0& 0& e_6& 0& 0& 0& 0\\ 0& 0& 0& 0& 0& 0& e_7& 0& 0& 0\\ 0& 0& 0& 0& 0& 0& 0& e_7& 0& 0\\ 0& 0& 0& 0& 0& 0& 0& 0& e_9& 0\\ 0& 0& 0& 0& 0& 0& 0& 0& 0& e_9\\ e_1& 0& 0& 0& 0& 0& 0& 0& 0& 0\\ 0& e_2& 0& 0& 0& 0& 0& 0& 0& 0\\ 0& 0& e_2& 0& 0& 0& 0& 0& 0& 0\\ 0& 0& 0& e_4& 0& 0& 0& 0& 0& 0\\ 0& 0& 0& 0& e_4& 0& 0& 0& 0& 0\end{array}\right).$$
Proof. The proof is similar to (2.4).
To prove (2.2), we consider first the case where both $`\chi _4`$ and $`\chi _5`$ are $`g`$-stable:
Lemma 2.6. Suppose that both $`\chi _4`$ and $`\chi _5`$ are $`g`$-stable.
(1) We have the following, where by $`d_1`$, etc. we mean $`d_1+d_1^{}`$ etc :
$$\begin{array}{c}\chi _{\text{top}}(X^{g^\pm })=2+\text{Tr}(g^{}|\chi _1\chi _1^{})+(d_1+d_3+e_1),\\ \chi _{\text{top}}(X^{g^1\tau ^{}})=\chi _{\text{top}}(X^{g\tau ^\pm })=2+\text{Tr}(g^{}|\chi _1\chi _1^{})+(d_12d_3+e_1),\\ \chi _{\text{top}}(X^{g^2})=2+(4d_1^2+5e_1^2),\\ \chi _{\text{top}}(X^{g^2\tau ^\pm })=2+(d_1^2e_1^2).\end{array}$$
(2) We have $`d_1^4=e_1^4=(d_1^{})^4=(e_1^{})^4=1`$ and $`d_3\{\pm d_1\}`$, $`d_3^{}\{\pm d_1^{}\}`$.
(3) Among six 4-th roots of 1: $`e_1`$, $`e_1^{}`$, $`d_i`$, $`d_i^{}`$ ($`i=1,3`$), either all six of them are primitive, or exactly $`e_1,e_1^{}`$ are primitive, or exactly the $`d_i`$, $`d_i^{}`$ ($`i=1,3`$) are primitive 4-th root of 1.
(4) (2.2) holds.
Proof. (1) and (2) follow from (2.4). For (3), the formula for $`\chi _{\text{top}}(X^{g^2})`$ in (1) and its upper bound $`18`$ in (1.2) imply that at least one of the six 4-th roots of 1 in (3) is primitive. Now (3) is a consequence of (2) and the description of $`\chi _{\text{top}}(X^g)`$ and $`\chi _{\text{top}}(X^{g\tau })`$ in (1) which and the difference (i.e., $`3d_3=3(d_3+d_3^{})`$) of which must be real numbers (indeed, integers).
To prove (4), we apply (3). If exactly these four: $`d_i,d_i^{}`$ ($`i=1,3`$) are primitive 4-th roots of 1, then $`\chi _{\text{top}}(X^{g^2\tau })=2+(2)2<0`$, contradicting (1.4). If all these six in (3) are primitive 4-th roots of 1, then $`\chi _{\text{top}}(X^g)`$ and $`\chi _{\text{top}}(X^{g\tau })`$, given in (1) and being real numbers, must all be equal to $`2+\text{Tr}(g^{}|\chi _1\chi _1^{})`$; hence they are all equal to 4 โ the only possible common value of these two, by (1.4); but then $`\chi _{\text{top}}(X^{g^2\tau })=2+(2)(2)=2<4=\chi _{\text{top}}(X^{g\tau })`$, a contradiction to (1.4).
Thus, exactly $`e_1,e_1^{}`$ are primitive 4-th root of 1, while $`d_i,d_i^{}\{\pm 1\}`$ ($`i=1,3`$). So (\*) : $`2\chi _{\text{top}}(X^g)8`$. Also $`\chi _{\text{top}}(X^{g^2})=2+4\times 2+5\times (2)=0`$ and $`\chi _{\text{top}}(X^{g^2\tau ^\pm })=2+2(2)=6`$. Now (1) implies that $`\chi _{\text{top}}(X^{g\tau ^\pm })+3d_3=\chi _{\text{top}}(X^g)=0`$ (mod 4) by (1.4), and also $`d_3=d_3+d_3^{}\{0,\pm 2\}`$ and $`\chi _{\text{top}}(X^{g\tau ^\pm })\{2,4,6\}`$ by (1.4). These and the (\*) above infer that the cases in (2.2) occur. The lemma is proved.
The first two assertions below are consequences of (2.4) \- (2.5) and an argument similar to (2.6).
Lemma 2.7. Suppose that $`g`$ switches $`\chi _4`$ with $`\chi _4^{}`$ but keeps $`\chi _5`$ (and $`\chi _5^{}`$) stable.
(1) We have the following, where $`\delta S_3=\sigma ,\tau `$ and by $`e_1`$ etc. we mean $`e_1+e_1^{}`$ etc :
$$\begin{array}{c}\chi _{\text{top}}(X^{g^1\delta ^1})=\chi _{\text{top}}(X^{g\delta })=2+\text{Tr}(g^{}|\chi _1\chi _1^{})+e_1,\\ \chi _{\text{top}}(X^{g^2})=2+8d_1d_5+5e_1^2,\\ \chi _{\text{top}}(X^{g^2\tau ^\pm })=2+2d_1d_5e_1^2.\end{array}$$
(2) We have $`e_1^4=(e_1^{})^4=(d_1d_5)^2=1`$. Either $`\{e_1,e_1^{}\}=\{\pm \sqrt{1}\}`$, or $`e_1,e_1^{}\{\pm 1\}`$.
(3) (2.2) holds.
Proof. To prove (3), by (1) $`\chi _{\text{top}}(X^g)`$ ($`=0`$ mod 4) and $`\chi _{\text{top}}(X^{g\tau })`$ ($`\{2,4,6\}`$) are equal (see (1.4)). Hence they are all equal to 4. If both $`e_1,e_1^{}`$ are in $`\{\pm 1\}`$, then $`\chi _{\text{top}}(X^{g^2\tau })=2+2d_1d_522<4=\chi (X^{g\tau })`$, contradicting (1.4). Thus, $`\{e_1,e_1^{}\}=\{\pm \sqrt{1}\}`$. By (1.4), we have $`4=\chi _{\text{top}}(X^{g\tau })\chi _{\text{top}}(X^{g^2\tau })=2+2d_1d_5+2`$, whence the latter equals $`6`$ and $`d_1d_5=1`$. Now $`\chi _{\text{top}}(X^{g^2})=2+8+5\times (2)=0`$. Therefore, the second case in (2.2) occurs. This proves the lemma.
Lemma 2.8. Suppose that $`\chi _4`$ (and $`\chi _4^{}`$ are) is $`g`$-stable but $`g`$ switches $`\chi _5`$ with $`\chi _5^{}`$.
(1) We have the following, where by $`d_1`$ etc. we mean $`d_1+d_1^{}`$ etc :
$$\begin{array}{c}\chi _{\text{top}}(X^{g^\pm })=2+\text{Tr}(g^{}|\chi _1\chi _1^{})+(d_1+d_3),\\ \chi _{\text{top}}(X^{g^1\tau ^{}})=\chi _{\text{top}}(X^{g\tau ^\pm })=2+\text{Tr}(g^{}|\chi _1\chi _1^{})+(d_12d_3),\\ \chi _{\text{top}}(X^{g^2})=2+4d_1^2+10e_1e_6,\\ \chi _{\text{top}}(X^{g^2\tau ^\pm })=2+d_1^22e_1e_6.\end{array}$$
(2) We have $`d_1^4=(d_1^{})^4=(e_1e_6)^2=1`$ and $`d_3\{\pm d_1\}`$, $`d_3^{}\{\pm d_1^{}\}`$.
(3) Either the four 4-th roots of 1: $`d_i,d_i^{}`$ ($`i=1,3`$) are all in $`\{\pm \sqrt{1}\}`$, or these four are all in $`\{\pm 1\}`$ (so $`e_1e_6=1`$ and $`\chi _{\text{top}}(X^{g^2})=0`$ by (1.2)).
(4) (2.2) holds.
Proof. (1) - (2) are consequences of (2.5) \- (2.6), while the proof of (3) - (4) are similar to the argument for the case of (2.6). Indeed, if the first (resp. second) situation in (3) occurs, then a contradiction (resp. (2.2) holds). This proves the lemma.
Lemma 2.9. Suppose that $`g`$ switches $`\chi _4`$ with $`\chi _4^{}`$ and $`\chi _5`$ with $`\chi _5^{}`$. Then (2.2) holds.
To be precise, we have the following, where $`\delta `$ is in $`S_3=\sigma ,\tau `$, where $`(d_1d_5)^2=(e_1e_6)^2=1`$:
$$\begin{array}{c}\chi _{\text{top}}(X^{g^1\delta ^1})=\chi _{\text{top}}(X^{g\delta })=2+\text{Tr}(g^{}|\chi _1\chi _1^{}),\\ \chi _{\text{top}}(X^{g^2})=2+8d_1d_5+10e_1e_6,\\ \chi _{\text{top}}(X^{g^2\tau ^\pm })=2+2d_1d_52e_1e_6.\end{array}$$
Proof. The formulae or equalities are consequences of (2.4) \- (2.5). As in (2.7), the formulae in (1) and (1.4) imply that $`\chi _{\text{top}}(X^g)=\chi _{\text{top}}(X^{g\tau })=4`$. The formula for $`\chi _{\text{top}}(X^{g^2\tau })`$ and its lower bounder $`4=\chi _{\text{top}}(X^{g\tau })`$ by (1.4), infer that it equals $`6`$ and $`d_1d_5=1`$, $`e_1e_6=1`$. This proves the lemma. The proof of (2.2) is completed.
ยง3. The proofs of Theorems A, B and C
In this section we shall prove Theorems A, B and C. We first prove the result below which is a consequence of (3.2)-(3.8) below.
Theorem 3.1.
(1) There is no faithful group action of the form $`A_5.\mu _4`$ (see (1.0)) on a $`K3`$ surface.
(2) If $`A_5.\mu _I`$ acts faithfully on a $`K3`$ surface, then $`I=1`$, or $`2`$.
(2) follows from (1), (1.1) and \[Z2, Theorem 3.1\]. Let us prove (3.1) (1). Suppose the contrary that $`G:=A_5.\mu _4`$ acts on a $`K3`$ surface $`X`$. Then $`G=A_5:\mu _4`$ and the unique group structure of such $`G`$ is given in (1.7). We use the notation in (2.1) and (2.2). First, we need:
Proposition 3.2. Suppose that $`G=A_5:\mu _4`$ acts on a $`K3`$ surface $`X`$. Then with the notation in (2.1), the fixed locus $`X^{g^2}=C_{i=1}^6D_i`$ is a disjoint union of a genus-$`7`$ curve $`C`$ (hence $`C^2=12`$) and six smooth rational curves. Both $`C`$ and $`_{i=1}^6D_i`$ are $`G`$-stable.
Proof. We apply (2.2). Then we always have $`\chi _{\text{top}}(X^{g^2})=0`$. Also (1.4) implies that $`X^{g^2}X^g\mathrm{}`$, so either $`X^{g^2}=_{i=1}^sE_i`$ with $`1s10`$ (by (1.2)) is a disjoint union of a few smooth elliptic curves $`E_i`$ (so $`X_{1\text{dim}}^g`$ is, if not empty, a disjoint union of some of the $`E_i`$โs, and hence $`n_g=0`$ in notation of (1.4)), or $`X^{g^2}=C_{i=1}^sD_i`$ is a disjoint union of a smooth curve $`C`$ and $`s`$ smooth rational curves $`D_i`$ with $`9s=g(C)11`$ (see (1.2)).
Consider the case where $`X^{g^2}=_{i=1}^sE_i`$. Then $`n_g=0`$ and $`(n_g,m_g;\chi _{\text{top}}(X^g),\chi _{\text{top}}(X^{g\tau })`$, $`\chi _{\text{top}}(X^{g^2\tau })`$, $`\chi _{\text{top}}(X^{g^2}))=(0,4;4,4,6,0)`$. Note that $`|X_{\text{isol}}^g|=m_g=4`$. We may assume that $`E_1`$ contains an isolated $`g`$-fixed point. Since the restriction $`g|E_1`$ is now of order 2, this $`E_1`$ contains all four isolated $`g`$-fixed points by (1.9). Now $`g`$ commutes with every element of $`\sigma ,\tau =S_3`$ as mentioned in (2.1), and hence there is a natural homomorphism $`S_3S_4`$ ($`=`$ the full symmetry group of the 4-point set $`X_{\text{isol}}^g`$). By (1.2) and (1.9), the restriction $`\tau |X_{\text{isol}}^g\text{id}`$. So the image of this homomorphism equals one of the four 1-point (say $`P_1`$) stabilizer subgroups ($`S_3`$) in $`S_4`$. This leads to that $`S_3<SL(T_{X,P_1})`$, contradicting (1.0C).
Next we consider the case where $`X^{g^2}=C_{i=1}^sD_i`$. We claim that $`s=1,5,6`$. Since $`g^2`$ is in the centre of $`G`$ by (2.1), our $`G`$ acts on $`X^{g^2}`$ and hence stabilizes $`C`$ and permutes $`D_i`$โs. Note that $`C`$ and the $`A_5`$-orbits of $`\{D_1,\mathrm{},D_s\}`$ will give linearly independent classes in $`S_X^{A_5}`$. Since the latter is of rank 2 by (1.6), this $`A_5`$ acts transitively on the set $`\{D_1,\mathrm{},D_s\}`$ and hence the image of the natural homomorphism $`A_5S_s`$ is a transitive subgroup of $`S_s`$. Now the claim follows from (1.8).
We assert that $`C`$ is not $`g`$-fixed. Indeed, let $`\delta =(13)(24)`$, then $`c_\delta (g)=g\sigma `$ with $`\sigma =(12)(34)`$ (because $`c_g=c_{(12)}`$ on $`A_5`$). Hence $`X^{g\sigma }=\delta (X^g)`$. So $`\delta (C)`$ is contained in $`X^{g\sigma }X^{g^2}`$ (noting that $`(g\sigma )^2=g^2`$), whence it equals the unique curve $`C`$ of genus $`2`$ in $`X^{g^2}`$. Thus $`C=\delta (C)`$ is pointwise $`g\sigma `$-fixed. However, $`C`$ is also pointwise $`g`$-fixed, whence is pointwise $`\sigma `$-fixed. This contradicts (1.2). So the assertion is proved.
We claim that $`s=1`$ is impossible. Consider the case $`s=1`$. Then $`G=A_5:g`$ acts on the set $`\{C,D_1\}`$ and hence stabilizes both $`C`$ and $`D_1`$. If $`D_1`$ is pointwise $`g`$-fixed, then as above, $`D_1`$ would be pointwise ($`g\sigma `$ and hence) $`\sigma `$-fixed, a contradiction. So the restriction $`g|D_1`$ is not identity. We consider the natural homomorphism $`f:S_5=A_5:\overline{g}=G/g^2\text{Aut}(D_1)`$ (see (2.1), where $`\overline{g}`$ is the coset in $`g/g^2`$ containing $`g`$. Clearly, the restriction $`f|A_5`$ is an injection by (1.2). Hence $`|\text{Ker}(f)|2`$ and $`\text{Ker}(f)`$ is normal in $`S_5`$. So $`\text{Ker}(f)=(1)`$ and $`S_5f(S_5)<\text{Aut}(^1)`$, contradicting (1.9).
We still have to rule out the case $`s=5`$. Since $`C`$ is not pointwise $`g`$-fixed as proved above, $`X_{1\text{dim}}^g`$ is (if not empty) a disjoint union of $`n_g/2`$ ($`0`$) of $`D_i`$โs. If $`\tau =(345)`$ stabilizes some $`D_j`$ then $`\tau `$ fixes exactly two points on $`D_j`$ by (1.2) and (1.9). Since $`|X^\tau |=6`$, this $`\tau `$ stabilizes at most three $`D_j^{}s`$. Thus we may assume that $`\tau `$ permutes $`D_1,D_2,D_3`$ while stabilizes $`D_4,D_5`$. Now the commutability of $`g`$ with $`\tau `$ implies that $`g`$ stabilizes each $`D_i`$ ($`i=1,2,3`$); also none of $`D_i`$ ($`i=1,2,3`$) is pointwise $`g`$-fixed, for otherwise all these three $`D_i`$ (forming one $`\tau `$-orbit) are pointwise $`g`$-fixed, whence $`n_g3`$, contradicting (2.2). Thus, $`m_g=|X_{\text{isol}}^g||_{i=1}^3|D_i^g|=6`$. So the first case in (2.2) occurs and $`n_g=1`$, $`m_g=6`$. Here $`n_g=1`$ implies that (after switching $`D_4`$ with $`D_5`$ if necessary) $`D_5`$ is poinwise $`g`$-fixed, and $`D_4`$ is $`g`$-stable but not $`g`$-fixed. This leads to $`6=|X_{\text{isol}}^g|_{i=1}^4|D_i^g|=8`$, a contradiction. So (3.2) is proved. Indeed, for the last part, note that $`g^2`$ is in the centre of $`G`$ by (2.1) and hence $`G`$ acts on $`X^{g^2}`$.
We continue the proof of (3.1) (1). In notation of (3.2), we set $`D=_{i=1}^6D_i`$ and $`L_0:=[C,D]`$. Then we have:
Lemma 3.3. Suppose that $`G=A_5:\mu _4`$ acts on a $`K3`$ surface $`X`$.
(1) $`L_0`$ is a sublattice (with intersection form $`\text{diag}[12,12]`$) of $`S_X^{A_5}`$ of finite index $`d_1`$. In particular, $`S_X^G=S_X^{A_5}`$, i.e., $`g^{}|S_X^{A_5}=\text{id}`$.
(2) If $`d_1>1`$, then $`d_1=2`$ and $`S_X^{A_5}`$ equals $`[u_1,u_2]`$ with $`u_1=\frac{1}{2}(C+D)`$ and $`u_2=\frac{1}{2}(CD)`$ and with the intersection form $`U(6)`$, i.e., $`u_i^2=0`$ and $`u_1.u_2=6`$.
Proof. (1) Clearly, $`S_X^{A_5}S_X^GL_0`$ by (3.2). Now (1) follows from the fact that $`\text{rank}S_X^{A_5}=2`$ by (1.6).
(2) Suppose that $`d_1>1`$. Let $`\theta =\frac{1}{12}(aC+bD)`$ be in $`S_X^{A_5}L_0^{}=\text{Hom}(L_0,)=[C/12,D/12]`$ but not in $`L_0`$. Since $`2b/12=\theta .D_1`$, we have $`6|b`$. This and $`(a^2b^2)/12=\theta ^2`$ imply that $`12`$ divides $`a^2`$, whence $`6|a`$. So modulo $`L_0`$, our $`\theta =C/2`$, or $`D/2`$ or $`(C+D)/2`$. Since $`\theta ^22`$, we have $`\theta =(C+D)/2`$ and hence $`S_X^{A_5}=[C,(C+D)/2]=[(C+D)/2,(CD)/2]`$. The lemma is proved.
Set $`L=H^0(X,)`$ which contains $`S_XT_X`$ as a sublattice of finite index. Also $`L^{A_5}`$ contains $`S_X^{A_5}T_X`$ as a sublattice of finite index $`d`$ by (1.0A-B).
Lemma 3.4. The quotient $`L^{A_5}/(S_X^{A_5}T_X)`$ is 2-elementary of order $`d`$ and isomorphic to $`(0)`$ ($`d=1`$), $`/(2)`$ ($`d=2`$) or $`(/(2))^2`$ ($`d=4`$).
Proof. For a lattice $`M`$, we denote by $`M^{}=\text{Hom}(M,)`$ the dual and $`A_M=M^{}/M`$ the discriminant group. Then we have, where $`\iota `$ is the inclusion:
$$\begin{array}{c}S_X^{A_5}T_XL^{A_5}(L^{A_5})^{}(S_X^{A_5})^{}T_X^{},\\ \iota :L^{A_5}/(S_X^{A_5}T_X)A_{S_X^{A_5}}A_{T_X}.\end{array}$$
Let $`pr_1`$ and $`pr_2`$ be the projections from $`A_{S_X^{A_5}}A_{T_X}`$ to its first and second summands, respectively. Since $`S_X^{A_5}`$ and $`T_X`$ are primitive in $`L^{A_5}`$, both compositions $`pr_i\iota `$ are injective. In particular, the quotient group in (3.4) is regarded as a subgroup of a bigger group $`A_{T_X}`$, whence is generated by 2 elements because the same is true for the bigger group (since $`\text{rank}T_X=2`$ by (1.1)). We still have to show that this quotient group is 2-elementary.
Take a coset $`\overline{\theta }`$ from the quotient group in (3.4). In notation of (1.5), we write
$$\theta =u+\frac{1}{2m}(at_1+bt_2)(S_X^{A_5})^{}T_X^{}.$$
Regarding $`\overline{\theta }`$ as an element of $`A_{S_X^{A_5}}`$ via the injection $`pr_1\iota `$, we have by (3.3), modulo $`S_X^{A_5}T_X`$, that
$$0=g^{}\theta \theta =\frac{1}{2m}[a(g^{}t_1t_1)+b(g^{}t_2t_2)]=\frac{1}{2m}[(a+b)t_1+(ab)t_2].$$
So $`2m`$ divides $`a+b`$, $`ab`$ (and hence $`2a`$ and $`2b`$) because $`T_X`$ is primitive in $`L`$. Thus $`m`$ divides $`a`$ and $`b`$ and we write $`a=ma^{}`$ and $`b=mb^{}`$ so that $`\theta =u+\frac{1}{2}(a^{}t_1+b^{}t_1)`$. Therefore, modulo $`T_X`$, we have $`2u=2\theta 2L^{G_N}L^{G_N}`$, whence $`2uL(S_X^{A_5})^{}=S_X^{A_5}`$ (because the latter is primitive in $`L`$). So $`2\overline{\theta }=0`$. The lemma is proved.
Lemma 3.5. One of the following cases occurs.
(1) We have $`m=5`$. Both the quotients $`S_X^{A_5}/L_0`$ and $`L^{A_5}/(S_X^{A_5}T_X)`$ are isomorphic to $`/(2)`$. Moreover, the discriminant form of $`(L^{A_5})^{}/L^{A_5}(/(30))^2`$ is given in \[Z2, Theorem 2.1\] (corresponding to the matrix $`M_1`$ there) and generated by the cosets $`\overline{\epsilon }_i`$ with $`\epsilon _1=e_1^{},\epsilon _2=e_2^{}+e_3^{}+e_4^{}`$ and the intersection form (note that $`\overline{\epsilon }_i^2`$ is in $`/2`$ while $`\overline{\epsilon }_1.\overline{\epsilon }_2`$ is in $`/`$):
$$(\overline{\epsilon }_i.\overline{\epsilon }_j)=\left(\begin{array}{cc}23/30& 1/5\\ 1/5& 35/30\end{array}\right).$$
(2) We have $`m=10`$, $`S_X^{A_5}/L_0/(2)`$ and $`L^{A_5}/(S_X^{A_5}T_X)(/(2))^2`$.
(3) We have $`m=5`$, $`L_0=S_X^{A_5}`$ and $`L^{A_5}/(S_X^{A_5}T_X)(/(2))^2`$.
Proof. In notation of (3.3) and (3.4), we have $`(12^2)(4m^2)=|L_0||T_X|=d_1^2d^2|L^{A_5}|`$. On the other hand, $`|L^{A_5}|=30^2,3\times 10^2,20^2,3\times 20^2,3\times 40^2`$ by the calculation in \[Z2, Theorem 2.1\]. Then the lemma follows easily.
Lemma 3.6. The case (3) in (3.5) does not occur.
Proof. Consider the case (3) in (3.5). Let $`\theta `$ be an element in $`L^{A_5}`$ but not in the smaller set $`S_X^{A_5}T_X`$. We claim that $`\theta ^22`$ implies that modulo this smaller set, our $`\theta `$ equals some $`\theta _j`$ below, where $`u_1:=C`$, $`u_2:=D`$ and $`T_X=[t_1,t_2]`$ as in (1.5). Here $`\theta _j:=\frac{1}{2}(t_1+t_2)+\frac{1}{2}u_j`$.
Indeed, since the quotient group in (3.5) (3) is 2-elementary, we can write, modulo the smaller set, that $`\theta =\frac{1}{2}(a_1t_1+a_2t_2+b_1u_1+b_2u_2)`$ with $`a_i`$, $`b_j`$ in $`\{0,1\}`$ but not all zero. Indeed, $`(a_1,a_2)(0,0)(b_1,b_2)`$ because both $`S_X^{A_5}`$ and $`T_X`$ are primitive in $`L`$. Now modulo $`2`$, we have the following, so the claim follows:
$$\frac{1}{2}(a_1^2+a_2^2)+b_1^2+b_2^2=\frac{2m}{4}(a_1^2+a_2^2)+\frac{12}{4}(b_1^2b_2^2)=\theta ^2=0.$$
Since $`\theta _1\theta _2`$ is not in $`L^{A_5}`$ (not in $`L`$ at all, by the primitivity of $`S_X^{A_5}`$ in $`L`$), at most one of $`\theta _j`$ is in $`L^{A_5}`$. So $`L^{A_5}/(S_X^{A_5}T_X)`$ is of order $`2`$, a contradiction.
We start anew. By (3.3) and (3.6), the lattice $`S_X^{A_5}`$ equals $`[u_1,u_2]`$ with $`u_1=\frac{1}{2}(C+D)`$ and $`u_2=\frac{1}{2}(CD)`$, and has the intersection form $`U(6)`$.
Lemma 3.7. The case (2) in (3.5) is impossible.
Proof. Take $`\theta `$ in $`L^{A_5}`$ but not in the smaller set $`S_X^{A_5}T_X`$. As in (3.6), $`\theta ^22`$ implies that modulo the smaller set, our $`\theta `$ is one of the following
$$\theta ^i=\frac{1}{2}t_i+\frac{1}{2}(u_1+u_2),\theta _j=\frac{1}{2}(t_1+t_2)+\frac{1}{2}u_j.$$
Since $`\theta ^1\theta ^2`$ is not in $`L^{A_5}`$ (not in $`L`$ at all), not both $`\theta ^i`$ are in $`L^{A_5}`$. By the same reasoning not both $`\theta _j`$ are in $`L^{A_5}`$. Since $`L^{A_5}/(S_X^{A_5}T_X)(/(2))^2`$ is generated by two elements, one of $`\theta ^i`$ ($`i=1,2`$) and one of $`\theta _j`$ ($`j=1,2`$) are in $`L^{A_5}`$. But $`\theta ^i.\theta _j=\frac{2m}{4}+\frac{6}{4}=\frac{13}{2}`$, which is not an integer. This is a contradiction.
Lemma 3.8. Suppose the case (1) in (3.5) occurs. Then we have:
(1) $`L^{A_5}`$ is generated by $`S_X,T_X`$ and $`\theta =\frac{1}{2}(t_1+t_2+u_1+u_2)`$.
(2) The discriminant group $`A_{L^{A_5}}=(L^{A_5})^{}/L^{A_5}`$ (with the dual $`(L^{A_5})^{}=\text{Hom}(L^{A_5},)`$) is generated by the cosets $`\overline{\delta }_j`$ ($`j=1,2`$) which (together with the intersection form) is given as follows: (where $`t_i^{}.t_j=\delta _{ij}`$, and $`u_i^{}.u_j=\delta _{ij}`$ in Kroneckerโs symbol):
$$\delta _1=t_2^{}+u_1^{}+2u_2^{}=\frac{1}{10}t_2+\frac{1}{6}(2u_1+u_2),\delta _2=t_1^{}+u_1^{}=\frac{1}{10}t_1+\frac{1}{6}u_2,$$
$$(\overline{\delta }_i.\overline{\delta }_j)=\left(\begin{array}{cc}23/30& 1/3\\ 1/3& 1/10\end{array}\right).$$
Proof. (1) can be proved as in (3.6), by making use of that $`\theta _1^22`$ for every $`\theta _1`$ in $`L^{A_5}`$.
(2) Since $`\delta _i.\theta `$, $`\delta _i.t_j`$ and $`\delta _i.u_j`$ are all in $``$ by a direct calculation, we see that both $`\delta _i`$ are in $`(L^{A_5})^{}`$. One checks easily that the subgroup $`\overline{\delta }_1,\overline{\delta }_2`$ of the discriminant group in (2) is isomorphic to $`(/(30))^2`$, whence this subgroup is indeed the whole discriminant group in (2) (because the latter is of order $`30^2`$ by (3.5)). This proves the lemma.
Here comes the punch line. By (3.5) \- (3.8), there is an isometry $`\phi :\overline{\epsilon }_1,\overline{\epsilon }_2\overline{\delta }_1,\overline{\delta }_2`$, so for some integers $`a,b,c,d`$, we can write $`(\phi (\overline{\epsilon }_1),\phi (\overline{\epsilon }_2))=(\overline{\delta }_1,\overline{\delta }_2)\left(\begin{array}{cc}a& c\\ b& d\end{array}\right)`$. Thus,
$$\begin{array}{c}23/30=\epsilon _1^2=\phi (\epsilon _1)^2=(a\delta _1+b\delta _2)^2=\frac{1}{30}(23a^2+3b^2+20ab)(\text{mod}\mathrm{\hspace{0.17em}\hspace{0.17em}2}),\\ 23=23a^2+3b^2+20ab(\text{mod}\mathrm{\hspace{0.17em}\hspace{0.17em}60}).\end{array}$$
The congruence above implies that modulo 4, we have $`1=a^2b^2`$, which is impossible. This completes the proof of (3.1) (1) and also the whole of (3.1).
We now prove Theorems A, B and C in the Introduction. In Theorem C, we have $`HG_N`$ by (1.0A); so $`H`$ is either one of $`A_5`$, $`L_2(7)`$, $`A_6`$ and $`M_{20}=C_2^4:A_5`$, by \[Xi, the list\]; if $`H=L_2(7)`$ then $`G_N=H`$ by \[Mu1\] and Theorem C follows from \[OZ3, Main Theorem\].
Therefore, we may assume that in all three theorems, $`G`$ is a finite group containing $`A_5`$ and acting faithfully on a $`K3`$ surface $`X`$. Write $`G=G_N.\mu _I`$ as in (1.0). By (1.0A) the $`A_5`$ in $`G`$ is contained in $`G_N`$. So $`G_N`$ is either one of $`A_5`$, $`S_5`$, $`A_6`$ and $`M_{20}=C_2^4:A_5`$, by \[Xi, the list\].
Consider the case $`G_N=A_5`$. Then $`I=1,2`$, by (1.1), \[Z2, Theorem 3.1\] and (3.1). If $`I=1`$, then $`G=A_5`$. If $`I=2`$, let $`\rho :GS_5\times \mu _2`$ ($`x(c_x,\alpha (x))`$) be the injection as in (1.7) so that $`\rho (A_5)=A_5\times 1`$; if the projection $`pr_1:S_5\times \mu _2S_5`$ maps $`\rho (G)`$ to $`A_5`$ (resp. to $`S_5`$), then $`G\rho (G)=A_5\times \mu _2`$ (resp. $`G\rho (G)pr_1(\rho (G))=S_5`$, by comparing the orders); see the argument below. Thus Theorems A, B and C are true.
Consider the case $`G_N=S_5`$. Let $`g`$ be in $`G`$ such that $`\alpha (g)`$ is a generator of $`\mu _I`$. Since $`\text{Aut}(S_5)=S_5`$ and $`xc_x`$ gives rise to an isomorphism $`S_5\text{Aut}(S_5)`$, we see that the map $`G\text{Aut}(S_5)=S_5`$ ($`xc_x`$) is surjective, and the conjugation maps $`c_g=c_s`$ on $`S_5`$, for some $`sS_5`$. Replacing $`g`$ by $`gs^1`$, we may assume that $`g`$ commutes with every element in $`G_N=S_5`$. So $`g^I\text{Ker}(\alpha )=G_N`$ is in the centre of $`G_N=S_5`$ (which is $`(1)`$), whence $`\text{ord}(g)=I`$, while $`\alpha (g)`$ is a generator of $`\mu _I`$. Thus $`G=S_5\times \mu _IA_5\times \mu _I`$. So $`I=1,2`$ by (1.1), \[Z2, Theorem 3.1\] and (3.1). Hence Theorems A, B and C are true.
Consider the case where $`G_N=A_6`$ or $`G_N=M_{20}=C_2^4:A_5`$. Then $`G_N`$ does not contain an $`A_5`$ as a normal subgroup (otherwise, in the latter case, $`M_{20}=C_2^4\times A_5`$, absurd). So $`A_5`$ is also not normal in $`G`$. Thus Theorems A and B are void this time. Now Theorem C follows from \[Ko2\] and \[KOZ1\].
References
\[AS2\] M. F. Atiyah and G. B. Segal, The index of elliptic operators. II. Ann. of Math. 87 (1968), 531โ545.
\[AS3\] M. F. Atiyah and I. M. Singer, The index of elliptic operators. III. Ann. of Math. 87 (1968) 546โ604.
\[Atlas\] J. H. Conway, R. T. Curtis, S. P. Norton, R. A. Parker and R. A. Wilson, Atlas of finite groups. Oxford University Press. Reprinted 2003 (with corrections).
\[CS\] J. H. Conway and N. J. A. Sloane, Sphere packings, lattices and groups. 3rd ed. Grundlehren der Mathematischen Wissenschaften, 290. Springer-Verlag, New York, 1999.
\[EDM\] Encyclopedic dictionary of mathematics. Vol. IโIV. Translated from the Japanese. 2nd ed. Edited by Kiyosi It . MIT Press, Cambridge, MA, 1987.
\[IOZ\] A. Ivanov, K. Oguiso and D. -Q. Zhang, The monster and $`K3`$ surfaces, in preparation.
\[KOZ1\] J. Keum, K. Oguiso and D. -Q. Zhang, The alternating group of degree 6 in geometry of the Leech lattice and $`K3`$ surfaces, Proc. London Math. Soc. 90 (2005), 371 - 394.
\[KOZ2\] J. Keum, K. Oguiso and D. -Q. Zhang, Extensions of the alternating group of degree 6 in geometry of $`K3`$ surfaces, math.AG/0408105, European J. Combinatorics: Special issue on Groups and Geometries, to appear.
\[Ko1\] S. Kondo, Niemeier lattices, Mathieu groups, and finite groups of symplectic automorphisms of $`K3`$ surfaces. Duke Math. J. 92 (1998), 593โ598.
\[Ko2\] S. Kondo, The maximum order of finite groups of automorphisms of $`K3`$ surfaces. Amer. J. Math. 121 (1999), 1245โ1252.
\[MO\] N. Machida and K. Oguiso, On $`K3`$ surfaces admitting finite non-symplectic group actions. J. Math. Sci. Univ. Tokyo 5 (1998), 273โ297.
\[Mu1\] S. Mukai, Finite groups of automorphisms of $`K3`$ surfaces and the Mathieu group. Invent. Math. 94 (1988), 183โ221.
\[Mu2\] Lattice-theoretic construction of symplectic actions on $`K3`$ surfaces, Duke Math. J. 92 (1998), 599โ603. As the Appendix to \[Ko1\].
\[Ni1\] V. V. Nikulin, Finite automorphism groups of Kahler $`K3`$ surfaces, Trans. Moscow Math. Soc. 38 (1980), 71โ135.
\[Ni2\] V. V. Nikulin, Factor groups of groups of automorphisms of hyperbolic forms with respect to subgroups generated by 2-reflections. Algebrogeometric applications. J. Soviet Math. 22 (1983), 1401โ1475.
\[Ni3\] V. V. Nikulin, Integer symmetric bilinear forms and some of their applications. Math. USSR Izvestija. 14 (1980), 103 โ 167.
\[Og\] K. Oguiso, A characterization of the Fermat quartic $`K3`$ surface by means of finite symmetries, math.AG/0308062. Compositio Math. to appear.
\[OZ1\] K. Oguiso and D. -Q. Zhang, On the most algebraic $`K3`$ surfaces and the most extremal log Enriques surfaces. Amer. J. Math. 118 (1996), 1277โ1297.
\[OZ2\] K. Oguiso and D. -Q. Zhang, On Vorontsovโs theorem on $`K3`$ surfaces with non-symplectic group actions. Proc. Amer. Math. Soc. 128 (2000), 1571โ1580.
\[OZ3\] K. Oguiso and D. -Q. Zhang, The simple group of order 168 and $`K3`$ surfaces. Complex geometry (Gottingen, 2000), Collection of papers dedicated to Hans Grauert, 165โ184, Springer, Berlin, 2002.
\[Sh1\] I. Shimada, Rational double points on supersingular $`K3`$ surfaces, Mathematics of Computation, 73 (2004), 1989โ2017.
\[Sh2\] I. Shimada, Lattices of algebraic cycles on Fermat varieties in positive characteristics, Proc. London Math. Soc. 82 (2001), 131โ172.
\[Sh3\] I. Shimada, On elliptic $`K3`$ surfaces, Michigan Math. J. 47 (2000), 423โ446.
\[Su\] M. Suzuki, Group theory. I. Translated from the Japanese by the author. Grundlehren der Mathematischen Wissenschaften 247. Springer-Verlag, Berlin-New York, 1982.
\[Vi\] E. B. Vinberg, The two most algebraic $`K3`$ surfaces. Math. Ann. 265 (1983), 1โ21.
\[Xi\] G. Xiao, Galois covers between $`K3`$ surfaces. Ann. Inst. Fourier (Grenoble) 46 (1996), 73โ88.
\[Z1\] D. -Q. Zhang, Quotients of $`K3`$ surfaces modulo involutions. Japan. J. Math. (N.S.) 24 (1998), 335โ366.
\[Z2\] D. -Q. Zhang, Niemeier lattices and K3 groups, math.AG/0408106, Proc. Intern. Conf. Alg. Geom. in honour of Prof. Dolgachev, Contemporary Math. Amer. Math. Soc. J. Keum (ed) to appear.
D. -Q. Zhang
Department of Mathematics
National University of Singapore
Singapore
E-mail : MATZDQMATH.NUS.EDU.SG
|
warning/0506/math0506545.html
|
ar5iv
|
text
|
# On a modification of a problem of Bialostocki, Erdลs, and Lefmann
## 1 Introduction
Ramsey type problems regarding colorings of the natural numbers are concerned with finding the minimum number $`N(r)`$, if it exists, for which every coloring of the integers in $`[1,N]`$ by $`r`$ colors contains some given monochromatic configuration. Traditionally, these configurations are solutions to systems of linear equations. The general theory developed by Rado in gave rise to the determination of $`N(r)`$ for certain monochromatic configurations, such as Schur numbers and Van der Waerden numbers . Other exact results of a similar kind were determined in , , , , and . The difficulty in computing such numbers led to the consideration of inequalities instead of equations. In particular, arithmetic progressions prompted Brown, Erdลs, and Freedman to define the notion of ascending waves. These and similar structures have been investigated in , , and .
Along similar lines, Bialostocki, Erdลs, and Lefmann considered in the following problem concerning monochromatic sets of nondecreasing diameter. For two positive integers $`m`$ and $`r`$, determine the minimum integer, $`f(m,r)`$, such that for every map $`\mathrm{\Delta }:\{1,\mathrm{},f(m,r)\{1,\mathrm{},r\}`$ there exist $`2m`$ integers
$$x_1<\mathrm{}<x_m<y_1<\mathrm{}<y_m$$
which satisfy the following conditions:
1. $`\mathrm{\Delta }(x_1)=\mathrm{}=\mathrm{\Delta }(x_m)`$,
2. $`\mathrm{\Delta }(y_1)=\mathrm{}=\mathrm{\Delta }(y_m)`$,
3. $`x_mx_1y_my_1`$.
They showed $`f(m,2)=5m3`$ and $`f(m,3)=9m7`$. Recently, Grynkiewicz proved $`f(m,4)=12m9`$ in . Bollobรกs, Erdลs, and Jin investigated in a closely related function, $`f^{}(2,r)`$, where strict inequality is required in 3 above. They determined $`f^{}(2,r)`$ for $`r=2^k`$.
In this paper we replace condition 3 by
1. $`2(x_mx_1)y_mx_1`$
and denote the corresponding function by $`g(m,r)`$. Notice that this is a relaxation of 3, since adding $`x_mx_1`$ to each side of the inequality $`x_mx_1y_my_1`$ yields $`2(x_mx_1)`$ on the left and $`y_my_1+x_mx_1<y_mx_1`$ on the right. It is not hard to see $`g(1,r)=2`$ for any $`r`$, and as such we assume throughout the sequel that $`m2`$.
The paper is organized as follows. In Section 2 we introduce some basic terms and develop a useful lemma that simplifies the construction of lower bounds. In Section 3 we determine $`g(m,r)`$ for $`r\{2,3,4\}`$. The main theorems appear in Section 4, where we develop tools that allow for either a bound or determination of $`g(m,r)`$ based upon the value of $`g(m,j)`$ for $`j<r`$. We conclude with some conjectures that arose from studying $`g(m,r)`$ using a computer program based on the theorems in Section 4.
## 2 Preliminaries
If $`S`$ is a nonempty set of integers and $`\mathrm{\Delta }:SR`$ is a mapping where $`|R|=r`$, then $`\mathrm{\Delta }`$ is called an $`r`$-coloring of $`S`$. For $`TS`$, we write $`\mathrm{\Delta }(T)=\{\mathrm{\Delta }(t):tT\}`$. We say $`T`$ is monochromatic if $`|\mathrm{\Delta }(T)|=1`$.
Throughout this paper an $`m`$-set, denoted $`Z=(z_1,\mathrm{},z_m)`$, is a sequence of $`m`$ distinct positive integers such that $`z_1<\mathrm{}<z_m`$. For a pair of $`m`$-sets $`X`$ and $`Y`$, we write $`XY`$ if $`x_m<y_1`$. Suppose $`XY`$; we define $`Y`$ to be $`X`$-admissible if $`2(x_mx_1)>y_mx_1`$. Furthermore, let $`\mathrm{\Delta }`$ be an $`r`$-coloring of a nonempty set $`S`$; we say $`\mathrm{\Delta }`$ is an $`L(r)`$-coloring of $`S`$ if for every pair of monochromatic $`m`$-sets $`X,YS`$, either $`XY`$ or $`Y`$ is $`X`$-admissible. That is, a coloring $`\mathrm{\Delta }`$ is an $`L(r)`$-coloring provided there are no two monochromatic $`m`$-sets $`X,YS`$ such that $`XY`$ and conditions 1,2, and 3 above are satisfied.
For an $`n`$-set $`X=(x_1,\mathrm{},x_n)`$ we use the following notation:
(i) $`int_i(X)=x_i`$ for $`in`$;
(ii) $`first_k(X)=\{x_1,\mathrm{},x_{\mathrm{min}\{k,n\}}\}`$; and
(iii) $`last_k(X)=\{x_{\mathrm{max}\{1,nk+1\}},\mathrm{},x_n\}`$.
For two integers $`a`$ and $`b`$ we use $`[a,b]`$ to denote the set of all integers $`i`$ such that $`ai`$ and $`ib`$, and refer to it as an interval. Note that if $`a>b`$ then $`[a,b]=\mathrm{}`$. Furthermore, for positive integers $`r`$,$`m`$, and $`s`$, where $`s2r(m1)+1`$, define the disjoint intervals $`I_1`$, $`I_2`$, and $`I_3`$ to be
1. $`I_1=[1,r(m1)+1]`$,
2. $`I_2=[r(m1)+2,2r(m1)]`$, and
3. $`I_3=[2r(m1)+1,s]`$.
Here we have used $`m2`$ to assume $`I_1I_3=\mathrm{}`$.
Since $`|I_1|=r(m1)+1`$ one sees that for an arbitrary $`r`$-coloring $`\mathrm{\Delta }`$ there must be some monochromatic $`m`$-set $`XI_1`$. The following proposition is immediate.
###### Proposition 2.1.
Let $`s2r(m1)+1`$ be a positive integer, and let $`\mathrm{\Delta }:[1,s][1,r]`$ be a coloring. If there exists a monochromatic $`m`$-set $`YI_2I_3`$ with $`y_mI_3`$ then $`\mathrm{\Delta }`$ is not an $`L(r)`$-coloring.
The following lemma simplifies the construction of $`L(r)`$-colorings by inducing an $`L(r)`$-coloring of $`I_1I_2`$ from an $`L(r)`$-coloring of $`I_2`$.
###### Lemma 2.2.
Let $`\mathrm{\Delta }:I_2[1,r]`$ be an $`L(r)`$-coloring. Then there exists an $`L(r)`$-coloring $`\mathrm{\Delta }_e`$ of $`I_1I_2`$ which is an extension of $`\mathrm{\Delta }`$. Further, $`\mathrm{\Delta }_e`$ satisfies $`\mathrm{\Delta }_e(1)=\mathrm{\Delta }_e(r(m1)+1)`$ and $`|\mathrm{\Delta }_e^1(t)[1,r(m1)]|=m1`$ for all $`t`$.
###### Proof.
Since $`|I_2|=r(m1)1`$, it follows that there is a color $`c`$ such that $`|\mathrm{\Delta }^1(c)I_2|<m1`$.
We define $`\mathrm{\Delta }_e:I_1I_2[1,r]`$ in two steps. First, we induce a coloring on $`I_2`$ and part of $`I_1`$ as described below:
$$\mathrm{\Delta }_e(x)=\{\begin{array}{cc}c,\hfill & \text{if }x=1\text{ or }x=r(m1)+1\hfill \\ \mathrm{\Delta }\left(x+r(m1)\right),\hfill & \text{if }x+r(m1)_{t=1}^rfirst_{m1}(\mathrm{\Delta }^1(t)I_2)\hfill \\ \mathrm{\Delta }(x),\hfill & \text{if }xI_2\hfill \end{array}$$
Second, we color the remaining integers of $`I_1`$ recursively as follows: suppose $`xI_1`$ and that $`\mathrm{\Delta }_e|_{[1,x1]}`$ is defined while $`\mathrm{\Delta }_e(x)`$ is not; then $`\mathrm{\Delta }_e(x)=i`$, where $`i=\mathrm{min}[1,r]`$ such that $`|\mathrm{\Delta }_e^1(i)[1,r(m1)]|<m1`$. From the definition of $`\mathrm{\Delta }_e`$ it is easy to verify that $`|\mathrm{\Delta }_e^1(t)[1,r(m1)]|=m1`$ for every $`t[1,r]`$.
It is left to show that $`\mathrm{\Delta }_e`$ is an $`L(r)`$-coloring of $`I_1I_2`$. Let $`X,YI_1I_2`$ be monochromatic $`m`$-sets with $`XY`$. If $`x_1I_2`$ then $`Y`$ is $`X`$-admissible since $`\mathrm{\Delta }_e|_{I_2}=\mathrm{\Delta }`$, which is an $`L(r)`$-coloring by assumption. Hence we may assume $`x_1I_1`$.
Case 1. Suppose $`\mathrm{\Delta }_e(x_1)=tc`$. Then since $`|\mathrm{\Delta }_e^1(t)I_1|=m1`$ for each $`tc`$, it follows that $`x_1=int_i(\mathrm{\Delta }_e^1(t)I_1)`$ for some $`i[1,m1]`$. Hence, since $`\mathrm{\Delta }_e(r(m1)+1)=ct`$, it follows that $`|\mathrm{\Delta }_e^1(t)[x_1,r(m1)+1]|=mi`$, and thus $`x_mint_i(\mathrm{\Delta }_e^1(t)I_2)`$ and $`|\mathrm{\Delta }^1(t)I_2|i`$. Remembering $`im1`$, it therefore follows from the definition of $`\mathrm{\Delta }_e`$ that
$$x_1=int_i(\mathrm{\Delta }_e^1(t)I_1)=int_i(\mathrm{\Delta }^1(t)I_2)r(m1)x_mr(m1),$$
so that $`x_mx_1r(m1)`$. Hence, since $`YI_1I_2`$,
$$2(x_mx_1)2r(m1)y_m>y_mx_1,$$
and $`Y`$ is $`X`$-admissible.
Case 2. Suppose $`\mathrm{\Delta }_e(x_1)=c`$. The argument above holds except in the case that $`x_1=1`$. In this case, we have $`x_mr(m1)+1=r(m1)+x_1`$, and $`x_mx_1r(m1)`$ as before. โ
In conjunction with Proposition 2.1, Lemma 2.2 shows there exists an $`L(r)`$-coloring on $`I_1I_2I_3`$ provided the existence of a coloring $`\mathrm{\Delta }:I_2I_3[1,r]`$ which is an $`L(r)`$-coloring on $`I_2`$ such that $`|\mathrm{\Delta }^1(c)(I_2I_3)|m1`$ for every $`c\mathrm{\Delta }(I_3)`$. Henceforth, we shall let the existence of $`\mathrm{\Delta }:I_2I_3[1,r]`$ which satisfies these conditions suffice to show the existence of an $`L(r)`$-coloring $`\mathrm{\Delta }_e:I_1I_2I_3[1,r]`$ without explicit construction.
## 3 The function $`g(m,r)`$ for $`r\{2,3,4\}`$
We first evaluate the function $`g(m,r)`$ for small values of $`r`$ and appropriate values of $`m`$. The case when $`r=2`$ is trivial.
###### Theorem 3.1.
Let $`m2`$ be an integer. Then, $`g(m,2)=5m4`$.
###### Proof.
The coloring $`\mathrm{\Delta }:[2m,5m5][1,2]`$ given by
$$1^{2m3}2^{m1}$$
shows that $`g(m,2)5m4`$.
Next we show that $`g(m,2)5m4`$. Let $`\mathrm{\Delta }:[1,5m4][1,2]`$ be an arbitrary $`2`$-coloring, and let $`P=[3m2,5m4]`$. Since $`|P|=2m1`$ there exists some monochromatic $`m`$-set $`YP`$. Furthermore, since $`|PI_2|=m1`$, it follows that $`YI_3\mathrm{}`$. Applying Proposition 2.1 completes the proof. โ
In evaluating $`g(m,3)`$ it will be beneficial to have the following
###### Lemma 3.2.
Let $`m4`$ be an integer, and let $`\mathrm{\Delta }:[1,3m4][1,3]`$ be a $`3`$-coloring. If $`|\mathrm{\Delta }^1(c)|3m\frac{m}{2}2`$ for some $`c[1,3]`$, then $`\mathrm{\Delta }`$ is not an $`L(3)`$-coloring.
###### Proof.
Let $`I=[1,3m4]`$ and $`t=|[1,int_1(\mathrm{\Delta }^1(c)I)1]|`$. Further, let $`s=3m4|\mathrm{\Delta }^1(c)|\frac{m}{2}2`$, the number of integers in the interval $`I`$ not colored by $`c`$. Finally, let $`w=|[int_1(\mathrm{\Delta }^1(c)I),int_m(\mathrm{\Delta }^1(c)I)]|m`$. It will be important later to note that
$$w+s2s2(\frac{m}{2}2)m3.$$
(1)
Let $`X=first_m(\mathrm{\Delta }^1(c)I)`$ (note that since $`\mathrm{\Delta }^1(c)3m\frac{m}{2}2>m`$, $`X`$ is in fact an $`m`$-set). By construction we have $`x_1=t+1`$ and $`x_m=t+w+m`$, so that $`x_mx_1=m+w1`$. Hence, if there is a monochromatic $`m`$-set $`Y`$ with $`y_mx_1+2(m1+w)=2m1+t+2w`$ and $`XY`$, then $`Y`$ is not $`X`$-admissible and the proof is complete. We show that
$$Y=last_m(\mathrm{\Delta }^1(c)I)$$
satisfies these conditions.
First, note that $`|\mathrm{\Delta }^1(c)|3m\frac{m}{2}22m`$ since $`m4`$, from which it follows that $`Y`$ is indeed an $`m`$-set and $`XY`$. We now show $`last_1(\mathrm{\Delta }^1(c)I)2m1+t+2w`$. Since there are exactly $`s(t+w)`$ integers $`z`$ with $`\mathrm{\Delta }(z)c`$ and $`z>x_m`$, it follows that $`last_1(\mathrm{\Delta }^1(c)I)3m4\left(s(t+w)\right)`$. Hence, recalling Equation 1, it follows that
$$\begin{array}{cc}\hfill last_1(\mathrm{\Delta }^1(c)I)& 3m4s+t+w\hfill \\ & 3m4(m3w)+t+w\hfill \\ & =2m1+t+2w,\hfill \end{array}$$
and the proof is complete. โ
###### Theorem 3.3.
Let $`m4`$ be an integer. Then, $`g(m,3)=7m+\frac{m}{2}6`$.
###### Proof.
One may verify that the coloring $`\mathrm{\Delta }:[3m1,7m+\frac{m}{2}7][1,3]`$ given by
$$1^{m\frac{m}{2}2}2^{\frac{m}{2}1}1^{2m1}2^{\frac{m}{2}}3^{m1}$$
shows $`g(m,3)7m+\frac{m}{2}6`$.
Next we show that $`g(m,3)7m+\frac{m}{2}6`$. Let $`\mathrm{\Delta }:[1,7m+\frac{m}{2}6][1,3]`$ be an arbitrary $`3`$-coloring. Since $`|I_2|=3(m1)1`$ it follows there exists some $`c[1,3]`$ such that $`|\mathrm{\Delta }^1(c)I_2|m1`$. If $`\mathrm{\Delta }^1(c)I_3\mathrm{}`$, then the proof is complete. We may therefore assume $`\mathrm{\Delta }^1(c)I_3=\mathrm{}`$ and thus $`|\mathrm{\Delta }(I_3)|2`$. Since $`|I_3|=m+\frac{m}{2}`$, if $`|I_2||\mathrm{\Delta }^1(c)I_2|\frac{m}{2}1`$ then it follows from the pigeonhole principle that some monochromatic $`m`$-set $`YI_2I_3`$ exists with $`YI_3\mathrm{}`$. In this case, an application of Proposition 2.1 completes the proof.
Finally, we are left to assume that $`|I_2||\mathrm{\Delta }^1(c)I_2|<\frac{m}{2}1`$, so that $`|\mathrm{\Delta }^1(c)I_2|3m\frac{m}{2}3`$. Translating $`I_2`$ to the interval $`[1,3m4]`$ and applying Lemma 3.2 completes the proof. โ
###### Lemma 3.4.
Let $`m3`$ be an integer, and let $`\mathrm{\Delta }:[1,4m5][1,4]`$ be a $`4`$-coloring. If $`|\mathrm{\Delta }^1(c)|3m3`$ for some $`c[1,4]`$, then $`\mathrm{\Delta }`$ is not an $`L(4)`$-coloring.
###### Proof.
The proof of Lemma 3.4 is similar to that of Lemma 3.2, and we omit it.โ
###### Theorem 3.5.
Let $`m3`$ be an integer. Then, $`g(m,4)=10m9`$.
###### Proof.
One may verify that the coloring $`\mathrm{\Delta }:[4m2,10m10][1,4]`$ given by
$$1^{m3}2^{m1}1^{2m1}3^{m1}4^{m1}$$
shows $`g(m,4)10m9`$.
Next we show that $`g(m,4)10m9`$. Let $`\mathrm{\Delta }:[1,10m9][1,4]`$ be an arbitrary $`4`$-coloring. Since $`|I_2|=4(m1)1`$, it follows that there exists $`c[1,4]`$ such that $`|\mathrm{\Delta }^1(c)I_2|m1`$. If $`\mathrm{\Delta }^1(c)I_3\mathrm{}`$ then the proof is complete. Otherwise we have $`\mathrm{\Delta }^1(c)I_3=\mathrm{}`$, and so $`|\mathrm{\Delta }(I_3)|3`$. Since $`|I_3|=2m1`$, if $`|I_2||\mathrm{\Delta }^1(c)I_2|m1`$ then it follows that some monochromatic $`m`$-set $`YI_2I_3`$ exists with $`YI_3\mathrm{}`$. In this case, an application of Proposition 2.1 completes the proof.
Finally, we are left to assume that $`|I_2||\mathrm{\Delta }^1(c)I_2|<m1`$, so that $`|\mathrm{\Delta }^1(c)I_2|3m3`$. Translating $`I_2`$ to the interval $`[1,4m5]`$ and applying Lemma 3.4 completes the proof. โ
## 4 Recursion in evaluating $`g(m,r)`$ when $`r5`$
Though the techniques used in the previous section may be duplicated in an attempt to solve $`g(m,r)`$ for $`r>4`$, the limitations of such an approach are easily seen. In this section we instead focus our attention on a more general argument which will allow us to solve $`g(m,r)`$ for certain infinite families of integers.
Developing this technique will require that we know certain properties of $`L(r)`$-colorings. The following two lemmas give some information concerning the structure of $`L(r)`$-colorings on the interval $`[1,g(m,r)k]`$.
###### Lemma 4.1.
Let $`m,r`$ be positive integers, and let $`\mathrm{\Delta }:[1,g(m,r)1][1,r]`$ be an $`r`$-coloring. If $`|\mathrm{\Delta }^1(c)[1,r(m1)]|m`$ for some $`c[1,r]`$, then $`\mathrm{\Delta }`$ is not an $`L(r)`$-coloring.
###### Proof.
Suppose for contradictionโs sake that $`X`$ is a monochromatic $`m`$-set with $`X[1,r(m1)]`$ and that $`\mathrm{\Delta }`$ is an $`L(r)`$-coloring of $`[1,g(m,r)1]`$. Then $`\mathrm{\Delta }|_{[x_m+1,g(m,r)1]}`$ is an $`L(r)`$-coloring such that no monochromatic $`m`$-set $`Y`$ exists with $`Y[x_m+1,g(m,r)1]`$ and $`Y[2x_mx_1,g(m,r)1]\mathrm{}`$. Applying Lemma 2.2 and its subsequent remark allows us to extend $`\mathrm{\Delta }|_{[x_m+1,g(m,r)1]}`$ to an $`L(r)`$-coloring $`\mathrm{\Delta }_e`$ of the interval $`[x_mr(m1),g(m,r)1]`$. Since $`x_mr(m1)`$, it follows that $`\mathrm{\Delta }_e|_{[0,g(m,r)1]}`$ is an $`L(r)`$-coloring, which after an appropriate translation contradicts the definition of $`g(m,r)`$. โ
###### Lemma 4.2.
Let $`m,r`$ and $`k`$ be positive integers, and let $`\mathrm{\Delta }:[1,g(m,r)k][1,r]`$ be an $`r`$-coloring. Let $`a=\mathrm{min}\left\{int_m(\mathrm{\Delta }^1(c))\right\}_{c[1,r]}`$. For each $`c[1,r]`$, let $`A_c(\mathrm{\Delta })=|\mathrm{\Delta }^1(c)[1,a1]|`$ and $`B_c(\mathrm{\Delta })=|\mathrm{\Delta }^1(c)[a+1,g(m,r)k]|`$. If
$$\underset{c[1,r]}{}\left(A_c(\mathrm{\Delta })+\mathrm{min}\{B_c(\mathrm{\Delta }),m1\}\right)r(2m2)k$$
(2)
then $`\mathrm{\Delta }`$ is not an $`L(r)`$-coloring.
###### Proof.
We use induction on $`k`$. Suppose $`k=1`$, and assume for contradictionโs sake that $`\mathrm{\Delta }`$ is an $`L(r)`$-coloring. By Lemma 4.1, it must be the case that $`a=r(m1)+1`$, so that $`[1,a]=I_1`$ and $`[a+1,g(m,r)1]=I_2I_3`$. Hence we have
$$\underset{c[1,r]}{}A_c(\mathrm{\Delta })=r(m1),$$
so by Equation 2 it must be the case that $`|\mathrm{\Delta }^1(c)(I_2I_3)|<m1`$ for some $`c[1,r]`$. Induce a coloring $`\mathrm{\Delta }_e:[1,g(m,r)][1,r]`$ defined by
$$\mathrm{\Delta }_e(x)=\{\begin{array}{cc}\mathrm{\Delta }(x),\hfill & \text{ for }x[1,g(m,r)1]\hfill \\ c,\hfill & \text{ for }x=g(m,r).\hfill \end{array}$$
By the definition of $`g(m,r)`$ there exist $`m`$-sets $`X,Y[1,g(m,r)]`$ with $`XY`$ and $`y_mx_12(x_mx_1)`$. Since $`\mathrm{\Delta }`$ is an $`L(r)`$-coloring, it follows that $`y_m=g(m,r)`$; furthermore $`y_1I_1`$ since $`|\mathrm{\Delta }^1(c)(I_2I_3)|<m1`$. Therefore, $`X[1,r(m1)]`$, a contradiction.
Assume the result holds for $`k`$; we show it also holds for $`k+1`$. Let $`\mathrm{\Delta }:[1,g(m,r)k1][1,r]`$ be such that
$$\underset{c[1,r]}{}A_c(\mathrm{\Delta })+\mathrm{min}\{B_c(\mathrm{\Delta }),m1\}r(2m2)k1.$$
(3)
We consider two cases.
Case 1. If $`a<r(m1)+1`$, then there must be some $`t[1,r]`$ such that $`|\mathrm{\Delta }^1(t)[1,a]|<m1`$. Induce a coloring $`\mathrm{\Delta }_e:[1,g(m,r)k][1,r]`$ defined by
$$\mathrm{\Delta }_e(x)=\{\begin{array}{cc}t,\hfill & \text{ for }x=1\hfill \\ \mathrm{\Delta }(x1),\hfill & \text{ for }x[2,g(m,r)k]\hfill \end{array}$$
Notice that for $`\mathrm{\Delta }_e`$ we have $`\mathrm{min}\{int_m(\mathrm{\Delta }^1(c)[1,g(m,r)k])\}_{c[1,r]}=a+1`$ so that
$$\underset{c[1,r]}{}A_c(\mathrm{\Delta }_e)+\mathrm{min}\{m1,B_c(\mathrm{\Delta }_e)\}r(2m2)k.$$
Hence, by induction there exist monochromatic $`m`$-sets $`X,Y`$ with $`XY`$ and
$$y_mx_12(x_mx_1).$$
(4)
If $`\mathrm{\Delta }`$ is an $`L(r)`$-coloring it follows that $`x_1=1`$; furthermore $`x_m>a+1`$ since $`|\mathrm{\Delta }_e^1(t)[1,a+1]|m1`$. Denoting the monochromatic $`m`$-set $`first_m(\mathrm{\Delta }_e(a+1)^1[1,a+1])`$ by $`Z`$, we therefore have $`x_1<z_1`$ and $`x_m>z_m`$. Along with Equation 4, this gives us
$$y_m+z_1>y_m+12x_m>2z_m,$$
from which it follows that $`y_mz_12(z_mz_1)`$, a contradiction. Therefore, $`\mathrm{\Delta }`$ is not an $`L(r)`$-coloring.
Case 2. If $`ar(m1)+1`$ (and hence $`a=r(m1)+1`$), we have
$$|\mathrm{\Delta }^1(c)[1,r(m1)]|=m1$$
(5)
for every $`c[1,r]`$. By Equation 3, there must be some $`t[1,r]`$ such that $`|\mathrm{\Delta }^1(t)(I_2I_3)|<m1`$. Induce a coloring $`\mathrm{\Delta }_e:[1,g(m,r)k][1,r]`$ defined by
$$\mathrm{\Delta }_e(x)=\{\begin{array}{cc}\mathrm{\Delta }(x),\hfill & \text{ for }x[1,g(m,r)k1]\hfill \\ t,\hfill & \text{ for }x=g(m,r)k.\hfill \end{array}$$
It is easily verified for $`\mathrm{\Delta }_e`$ that
$$\underset{c[1,r]}{}A_c(\mathrm{\Delta }_e)+\mathrm{min}\{B_c(\mathrm{\Delta }_e),m1\}r(2m2)k.$$
Hence, by induction there exist monochromatic $`m`$-sets $`X,Y`$ with $`XY`$ such that $`y_mx_12(x_mx_1)`$. If $`\mathrm{\Delta }`$ is an $`L(r)`$-coloring it follows that $`y_m=g(m,r)k`$; furthermore $`y_1r(m1)+1`$ since $`|\mathrm{\Delta }_e^1(t)(I_2I_3)|m1`$. Hence, $`X[1,r(m1)]`$, a contradiction.โ
We now develop a recursive technique for evaluating $`g(m,r)`$ given values of $`g(m,j)`$, $`j<r`$. The first theorem provides the means for evaluating $`g(m,r)`$ when $`r`$ belongs to the family of integers defined by the recurrence relation $`r_n=3r_{n1}r_{n2}`$ with particular initial conditions.
###### Theorem 4.3.
Let $`m,j`$ and $`r`$ be positive integers, with $`m2`$ and $`j<r`$. If $`r(m1)g(m,j)r(m1)+n`$ for $`mm_0`$, where $`r,n`$, and $`m_0`$ are positive integers, then
$$g(m,r)=(3rj)(m1)+1$$
for $`m\mathrm{max}\{m_0,n+1\}`$.
###### Proof.
By hypothesis there exists $`\mathrm{\Delta }_j:[r(m1)+2,2r(m1)][1,j]`$ which is an $`L(j)`$-coloring for $`mm_0`$. For convenience, let
$$_i=[(2r+i1)(m1)+1,(2r+i)(m1)]$$
for $`i[1,rj]`$. Define the function $`\mathrm{\Delta }_r:[r(m1)+2,(3rj)(m1)][1,r]`$ as follows
$$\mathrm{\Delta }_r(x)=\{\begin{array}{cc}\mathrm{\Delta }_j(x),\hfill & \text{ for }x[r(m1)+2,2r(m1)]\hfill \\ j+i,\hfill & \text{ for }x_i,i[1,rj].\hfill \end{array}$$
That $`\mathrm{\Delta }_r`$ is an $`L(r)`$-coloring follows since $`\mathrm{\Delta }_j`$ is an $`L(j)`$-coloring. Since for each $`c\mathrm{\Delta }(I_3)`$ we have $`|\mathrm{\Delta }^1(c)(I_2I_3)|=m1`$, we see that $`g(m,r)>(3rj)(m1)`$ for $`mm_0`$.
Now, let $`\mathrm{\Delta }:[1,(3rj)(m1)+1][1,r]`$ be an arbitrary $`r`$-coloring and $`m\mathrm{max}\{m_0,n+1\}`$. Let $`\mathrm{\Delta }(I_3)=C`$ and $`k=|C|`$. We proceed to show that $`\mathrm{\Delta }`$ is not an $`L(r)`$-coloring by case analysis of $`k`$.
Case 1. Suppose $`krj`$. Since $`|I_3|=(rj)(m1)+1`$, it follows that there must be some $`c[1,r]`$ such that $`|\mathrm{\Delta }^1(c)I_3|m`$, whence $`\mathrm{\Delta }`$ is not an $`L(r)`$-coloring by Proposition 2.1.
Case 2. Suppose $`k>rj`$. Let $`S=\mathrm{\Delta }^1(C)(I_2I_3)`$ and let $`U=SI_2`$. Let $`t=|\mathrm{\Delta }(I_2)||\mathrm{\Delta }(I_2)C|`$, so that $`trk<j`$. Assume without loss of generality that $`\mathrm{\Delta }(I_2)\{\mathrm{\Delta }(I_2)C\}=[1,t]`$. Furthermore, we may assume $`|S|k(m1)`$, since otherwise some monochromatic $`m`$-set $`Y`$ exists with $`YI_2I_3`$ and $`YI_3\mathrm{}`$ and we are done. Hence, since $`|I_3|=(rj)(m1)+1`$, we have that $`|U|=|S||I_3|(kr+j)(m1)1`$.
Let $`๐ซ`$ be a partition of $`U`$ into $`p=jtkr+j`$ sets $`\gamma _1,\mathrm{},\gamma _p`$ such that $`|\gamma _i|m1`$ for each $`i[1,p]`$. Define a coloring $`\widehat{\mathrm{\Delta }}:I_2[1,j]`$ as follows:
$$\widehat{\mathrm{\Delta }}(x)=\{\begin{array}{cc}\mathrm{\Delta }(x),\hfill & \text{ for }\mathrm{\Delta }(x)[1,t]\hfill \\ t+i,\hfill & \text{ for }x\gamma _i,i[1,jt].\hfill \end{array}$$
Using the notation of Lemma 4.2 and the fact that $`A_c(\mathrm{\Delta })+B_c(\mathrm{\Delta })|\mathrm{\Delta }^1(c)|`$ when $`|\mathrm{\Delta }^1(c)|m1`$, we note that
$$\begin{array}{cc}\hfill \underset{c[1,j]}{}A_c(\widehat{\mathrm{\Delta }})+\mathrm{min}\{B_c(\widehat{\mathrm{\Delta }}),m1\}& t(2m2)+|U|\hfill \\ & (2t+kr+j)(m1)1.\hfill \end{array}$$
Since $`g(m,j)n1r(m1)1`$, Lemma 4.2 implies that $`\widehat{\mathrm{\Delta }}`$ is not an $`L(j)`$-coloring if $`(2t+kr+j)(m1)1j(2m2)n1`$. Using $`trk<j`$, this inequality is easily verified for $`mn+12`$. Hence there exist monochromatic $`m`$-sets $`X,YI_2`$ where $`XY`$ and $`2(x_mx_1)y_mx_1`$. Moreover, $`\widehat{\mathrm{\Delta }}(X)[1,t]`$ and $`\widehat{\mathrm{\Delta }}(Y)[1,t]`$ since $`|\widehat{\mathrm{\Delta }}^1(t+i)|<m`$ for each $`i[1,p]`$. Thus, $`X`$ and $`Y`$ are monochromatic in $`\mathrm{\Delta }`$, and the proof is complete. โ
###### Example 1.
Consider the alternate proof that $`g(m,2)=5(m1)+1`$ for $`m3`$: note that $`g(m,1)`$ is trivially $`2m=2(m1)+2`$ for all positive $`m`$; by the previous proof, we have $`g(m,2)=5(m1)+1`$ for all $`m3`$.
As another example, we have seen in Theorem 3.1 that $`g(m,2)=5(m1)+1`$ for all $`m2`$. By the previous theorem, this implies $`g(m,5)=13(m1)+1`$ for $`m2`$, which in turn implies $`g(m,13)=34(m1)+1`$ for $`m2`$.
Likewise, we have see in Theorem 3.5 that $`g(m,4)=10(m1)+1`$ for all $`m3`$. The previous theorem gives $`g(m,10)=26(m1)+1`$ for $`m3`$, which in turn implies $`g(m,26)=68(m1)+1`$ for $`m3`$.
More explicitly, Theorems 3.1 and 3.5 can be used in conjunction with Theorem 4.3 to solve $`g(m,r_n)`$, when $`r_n`$ is in the family of integers generated by the recurrence relation
$$r_n=3r_{n1}r_{n2}$$
(6)
with initial conditions $`r_0=2,r_1=5`$ from Theorem 3.1 or $`r_0=4,r_1=10`$ from Theorem 3.5 .
One can solve these recurrence relations in terms of the Fibonacci numbers. In particular the initial value set $`r_0=2,r_1=5`$ gives $`r_n=5f_{2n}2f_{2n2}`$, where $`f_0=0`$ and $`f_1=1`$ are the first two Fibonacci numbers. Using properties of Fibonacci sequence simplifies this expression to $`r_n=f_{2n+3}`$. Of course the recurrence relation with initial conditions $`r_0=4`$ and $`r_1=10`$ then has general solution $`r_n=2f_{2n+3}`$.
$`\mathrm{}`$
Our ultimate goal is to evaluate $`g(m,r)`$ for as many $`r`$ as possible. Although Theorem 4.3 is an important step in that direction, it is of no use without the proper asymptotic value $`g(m,r_0)=r_1(m1)+n`$. We shall need another result to provide a bound on $`g(m,r)`$ so that we may apply Theorem 4.3.
###### Theorem 4.4.
Let $`m,j`$ and $`r`$ be positive integers, with $`m2`$ and $`j+1<r`$. If $`(r2)(m1)g(m,j)`$ for $`mm_0`$ and $`g(m,j+1)(r+1)(m1)+n`$ for $`mm_1`$, where $`r,n,m_0`$, and $`m_1`$ are positive integers, then
$$(3rj1)(m1)<g(m,r)(3rj1)(m1)+n$$
for $`m\mathrm{max}\{m_0,m_1\}`$.
###### Proof.
By hypothesis there exists $`\mathrm{\Delta }_j:[(r+1)(m1)+2,(2r1)(m1)][1,j]`$ which is an $`L(j)`$-coloring for $`mm_0`$. Define $`\mathrm{\Delta }_{j+1}:[r(m1)+2,2r(m1)][1,j+1]`$ as follows
$$\mathrm{\Delta }_{j+1}(x)=\{\begin{array}{cc}j+1,\hfill & \text{ for }x[r(m1)+2,(r+1)(m1)+1]\hfill \\ & \text{ or }x[(2r1)(m1)+1,2r(m1)]\hfill \\ \mathrm{\Delta }_j(x),\hfill & \text{ otherwise. }\hfill \end{array}$$
Since $`\mathrm{\Delta }_j`$ is an $`L(j)`$-coloring it follows immediately that $`\mathrm{\Delta }_{j+1}`$ is an $`L(j+1)`$-coloring.
As before, let
$$_i=[(2r+i1)(m1)+1,(2r+i)(m1)]$$
for $`i[1,rj1]`$. Define the function $`\mathrm{\Delta }_r:I_2[2r(m1)+1,(3rj1)(m1)][1,r]`$ as follows
$$\mathrm{\Delta }_r(x)=\{\begin{array}{cc}\mathrm{\Delta }_{j+1}(x),\hfill & \text{ for }xI_2\hfill \\ j+1+i,\hfill & \text{ for }x_i,i[1,rj1].\hfill \end{array}$$
From Lemma 2.2 and its subsequent remark, $`\mathrm{\Delta }_r`$ can be extended to an $`L(r)`$-coloring of $`[1,(3rj1)(m1)]`$, and so $`g(m,r)>(3rj1)(m1)`$.
Let $`\mathrm{\Delta }:[1,(3rj1)(m1)+n+1][1,r]`$ be a given $`r`$-coloring, and let $`m\mathrm{max}\{m_0,m_1\}`$. Let $`\mathrm{\Delta }(I_3)=C`$ and $`k=|C|`$. We proceed to show that $`\mathrm{\Delta }`$ is not an $`L(r)`$-coloring by case analysis of $`k`$.
Case 1. Suppose $`krj1`$. Since $`|I_3|=(rj1)(m1)+n`$ where $`n1`$ it follows that there must be some $`c[1,r]`$ such that $`|\mathrm{\Delta }^1(c)I_3|m`$, whence $`\mathrm{\Delta }`$ is not an $`L(r)`$-coloring by Proposition 2.1.
Case 2. Suppose $`k>rj1`$. Let $`S=\mathrm{\Delta }^1(C)(I_2I_3)`$ and let $`U=SI_2`$. Let $`t=|\mathrm{\Delta }(I_2)||\mathrm{\Delta }(I_2)C|`$, so that $`trk<j+1`$. Assume without loss of generality that $`\mathrm{\Delta }(I_2)\{\mathrm{\Delta }(I_2)C\}=[1,t]`$. Furthermore, we may assume $`|S|k(m1)`$, since otherwise some monochromatic $`m`$-set $`Y`$ exists with $`YI_2I_3`$ and $`YI_3\mathrm{}`$, and we are done. Since $`|I_3|=(rj1)(m1)+n`$, we have that $`|U|=|S||I_3|(kr+j+1)(m1)n`$.
Let $`๐ซ`$ be a partition of $`U`$ into $`p=j+1tkr+j+1`$ sets $`\gamma _1,\mathrm{},\gamma _p`$ such that $`|\gamma _i|m1`$ for each $`i[1,p]`$. Define a coloring $`\widehat{\mathrm{\Delta }}:I_2[1,j+1]`$ as follows:
$$\widehat{\mathrm{\Delta }}(x)=\{\begin{array}{cc}\mathrm{\Delta }(x),\hfill & \text{ for }\mathrm{\Delta }(x)[1,t]\hfill \\ t+i,\hfill & \text{ for }x\gamma _i,i[1,j+1t].\hfill \end{array}$$
Using the notation of Lemma 4.2 and the fact that $`A_c(\mathrm{\Delta })+B_c(\mathrm{\Delta })|\mathrm{\Delta }^1(c)|`$ when $`|\mathrm{\Delta }^1(c)|m1|`$, we have
$$\begin{array}{cc}\hfill \underset{c[1,j+1]}{}A_c(\widehat{\mathrm{\Delta }})+\mathrm{min}\{B_c(\widehat{\mathrm{\Delta }}),m1\}& t(2m2)+|U|\hfill \\ & (2t+kr+j+1)(m1)n.\hfill \end{array}$$
Since $`g(m,j+1)mnr(m1)1`$, Lemma 4.2 implies that $`\widehat{\mathrm{\Delta }}`$ is not an $`L(j+1)`$-coloring if $`(2t+kr+j)(m1)n(j+1)(2m2)mn`$. Using $`trk<j+1`$, this is easily verified for all $`m2`$. Hence there exist monochromatic $`m`$-sets $`X,YI_2`$ where $`XY`$ and $`2(x_mx_1)y_mx_1`$. Moreover, $`\widehat{\mathrm{\Delta }}(X)[1,t]`$ and $`\widehat{\mathrm{\Delta }}(Y)[1,t]`$ since $`|\widehat{\mathrm{\Delta }}^1(t+i)|<m`$ for each $`i[1,p]`$. Thus, $`X`$ and $`Y`$ are monochromatic in $`\mathrm{\Delta }`$, and the proof is complete.โ
###### Example 2.
From Theorems 3.1 and 3.3 we have that $`g(m,2)=5(m1)+1`$ for $`m2`$ and $`g(m,3)8(m1)+1`$ for $`m4`$. We see from Theorem 4.4 that $`g(m,7)=18(m1)+1`$ for $`m4`$. Repeated use of Theorem 4.3 provides another infinite family $`\{r_n\}`$ for which $`g(m,r_n)=r_{n+1}(m1)+1`$. Here the elements $`r_n`$ satisfy Equation 6 with initial conditions $`r_0=7,r_1=18`$. This family can also be expressed in terms of the Fibonacci numbers, with
$$r_n=18f_{2n}7f_{2n2}.$$
Likewise, from Theorems 3.3 and 3.5 we have that $`g(m,3)>7(m1)+1`$ for $`m4`$ and $`g(m,4)=10(m1)+1`$ for $`m3`$. Applying Theorem 4.3, we have $`g(m,9)=23(m1)+1`$ for $`m4`$. Again, repeated use of Theorem 4.4 solves $`g(m,r_n)=r_{n+1}(m1)+1`$, where here
$$r_n=23f_{2n}9f_{2n2}.$$
$`\mathrm{}`$
The next result gives a fairly loose bound for $`g(m,r)`$ given values of $`g(m,j)`$, $`j<r`$. However, it bounds the function $`g(m,r)`$ such that Theorem 4.4 may be invoked.
###### Theorem 4.5.
Let $`m,j`$ and $`r`$ be positive integers, with $`m2`$ and $`j<r`$. If $`(r1)(m1)+1g(m,j)<r(m1)`$ for $`mm_0`$, where $`r`$ and $`m_0`$ are positive integers, then
$$(3rj1)(m1)+1<g(m,r)(3rj)(m1)$$
for $`mm_0`$.
###### Proof.
We start with the lower bound. By hypothesis there exists $`\mathrm{\Delta }_j:[(r+1)(m1)+1,2r(m1)][1,j]`$ which is an $`L(j)`$-coloring for $`mm_0`$. As before, let
$$_i=[(2r+i1)(m1)+1,(2r+i)(m1)]$$
for $`i[1,rj1]`$. Define the function $`\mathrm{\Delta }_r:[r(m1)+2,(3rj1)(m1)+1][1,r]`$ as follows
$$\mathrm{\Delta }_r(x)=\{\begin{array}{cc}j+1,\hfill & \text{ if }x[r(m1)+2,(r+1)(m1)]\hfill \\ & \text{ or }x=(3rj1)(m1)+1\hfill \\ \mathrm{\Delta }_j(x),\hfill & \text{ for }x[(r+1)(m1)+1,2r(m1)]\hfill \\ j+1+i,\hfill & \text{ for }x_i,i[1,rj1].\hfill \end{array}$$
It is not difficult to see that $`\mathrm{\Delta }_r`$ is an $`L(r)`$-coloring on $`I_2`$ such that there is no monochromatic $`m`$-set $`YI_2I_3`$ with $`y_mI_3`$. Thus, it follows from Proposition 2.1 and Lemma 2.2 that $`g(m,r)>(3rj1)(m1)+1`$ for every $`mm_0`$.
To show that $`g(m,r)(3rj)(m1)`$, let $`\mathrm{\Delta }:[1,(3rj)(m1)][1,r]`$ be an arbitrary $`r`$-coloring. Let $`\mathrm{\Delta }(I_3)=C`$ and $`k=|C|`$. We proceed to show that $`\mathrm{\Delta }`$ is not an $`L(r)`$-coloring by case analysis of $`k`$.
Case 1. Suppose $`k<rj`$. Since $`|I_3|=(rj)(m1)`$, it follows that there must be some $`c[1,r]`$ such that $`|\mathrm{\Delta }^1(c)I_3|m`$, whence $`\mathrm{\Delta }`$ is not an $`L(r)`$-coloring by Proposition 2.1.
Case 2. Suppose $`k=rj`$. Since $`g(m,j)<r(m1)`$ and $`|I_2|=r(m1)1`$, if $`|\mathrm{\Delta }(I_2)|j`$ then $`\mathrm{\Delta }`$ is not an $`L(j)`$-coloring. Hence $`\mathrm{\Delta }(I_2)>j`$ so that $`\mathrm{\Delta }(I_2)\mathrm{\Delta }(I_3)\mathrm{}`$, and it follows that there exists some $`z\mathrm{\Delta }^1(C)I_2`$. Since $`|I_3\{z\}|=(rj)(m1)+1`$, there must be some monochromatic $`m`$-set $`Y`$ such that $`YI_2I_3`$ and $`y_mI_3`$. Applying Proposition 2.1 completes the proof.
Case 3. Suppose $`k>rj`$. Let $`S=\mathrm{\Delta }^1(C)(I_2I_3)`$ and let $`U=SI_2`$. Let $`t=|\mathrm{\Delta }(I_2)||\mathrm{\Delta }(I_2)C|`$, so that $`trk`$. Assume for simplicity that $`\mathrm{\Delta }(I_2)\{\mathrm{\Delta }(I_2)C\}=[1,t]`$. Furthermore, we may assume $`|S|k(m1)`$, since otherwise some monochromatic $`m`$-set $`Y`$ exists with $`YI_2I_3`$ and $`YI_3\mathrm{}`$. Hence, since $`|I_3|=(rj)(m1)`$, we have that $`|U|(kr+j)(m1)`$.
Let $`๐ซ`$ be a partition of $`U`$ into $`p=jtkr+j`$ sets $`\gamma _1,\mathrm{},\gamma _p`$ such that $`|\gamma _i|m1`$ for each $`i[1,p]`$. Define a coloring $`\widehat{\mathrm{\Delta }}:I_2[1,j]`$ as follows
$$\widehat{\mathrm{\Delta }}(x)=\{\begin{array}{cc}\mathrm{\Delta }(x),\hfill & \text{ for }\mathrm{\Delta }(x)[1,t]\hfill \\ t+i,\hfill & \text{ for }x\gamma _i,i[1,jt].\hfill \end{array}$$
Since $`g(m,j)<r(m1)`$ and $`|I_2|=r(m1)1`$, there exist monochromatic $`m`$-sets $`X,YI_2`$ where $`XY`$ and $`2(x_mx_1)y_mx_1`$. Moreover, $`\widehat{\mathrm{\Delta }}(X)[1,t]`$ and $`\widehat{\mathrm{\Delta }}(Y)[1,t]`$ since $`|\widehat{\mathrm{\Delta }}^1(t+i)|<m`$ for each $`i[1,p]`$. Thus, $`X`$ and $`Y`$ are monochromatic in $`\mathrm{\Delta }`$, and the proof is complete. โ
###### Example 3.
By Theorem 3.1 we have $`g(m,2)=5(m1)+1`$ for $`m2`$. Applying Theorem 4.5 we have
$$15(m1)+1<g(m,6)16(m1)$$
for $`m2`$.
Likewise, by Theorem 3.3 we have $`7(m1)+1g(m,3)<8(m1)`$ for $`m5`$. From this we see
$$20(m1)+1<g(m,8)21(m1)$$
for $`m5.`$
$`\mathrm{}`$
## 5 Conclusion and Conjectures
In the previous two sections we gave either an exact solution to or a bound on $`g(m,r)`$ for all $`r[2,10]`$ and sufficiently large $`m`$. Of course, we could use Theorems 4.3, 4.4, and 4.5 to solve or bound $`g(m,r)`$ for many $`r>10`$. We conjecture that for each positive integer $`r`$ one may find a positive integer $`j_r`$ such that one of Theorems 4.3, 4.4, or 4.5 may be used to solve or bound $`g(m,r)`$.
We have verified by computer the existence of some $`j_r`$ for each $`r10^5`$. This program was also used to calculate the proportions in which exact or bounded results appear in these first $`10^5`$ integers, finding that approximately $`38.2\%`$ of integers have exact solutions (generated by Theorem 4.3), $`23.6\%`$ are bounded by a constant (generated by Theorem 4.4), and the remaining $`38.2\%`$ are bounded by a coefficient on $`m`$ (generated by Theorem 4.5). Furthermore, these proportions are represented in much smaller samples, perhaps suggesting that these values are near the asymptotic proportions.
## Acknowledgement
The author wishes to express his thanks to Professor A. Bialostocki for his kind supervision and to D. Grynkiewicz for fruitful discussions. He would also like to thank two anonymous referees for their careful corrections and excellent suggestions.
|
warning/0506/gr-qc0506063.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
We begin by briefly reviewing the role of scalars and scalar fields in physics. Before Einstein, the basic relativity principle in Galilean-Newtonian physics required invariance in form of the laws of physics under transformations of the Galilean group. Restricting ourselves to Newtonian gravity and Maxwellโs electromagnetism in this context, we can easily find examples of scalars, such as mass, electric charge, energy, etc., under the static (excluding velocity transformations) affine subgroup of the full Galilean group. However, when we allow constant velocity transformations, the notion of scalars becomes a little less obvious. For example, under the (constant) rotation group spatial intervals are clearly scalars, but this is not the case for non-trivial velocity transformations, for which the spatial distance between two events at the origin of one frame is zero as measured in that frame, but not zero in another. Similar considerations apply to speed, and thus kinetic energy. Clearly, the kinetic energy of a particle is not a scalar under non-trivial velocity transformations. Similarly, when we try to understand Maxwellโs electromagnetism in terms of a โscalarโ and a โvectorโ potential, we find ourselves not able to formulate a consistent theory invariant under constant velocity transformations, and must rely on some fixed rest frame such as the ether. Of course, these considerations are precisely those that led from Galilean to Einsteinian special relativity, and a formulation of Maxwellโs electromagnetism in terms of a four-vector potential, with the complete elimination of any scalar component of the electromagnetic potential. The next step, from special to general relativity describes gravity in terms of a tensor, not a scalar, field.
Thus, while scalars (constants) naturally abound, fundamental, i.e., not derived, scalar fields are only hypothetical to date. In cosmology, pure Einstein theory uses only a 2-tensor, while in quantum theory, the observed โparticles,โ such as quarks, leptons, are represented by fermi spinors and the โgauge forcesโ are carried by boson vector fields.
* As of 2004, fundamental scalars appear only as hypothetical, as yet unobserved, fields related possibly to the gravitational or cosmological โconstants,โ dark energy, inflatons, dilatons, or Higgs fields.
In other words, nature seems to abhor using fields which have the same value in all reference frames. This is surely a curious fact.
## 2 Special Relativistic gravity
In the early days of special relativity, Einsteinโs first successful field theory was a special relativistic re-formulation of Maxwellโs electromagnetic theory. Newtonian mechanics could be reformulated in terms of force as a four-vector, $`^\alpha ,`$ leading to a fully Lorentz covariant theory of mechanics describing motion parameterized by proper time, $`x^\alpha (\tau )`$
$$\frac{d}{d\tau }(m_i\frac{dx^\alpha }{d\tau })=^\alpha .$$
(1)
where $`m_i`$ is a constant inertial mass, and of course,
$$\eta _{\alpha \beta }\frac{dx^\alpha }{d\tau }\frac{dx^\beta }{d\tau }=1.$$
(2)
The constancy of $`m_i`$ and the consistency of (1) and (2) then require that the four-force be four-orthogonal to the velocity,
$$\eta _{\alpha \beta }\frac{dx^\alpha }{d\tau }^\beta =0.$$
(3)
This is clearly satisfied identically for the electromagnetic four-force, $`^\alpha =F^{\alpha \beta }\eta _{\beta \gamma }\frac{dx^\gamma }{d\tau }`$
But what of a gravitational field theory?
* How does gravity fit into special relativity?
First, recall Newtonโs formulation in terms of a Galilean scalar gravitational potential field:
$$^2\varphi =\frac{\kappa }{2}\rho _{ag},$$
(4)
where $`\rho _{ag}`$ is mass density, with the $`ag`$ subscript indicating that here the mass is acting as a source for gravity. Also, $`\kappa 8\pi G,`$ and $`G`$ the usual Newton constant. Using Galilean three-vector notation,
$$๐_g=\varphi ,$$
(5)
the equations of motion become
$$\frac{d}{dt}\left(m_i\frac{d๐ซ}{dt}\right)=m_{pg}๐_g.$$
(6)
Here the $`i`$ subscript indicates inertial mass, while $`pg`$ corresponds to passive gravitational mass. Of course, it was and is common to simply assume
$$\frac{m_{ga}}{m_{gp}}=1,$$
(7)
and
$$\frac{m_{gp}}{m_i}=1.$$
(8)
It is fairly easy to give an argument that momentum conservation requires the satisfaction of (7). On the other hand, (8) is less trivial, and corresponds to the operationally significant
* Weak principle of equivalence:Gravitational acceleration at a given point is independent of mass.
So, as was common around 1900, let us temporarily assume
$$m_{ag}=m_{pg}=m_i=m=constant.$$
(9)
Finally, before leaving pre-Einsteinian gravity, we note that this potential, $`\varphi ,`$ has units of velocity squared, so that in the standard relativistic choice used in this paper, $`c=1`$, $`\varphi `$ is dimensionless.
So, how do Einstein and his colleagues attack the problem of integrating Newtonian gravity into special relativity? Fortunately there are excellent, easily readable, accounts of this process, , . What might seem to be the most natural way to proceed? Simply assume that gravity in special relativity will be described by a 4-scalar, $`\varphi ,`$ satisfying
$$\text{o}^2\varphi =\frac{\kappa }{2}\rho ,$$
(10)
$$_g^\alpha =m\varphi ^{,\alpha },$$
(11)
(1) as equation of motion. However, (3), applied to (11) results in
$$\frac{dx^\alpha }{d\tau }\frac{\varphi }{x^\alpha }=\frac{d\varphi }{d\tau }=0.$$
(12)
In other words, if we use (11) and (3) the potential must constant along the path of every particle, so the gravitational force must necessarily be zero on every particle! Clearly, something is wrong here.
Consider the problem from the viewpoint of an action. A point particle with path $`z^\mu (\tau ),`$ and density, $`\delta ^4(x^\nu z^\nu (\tau ).`$ Here $`\tau `$ is proper time, so
$$\eta _{\mu \nu }\frac{dx^\mu }{d\tau }\frac{dx^\nu }{d\tau }=1.$$
(13)
As a guide, look at the electromagnetic equations, field and particle motion, as derived from particle, field, and interaction parts,
$$\begin{array}{cc}A_p+A_{em}+A_I\hfill & =\left(m\sqrt{\dot{z}^\mu \dot{z}_\mu }\delta ^4(x^\mu z^\mu (\tau ))๐\tau \right)d^4x\hfill \\ & \frac{1}{16\pi }(A_{\mu ,\nu }A_{\nu ,\mu })(A^{\mu ,\nu }A^{\nu ,\mu })d^4x+\hfill \\ & q\left(\dot{z}^\mu (\tau )A_\mu \delta ^4(x^\nu z^\nu (\tau ))๐\tau \right)d^4x.\hfill \end{array}$$
(14)
Now consider a scalar gravitational modification of such a formalism,
$$\begin{array}{cc}A_p+A_\varphi +A_I\hfill & =\left(m\sqrt{\dot{z}^\mu \dot{z}_\mu }\delta ^4(x^\mu z^\mu (\tau ))๐\tau \right)d^4x\hfill \\ & \frac{1}{\kappa }\varphi _{,\mu }\varphi ^{,\mu }d^4x\hfill \\ & \varphi \left(m\sqrt{\dot{z}^\mu \dot{z}_\mu }\delta ^4(x^\mu z^\mu (\tau ))๐\tau \right)d^4x.\hfill \end{array}$$
(15)
The field variation of this action results in (10) with $`\rho (x^\mu )=m\delta ^4(x^\mu z^\mu (\tau ))๐\tau `$ in the conformal gauge, (13). However, the variation over the particleโs variables, $`z^\mu (\tau ),\dot{z}^\mu (\tau )`$ results in something quite new, namely,
$$\frac{d}{d\tau }\left(m(1+\varphi )\dot{z}^\mu (\tau )\right)=m\varphi ^{,\mu }.$$
(16)
When the four-vector equations of motion (16) are expressed in terms of local coordinate time, it is clear that local coordinate acceleration of a particle will depend not only on the the particleโs kinetic energy, but also on a modified inertial mass, $`m(1+\varphi )`$, thus violating the equal acceleration principle, WEP. Neglecting the $`\frac{d}{d\tau }\varphi `$ term, the usual expansion of the left side of (16) into local coordinate expressions gives
$$\frac{d^2๐ซ}{dt^2}(1v^2)\varphi .$$
(17)
Thus, the gravitational acceleration would depend on the velocity, so spinning bodies would have smaller accelerations in a gravitational field than non-spinning identical ones, hot bodies than cold, etc. Of course, this effect was too small to be noticed by early 20th century technology, but naturally Einstein was disturbed by the dependence of gravitational acceleration on internal structure of the falling body occurring in this initial attempt to โrelativizeโ gravity.
In a parallel vein, von Laue was looking into the models of internal stress in extended bodies and found what we now call the four dimensional stress-energy tensor, with $`T^{00}`$ identified with energy density, and $`T^{ij}=p^{ij}`$ the components of the spatial stresses on the body. However, the application of a Lorentz velocity transformation to such a tensor would mix the purely spatial stress components into the energy density, so that the energy density of a moving body would depend on its internal stress. Thus, these internal stress components should contribute to the gravitational mass.
It might have been something along these lines that motivated Einstein in 1907 to discount the appropriateness of a scalar special relativistic theory of gravitation because it did not allow โโฆthe inert mass of a body to depend on the gravitational potential.โ A related critique was formulated by Abraham, .
Actually, Nordstrรถm, suggested that the inertial mass might depend on $`\varphi `$,
$$m=m_0\mathrm{exp}\varphi ,$$
(18)
or for a weak field
$$m=m_0(1+\varphi ).$$
(19)
In fact, (16) is related to Nordstrรถmโs (18), with $`\mathrm{exp}(\varphi )1+\varphi ,`$ to first order in $`\varphi .`$ Most importantly, the resulting equation of motion, (16), is consistent with (13), as well as the suggested field and force equations, (10) and (11).
On the other hand, why associate $`\varphi `$ with the mass? Why not associate it with the metric, replacing
$$d\tau ^2=(dt^2dx^2dy^2dz^2),$$
(20)
with
$$d\tau ^2=\mathrm{exp}(2\varphi )(dt^2dx^2dy^2dz^2).$$
(21)
This is the direction taken by Einstein leading to his full general relativistic field equations using a 2-tensor, the metric as the potential. In the spirit of this paper this early scalar form for metric gravity, with the scalar appearing as a metric โdilationโ was very notable.
Historically, however, after its brief, but passing, appearance in a Nordstrรถm model, (21), there seemed to be no room in physics for a scalar field. But Nordstrรถmโs suggestion (18), led Einstein to further pursue a Machโs Principle in the sense of having inertial mass depend on the gravitational interaction of all of the other masses in the universe. We will briefly return to this later.
Of course, in parallel to relativity theory, quantum theory was being developed, and scalar fields again appear in the context of the Klein-Gordon equation. In turn, this equation, and its corresponding scalar field were replaced by Dirac equations. As we now understand observed quantum physics, elementary particles are fermions, satisfying Dirac equations, while forces correspond to gauge fields which, while bosons, are spacetime vectors rather than scalars. When we go beyond present day observation, however, scalar fields may indeed return, as Higgs bosons, dilatons, etc. We will mention these later.
As of the beginning of the 21st century, fundamental scalar fields exist only as hypothetical structures in physics, such as outlined in the following:
* Hypothetical non-quantum scalar fields
+ Scalar-tensor fields, such as the JBD determinant of $`G`$,
+ Inflatons, scalar field to give rise to observed anomalies of cosmological expansion,
* Hypothetical quantum scalar fields
+ Higgs particle, quantum scalar field providing mass by interactions with massless particles.
+ Dilatons, etc., quantum fields appearing in superstring and M theory.
## 3 The First Searches for Unified Field Theories
The hunt for a theory unifying gravity and electromagnetism began in the very earliest days of Einsteinโs general relativity. For our purpose, the most significant was the 5-dimensional versions associated with the names of Kaluza and Klein. Applequist et al have prepared a convenient review and reprints of original papers on the subject.
Briefly, KK theories enlarge the dimension of spacetime by one, so that the metric has a form
$$\gamma _{AB}=\left(\begin{array}{cc}V^2& V^2A_\beta \\ V^2A_\alpha & g_{\alpha \beta }+V^2A_\alpha A_\beta \end{array}\right).$$
(22)
By restricting the five dimensional transformation group appropriately, $`A_\alpha `$ appear as the components of a spacetime 4-vector, with $`V`$ a spacetime scalar<sup>2</sup><sup>2</sup>2Jordan et al, , were able to characterize these transformations as those of a projective group.. Furthermore, these transformations could also account for electromagnetic gauge transformations. Formally, this unification of spacetime and gauge transformations made this sort of formalism highly attractive, although the unobserved extra dimension was generally regarded as an obstacle to serious consideration of the model. Also, there was the question of the appearance of the unwanted scalar field, which Kaluza in 1921 described as โnoch ungedeutet.โ
But what are the field equations? By apparently natural extension of the four dimensional Einstein equations, consider
$$\delta d^5x\sqrt{|g^{(5)}|}=0.$$
(23)
These lead to spacetime four dimensional equations as well as 4 spacetime-fifth dimension equations and a single 5-5 equation, involving only derivatives of $`V=g_{55}.`$ The spacetime equations are
$$R_{\alpha \beta }\frac{1}{2}g_{\alpha \beta }R=\frac{V^2}{2}(F_{\alpha \mu }F_\beta ^\mu +\frac{\eta _{\alpha \beta }}{4}F_{\mu \nu }F^{\mu \nu })(\frac{V_{,\alpha ;\beta }}{V}\frac{\eta _{\alpha \beta }\text{o}^2V}{V}),$$
(24)
with $`F_{\alpha \beta }`$ the electromagnetic components derived from the potentials $`A_\alpha `$ as usual. These are the standard Einstein equations with electromagnetic stress tensor source, if we identify $`V^2`$ with 4 times the usual Newtonian gravitational constant, $`G`$. However, in these equations $`V`$ may not be constant and its derivatives also contribute.
* Equations (24) are the first hint of a varying gravitational constant.
## 4 Diracโs numbers
Meanwhile, Dirac , building on the work of Eddington and Milne, became intrigued by apparently โcoincidentalโ approximate equality between important physical quantities expressed in dimension free manner. Atomic scales can be deduced from $`\mathrm{},e`$ and $`m_p`$, say the mass of the proton. Then an atomic time(distance) scale is supplied by $`T_a=R_a=e^2/m_a.`$ On the other hand we have the age(distance scale) and the mass of the universe, $`T_u=R_u,M_u`$ as cosmological scales. Finally, we have the gravitational constant, $`\kappa `$. The resulting dimensionless quantities could be approximately grouped into powers of the incredibly large number, $`10^{40}`$,
$$\begin{array}{cc}\hfill \alpha =e^2/\mathrm{}& \hfill 10^0,\\ \hfill T_u/T_a& \hfill 10^{40},\\ \hfill T_a/\kappa & \hfill 10^{40},\\ \hfill M_u/m_p& \hfill 10^{80}.\end{array}$$
(25)
For our purposes, the combination of these equations in the following form is most important
$$\frac{1}{\kappa }M_u/R_u.$$
(26)
## 5 Scalar-Tensor Theories
Perhaps the earliest work in this direction was pursued independently by Jordan in Germany and Einstein and Bergmann in the US beginning in the late 1930โs. Of course, this work proceeded under all of the terrible constraints of the second world war and the resulting isolation of the two groups. Actually, Einstein and Bergmann apparently decided not to proceed with the variable gravitational constant idea to the point of publication. Bergmann reviews these parallel efforts in his paper, โUnified field theory with fifteen field variables,โ from which we now quote:
> In the spring of 1946, Professor W. Pauli turned over to the author of this paper galleys of a paper by P. Jordan entitled โGravitationstheorie mit veranderlicher Gravitationszahlโ, which was to have appeared in the Physikalische Zeitschrift sometime in 1945, but which was, of course, never published because the Phys. Zeitschrift in the meantime ceased publication. In this paper, Jordan attempted to generalize Kaluzaโs five dimensional unified field theory by retaining $`g_{55}`$ as a fifteenth field variable. Professor Einstein and the present author had worked on that same idea several years earlier, but had finally rejected it and not published that abortive attempt. The fact that another worker in this field has proposed the same idea, and independently, is an indication of its inherent plausibility. Therefore, it seemed worthwhile to review these attempts to โvary the constant of gravitationโ and to discuss the possibilities inherent in geometries of this kind.
Thus, independently of Einstein and Bergmann in the USA, Pascual Jordan and his colleagues in Germany began an extensive look at Kaluza-Klein theories with special concern for the possibility that the new five-dimensional metric component, a spacetime scalar, might play the role of a varying gravitational โconstant,โ as suggested by Diracโs (26). Certainly the resulting four-dimensional form of the field equations can interpreted this way. However, Jordan and his colleagues went beyond the 5-dimensional origins of this scalar and proposed purely four dimensional field equations involving a scalar field related to Newtonโs constant. Later Brans and Dicke independently arrived at similar point. However, for Brans and Dicke, Machโs ideas on inertial induction, that the total mass distribution in the universe should determine local inertial properties, were of prime concern. In fact, Sciama had earlier proposed a model theory of inertial induction.
Sciamaโs work provided a theoretical model in which inertial forces felt during acceleration of a reference frame relative to the โfixed stars,โ are of gravitational origin. From this assumption, Dicke argued that Machโs principle would manifest itself in having the ratio of inertial to gravitational mass depend on the average distribution of mass in the universe. That is,
Dickeโs form of Machโs Principle: The gravitational constant, $`\kappa ,`$ should be a function of the mass distribution in the universe.
Because of Diracโs large number hypothesis in the form
$$1/\kappa M/R,$$
(27)
it seems that the reciprocal of the gravitational constant will likely be the field quantity. In other words, $`1/\kappa `$ itself might be a field variable and satisfy a field equation with mass as a source, something like
$$\text{โ }\text{o}\text{ โ}\frac{1}{\kappa }=\rho .$$
(28)
So, introduce a scalar field, $`\varphi `$, which will play the role, at least locally and approximately, of the reciprocal Newtonian gravitational constant, $`\kappa `$.
The usual Lagrangian for Einstein theory including matter contains $`\kappa `$ directly multiplying the matter contributions. Keeping the field directly coupled to matter would then inevitably lead to changes in the local behavior of matter, the local equations of motion, as a result of variations in $`\varphi .`$ So, in order to incorporate a Machโs principle by way of a variable gravitational โconstant,โ we need to look at further modifications in the form of the general relativistic action. Begin with the standard Einstein action as
$$\delta d^4x\sqrt{g}(R+\kappa L_m)=0,$$
(29)
where $`L_m`$ is the โusualโ matter Lagrangian, a priori derived from some particular classical or quantum model. Equation (29) is clearly not enough since it provides no field equation for the new field, $`\kappa .`$
Before proceeding, we need to review some aspects of the famous (or infamous) โprinciple of equivalence.โ As usual we neglect tidal forces, extended bodies, etc., in these idealized models. Dicke often pointed out that we need to distinguish two forms:
* WEP. One form asserts that all bodies at the same spacetime point in a given gravitational field will undergo the same acceleration. We will refer to this as the โweakโ equivalence principle, WEP. As it stands, this does not exclude possible effects of gravity other than acceleration.
* SEP. A stronger statement, which actually is important to Einsteinโs general relativistic theory of gravity, is that the only influence of gravity is through the metric, and can thus (apart from tidal effects) be locally, approximately transformed away, by going to an appropriately accelerated reference frame. This is the โstrongโ principle, SEP.
An action of the form in (29) with variable $`\kappa ,`$ will clearly change the geodesic equation for test particles, thus, possibly the WEP, and even mass conservation. As Dicke noted, the Eรถtvรถs experiment verifies the WEP (but not the SEP), so, in the 1960โs, we would like to at least modify (29) to agree with the WEP. To ensure the geodesic equations for point particles we isolate $`\kappa `$ from matter in the original (29) by dividing by it,
$$\delta d^4x\sqrt{g}(\varphi R+L_m)=0,$$
(30)
where we have replaced $`\kappa 1/\varphi .`$ However, we should note the following. While we seem to have saved the geodesic equations for test particles, the motion of composite bodies is more complex. It turns out that the coupling of a new, universal field, $`\varphi `$ directly to the gravitational field gives rise to potentially observable effects for the motion of matter configurations to which gravitational energy contributes significantly. This is now known as theโDicke-Nordtvedtโ effect and has been investigated in the earth-moon system with the lunar laser reflector, leading to possible violations of even the WEP for extended masses. These possibilities were not considered in the early days. So, let us proceed to see what follows from (29). We need field equations for $`\varphi `$ so some action for this new field must be supplied,
$$\delta d^4x\sqrt{g}(\varphi R+L_m+L_\varphi )=0.$$
(31)
We must note, that by allowing a new, scalar, field, we are opening the door to other consequences. Since gravity is universally coupled to all physics, the direct coupling of $`\varphi `$ to geometry, $`\varphi R`$, means that $`\varphi `$ is universally coupled in some sense.
Consequences of (31): We are allowing for a possible violation of the SEP, since gravity, the universal interaction of mass, can influence local physics, not only through geometry, but also by changing the local universally coupled $`\varphi `$, thus changing internal gravitational structure.
The usual requirement that the field equations be second order leads to
$$L_\varphi =L(\varphi ,\varphi _{,\mu }).$$
(32)
Apart from this, there seem to be few a priori restrictions on $`L_\varphi .`$ The standard choice for a scalar field,
$$L_\varphi =\omega \varphi _{,\mu }\varphi _{,\nu }g^{\mu \nu },$$
(33)
results in a wave equation for $`\varphi `$ with $`R`$ as source seems natural. However, the coupling constant $`\omega `$ would itself then need to have the same dimensions as the gravitational $`\kappa `$ that the new field is to replace! But, one of the motivations for extending Einsteinโs theory is to eliminate the dimensional constant, $`\kappa .`$ So we require that any new coupling constant appear as dimensionless. An obvious natural minimal choice is
$$L_\varphi =\omega \varphi _{,\mu }\varphi _{,\nu }g^{\mu \nu }/\varphi ,$$
(34)
in which the field $`\varphi `$ has dimensions of inverse gravitational constant,
$$[\varphi ]=[\kappa ^1].$$
(35)
In fact, we will soon see, (41), that this results in field equations suggestive of (28). The form (34) leads to an action which is often referred to as the โJordan-Brans-Dicke,โ JBD, action,
$$\delta d^4x\sqrt{g}(\varphi R+L_m\frac{\omega }{\varphi }\varphi _{,\mu }\varphi _{,\nu }g^{\mu \nu })=0.$$
(36)
The variational principle, with standard topological and surface term assumptions, results in
$$\delta _m๐x^4\sqrt{g}L_m=0,$$
(37)
$$\varphi S_{\alpha \beta }=T_{(m)\alpha \beta }+\varphi _{;\alpha ;\beta }g_{\alpha \beta }\text{o}\varphi +\frac{\omega }{\varphi }(\varphi _{,\alpha }\varphi _{,\beta }\frac{1}{2}g_{\alpha \beta }\varphi _{,\lambda }\varphi ^{,\lambda }),$$
(38)
$$\omega (\frac{2\text{o}\varphi }{\varphi }\frac{\varphi _{,\lambda }\varphi ^{,\lambda }}{\varphi ^2})=R.$$
(39)
The first of these, (37), is the standard variational principle for matter, leading to the same expression for matter motion in terms of the metric. It thus (apparently) satisfies the weak equivalence principle. For example test particles, (37), follow geodesics. However, if the matter is extended, not a point particle, this is may no longer be true, even in standard general relativity. However, for scalar-tensor model, there is a second order interaction of matter through the scalar-metric coupling. This thus gives rise to violations of the weak equivalence principle. In other words, extended bodies of different mass may have different gravitational accelerations at the same point in a gravitational field. Of course, we do have the standard energy tensor for matter and resulting matter conservations laws. This is a result of the choice because of the free standing $`L_m`$ in (36), not directly coupled to $`\varphi `$,
$$T_{(m)\alpha ;\beta }^\beta =0.$$
(40)
We can couple $`\varphi `$ directly to matter by taking the trace of (38), solving for $`R`$. The result is another form for (39),
$$\text{o}\varphi =\frac{1}{(2\omega +3)}T_{(m)},$$
(41)
in which $`T_{(m)}`$ is the trace of the ordinary matter tensor. It should be noted that traceless matter, such as null electromagnetic fields, do not directly couple to $`\varphi .`$
Now, to look at the possible satisfaction of Diracโs (27), consider a weak field model situation with a static spherical shell of mass $`M`$, radius $`R`$ and otherwise empty universe this equation. The result is
$$\varphi \varphi _{\mathrm{}}+\frac{1}{4\pi (2\omega +3)}\frac{M}{R}.$$
(42)
Dividing equation (38) by $`\varphi `$ results in an equation in which the โordinaryโ matter tensor, $`T_{(m)\alpha \beta }`$ is divided by $`\varphi `$, which thus can be identified with the local reciprocal gravitational constant. Also, of course, the $`\varphi `$ contributes its own stress energy matter tensor to the right side of (38). If $`\varphi _{\mathrm{}}`$ is set zero as a default asymptotic condition, then (42) is seen to be consistent with the Dirac coincidence, (27). A natural approximation to (41) is to consider the effect of local matter over some background $`\varphi _0`$ equal to the present observed value,
$$\varphi \varphi _0+\frac{1}{4\pi (2\omega +3)}\underset{\text{local m}}{}\frac{m}{r}.$$
(43)
This can be regarded as an extension of Diracโs (27).
In equation (38), $`T_{(m)\alpha \beta }`$ are the components of the stress-energy tensor for matter derived from the matter Lagrangian $`L_m`$ in the standard fashion. Grouping this term with the $`\varphi `$ ones, results in an interpretation of
$$S_{\alpha \beta }=(1/\varphi )(T_{(m)\alpha \beta }+T_{(\varphi )\alpha \beta }),$$
(44)
as the total source for the Einstein tensor in(38). So, $`(1/\varphi )`$ does indeed act as a generalized gravitational โconstantโ, with both ordinary matter and the field $`\varphi `$ itself serving as sources for the metric. Actually the the $`\text{o}\varphi `$ term on the right hand side of (38), together with (41) results in two occurrences of the matter tensor as a source. Thus there could be some argument for renormalizing $`1/\varphi `$ as the โgravitational constantโ multiplying ordinary matter as it contributes to the Einstein tensor.
Pascual Jordan and his collaborators were the earliest serious investigators of equations of this sort. Most of the work by Jordan and his group is summarized in Jordanโs book, . See also a more recent review by Schรผcking . In addition to surveying the projective UFTโs motivation, Jordanโs book contains thorough studies of the static, spherically symmetric generalizations (the Heckmann solutions) of the Schwarzschild solution as well as cosmological solutions and other topics.
Equation (36) the brings to mind actions obtained by conformal changes of the metric. So, it is natural to look at the action of the local โconformal groupโ on the representations of the theory. Replace the metric, $`g_{\mu \nu }\overline{g}_{\mu \nu }=\psi g_{\mu \nu }`$. Discarding the surface (topological) part, (36) becomes
$$\delta d^4x\sqrt{\overline{g}}(\frac{\varphi }{\psi }\overline{R}+\frac{3\varphi }{2}\frac{|\overline{}\psi |^2}{\psi ^3}3\overline{}\psi \overline{}\varphi /\psi ^2+L_m/\psi ^2\frac{\omega }{\varphi \psi }|\overline{}\varphi |^2)=0.$$
(45)
If $`\psi `$ is chosen to be $`\varphi `$, (45) becomes
$$\delta d^4x\sqrt{\overline{g}}(\overline{R}(\omega +\frac{3}{2})|\overline{\alpha }|^2+e^{2\alpha }L_m(\overline{g}))=0,$$
(46)
where $`\varphi =e^\alpha `$. It is easy to see that this variational principle is just the Einstein one for a massless scalar field(dimensionless), $`\alpha `$, but universally coupled to all other matter through the $`e^{2\alpha }`$ factor. These conformal rescalings of the metric constitute the โmetric gauge group.โ Thus (46) is an expression of the theory in the โEinstein gauge,โ as opposed to the original (36), the โJordanโ gauge. But there is more to the conformal scaling than merely the formal expression of the equations. Most significantly, the universal coupling of $`\alpha `$ to all matter in (45) or (46) means that, in this metric, test particles will not follow geodesics, nor have conserved inertial mass, etc., in the Einstein gauge. In other words, conservation laws derived from the matter tensor depend on the construction of that tensor from the function multiplying $`\sqrt{g}`$ in the action, (45) or (46). Choosing the Einstein metric in (46) as the โphysicalโ metric leads to significant and observable violations of mass conservation and the WEP.
In the 1960โs and 1970โs, Bob Dicke was a leading influence influence in the push to experimentally test general relativity in Einsteinโs original form as well as alternatives such scalar-tensor generalizations. In fact, the explosion of interest in relativity and gravitational theories and tests was prompted at least in part by the presence of theoretically viable alternatives to standard Einstein theory, and Dickeโs energetic promotion of them. Also NASA was coming of age and searching for space related experiments of fundamental importance. The important bridge between theory and experiment in gravitational theories was developed by Thorne, Nordtvedt, Will and others . Their work provided rigorous underpinnings for the operational significance of various theories, especially in solar system context. An important tool is the parameterized post Newtonian (PPN) formalism which provides theoretical standard for expressing the predictions of relativistic gravitational theories in terms which can be directly related to experimental observations.
From (38), it appears that the equations of scalar-tensor theory approach those of standard Einstein theory as $`\varphi `$ approaches a constant. From (42) this would seem to occur in the limit of large $`\omega `$. Of course, this equation is just an approximation to a solution of (41) for an asymptotically empty universe, with $`\varphi 0`$ as boundary condition. Actually, these comments obscure the need for rigorous analysis for the action (36) as $`\omega \mathrm{}.`$ This is not surprising since the limiting dependence of solutions of field equations on parameters in these equations is in general a complex problem with all of the subtleties associated with the topology of a space of functions. However, it is true that
Approach to standard Einstein: In the realm of solar system experiments, the predictions of a theory of the form (36) approach those of standard Einstein theory as $`\omega \mathrm{}.`$
So, tests of such theories are often expressed as providing lower limits for $`\omega .`$ For more details, see .
As the experimental data on solar system gravitational measurements come in, new limits on the value of the parameter $`\omega `$ have become so large as to make the predictions of this theory essentially equivalent to those of standard Einstein theory. In other words, from solar system experimentation it seems that scalar-tensor modifications of standard Einstein theory would necessarily differ insignificantly from the standard.
Gravitational radiation provides another arena for experimental studies of gravitation. In 1975 the Hulse-Taylor binary pulsar decay data showed that gravitational radiation can provide another tool for testing gravitational theories. More recently, part of the justification for the LIGO gravitational radiation study is to provide further comparison of standard Einstein to alternative theories.
In spite of the apparently unpromising solar system experimental results, it turns out that universally coupled, thus gravitational, scalar fields continue to play important roles in contemporary physics. David Kaiser has given a review of this topic, comparing JBD and Higgs fields, for example. We will briefly consider some of these possibilities in the following sections.
## 6 Dilatons
As discussed in the introduction, it is surprising that scalar fields do not seem to occur naturally in special relativistic, pre-quantum physics. However, from the earliest days of quantum theory, scalar quantum fields were prevalent, first as the pre-relativistic Schroedinger wave function, then as the Klein-Gordon boson field, providing an early, but later discarded, model of a โmesonโ. Of course, Diracโs spinor took over as the basic field for โpermanentโ particles as fermions, with force-field carriers such as photons, being bosons. Of course, the photon field is a vector, not a scalar. However, investigations of internal spaces for particle symmetries directly involve gauge theories of force fields. In this model the internal symmetry spaces for families of fields have interesting transformation properties from the internal gauge group viewpoint but are nonetheless spacetime scalars. Some of the earliest are the $`SO(N)`$ bosons of the dual model, the Nambu-Goldstone bosons and the famous Higgs fields. Of course, the motivations for considering quantum scalar fields is certainly very different from those leading to the scalar field in scalar-tensor theories. Nevertheless, certain forms of the quantum formalism, and perhaps its macroscopic manifestations may turn out to be not too different from the classical scalar fields. Such comparisons may be most evident in the context of cosmological quantum particle models, . We begin with the quantum origin of โdilatons.โ
The late 1960โs and early 1970โs saw the birth of quantum dual models, which eventually led to string theory and later superstring theory, . These theoretical models quite naturally lead to a scalar field referred to as a โdilaton.โ This field couples directly to the trace of the two-dimensional string stress tensor. This coupling breaks the Weyl conformal symmetry of the string. Since conformal metric transformations are dilations, we arrive at the word โdilaton.โ The dilaton turns out to be what is needed to balance the quantum anomalies of this tensor by way of beta functionals of this tensor. In this analysis, the Einstein equations for the enveloping spacetime metric are โderivedโ as the beta functions. This rather involved arguments is discussed in the first volume of the book which thus provides useful description of the origin and role of dilatons. Here we only briefly summarize the argument in the following.
Start with a string action as a natural generalization of a point particle action. Given a background metric, $`g_{\alpha \beta },`$ an obvious choice is
$$S_1=\frac{1}{4\pi \alpha ^{}}d^2\sigma \sqrt{|h|}h^{ab}_aX^\alpha _bX^\beta g_{\alpha \beta }(X^c),$$
(47)
with internal coordinate area $`d^2\sigma ,`$ internal string metric, $`h_{ab},a,b\mathrm{}=1,2,`$ and $`\alpha ^{}`$ a tension related coupling parameter. Comparing $`S_1`$ to a relativistic point particle action, we see the need for an intrinsic surface metric, $`h_{ab}`$ for the string that is not present for point particle. Now, assume that the derived physics should be independent of the internal parameterization, that is the choice of string metric. However, any two-dimensional metric is conformally flat (but only locally, in general!),
$$h_{ab}=\varphi \eta _{ab},$$
(48)
with constant $`\eta _{ab}.`$ So the surface element appearing in (47) reduces to the flat one,
$$d^2\sigma \sqrt{|h|}h^{ab}=d^2\sigma \eta ^{ab}.$$
(49)
In addition to $`S_1`$, other terms have been proposed. One of these makes use of the string geometry through its curvature scalar,
$$\chi =\frac{1}{4\pi }d^2\sigma \sqrt{|h|}R^{(2)}.$$
(50)
Of course, one of the earliest discoveries relating geometry and topology was that this integral depends only on the topology of the string surface, and not the particular geometry. In fact, (50) defines the first Chern class for two dimensions. The value for $`\chi `$ is the Euler number of the surface, and cannot be a dynamical variable. However, dynamics can be restored by modifying the form of (50) by adding to (50) a scalar field factor, the โdilaton,โ $`\mathrm{\Phi }`$, giving
$$S_2=\frac{1}{4\pi }d^2\sigma \sqrt{|h|}\mathrm{\Phi }(X^c)R^{(2)}.$$
(51)
Classically this term breaks the conformal invariance. However, perhaps surprisingly, it is precisely this term which can restore conformal invariance after quantization. When the action $`S=S_1+S_2`$ is quantized, conformal invariance is broken (an anomaly) unless the external fields satisfy three equations. This argument is described in detail in GSW, volume 1, page 180. Here we drop the $`B_{\alpha \beta }`$ for simplicity and get (in the magical string dimension 26!) Einstein-like equations,
$$0=R_{\alpha \beta }2\mathrm{\Phi }_{;\alpha ;\beta },$$
(52)
$$0=4\mathrm{\Phi }_{,\alpha }\mathrm{\Phi }^{,\alpha }4\mathrm{\Phi }_{;\alpha }^{;\alpha }+R.$$
(53)
This โderivationโ of the Einstein equations from string theory was one of the attractive features of string theory. Recall, however, that this required the introduction of a dilaton, spacetime scalar, field to break conformal invariance, which is later restored only if Einstein-like equations are satisfied.
Now, without regard for their string theory origins, field equations can be derived from an โeffective action,โ
$$\delta d^DXe^{2\mathrm{\Phi }}(R4\mathrm{\Phi }_{,\alpha }\mathrm{\Phi }^{,\alpha })=0.$$
(54)
Of course, this action is nothing but a special case of the vacuum scalar-tensor one, (36), with $`2\mathrm{\Phi }=\mathrm{ln}\varphi ,`$ and $`\omega =1.`$ While the motivation and physics of the scalar field in the classical, pre-quantum, scalar-tensor theories is vastly different from the dilaton scalar field, it is difficult not to notice the close parallel between the universally coupled scalar of the old scalar-tensor theories and the new dilaton.
## 7 Inflatons
We will not attempt to review the rapidly expanding field of rapidly expanding (accelerating) cosmological models, but end this paper with a few comments about the early days of inflationary cosmology.
Standard general relativity has long been known to have difficulties in its application to observed cosmological facts. For example, standard general relativity requires that the initial big bang conditions be fantastically fine-tuned in order to result in the universe as we now see it some $`10^{11}`$ years later. See for example Peebles , Linde . Look at the standard Robertson-Walker isotropic homogeneous metric model,
$$ds^2=dt^2+R(t)^2d\sigma _ฯต^2,$$
(55)
where the three-space metric, $`d\sigma _ฯต^2`$, is hyperbolic, flat, or spherical depending on whether $`ฯต`$ is -1, 0 or +1. The Einstein equations result in
$$\left(\frac{\dot{R}}{R}\right)^2=\kappa \rho /3+\frac{ฯต}{R(t)^2}+\mathrm{\Lambda }/3.$$
(56)
Defining the Hubble variable as usual, this can be rewritten,
$$1=\mathrm{\Omega }+ฯต\mathrm{\Omega }_R+\mathrm{\Omega }_\mathrm{\Lambda },$$
(57)
where
$$\mathrm{\Omega }\frac{\kappa \rho }{3H^2},$$
(58)
$$\mathrm{\Omega }_R\frac{1}{(RH)^2},$$
(59)
and
$$\mathrm{\Omega }_\mathrm{\Lambda }\frac{\mathrm{\Lambda }}{3H^2}.$$
(60)
As of the 1980โs these three quantities were measured to be each in the ballpark of one. In fact,
$$\mathrm{\Omega }(now)\frac{\kappa M}{R}10^0,$$
(61)
which is one of Diracโs large number coincidences which was so instrumental in leading to the scalar-tensor theories. However, if we stick to standard GR, not a scalar-tensor variation, $`\kappa `$ is constant, (61) is valid only now, and takes this value now only if the universe evolves from very finely tuned earlier values. For example, in the present matter dominated era the equation of state leads to
$$\rho R^3=Mconst,$$
(62)
whereas in an earlier radiation dominated state
$$\rho R^4const.$$
(63)
An analysis of the time evolution of these quantities in standard general relativity under drastically different regimes show that an extremely small variation of the values of the $`\mathrm{\Omega }`$โs at early times would result in drastically different values now. But this is not the only conceptual problem. For example, there are questions of how the universe could have homogenized itself from random early data (the โhorizonโ problem), and others, ,
Guth pointed out that this myriad of difficulties could be at least partially resolved if the early stages of evolution were โinflationary,โ that is
$$R(t)=R(0)e^{Ht},$$
(64)
with constant $`H.`$ Such a model is consistent with (56) for $`\rho =ฯต=0,\mathrm{\Lambda }0.`$ Of course, this is not consistent with present data, so something other than a cosmological constant is needed. One way to achieve it is to introduce a new massless scalar field, the โinflaton,โ $`\varphi ,`$ with Lagrangian density,
$$=g^{\alpha \beta }\varphi _\alpha \varphi _\beta V(\varphi ).$$
(65)
This field contributes an effective mass density and pressure given by
$$\rho _\varphi =\dot{\varphi }^2/2+V,p_\varphi =\dot{\varphi }^2/2V.$$
(66)
The introduction of $`\varphi `$ and its potential, $`V`$, can be used to resolve at least some, but certainly not all, of the problems discussed above. In some models, this inflaton has a dilaton-like nature, in others it is reminiscent of the $`\varphi `$ in the old scalar-tensor theories. The problems of scalar-tensor field theories with solar-system sized observations require that $`\omega `$ be very large. However, this restriction need not diminish the significance of the inflaton field in earlier cosmological contexts.
Of course, as of the beginning of the 21st century, cosmological observations and theory have expanded well past these early inflationary models, but we will stop here, and remind the reader that universally coupled, thus gravitational, scalar fields are still active players in contemporary theoretical physics. So perhaps we can say that the scalar field is still alive and active, if not always well, in current gravity research.
|
warning/0506/astro-ph0506629.html
|
ar5iv
|
text
|
# The anatomy of a quadruply imaged gravitational lens system
## 1 Introduction
Strong gravitational lens systems provide a tool for measuring cosmological parameters. With the measured relative arrival time delays between the multiple images of the lensed source and a model of the lens potential, one can deduce a value of the Hubble constant. In addition, strong gravitational lens systems can be used to probe galaxy mass distributions, including dark matter, since the lens potential is directly related to the lens mass distribution. (Refsdal, 1964)
Several strong gravitational lenses with either two images (โdoublesโ) or four images (โquadsโ) have been observed. The ones with extended source distributions are of special interest since they provide additional constraints for the lens potential due to surface brightness conservation. The traditional approach to modelling the lens mass distribution is to postulate a parametric form for the lens distribution and minimize some chi-square to fit the data. The method is limited by the choice of the parameters; as the observational quality improves, the original parametric model generally becomes inadequate to fit the data (Williams & Saha (2000) consider a pixellated mass distribution which is non-parametric, but use only the nuclear image positions and not information from the extended source to constrain the distribution). Ideally, we want a method that employs the extended source information to obtain a non-parametric form of the lens potential whose accuracy is limited solely by the observational noise in the data. Koopmans (2005) has also taken this approach.
In section 2, we study in detail a quadruply imaged gravitational lens system with an extended source, B1608+656, showing the various criteria that the isophotes of the extended source must satisfy. In section 3, we examine a method of potential reconstruction proposed by Blandford et al. (2001) to correct the potential values pixel by pixel from a starting perturbed potential model.
## 2 Properties of a Quadruply Lensed System
We will focus on the quadruply imaged gravitational lens system B1608+656 in this section. Fig. 1 shows an image of the system taken by the Hubble Space Telescope through the F814W filter (Surpi & Blandford, 2003). The source is at a redshift of $`z_s=1.39`$ (Fassnacht et al., 1996) and its images are labelled by A, B, C, and D. The system has two lens galaxies, G1 (the primary lens) and G2 (the secondary lens), that are at a redshift of $`z_d=0.63`$ (Myers et al., 1995).<sup>1</sup><sup>1</sup>1The quoted redshift is that of G1. We assume that G2 is at the same redshift as G1 since G2 is too faint for its redshift to be measured. The galaxy G1 is about five times more massive than G2. B1608+656 is unique in that all three relative time delays between the four images are determined with accuracies of a few percent. The time delays relative to image B for images A, C, and D are $`31.5\pm 1.5`$ days, $`36.0\pm 1.5`$ days, and $`77.0\pm 1.5`$ days, respectively (Fassnacht et al., 1999, 2002).
Section 2.1 that follows is a review of the theory of gravitational lensing. Readers familiar with lensing may wish to proceed directly to Section 2.2, which analyses the Koopmans et al. (2003) model of B1608+656.
### 2.1 Gravitational lensing
Readers familiar with gravitational lensing may wish to skip this section. We follow Kochanek, Schneider & Wambsganss (2004) for the theory of gravitational lensing.
Let us denote the angular coordinates on the source and image planes by $`๐ท=(\beta _1,\beta _2)`$ and $`๐ฝ=(\theta _1,\theta _2)`$, respectively. The lens equation governing the deflection of light rays is
$$๐ท=๐ฝ๐ถ(๐ฝ),$$
(1)
where $`๐ถ(๐ฝ)`$ is the scaled deflection angle that is the gradient of a scalar function called the lens (or deflection) potential:
$$๐ถ(๐ฝ)=\mathbf{}\psi (๐ฝ).$$
(2)
In terms of the dimensionless surface mass density, denoted by $`\kappa (๐ฝ)`$, the lens potential is
$$\psi (๐ฝ)=\frac{1}{\pi }_\mathrm{}^2d^2\theta ^{}\kappa (๐ฝ^{\mathbf{}})\mathrm{ln}|๐ฝ๐ฝ^{\mathbf{}}|.$$
(3)
The time delay function relative to the case of no lensing is
$$T(๐ฝ,๐ท)=\frac{1}{c}\frac{D_dD_s}{D_{ds}}(1+z_d)\left[\frac{(๐ฝ๐ท)^2}{2}\psi (๐ฝ)\right],$$
(4)
where $`D_d`$, $`D_s`$, and $`D_{ds}`$ are, respectively, the angular diameter distance from us to the lens, from us to the source, and from the lens to the source.
The constant coefficient in equation (4) is proportional to the angular diameter distance and hence inversely proportional to the Hubble constant in a flat $`\mathrm{\Lambda }`$-CDM universe. Therefore, by measuring the relative time delays between the various images, we can in principle deduce the value of the Hubble constant if we know the source position ($`๐ท`$) and the lens potential ($`\psi (๐ฝ)`$).
To characterise the magnifications of images in gravitational lensing, a Hessian is used
$$\text{A}(๐ฝ)=\frac{๐ท}{๐ฝ}.$$
(5)
Using the lens equation (1), the above equation can be written as
$`\text{A}(๐ฝ)`$ $`=`$ $`\left(\begin{array}{cc}1\psi _{11}(๐ฝ)& \psi _{12}(๐ฝ)\\ \psi _{12}(๐ฝ)& 1\psi _{22}(๐ฝ)\end{array}\right),`$ (8)
where the subscript 1 (or 2) in $`\psi `$ indicates a derivative with respect to $`\theta _1`$ (or $`\theta _2`$). The magnification matrix is defined as $`๐=\text{A}^1`$, and the associated magnification factor is
$$\mu (๐ฝ)=\frac{1}{det\text{A}(๐ฝ)}.$$
(9)
According to equation (9), the positions $`๐ฝ`$ with $`det\text{A}(๐ฝ)=0`$ have divergent magnification; the loci of such points on the image plane define the critical curves. Using the lens equation (1), critical curves on the image plane are mapped to caustic curves (or simply caustics) on the source plane. The caustic curves separate regions of different image multiplicities.
### 2.2 Gravitational lens B1608+656
To investigate the anatomy of the quad B1608+656, we use the mass distribution model proposed by Koopmans et al. (2003). The parametric form of the dimensionless surface mass density for each of the two lens galaxies is a singular power law ellipsoid:
$$\kappa (\theta _{gal_1},\theta _{gal_2})=b\left[\theta _{gal_1}^2+\left(\frac{\theta _{gal_2}}{q_l}\right)^2\right]^{\frac{1\gamma ^{}}{2}},$$
(10)
where $`(\theta _{gal_1},\theta _{gal_2})`$ are coordinates relative to the galaxy centre and $`b`$, $`q_l`$, and $`\gamma ^{}`$ are parameters to fit the data. The origin of coordinate $`๐ฝ`$ is set at the position of image A. Each of the lens galaxies is centred at the coordinates $`(\theta _{l1},\theta _{l2})`$ and is rotated by a major-axis position angle $`\theta _{PA}`$ that is measured from north to east (top to left). There is an additional external shear centred on G1 whose contribution to the lensing potential, in polar coordinates relative to the shear centre ($`(r,\varphi )`$ with $`\theta _{sh_1}=r\mathrm{cos}(\varphi )`$ and $`\theta _{sh_2}=r\mathrm{sin}(\varphi )`$), is
$$\psi _{ext}(๐ฝ_{๐๐})=\frac{1}{2}\gamma _{ext}r^2\mathrm{cos}(2\varphi ),$$
(11)
where $`\gamma _{ext}`$ is a parameter characterising the shear strength. The rotation of the external shear is given by the position angle $`\theta _{ext}`$. We adopt the parameter values of the SPLE1+D(isotropic) model in Koopmans et al. (2003) and list them in Table 1.
#### 2.2.1 Critical and caustic curves
The critical curves on the image plane and the caustic curves on the source plane of the SPLE1+D(isotropic) model in Koopmans et al. (2003) are shown in Fig. 2 in the middle panel and the left panel, respectively. The locations of the lens galaxies are indicated by open triangles on the image plane. The marked source and image locations will be discussed in the next section. With the two elliptical lens galaxies, the large critical curve loop is a deformed version of an elliptical curve of one singular power law ellipsoid (equation (10)). The corresponding diamond shaped caustic curve, known as an astroid, is typical for elliptical mass distributions. An astroid is composed of four folds (branches of smooth curves) joining at four cusps. An individual power law ellipsoid has an astroid that is symmetrical with respect to the semi-major and semi-minor axis of the lens. With the two lens galaxies in the SPLE1+D(isotropic) model, we have an asymmetry in the astroid and two additional small triangular caustics, called the deltoids, that map into the small loops on the image plane.
#### 2.2.2 Image positions and time delay surface
It is instructive to see how the images move on the image plane as the source is displaced. Understanding such movements is important for analysing quads and for defining the limit curves in the next section. Fig. 2 shows the locations of the images, labelled by A, B, C, D, and E (middle panel), when the source is at the centre of the astroid caustic (left panel). Despite having five images, the system is called a quad because the central image is usually de-magnified and lies near the lens galaxies, making it nearly observationally invisible<sup>2</sup><sup>2</sup>2We refer the reader to Winn, Rusin & Wambsganss (2004) for candidates of central image detections in gravitational lens systems.. The arrival time delay contours in the right panel show that the image locations are at the time delay extrema or saddles, except for the extrema where the surface mass densities are non-smooth (Kochanek et al., 2004). At the centroids of G1 and G2 whose locations are given in Table 1, the time delay achieves local maxima, but there are no corresponding images because the surface mass densities are singular at the galaxy centroids in the model described by equation (10). Ignoring the central image (E, which is finitely de-magnified), the two images (C and D) inside the critical curve are time delay saddles, and the two images (A and B) outside the critical curve are time delay minima. This is true in general for quads.
Fig. 3 shows the image locations and the time delay contours as the source moves across a fold from within the caustic. As the source approaches a fold, two of the images (B and C for the upper fold of interest) that are separated by the critical curve come together. When the source is on the fold, the two images merge to become one at the corresponding point on the critical curve. Finally, when the source moves across the fold, the merged image disappears. The merging and disappearance of the two images can be explained using the lemniscate time delay contour (the saddle with two minima) in the right panels. When the source approaches a fold, the time delay saddle of the lemniscate joins with one of its two associated local minima; after the source crosses the fold, only one time delay minimum remains.
Fig. 4 shows the image locations and the arrival time delay contours as the source moves from within the astroid caustic across a cusp in a direction that is roughly along the semi-major axis of the lens distribution. As the source approaches the cusp, three of the images (A, B, and C in this case) come together. Two images (A and B) are outside and one image (C) is inside the critical curve. When the source is on the cusp, the three images become one on the critical curve. Finally, when the source moves across the cusp, one image remains outside the critical curve. (We label the remaining image by the one that comes alphabetically first among the three merging images.) The time delay contours in the right panels depict this behaviour: the time delay saddle of a lemniscate merges simultaneously with both of its two minima and leaves a single minimum in the end.
Fig. 5 is similar to Fig. 4 but with the source displacing toward a cusp that is roughly along the semi-minor axis of lens distribution. The three merging images now have one image (B) outside and two images (C and D) inside the critical curve (shown in middle panels). In terms of the time delay contours (right panels), this corresponds to the simultaneous merging of the saddle of the lemniscate with one of its minima and with the saddle of the enclosing limaรงon, leaving only the limaรงon saddle in the end.
#### 2.2.3 Inner and outer limits
The movements of the image locations shown in Figs. 2 to 5 allow us to define limit curves (Blandford & Narayan, 1986). Consider moving a hypothetical point source on the caustic curve. As the source traces around the folds of the caustic, the two non-merging images trace out the limit curves. For the astroid, the non-merging image inside the critical curve is on the inner limit and the image outside the critical curve is on the outer limit. For the deltoids, both non-merging images are outside the corresponding critical curves. The deltoids thus have only outer limits composed of two images and no inner limits. Fig. 6 is the plot of the limit curves for the SPLE1+D(isotropic) model in Koopmans et al. (2003). The inner limit and outer limit for the astroid are shown in green and orange, respectively. The outer limit for the deltoids are shown in cyan.
We focus only on the limit curves of the astroid since they are typical for elliptical lens mass distributions. Both the inner and the outer limits are tangent to the critical curve twice, corresponding to source placement at the cusps of the caustic. The limit curves mark the boundary of the region containing four images.
#### 2.2.4 Isophotal separatrices
An isophote is an intensity contour. We assume the source intensity distribution has a single maximum with nested, non-crossing contours. An isophotal separatrix on the image plane corresponds to a source intensity contour that is tangent to the caustic curve. The isophotes must cross at the critical curve and be tangent to the limit curves as we explain below.
Consider an extended elliptical source intensity distribution centred at $`(\beta _{s1},\beta _{s2})=(0.088,1.069)`$ with an axis ratio of 0.634 and a semi-major axis position angle of 22.1 degrees<sup>3</sup><sup>3</sup>3This source model differs from the Koopmans et al. (2003) source model in the position angle, but the difference is irrelevant for the purpose of describing isophotal separatrices.. The left panel in Fig. 7 shows four coloured intensity contours of the extended source. The two intermediate isophotes are very close together (light blue and dark blue). The right panel in Fig. 7 shows the mapped isophotes (same colours) with the critical curves (black) and limit curves (red). Each coloured set of isophotes must intersect at the critical curve and be tangent to the inner and outer limit. This is shown most clearly by the purple isophotes that consist of a lemniscate (separatrix) with two elliptical satellite isophotes on the image plane. The lemniscate isophote must cross at the critical curve, and the two satellite isophotes must each be tangent to either the inner or the outer limit.
To explain the crossing and tangency conditions, let us consider the purple isophotes in detail. The crossing point of the lemniscate on the critical curve corresponds to the tangency point of the source isophote to the astroid caustic curve. Recall from section 2.2.2 that two of the four images of a hypothetical point source merge on the critical curve as the source moves across the fold from within. Therefore, a segment of the source isophote to either side of the caustic tangency point will map to two segments on the image plane, one inside and one outside the critical curve, that connect at the critical curve. The entire source isophote that is within the caustic will thus correspond to a lemniscate crossing the critical curve on the image plane with one lobe inside and one lobe outside the critical curve. The tangencies of the image isophotes to the limit curves can be understood based on the definition of limit curves, which are the inner and outer boundaries of the four-image region that are marked by the two non-merging images as a hypothetical source traces around the caustic. The two satellite isophotes correspond to image isophotes traced by the two non-merging images that must touch the inner and outer limits when the source isophote is tangent to the caustic. Since the inner and outer limits are the four-image boundaries, the touchings of the satellite isophotes to the limit curves become tangencies. Similar reasoning applies to the crossings and tangencies of the other three sets of isophotes.
The crossing of the isophotes at the critical curves and the tangency of the isophotes to the limit curves provide qualitative tests on how good a lens model is.
#### 2.2.5 Observational Data
We use the result of section 2.2.4 to qualitatively test the SPLE1+D(isotropic) model in Koopmans et al. (2003) by superimposing the critical and limit curves of the model on the intensity contours of the observational data. Fig. 8 shows the isophotal separatrices (in black in various line styles) of the deconvolved residual HST/F814W image of B1608+656 (Koopmans et al., 2003) with the critical curves (red) and limit curves (green, orange, cyan). We check the crossing and tangency conditions for each of the four sets of isophotal separatrices, using Fig. 7 as a guide for the approximate crossing and tangency locations. For the dashed isophotes, the conditions for the crossing of the separatrix at the critical curve and the tangency to the limit curves are violated. For the solid isophotes, the crossing at $`(\theta _1,\theta _2)(0.8,1.1)`$ is not at the critical curve, but the tangency requirements at $`(0.9,1.4)`$ and $`(0.4,0.3)`$ are satisfied within the noise. For the dotted isophotes, the crossing at $`(0.7,1.9)`$ is at the critical curve within the noise, but the isophotes near $`(0.5,0.9)`$ and $`(1.1,0.2)`$ are not tangent to the limit curves. Lastly, for the long-dashed isophotes, the crossing at $`(1.3,0.6)`$ is on the critical curve, and the isophotes near $`(0.5,0.6)`$ and $`(0.3,2.4)`$ are tangent to the limit curves, within the noise. Therefore, the SPLE1+D(isotropic) model proposed by Koopmans et al. (2003) satisfies the crossing and tangency conditions stated in section 2.2.4 for some, but not all, of the isophotal separatrices. As a result, the model proposed by Koopmans et al. (2003) must not represent the true lens potential of the system, especially in the regions where the crossings and tangencies fail. Recall that we need an accurate model of the lens potential to calculate the Hubble constant. In the next section, we examine a method of potential correction.
## 3 Potential Reconstruction
### 3.1 Theory of potential reconstruction
The method of potential reconstruction was first suggested by Blandford et al. (2001). Following the notation in section 2.1, let $`I(๐ฝ)`$ be the observed image intensity of a gravitational lens system with an extended source. For a given potential model, $`\psi (๐ฝ)`$, one can obtain the best-fitting source intensity distribution (Warren & Dye, 2003). Let $`I(๐ท)`$ be the source intensity translated to the image plane via the potential model, $`\psi (๐ฝ)`$. We define the intensity deficit on the image plane by
$$\delta I(๐ฝ)=I(๐ฝ)I(๐ท).$$
(12)
The intensity deficit is zero everywhere with the true lens potential distribution.
Consider a lens potential model that is perturbed from the true potential, $`\psi _0(๐ฝ)`$, by $`\delta \psi (๐ฝ)`$:
$$\psi (๐ฝ)=\psi _0(๐ฝ)+\delta \psi (๐ฝ).$$
(13)
We can correct the potential model perturbatively by solving for the perturbation $`\delta \psi (๐ฝ)`$. For a given image (fixed $`๐ฝ`$ and $`I(๐ฝ)`$), we can relate a change in position on the source plane, $`\delta ๐ท`$, to the potential perturbation using the lens equation (1):
$$\delta ๐ท=\frac{\delta \psi (๐ฝ)}{๐ฝ}.$$
(14)
Expanding $`I(๐ท)`$ to first order in $`\delta ๐ท`$ and using equation (14) in equation (12), we obtain
$$\delta I(๐ฝ)=\frac{I(๐ท)}{๐ท}\mathbf{}\delta ๐ท=\frac{I(๐ท)}{๐ท}\mathbf{}\frac{\delta \psi (๐ฝ)}{๐ฝ}.$$
(15)
The source intensity gradient $`\frac{I(๐ท)}{๐ท}`$ implicitly depends on the potential model $`\psi (๐ฝ)`$ since the source position $`๐ท`$ (where the gradient is evaluated) is related to $`\psi (๐ฝ)`$ via the lens equation (1). To first order, using the perturbed model $`\psi (๐ฝ)`$ is equivalent to using the true model $`\psi _0(๐ฝ)`$ in the evaluation of the source intensity gradient $`\frac{I(๐ท)}{๐ท}`$.
We can solve equation (15) for the potential correction, $`\delta \psi (๐ฝ)`$, provided that we start at a potential model that is close to the true potential. (We quantify what โcloseโ means in the next section.) One method to solve for the potential correction is to integrate along the characteristics of the partial differential equation (15). The solution is
$$\delta \psi (๐ฝ)=\delta \psi (๐ฝ_๐จ)+_{๐ฝ_๐จ}^๐ฝ\frac{d\theta _s\delta I(๐ฝ)}{\left|\frac{I(๐ท)}{๐ท}\right|},$$
(16)
where
$$d\theta _s=\left(d\theta _1^2+d\theta _2^2\right)^{1/2},$$
(17)
$$\left|\frac{I(๐ท)}{๐ท}\right|=\sqrt{\left(\frac{I(๐ท)}{\beta _1}\right)^2+\left(\frac{I(๐ท)}{\beta _2}\right)^2},$$
(18)
and $`๐ฝ_๐จ`$ is an arbitrary reference point that is conveniently chosen to be at the location of one of the images, say A. (The reference point is arbitrary because the potential is determined up to a constant.) The characteristic curves, on which we must integrate to obtain the potential correction, are given by curves that satisfy
$$\frac{d\theta _1}{d\theta _2}=\frac{I/\beta _1}{I/\beta _2}.$$
(19)
Each point on a characteristic curve thus follows the source intensity gradient (evaluated at the corresponding source location given by the lens equation (1)) that is directly translated to the image plane without distortions via the magnification matrix. Due to the direct translation of the source intensity gradient, the characteristic curves differ from the curves on the image plane that map to the source intensity gradient curves. The structure of the characteristic curves allows us to determine whether the potential solution given by equation (16) is unique. This is demonstrated in the next section.
We can repeat the process for a perturbative and iterative potential reconstruction method. We expect the potential to be closer to the true potential after each iteration, which is indicated by a decrease in the magnitude of the intensity deficit.
The potential reconstruction method is non-parametric. We can pixellate the potential distribution to match the observed image pixellation, and the potential correction at each pixel is given by equation (16).
To summarise, the four steps for the method are: (i) start with a potential model close to the true potential, (ii) calculate the intensity deficit (equation (12)) of each pixel, (iii) calculate the potential correction of each pixel (equation (16)) by integrating along the characteristics (equation (19)), (iv) obtain the corrected potential and repeat the process (steps (ii) to (iv)) until the intensity deficit approaches zero. In the next section, we examine a quadruply imaged toy model to test the method of potential reconstruction.
### 3.2 Example Toy Model
To demonstrate the method of potential reconstruction discussed in the previous section, we consider a toy model with a simple lens potential that produces a quad like B1608+656.
The toy system has a non-singular isothermal ellipsoid lens whose potential takes the form:
$$\psi (\theta _1,\theta _2)=(\theta _1^2+2\theta _2^2+0.1)^{1/2}$$
(20)
We take the perturbed potential to be the original potential that is rotated clockwise by 1.1 degrees. The source intensity distribution has elliptical contours with axis ratio of 0.634 and position angle of 147.2 degrees. The source nucleus is located at $`(\beta _{1s},\beta _{2s})=(0.1,0.05)`$ and has an intensity peak of 100, in arbitrary units. We assume the data is perfect with no noise, but we discretize the image plane region \[-2,2\]$`\times `$\[-2,2\] into a 201$`\times `$201 grid in order to correct for the perturbation of every pixel. In Fig. 9, the left panel shows the caustic curves (dashed) of the original potential and the source intensity contours (dotted), and the right panel shows the corresponding critical curves (dashed) and image intensity contours (dotted). Analogous to B1608+656, there is an astroid caustic in the left panel. The additional elliptical caustic curve is due to the non-singular nature of the lens potential. Different regions separated by the caustic curves have different image multiplicities. In the enclosed region intersected by the astroid and elliptical caustic curves, a source has five images on the image plane. In the region within the caustic curves excluding the intersection, a source has three images. In the region outside the caustic curves, a source has one image. The astroid caustic is mapped to the outer critical curve and the elliptical caustic is mapped to the inner critical curve. As for B1608+656, we focus on the astroid caustic and the outer critical curve. Among the isophotes in the right panel, the four isophotal separatrices that are shown match to the four isophotes tangent to the astroid caustic in the left panel. The separatrices intersect at the outer critical curve, as required (section 2.2.4). Fig. 10 shows the arrival time delay contour of the source nucleus of the toy model. The quad has similar time delay extrema (two saddles within the critical curve and two minima outside the critical curve) to the SPLE1+D(isotropic) model of B1608+656.
We simplify the potential correction method by using the original source intensity distribution and the characteristic fields of the original potential (instead of reconstructing from the perturbed potential). In reality, we would have to use the reconstructed source (Warren & Dye, 2003) and the characteristic fields of the perturbed potential. This would involve simultaneously determining the source and lens potential distributions and investigating the partial degeneracy between them, which are beyond the scope of this paper. We use the simplifying assumptions on the source intensity and characteristic curves as the first step to testing the method of potential reconstruction via integration along characteristics. Only if the method works robustly in this simplified regime is the consideration of the more general problem relevant.
Fig. 11 shows the characteristic field given by equation (19). The field has โattractorsโ (where field lines come together) and โrepellorsโ (where field lines curve away) at the image locations of the source nucleus. Using equation (4) and noting that the Jacobian matrix of $`T(๐ฝ,๐ท)`$ with respect to $`๐ฝ`$ is equivalent to A in equation (5) up to a constant coefficient, one can show that the attractors (or repellors) are associated to time delay minima/maxima (or saddles) for a source distribution that has non-crossing intensity contours. A comparison between Fig. 10 and Fig. 11 confirms this fact.
We need to follow along the characteristics to correct for the potential perturbation given by equation (16). In Fig. 11, almost all of the characteristic curves end at one of the three attractors; but there are special characteristic curves that connect the attractors and repellors. These four connecting characteristics between the four images (excluding the central image), shown in the right panel of Fig. 9 in solid lines, allow us to fix the potential offsets between the images and hence uniquely determine the potential up to a constant. The left panel of Fig. 9 shows the mapping of these connecting characteristics onto the source plane (solid lines). As one may expect, the mapping of each of the connecting characteristics between an attractor and a repellor is a loop on the source plane that is tangent to the astroid caustic curve due to the connecting characteristics intersecting the outer critical curve.
In addition to the characteristic curves, the intensity deficit is required for potential correction in equation (16). To get the intensity deficit defined in equation (12) for the pixels on the image plane, first we use the perturbed potential model, the lens equation (1), and the original source intensity distribution to get $`I(๐ท)`$, then we subtract it from $`I(๐ฝ)`$ obtained from the original potential. Fig. 12 shows the initial intensity deficit and the initial potential perturbation ($`\delta \psi (๐ฝ)`$ in equation (13)) before the perturbative and iterative potential correction, in the top left and bottom left panels, respectively. We use plots of $`\delta \psi (๐ฝ)`$ to check that the perturbation approaches zero after corrections.
In each potential reconstruction iteration, we use the current perturbed potential model to obtain the intensity deficit ($`\delta I(๐ฝ)`$) and the source intensity gradient ($`\left|\frac{I(๐ท)}{๐ท}\right|`$) at every pixel on the image plane; we then use equation (16) to correct the perturbed potential by integrating along the characteristic curves of the original potential model. Two iterations are performed and the resulting intensity deficit and potential perturbation after each iteration are shown in Fig. 12. The middle and right panels show the intensity deficit (potential perturbation) in the top (bottom) after 1 and 2 iterations, respectively. The middle and right panels are plotted on the same scales as that in the left panels. Comparing the right panels to the left panels, the intensity deficit and potential perturbation converge to zero after two iterations (apart from numerical error), signifying that the method of potential reconstruction along characteristics works in theory with perfect data.
A possible limitation to this method is that the intensity deficit needs to be zero at the image locations; otherwise, according to equation (16), the integrand diverges at the image locations, which are the end points of integration. For the example above, we are saved from this divergence by discretizing the image plane and thus only reaching the image points within some tolerance, but never ending at the image (divergent) points. The potential correction is most significant near the image points for any non-zero intensity deficit in the region. Therefore, integrating along the characteristics may place limitations on the magnitude of potential perturbation that we can correct, which we discuss in the next section.
This method of potential reconstruction works only for small potential perturbations like the example we considered where the perturbation magnitude is on average (over the image grid) 0.5 per cent of the original potential. By increasing the rotation of the original potential distribution to get the perturbed potential (that is, increasing the perturbation), we require more iterations for convergence, as expected. When the rotation the original potential gets to $`4.5`$ degrees, which corresponds to an average potential perturbation magnitude of $`1.5`$ per cent, the method ceases to converge. Therefore, the method of potential correction by integrating equation (16) along characteristics works in theory with perfect data with a small ($``$ 1 per cent) potential model error. Unless a better algorithm is found for treating larger potential perturbation and real data with noise, the method proposed by Blandford et al. (2001) will not in practice be useful.
The example toy model considered provides a practical insight into the theory of potential reconstruction. In reality, we do not have useful data everywhere due to the presence of noise; for an extended source, we can observe emission in an Einstein ring connecting the four images. Based on the analysis of this section, the Einstein ring must be large enough to enclose the connecting characteristics in order to obtain proper potential offsets between the images. This condition must hold for any potential reconstruction algorithm based on equation (15).
## 4 Conclusions and further work
We have considered the gravitational lens system B1608+656 to investigate the properties of quads. We have defined limit curves as the loci of non-merging images as the source traces the caustic curve. For the typical astroid caustic curve of a quad, there are inner and outer limits (relative to the critical curve) that are each tangent to the critical curve twice. We have shown that isophotes that are tangent to the astroid caustic curve on the source plane map to isophotal separatrices on the image plane. These separatrices must intersect on the critical curve and their associated satellite isophotes must be tangent to the limit curves. We have shown that the current model proposed by Koopmans et al. (2003) for B1608+656 does not satisfy these qualitative constraints for some of the isophotal separatrices.
We have investigated a perturbative and iterative method of potential reconstruction proposed by Blandford et al. (2001). The method requires solving a partial differential equation for the potential correction, which we have done by integrating along the characteristic curves. We have used a toy model that is a quad like B1608+656 to test the method, assuming perfect data. For small perturbations whose magnitudes are on average $``$ 1 per cent of the original potential, the method has worked and we have had the perturbed potential converging to the true potential. However, the method has failed to converge when the perturbation magnitude increases to around 1.5 per cent of the original potential. This may be due to the non-zero intensity deficit near the image locations which restricts the integration along characteristics. We hope to use the knowledge we have acquired about the anatomy of the quads and the characteristic fields of the potential correction equation to find a more robust method of potential correction that can be applied to real data with noise.
## Acknowledgement
We thank C. Fassnacht, L. Koopmans, P. Marshall, and T. Treu for useful discussions and encouragement. We thank the referee for comments that improved the presentation of the paper. SS thanks I. Mandel for commenting on the manuscript. This work was supported by the NSF under award AST04-44059 and in part by the U.S. Department of Energy under contract number DE-AC02-76SF00515. SS acknowledges the support of the NSERC (Canada) through the Postgraduate Scholarship.
|
warning/0506/hep-ph0506316.html
|
ar5iv
|
text
|
# ๐ต_๐ โ๐โข(๐)โข๐พ^โ with perturbative QCD approach
## 1 Introduction
Exclusive nonleptonic B decays have provided a fertile field to investigate the CP violation and search for new physics. The hadronic matrix elements of the effective operators play a key role in the study of B meson decays, but it is difficult to calculate them precisely due to the long distance QCD dynamics. The factorization approach (FA) based on the color transparency mechanism has been applied to many decay modes, and it works well in many channels. But it suffers from some problems such as infrared-cutoff and scale dependence. To solve these problems and make more accurate predictions, the perturbative QCD approach (PQCD) , the QCD improved factorization (QCDF) as well as the Soft-collinear effective theory (SCET) have been developed in the recent years.
PQCD is based on $`k_T`$ factorization theorem . The decay amplitude is factorized into the convolution of the mesonsโ light-cone wave functions (see $`AppendixA`$), the hard scattering kernels and the Wilson coefficients, which stand for the soft, hard and harder dynamics respectively. The transverse momentum is introduced so that the endpoint singularity which will break the collinear factorization is regulated and the large double logarithm term appears after the integration on the transverse momentum, which is then resummed into the Sudakov form factor. The formalism can be written as:
$``$ $`{\displaystyle }dx_1dx_2dx_3b_1db_1b_2db_2b_3db_3Tr[C(t)\mathrm{\Phi }_B(x_1,b_1)\mathrm{\Phi }_K^{}(x_2,b_2)\mathrm{\Phi }_\rho (x_3,b_3)`$ (1)
$`H(x_i,b_i,t)S_t(x_i)e^{S(t)}],`$
where the $`b_i`$ is the conjugate space coordinate of the transverse momentum, it denotes the transverse interval of the meson. $`t`$ is the energy scale in hard function $`H`$. The jet function $`S_t(x_i)`$ comes from the summation of the double logarithms $`\mathrm{ln}^2x_i`$ near the endpoint, called threshold resummation . The factorization theorem guarantees the infrared safety and the gauge invariance of the hard kernel and has been proved to all order of $`\alpha _s`$ .
Many hadronic two body $`B`$ decays have been studied in PQCD approach . Most predictions are consistent with the current experiments. The $`B_s`$ decays are important to extract CKM phase angles and study the CP violation. As $`B_s`$ meson is not in the energy scale of the high luminosity B factories SLAC and KEK, it is more difficult to be produced and measured now. We can study the $`B_s`$ decays more precisely in the very near future with the increase of luminosity at TEVATRON and the upcoming Large Hadron Collider (LHC).
$`B_s`$ meson is different from B meson due to the heavier strange quark (compare to $`u`$, $`d`$ quark) which induces the SU(3) symmetry-breaking effect. This effect is considered to be small and the distribution amplitude of $`B_s`$ meson(given in the following formula) should be similar to that of the B meson,
$`\varphi _{B_s}(x)=N_{B_s}x^2(1x)^2\mathrm{exp}\left[{\displaystyle \frac{1}{2}}({\displaystyle \frac{xM_{B_s}}{\omega _{B_s}}})^2{\displaystyle \frac{\omega _{B_s}^2b^2}{2}}\right].`$ (2)
The upper limit of the $`B_s\pi K`$ branching ratio is $`7.5\times 10^6`$ , which constrain the parameter $`\omega _{B_s}`$ to a lower limit of about 0.5 . Moreover, in order to fit the branching ratio measured in the $`B_s\varphi \varphi `$ decay , we constrain $`\omega _{B_s}`$ to about 0.55 , then we can see that the SU(3) symmetry-breaking is not negligible. Here we integrate out the variable $`b`$ and show the distribution amplitude of $`B`$ and $`B_s`$ meson in Fig.1. We can see that the peak point of the curve of $`B_s`$ mesonโs distribution amplitude prefers a larger $`x`$ ($`x`$ denotes the momentum fraction carried by the light quark) region comparing to the B meson. This is consistent with the fact that the s quark much heavier than the d (u) quark, should carry more momentum. Later in this paper, we will see that the branching ratios of $`B_s`$ decays are very sensitive to this parameter. If measured by experiments, radiative leptonic decays of $`B_s`$ meson can provide information of this parameter .
In this paper, we study $`B_s\rho (\omega )K^{}`$ decays in the PQCD approach. Hopefully the branching ratio is not too small and can be detected by the TEVATRON or LHCb experiments, then it may allow us to determine the $`B_s`$ distribution amplitude and SU(3) breaking effects with much more precision. Moreover, we can also constrain $`\alpha `$ with fewer pollution from this channel.
## 2 Calculation and Numerical analysis
We use the effective Hamiltonian for the process $`B_s\rho (\omega )K^{}`$ given by
$`_{eff}={\displaystyle \frac{G_F}{\sqrt{2}}}\left\{V_u\left[C_1(\mu )O_1(\mu )+C_2(\mu )O_2(\mu )\right]V_t{\displaystyle \underset{i=3}{\overset{10}{}}}C_i(\mu )O_i^{(q)}(\mu )\right\},`$ (3)
where $`V_u=V_{ud}^{}V_{ub}`$, $`V_t=V_{td}^{}V_{tb}`$, $`C_i(\mu )`$ are the Wilson coefficients, and the operators are
$`O_1=(\overline{d}_iq_j)_{VA}(\overline{q}_jb_i)_{VA},O_2=(\overline{d}_iq_i)_{VA}(\overline{q}_jb_j)_{VA},`$
$`O_3=(\overline{d}_ib_i)_{VA}{\displaystyle \underset{q}{}}(\overline{q}_jq_j)_{VA},O_4=(\overline{d}_ib_j)_{VA}{\displaystyle \underset{q}{}}(\overline{q}_jq_i)_{VA},`$
$`O_5=(\overline{d}_ib_i)_{VA}{\displaystyle \underset{q}{}}(\overline{q}_jq_j)_{V+A},O_6=(\overline{d}_ib_j)_{VA}{\displaystyle \underset{q}{}}(\overline{q}_jq_i)_{VA},`$
$`O_7={\displaystyle \frac{3}{2}}(\overline{d}_ib_i)_{VA}{\displaystyle \underset{q}{}}e_q(\overline{q}_jq_j)_{V+A},O_8={\displaystyle \frac{3}{2}}(\overline{d}_ib_j)_{VA}{\displaystyle \underset{q}{}}e_q(\overline{q}_jq_i)_{V+A},`$
$`O_9={\displaystyle \frac{3}{2}}(\overline{d}_ib_i)_{VA}{\displaystyle \underset{q}{}}e_q(\overline{q}_jq_j)_{VA},O_{10}={\displaystyle \frac{3}{2}}(\overline{d}_ib_j)_{VA}{\displaystyle \underset{q}{}}e_q(\overline{q}_jq_i)_{VA}.`$ (4)
Here i and j stand for $`SU(3)`$ color indices.
The decay width for these channels is :
$`\mathrm{\Gamma }={\displaystyle \frac{G_F^2|๐ฉ|}{16\pi M_B^2}}{\displaystyle \underset{\sigma =L,T}{}}^\sigma ^\sigma `$ (5)
where $`๐ฉ`$ is the 3-momentum of the final state mesons, $`|๐ฉ|=\frac{M_B}{2}(1r_K^{}^2r_{\rho (\omega )}^2)`$, and $`r_{K^{}(\rho ,\omega )}=m_{K^{}(\rho ,\omega )}/m_{B_s}`$. $`^\sigma `$ is the decay amplitude , which will be calculated later in PQCD approach. The subscript $`\sigma `$ denotes the helicity states of the two vector mesons with the longitudinal (transverse) components L(T). According to Lorentz structure analysis, the amplitude can be decomposed into:
$`^\sigma =M_{B_s}^2_L+M_{B_s}^2_Nฯต_2^{}(\sigma =T)ฯต_3^{}(\sigma =T)+i_Tฯต_{\mu \nu \rho \sigma }ฯต_2^\mu ฯต_3^\nu P_2^\rho P_3^\sigma .`$ (6)
We can define the longitudinal $`H_0`$, transverse $`H_\pm `$ helicity amplitudes
$$H_0=M_{B_s}^2_L,H_\pm =M_{B_s}^2_NM_K^{}^2\sqrt{r^21}_T,$$
(7)
where $`r^{}=\frac{P_2P_3}{M_K^{}M_{\rho (\omega )}}`$. They satisfy the relation
$$\underset{\sigma =L,R}{}^\sigma ^\sigma =|H_0|^2+|H_+|^2+|H_{}|^2.$$
(8)
The leading order diagrams in PQCD approach are shown in Fig.2. The amplitudes for $`B_s\rho K^0`$ and $`\overline{B}_s\rho \overline{K}^0`$are written as
$`_H`$ $`=`$ $`V_u^{}T_HV_t^{}P_H,`$ (9)
$`\overline{}_H`$ $`=`$ $`V_uT_HV_tP_H,`$ (10)
respectively, where the subscript $`H`$ denote different helicity amplitudes $`L,N`$ and $`T`$, and $`T_H`$ and $`P_H`$ are the amplitudes from tree and penguin diagrams respectively. The detailed formulae of $`T_H`$ and $`P_H`$ are similar to those in $`BK^{}K^{}`$ and $`B\varphi K^{}`$ , so we will not show them here.
The parameters used in our calculations are: the Fermi coupling constant $`G_F=1.16639\times 10^5GeV^2`$, the meson masses $`M_{B_s}=5.37GeV,M_K^{}=0.89GeV,M_{\rho (\omega )}=0.77GeV`$ , the decay constants $`f_K^{}=0.217GeV,f_K^{}^T=0.16GeV,f_\rho =0.205GeV,f_\rho ^T=0.155GeV,f_\omega =0.195GeV,f_\omega ^T=0.14GeV`$ , the central value of the CKM matrix elements $`\alpha =95^{}`$, $`|V_{td}|=0.0075,|V_{tb}|=0.9992,|V_{ud}|=0.9745`$ , $`|V_{ub}|=0.0047`$ and the $`B_s`$ meson lifetime $`\tau _{B_s}=1.461ps`$ .
As we mentioned before, the decay $`B_s\rho (\omega )K^{}`$ can be used to determine the $`B_s`$ meson wave function parameter $`\omega _{B_s}`$, or $`\omega _{B_s}`$ can influence our predictions of $`B_s\rho (\omega )K^{}`$ decay. So we show the results in Table 1 according to 3 different values of $`\omega _{B_s}`$. From the table we can easily find out the averaged branching ratio for $`B_s(\overline{B}_s)\rho ^\pm K^{}`$ is much larger than the other two, for $`B_s(\overline{B}_s)\rho ^\pm K^{}`$ involve large Wilson coefficient $`C(2)+C(1)/3`$ for the factorizable part while the other two ($`B_s(\overline{B}_s)\rho ^0\overline{K}^0(K^0)`$ and $`B_s(\overline{B}_s)\omega \overline{K}^0(K^0)`$) involve a much smaller Wilson coefficient $`C(1)+C(2)/3`$ (color-suppressed) for the factorizable part of the emission diagram. As a result the first one is tree dominated and has a large branching ratio and small direct CP asymmetry. While referring to the other two, the contributions from penguin and tree diagrams are at the same order ($`Z_H0.51.5`$), hence we can expect a large direct CP asymmetry from eqs.(13).
The polarization fraction difference of these channels are also due to that the main contribution of each channel comes from different topology. $`B_s(\overline{B}_s)\rho ^\pm K^{}`$ is tree dominated. The main contribution comes from the factorizable part of the emission diagram, where transverse polarization amplitude is suppressed by a factor $`r_\rho ^2(0.77/5.37)^2`$ (see formulas in ), so the longitudinal polarization dominates and contributes more than $`90\%`$ of the total branching ratio. But in $`B_s(\overline{B}_s)\rho ^0\overline{K}^0(K^0)`$ and $`B_s(\overline{B}_s)\omega \overline{K}^0(K^0)`$ decays, tree emission (factorizable) diagram contribution is suppressed due to the cancellation of Wilson coefficients $`C_1+C_2/3`$. The left dominant contribution is the non-factorizable diagrams of tree operators and penguin diagrams. both of these contributions equally contribute to longitudinal and transverse polarizations. The transverse polarization is not suppressed in those cases, therefore numerically we get a small longitudinal fraction of about 0.4. This similar situation is also found in $`B\rho \rho (\omega )`$ decays , which are related by SU(3) symmetry to our $`B_s\rho K^{}`$ decays.
To extract the CP violation parameters and dependence on CKM phase angle $`\alpha `$ of these decays, we rewrite the helicity amplitudes in (9,10) as the functions of $`\alpha `$:
$`_H^+`$ $`=`$ $`V_u^{}T_HV_t^{}P_H`$ (11)
$`=`$ $`V_u^{}T_H(1+Z_He^{i(\alpha +\delta _H)})`$
$`_H^{}`$ $`=`$ $`V_uT_HV_tP_H`$ (12)
$`=`$ $`V_uT_H(1+Z_He^{i(\alpha +\delta _H)})`$
where $`Z_H=|V_t^{}/V_u^{}||P_H/T_H|`$, and $`\delta `$ is the relative strong phase between tree ($`T`$) and penguin ($`P`$) diagrams. Here in PQCD approach, the strong phase comes from the nonfactorizable diagrams and annihilation diagrams. This is different from Beneke-Buchalla-Neubert-Sachrajda approach. In that approach, annihilation diagrams are not taken into account, strong phases mainly come from the so-called Bander-Silverman-Soni mechanism . As shown in , these effects are in fact next-to-leading-order ($`\alpha _s`$ suppressed) elements and can be neglected in PQCD approach. We give the averaged branching ratios of $`B_s(\overline{B}_s)\rho ^0K^0(\overline{K}^0)`$ and $`B_s(\overline{B}_s)\omega K^0(\overline{K}^0)`$ as a function of $`\alpha `$ in Fig.3, and the averaged branching ratios of $`B^0(\overline{B}^0)\rho ^\pm K^{}`$ in Fig.4.
Using Eqs.(11,12), the direct CP violating parameter is
$`A_{CP}^{dir}`$ $`=`$ $`{\displaystyle \frac{|M|^2|\overline{M}|^2}{|M|^2+|\overline{M}|^2}}`$ (13)
$`=`$ $`{\displaystyle \frac{2sin\alpha \left(T_L^2sin\delta _L+2T_N^2sin\delta _N+2T_T^2sin\delta _T\right)}{T_L^2(1+Z_L^2+2Z_Lcos\alpha cos\delta _L)+2_{i=N,T}T_i^2\left(1+Z_i^2+2Z_icos\alpha cos\delta _i\right)}}.`$
Notice the CP asymmetry for these channels are sensitive to CKM angle $`\alpha `$, we show the direct CP asymmetry as a function of $`\alpha `$ in Fig.5. It is easy to see that the $`B_s\rho ^0K^0`$ and $`\omega K^0`$ have large direct CP asymmetries up to 50%, with a relative minus sign. On the other hand, the $`B_s\rho ^\pm K^{}`$ decay has small direct CP asymmetry due to only one large tree contribution in this decay. The uncertainty shown at this table is only from the $`B_s`$ meson wave function parameter dependence. In fact, since CP asymmetry is sensitive to many parameters, the line should be more broadened by uncertainties. The mixing induced CP asymmetry is complicated and requires angular distribution study, similar study may be found in .
At last, if we compare our predictions with those of naive factorization
$`BR(B_s(\overline{B}_s)\rho ^0\overline{K}^0(K^0))=5.5\times 10^7,`$ (14)
$`BR(B_s(\overline{B}_s)\omega \overline{K}^0(K^0))=6.0\times 10^7,`$ (15)
$`BR(B_s(\overline{B}_s)\rho ^\pm K^{})=1.7\times 10^5,`$ (16)
and ones of QCDF
$`BR(B_s(\overline{B}_s)\rho ^0\overline{K}^0(K^0))=5.3\times 10^7,`$ (17)
$`BR(B_s(\overline{B}_s)\omega \overline{K}^0(K^0))=3.1\times 10^7,`$ (18)
$`BR(B_s(\overline{B}_s)\rho ^\pm K^{})=1.8\times 10^5.`$ (19)
We can see that they are consistent. It should be noticed that the branching ratios in FA and QCDF strongly depend on form factors. While in PQCD, the branching ratios and form factors depend on wave functions, especially the $`B_s`$ meson wave function. Nowadays, very few $`B_s`$ meson decays have been measured, so we can only give rough constraints on the parameters from other channel and permit large errors. More experimental data can help to constrain the form factors and wave functions, then we can give more precise predictions and the different methods can be tested by the experiments. Although similar results are got by different methods for branching ratios, the polarization fractions are quite different. The QCDF and naive factorization give only several percent transverse polarization for all three decay modes , while Table 1 shows large transverse contribution for $`B_s\rho ^0(\omega )K^0`$ decays in our PQCD approach. The direct CP asymmetry are not given in ref., but they probably also differ from PQCD approach as it happened in $`B\pi \pi `$ and $`K\pi `$ case .
The numerical results shown here are only leading order ones. For the tree dominated channel $`B_s\rho ^\pm K^{}`$, the leading order diagrams should give the main contribution. But for the other two decays, with a branching ratio as small as $`10^7`$, the next-to-leading order and power suppressed contributions should not be negligible, the results may suffer from large corrections when the next to leading order corrections are included .
Current experiments only give the upper limit for the decay
$`BR(B_s\rho ^0\overline{K}^0)<7.6\times 10^4.`$ (20)
More data are needed to test our calculations.
## 3 Summary
In this paper we calculate the branching ratios, polarization fraction and CP asymmetries of $`B_s\rho (\omega )K^{}`$ modes using PQCD theorem in SM. We perform all leading order diagrams to next to leading twist wave functions. We also study the dependence of their averaged branching ratios and the CP asymmetry on the CKM angle $`\alpha `$. At last we compare our predictions with values from other approaches.
## Acknowledgments
This work is partly supported by the National Science Foundation of China under Grant (No.90103013, 10475085 and 10135060). We thank G.-L. Song for reading our manuscript and giving us many helpful suggestions. We also thank J-F Cheng, H-n Li, Y. Li, and X-Q Yu for helpful discussions. We thank the Institute for Nuclear Theory at the University of Washington for its hospitality and the Department of Energy for partial support during the completion of this work.
## Appendix A wave function
For longitudinal polarized $`K^{}`$ meson, the wave function is written as
$`{\displaystyle \frac{1}{\sqrt{2N_c}}}[M_K^{}\overline{)}ฯต_{2L}\varphi _K^{}(x)+\overline{)}ฯต_{2L}\overline{)}P_2\varphi _K^{}^t(x)+M_\varphi I\varphi _K^{}^s(x)],`$ (21)
and the wave function for transverse polarized $`K^{}`$ meson reads
$`{\displaystyle \frac{1}{\sqrt{2N_c}}}[M_K^{}\overline{)}ฯต_{2T}\varphi _K^{}^v(x)+\overline{)}ฯต_{2T}\overline{)}P_2\varphi _K^{}^T(x)+{\displaystyle \frac{M_K^{}}{P_2n_{}}}iฯต_{\mu \nu \rho \sigma }\gamma _5\gamma ^\mu ฯต_{2T}^\nu P_2^\rho n_{}^\sigma \varphi _K^{}^a(x)].`$ (22)
The $`K^{}`$ meson distribution amplitudes up to twist-3 are given by ref. with QCD sum rules.
$`\varphi _K^{}(x)={\displaystyle \frac{3f_K^{}}{\sqrt{2N_c}}}x(1x)[1+0.57(12x)+0.07C_2^{3/2}(12x)],`$ (23)
$`\varphi _K^{}^t(x)={\displaystyle \frac{f_K^{}^T}{2\sqrt{2N_c}}}\{0.3(12x)[3(12x)^2+10(12x)1]+1.68C_4^{1/2}(12x)`$
$`+0.06(12x)^2[5(12x)^23]+0.36\{12(12x)[1+\mathrm{ln}(1x)]\}\},`$ (24)
$`\varphi _K^{}^s(x)={\displaystyle \frac{f_K^{}^T}{2\sqrt{2N_c}}}\{3(12x)[1+\mathrm{0..2}(12x)+0.6(10x^210x+1)]`$
$`0.12x(1x)+0.36[16x2\mathrm{ln}(1x)]\},`$ (25)
$`\varphi _K^{}^T(x)={\displaystyle \frac{3f_K^{}^T}{\sqrt{2N_c}}}x(1x)[1+0.6(12x)+0.04C_2^{3/2}(12x)],`$ (26)
$`\varphi _K^{}^v(x)={\displaystyle \frac{f_K^{}^T}{2\sqrt{2N_c}}}\{{\displaystyle \frac{3}{4}}[1+(12x)^2+0.44(12x)^3]`$
$`+0.4C_2^{1/2}(12x)+0.88C_4^{1/2}(12x)+0.48[2x+\mathrm{ln}(1x)]\},`$ (27)
$`\varphi _K^{}^a(x)={\displaystyle \frac{f_K^{}^T}{4\sqrt{2N_c}}}\{3(12x)[1+0.19(12x)+0.81(10x^210x+1)]`$
$`1.14x(1x)+0.48[16x2\mathrm{ln}(1x)]\},`$ (28)
where the Gegenbauer polynomials are
$`C_2^{\frac{1}{2}}(\xi )={\displaystyle \frac{1}{2}}(3\xi ^21),`$ (29)
$`C_4^{\frac{1}{2}}(\xi )={\displaystyle \frac{1}{8}}(35\xi ^430\xi ^2+3),`$ (30)
$`C_2^{\frac{3}{2}}(\xi )={\displaystyle \frac{3}{2}}(5\xi ^21).`$ (31)
For $`\rho `$ and $`\omega `$ meson, we employ $`\rho ^0=\frac{1}{\sqrt{2}}(u\overline{u}d\overline{d})`$ and $`\omega =\frac{1}{\sqrt{2}}(u\overline{u}+d\overline{d})`$. Their Lorentz structures are similar to $`K^{}`$ meson, the distribution amplitudes are the same for $`\rho `$ and $`\omega `$ and given as :
$`\varphi _\rho (x)`$ $`=`$ $`{\displaystyle \frac{3f_\rho }{\sqrt{2N_c}}}x(1x)\left[1+0.18C_2^{3/2}(12x)\right],`$ (32)
$`\varphi _\rho ^t(x)`$ $`=`$ $`{\displaystyle \frac{f_\rho ^T}{2\sqrt{2N_c}}}\{3(12x)^2+0.3(12x)^2[5(12x)^23]`$ (33)
$`+0.21[330(12x)^2+35(12x)^4]\},`$
$`\varphi _\rho ^s(x)`$ $`=`$ $`{\displaystyle \frac{3f_\rho ^T}{2\sqrt{2N_c}}}(12x)\left[1+0.76(10x^210x+1)\right],`$ (34)
$`\varphi _\rho ^T(x)`$ $`=`$ $`{\displaystyle \frac{3f_\rho ^T}{\sqrt{2N_c}}}x(1x)\left[1+0.2C_2^{3/2}(12x)\right],`$ (35)
$`\varphi _\rho ^v(x)`$ $`=`$ $`{\displaystyle \frac{f_\rho }{2\sqrt{2N_c}}}\{{\displaystyle \frac{3}{4}}[1+(12x)^2]+0.24[3(12x)^21]`$ (36)
$`+0.12[330(12x)^2+35(12x)^4]\},`$
$`\varphi _\rho ^a(x)`$ $`=`$ $`{\displaystyle \frac{3f_\rho }{4\sqrt{2N_c}}}(12x)\left[1+0.93(10x^210x+1)\right].`$ (37)
|
warning/0506/astro-ph0506726.html
|
ar5iv
|
text
|
# The INT Photometric H๐ผ Survey of the Northern Galactic Plane (IPHAS)
## 1 Introduction
The astronomical significance of $`H\alpha `$ spectral line emission is that it both traces diffuse ionized nebulae and is commonly prominent in the spectra of pre- and post-main-sequence stars and binaries. Since these are objects in relatively short-lived phases of evolution, they are a minority in a mature galaxy like our own. Their scarcity has in turn acted as a brake on our understanding of these crucial evolutionary stages that in youth help shape the growth of planetary systems, and in old age determine stellar end states and the recycling of energy and chemically-enriched matter back into the galactic environment. The major groups of emission line stars include all evolved massive stars (supergiants, luminous blue variables, Wolf-Rayet stars, various types of Be star), post-AGB stars, pre-main-sequence stars at all masses, active stars and interacting binaries. This last group most likely harbours SN Ia progenitors within it, in guises that are still subject to considerable debate (Hillebrandt & Niemeyer 2000; Uenishi, Nomoto & Hachisu 2003).
Existing catalogues of emission line objects contain anything from a few, to a few hundred, sources. Within the least populous object classes (e.g. the luminous blue variables and supersoft X-ray binaries, with just a few of each known in the Galaxy) there can be a confusing mรชlรฉe of โspecial casesโ that inhibit confident identification of essential and general behaviours. In effect, stellar evolutionary studies have been bedevilled by small number statistics and a lack of good demographics. The remedy for this problem is to exploit the technical developments of recent years that have boosted both the efficiency with which large scale astronomical surveys can be performed, and the quality that can be achieved. In particular, large area CCD-mosaic detectors offering good spatial resolution have now completely supplanted the photographic techniques of the last century. In this paper, we describe the Isaac Newton Telescope (INT) Photometric H$`\alpha `$ Survey of the Northern Galactic Plane (IPHAS), a programme that began taking data with the INT Wide Field Camera in the second half of 2003.
The goal of IPHAS is to survey the entire northern Galactic Plane in the latitude range $`5^\mathrm{o}<b<+5^\mathrm{o}`$ โ a sky area of 1800 sq.deg. The choice of latitude range was tensioned between the rising total telescope time requirement and the expected fall off in discoveries to be made with increasing latitude (see below). The 10-degree wide strip requires in the region of 22 weeks clear time, and the hope is to complete the observations before the end of 2006. The data obtained will be mined both for spatially-resolved nebulae and for unresolved emission line stars. For point sources, the magnitude range will be $`13<r^{}<20`$. Here we will focus on presenting the basic features of the survey, together with the extraction of point source data and the analysis of photometric colour information. The different technical issues relating to the identification and measurement of resolved H$`\alpha `$-emitting nebulae will be presented in a later paper. For now, we just point to the opportunity that IPHAS presents both for making new discoveries and for high quality H$`\alpha `$ emission mapping on large angular scales.
To place this new northern hemisphere survey in context, it is appropriate to review the scale and character of the emission line star population that previous Galactic H$`\alpha `$ surveys have revealed. Kohoutek & Wehmeyer (1999, hereafter KW99) have added their own discoveries within the latitude range $`10^\mathrm{o}<b<+10^\mathrm{o}`$ (1979 objects; data obtained in the years 1964-1970) to those of a wide range of independent searches: these go back as far as the original work of Merrill & Burwell that resulted in the Mount Wilson Catalogue (MWC, see Merrill & Burwell 1933). The total number of KW99 objects is 4174. In many cases the source observations are spectra obtained using objective prism facilities. For sources in the northern hemisphere, this compilation supercedes that due to Wackerling (1970). Three-quarters of the stars listed in KW99 are assigned a photovisual magnitude, $`m_{\mathrm{pv}}<13`$, and it is surmised that this is roughly the catalogueโs completeness limit. They also note that over 80 percent of all the objects they list in the $`10^\mathrm{o}<b<+10^\mathrm{o}`$ band fall within the narrower $`5^\mathrm{o}<b<+5^\mathrm{o}`$ band. At the fainter magnitudes we are exploring, we might expect this concentration toward the Galactic Equator to become even more pronounced. Naive extrapolation of the bright-end ($`m_{\mathrm{pv}}<13`$) magnitude distribution of the KW99 emission line stars to span $`13<m_{\mathrm{pv}}<20`$ would suggest that our survey should uncover 8000โ10000 new objects.
This is probably an underestimate for a number of reasons. First, we can check this extrapolation of KW99 against the same quantity derived from the Stephenson & Sanduleak (1971, hereafter SS71) southern Galactic Plane survey. The SS71 completeness limit is shallower at $`m_{\mathrm{pg}}11`$. We find, even on excluding the Galactic Bulge region located exclusively in the southern sky, that the prediction rises to $``$40000. This dramatic difference can have a number of origins โ beginning with simple differences in the Galactic stellar populations accessible from the northern and southern hemispheres, and ending with issues of experimental technique. Nevertheless a parallel between the KW99 and SS71 catalogues is that the bright magnitudes sampled strongly favour early-type, intrinsically luminous stars (such objects account for three-quarters of the KW99 catalogue). On going to much fainter magnitudes the sampled emission line star population is likely to broaden in character as intrinsically fainter object types (e.g. young and active stars, interacting binaries) become included.
An immediate precursor to IPHAS and, indeed, a prompt for the need for a northern survey, is the AAO/UKST narrow-band H$`\alpha `$ Survey of the Southern Galactic Plane and Magellanic Clouds. This was the last photographic sky survey carried out on the UK Schmidt Telescope (UKST). It was completed in 2003 and is now available as digital survey data derived from SuperCOSMOS scans of the original survey films (the SHS database, located at http://www-wfau.roe.ac.uk/sss/halpha/). A description of this survey is presented by Parker et al (2005): important points to note are its high spatial resolution ($`1`$ arcsec) and its areal completeness โ the entire southern Galactic Plane was imaged within the latitude range $`10{}_{}{}^{}<b<+10^{}`$. Each of the 233 Galactic Plane fields observed had an effective dimension projected on the sky of $`4{}_{}{}^{}\times 4^{}`$. The southern survey has provided the source material for a variety of continuing research projects (see e.g. Morgan, Parker & Russeil 2001; Parker & Morgan 2003; Drew et al 2004). For the detection of point sources, IPHAS betters both the SHS sensitivity and spatial resolution, and offers the advantage of CCD dynamic range and linearity. The sensitivity of the two surveys to spatially-resolved H$`\alpha `$ emission is comparable.
We begin the description of IPHAS in the next section with a presentation of the filters used, and our observing and data reduction techniques. Following this, in section 3, we discuss the use of the H$`\alpha `$,$`r^{}`$,$`i^{}`$ filter photometry in the diagnostic $`(r^{}H\alpha )`$ versus $`(r^{}i^{})`$ diagrams that can be constructed from the survey data for point sources. Specifically, we introduce simulated colour-colour tracks for both normal solar-metallicity stars occupying the main stellar locus and for emission line objects. We then provide examples of $`(r^{}H\alpha ,r^{}i^{})`$ diagrams in three contrasting northern Galactic plane locations (sections 4 to 6). The fields discussed are identified in Table 1. In section 4, we illustrate the application of the simulated tracks for normal stars with reference to fields in Aquila; in section 5 we perform a consistency check of IPHAS photometry of a Taurus field obtained on a photometric night; and in section 6 we present some follow-up spectroscopy relating to a field in Cepheus that illustrates the high success rate achieved in the confirmation of candidate emission line objects. In Section 7, we outline the application of IPHAS to imaging spatially-resolved nebulae, and illustrate this with the beautiful example of the supernova remnant, S 147. The paper ends with a summarising discussion (Section 8).
## 2 Survey observations and data extraction
### 2.1 IPHAS observations
The Wide Field Camera, mounted on the 2.5-metre Isaac Newton Telescope, is an imager comprising 4 AR-coated, thinned 4K $`\times `$ 2K EEV CCDs arranged in an L shape, capturing data from an on-sky area of approximately 0.3 of a square degree. With a pixel dimension of 13.5 $`\mu `$m, corresponding on-sky to 0.333$`\times `$0.333 arcsec<sup>2</sup>, the instrument is appropriately configured to fully exploit the high quality sub-arcsecond seeing frequently encountered at the Roque de los Muchachos Observatory in La Palma. Adequately sampled $``$1 arcsec resolution is particularly useful given that lower-reddening Galactic Plane star fields, observed down to $``$20th magnitude, are at times very crowded.
Not accounting for the geometric consequences of the L-shaped detector arrangement, the total number of pointings required to span the 10$`\times `$180 deg<sup>2</sup> survey area would be 6000. On accounting for the detector outline and requiring a little overlap between pointings, we have chosen to fix the number of field centres at a total of 7635. Furthermore, each pointing is paired with a second pointing at an offset of 5 arcmin W and 5 arcmin S, such that the number of quality-controlled sets of exposures expected to be compiled into the final survey database is 15270. Stars falling into a gap between the mosaiced CCDs in one exposure are captured in a partner exposure set. Nevertheless the great majority of Galactic Plane sources will be imaged at least twice. The pairing and offsetting, together with the chosen tessellation, comes very close to complete coverage of the northern Plane ($`>`$ 99 %).
Since H$`\alpha `$ falls in the red part of the spectrum, IPHAS was conceived of as a large-scale programme that could readily make use of less heavily subscribed bright and grey nights. To ensure this scheduling flexibility, whilst obtaining the associated continuum-band observations required for establishing unambiguous H$`\alpha `$ excesses, it was decided to restrict our broadband choices to red or longer wavelengths. This also has the effect of increasing the penetration of the survey for a given exposure time since these longer wavelengths are also less subject to Galactic dust obscuration than UBV bands. This stands in contrast to the somewhat bluer emphases of the older H$`\alpha `$ catalogues (e.g. SS71, where the objective prism data spanned $`3300<\lambda `$(ร
)$`<6800`$). The particular choice we made was to obtain the H$`\alpha `$ exposures alongside Sloan $`r^{}`$ and $`i^{}`$ filter observations.
All three filter profiles are plotted as Fig. 1. The Sloan filters have been preferred over Harris alternatives because of their squarer transmission profiles. The Sloan $`r^{}`$ filter is the most blue-sensitive of the three (central wavelength 6240 ร
), with the H$`\alpha `$ filter positioned toward the red end of its bandpass (central wavelength 6568 ร
). With a full-width at half-maximum (FWHM) transmission of 95 ร
, the H$`\alpha `$ filter is more than broad enough to capture all likely Doppler shifts due to Galactic motions of up to a few hundred km s<sup>-1</sup> or $``$10 ร
, as well as blueshifts of up to an additional $``$10 ร
due to the converging beam of the INT/WFC. The central wavelength of the Sloan $`i^{}`$ filter is 7743 ร
. We have added two broad band filters to this survey in order to give a continuum-dominated colour with which the $`(r^{}H\alpha )`$ excess measurement can be compared. It has been shown in the past (see e.g. Robertson & Jordan 1989) that this is important for distinguishing between a genuine H$`\alpha `$ emission excess and a molecular band dominated late type stellar spectrum โ in such cases large $`(r^{}H\alpha )`$ โcolourโ will correlate with relatively extreme $`(r^{}i^{})`$. In truth, the diagnostic value of this strategy is wider than this, as shall become apparent in Section 3.
The exposure times in the three filters were set at 120 sec ($`H\alpha `$) and 10 sec ($`r^{}`$ and $`i^{}`$) for the first seasonโs observing in 2003. Evaluation of these data, once extracted, led us to increase the $`r^{}`$ band exposure to 30 secs, from the start of the 2004 observing season, to compensate better for their typically higher moonlit background. This adjustment also acknowledges the pivotal role the $`r^{}`$ band exposures must play in the surveyโs exploitation โ it is important that errors in this band, appearing in both the H$`\alpha `$ excess and the broadband colour measurement, are minimised.
For the purpose of photometric calibration, each nightโs observations includes standard fields, obtained in twilight and at intervals of approximately 2 hours through the night. The standards are chosen from a list including the Landolt equatorial fields (Landolt 1992), Sloan (Smith et al 2002) and Stetson standards (at the Canadian Astronomy Data Centre, http://cadcwww.dao.nrc.ca/standards/). Nightly observations are also acquired of spectrophotometric standards with a view to assisting the final calibration of the narrow-band H$`\alpha `$ data. A programme of supporting spectrophotometric observations is planned with a view to placing the H$`\alpha `$ calibration on the desired firm footing in the longer term.
### 2.2 Data processing
Processing of IPHAS INT WFC data generally follows the pipeline procedure devised by Irwin and Lewis (2001) for dealing with optical mosaic camera data. The two-dimensional instrumental signature removal includes provision for: non-linearity correction at the detector level; bias and overscan correction prior to trimming to the active detector areas; flatfielding; and fringe removal in the $`i^{}`$ passband.
Flatfielding in all bands is accomplished by stacking suitable twilight flatfield exposures taken over the course of each typically one-week observing run to create master calibration flats. These have been found to be stable on this timescale provided no filter changes, or other instrumental setup changes, occur in the middle of the run. The gain differences between each detector in each passband are removed by normalising a robust measure of the average sky level for each detector to a common system (in this case the sky level on CCD no. 1). The flatfielded $`i^{}`$ data, even for the short exposures (10 s) used here, show measurable fringing. At the same time, the data taken for the IPHAS project are not themselves suitable (short exposures in crowded Galactic Plane regions) to construct good quality fringe maps for correcting this problem. Since the fringing in the $`i^{}`$ band is relatively stable with time, we make use of a library of $`i^{}`$ band fringe maps taken from other observing runs using the INT WFC. These have been found to reduce the level of fringing to an acceptable level when used with the defringing algorithm in the pipeline.
Each master flat, in conjunction with a previously defined bad column list, is also used to construct confidence maps for each passband. These are used during the catalogue generation to flag less reliable pixels in each image by providing a measure of the inverse variance weight for each pixel e.g. bad pixels have zero weight, heavily vignetted regions have low weight, poor DQE pixels have lower weight, and so on. These confidence measures are used directly to weight the image detection part of the catalogue generation algorithm and help avoid generating excessive numbers of spurious images around defects and other excessively noisy regions.
Catalogue generation follows the precepts outlined by Irwin (1985, 1997) and includes the facility to: automatically track any background variations on scales of typically 20โ30 arcsec; detect and deblend images or groups of images; and parameterise the detected images to give various (soft-edged) aperture fluxes, position and shape measures. The generated catalogues start with an approximate World Coordinate System (WCS) defined by the known telescope and camera properties (eg. WCS distortion model) and are then progressively refined using all-sky astrometric catalogues (eg. USNO, APM, 2MASS) to give internal precision generally better than 0.1 arcsec and global external precision of 0.25 arcsec with respect to USNO and APM, and 0.1 arcsec with respect to 2MASS. These latter numbers are solely dependent on the accuracy of the astrometric catalogues used in the refinement.
All catalogues for all CCDs for each pointing are then processed using the image shape parameters for morphological classification in the main categories: stellar; non-stellar; noise-like. A sampled curve-of-growth for each detected object is derived from a series of aperture flux measures as a function of radius. The classification is then based on comparing the curve-of-growth of the flux for each detected object with the well-defined curve-of-growth for the general stellar locus. This latter is a direct measure of the integral of the point spread function (PSF) out to various radii and is independent of magnitude, if the data are properly linearised, and if saturated images are excluded. The average stellar locus on each detector is clearly defined and is used as the basis for a null hypothesis stellar test for use in classification. The curve-of-growth for stellar images is also used to automatically estimate frame-based aperture corrections for conversion to total flux. <sup>1</sup><sup>1</sup>1We note that in regions of intense nebular emission with increasingly short spatial scale variations of the โbackgroundโ, automatic detection, parameterisation and classification of objects becomes progressively more unreliable. In such regions continuum subtraction via difference imaging will yield better results.
Any photometric standards observed during the run (mainly Landolt 1992 and spectrophotmetric standards) are automatically located in a standards database and used to estimate the zero-point in each passband for every pointing containing any of these standards. The trend in the derived zero-points is then used to assign a photometric quality index for each night and also as a first pass estimate for the magnitude calibration for all the observations. The H$`\alpha `$ filter is treated as equivalent to a standard Johnson-Cousins R-band filter to obtain a Vega-like magnitude which is used as an initial calibration (to be refined later, as mentioned above).
Various quality control plots are generated by the pipeline and these are used to monitor characteristics, such as:- the seeing; the average stellar image ellipticity (to measure trailing); the sky brightness and sky noise; the size of aperture correction for use with the โoptimalโ aperture flux estimates (here โoptimalโ refers to the well-known property that soft-edged apertures of roughly the average seeing radius provide close to profile fit accuracy eg. Naylor 1998). The โoptimalโ catalogue fluxes for the $`r^{}`$, $`i^{}`$ and H$`\alpha `$ filters for each field are then combined to produce a single matched merged catalogue from which diagnostic colour-magnitude diagrams and two-colour diagrams may be produced. These merged catalogues โ the fundamental IPHAS product โ contain flux, classification and match position error for each object in each passband.
The IAU-registered naming convention for all point sources derived from these catalogues is IPHAS JHHMMSS.ss$`+`$DDMMSS.s โ thereby encoding the 2000 object co-ordinates into the name.
To give an impression of the internal magnitude errors in the catalogued magnitudes and derived colours we plot, in Fig. 2, the rms deviation between the magnitudes measured in each filter, and the associated colours, for point sources common to two overlapping exposure sets (fields 2540 and 2540o, discussed again in Section 5). These were obtained on a photometric night in November 2003 as the moon was setting. Calculated empirically as $`\sqrt{<(m_{2540}m_{2540\mathrm{o}})^2>}`$ or its colour equivalent, over a range in mean magnitude $`\mathrm{\Delta }m=0.5`$ or $`\mathrm{\Delta }m=0.25`$, the error is corrected back to a representative single field measurement error by dividing by $`\sqrt{2}`$. The bright-end errors in the magnitudes themselves, in plots such as these, are typically dominated by calibration offsets of a few hundredths that will be removed when a final uniform survey calibration is devised. In the case of fields 2540 and 2540o the offsets were all small (less than $`0.01`$). From 2004 on, when the $`r^{}`$ exposures were increased to 30 sec, the faint-end $`r^{}`$ errors drop to around 60% of those for 2003 (for the same sky conditions). This carries through to the colour errors falling to 80% or less of their 2003 levels. Altogether, this significantly raises the fraction of catalogued objects that will meet the quality target of $`\mathrm{\Delta }r^{}0.1`$ for $`r^{}20`$.
To date roughly 3 Tbytes of raw data from the first two seasons of IPHAS observing have been processed this way. This corresponds to well over 100,000 4kx2k CCD images and over 40 million objects have been catalogued. All of this processed data is also available at the individual frame and catalogue level via a PostgreSQL database interface which allows users to: postage stamp browse for candidate verification; construct image catalogue overlays, including on-the-fly matching with other catalogues such as the 2MASS point source catalogue; perform on-demand continuum image subtraction and mosaicing of larger areas; access all the quality control information; and more (see Irwin et al 2005). The database interface is available on the Cambridge Astronomical Survey Unit (CASU) website, at http://apm2.ast.cam.ac.uk/cgi-bin/wfs/dqc.cgi. Co-ordinates of the centres of the observed IPHAS fields are obtainable there. <sup>2</sup><sup>2</sup>2These are identified via object names taking the form intphas\_nnnn$``$, where nnnn is a 4-digit number up to 7635, and $``$ is the wild card for further characters identifying exposure type.
## 3 Simulation of the IPHAS colour-colour plane
The three bandpasses of the survey provide the basis for the construction of a number of magnitude-colour diagrams and a colour-colour diagram to describe any chosen region in the northern Galactic Plane. Using just the two $`r^{}`$, $`i^{}`$, broad bandpasses, one may derive colour-magnitude diagrams that can in principle reveal different sequences at different reddenings that may be present in the field under investigation.
Full exploitation of IPHAS hinges on the colour-colour plane involving all three bands. The combination of magnitudes we use is $`(r^{}i^{})`$ as abscissa and $`(r^{}H\alpha )`$ as ordinate, so that objects with H$`\alpha `$ band excesses appear higher within the diagram, while intrinsically redder or more highly reddened objects are over to the right. The most straightforward use that can be made of such diagrams is to pick out for spectroscopic follow-up those objects whose $`(r^{}H\alpha )`$ colour places them clearly above the main locus of non-emission line objects. Additional information contained within the colour-colour diagrams can lead to identification of more subtle candidate emission line stars and also to a characterisation of the stellar populations distributed along the line of sight. In this second sense, IPHAS can also be seen as providing a far-red map of stellar populations in the northern Galactic Plane.
Fig. 3 is a composite $`(r^{}H\alpha ,r^{}i^{})`$ plot derived from data obtained in three paired IPHAS fields (fields 4090/4090o, 4095/4095o and 4199/4199o: see Table 1). In each case, catalogues of sources classified by the CASU pipeline as either โdefinitely stellarโ or โprobably stellarโ were extracted from within a 30 x 30 arcmin<sup>2</sup> box spanning most of the overlap region between the two pointings. For each extracted object, the datum plotted is the mean of the colours derived independently from each of the two exposures making up the field pair. The data shown are limited to the magnitude range $`13<r^{}<20`$, where the error in either colour is kept to less than $``$0.05 magnitudes. These are representative of the better data in the IPHAS database in that they were obtained on photometric nights in June 2004 at times of $``$1 arcsec seeing and low sky background.
All three fields are located in the Aquila Rift region, and sample sightlines that pass through the outer parts of the molecular cloud. Dame & Thaddeus (1985) noted that this is a nearby ($``$200 pc) and not particularly opaque cloud system, presenting around 2 magnitudes of visual extinction only. This is a modest addition to the reddening through the remaining Galaxy beyond โ the reddening data of Schlegel, Finkbeiner & Davis (1998) indicate maximum visual extinctions, $`A_V`$ ranging from $`5`$, in 4095, up to $`10`$, in 4199. The nearby rift cloud is responsible for the lightly-populated gap, seen in Fig. 3, between the upper sequence and the lower, but much more densely populated strip. The existence of this separation allows a clear demonstration of how well theoretically-synthesised tracks compare with and make sense of the photometry.
To achieve an understanding of the behaviours seen in the colour-colour domain, we have constructed two types of synthetic tracks: the first type concerns the properties of normal stars without H$`\alpha `$ emission, while the second explores the impact of adding narrow H$`\alpha `$ emission to generic stellar spectral energy distributions (SEDs). We present these tracks below, using the Aquila fields to illustrate the former in section 4.
### 3.1 The IPHAS colours of normal stars
For simulating the $`(r^{}H\alpha )`$ and $`(r^{}i^{})`$ colours of normal stars, we have used the library of stellar spectral energy distributions (SEDs) due to Pickles (1998, hereafter P98). At a final binning of 5 ร
the spectra in this library are well enough sampled that we may use them to compute narrow-band $`H\alpha `$ relative magnitudes with confidence, alongside the analogous broadband $`r^{}`$ and $`i^{}`$ quantities. The required numerical filter transmission profiles, shown in Fig. 1, are available via the ING WFC web pages (http://www.ing.iac.es/Astronomy/instruments/wfc/), as is a mean Wide Field Camera CCD response curve. To ensure compliance with the Vega-based zero magnitude scale, we have defined synthetic colour as follows:
$$(r^{}i^{})=2.5\mathrm{log}(\frac{\mathrm{\Sigma }T_r^{}F_\lambda \mathrm{\Delta }\lambda }{\mathrm{\Sigma }T_r^{}F_{\lambda ,V}\mathrm{\Delta }\lambda })+2.5\mathrm{log}(\frac{\mathrm{\Sigma }T_i^{}F_\lambda \mathrm{\Delta }\lambda }{\mathrm{\Sigma }T_i^{}F_{\lambda ,V}\mathrm{\Delta }\lambda })$$
(1)
where $`T_r^{}`$ and $`T_i^{}`$ are the $`r^{}`$ and $`i^{}`$ numerical transmission profiles, after multiplying by the mean WFC CCD response curve, and rebinning to match the P98 spectral library sampling. The SED for Vega, $`F_{\lambda ,V}`$, is the appropriately resampled version of that due to Hayes (1985). The $`(r^{}H\alpha )`$ colour is evaluated in the same way, after substituting the $`H\alpha `$ numerical profile in place of the $`i^{}`$ profile. Since Vega is an A0V star, its SED at H$`\alpha `$ incorporates a strong absorption line feature. Currently, because the CASU pipeline uses broad-band standard fields to calibrate measured source magnitudes, there is an offset in $`(r^{}H\alpha )`$ colour between the catalogue data and our simulations. The dominant spectral type in the standard fields will be appreciably later than Vegaโs A0, with the consequence that the standard-star SEDs will both be redder and less eroded by H$`\alpha `$ line absorption. On this basis one would expect, and we do find, that zero $`(r^{}H\alpha )`$ for unreddened main sequence stars corresponds to $`(r^{}i^{})0.3`$ (late F), rather than to $`(r^{}i^{})=0`$ (Vega, A0V), in plots of IPHAS data obtained in photometric conditions. Hence, on comparing simulated tracks with observation it is necessary to correct for this. The best way to do this is to assume that only a shift in $`(r^{}H\alpha )`$ is required, given that the calibration of the broadband-only $`(r^{}i^{})`$ colour should be secure enough (see Section 5 and the discussion of Fig. 9). The shift that needs to be applied to theoretical $`(r^{}H\alpha )`$ values to match them to observation is then always downwards, varying in amount between about $``$0.10 and $``$0.25.
We have simulated tracks for main sequence stars (luminosity class V), giant stars (class III) and supergiants (class I) using, for simplicity only, solar metallicity P98 spectra. Each sequence has been calculated for a range of reddenings using an $`R=3.1`$ optical-IR extinction law in the form given by Howarth (1983). The colours derived are given for $`E_{BV}=0`$, 1, 2, 3 and 4 in Tables 2 and 3. They are specified at this level of detail because there is no single reddening vector that translates all of a sequence directly onto its reddened counterpart. The unreddened dwarf, giant and supergiant sequences are compared in Fig. 4, where it can be seen that there is a gradual decrease in track gradient with increasing luminosity class. Nevertheless between mid-F and late-K spectral types there is minimal distinction between the luminosity classes โ this is the regime where H$`\alpha `$ absorption is weak and there is not yet any marked development of the molecular band structure that typifies M-type stars.
The effect of interstellar extinction on the sequences is illustrated in Fig. 5. With increasing reddening, the range of $`(r^{}H\alpha )`$ colour spanned by each luminosity-class sequence diminishes. This shrinkage in dynamic range is due to effective wavelength of the $`r^{}`$ bandpass lengthening, and moving closer to the $`H\alpha `$ bandpass as reddening becomes ever more extreme. It may be seen in Fig. 5 that the supergiant sequence essentially reddens along itself, while the main and giant sequences shift across in a manner that sweeps out area in the colour-colour plane. Also shown in this figure is the reddening locus for A0 dwarfs โ for non-degenerate stars and binaries this line amounts to an important boundary in that none should drop below it. In practice, stars will be carried into the forbidden domain by a range of types of observational error. Degenerate dwarfs and related objects with stronger H$`\alpha `$ absorption than early A stars would also fall below this line.
### 3.2 The IPHAS colours of emission line stars
To assess the impact on $`(r^{}H\alpha )`$ and $`(r^{}i^{})`$ colours of increasingly strong H$`\alpha `$ emission, we have represented underlying stellar SEDs using simple power laws or, at later spectral types, blackbodies. At the acceptable price of some approximation, such as ignoring stronger spectral features like the Paschen limit, this approach allows us to explore a broad range of SEDs flexibly and to quantify thresholds for the straightforward detection of H$`\alpha `$ emission. We present results for 4 simple SEDs: three power laws of the form $`F_\lambda \lambda ^\beta `$ with $`\beta `$ set equal to 4 (Rayleigh-Jeans case, relevant to the hottest O stars), 3 (appropriate to $``$A0 stars) and 2.3 (the optically thick accretion disk case); a Planck function at a temperature of 5900 K that is a good match to the G2V SED in the P98 library.
H$`\alpha `$ emission, where present, takes on a wide range of profiles in stellar spectra, and can be practically any width โ with FWHM anywhere in the range from a few 10s of km s<sup>-1</sup> up to 1000s. But, for now, we treat the simple limiting case of H$`\alpha `$ emission that is well-contained within the width of the INT/WFC narrow-band H$`\alpha `$ filter. The particular realisation used is of a rectangular profile of breadth 25 ร
centred at 6570 ร
(note that the effective H$`\alpha `$ filter bandpass is blueshifted for objects observed off-axis, which is why the central wavelength of the WFC filter is accordingly specified as 6568 ร
). This yields results negligibly different from using a Gaussian profile.
The synthesis of colours consists of the following steps: an underlying stellar SED is chosen; a rectangular H$`\alpha `$ emission profile of the desired equivalent width (EW) is superimposed; the resultant artificial spectrum is then reddened as required using the reddening law specified in section 3.1; finally, the reddened SED is multiplied by the product of the survey filter profiles and WFC response and integrated to form colours as in equation 1. We have not attempted to apply this procedure to very late type SEDs dominated by molecular bands โ in these stars, neither can the SED be easily parameterised, nor is an objective definition of H$`\alpha `$ EW straightforward.
On the basis of this procedure we have synthesised colours for the same set of reddenings ($`E(BV)=0`$ to $`4`$ in steps of $`1`$) for each of the 4 adopted stellar SEDs. Our results are presented in Fig. 6 and in Table 4. We find that there is a practical degeneracy between reddening and underlying SED such that a very-nearly unique locus is traced at each adopted H$`\alpha `$ emission EW. This means that, in principle, a given location in the colour-colour plane, above the main stellar locus, is associated with a particular H$`\alpha `$ EW.
A further property of the SED-specified tracks is that at EW up to $``$100 ร
, the trend with increasing EW is nearly vertical. But as H$`\alpha `$ EW becomes very large, the tracks bend toward smaller $`(r^{}i^{})`$ as the H$`\alpha `$ emission becomes a more significant contributor to the $`r^{}`$ flux. Indeed beyond an EW of 1000 ร
as the switch from a โstellarโ to a โnebularโ spectrum with little discernable continuum takes place, the bending becomes very extreme. In reality $`(r^{}i^{})`$ in the nebular case will also depend somewhat on the relative strength of line emission in the $`i^{}`$ band โ left out of consideration here. The limiting value of $`(r^{}H\alpha )`$ in the absence of any continuum is $`3.24`$ for our synthetic system referred to the Hayes (1985) SED for Vega (see Table 4). At the present time, without a properly defined zero point to the H$`\alpha `$ filter magnitudes, this translates to an effective observed upper limit on $`(r^{}H\alpha )`$ of around 3.1 โ any value appreciably above this signals a problem with the individual objectโs photometry.
Finally an important feature to note in the trend in IPHAS colours with respect to both emission EW and reddening is that the threshold for the detection of H$`\alpha `$ emission is lowest for bluer and/or less reddened objects (see Fig. 6). This implies, for instance, that IPHAS will pick out faint, nearby accreting objects very well indeed down to just a few Angstroms EW. Conversely, in the worst case of a densely populated main stellar locus spanning a wide range of reddenings, the EW threshold on the straightfoward detection of classical T Tau stars at $`E(BV)2`$ ($`A_V6`$) is around 30 ร
. Not infrequently, however, at larger $`(r^{}i^{})`$ ($`>2`$) the colour-colour plane below the unreddened main sequence may be sparsely populated โ the few objects located here could be the result of a combination of anomalous reddening and line emission. Indeed it is generally the case that objects, checked as having reliable photometry, lying outside the bounds of the densely populated main stellar locus for their field, have a relatively high probability of being interesting in one way or another.
## 4 Simulated and observed IPHAS colour-colour diagrams compared โ fields in Aquila
With the assistance of the synthetic tracks derived in the preceding section it is possible to begin to make sense of the morphologies appearing in IPHAS $`(r^{}H\alpha ,r^{}i^{})`$ diagrams โ with a view to their fuller exploitation. For this purpose we have selected some of the highest quality IPHAS observations obtained from pointings in the Aquila Rift region, allowing us to exploit its distinctive and easily interpreted colour-colour domain morphologies.
We begin with field 4095 in Aquila, included in Fig. 3 as the blue data points. This field is roughly centred on $`\mathrm{}=32.5^\mathrm{o},b=+4.8^\mathrm{o}`$, sampling a region close to the surveyโs Galactic latitude upper limit. The reddening data of Schlegel et al (1998) indicates that $`E_{BV}`$ typically does not exceed 1.6 in this direction. This makes it the least obscured of the three fields and, correspondingly, the field presenting the highest apparent density of stars (7097 out of the 13818 in Fig. 3). In Fig. 7 the data on extracted point sources are compared with selected synthetic tracks that have all been shifted downwards in $`(r^{}H\alpha )`$ by 0.17 to best match them to the data.
The upper panel in the Fig. 7 shows the brighter end of the magnitude range ($`13<r^{}<18`$) that includes a modest number of very nearly unreddened M dwarfs and a much larger number of mainly giant stars. Indeed the M giants form a particularly tight sequence at $`(r^{}i^{})>2.0`$. This suggests that most of the Galactic reddening along this sight line accumulates nearby because, if it were not, we would expect to see a more smeared giant distribution. The simulated giant tracks for $`E_{BV}=1.4`$ and $`1.6`$ are compared with this very well-defined feature. For $`(r^{}i^{})<2`$ many of the brighter objects will be giants at a plausible reddening; but at $`(r^{}i^{})>2`$, the synthesised tracks for M2-5 III stars fall too low by $`0.05`$ in $`(r^{}H\alpha )`$. A similar problem affects comparisons between synthesised and observed tracks for M dwarfs also (see below). At $`r^{}<18`$, only one object falls significantly below the early-A reddening line - it is likely to be a white dwarf or related object.
The lower panel in Fig. 7 presents the faint end of the $`r^{}`$ magnitude range, with the synthetic main sequence and giant tracks, reddened to $`E(BV)=1.6`$, superimposed. The main locus of observed objects is now a little more steeply angled, indicating that these fainter stars include a much increased component of main sequence objects. However, at $`r^{}20`$, stars later in spectral type than mid-K are only detectable at $`E(BV)<0.8`$, as evidenced by the scatter of points extending the main stellar locus up to $`(r^{}H\alpha )0.9`$. The small proportion of the plotted objects falling below the early-A reddening line can be presumed consistent with observational error. The 0.17 offset of the synthesised tracks was determined by optimising the positioning of both this notional line and the unreddened main sequence with respect to the data for $`r^{}18`$. These particular IPHAS observations have captured objects out to the limits of the Galactic disc population, such as reddened mid-M giants at $`10`$ kpc, located around 800 pc above the mid-plane at about the location of the far Sagittarius-Carina arm.
The more highly reddened Aquila field 4199, centred at $`\mathrm{}=34.3^\mathrm{o}`$, $`b=+1.8^\mathrm{o}`$, provides some degree of contrast with 4095 and is illustrated in Fig. 8. The reddening here is more variable with position, as well as more extreme. This shows itself directly in the broader red giant locus. Superimposed on the colour-colour plot are synthesised giant-star tracks for $`E_{BV}=2.4`$, and $`3.0`$. The latter value was selected because the maximum Galactic extinction for this field, derived from the Schlegel et al 1998 mapping data, is $`E_{BV}3.0`$. There is a rough consistency here with the findings from field 4095, in that the M giant track synthesised for this extinction falls a bit below the observed thinning of putative M giants (as it did for field 4095). Down to $`r=20`$, the colour-colour data suggest the presence of main sequence stars of K and earlier type at reddenings in the range $`1.2<E(BV)<2.4`$: the main sequence tracks for these limits are drawn to illustrate this (Fig. 8). This field again bears the imprint of the Aquila Rift in the relative deficit of stars between the lightly populated zero-extinction main sequence and the dense locus of stars at $`E_{BV}>1.2`$.
In field 4199 it is more apparent that the observed unreddened M dwarfs tend to maintain the locus gradient defined at earlier spectral types, rather than begin to turn over as the synthesised track indicates they should. This is most likely another symptom of the problem behind the M-giant discrepancy. At the present time, the available conversion between Landolt $`(RI)`$ colours, appropriate to the standard star fields, and Sloan $`(r^{}i^{})`$ are not properly defined for M dwarf colours (see Smith et al 2002). Similarly, the existing conversion used in the CASU pipeline is not validated for $`(r^{}i^{})>1.5`$. Clearly this will need to be corrected in the future. In the mean time, the comparison between observations of M stars and synthesised data will be increasingly qualitative as $`(r^{}i^{})`$ increases beyond the validation limit.
In neither field 4095 nor field 4199 do supergiants stand out in any obvious morphological way. This is likely to be both a consequence of their relative rarity and of the way in which their locus shifts almost along itself with increasing reddening. In principle, extremely red, isolated objects located below the red giant locus could be picked out as candidate reddened supergiants โ or as potential examples of other interesting object types. Indeed, we find that IPHAS J184644.25+015324.6, the one isolated object in this part of the 4199 colour-colour plane at $`(r^{}i^{},r^{}H\alpha )=(3.36\pm 0.02,0.499\pm 0.03)`$, cannot be a reddened supergiant. One reason is that the reddening ($`E(BV)>4`$) required to explain its position in these terms is excessive relative to the maximum expected for the field. Another is that the 2MASS point source within 0.2 arcsec of this objectโs position exhibits very bright $`JHK`$ magnitudes with unusual colours ($`K=9.55\pm 0.02`$, $`(JH)=1.68\pm 0.03`$, $`(HK)=0.76\pm 0.03`$). At $`r^{}=18.75\pm 0.01`$, there are no grounds for doubting the reliability of the IPHAS photometry and the reality of the source. The absence of any significant proper motion rules this object out as a nearby brown dwarf. This object is largely absent from pre-existing photographic surveys, except that there is a detection of it in the UKST Infrared (IVN) Survey reported on the SuperCOSMOS Sky Survey website (http://www-wfau.roe.ac.uk/sss/): it is reported there at an $`I`$ magnitude of 18.383, around 3 magnitudes fainter than the IPHAS $`i^{}`$ magnitude. The IVN plate was obtained in 1981. This variability, the anomalously low $`(r^{}H\alpha )`$ colour, together with its NIR colours, point towards a carbon star at a reddening corresponding to $`E(BV)1.4`$ (see e.g. Bessell & Brett 1988). A faint-end absolute $`K`$ magnitude for such a star, if on the AGB, would be -6.5 (Claussen et al 1987). This places it at $``$15 kpc.
## 5 A comparison between flux-calibrated spectra and IPHAS photometry โ a field in Taurus
Up to this point the interpretation of the $`(r^{}H\alpha ,r^{}i^{})`$ plane has been based on synthetic photometry derived from P98 library spectra. Early in February 2004, we obtained WHT/ISIS service spectra of 6 stars selected from IPHAS data on fields 2540 and 2540o in Taurus. These IPHAS images, obtained on 5th November 2003, were among the first to be pipeline-processed and were picked for closer investigation as examples of apparently good quality data obtained in good seeing and photometric conditions. The aim of the follow-up service spectra was to obtain relative spectrophotometry in order to ascertain optical SEDs, spectral types and reddenings for the sample stars as a retrospective check on the IPHAS colours, and the typical errors in them. This exercise gives an impression both of the current state of the photometric calibration of the data and of the quality of the synthetic colour comparisons derived from P98.
Field 2540 is centred on RA 05 33 49 Dec $`+`$25 15 00 (2000) in Taurus, only a few degrees from the Galactic anticentre direction. The characteristics of this sky position are very different from those in Aquila: here, the maximum Galactic reddening is modest and more smoothly varying, ranging from $`E(BV)0.8`$ in the NE of the $`30\times 30`$ arcmin<sup>2</sup> extracted region up to $`1.1`$ in its SW. The other obvious difference, which stands out in the $`(r^{}H\alpha ,r^{}i^{})`$ plot for 2540/2540o in Fig. 9, is the absence of any red giants within the magnitude range shown ($`13<r^{}<20`$). This absence is not just a consequence of the imposed magnitude limits since a K/M giant at 10 kpc viewed through $``$3 visual magnitudes of extinction should be detected at $`r^{}17`$: it must be a real absence. In the example of solar and lower metallicity isochrones presented by Bertelli et al (1994), a red giant branch is only well-developed from around 100 million years of age onwards โ suggesting that the stellar populations sampled in this part of the outer Galaxy are younger than this.
8 stars were initially selected for ISIS spectroscopy from the colour-colour diagram for fields 2540/2540o on the basis that they were not too faint ($`r^{}<18`$) and lay on the outer boundary of the main stellar locus (dark blue asterisks in Fig. 9). These criteria biased the selection in favour of evolved spectral types. In the event, 6 of the 8 stars (which we refer to as stars AโF) were observed during a service night of mediocre weather. Both the blue and red arms of ISIS were used, with the R600B and R316R gratings installed, delivering spectra spanning 3500โ5000 ร
and 6000โ8700 ร
for each star. To further the aim of relative spectrophotometry, the slit width was set fairly wide at 1.8 arcsec, while the slit orientation tracked the parallactic angle. The resolution of the spectra is $`3.6`$ ร
. An observation of the white dwarf G191$``$B2B was also obtained to serve as a spectrophotometric flux standard.
The data were extracted from the CCD frames and then wavelength- and flux-calibrated using routines from the software package, FIGARO. The extracted 1-D spectra were then imported to the software tool, DIPSO, in order to determine approximate spectral types and reddenings by comparing them with P98 library spectra. In every case, the spectral type determination rested on matching absorption line characteristics. This matching was performed using the blue spectra for all but the M4V star (star D) โ for this object the red spectrum was more appropriate. For each star, the P98 library spectrum of the appropriate spectral type was progressively reddened, using the mean Galactic extinction law (Howarth 1983), to identify the best fitting colour excess. The observed spectra and the best fits to them derived in this manner are shown in Fig. 10. The positions, magnitudes and further data on the 6 stars appear in Table 5. For star F, the data are not of sufficient quality to provide a reasonable fix on luminosity class: but we suspect that its H$`\alpha `$ profile indicates a lower gravity than a main sequence A5 star. In order to: smooth errors due to irregularities in the spectrophotometric flux calibration; avoid the need to correct for telluric absorption; provide a good extrapolation of the observed spectral energy distributions (SEDs) to cover the full spectral range, the final step of deriving photometric colours used the closest matching, appropriately-reddened P98 library spectra in the $`r^{}`$ and $`i^{}`$ bands, rather than the calibrated observations. Only the H$`\alpha `$ fluxes were computed by multiplying the H$`\alpha `$ filter profile directly with the calibrated ISIS spectra.
Both the original IPHAS colours and the colours derived from the fits to the spectrophotometry are listed in Table 5. As there is not yet a uniform and fully-verified zero-point calibration for all IPHAS frames, we have to shift the SED-based colours on to the IPHAS colours. The shifts that minimise the mean differences in each of $`(r^{}H\alpha )`$ and $`(r^{}i^{})`$ are $`0.14`$ and $`+0.03`$, with final rms deviations between the 6 pairs of observed and predicted colours of 0.015 and 0.026, respectively. The size and sense of shift in $`(r^{}H\alpha )`$ is as expected (see section 3). That the shift in $`(r^{}i^{})`$ is small, but apparently finite, indicates that the night the IPHAS imaging was obtained was not perfectly photometric. On the basis of the errors estimated for the IPHAS photometry (see Fig. 2), we would expect the rms deviations between the catalogued and observed colours to be $``$0.01. The somewhat larger values of 0.015 and 0.026 obtained here turn out to be determined mainly by errors in the relative spectrophotometry and its analysis: for example, the uncertainty in the $`E(BV)`$ estimates is typically 0.05 and translates into a $`(r^{}i^{})`$ error of $``$0.03. In $`(r^{}H\alpha )`$, the discrepancies are smaller and mainly arise in the $`r^{}`$ band integration. The circle from photometry to spectroscopy, back to photometry, closes satisfactorily.
At the end of this process, it can be seen in Fig. 9 that star A (A2V) falls a little below the early-A reddening line, rather than just above, while the reddening line itself lies $`0.02`$ magnitudes above the bottom edge of the main stellar locus. A part of the reason for this may be illustrated in Fig. 11 where the P98 A2V spectrum is superimposed on the ISIS observation of star A: it is possible that the limited resolution of the P98 library spectra ($`R500`$ or $`\mathrm{\Delta }\lambda 13`$ ร
at H$`\alpha `$) leads to the H$`\alpha `$ in-band fluxes of early A-type stars being overestimated, very slightly. Another factor will be linked to the question of the mean H$`\alpha `$ absorption EW and its variance for early-A stars as a function of sub-type and metallicity. The H$`\alpha `$ absorption EW for star A is $`14.5\pm 0.4`$ ร
, while that for the P98 A2V star is $`10.8\pm 0.5`$. These numbers for A2V may be compared with the P98 A0V and Hayes (1985) Vega H$`\alpha `$ EWs, that are both close to 13.0 ร
. Finally we note the contrast between the statement by Jaschek & Jaschek (1987) that the Balmer lines are strongest at A2 (see their Table 10.1) and the maximum in P98 for near-solar metallicity dwarfs at A0. The underlying practical difficulty here is the measurement and calibration challenge of the very broad H$`\alpha `$ absorption wings in A-type spectra. As the IPHAS survey completes and a uniform photometric calibration is constructed, it will then be appropriate to sort the issue out and improve the absolute registration of this lower boundary. For the present, the A0V spectrum in the P98 library defines the shape of the early-A reddening line well enough to allow it to be used in a relative manner.
The H$`\alpha `$ profile for star B is also presented in Fig. 11. This object shows a distinct central reversal in H$`\alpha `$ which allows it to be described as a weak emission line object. The excess (emission) equivalent width with respect to the library A0V spectrum is 12 ร
. Indeed it was included in the list of ISIS service targets because of the suspicion that this might be the case: when it is plotted on the colour-colour plane with only neighbouring stars, within a 10$`\times `$10 arcmin<sup>2</sup> box, it sits clearly separated in $`(r^{}H\alpha )`$ just above the local main stellar locus (see Fig. 12). This is an example of the greater coherence of mean conditions (reddening, nature of population) within a smaller sky area leading to greater success in identifying an โunusualโ object.
## 6 Spectroscopic trawling for emission-line and other rare objects โ fields in Cepheus
We now present some results from early spectroscopic follow-up of IPHAS in order to give a concrete example of the yields of different object types from IPHAS data. In this case, the choice of sky area has been dictated mainly by observational convenience, rather than by data quality considerations. The results presented here rest on more typical IPHAS photometry.
### 6.1 The MMT/HectoSpec observations
In June 2004 we obtained spectra with the Mount Hopkins 6.5-metre MMT in F/5 configuration using the recently commissioned HectoSpec facility, a multi-object spectrograph fitted with 300 fibres that can be deployed across a field, 1 degree in diameter (Fabricant et al 2004). The fibre positioner is mounted at Cassegrain. The 270 groove/mm grating used delivers broad wavelength coverage (4488 โ 8664 ร
) at 6.2 ร
resolution. Over two nights, six different fields were observed using two fibre configurations per pointing. The target stars selected for this programme fell mainly in the magnitude range $`17r^{}20`$. The total on-source exposure times were 1200 secs. Spectra were extracted by the instrument pipeline that includes CCD bias and gain corrections, flat-fielding using domeflats as well as a sensitivity correction for the individual fibres using twilight flats. Individual fibre spectra were then extracted and wavelength calibrated using FeNeAr-lamp exposures. Finally, a mean sky spectrum derived from the sky fibres was subtracted. Due to spatial variations in the sky background across the field of view, sky subtraction using sky fibres is not always perfect and care must be taken with spectra displaying weak, unresolved H$`\alpha `$ emission components. Sky subtraction will be improved in the future using offset sky exposures that sample the sky for each fibre close to the target position, in conjunction with an optimal scaling correction using the strongest sky lines.
### 6.2 Selecting targets for the Cepheus field
We report the results from one of two pointings in the constellation of Cepheus. Centred on RA 22 17 00, Dec $`+`$61 33 37 (2000) ($`\mathrm{}=105.6^\mathrm{o}`$, $`b=4.0^\mathrm{o}`$), this position was picked because it contains a strip included in the Spitzer Galactic First Look Survey (http://ssc.spitzer.caltech.edu/fls/galac/). Despite its relatively high Galactic latitude, this area of sky presents significant and locally-variable interstellar extinction (ranging from $`A_V4`$ up to $``$7 magnitudes). This shows up in the IPHAS colour data for the region as somewhat broadened main stellar loci. An example is shown as Fig. 13, where colour data extracted from the IPHAS field pair, 7012/7012o, are plotted. At magnitudes brighter than $`r^{}19`$ (top panel), the colour uncertainties are less than $``$0.05 and are less significant than environmental factors in the smearing of the main stellar locus โ this reverses at fainter magnitudes (lower panel) where the typical errors are $`0.1`$.
The HectoSpec 1<sup>o</sup>-diameter field spans then IPHAS field positions, not including offsets. The largest contributions, however, are from IPHAS fields 6985, 6993, 7012 and 7019 (see Table 1). To speed up the compilation of the target lists, we chose to merge the data from all the relevant pointings first, before proceeding to target selection. In merging the data, corrections for photometric shifts between different WFC exposures had to be applied. These were calculated using the mean magnitude offsets for sources located in field overlaps. Inevitably, this merging blurred the main stellar locus in the colour plane some more โ compare the lower panel in Fig. 13 with the plot for the full HectoSpec field in Fig. 14. For the future, we are re-ordering the algorithm, in order to give more emphasis to selection at the individual field level (cf the discussion of โstar Bโ at the end of Section 5). Indeed a further tactic that can be applied in order to minimise the spread of the main stellar locus is to select from within a number of narrowly-set $`r^{}`$ magnitude ranges. On this occasion, after merging the catalogues for the relevant individual fields together, the main target selection was performed within the full magnitude range to be observed, $`17r^{}20`$. Finally a few promising emission line star candidates down to $`r^{}=20.5`$ were added by hand.
The goal of this first round of HectoSpec observations was to explore the complete IPHAS colour-colour plane while trying to give high priority to objects that are outliers or near the edge of the general distribution of objects. This naturally includes all emission line star candidates, which lie above the main stellar locus.
To achieve this sampling, the following selection algorithm was applied. The colour-colour plane was split into bins of 0.1 magnitudes in width and height. Targets were then selected based on the number of objects in each bin: from 1 to 3 objects in a bin, all were selected and given the highest fibre allocation priority; between 4 to 9 objects in a bin, a random selection of 75% of the bin members was chosen with a slightly lower fibre allocation priority; between 10 and 50 objects, a random fraction falling linearly from 50% to 10% was selected and given a low fibre selection priority based on the number of objects in the bin; $`>`$50 objects in a box, 10% of the objects were selected randomly and given a low fibre selection priority based on the number of objects in the bin. The maximum number of objects that could be selected in a box was capped at 10. Finally, if a bin had 4 neighbouring bins in the cardinal directions with 10 or more objects, then only one object was selected for spectroscopy. This rule was introduced to further reduce the number of objects picked for spectroscopic follow up that lie in the densest โ and probably most uninteresting โ parts of the stellar distribution.
If the selection algorithm gave less than the requested number of objects ($`700`$ for 2 HectoSpec configurations) then 1 object per bin was added, starting with the least populated bins until the required number of objects was reached. If more than the requested number of objects was chosen, then 1 object was removed from each bin with more than 1 object selected โ starting with the most populated bin until the requested number of objects was reached. By this means we were able to assign fibres to between 450 and 540 targets, depending on the field observed. Each set was then split into two configurations.
The outcome of this process and the HectoSpec observation of the Cepheus field was a collection of 496 stellar spectra.
### 6.3 Results of the MMT/HectoSpec spectroscopy in Cepheus
First and foremost we find that essentially every target star located in the colour-colour plane clearly above the main stellar locus is confirmed as an emission line star. Altogether 29 objects are confirmed as having H$`\alpha `$ in emission, with 6 or 7 lying on or just below the upper bound of the highly populated region (green encircled points in Fig. 15). The one object not encircled in green, at $`(r^{}i^{})1.1`$ and $`(r^{}H\alpha )1.5`$, is probably an emission line object also, but remains ambiguous because it is very faint and its spectrum is correspondingly noisy. There is a still larger group, numbering 47, of probable dMe stars. They are โprobableโ because the sky subtraction may have left a false residue of H$`\alpha `$ emission. The range of H$`\alpha `$ emission equivalent widths in this group is from a few up to 10โ20ร
. Quite plausibly, most of these objects lie mixed in with non-emission dwarf M stars (of which there are 90 or more). The one โprobable dMeโ stars (at $`(r^{}i^{})2.50`$, $`(r^{}H\alpha )1.3`$) above the main locus can be viewed as dMe with the greatest confidence because time-variable H$`\alpha `$ emission is characteristic of dMe stars โ presumably at the time of the IPHAS imaging, its H$`\alpha `$ emission was brighter than 9 months later, at the time of the spectroscopy.
The combination of moderate spectral resolution and short exposure time has meant that many of the more routine objects, without H$`\alpha `$ in emission or in marked absorption, are more challenging to sort into spectral classes. This large group of 214 objects will be dominated by late-A to mid K stars, but will also include some non-emission OB stars. The spectra of a further 89 stars are so faint and noisy that no comment can be made about them. The stand-out objects towards the lower boundary of the main stellar locus in Fig. 15 are the stars with H$`\alpha `$ strongly in absorption. There are 15 of these. Two of them are well-separated from the main locus at much lower $`(r^{}i^{})`$ and also much lower $`(r^{}H\alpha )`$ โ they are both white dwarfs. Similarly placed objects in other IPHAS fields for which we have MMT/hectospec spectra have turned out to be white dwarfs too. The remaining thirteen stars with strong H$`\alpha `$ absorption are early A stars. Tighter classification at this time is not feasible.
We now present a selection of 8 objects and their spectra for more detailed discussion. These are identified in the colour-colour plane shown as Fig. 15. The data on them, given in Table 6, includes estimates of their $`r^{}`$ magnitudes and $`(r^{}i^{})`$, $`(r^{}H\alpha )`$ colours derived from the highest quality IPHAS exposures currently available. Note that the colours are, typically, different from those plotted in Figs. 15 โ this is due to the colour shifts applied in combining IPHAS fields before MMT/HectoSpec target selection. Since they assist in assigning broad object class, we also include in the table 2MASS $`(JH)`$, $`(HK)`$ colours and $`K`$ magnitudes.
Stars 1 โ 3 (Fig. 16) are most likely to be young stellar objects (YSOs) of Herbig or T Tau type. This object class assignment is easiest for star 1 since the veiling is not so extreme as to hide the underlying M-star spectrum. Indeed, a comparison between stars 1 and 4 in Fig. 16 suggests that these objectsโ M spectral sub-types are likely to be very similar. The sky around star 1 has been imaged in all four IRAC bands by the Spitzer Space Telescope First Look Survey. We downloaded the calibrated images from the Spitzer Science Archive and carried out point source extractions. Star 1 was detected at 3.6, 4.5 and 5.8 $`\mu `$m, with fluxes corresponding to magnitudes of 12.47$`\pm `$0.11, 12.12$`\pm `$0.12 and 11.93$`\pm `$0.14, respectively. Using the observed IPHAS, 2MASS and IRAC fluxes, we find that the SED of Star 1 from the r-band to 5.8 $`\mu `$m happens to fit quite well to a 2000 K blackbody (although the SED must include components due to a reddened stellar photosphere, an accretion disk and warm dust).
Star 2 has the highest contrast and richest emission line spectrum of the three stars โ in this object, even the Na i D lines are thrown into emission and some forbidden line emission is present. Star 3 is intermediate between 1 and 2, both in terms of the contrast of its emission spectrum, and that it is just possible to pick up late-type photospheric absorption against the continuum (e.g. the blend at 6495 ร
strong in G/early-K stars). The NIR colours of all 3 objects imply modest NIR continuum excesses that are not out of place for Herbig or T Tau stars, with star 2 showing the most marked excess.
It seems likely that these stars 1โ3 are associated with LDN 1188 ($`\mathrm{}=105.7`$, $`b=+4.2`$), a dark cloud less than $`2^\mathrm{o}`$ away from the well-known Sh 2-140 region of star formation. This proximity suggests a similar distance to both regions, which has been given as 910 pc for Sh 2-140 by Crampton & Fisher (1974). However both these nebulae lie on the periphery of Cep OB2, to which the distance appears to be rather less ($``$600 pc, de Zeeuw et al 1999). In a study of LDN 1188, Abraham et al (1995) reported the discovery of a number of emission line stars in objective prism data obtained at Konkoly Observatory at brighter magnitudes than our HectoSpec selection. RNO 140 and RNO 141 (Cohen 1980) are also in this neighbourhood.
Star 4 can be presumed to be in the foreground with respect to LDN 1188 and stars 1โ3. Hence its $`JHK`$ colours, combined with its spectral type can be used to estimate a minimum interstellar extinction towards LDN 1188. This works out at $`E(BV)1.3`$ or $`A_V4`$ (for $`R=3.1`$, and using data from Bessell & Brett 1988). We can now see if this marries up with the implications of the IPHAS colours of stars 1 โ 3 by comparing their catalogue values with reddened synthetic estimates. This is accomplished via Fig. 17 in which the IPHAS colours for stars 1 to 3 (Table 6) are compared with synthetic tracks (Table 4).
If star 2, with its generally high contrast emission line spectrum is a Herbig or T Tau star with an accretion dominated SED, its reddening would correspond to $`E(BV)2`$: for, in Fig. 17, it lies just to the right of the track for $`F_\lambda \lambda ^{2.3}`$ and $`E(BV)=2`$. A similar, or somewhat lower, reddening would appear plausible for star 1 in that its optical SED should be somewhat redder, intrinsically, than that of star 2. The intrinsic optical SED of star 3 should be intermediate between stars 1 and 2 (given the marginal detection of late-type photospheric absorption), and yet it is observed to be โbluerโ than either. The highest likely reddening of star 3 is $`E(BV)1.5`$: this reddening would apply in the limiting case of an accretion-dominated SED, where the contribution of the G/K star is small.
On the basis of its $`(r^{}i^{})`$ colour and spectral type, and after correction for its H$`\alpha `$ emission, star 4 would be assigned $`E(BV)1`$. This is a somewhat lower estimate than the estimate based on 2MASS NIR colours ($`E(BV)1.3`$) but not so large a discrepancy that either the NIR or optical photometrically calibration must be called into question. In conclusion, we find that the order of increasing reddening appears to be: star 4 in the foreground, star 3, and then star 1 and star 2, spanning the range $`1<E(BV)<2`$. The Schlegel et al (1998) Galactic reddening maps indicate maximal reddenings of $`E(BV)2.5`$ for this part of the Plane. We have rough consistency and a first indication of patchy reddening toward the young objects in the vicinity of LDN 1188 . If half the $`r^{}`$ flux of star 1 is attributed to an $``$M3 stellar photosphere, and $`A_V6`$, one may deduce a stellar radius of around 4 times the M3V main sequence radius, for a distance of $`600`$ pc.
The IPHAS colours for stars 2 and 3, picked out in Fig. 17, are broadly consistent with their observed H$`\alpha `$ EWs (200 ร
and 80 ร
), in that their colours โpredictโ EWs of $``$200 ร
and $``$100 ร
, respectively. Only star 1 is discrepant in that its observed EW ($``$190 ร
) is distinctly low compared with the implications of its IPHAS colours (suggesting an enormous EW of about 250 ร
). Given that H$`\alpha `$ EW is well known to be a time-variable quantity in most classes of emission line object, consistency for 2 out of 3 objects is acceptable.
The spectra of 4 non emission line objects, star 5โ8, are shown in Fig. 18. These draw attention to what can be called H$`\alpha `$ deficit positions in the IPHAS colour-colour plane (cf Fig. 15). At the blue end, objects 5 and 6 are examples of extreme and strong H$`\alpha `$ absorption objects: respectively a white dwarf and an early A star. At the red end, objects 7 and 8 are respectively a normal, somewhat reddened mid-M giant, while object 8 has the distinctive CN band structure of a carbon star in its spectrum. Object 7 is typical of the stars populating the red end of the giant strip in the IPHAS plane. The NIR colours and rough spectral type ($``$M4III) suggest a reddening corresponding to $`E(BV)`$ of about 2. Carbon stars like object 8 will usually fall below the red giant strip (lacking the TiO bands that, in M-type spectra, lead to the seeming flux maximum in the H$`\alpha `$ region). In terms of its NIR colours, star 8 is a more extreme object than the suspected carbon star mentioned at the end of section 5 (although reddening gives both objects more extreme $`(JH,HK)`$ colours than seen in the local Galactic carbon star sample of Claussen et al, 1987). Indeed Object 8 appears to be a reddened ($`A_V5`$) version of 2MASSI J0326599+143957, described by Liebert et al (2000) as a luminous, very cool, late N type carbon star.
In summary, our early MMT/Hectospec observations of this field in Cepheus sample a diverse range of objects. Sources lying clearly above the stellar locus in the $`(r^{}H\alpha ,r^{}i^{})`$ colour-colour plane have indeed been confirmed as true emission line objects with H$`\alpha `$ EWs ranging from a few to 200 ร
. Many of them are likely to be Herbig or T Tau stars. In addition several H$`\alpha `$ deficit sources have been identified. Multi-object spectroscopic follow-up will remain a key part of our efforts to mine the IPHAS database.
## 7 IPHAS opportunities for H$`\alpha `$ imaging
The power of the survey for detecting stellar H$`\alpha `$ emission has already been described. A second target for the survey are the spatially-resolved emission-line nebulae. These nebulae are โ like emission-line stars โ associated with early and late phases of stellar evolution, and indicate ionization of circumstellar gas, in the form of H ii regions, planetary nebulae and supernova remnants. Already, in the southern hemisphere, the UKST SHS has shown the remarkable incompleteness of existing catalogues by doubling the number of known planetary nebulae (Parker et al 2003).
The Galactic Plane shows ubiquitous diffuse H$`\alpha `$ emission as well as reflection nebulae, which need to be distinguished from circumstellar ionized nebulae. Since reflection nebulae are the product of continuum scattering, they can be removed by comparison of H$`\alpha `$ and $`r^{}`$ images. Diffuse H$`\alpha `$ emission occurs on large size scales (10 arcminutes to degrees) and lacks the usual symmetry of circumstellar nebulae.
Imaging of extended nebulae requires well behaved background on the CCDs making up the WFC. Because of this, the preferred image properties for nebular studies are very different from those for point source extraction: high background is acceptable for the latter, while poor seeing is acceptable for the former. Small nebulae covered well within a single CCD do not require special reduction. Each object will usually have been covered in at least two pointings by the time the survey is complete, giving improved S/N. For larger objects, mosaics need to be made. Subtracting an $`r^{}`$ frame removes stars: for small areas this can be done using PSF matching techniques, but this is very computationally intensive โ such that for larger fields, a direct subtraction is used which typically leaves larger residuals.
The limitations of the technique are largely due to background variations. The observations are generally taken in grey and bright time. Different fields will therefore present with very different background sky levels. The sky subtracts fairly well in an $`H\alpha r^{}`$ image (unless the background is variable under non-photometric conditions), but if there is smooth extended H$`\alpha `$ emission over large angular scales, its contribution is currently not separately determined from the sky contribution. Internal reflections are seen in some images, from bright stars. In some locations, a bright star just outside the field of view gives a flare-like feature on the edge of a nearby frame.
A photometric calibration is determined for point sources, but not for extended emission. To correctly calibrate emission nebulae, assuming the continuum background is fully subtracted, the filter response curve needs to be precisely known and needs to be stable over the likely Doppler wavelength shifts (Ruffle et al 2004). \[N ii\] will also intrude into the flux. Currently, calibration is best performed using known planetary nebulae located in the imaged area. Note that stellar H$`\alpha `$ sources cannot be used for calibration as their line flux tends to be time variable.
### 7.1 A supernova remnant
As an example of the possiblities regarding extended nebulae, we present an image of the supernova remnant S 147 (Shajn 147, or Simeis 147, in full โ not Sh 2-147, with which it is confused in SIMBAD). This is a near-perfect remnant of an approximately spherical shape, showing a typical filamentary structure. It is positioned just overlapping the anti-galactic centre, its own centre being at $`\mathrm{}180.1^\mathrm{o}`$, $`b1.6^\mathrm{o}`$. But due to its large extent, spanning several degrees, only photographic images have been published so far (Van den Bergh, Marscher & Terzian, 1973). See also the 24th March 2005 โAstronomy Picture of the Dayโ due to R. Gendler (http://antwrp.gsfc.nasa.gov/apod/ap050324.html).
An IPHAS image of S 147 was produced by combining approximately 250 pointings. The pipeline mosaicing procedure was found to be inadequate for combining many fields taken under widely varying conditions. We therefore used the pipeline only to produce reduced images of the individual CCDs for each pointing. Then, for each image, we subtracted the $`r^{}`$ image from the H$`\alpha `$ data and smoothed to a pixel size of 5 arcsec (a binning factor of 15). The background per image was approximately nulled by subtraction of the median: in fields with bright and extended emission this required manual selection of areas for background definition โ otherwise the median over all four CCDs of one pointing was used. One corner of CCD no. 3 is affected by scattered light in conditions of bright Moon light (amplified if cirrus was present): this corner was always blanked out. All CCDs tend to show a gradient along the long axis of about 1 ADU in the subtracted image: this could be due to a charge-transfer efficiency limitation, but its exact cause is not known. It does not subtract automatically because the $`r^{}`$ image is scaled first โ accordingly, as a final step, this remaining gradient was subtracted from all images.
The resulting images were combined in a single mosaic using the Virtual Observatory software package Montage. This involved regridding each frame, and determining background corrections by comparing areas in common between different images. We finally produced an image covering 25 square degrees. We note that extended emission on scales of a degree or more may not be well represented, as it can be affected by the background subtraction procedure. However, the filamentary structures are very well recovered. The background fitting was found to be insufficiently constrained at the outer edges of the imaged area, leaving some negative areas. To deal with this, a linear gradient was fitted to the background in empty regions of the full field, and subtracted.
The image of S 147 is shown as Fig. 19. A bigger version is to be found in the ING Newletter article describing IPHAS (Drew et al 2005). The improvement is very marked in comparison with the photographic image shown by Van den Bergh et al (1973): the huge increase in dynamic range brings with it a subtlety of detail missing from the old imagery. Furthermore, these data allow smaller areas to be imaged to much higher resolution. But even at this 5 arcsec rebinning the structure is clear. We note that a blow-out is obvious on two sides of the remnant: left and right (to the E and W). The full extent of S 147 is essentially as reported by Van den Bergh et al, at a mean diameter of just over 3<sup>o</sup> โ several times that of the Moon.
At the eastern edge of the image mosaic, and south of centre, lies another prominent but much more compact nebulosity, Sh 2-242. This very bright smudge in Fig. 19 has a mean H$`\alpha `$ diameter of 8.0 arcmin. As a contrast to the very large scale structure of S 147, we show as Fig. 20, a full resolution image of this H ii region.
## 8 Summarising discussion
The main aim of this paper has been to introduce IPHAS, the INT Photometric H$`\alpha `$ Survey of the Northern Galactic Plane. By the beginning of 2005, 55% of the imaging observations had been obtained. It is expected that the survey will reach completion in 2006.
Given that previous surveying of the northern Galactic Plane for emission line objects rarely reached deeper than $`V13`$ (see KW99), while the sensitivity limit of IPHAS is $`r^{}20`$, there is no doubt that a huge domain is being opened up for exploration for the first time. It is difficult to predict the numbers of emission line objects that will be discovered, even now with the survey in progress, because the distribution of such objects along the Galactic Plane is extremely uneven (see KW99). The small number of IPHAS fields discussed in this paper fit in with this impression: no emission line objects were evident in the colour-colour plane for the Taurus field (2540/2540o in section 5 โ but a very modest example was found on closer examination); a handful were evident in the Aquila fields (section 4), and upwards of 20, all fainter than $`r^{}=17`$, were picked up in the MMT/HectoSpec pointing toward LDN 1188 in Cepheus (section 6). Crudely averaging the experience to date, it is likely IPHAS will uncover around 10 emission line objects per square degree in the range $`13<r^{}<20`$ (roughly 2 to 3 per IPHAS field), and hence no less than $``$20,000 altogether.
Through its use of both narrow-band H$`\alpha `$ and broadband $`r^{}`$ and $`i^{}`$ CCD photometry, IPHAS has the capability to pick out H$`\alpha `$ deficit objects โ a possibility typically beyond objective prism spectroscopy, the traditional tool of emission line star hunting. For example, unreddened white dwarfs are easily spotted as blue objects with small or negative $`(r^{}H\alpha )`$, separated from the main stellar locus in the IPHAS $`(r^{}H\alpha ,r^{}i^{})`$ plane. It is a reasonable guess that over a thousand will be discovered in IPHAS data, just as have been discovered in a comparable sky area around the north Galactic cap via the Sloan survey (Kleinman et al 2004). Very red H$`\alpha `$ deficit objects can be either brown dwarfs or carbon stars โ classes of star that both lack the TiO band absorption responsible for raising $`(r^{}H\alpha )`$ in normal M stars โ or they can be very reddened, rare examples of late-type supergiants. Two carbon stars (one probable, the other confirmed) have been reported here.
A further role for IPHAS is that it can help trace the way in which the stellar populations making up the Galactic Plane vary across the northern sky. It has been shown here that not all sightlines look the same in IPHAS colours: of particular note is the sharp contrast between the red-giant deficient Taurus field ($`\mathrm{}=181.7^\mathrm{o}`$, section 5), in the Galactic Anticentre region, versus the Aquila fields ($`\mathrm{}33^\mathrm{o}`$) with their prominent giant-star populations, sampling the inner Galaxy. Before now, rather little has been known about the far reaches of the Milky Way outside the Solar Circle. For example, the recent study of Galactic spiral structure by Russeil (2003) reaches only to $`6`$ kpc, outward from the Sun. The sensitivity of IPHAS is more than adequate for extending our knowledge much further: even relatively humble A0โA3 V stars โ easily picked out around what has been dubbed the early-A reddening line in the IPHAS colour-colour plane โ are potentially accessible out to distances of $``$20 kpc in the direction of the lightly reddened Galactic Anticentre.
We have laid out the character of the $`(r^{}H\alpha ,r^{}i^{})`$ colour-colour plane that is unique to this survey, and have established a grid of simulated colours that will be of use in the analysis of IPHAS observations. In doing this, we have inevitably identified items of work for the future. For example, a job remains to be done to achieve as good a quantitative match between simulations and observations of M stars as is achieved for earlier spectral types. A related problem has been noted by those developing Sloan Digital Sky Survey photometric calibrations (Smith et al 2002). The reward for solving this problem will be the chance to fully realise the potential of IPHAS as an unparalleled resource for the statistical analysis of M-dwarf activity. Another task is to gather a a range of early A star spectra that can be used to achieve better definition of the early-A reddening line and its dependences. As specified here, it is unlikely to be more than $``$0.02 magnitudes from its correct registration in $`(r^{}H\alpha )`$.
Bigger issues to be dealt with are that the photometric calibration of IPHAS data will need to be made uniform across the survey, and that a proper definition of H$`\alpha `$ magnitude zero point (referred to Vega) is needed that can be applied within the pipeline processing. Until these calibration requirements have been met, it will remain necessary to derive colour offsets between observed and simulated data independently for every IPHAS field. Typically, offsets in $`(r^{}i^{})`$ will be small โ but for $`(r^{}H\alpha )`$, the differences between catalogue and simulated colours may vary from $``$0.1 up to $``$0.2 magnitudes. In most cases it is easy to gauge the required offset by matching the synthesised unreddened main sequence track and early-A reddening line to the upper and lower boundaries of the main stellar locus in the colour-colour plane.
Simulations have also been performed that show how the equivalent width threshold for the detection of H$`\alpha `$ emission will change with observed $`(r^{}i^{})`$ colour (section 3.2 and Fig. 6). The photometric accuracy of IPHAS is such that $`(r^{}H\alpha )`$ differences of 0.05โ0.1 are significant down to $`r^{}20`$. Re-expressed in terms of an H$`\alpha `$ equivalent width, and for magnitudes brighter than $``$19, this corresponds to a threshold emission EW of roughly 5 ร
.
It has been shown that the typical morphology of the main stellar locus in the colour-colour plane permits the selection of candidate emission line stars presenting with threshold H$`\alpha `$ emission at low $`(r^{}i^{})`$ only. In practise the IPHAS bright magnitude limit ($`r^{}13`$) has the effect of all but eliminating normal stars with colours bluer than $`(r^{}i^{})0.5`$ from the point source catalogue (note Figs. 3, 9 and 14) โ with the consequence that lightly-reddened subluminous accreting objects, with or without H$`\alpha `$ emission, will usually lie comfortably outside the main stellar locus. IPHAS can therefore be used straightforwardly to identify all such objects (in addition to many non-interacting white dwarfs, found as โdeficitโ objects).
The threshold for H$`\alpha `$ emission high-confidence detection rises from $`10`$ ร
equivalent width at $`(r^{}i^{})1`$ up to $`50`$ ร
at $`(r^{}i^{})2.5`$ โ beyond this the colour-colour plane typically becomes sparsely populated again. This has important implications for the detectability of T Tau and other young emission line objects: based on equivalent width data collected by Reipurth, Pedrosa & Lago (1996, their Table 1 and Fig. 10), one-third to a half of such objects would be immediately identifiable as emission line objects in the IPHAS database at $`E(BV)3`$. At lower reddenings the fraction would be higher. This suggests that the roles for IPHAS with respect to young stellar populations are (i) finding extreme examples anywhere (here we have presented two with H$`\alpha `$ emission EWs of $`200`$ ร
), and (ii) picking out new associations through the detection of its most active members. Once a new association is identified, a more detailed exploration of catalogued IPHAS sources, exploiting the tricks of e.g. apparent magnitude binning, or small area searches, would be likely to uncover further candidate emission line sources.
Evolved high mass stars, with H$`\alpha `$ emission, that find their way into the IPHAS database will do so because they are very distant and significantly reddened (they are otherwise too bright). The more extreme, least well-understood, and therefore more interesting groups (Wolf-Rayet stars, luminous blue variables, B\[e\] stars, yellow hypergiants) usually present with very high EW H$`\alpha `$ emission ($`100`$ ร
and more) and so will not be missed. The most extreme point-source emission line objects of all, compact nebulae, are most likely to appear to be relatively blue in $`(r^{}i^{})`$ or even evade detection in the $`i^{}`$ band, and will have the most extreme values of $`(r^{}H\alpha )`$ possible, i.e. not more than 3 to 3.1, in practice.
This connects naturally to a topic only touched on in this paper โ the exploitation of IPHAS in the study of spatially-resolved nebulae (deferred to a later paper). A search for PNe is in progress, in which several tens of candidates have been identified and a few have been studied spectroscopically. A paper on the study of an intriguing quadrupolar nebula located well-outside the Solar Circle is in preparation (Mampaso et al). Here we have simply drawn attention to the flexibility IPHAS presents for the investigation of a wide range of spatial scales, ranging from arcseconds to several degrees.
Further power to diagnose either particular object types or complete stellar populations will come from pooling IPHAS $`r^{}`$, $`H\alpha `$ and $`i^{}`$ data with data from surveys in different wavebands. In the future of an astronomy conducted via virtual observatories, a survey as comprehensive as IPHAS โ with its particular exploitation of narrowband H$`\alpha `$ data โ will be a major resource. Currently there is an obvious synergy with the all-sky NIR survey 2MASS, although this is limited to the reddened parts of the northern Galactic Plane since 2MASS reaches only to $`K15`$. The gap this leaves should soon be plugged by the UKIDSS Galactic Plane Survey, reaching to $`K=19`$ (see http://www.ukidss.org/). Beyond IPHAS, comprehensive optical surveying of the northern Galaxy using linear detectors is still lacking. However it is interesting to note the recent release by SDSS of uโgโrโiโzโ data on a number of low Galactic latitude fields (Finkbeiner et al 2004).
We finish with a comment on the plans for making IPHAS data available to the community. At the present time there is open access to the reduced images held at CASU, from a year after the images have been processed. Accordingly the images obtained in 2003 can already be accessed by anyone able to reach the CASU web interface (http://apm2.ast.cam.ac.uk/cgi-bin/wfs/dqc.cgi). Access immediately after processing is available to those working in the ING partner countries: the United Kingdom; Spain and The Netherlands. The point source catalogues are to be released in two stages: in the first half of 2006 we aim to release as many as are available, calibrated only at the individual exposure level (as they have been described here); we intend to follow this up with a second release, after the survey is complete, and when a uniform calibration has been established across all fields. We anticipate that the final catalogue will contain photometry on around 80 million point sources, capturing data on roughly 1 for every 1000 stars estimated as existing in the northern Milky Way.
## Acknowledgments
This paper makes use of data from both the Isaac Newton and William Herschel Telescopes, operated on the island of La Palma by the Isaac Newton Group in the Spanish Observatorio del Roque de los Muchachos of the Instituto de Astrofisica de Canarias. The multi-object spectroscopic observations reported here were obtained at the MMT Observatory, a joint facility of the Smithsonian Institution and the University of Arizona. We would like to thank the HectoSpec team for their assistance: in particular, Nelson Caldwell and Perry Berlind for their help with the data acquisition, and Susan Tokarz for the pipeline reduction products. We also acknowledge use of data products from the Two Micron All Sky Survey (2MASS), which is a joint project of the University of Massachusetts and the Infrared Processing and Analysis Center/California Institute of Technology (funded by the USAโs National Aeronautics and Space Administration and National Science Foundation). DS acknowledges a Smithsonian Astrophysical Observatory Clay Fellowship.
|
warning/0506/math0506486.html
|
ar5iv
|
text
|
# Formal deformations of morphisms of associative algebras
## 1. Homotopical deformations
Let $`\gamma :AB\text{Mor}\text{Alg}(๐ค)`$, we will denote it simply by $`\gamma `$. Classically (\[GS3\]) deformations of $`\gamma `$ are described by a functor $`\text{Def}^\gamma :\text{art}(๐ค)\text{Grp}`$, defined by
$$\text{Def}^\gamma (R):=(\text{Mor}^{gr,fl}\text{Alg}(R),\pi _R^{},\gamma ),$$
where the image is the comma category of $`\pi _R^{}`$ over $`\gamma `$, with $`\text{Mor}^{gr,fl}\text{Alg}(R)`$ being the underlying groupoid of the category of morphisms between flat $`R`$-algebras. If we define equivalences between functors from $`\text{art}(๐ค)`$ to Grp as natural transformations, that are objectwise equivalences of groupoids, then $`\text{Def}^\gamma `$ is equivalent to $`(\pi ^{})^1(Id_\gamma )`$, that maps $`R`$ to the pre-image in $`\text{Mor}^{gr,fl}\text{Alg}(R)`$ under $`\pi _R^{}`$ of the identity automorphism of $`\gamma `$. This is an instance of the more general weak equivalences of functors to the category of simplicial groupoids $`\underset{ยฏ}{\text{Grp}}`$.
To define homotopical deformations of $`\gamma `$ we consider it as an object in $`\text{Mor}\text{DGAlg}(๐ค)`$ and extend the parameter category to $`\text{dgart}(๐ค)`$. Weakly equivalent objects should have weakly equivalent deformations (weak equivalences of morphisms are pairs of quasi-isomorphisms). One could redefine $`\text{Def}^\gamma (R)`$ as the pre-image of the identity automorphism of $`\gamma `$ with respect to the functor between the localizations by quasi-isomorphisms of $`\text{Mor}\text{DGAlg}(R)`$ and $`\text{Mor}\text{DGAlg}(๐ค)`$, however this will not capture all of the homotopical structure. Instead we will use the left derived functor of the localization - simplicial localization (\[DK3\]). So homotopical deformations are described by a functor from $`\text{dgart}(๐ค)`$ to $`\underset{ยฏ}{\text{Grp}}`$.
Identifying deformations of weakly equivalent objects we are forced to consider simplicial groupoids up to weak equivalences. In order to be able to do homotopy theory we use a closed model structure on the category of simplicial categories, and then consider the subcategory of simplicial groupoids. Recall that by simplicial categories we mean categories enriched over simplicial sets. If $`\underset{ยฏ}{C}`$ is a simplicial category, we denote by $`\pi _0(\underset{ยฏ}{C})`$ the category with the same set of objects and morphisms being the sets of connected components of the corresponding spaces of maps in $`\underset{ยฏ}{C}`$, and by $`\pi _0(\underset{ยฏ}{F})`$ of a simplicial functor $`\underset{ยฏ}{F}`$, the corresponding functor between $`\pi _0`$ of the categories.
###### Proposition 1.
(\[Ber\]) Let $`\underset{ยฏ}{C_1},\underset{ยฏ}{C_2}\underset{ยฏ}{\text{Cat}}`$ be simplicial categories. Call a functor $`\underset{ยฏ}{F}:\underset{ยฏ}{C_1}\underset{ยฏ}{C_2}`$ a weak equivalence if $`\pi _0(\underset{ยฏ}{F}):\pi _0(\underset{ยฏ}{C_1})\pi _0(\underset{ยฏ}{C_2})`$ is an equivalence of categories and for any $`c_1,c_2\underset{ยฏ}{C}`$ the map of $`\underset{ยฏ}{Hom}(c_1,c_2)`$ to $`\underset{ยฏ}{Hom}(\underset{ยฏ}{F}(c_1),\underset{ยฏ}{F}(c_2))`$ is a weak equivalence of simplicial sets. Call $`\underset{ยฏ}{F}`$ a fibration if for any $`c_1,c_2\underset{ยฏ}{C}`$ the map of $`\underset{ยฏ}{Hom}(c_1,c_2)`$ to $`\underset{ยฏ}{Hom}(\underset{ยฏ}{F}(c_1),\underset{ยฏ}{F}(c_2))`$ is a fibration of simplicial sets, and for any $`c_1\underset{ยฏ}{C_1}`$, $`x\underset{ยฏ}{C_2}`$ and any $`f\underset{ยฏ}{Hom}_0(\underset{ยฏ}{F}(c_1),x)`$, s.t. $`\pi _0(f)`$ is invertible, there is an $`f^{}\underset{ยฏ}{Hom}_0(c_1,c_2)`$, s.t. $`\pi _0(f^{})`$ is invertible and $`\underset{ยฏ}{F}(f^{})=f`$. Weak equivalences and fibrations so defined are a part of the structure of a closed model category on $`\underset{ยฏ}{\text{Cat}}`$.
Note that the weak equivalences in the closed model structure on $`\underset{ยฏ}{\text{Cat}}`$ described in proposition 1 are different from those described in \[Hin1\](5.1.3), where they are required only to induce weak equivalences on the nerves of the categories of connected components. However for simplicial groupoids these definitions of weak equivalences coincide since a functor between groupoids is an equivalence if and only if it induces a weak equivalence of the corresponding nerves.
The category $`\text{dgart}(๐ค)`$ has a subcategory of weak equivalences consisting of morphisms, that induce isomorphisms on cohomology. Let $`Hom(\text{dgart}(๐ค),\underset{ยฏ}{\text{Grp}})`$ be the category of functors from $`\text{dgart}(๐ค)`$ to $`\underset{ยฏ}{\text{Grp}}`$ that map weak equivalences to weak equivalences. Simplicial groupoid describing homotopical deformations of $`\gamma `$ will be an object in this category.
Let $`\underset{ยฏ}{C}`$ be a simplicial category. Following \[DK1\](6.3) we introduce the underlying simplicial groupoid of $`\underset{ยฏ}{C}`$. Recall that simplicial groupoid means a simplicial category whose $`\pi _0`$ is a groupoid.
###### Definition 1.
The homotopy groupoid $`\underset{ยฏ}{\mathrm{C}}^{\mathrm{g}\mathrm{r}}`$ of $`\underset{ยฏ}{\mathrm{C}}`$ is the maximal simplicial subcategory of $`\underset{ยฏ}{\mathrm{C}}`$, which is a simplicial groupoid, i.e. objects of $`\underset{ยฏ}{\mathrm{C}}^{\mathrm{g}\mathrm{r}}`$ are those of $`\underset{ยฏ}{\mathrm{C}}`$, and for any two of them $`\underset{ยฏ}{\mathrm{H}\mathrm{o}\mathrm{m}}_{\underset{ยฏ}{\mathrm{C}}^{\mathrm{g}\mathrm{r}}}(\mathrm{c}_1,\mathrm{c}_2)`$ consists of the connected components of $`\underset{ยฏ}{\mathrm{H}\mathrm{o}\mathrm{m}}_{\underset{ยฏ}{\mathrm{C}}}(\mathrm{c}_1,\mathrm{c}_2)`$, whose classes in $`\mathrm{\pi }_0(\underset{ยฏ}{\mathrm{C}})`$ are invertible.
Let $`L\text{Mor}\text{DGAlg}(R)`$ be the simplicial localization (\[DK3\]) of $`\text{Mor}\text{DGAlg}(R)`$ with respect to quasi-isomorphisms. Any map $`f:R_1R_2`$ in $`\text{dgart}(๐ค)`$ induces a functor $`f^{}`$ from the category of morphisms of algebras over $`R_1`$ to those over $`R_2`$. In general this functor does not preserve weak equivalences, but it does so for weak equivalences between cofibrant objects (\[Hin3\](3.3)). Also we have a functorial cofibrant replacements for objects in $`\text{Mor}\text{DGAlg}(R)`$ (\[Hin3\](7.4.2)), combining $`f^{}`$ with these replacements we get a functor that preserves all weak equivalences and hence induces a functor $`Lf^{}`$ between the corresponding simplicial localizations. If $`f`$ is a weak equivalence itself, $`f^{}`$ is a part of a Quillen equivalence, and hence $`Lf^{}`$ is a weak equivalence between simplicial categories (\[DK1\](3.6)).
###### Definition 2.
Homotopical deformations of $`\mathrm{\gamma }`$ are described by the functor $`\mathrm{L}\text{Def}^\mathrm{\gamma }\mathrm{H}\mathrm{o}\mathrm{m}(\text{dgart}(๐ค),\underset{ยฏ}{\text{Grp}})`$, defined by
$$L\text{Def}^\gamma (R):=\text{h-fib}_{\underset{ยฏ}{\gamma }}(L\pi _R^{}:(L\text{Mor}\text{DGAlg}(R))^{gr}(L\text{Mor}\text{DGAlg}(๐ค))^{gr}),$$
where $`\text{h-fib}_{\underset{ยฏ}{\gamma }}`$ stands for homotopy fiber at $`\underset{ยฏ}{\gamma }`$, and $`\underset{ยฏ}{\gamma }`$ is the image in $`L\text{Mor}\text{DGAlg}(R)`$ of the final object in $`\underset{ยฏ}{\text{Cat}}`$, given by $`\gamma `$ and the trivial simplicial set consisting of $`Id_\gamma `$.
Definition 2 uses simplicial localization, which is hardly computable. In order to describe homotopical deformations effectively we will use the simplicial model structure on $`\text{Mor}\text{DGAlg}(R)`$. Let $`\underset{ยฏ}{\text{DGAlg}}(R)`$ be the simplicial model category (\[Hin3\](4.8)) with the same objects as $`\text{DGAlg}(R)`$ and the spaces of maps defined by
$$\underset{ยฏ}{Hom}_n(A,B):=Hom(A,B_n),A,B\text{DGAlg}(R),B_n:=B_R(R_๐ค\mathrm{\Omega }_n),$$
where $`\mathrm{\Omega }_n`$ is the algebra of polynomial forms on the $`n`$-simplex (\[Bou\](1.)), i.e. it is the free commutative unital $`๐ค`$-algebra generated by $`\{t_0^n,\mathrm{},t_n^n,dt_0^n,\mathrm{},dt_n^n\}`$ with $`\text{deg}(t_i^n)=0`$, $`\text{deg}(dt_i^n)=1`$, and the differential defined by $`d(t_i^n):=dt_i^n`$. The set $`\{\mathrm{\Omega }_n\}_{n0}`$ is a simplicial algebra, with face and degeneracy maps $`\delta _i^n:\mathrm{\Omega }_n\mathrm{\Omega }_{n1}`$, $`\sigma _i^n:\mathrm{\Omega }_n\mathrm{\Omega }_{n+1}`$ given by
$$\delta _i^n(t_j^n)=\{t_{j1}^{n1}\text{ if }i<j,\text{ 0 if }i=j,\text{ }t_j^{n1}\text{ else}\},$$
$$\sigma _i^n(t_j^n)=\{t_{j+1}^{n+1}\text{ if }i<j,\text{ }t_j^{n+1}+t_{j+1}^{n+1}\text{ if }i=j,\text{ }t_j^{n+1}\text{ else}\}.$$
Since category of morphisms of algebras is the category of diagrams in a cofibrantly generated simplicial model category, it also has a simplicial model structure (\[Hir\](11.7) and \[ShSh\] for the cofibrant generation), defined as follows. Let $`\gamma :AB`$, $`\gamma ^{}:A^{}B^{}`$ be two objects of $`\text{Mor}\text{DGAlg}(R)`$. A morphism from $`\gamma `$ to $`\gamma ^{}`$ is a pair $`\varphi :AA^{}`$, $`\psi :BB^{}`$. It is a weak equivalence (fibration) if both $`\varphi `$ and $`\psi `$ are. Let $`\text{Mor}\underset{ยฏ}{\text{DGAlg}}(R)`$ be the simplicial model category with the same objects as $`\text{Mor}\text{DGAlg}(R)`$ and the mapping spaces defined by
$$\underset{ยฏ}{Hom}_n(\gamma ,\gamma ^{}):=Hom(\gamma ,\gamma _n^{}),\gamma _n^{}:=(\gamma ^{}_RId):A_n^{}B_n^{}.$$
Classical definition was equivalent to a fiber in the subcategory of flat $`R`$-algebras and isomorphisms. In homotopy theory flatness condition is expressed by cofibrant objects and isomorphisms by weak equivalences between cofibrant objects. Let $`\text{Mor}^{w,c}\underset{ยฏ}{\text{DGAlg}}`$ be the simplicial subcategory of $`\text{Mor}\underset{ยฏ}{\text{DGAlg}}(R)`$ consisting of cofibrant objects, and the morphisms being
$$\underset{ยฏ}{Hom}_n^{w,c}(\gamma ,\gamma ^{}):=\{\text{weak equivalences from }\gamma \text{ to }\gamma _n^{}\}.$$
Let $`\underset{ยฏ}{\pi ^{}}`$ be the simplicial extension of $`\pi ^{}`$. Let $`๐\gamma `$ be a cofibrant replacement of $`\gamma `$. The trivial simplicial set $`\{(Id_{๐\gamma }_RId):๐\gamma ๐\gamma _n\}_{n0}`$ together with $`๐\gamma `$ itself is an image in $`\text{Mor}^{w,c}\underset{ยฏ}{\text{DGAlg}}(R)`$ of the final object on $`\underset{ยฏ}{\text{Cat}}`$. We will denote it by $`\underset{ยฏ}{๐\gamma }`$.
###### Definition 3.
(\[Hin1\]) Homotopical deformations of $`\gamma `$ are described by the functor $`\underset{ยฏ}{\text{Def}}^\gamma Hom(\text{dgart}(๐ค),\underset{ยฏ}{\text{Grp}})`$, defined by
$$\underset{ยฏ}{\text{Def}}^\gamma (R):=\text{h-fib}_{\underset{ยฏ}{๐\gamma }}(\underset{ยฏ}{\pi ^{}}_R:\text{Mor}^{w,c}\underset{ยฏ}{\text{DGAlg}}(R)\text{Mor}^{w,c}\underset{ยฏ}{\text{DGAlg}}(๐ค)),$$
where $`\text{h-fib}_{\underset{ยฏ}{๐\gamma }}`$ stands for homotopy fiber at $`\underset{ยฏ}{๐\gamma }`$.
As in the case of definition 2, $`\underset{ยฏ}{\text{Def}}^\gamma `$ is an object of $`Hom(\text{dgart}(๐ค),\underset{ยฏ}{\text{Grp}})`$ because for any morphism $`f:R_1R_2`$ in $`\text{dgart}(๐ค)`$ the functor $`f^{}:\text{Mor}\text{DGAlg}(R_1)\text{Mor}\text{DGAlg}(R_2)`$ is a left Quillen functor, i.e. maps cofibrations (weak equivalences between cofibrants) to the like. Hence it induces a functor $`\underset{ยฏ}{f^{}}:\text{Mor}^{w,c}\underset{ยฏ}{\text{DGAlg}}(R_1)\text{Mor}^{w,c}\underset{ยฏ}{\text{DGAlg}}(R_2)`$. Finally $`\underset{ยฏ}{\text{Def}}^\gamma `$ maps weak equivalences in $`\text{dgart}(๐ค)`$ to weak equivalences because Quillen equivalences induce weak equivalences of function complexes (\[Hir\](17.4.16)).
Since $`๐`$ is functorial and there is a natural transformation $`๐Id_{\text{Mor}\text{DGAlg}(R)}`$, it induces a weak self-equivalence on $`L\text{Mor}\text{DGAlg}(R)`$, hence $`L\text{Def}^\gamma `$ and $`L\text{Def}^{๐\gamma }`$ are weakly equivalent (homotopical invariance of h-fib). According to \[DK2\](4.8) and \[DK1\](2.2) $`L\text{Mor}\text{DGAlg}(R)`$ and $`\text{Mor}\underset{ยฏ}{\text{DGAlg}}(R)`$ are naturally weakly equivalent. Therefore the corresponding homotopical groupoids are weakly equivalent, and since the โ2 out of 3โ axiom implies that $`\text{Mor}^{w,c}\underset{ยฏ}{\text{DGAlg}}(R)`$ is the homotopical groupoid on the cofibrant objects, we conclude that $`L\text{Def}^{๐\gamma }`$ and $`\underset{ยฏ}{\text{Def}}^\gamma `$ are weakly equivalent. So these two definitions of homotopical deformations coincide.
###### Lemma 1.
The fiber of $`\underset{ยฏ}{\pi ^{}}_R:\text{Mor}^{w,c}\underset{ยฏ}{\text{DGAlg}}(R)\text{Mor}^{w,c}\underset{ยฏ}{\text{DGAlg}}(๐ค)`$ at $`\underset{ยฏ}{๐\gamma }`$ is weakly equivalent to the homotopy fiber.
Proof: As it was noted before, for simplicial groupoids the notion of weak equivalences from proposition 1 coincides with that of \[Hin1\](5.). Also fibrations between simplicial groupoids according to proposition 1 are fibrations according to \[Hin1\], indeed right lifting property with respect to adding an ingoing or an outgoing arrow (\[Hin1\](5.1.4)) is held by fibrations between simplicial groupoids according to proposition 1, because those functors map $`\underset{ยฏ}{Hom}_0`$โs surjectively. Therefore according to \[Hin1\](5.3.2) the nerve functor maps weak equivalences and fibrations (as in proposition 1) between simplicial groupoids to the like between simplicial sets.
$`\underset{ยฏ}{\pi ^{}}_R:\text{Mor}^{w,c}\underset{ยฏ}{\text{DGAlg}}(R)\text{Mor}^{w,c}\underset{ยฏ}{\text{DGAlg}}(๐ค)`$ is a fibration of simplicial groupoids (\[Hin1\](4.2.1)). Indeed, every morphism in $`\text{Mor}^{w,c}\underset{ยฏ}{\text{DGAlg}}(๐ค)`$ has a pre-image in $`\text{Mor}^{w,c}\underset{ยฏ}{\text{DGAlg}}(R)`$, for example $`R`$-linear extension. If $`\gamma _1,\gamma _2\text{Mor}^{w,c}\underset{ยฏ}{\text{DGAlg}}(R)`$, then $`\underset{ยฏ}{\pi ^{}}:\underset{ยฏ}{Hom}(\gamma _1,\gamma _2)\underset{ยฏ}{Hom}(\underset{ยฏ}{\pi ^{}}(\gamma _1),\underset{ยฏ}{\pi ^{}}(\gamma _2))`$ is a fibration because it coincides with the map $`\underset{ยฏ}{Hom}(\gamma _1,\gamma _2)\underset{ยฏ}{Hom}(\gamma _1,\gamma _2_R๐ค)`$ given by the morphism $`\gamma _2\gamma _2_R๐ค`$, and this last map is a fibration whereas $`\gamma _1`$ is a cofibrant object (\[Hir\](9.3.1)).
The nerve functor has a left adjoint (\[Hin1\](5.3.1)), hence it preserves limits and therefore maps the nerve of the fiber of a functor to the fiber of the nerve of that functor. Since SSet is a proper model category, taking fibers of fibrations is equivalent to taking homotopy fibers (\[Hir\](13.4.6)). Hence we conclude that the canonical map from the homotopy fiber of $`\underset{ยฏ}{\pi ^{}}_R:\text{Mor}^{w,c}\underset{ยฏ}{\text{DGAlg}}(R)\text{Mor}^{w,c}\underset{ยฏ}{\text{DGAlg}}(๐ค)`$ to the fiber induces weak equivalence of the nerves, hence it is a weak equivalence of the groupoids themselves according to \[DK3\](9.6) (we can consider the two fibers to have the same set of objects because the functor between them induces equivalence on the underlying categories of connected components). $`\mathrm{}`$
Lemma 1 reduces the question of computing deformation groupoid to computing a fiber of a functor. As it is explained in \[Hin2\], (the nerve of) such functor can be described by the nerve of a dg Lie algebra. Usually one looks for such dg Lie algebra appearing naturally from the structure to be deformed. In case of deformation of one algebra it would be the dg Lie algebra of derivations of a cofibrant replacement. Equivalences between solutions of the Maurer-Cartan equation in this dg Lie algebra are represented by infinitesimal automorphisms of the cofibrant replacement, and when we consider deformations over graded Artin rings, all elements of the dg Lie algebra represent (graded) automorphisms. However, in case of deformations of morphisms, not all derivations are infinitesimal automorphisms, and the algebra is not Lie but an $`_{\mathrm{}}`$-algebra. In such more general situations we can have an $`_{\mathrm{}}`$-algebra $`๐ค`$ and an $`_{\mathrm{}}`$-subalgebra $`๐ฅ`$, that represents infinitesimal automorphisms. Definition of the Deligne groupoid of a pair $`(๐ฅ,๐ค)`$ is given in definition 8.
###### Problem 1.
Let $`\gamma `$ be a morphism of dg associative algebras. The deformation problem defined by $`\gamma `$ is to find a pair of $`_{\mathrm{}}`$-algebras $`(๐ฅ,๐ค)`$, s.t. the corresponding simplicial Deligne groupoid $`\underset{ยฏ}{\text{Del}}^\gamma `$ is weakly equivalent to $`\underset{ยฏ}{\text{Def}}^\gamma `$.
In case $`\gamma `$ is a morphism of non-positively graded algebras, solution to this problem is given in section 3. We will do it by expressing this deformation problem in the language of $`๐_{\mathrm{}}`$-structures. That will allow us to compare the result with the Lie algebra on cohomology defined in \[GS2\] and \[GS3\].
## 2. Deformation of $`๐_{\mathrm{}}`$-structures
### 2.1. Definition of $`๐_{\mathrm{}}`$-structures
We will define an $`๐_{\mathrm{}}`$-structure on a pair of modules as a morphism between two coassociative codifferential coalgebras. This requires a choice of identification of the tensor algebra on a module with the one on its suspension. We make the following choice (\[Pen\](5.)). Let $`M\text{dgmod}(R)`$, then $`(sM)^k:=M^{k+1}`$. Let $`T(M):=\underset{n>0}{}M^{_{R}^{}{}_{}{}^{n}}`$, then we define
$$s:T(M)T(sM),s(x_1_R\mathrm{}_Rx_n):=(1)^ฯตsx_1_R\mathrm{}_Rsx_n,$$
$$ฯต=(n1)x_1+(n2)x_2+\mathrm{}+x_{n1}.$$
Recall that by $`\widehat{\alpha }`$ we mean the coderivation of a cofree coalgebra, cogenerated by a map $`\alpha `$, and by $`\stackrel{~}{\gamma }`$ we mean the coalgebra morphism, cogenerated by $`\gamma `$.
###### Definition 4.
(e.g. \[Pen\](5.)) Let $`R\text{dgart}(๐ค)`$. Let $`A,B\text{dgmod}(R)`$. An $`๐_{\mathrm{}}`$-structure on the pair $`(A,B)`$ is a triple $`(\widehat{\alpha },\widehat{\beta },\stackrel{~}{\gamma })`$, where $`\widehat{\alpha },\widehat{\beta }`$ are coderivations of degree 1 on the cofree coassociative coalgebras $`T(sA),T(sB)`$ respectively and $`\stackrel{~}{\gamma }:T(sA)T(sB)`$ is a degree 0 morphism of coalgebras, such that
$$(\widehat{\alpha }+\widehat{\delta }_A)^2=0,(\widehat{\beta }+\widehat{\delta }_B)^2=0,\stackrel{~}{\gamma }(\widehat{\alpha }+\widehat{\delta }_A)(\widehat{\beta }+\widehat{\delta }_B)\stackrel{~}{\gamma }=0,$$
where $`\widehat{\delta }_A,\widehat{\delta }_B`$ are the coderivations cogenerated by differentials on $`A,B`$.
Since $`T(sA),T(sB)`$ are cofree coalgebras, $`\widehat{\alpha },\widehat{\beta },\stackrel{~}{\gamma }`$ are determined by their corestrictions to cogenerators
$$\{\alpha _n:(sA)^{_{R}^{}{}_{}{}^{n}}sA\}_{n>0},\{\beta _n:(sB)^{_{R}^{}{}_{}{}^{n}}sB\}_{n>0},\{\gamma _n:(sA)^{_{R}^{}{}_{}{}^{n}}sB\}_{n>0}.$$
Using identification $`s`$ we can translate $`\alpha ,\beta ,\gamma `$ to tensor operations on $`A,B`$, namely
$$\widehat{\mu }:=s^1\widehat{\alpha }s,\widehat{\nu }:=s^1\widehat{\beta }s,\stackrel{~}{\varphi }:=s^1\stackrel{~}{\gamma }s.$$
Applying this to the corestrictions we get the sequences of Hochschild cochains $`\{\mu _nC^n(A,A)\}_{n>0}`$, $`\{\nu _nC^n(B,B)\}_{n>0}`$, $`\{\varphi _nC^n(A,B)\}_{n>0}`$, such that in the case $`A,B`$ have trivial differentials
$$\text{deg}(\mu _n)=\text{deg}(\nu _n)=2n,\text{deg}(\varphi _n)=1n,$$
$$\underset{\underset{0ink}{k+l=n+1}}{\mathrm{\Sigma }}(1)^{ฯต_1}\mu _l(a_1\mathrm{}\mu _k(a_{i+1}\mathrm{}a_{i+k})\mathrm{}a_n)=0,ฯต_1=i(k1)+k(nk+a_1+\mathrm{}+a_i),$$
$$\underset{\underset{0ink}{k+l=n+1}}{\mathrm{\Sigma }}(1)^{ฯต_2}\nu _l(b_1\mathrm{}\nu _k(b_{i+1}\mathrm{}b_{i+k})\mathrm{}b_n)=0,ฯต_2=i(k1)+k(nk+b_1+\mathrm{}+b_i),$$
$$\underset{1inm}{\underset{m+k=n+1}{\mathrm{\Sigma }}}(1)^{ฯต_3}\varphi _k(a_1\mathrm{}\mu _m(a_{i+1}\mathrm{}a_{i+m})\mathrm{}a_n)=$$
$$=\underset{i_1+\mathrm{}+i_r=n}{\underset{1rn}{\mathrm{\Sigma }}}(1)^{ฯต_4}\nu _r(\varphi _{i_1}(a_1\mathrm{}a_{i_1})\mathrm{}\varphi _{i_r}(a_{ni_r}\mathrm{}a_n)),$$
$$ฯต_3=i(m1)+m(nm)+m(a_1+\mathrm{}+a_i),ฯต_4=\underset{1t<r}{\mathrm{\Sigma }}\varphi _{i_{t+1}}\underset{s=1}{\overset{i_1+\mathrm{}+i_t}{\mathrm{\Sigma }}}a_s+\underset{1t<r}{\mathrm{\Sigma }}(rt)\varphi _t,$$
this is the usual definition of $`๐_{\mathrm{}}`$-algebras and $`๐_{\mathrm{}}`$-morphisms (e.g. \[Kel\](3.1,3.4)).
### 2.2. $`_{\mathrm{}}`$-algebra on the cochain complex
The right hand side of the last equation contains compositions of the type $`\nu _n(\varphi _{i_1}_R\mathrm{}_R\varphi _{i_n})`$. It is these compositions that make the Hochschild complex of an $`๐_{\mathrm{}}`$-structure not a Lie algebra but an $`_{\mathrm{}}`$-algebra.
###### Definition 5.
(e.g. \[Pen\](6.)) Let $`R\text{dgart}`$, $`M\text{dgmod}(R)`$. The structure of an $`_{\mathrm{}}`$-algebra on $`M`$ is given by a degree 1 coderivation $`\widehat{d}`$ on the cofree cocommutative non-counital coalgebra (cofree in the category of connected coalgebras, see e.g. \[LM\] page 2150) cogenerated by the suspension $`sM`$ of $`M`$, such that $`(\widehat{d}+\widehat{\delta }_M)^2=0`$, where $`\widehat{\delta }_M`$ is the coderivation cogenerated by the differential on $`M`$.
Since a cocommutative coalgebra is also coassociative there is a canonical embedding of it into $`T(sM)`$, given by the universal property of $`T(sM)`$. Let $`S(sM)`$ denote this coassociative sub-coalgebra of $`T(sM)`$. Explicitly
$$S(sM)(sM)^{_{R}^{}{}_{}{}^{n}}=<\underset{๐}{\mathrm{\Sigma }}(1)^{ฯต(\sigma ;sx_1,\mathrm{},sx_n)}sx_{\sigma (1)}_R\mathrm{}_Rsx_{\sigma (n)}>,$$
where the r.h.s. is the $`R`$-submodule generated by symmetrizations of all elements of $`(sM)^{_{R}^{}{}_{}{}^{n}}`$, and $`ฯต(\sigma ;sx_1,\mathrm{},sx_n)`$ is the usual sign of a permutation. The following lemma, generalizing the pre-Lie algebra technique (\[GS1\](10.1)), is obvious (\[vdL\](3.8)).
###### Lemma 2.
Let $`M\text{dgmod}(R)`$ and let $`\widehat{d}`$ be a degree 1 coderivation on $`T(sM)`$, that commutes with $`\widehat{\delta }_M`$. If $`\widehat{d}^2`$ vanishes on $`S(sM)`$, it defines the structure of an $`_{\mathrm{}}`$-algebra on $`M`$.
We will denote $`_{\mathrm{}}`$-algebra $`(S(sM),\widehat{d})`$ by $`(M,\widehat{d})`$. From now on we fix $`A,B\text{dgmod}(๐ค)`$. The $`_{\mathrm{}}`$-algebra, that describes deformations of $`๐_{\mathrm{}}`$-structures on the pair $`(A,B)`$ is built on
$$๐ค:=Hom(T(sA),sA)Hom(T(sB),sB)s^1Hom(T(sA),sB).$$
Differentials on $`A,B`$ induce codifferentials on $`T(sA),T(sB)`$, and we consider $`๐ค`$ together with the induced differential. We denote an element $`g`$ of $`๐ค`$ by $`\alpha +\beta +s^1\gamma `$. Then $`\widehat{\alpha }`$, $`\widehat{\beta }`$ denote the coderivations on $`T(sA)`$, $`T(sB)`$, cogenerated by $`\alpha `$, $`\beta `$, and $`\stackrel{~}{\gamma }`$ denotes the morphism of coalgebras, cogenerated by $`\gamma `$.
###### Definition 6.
Let the coderivation $`\widehat{d}`$ on $`T(s๐ค)`$ be defined by its corestriction to cogenerators as follows
$$d:=\chi _A+\chi _B+\lambda +\underset{m1}{\mathrm{\Sigma }}\rho _m.$$
$$\chi _A(s\alpha _1_๐คs\alpha _2):=(1)^{\alpha _1}s(\alpha _1\widehat{\alpha }_2),\chi _B(s\beta _1_๐คs\beta _2):=(1)^{\beta _1}s(\beta _1\widehat{\beta }_2),$$
$$\lambda (\gamma _๐คs\alpha ):=(1)^\gamma \gamma \widehat{\alpha },\rho _m(s\beta _๐ค\gamma _1_๐ค,\mathrm{},_๐ค\gamma _m):=\beta (\gamma _1_๐ค\mathrm{}_๐ค\gamma _m),$$
where $`m1`$ and $`(\gamma _1_๐ค\mathrm{}_๐ค\gamma _m)`$ is the linear map $`T(sA)(sB)^{_{๐ค}^{}{}_{}{}^{m}}`$ with the sign, given by the Koszul sign rule. On the rest of $`T(s๐ค)`$ the maps $`\chi _A,\chi _B,\lambda ,\{\rho _m\}_{m1}`$ are defined to be zero.
###### Proposition 2.
Degree of $`\widehat{d}`$ is 1 and its square vanishes on $`S(s๐ค)`$.
Proof: Composition of operations has degree zero and the l.h.s. of the defining equations of $`\chi `$, $`\lambda `$, $`\rho `$ have 1 more application of the suspension than the corresponding r.h.s., therefore $`d`$ and hence $`\widehat{d}`$ have degree 1.
Since $`\text{deg}(\widehat{d})=1`$, $`\widehat{d}^2`$ is a coderivation. Therefore it is enough to check that its corestriction to the cogenerators of $`S(s๐ค)`$ vanishes. From the definition it follows that this corestriction is
$$\chi _A\widehat{\chi }_A+\chi _B\widehat{\chi }_B+\lambda \widehat{\chi }_A+\lambda \widehat{\lambda }+\lambda \widehat{\rho }_k+\rho _k\widehat{\lambda }+\rho _k\widehat{\chi }_B+\rho _l\widehat{\rho }_m,$$
the rest of the compositions vanish identically on $`T(s๐ค)`$. We will prove that the above sum is zero on $`S(s๐ค)`$ by dividing it into 4 summands.
* $`\underset{ยฏ}{\chi _A\widehat{\chi }_A+\chi _B\widehat{\chi }_B}`$
$$(\chi _A\widehat{\chi }_A)(s\alpha _1_๐คs\alpha _2_๐คs\alpha _3)=(1)^{\alpha _1+(\alpha _1+\alpha _2)}s((\alpha _1\widehat{\alpha }_2)\widehat{\alpha }_3)+$$
$$+(1)^{(\alpha _1+1)+\alpha _2+\alpha _1}s(\alpha _1(\widehat{\alpha }_2\widehat{\alpha }_3)).$$
Looking at the signs one sees that the part that is not zero is when $`\alpha _3`$ is not โpluggedโ into $`\alpha _2`$. But that is taken care of by interchanging $`\alpha _2`$ and $`\alpha _3`$ (with the correct signs). So $`\chi _A\widehat{\chi }_A`$ vanishes on the symmetrization of $`s\alpha _1_๐คs\alpha _2_๐คs\alpha _3`$ and therefore on the whole of $`S(s๐ค)`$. The same for $`\chi _B\widehat{\chi }_B`$.
* $`\underset{ยฏ}{\lambda \widehat{\chi }_A+\lambda \widehat{\lambda }}`$
$$(\lambda \widehat{\chi }_A)(\gamma _๐คs\alpha _1_๐คs\alpha _2)=(1)^{\gamma +\alpha _1+\gamma +1}\gamma (\widehat{\alpha }_1\widehat{\alpha }_2),$$
$$(\lambda \widehat{\lambda })(\gamma _๐คs\alpha _1_๐คs\alpha _2)=(1)^{\gamma +1+(\gamma +\alpha _1)+1}(\gamma \widehat{\alpha }_1)\widehat{\alpha }_2.$$
As in the step 1. the part that is not zero vanishes after symmetrization, i.e. $`\lambda \widehat{\chi }_A+\lambda \widehat{\lambda }`$ is zero on $`\gamma _๐คs\alpha _1_๐คs\alpha _2+(1)^{(\alpha _1+1)(\alpha _2+1)}\gamma _๐คs\alpha _2s\alpha _1`$. On the rest of the permutations of $`\gamma _๐คs\alpha _1_๐คs\alpha _2`$ it vanishes by definition of the maps $`\chi _A,\lambda `$.
* $`\underset{ยฏ}{\lambda \widehat{\rho }_m+\rho _m\widehat{\lambda }}`$
$$(\lambda \widehat{\rho }_m)(s\beta _๐ค\gamma _1_๐ค\mathrm{}_๐ค\gamma _ms\alpha )=(1)^r(\beta (\gamma _1_๐ค\mathrm{}_๐ค\gamma _m))\widehat{\alpha },$$
$$(\lambda \widehat{\rho }_m)(s\beta _๐ค\gamma _1_๐ค\mathrm{}_๐ค\gamma _i_๐คs\alpha _๐ค\gamma _{i+1}_๐ค\mathrm{}_๐ค\gamma _m)=0ifi<m,$$
$$(\rho _m\widehat{\lambda })((1)^{(\alpha +1)(\gamma _{i+1}+\mathrm{}+\gamma _m)}s\beta _๐ค\gamma _1_๐ค\mathrm{}_๐ค\gamma _i_๐คs\alpha _๐ค\gamma _{i+1}_๐ค\mathrm{}_๐ค\gamma _m)=$$
$$=(1)^t\beta (\gamma _1_๐ค\mathrm{}_๐ค(\gamma _i\widehat{\alpha })_๐ค\mathrm{}_๐ค\gamma _m),$$
where
$$r=\beta +\gamma _1+\mathrm{}+\gamma _m+1,$$
$$t=\beta +1+\gamma _1+\mathrm{}+\gamma _{i1}+\gamma _i+1+(\alpha +1)(\gamma _{i+1}+\mathrm{}+\gamma _m).$$
The difference is $`\alpha (\gamma _{i+1}+\mathrm{}+\gamma _m)+1`$, but $`(1)^{\alpha (\gamma _{i+1}+\mathrm{}+\gamma _m)}`$ is exactly the sign that appears when we apply first $`\widehat{\alpha }`$ and then $`\gamma _1_๐ค\mathrm{}_๐ค\gamma _m`$ or $`\gamma _1_๐ค\mathrm{}_๐ค(\gamma _i\widehat{\alpha })_๐ค\mathrm{}_๐ค\gamma _m`$ directly. It means that $`\lambda \widehat{\rho }_m+\rho _m\widehat{\lambda }`$ vanishes on the symmetrization of $`s\beta _๐ค\gamma _1_๐ค\mathrm{}_๐ค\gamma _ms\alpha `$.
* $`\underset{ยฏ}{\rho _m\widehat{\chi }_B+\rho _l\widehat{\rho }_n}`$
$$(\rho _m\widehat{\chi }_B)(s\beta _1_๐คs\beta _2_๐ค\gamma _1_๐ค\mathrm{}_๐ค\gamma _m)=(1)^{\beta _1}(\beta _1\widehat{\beta }_2)(\gamma _1_๐ค\mathrm{}_๐ค\gamma _m),$$
$$(\rho _m\widehat{\chi }_B)(s\beta _1_๐ค\gamma _1_๐ค\mathrm{}_๐ค\gamma _i_๐คs\beta _2_๐ค\mathrm{}_๐ค\gamma _m)=0ifi>0,$$
$$(\rho _l\widehat{\rho }_n)((1)^{(\beta _2+1)(\gamma _1+\mathrm{}+\gamma _i)}s\beta _1_๐ค\gamma _1_๐ค\mathrm{}_๐ค\gamma _i_๐คs\beta _2_๐ค\mathrm{}_๐ค\gamma _m)=$$
$$=(1)^r\beta _1(\gamma _1_๐ค\mathrm{}\gamma _i_๐ค(\beta _2(\gamma _{i+1}_๐ค\mathrm{}_๐ค\gamma _{i+n}))_๐ค\mathrm{}_๐ค\gamma _m),$$
where
$$r=(\beta _2+1)(\gamma _1+\mathrm{}+\gamma _i)+\beta _1+1+\gamma _1+\mathrm{}+\gamma _i,l+n=m+1.$$
By definition of $`\widehat{\beta }_2`$, application of $`\rho _n\widehat{\chi }_B`$ as in the first formula is a sum and its summands are exactly the results of application of $`\rho _l\widehat{\rho }_m`$ as in the last formula. Then if the signs are opposite we would conclude that $`\rho _m\widehat{\chi }_B+\rho _l\widehat{\rho }_n`$ vanishes on the symmetrization of $`s\beta _1_๐คs\beta _2_๐ค\gamma _1_๐ค\mathrm{}_๐ค\gamma _m`$ (and therefore on the whole of $`S(s๐ค)`$). As it is written, the difference in signs is
$$(1)^{\beta _2(\gamma _1+\mathrm{}+\gamma _i)+1},$$
but this is exactly opposite to the difference in signs that appears if we apply first $`\gamma _1_๐ค\mathrm{}_๐ค\gamma _m`$ and then $`\beta _2`$ at position $`i+1`$ or we apply
$$\gamma _1_๐ค\mathrm{}_๐ค(\beta _2(\gamma _{i+1}_๐ค\mathrm{}_๐ค\gamma _{i+n}))_๐ค\mathrm{}_๐ค\gamma _m$$
directly.
$`\mathrm{}`$
### 2.3. Deligne groupoid
Proposition 2 together with lemma 2 imply that $`(๐ค,d)`$ is an $`_{\mathrm{}}`$-algebra ($`dHom(S(s๐ค),sg)`$). Indeed operations in the definition of $`d`$ are defined using composition of maps. The operation of composition is a cocycle for the differential on $`๐ค`$, hence $`d`$ commutes with that differential. We will denote the part of $`d`$ lying in $`Hom((s๐ค)^{_{๐ค}^{}{}_{}{}^{m}},s๐ค)`$ by $`d_m`$.
Let $`R\text{dgart}(๐ค)`$, let $`(๐ค_๐คR,d)`$ be the $`R`$-linear extension of $`(๐ค,d)`$, it is an $`_{\mathrm{}}`$-algebra over $`R`$. We will denote the corestriction of $`d`$ to $`s๐ค_๐คR`$ again by $`\underset{m>0}{\mathrm{\Sigma }}d_m`$.
In order to describe solutions of the Maurer-Cartan equation as points in a coalgebra we have to complete $`S(s๐ค_๐คR)`$. Define $`\overline{S}(s๐ค_๐คR):=\underset{m>0}{\mathrm{\Sigma }}(S(s๐ค(R))(s๐ค_๐คR)^{_{R}^{}{}_{}{}^{m}})`$, it is a cocommutative coalgebra. A point in $`\overline{S}(s๐ค_๐คR)`$ is an element $`\overline{g}_R`$, s.t. $`\mathrm{\Delta }(\overline{g}_R)=\overline{g}_R_R\overline{g}_R`$, where $`\mathrm{\Delta }`$ is the comultiplication, this name comes from considering $`_{\mathrm{}}`$-algebras as formal dg manifolds (\[Kon\]).
Clearly $`C^{}(A_๐คR,A_๐คR)C^{}(B_๐คR,B_๐คR)C^{}(A_๐คR,B_๐คR)`$ is graded by the number of the arguments of a cochain, $`๐ค_๐คR`$ is the completion with respect to the associated filtration, and $`d`$ is continuous with respect to this filtration (because it is given by composition of the cochains). Therefore if we denote $`\overline{d}:=\underset{m>0}{\mathrm{\Sigma }}d_m`$, it is well defined as a function on $`\overline{S}(s๐ค_๐คR)`$ with values in $`s๐ค_๐คR`$, indeed values of $`d_m`$, for $`m>2`$, have at least $`m`$ arguments.
###### Definition 7.
A solution of the Maurer-Cartan equation (MCE) in $`(๐ค_๐ค\mathrm{R},\mathrm{d})`$ is a degree 0 element $`\mathrm{s}\mathrm{g}_\mathrm{R}`$, such that the following equation holds
$$\delta _R(sg_R)+\underset{m>0}{\mathrm{\Sigma }}d_m((sg_R)^{_{R}^{}{}_{}{}^{m}})=0,$$
where $`\delta _R`$ is the differential on $`๐ค_๐คR`$.
Note that in \[Kon\](4.3) in the definition of the Maurer-Cartan equation for $`_{\mathrm{}}`$-algebras there is a coefficient $`\frac{1}{m!}`$ before $`d_m`$. We do not have them here because the canonical embedding of the cofree cocommutative coalgebra cogenerated by $`s๐ค_๐คR`$ into the cofree coassociative one maps $`(sg_R)^{_R^m}`$ to $`m!(sg_R)^{_{R}^{}{}_{}{}^{m}}`$, when $`g_R`$ is odd.
Since $`\overline{S}(s๐ค_๐คR)`$ is the completion of a cofree coalgebra, every degree 0 point is $`\overline{g}_R=\underset{m>0}{\mathrm{\Sigma }}(sg_R)^{_{R}^{}{}_{}{}^{m}}`$, for some $`sg_Rs๐ค_๐คR`$ of degree 0, and for $`sg_R`$ to be a solution of MCE is equivalent to $`\overline{g}_R`$ being a cocycle for $`\overline{d}+\delta _R`$. We will sometimes represent solutions of MCE by the corresponding points.
###### Proposition 3.
There is a bijection between the set of $`๐_{\mathrm{}}`$-structures on the pair $`A_๐คR,B_๐คR`$ and the set of solutions of MCE in $`(๐ค_๐คR,d)`$.
Proof: Let $`\overline{g}_R=\underset{m>0}{\mathrm{\Sigma }}(sg_R)^{_{R}^{}{}_{}{}^{m}}`$ be a solution of MCE. Then $`g_R=\alpha _R+\beta _R+s^1\gamma _R`$, where $`\alpha _RHom(T(sA_๐คR),sA_๐คR)`$, $`\beta _RHom(T(sB_๐คR),sB_๐คR)`$ of degree 1, $`\gamma _RHom(T(sA_๐คR),sB_๐คR)`$ of degree 0, and they satisfy $`(\widehat{\alpha }_R+\widehat{\delta }_A)^2=0`$, $`(\widehat{\beta }_R+\widehat{\delta }_B)^2=0`$, $`\gamma _R(\widehat{\alpha }_R+\widehat{\delta }_A)(\beta _R+\widehat{\delta }_B)\stackrel{~}{\gamma }_R=0`$ (recall that $`\widehat{\alpha },\widehat{\beta }`$ are the coderivations, $`\stackrel{~}{\gamma }`$ is the coalgebra morphism, cogenerated by $`\alpha ,\beta ,\gamma `$ respectively), where $`\widehat{\delta }_A`$ is the coderivation, cogenerated by the differential on $`sA_๐คR`$. So they comprise an $`๐_{\mathrm{}}`$-structure on $`A_๐คR,B_๐คR`$. Conversely, any such $`๐_{\mathrm{}}`$-structure consists of an element of
$$Hom(T(sA_๐คR),sA_๐คR)Hom(T(sB_๐คR),sB_๐คR)s^1Hom(T(sA_๐คR),sB_๐คR),$$
and the equations of definition 4 translate to the MCE. $`\mathrm{}`$
Solutions of MCE in $`(๐ค_๐คR,d)`$ represent all of $`๐_{\mathrm{}}`$-structures on $`A_๐คR,B_๐คR`$. We are interested in those, whose reduction modulo $`\text{m}_R`$ is the given one $`\gamma `$ ($`\text{m}_R`$ stands for the maximal ideal in $`R`$). Such $`\gamma `$ is represented by a solution $`g`$ of MCE in $`(๐ค,d)`$, and a general procedure associates to it a new $`_{\mathrm{}}`$-algebra $`(๐ค,d^g)`$, that we will use to represent $`๐_{\mathrm{}}`$-structures with the correct reduction modulo $`\text{m}_R`$. This is the $`_{\mathrm{}}`$-version of the usual technique in dg Lie algebra: changing the differential by adding to it a bracket with an odd cocycle. The following lemma describes how a solution of MCE can be used to deform the $`_{\mathrm{}}`$-algebra.
###### Lemma 3.
Let $`(M,d=\underset{n>0}{\mathrm{\Sigma }}d_n)`$ be an $`_{\mathrm{}}`$-algebra ($`M\text{dgmod}(R)`$ for some $`R\text{dgart}(๐ค)`$). Let $`xM`$ be a solution of MCE in $`(M,d)`$. Define
$$d^x:T(sM)sM,d_n^x(sy_1_R\mathrm{}_Rsy_n):=\underset{m0}{\mathrm{\Sigma }}d_{m+n}Sh((sx)^{_{R}^{}{}_{}{}^{m}},sy_1_R\mathrm{}_Rsy_n),$$
where $`Sh`$ denotes all shuffles of $`(sx)^{_{R}^{}{}_{}{}^{m}}`$ in $`sy_1_R\mathrm{}_Rsy_n`$ (there is no sign change because $`sx`$ is even). Then $`(M,d^x=\underset{n>0}{\mathrm{\Sigma }}d_n^x)`$ is an $`_{\mathrm{}}`$-algebra.
Proof: Since tensoring with $`(sx)^{_{R}^{}{}_{}{}^{m}}`$ has degree 0 and $`d`$ is of degree 1, the composition has degree 1.
Consider application of $`(d^x)^2`$ to $`\underset{\sigma S_n}{\mathrm{\Sigma }}(1)^{ฯต(\sigma ;sy_1,\mathrm{},sy_n)}sy_{\sigma (1)}_R\mathrm{}_Rsy_{\sigma (n)}`$. The result is the sum
$$\underset{l+m=n+1}{\mathrm{\Sigma }}d_l^x\widehat{d_m^x}(\underset{\sigma S_n}{\mathrm{\Sigma }}(1)^{ฯต(\sigma ;sy_1,\mathrm{},sy_n)}sy_{\sigma (1)}_R\mathrm{}_Rsy_{\sigma (n)})=$$
$$=\underset{l+m=n+1}{\mathrm{\Sigma }}\underset{\underset{i+j=k}{k0}}{\mathrm{\Sigma }}d_{l+i}\widehat{d_{m+j}}(\underset{\sigma S_n}{\mathrm{\Sigma }}(1)^{ฯต(\sigma ;sy_1,\mathrm{},sy_n)}Sh((sx)^{_{R}^{}{}_{}{}^{k}},sy_{\sigma (1)}_R\mathrm{}_Rsy_{\sigma (n)})).$$
This equality is true because the point in $`\overline{S}(sM)`$, generated by $`sx`$, is a cocycle for $`\overline{d}`$. For each $`k0`$ we have
$$\underset{\underset{i+j=k}{l+m=n+1}}{\mathrm{\Sigma }}d_{l+i}\widehat{d_{m+j}}(\underset{\sigma S_n}{\mathrm{\Sigma }}(1)^{ฯต(\sigma ;sy_1,\mathrm{},sy_n)}Sh((sx)^{_{R}^{}{}_{}{}^{k}},sy_{\sigma (1)}_R\mathrm{}_Rsy_{\sigma (n)}))=$$
$$=\underset{p+q=n+k+1}{\mathrm{\Sigma }}\frac{1}{k!}\underset{\sigma S_{n+k}}{\mathrm{\Sigma }}(1)^{ฯต(\sigma ;sx,\mathrm{},sx,sy_1,\mathrm{},sy_n)}d_p\widehat{d_q}(\sigma ((sx)^{_{R}^{}{}_{}{}^{k}}_Rsy_1_R\mathrm{}_Rsy_n)),$$
where $`\frac{1}{k!}`$ appears because $`sx`$ is even and interchanging $`sx`$โs does not affect the sign of the permutation, whereas on the l.h.s. of the equation the permutations of $`sx`$โs are absent. The r.h.s. is obviously $`\frac{1}{k!}`$ times $`\widehat{d}^2`$, applied to the symmetrization of $`(sx)^{_{R}^{}{}_{}{}^{k}}_Rsy_1_R\mathrm{}_Rsy_n`$, therefore for each $`k0`$ the corresponding sum is 0. $`\mathrm{}`$
Now we fix an $`๐_{\mathrm{}}`$-structure $`A\stackrel{๐พ}{}B`$ (we will denote it simply by $`\gamma `$). Let $`g`$ be the corresponding solution of MCE in $`(๐ค,d)`$ (proposition 3), we extend $`R`$-linearly the $`_{\mathrm{}}`$-structure from $`(๐ค,d^g)`$ to $`(๐ค_๐คR,d^g)`$ and then consider the $`_{\mathrm{}}`$-subalgebra $`(๐ค_๐ค\text{m}_R,d^g)`$, it is an $`_{\mathrm{}}`$-algebra in $`\text{dgmod}(R)`$.
###### Proposition 4.
Let $`g`$ be the solution of MCE in $`(๐ค,d)`$, that corresponds to $`\gamma `$. There is a bijection between the set of solutions of MCE in $`(๐ค_๐ค\text{m}_R,d^g)`$ and the set of $`๐_{\mathrm{}}`$-structures on $`(A_๐คR,B_๐คR)`$, whose reduction modulo $`\text{m}_R`$ is $`\gamma `$.
Proof: Let $`g^{}`$ be a solution of MCE in $`(๐ค_๐ค\text{m}_R,d^g)`$, then $`g+g^{}๐ค_๐คR`$ and we have
$$d_n((s(g+g^{}))^{_{R}^{}{}_{}{}^{n}})=d_n((sg)^{_{๐ค}^{}{}_{}{}^{n}})+\underset{0l<n}{\mathrm{\Sigma }}d_n(Sh(((sg)^{_{๐ค}^{}{}_{}{}^{l}}_๐คR),(sg^{})^{_{R}^{}{}_{}{}^{nl}})).$$
i.e.
$$\overline{d}(\underset{n>0}{\mathrm{\Sigma }}(sg+sg^{})^{_{R}^{}{}_{}{}^{n}})=\overline{d}(\underset{n>0}{\mathrm{\Sigma }}(sg)^{_{๐ค}^{}{}_{}{}^{n}})+\overline{d^g}(\underset{n>0}{\mathrm{\Sigma }}(sg^{})^{_{R}^{}{}_{}{}^{n}}).$$
Therefore $`g+g^{}`$ is a solution of MCE in $`(๐ค_๐คR,d)`$. Let $`\gamma +\gamma ^{}`$ be the $`๐_{\mathrm{}}`$-structure that corresponds to $`g+g^{}`$ (proposition 3). The cochains that generate this $`๐_{\mathrm{}}`$-structure are given by $`g+g^{}`$, and since $`g^{}๐ค_๐ค\text{m}_R`$, reduction modulo $`\text{m}_R`$ of $`\gamma +\gamma ^{}`$ is obviously generated by $`g`$.
Conversely, let $`\gamma +\gamma ^{}`$ be an $`๐_{\mathrm{}}`$-structure whose reduction modulo $`\text{m}_R`$ is $`\gamma `$. Again by proposition 3 there is a corresponding solution $`g+g^{}`$ of MCE in $`(๐ค_๐คR,d)`$. Reduction modulo $`\text{m}_R`$ of $`g+g^{}`$ has to be $`g`$ and hence $`g^{}๐ค_๐ค\text{m}_R`$. Using identification of MCE with the defining equations of $`๐_{\mathrm{}}`$-structures (proposition 3) we conclude that $`g^{}`$ is a solution of MCE in $`(๐ค_๐ค\text{m}_R,d^g)`$. $`\mathrm{}`$
In case of a deformation of one algebra (say $`A`$), $`C^{}(A,A)`$ is a dg Lie algebra, and given a dg Artin algebra $`R`$, solutions of MCE in $`C^{}(A,A)_๐ค\text{m}_R`$ are equivalent iff the corresponding structures of an $`๐_{\mathrm{}}`$-algebra on $`A_๐คR`$ are connected by an invertible $`๐_{\mathrm{}}`$-morphism, whose reduction modulo $`\text{m}_R`$ is the identity automorphism. Hence on the set of solutions acts the group $`(C^{}(A,A)_๐ค\text{m}_R)_0`$ (with the Campbell-Hausdorff multiplication).
In case of a deformation of the morphism $`AB`$, equivalences between $`๐_{\mathrm{}}`$-structures are given by pairs of $`๐_{\mathrm{}}`$-morphisms $`AA`$ and $`BB`$, therefore elements of the subspace $`๐คs^1Hom(T(sA),sB)`$ do not represent infinitesimal automorphisms, instead these are given by the following subspace:
$$๐ฅ:=๐ค(Hom(T(sA),sA)Hom(T(sB),sB)).$$
From definition 6 it follows that on $`๐ฅ`$ all ternary and higher operations vanish (indeed the operations that involve 3 and more elements require elements of $`s^1Hom(T(sA),sB)`$ as inputs), hence, if we forget the differential, $`๐ฅ_๐ค\text{m}_R`$ is a nilpotent Lie algebra. Let $`H_R`$ be the group, defined on the degree 0 part of $`๐ฅ_๐ค\text{m}_R`$, with the Campbell-Hausdorff multiplication. We are going to define an action of $`H_R`$ on the set of solutions of MCE in $`(๐ค_๐ค\text{m}_R,d^g)`$, where $`g`$ is the solution of MCE in $`(๐ค,d)`$, corresponding to $`\gamma `$. This action is defined through the adjoint representation of $`๐ค`$ on itself (as an $`_{\mathrm{}}`$-algebra). This representation comes from the structure of a left $`_{\mathrm{}}`$-module of $`๐ค`$ over itself.
###### Lemma 4.
Let $`g^{}`$ be a solution of MCE in $`(๐ค_๐ค\text{m}_R,d^g)`$. For an $`hH_R`$ define
$$ad_hg^{}:=d_1^{g+g^{}}(sh)+\delta _R(sh)=\underset{n0}{\mathrm{\Sigma }}d_{n+1}(Sh((sg+sg^{})^{_{R}^{}{}_{}{}^{n}},sh))+\delta _R(sh),$$
where $`\delta _R`$ is the differential on $`๐ค_๐ค\text{m}_R`$, then $`\underset{k0}{\mathrm{\Sigma }}\frac{1}{k!}(ad_h)^kg^{}`$ is also a solution of MCE in $`(๐ค_๐ค\text{m}_R,d^g)`$ and this defines an action of $`H_R`$ on the set of solutions.
Proof: Consider the groups $`G_1,G_2`$ that are the subgroups of $`Aut(T(sA_๐คR))`$, $`Aut(T(sB_๐คR))`$, consisting of the elements whose reduction modulo $`\text{m}_R`$ are identities on $`T(sA),T(sB)`$. There is an action of $`G_1\times G_2`$ on the set of solutions of MCE in $`(๐ค_๐ค\text{m}_R,d^g)`$, given by
$$\widehat{\alpha }+\widehat{\alpha }^{}\varphi (\widehat{\alpha }+\widehat{\alpha }^{}+\widehat{\delta }_R)\varphi ^1,\widehat{\beta }+\widehat{\beta }^{}\psi (\widehat{\beta }+\widehat{\beta }^{}+\widehat{\delta }_R)\psi ^1,\stackrel{~}{\gamma +\gamma ^{}}\psi (\stackrel{~}{\gamma +\gamma ^{}})\varphi ^1,$$
where $`(\varphi ,\psi )G_1\times G_2`$, $`g=\alpha +\beta +s^1\gamma `$, $`g^{}=\alpha ^{}+\beta ^{}+s^1\gamma ^{}`$. For any such $`\varphi ,\psi `$ there are $`\mu ,\nu ๐ค_๐ค\text{m}_R`$, s.t. $`\varphi =e^{\widehat{\mu }},\psi =e^{\widehat{\nu }}`$, where $`\widehat{\mu },\widehat{\nu }`$ are the coderivations cogenerated by $`\mu ,\nu `$. We have
$$\psi (\stackrel{~}{\gamma +\gamma ^{}})\varphi ^1=e^{\widehat{\nu }}(\stackrel{~}{\gamma +\gamma ^{}})e^{\widehat{\mu }}=\underset{k0}{\mathrm{\Sigma }}\frac{1}{k!}(ad_{\widehat{\mu }+\widehat{\nu }})^k(\stackrel{~}{\gamma +\gamma ^{}}),$$
where $`ad_{\widehat{\mu }+\widehat{\nu }}(\stackrel{~}{\gamma +\gamma ^{}}):=\widehat{\nu }(\stackrel{~}{\gamma +\gamma ^{}})(\stackrel{~}{\gamma +\gamma ^{}})\widehat{\mu }`$. Summands on the r.h.s. of the last equation are $`\stackrel{~}{\gamma +\gamma ^{}}`$-coderivations from $`T(sA_๐คR)`$ to $`T(sB_๐คR)`$, therefore their sum is determined by its corestriction to the cogenerators of $`T(sB_๐คR)`$, and this is exactly the projection of $`ad_{\mu +\nu }g^{}`$ onto $`s^1Hom(T(sA),sB)_๐ค\text{m}_R`$, similarly for $`\widehat{\alpha },\widehat{\beta }`$. So $`\underset{k0}{\mathrm{\Sigma }}\frac{1}{k!}(ad_h)^k`$ represents the action of $`G_1\times G_2`$ on the set of solutions and hence it defines an action of $`H_R`$, since Campbell-Hausdorff multiplication represents composition in the corresponding group. $`\mathrm{}`$
Using lemma 4 we can represent morphisms between $`๐_{\mathrm{}}`$-structures by elements of the Lie algebra $`๐ฅ`$. However, we are interested in the whole spaces of maps. Let $`\mathrm{\Omega }_n`$ be the algebra of polynomial forms on the $`n`$-simplex described in section 1. Let $`g`$ be the solution of MCE in $`(๐ค,d)`$, corresponding to $`\gamma `$.
###### Definition 8.
(\[Hin1\](3.1)) The simplicial Deligne groupoid $`\underset{ยฏ}{\text{Del}}^\gamma (R)`$ is given by
$$Obj(\underset{ยฏ}{\text{Del}}^\gamma (R)):=\{\text{ the set of solutions of MCE in }(๐ค_๐ค\text{m}_R,d^g)\},$$
$$\underset{ยฏ}{Hom}_n(g_1,g_2):=\{h๐ฅ_๐ค\text{m}_R_๐ค\mathrm{\Omega }_n\text{ s.t. }\underset{k0}{\mathrm{\Sigma }}\frac{1}{k!}(ad_h)^kg_1=g_2\},$$
where we extend $`g_1,g_2`$ linearly to solutions in $`๐ค_๐ค\text{m}_R_๐ค\mathrm{\Omega }_n`$. Simplicial structure on $`\underset{ยฏ}{Hom}(g_1,g_2)`$ is given by the one on $`\{\mathrm{\Omega }_n\}_{n0}`$.
From lemma 4 it follows that $`\underset{ยฏ}{\text{Del}}^\gamma (R)`$ is indeed a simplicial groupoid.
Now consider in general a situation like in lemma 4: an $`_{\mathrm{}}`$-algebra $`๐ค`$, s.t. $`๐ค=M๐ฅ`$, where, if we forget the differential, $`๐ฅ`$ is a Lie subalgebra. Suppose we have a morphism of $`_{\mathrm{}}`$-algebras $`f:๐ค_1๐ค_2`$, such that $`f(M_1)M_2`$ and $`f(๐ฅ_1)๐ฅ_2`$.
###### Lemma 5.
If $`f`$ is a quasi-isomorphism, it induces a weak equivalence between the simplicial Deligne groupoids, corresponding to $`(๐ค_1,๐ฅ_1)`$ and $`(๐ค_2,๐ฅ_2)`$.
Proof: Applying cobar construction if necessary, we can assume that $`๐ค_1,๐ค_2`$ are dg Lie algebras. Suppose first that $`f`$ is an acyclic fibration. Then according to \[Hin1\](3.3.1), the corresponding functor $`F`$ between the Deligne groupoids of $`๐ค_1`$ and $`๐ค_2`$ is an acyclic fibration. That means in particular that $`\pi _0(F)`$ is an equivalence of categories and for any pair of objects $`P,Q`$ in $`\underset{ยฏ}{\text{Del}}(๐ค_1)`$, the map of simplicial sets $`\underset{ยฏ}{Hom}(P,Q)\underset{ยฏ}{Hom}(F(P),F(Q))`$ is an acyclic fibration. The simplicial set $`\underset{ยฏ}{Hom}(P,Q)`$ has a subset $`\underset{ยฏ}{Hom}_{๐ฅ_1}(P,Q)`$, consisting of the maps, defined by elements of $`๐ฅ_1`$. Since $`f`$ maps $`๐ฅ_1`$ to $`๐ฅ_2`$ and $`M_1`$ to $`M_2`$, it is clear that $`F`$ maps $`\underset{ยฏ}{Hom}_{๐ฅ_1}(P,Q)`$ to $`\underset{ยฏ}{Hom}_{๐ฅ_2}(F(P),F(Q))`$, and moreover the inverse image under $`F`$ of $`\underset{ยฏ}{Hom}_{๐ฅ_2}(F(P),F(Q))`$ is $`\underset{ยฏ}{Hom}_{๐ฅ_1}(P,Q)`$. Therefore, since pullback of an acyclic fibration is again an acyclic fibration, we conclude that $`F`$ defines an acyclic fibration from $`\underset{ยฏ}{Hom}_{๐ฅ_1}(P,Q)`$ to $`\underset{ยฏ}{Hom}_{๐ฅ_2}(F(P),F(Q))`$, and these are the mapping spaces in $`\underset{ยฏ}{\text{Del}}(๐ค_1,๐ฅ_1)`$ and $`\underset{ยฏ}{\text{Del}}(๐ค_2,๐ฅ_2)`$.
If $`f`$ is not an acyclic fibration, it splits $`f:g_1\stackrel{๐}{}๐ค_3\stackrel{๐}{}๐ค_2`$, where $`p`$ is an acyclic fibration and the $`i`$ is an acyclic cofibration. Let $`๐ฅ_3,M_3`$ be the inverse images of $`๐ฅ_2,M_2`$ under $`p`$. Clearly $`i^1(๐ฅ_3)=๐ฅ_1`$. From the construction of $`๐ค_3`$ (see e.g. \[Hin3\](2.2.4)) it is clear that one can define a quasi-isomorphism $`๐ค_3๐ค_1`$, left inverse to $`i`$, s.t. the image of $`๐ฅ_3`$ is in $`๐ฅ_1`$ and of $`M_3`$ is in $`M_1`$ (indeed just send all joined boundaries and coboundaries to $`0`$). Obviously this quasi-isomorphism is an acyclic fibration. $`\mathrm{}`$
###### Proposition 5.
Let $`\gamma :AB`$ and $`\gamma ^{}:A^{}B^{}`$ be morphisms of associative algebras. Let $`q:AA^{},p:BB^{}`$ be a quasi-isomorphism from $`\gamma `$ to $`\gamma ^{}`$. Then the Deligne groupoids $`\underset{ยฏ}{\text{Del}}^\gamma `$ and $`\underset{ยฏ}{\text{Del}}^\gamma ^{}`$ are weakly equivalent.
Proof: The map $`p`$ defines two maps: $`C^{}(B^{},B^{})C^{}(B,B^{})`$ and $`C^{}(B,B)C^{}(B,B^{})`$. Let $`๐ค`$ be the corresponding fiber product of $`C^{}(B^{},B^{}),C^{}(B,B)`$ over $`C^{}(B,B^{})`$ (as vector spaces). We can identify $`๐ค`$ with the subspace of $`C^{}(B^{},B^{})\times C^{}(B,B)`$, consisting of pairs of elements, whose images in $`C^{}(B,B^{})`$ coincide. Componentwise operations define of $`๐ค`$ the structure of a dg Lie algebra. We have canonical $`f:๐คC^{}(B,B)`$ and $`f^{}:๐คC^{}(B^{},B^{})`$, and since $`p`$ is a quasi-isomorphism, these maps are quasi-isomorphisms. We extend by the means of $`f,f^{}`$ the $`_{\mathrm{}}`$ structures from $`C^{}(A,A)s^1C^{}(A,B)C^{}(B,B)`$ and $`C^{}(A,A)s^1C^{}(A,B^{})C^{}(B^{},B^{})`$ to $`C^{}(A,A)s^1C^{}(A,B)๐ค`$ and $`C^{}(A,A)s^1C^{}(A,B^{})๐ค`$ respectively. We have three quasi-isomorphisms:
$$f:C^{}(A,A)s^1C^{}(A,B)๐คC^{}(A,A)s^1C^{}(A,B)C^{}(B,B),$$
$$f^{}:C^{}(A,A)s^1C^{}(A,B^{})๐คC^{}(A,A)s^1C^{}(A,B^{})C^{}(B^{},B^{}),$$
$$p_{}:C^{}(A,A)s^1C^{}(A,B)๐คC^{}(A,A)s^1C^{}(A,B^{})๐ค.$$
So by lemma 5, the simplicial Deligne groupoids that correspond to $`\gamma `$ and $`p\gamma `$ are weakly equivalent. Doing the same thing with $`q`$ we get the final result. $`\mathrm{}`$
## 3. Solution of the deformation problem in case of non-positively graded algebras
In this section we prove that if $`\gamma :AB`$ is a morphism of non-positively graded algebras, then $`\underset{ยฏ}{\text{Def}}^\gamma `$ is weakly equivalent to $`\underset{ยฏ}{\text{Del}}^\gamma `$. We make this requirement because all non-positively graded almost free algebras are cofibrant, whereas in general it is not true for $``$-graded algebras.
Let $`(M,\delta )`$ be a differential $`_0`$-graded associative $`R`$-algebra. The bar construction of $`(M,\delta )`$ is the codifferential coassociative coalgebra $`(๐
M,๐
\delta )`$, where $`๐
M:=T(sM)`$ and $`๐
\delta =s(\delta +\mu )s^1`$, where $`\mu `$ is the multiplication on $`M`$. In turn for a codifferential coassociative coalgebra $`(Z,\delta )`$ the co-bar construction is the dg associative algebra $`(\mathrm{\Omega }Z,\mathrm{\Omega }\delta )`$, where $`\mathrm{\Omega }Z:=T(s^1Z)`$, $`\mathrm{\Omega }\delta :=s^1(\delta +\mathrm{\Delta })s`$, where $`\mathrm{\Delta }`$ is the comultiplication on $`Z`$. We will denote the co-bar construction on the bar construction of $`M`$ by $`\mathrm{\Omega }๐
M`$.
$`\mathrm{\Omega }๐
M`$ is an almost free non-positively graded algebra, and it is cofibrant if $`M`$ is non-positively graded. Indeed almost free algebras are cofibrant in the category of non-positively graded dg associative algebras (\[Get\](4.6)), and an acyclic fibration of $``$-graded algebras induces an acyclic fibration of their truncations at 0, with 0-part consisting of cocycles. Left lifting property then goes over to the category of all $``$-graded algebras. There is a natural transformation $`ฯต:\mathrm{\Omega }๐
Id_{\text{DGAlg}(R)}`$.
Clearly $`\mathrm{\Omega }๐
`$ is a functor, hence it extends to $`\text{Mor}\text{DGAlg}(R)`$, and we will denote the extension again by $`\mathrm{\Omega }๐
`$. If $`M`$ and $`N`$ in $`\varphi :MN\text{Mor}\text{DGAlg}(R)`$ are non-positively graded algebras, then $`\mathrm{\Omega }๐
M,\mathrm{\Omega }๐
N`$ are cofibrant, and if in addition $`\varphi `$ is a cofibration, then $`\mathrm{\Omega }๐
\varphi `$ is a cofibration. Indeed, if $`\varphi `$ is injective then, since $`\mathrm{\Omega }๐
\varphi `$ maps generators of the domain injectively to generators of the co-domain, it is a cofibration (\[Get\] page 42), if $`\varphi `$ is not injective we can split it into an injective cofibration, followed by an acyclic fibration, and then $`\mathrm{\Omega }๐
\varphi `$ is a retract of a cofibration. Natural transformation $`ฯต`$ extends to $`\mathrm{\Omega }๐
Id_{\text{Mor}\text{DGAlg}(R)}`$.
Let $`C(R)`$ be the simplicial subcategory of $`\text{Mor}\underset{ยฏ}{\text{DGAlg}}(R)`$, consisting of cofibrations between non-positively graded, cofibrant algebras. As noted above $`\mathrm{\Omega }๐
`$ maps $`C(R)`$ into itself, and there is a natural transformation $`ฯต:\mathrm{\Omega }๐
Id_{C(R)}`$.
###### Lemma 6.
$`\mathrm{\Omega }๐
:C(R)\mathrm{\Omega }๐
(C(R))`$ and the inclusion of $`\mathrm{\Omega }๐
(C(R))`$ in $`C(R)`$ induce a weak equivalence of the nerves of these two categories.
Now consider a morphism $`\gamma :AB\text{Mor}\text{DGAlg}(๐ค)`$ between non-positively graded algebras that we want to deform. By proposition 5, for the purpose of comparing $`\underset{ยฏ}{\text{Del}}^\gamma `$ with $`\underset{ยฏ}{\text{Def}}^\gamma `$, we can consider $`\gamma `$ as a cofibration between cofibrant objects, i.e. $`\gamma C(R)`$.
We have the $`_{\mathrm{}}`$-algebra $`(๐ค,d^g)`$ that describes deformations of $`\gamma `$ as an $`๐_{\mathrm{}}`$-structure. These deformations consist of codifferentials on $`๐
(A_๐คR),๐
(B_๐คR)`$ and coalgebra morphisms $`๐
(A_๐คR)๐
(B_๐คR)`$. Applying $`\mathrm{\Omega }`$ to them we get objects in $`C(R)`$, whose reduction modulo $`\text{m}_R`$ is $`\mathrm{\Omega }๐
\gamma `$. In this way for every solution $`g^{}`$ of MCE in $`(๐ค_๐ค\text{m}_R,d^g)`$ there is a corresponding object $`\underset{ยฏ}{F}_R(g^{})`$ in $`\underset{ยฏ}{\text{Def}}^\gamma `$. Moreover, mapping spaces in the Deligne groupoid represent morphisms between morphisms of coalgebras (lemma 4), therefore since the cobar construction is a functor, $`\underset{ยฏ}{F}_R`$ is actually a simplicial functor $`\underset{ยฏ}{\text{Del}}^\gamma (R)\underset{ยฏ}{\text{Def}}^\gamma (R)`$. Clearly $`\mathrm{\Omega }`$ is functorial in $`R`$, hence $`\underset{ยฏ}{F}`$ is a morphism in $`Hom(\text{dgart}(๐ค),\underset{ยฏ}{\text{Grp}})`$.
###### Theorem 1.
$`\underset{ยฏ}{F}_R`$ is a weak equivalence $`\underset{ยฏ}{\text{Del}}^\gamma (R)\underset{ยฏ}{\text{Def}}^\gamma (R)`$.
Proof: As noted above, $`\underset{ยฏ}{F}_R`$ maps $`\underset{ยฏ}{\text{Del}}^\gamma (R)`$ to $`C(R)`$, and the latter is a full simplicial subcategory of the category of cofibrant objects in $`\text{Mor}\underset{ยฏ}{\text{DGAlg}}(R)`$. Clearly the image of $`\underset{ยฏ}{F}_R`$ is in the fiber of $`\underset{ยฏ}{\pi ^{}}_R`$ over $`\underset{ยฏ}{\mathrm{\Omega }๐
\gamma }`$ (see definition 3). To prove that $`\underset{ยฏ}{F}_R`$ is a weak equivalence of $`\underset{ยฏ}{\text{Del}}^\gamma (R)`$ with this fiber note that $`\underset{ยฏ}{F}_R`$ maps $`\underset{ยฏ}{\text{Del}}^\gamma (R)`$ identically on the fiber of $`\mathrm{\Omega }๐
(C(R))\mathrm{\Omega }๐
(C(๐ค))`$ at $`\underset{ยฏ}{\mathrm{\Omega }๐
\gamma }`$. To compare this fiber with the one of $`\underset{ยฏ}{\pi ^{}}_R`$, note that a map between simplicial groupoids is a weak equivalence if and only if it induces a weak equivalence of the nerves. Moreover, since for fibrations between simplicial categories, the nerve of a fiber is equivalent to the fiber of the nerve (see proof of lemma 1), from lemma 6 we conclude that the fiber of $`\mathrm{\Omega }๐
(C(R))\mathrm{\Omega }๐
(C(๐ค))`$ at $`\underset{ยฏ}{\mathrm{\Omega }๐
\gamma }`$ is indeed weakly equivalent to the fiber of $`\underset{ยฏ}{\pi ^{}}_R`$ at $`\underset{ยฏ}{\mathrm{\Omega }๐
\gamma }`$. $`\mathrm{}`$
### 3.1. Cohomology
If we start with a pair of associative algebras $`A,B`$, concentrated in degree 0, and an associative algebra morphism $`\gamma :AB`$, then the differential $`d_1^g`$ on $`๐ค`$ is as follows ($`\alpha C^{}(A,A)`$)
$$d_1^g(\alpha )(a_1_๐ค\mathrm{}_๐คa_{n+1})=(1)^{n+1}(a_1\alpha (a_1_๐ค\mathrm{}_๐คa_{n+1})+$$
$$+\underset{1in}{\mathrm{\Sigma }}(1)^{i+1}\alpha (a_1_๐ค\mathrm{}_๐คa_ia_{i+1}_๐ค\mathrm{}_๐คa_{n+1})+(1)^n\alpha (a_1_๐ค\mathrm{}_๐คa_n)a_{n+1})$$
$$\gamma (\alpha (a_1_๐ค\mathrm{}_๐คa_n)).$$
In other words
$$d_1^g(\alpha )=(1)^{\alpha +1}HD(\alpha )\gamma _{}(\alpha ),$$
where $`HD`$ is the Hochschild differential and the degree of $`\alpha `$ is the number of its arguments minus 1. For a $`\gamma ^{}C^{}(A,B)`$ we have
$$d_1^g(\gamma ^{})(a_1_๐ค\mathrm{}_๐คa_{n+1})=\gamma ^{}(a_1_๐ค\mathrm{}_๐คa_n)\gamma (a_{n+1})+$$
$$+(1)^{n+1}\gamma (a_1)\gamma ^{}(a_2_๐ค\mathrm{}_๐คa_{n+1})+(1)^n\underset{1in}{\mathrm{\Sigma }}(1)^{i+1}\gamma ^{}(a_1_๐ค\mathrm{}_๐คa_ia_{i+1}_๐ค\mathrm{}_๐คa_{n+1}),$$
that is
$$d_1^g(\gamma ^{})=(1)^\gamma ^{}HD(\gamma ^{}).$$
The case of $`\beta C^{}(B,B)`$ is similar to $`\alpha C^{}(A,A)`$. In total we can describe the differential complex
$$(C^{}(A,A)C^{}(B,B)C^{}(A,B),d_1^g)$$
as the cone of the morphism of complexes:
$$(C^{}(A,A),(1)^nHD)(C^{}(B,B),(1)^nHD)(C^{}(A,B),(1)^nHD),$$
where $`\alpha +\beta `$ is mapped to $`\gamma ^{}(\beta )\gamma _{}(\alpha )`$. Gerstenhaber and Schack (\[GS2\] page 249, \[GS3\] page 8) have constructed cohomology of a morphism using the same complexes with untwisted Hochschild differentials and the opposite morphism of complexes. The cohomology in their construction is the same as in ours.
Cohomology of an $`_{\mathrm{}}`$-algebra has a canonical structure of a Lie algebra (induced by the binary operation of the $`_{\mathrm{}}`$-structure). The Lie structure on cohomology appearing in \[GS2\],\[GS3\] is induced by the $`_{\mathrm{}}`$-structure on cochains that we have described, indeed the binary operation of this $`_{\mathrm{}}`$-structure coincides with the commutator of the circle operation on cochains that is used in \[GS2\],\[GS3\].
|
warning/0506/cond-mat0506494.html
|
ar5iv
|
text
|
# Multi-species grandcanonical models for networks with reciprocity
\[
## Abstract
Reciprocity is a second-order correlation that has been recently detected in all real directed networks and shown to have a crucial effect on the dynamical processes taking place on them. However, no current theoretical model generates networks with this nontrivial property. Here we propose a grandcanonical class of models reproducing the observed patterns of reciprocity by regarding single and double links as Fermi particles of different โchemical speciesโ governed by the corresponding chemical potentials. Within this framework we find interesting special cases such as the extensions of random graphs, the configuration model and hidden-variable models. Our theoretical predictions are also in excellent agreement with the empirical results for networks with well studied reciprocity.
\]
The topological properties of networks are known to affect crucially the outcomes of dynamical processes taking place on them . A particularly important role is played by the (second-order) correlations between vertex degrees, which have strong effects on many processes including percolation and epidemic spreading . Directed networks have been recently shown to display an additional type of second-order correlation: the nonrandom presence of mutual links between vertices, or *reciprocity*. Nontrivial reciprocity is found in all real networks, and provides new insights into their topology . Moreover, reciprocity was recently shown to change dramatically the properties of percolation and epidemic spreading , and that these unexpected dynamical properties are triggered even by a small fraction of bidirectional links . However, despite its ubiquity and relevance, reciprocity is currently not reproduced by any model.
In this Letter we introduce a general theory for networks displaying nontrivial reciprocity by extending various important models (including random graphs , the configuration model and the whole class of hidden-variable models ). Since all these models can be obtained as particular cases of the general class of โexponential modelsโ , a unifying approach is to extend the latter to include reciprocity, so that all the particular cases are automatically modified accordingly. Therefore we first reformulate the standard results for the exponential model defined by the โgraph Hamiltonianโ
$$H=\underset{ij}{}ฯต_{ij}a_{ij}$$
(1)
where $`a_{ij}=1`$ if a link from $`i`$ to $`j`$ is there (and 0 otherwise), and $`ฯต_{ij}`$ is the โenergyโ (or โcostโ) of such a link. The grand partition function and grand potential read
$`๐ต`$ $``$ $`{\displaystyle \underset{\{a_{ij}\}}{}}e^{\mu LH}={\displaystyle \underset{ij}{}}{\displaystyle \underset{a_{ij}=0,1}{}}e^{(\mu ฯต_{ij})a_{ij}}={\displaystyle \underset{i<j}{}}๐ต_{ij}`$ (2)
$`\mathrm{\Omega }`$ $``$ $`\mathrm{ln}๐ต={\displaystyle \underset{ij}{}}\mathrm{ln}๐ต_{ij}={\displaystyle \underset{ij}{}}\mathrm{\Omega }_{ij}`$ (3)
where $`\mu `$ is the chemical potential and
$$๐ต_{ij}1+e^{\mu ฯต_{ij}}\mathrm{\Omega }_{ij}\mathrm{ln}๐ต_{ij}$$
(4)
We note that $`\mu `$ is not considered explicitly in the literature since its role can be played by an additional constant term in $`H`$. However, since in what follows we shall introduce more โchemical speciesโ, we prefer to adopt the inverse strategy to keep $`\mu `$ and reabsorbe any constant energy term into it (this point will be made clearer below). This also allows us to obtain many expected topological properties as derivatives of $`\mathrm{\Omega }`$ with respect to $`\mu `$. For instance, the probability $`p_{ij}`$ of a directed link from $`i`$ to $`j`$ and the expected number $`L`$ of directed links read
$$p_{ij}=a_{ij}=\frac{\mathrm{\Omega }_{ij}}{\mu }=\frac{1}{1+e^{ฯต_{ij}\mu }}L=\frac{\mathrm{\Omega }}{\mu }=\underset{ij}{}p_{ij}$$
(5)
Many static models can be recovered as particular cases of this formalism . For instance, the case $`ฯต_{ij}=ฯต`$ is the directed version of the random graph model:
$$H=ฯต\underset{ij}{}a_{ij}=ฯตLp_{ij}=p=\frac{1}{1+e^\mu }$$
(6)
where we have reabsorbed $`ฯต`$ in a redefinition of $`\mu `$. Another interesting case is the additive one $`ฯต_{ij}=\alpha _i+\beta _j`$, which corresponds to the grandcanonical version of the directed configuration model :
$$H=\underset{i}{}(\alpha _ik_i^{out}+\beta _ik_i^{in})p_{ij}=\frac{zx_iy_j}{1+zx_iy_j}$$
(7)
where we have introduced the โfugacityโ $`ze^\mu `$ and the โfitness valuesโ $`x_ie^{\alpha _i}`$, $`y_je^{\beta _j}`$. We finally note that, while the above case corresponds to the choice $`ฯต_{ij}=\mathrm{ln}(x_iy_j)`$, the general form $`ฯต_{ij}=ฯต(x_i,y_j)`$ is equivalent to the whole class of (directed) hidden-variable models defined by the corresponding fitness-dependent probability $`p(x_i,y_j)`$. Therefore all the most relevant static network models can be recovered as particular cases of eq.(1). However, in all such cases the expected reciprocity is trivial, since the probability of having a link from $`i`$ to $`j`$ is independent on the probability of having the reciprocal link from $`j`$ to $`i`$ . In other words, if we write the reciprocity $`r`$ and its random value $`r_{rand}`$ as
$$r\frac{L^{}}{L}r_{rand}=\overline{a}=\frac{L}{N(N1)}$$
(8)
(where $`L^{}_{ij}a_{ij}a_{ji}`$ is the number of reciprocated links), all the above models display the trivial expected value $`r=r_{rand}`$. The only way to have a nontrivial value of $`r`$ is adding an extra term to eq.(1):
$$H=\underset{ij}{}ฯต_{ij}a_{ij}+H^{}H^{}=\frac{\lambda }{2}L^{}$$
(9)
This choice defines the *reciprocity model* first proposed in ref. and recently studied analytically by Park and Newman in the case $`ฯต_{ij}=ฯต`$ by treating $`H^{}`$ as a perturbation. They found that in this particular case the perturbation expansion can be resummed to all orders to give an exact expression for $`๐ต`$ . In our notation with $`ฯต`$ absorbed in $`\mu `$, the final expression for $`r`$ is
$$r=\frac{1}{1+e^{\mu \lambda }}$$
(10)
and $`rr_{rand}`$ whenever $`\lambda 0`$ . Therefore the reciprocity of this model can be tuned to any desired value, however other fundamental topological properties (such as scale-free behaviour) discovered more recently are not reproduced. Moreover, the perturbative approach is analytically complicated and yields exact results only in the particular case described above.
A non-perturbative theoretical model which reproduces all the relevant topological properties including the reciprocity is therefore missing. We now define a general class of such models by regarding reciprocated and non-reciprocated links as different โchemical speciesโ, each governed by the corresponding chemical potential. In particular, we consider each pair of vertices $`i`$, $`j`$ only once (say, with $`i<j`$) and regard a non-reciprocated link from $`i`$ to $`j`$ as a โparticleโ of the chemical species labeled by the symbol $`()`$, a non-reciprocated link from $`j`$ to $`i`$ as a particle of species $`()`$, and two mutual links between $`i`$ and $`j`$ as a single particle of type $`()`$. We denote the numbers of particles of such chemical species by $`n^{}`$, $`n^{}`$ and $`n^{}`$ respectively, and the corresponding chemical potentials by $`\mu ^{}`$, $`\mu ^{}`$ and $`\mu ^{}`$. The number of reciprocated links is therefore $`L^{}=2n^{}`$, and the number of non-reciprocated links is $`LL^{}=n^{}+n^{}`$, so that $`L=n^{}+n^{}+2n^{}`$. Our formalism corresponds to the decomposition of any directed graph with adjacency matrix $`a_{ij}`$ into three distinct graphs, with adjacency matrices $`a_{ij}^{}a_{ij}(1a_{ji})`$, $`a_{ij}^{}a_{ji}(1a_{ij})`$ and $`a_{ij}^{}a_{ij}a_{ji}`$. We can now generalize the Hamiltonian defined in eq.(1) to the case with three chemical species:
$$H=\underset{i<j}{}(ฯต_{ij}^{}a_{ij}^{}+ฯต_{ij}^{}a_{ij}^{}+ฯต_{ij}^{}a_{ij}^{})$$
(11)
The grand partition function is now
$`๐ต`$ $`{\displaystyle \underset{\{a_{ij}^{}\}}{}}{\displaystyle \underset{\{a_{ij}^{}\}}{}}{\displaystyle \underset{\{a_{ij}^{}\}}{}}e^{(\mu ^{}n^{}+\mu ^{}n^{}+\mu ^{}n^{}H)}`$ (12)
$`=`$ $`{\displaystyle \underset{\{a_{ij}^{}\}}{}}{\displaystyle \underset{\{a_{ij}^{}\}}{}}{\displaystyle \underset{\{a_{ij}^{}\}}{}}{\displaystyle \underset{i<j}{}}e^{[(\mu ^{}ฯต_{ij}^{})a_{ij}^{}+(\mu ^{}ฯต_{ij}^{})a_{ij}^{}+(\mu ^{}ฯต_{ij}^{})a_{ij}^{}]}`$ (13)
$`=`$ $`{\displaystyle \underset{i<j}{}}\left[1+e^{(\mu ^{}ฯต_{ij}^{})}+e^{(\mu ^{}ฯต_{ij}^{})}+e^{(\mu ^{}ฯต_{ij}^{})}\right]={\displaystyle \underset{i<j}{}}๐ต_{ij}`$ (14)
where we have defined the vertex-pair partition function
$$๐ต_{ij}1+e^{(\mu ^{}ฯต_{ij}^{})}+e^{(\mu ^{}ฯต_{ij}^{})}+e^{(\mu ^{}ฯต_{ij}^{})}$$
(15)
Note that, when exchanging sums and products in eq.(12), we have replaced the sum over the configurations $`\{a_{ij}^{}\},\{a_{ij}^{}\},\{a_{ij}^{}\}`$ with a sum over the allowed states $`(a_{ij}^{},a_{ij}^{},a_{ij}^{})=\{(0,0,0),(0,0,1),(0,1,0),(1,0,0)\}`$, nonzero adjacency matrix elements being mutually excluding. The grand potential is
$$\mathrm{\Omega }\mathrm{ln}๐ต=\underset{i<j}{}\mathrm{ln}๐ต_{ij}=\underset{i<j}{}\mathrm{\Omega }_{ij}$$
(16)
where $`\mathrm{\Omega }_{ij}\mathrm{ln}๐ต_{ij}`$. Our model is completely defined. For each unrepeated pair of vertices $`i<j`$, the probabilities of having a non-reciprocated link from $`i`$ to $`j`$, a non-reciprocated link from $`j`$ to $`i`$, two reciprocated links between $`i`$ and $`j`$, or no link at all are given by
$`p_{ij}^{}`$ $`=`$ $`a_{ij}^{}={\displaystyle \frac{\mathrm{\Omega }_{ij}}{\mu ^{}}}={\displaystyle \frac{e^{(\mu ^{}ฯต_{ij}^{})}}{๐ต_{ij}}}`$ (17)
$`p_{ij}^{}`$ $`=`$ $`a_{ij}^{}={\displaystyle \frac{\mathrm{\Omega }_{ij}}{\mu ^{}}}={\displaystyle \frac{e^{(\mu ^{}ฯต_{ij}^{})}}{๐ต_{ij}}}`$ (18)
$`p_{ij}^{}`$ $`=`$ $`a_{ij}^{}={\displaystyle \frac{\mathrm{\Omega }_{ij}}{\mu ^{}}}={\displaystyle \frac{e^{(\mu ^{}ฯต_{ij}^{})}}{๐ต_{ij}}}`$ (19)
$`p_{ij}^{}`$ $`=`$ $`1p_{ij}^{}p_{ij}^{}p_{ij}^{}`$ (20)
respectively. Note that formally eqs.(17-20) are undefined for $`i>j`$. However, since $`()`$ and $`()`$ are actually the same chemical species which has been โsplitโ in order to consider each pair of vertices only once, we require $`p_{ij}^{}=p_{ji}^{}`$ for $`i>j`$. Similarly, we require $`p_{ij}^{}=p_{ji}^{}`$ and $`p_{ij}^{}=p_{ji}^{}`$. This is realized by setting $`\mu ^{}=\mu ^{}`$, $`ฯต_{ij}^{}=ฯต_{ji}^{}`$ and $`ฯต_{ij}^{}=ฯต_{ji}^{}`$ for $`i>j`$. Thus it is enough to specify $`p_{ij}^{}`$ and $`p_{ij}^{}`$ to define the model completely. We can write the ordinary (unconditional) probability $`p_{ij}`$ and the conditional probability $`r_{ij}`$ introduced in ref. explicitly as
$`p_{ij}p(ij)=p_{ij}^{}+p_{ij}^{}`$ (21)
$`r_{ij}p(ij|ji)={\displaystyle \frac{p_{ij}^{}}{p_{ji}}}={\displaystyle \frac{1}{1+e^{(\mu ^{}ฯต_{ji}^{}\mu ^{}+ฯต_{ij}^{})}}}`$ (22)
and the expected values $`r`$ and $`r_{rand}`$ can be obtained either as the average values $`r_{rand}=_{ij}p_{ij}/N(N1)`$ and $`r=_{ij}r_{ij}/N(N1)`$ or from
$$n^{}=\frac{\mathrm{\Omega }}{\mu ^{}}n^{}=\frac{\mathrm{\Omega }}{\mu ^{}}n^{}=\frac{\mathrm{\Omega }}{\mu ^{}}$$
(23)
Note that $`r=r_{rand}`$ if $`r_{ij}=p_{ij}`$. Unlike the model defined in eq.(9), in our multi-species formalism all the terms of the Hamiltonian are equally important, with no โperturbationsโ. As a consequence, our results are obtained in a non-perturbative way and are exact for all choices of $`H`$. This remarkable advantage allows to perform otherwise complicated analytical calculations in a very simple and direct way.
Now we consider various special cases of our model. The simplest choice $`ฯต_{ij}^{}=ฯต^{}`$ and $`ฯต_{ij}^{}=ฯต^{}`$ yields
$$H=ฯต^{}(n^{}+n^{})+ฯต^{}n^{}=ฯต^{}L+\frac{ฯต^{}2ฯต^{}}{2}L^{}$$
(24)
which is the reciprocity model defined by eq.(9) in the case $`ฯต_{ij}=ฯต`$ with the identification $`ฯต^{}=ฯต`$ and $`ฯต^{}=2ฯต\lambda `$. After $`ฯต^{}`$ and $`ฯต^{}`$ are reabsorbed in $`\mu ^{}`$ and $`\mu ^{}`$, this identification becomes $`\mu ^{}=\mu `$ and $`\mu ^{}=2\mu +\lambda `$, or
$$\mu ^{}=2\mu ^{}+\lambda $$
(25)
and we expect to recover eq.(10) through it. We find
$`p_{ij}^{}={\displaystyle \frac{e^\mu ^{}}{1+2e^\mu ^{}+e^\mu ^{}}}p_{ij}^{}={\displaystyle \frac{e^\mu ^{}}{1+2e^\mu ^{}+e^\mu ^{}}}`$ (26)
$`r=r_{ij}={\displaystyle \frac{1}{1+e^{(\mu ^{}\mu ^{})}}}`$ (27)
and eq.(27) is indeed equivalent to eq.(10) through eq.(25). Therefore we recover the results by Park and Newman much more directly, and we can also generalize them immediately to more realistic Hamiltonians.
Another case is the additive choice $`ฯต_{ij}^{}=\alpha _i+\beta _j`$, $`ฯต_{ij}^{}=\gamma _i+\gamma _j`$. If we define the *non-reciprocated out- and in-degrees* $`k_i^{}`$, $`k_i^{}`$ and the *reciprocated degree* $`k_i^{}`$ as
$$k_i^{}\underset{j}{}a_{ij}^{}k_i^{}\underset{j}{}a_{ij}^{}k_i^{}\underset{j}{}a_{ij}^{}$$
(28)
then we can rewrite the Hamiltonian as
$$H=\underset{i}{}(\alpha _ik_i^{}+\beta _ik_i^{}+\gamma _ik_i^{})$$
(29)
Equation (29) should be compared with eq.(7). While in the โordinaryโ configuration model the degree sequences $`\{k_i^{in}\}`$, $`\{k_i^{out}\}`$ appear in $`H`$ and are preserved while higher-order properties are randomized, here the same happens for the three degree sequences $`\{k_i^{}\}`$, $`\{k_i^{}\}`$ and $`\{k_i^{}\}`$. We can therefore denote this case as the *configuration model with reciprocity*. The difference in terms of the statistical weight of graphs in the ensemble is shown in fig.1. The graphs $`G_1`$ and $`G_2`$ have the same $`\{k_i^{in}\}`$ and $`\{k_i^{out}\}`$ and the same $`\{k_i^{}\}`$, $`\{k_i^{}\}`$ and $`\{k_i^{}\}`$, and are equiprobable in both models. The same occurs for $`G_3`$ and $`G_4`$.
By contrast, $`G_5`$ and $`G_6`$ have the same $`\{k_i^{in}\}`$, $`\{k_i^{out}\}`$ but different $`\{k_i^{}\}`$, $`\{k_i^{}\}`$ and $`\{k_i^{}\}`$. Therefore in the ordinary configuration model they are equiprobable, while in our model they are not. Transforming $`G_1`$ into $`G_2`$ and $`G_3`$ into $`G_4`$ (but not $`G_5`$ into $`G_6`$) can also be considered as the allowed generalizations of the โlocal rewiring algorithmโ randomizing a network to detect higher-order correlations. Here the reciprocity is preserved while randomizing the network. This is possible since we have *two* fugacities $`z^{}e^\mu ^{}`$, $`z^{}e^\mu ^{}`$ and *three* fitness variables $`x_ie^{\alpha _i}`$, $`y_ie^{\beta _i}`$, $`w_ie^{\gamma _i}`$ determining the probabilities
$`p_{ij}^{}`$ $`=`$ $`{\displaystyle \frac{z^{}x_iy_j}{1+z^{}x_iy_j+z^{}x_jy_i+z^{}w_iw_j}}`$ (30)
$`p_{ij}^{}`$ $`=`$ $`{\displaystyle \frac{z^{}w_iw_j}{1+z^{}x_iy_j+z^{}x_jy_i+z^{}w_iw_j}}`$ (31)
$`r_{ij}`$ $`=`$ $`{\displaystyle \frac{z^{}w_iw_j}{z^{}w_iw_j+z^{}x_iy_j}}`$ (32)
and governing separately the various expected degrees
$$k_i^{}=\underset{j}{}p_{ij}^{}k_i^{}=\underset{j}{}p_{ij}^{}k_i^{}=\underset{j}{}p_{ij}^{}$$
(33)
The possibility to control the above degrees independently of each other is a remarkable advantage of our model. Even if all real directed networks display a nontrivial reciprocity structure , the modeling of dynamical processes is mostly performed on purely directed or purely undirected networks . However, it has been recently shown that the dynamics of percolation and epidemic spreading crucially depends on the degree sequencies $`\{k_i^{}\}`$, $`\{k_i^{}\}`$ and $`\{k_i^{}\}`$, its general properties being different from the simpler behaviour studied on purely undirected networks (where $`k_i^{}=k_i^{}=0`$ $`i`$) or purely directed networks (where $`k_i^{}=0`$ $`i`$). It has also been shown that on scale-free networks bidirectional links act as โpercolation catalystsโ , since even an infinitely small fraction of them determines a phase transition with the onset of a giant strongly connected component. Thus our model provides a way to generate random networks with an explicit reciprocity structure, where dynamical processes can be studied more realistically.
The most general case with arbitrary fitness-dependent probabilities $`p_{ij}^{}=p^{}(x_i,y_j)`$, $`p_{ij}^{}=p^{}(w_i,w_j)`$ represents the *hidden variable model with reciprocity*. Each vertex is now characterized by *three* quantities $`x`$, $`y`$, $`z`$ determining its expected degrees through eqs.(33). As a first example, we consider $`x_i=y_i=w_ii`$ and
$$p_{ij}^{}=\frac{z^{}x_ix_j}{1+(2z^{}+z^{})x_ix_j}p_{ij}^{}=\frac{z^{}x_ix_j}{1+(2z^{}+z^{})x_ix_j}$$
(34)
where $`z^{}`$ and $`z^{}`$ are defined as above. This model reproduces perfectly all the topological properties of the World Trade Web (WTW), where $`x_i`$ is identified with the total Gross Domestic Product of each country $`i`$. To see this, note that in this case the conditional probability (22) turns out to be constant and equal to the reciprocity $`r`$ of the network, which is an empirical property observed in each snapshot of the real WTW :
$$r_{ij}=r=\frac{1}{1+e^{(\mu ^{}\mu ^{})}}=\frac{z^{}}{z^{}+z^{}}$$
(35)
Then note that if we regard the network as undirected by drawing an undirected link between two vertices $`i`$ and $`j`$ if they are connected by *at least* one directed link in *any* direction, the probability of such an undirected link is
$$q_{ij}=p_{ij}^{}+p_{ij}^{}+p_{ij}^{}=\frac{(2z^{}+z^{})x_ix_j}{1+(2z^{}+z^{})x_ix_j}$$
(36)
and the above expression is exactly the one which in ref. we showed to reproduce all the topological properties of the undirected WTW. In other words, eqs.(34) describe completely the topology of the WTW, including its reciprocity. We recall that, having the form of a Fermi function, $`q_{ij}1`$ for large $`x_ix_j`$, which implies the โquantum effectโ that the WTW is not scale-free even if $`x`$ is empirically found to be power-law distributed . Note that this model can also be obtained from eq.(29) setting $`\alpha _i=\beta _i=\gamma _i`$ and introducing the *undirected degree* $`k_i`$ measured on the undirected version of the graph:
$$H=\underset{i}{}\alpha _i(k_i^{}+k_i^{}+k_i^{})=\underset{i}{}\alpha _ik_i$$
(37)
Our second example concerns shareholding networks (SN). In ref. it has been shown that the โordinaryโ hidden-variable model successfully reproduces the properties of SN if the unconditional probability has the form $`p_{ij}=p(x_i,y_j)=y_j^\beta f(x_i)`$ where $`y_j`$ is the wealth invested by the agent $`j`$ and $`x_i`$ is the information associated to the asset $`i`$. On the other hand, in ref. we showed that the SN for NYSE and NASDAQ have no reciprocated links, a property not reproduced by the above form for $`p_{ij}`$. Our model reproduces all these properties by setting
$$p_{ij}^{}=y_j^\beta f(x_i)p_{ij}^{}=0$$
(38)
We recall that, unlike the WTW, SN are in the โclassical limitโ where the empirical power-law distribution of $`y`$ is reflected in a scale-free degree distribution .
We finally propose an interpretation of eq.(25) in terms of a โchemical reactionโ converting the chemical species $`()`$, $`()`$ and $`()`$ into each other. Let us first consider our system when $`\lambda =0`$. Since in this case the graphs $`G_5`$ and $`G_6`$ in fig.1 have the same statistical weight, their โparticlesโ must be connected through the following chemical reaction which is at equilibrium:
$$(AB)+(CD)=(AB)+(BD)+(AC)$$
(39)
The condition for equilibrium is obtained by replacing in the above expression each chemical species with its chemical potential, which gives $`\mu ^{}=2\mu ^{}`$, consistently with eq.(25) since $`\lambda =0`$. When $`\lambda 0`$ the graphs $`G_5`$ and $`G_6`$ have different statistical weights, meaning that the above chemical reaction occurs with the release of an additional โenergyโ $`\lambda `$ such that $`\mu ^{}=2\mu ^{}+\lambda `$ as in eq.(25). When $`\lambda >0`$ the reaction is โesothermicโ and the production of reciprocated links is energetically favoured, while when $`\lambda <0`$ the reaction is โendothermicโ and the production of reciprocated links is suppressed.
We have introduced the first theoretical model reproducing the nontrivial properties of real networks including their reciprocity. Our results provide an improved characterization of the network topology and a basis for the investigation of the effects of reciprocity on network dynamics. These ideas, inspired by the Fermi statistics of multispecies systems, can be directly generalized to networks with different types of links (in preparation).
|
warning/0506/math0506109.html
|
ar5iv
|
text
|
# ON O-MINIMALITY OF EXTENSIONS OF โ BY RESTRICTED GENERIC SMOOTH FUNCTIONS.
( 26.4.2005)
## Abstract
It is shown that the extension of $``$ by a generic smooth function restricted to the unit cube is o-minimal. The generalization to countably many generic smooth functions is indicated. Possible applications are sketched.
0. Introduction.
A key result in the theory of subanalytic sets is Gabrielovโs theorem (\[Gab\]), which states that the complement of a subanalytic set is again subanalytic. One form to state this fact in the framework of mathematical logic, is as follows. Denote by $``$ the language $`\{0,1,+,,,<\}`$. For each function $`f`$ which is real analytic on a neighbourhood of the closed unit cube of $`^n`$ for some $`n`$, add to $``$ the symbol $`\widehat{f}`$. Denote the language obtained by $`L_{an}`$, and denote by $`_{an}`$ the set of the real numbers seen as a $`L_{an}`$ structure, with each of the symbols $`\widehat{f}`$ being interpreted as a function which is equal to $`f`$ on the unit cube and to $`0`$ outside. Denote by $`T_{an}`$ the theory of $`_{an}`$. Gabrielov theorem implies
Theorem \[vdD1\]. $`T_{an}`$ is model complete and o-minimal.
Our purpose here is to prove an analogue of this theorem for extensions of $``$ by (countably many) generic smooth functions restricted to the unit cube. To avoid overly complex notations, the argument is written down in detail for the case of extension by a single generic smooth function. We then comment on the generalization to countably many generic smooth functions.
The initial motivation for obtaining such a theorem came from trying to answer some questions about generic smooth control systems; the potential usefullness of such result comes from the fact that o-minimality implies strong regularity properties of definable sets and maps (\[vdD2\]). In section 4, another possible application is sketched.
To state our theorem precisely we introduce some notations. Fix $`n^+`$, and let $`\widehat{D^\alpha f}`$, $`\alpha =\alpha _1..\alpha _n(^0)^n`$, be $`n`$-ary function symbols indexed by the multiindex $`\alpha `$. Let $`_{\widehat{Df}}`$ denote the language obtained by adjoining the symbols $`\widehat{D^\alpha f}`$, $`\alpha =\alpha _1..\alpha _n(^0)^n`$, to the language $`=\{0,1,+,,,<\}`$. Let $`fC^{\mathrm{}}(^n,)`$. We make $``$ into $`_{\widehat{Df}}`$-structure by giving the symbols $`0,1,+,,,<`$ the usual interpretation, and by interpreting $`\widehat{D^\alpha f}`$, $`\alpha =\alpha _1..\alpha _n(^0)^n`$, as functions which are equal to the corresponding partial derivatives of the given function $`f`$ on $`[1,1]^n`$, and are equal to zero outside $`[1,1]^n`$. We denote this structure and its theory by $`_{\widehat{Df}}`$, $`T_{\widehat{Df}}`$ respectively. We call a subset of a Baire topological space $`X`$ residual if it contains a countable intersection of open dense subsets of $`X`$. Our result is
Theorem A. There exists a residual subset $`RC^{\mathrm{}}(^n,)`$, such that for each $`fR`$, the theory $`T_{\widehat{Df}}`$ is model complete and o-minimal.
Moreover, let $`\widehat{D^\alpha f_1},\widehat{D^\alpha f_2},..`$ be $`n_1,n_2,..`$-ary respectively function symbols for each $`\alpha =\alpha _1..\alpha _n(^0)^n`$, and denote by $`_{(\widehat{Df_i})_i}`$, the language obtained by adjoining to $`\{0,1,+,,,<\}`$ these function symbols. Let $`f_iC^{\mathrm{}}(^{n_i},)`$, $`i=1,2,..`$ . We make $``$ into $`_{(\widehat{Df})_i}`$-structure, and we denote by $`T_{(\widehat{Df})_i}`$ the theory of this structure. One may generalize the proof of Theorem A and obtain
Theorem B. There exists a residual subset $`R_iC^{\mathrm{}}(_i^n,)`$, such that $`(f_i)_iR`$, $`T_{(\widehat{Df})_i}`$ is model complete and o-minimal.
Let us make a comment on the proof. In section 5 of \[IY\], a related problem was considered, roughly corresponding to estimating the complexity of preimages of semialgebraic sets under jet extensions of generic smooth maps. For their purposes, the authors construct a special equisingular Whitney stratification of such preimage, and obtain the estimate sought from a certain numerical characteristic of this stratification. This characteristic is shown to be well defined by using Thomโs First Isotopy Lemma. We use here similar considerations, but develop them further: it is first established that for a generic function $`f`$, subsets which are defined by quantifier free $`_{\widehat{Df}}`$ formulas admit equisingular, in some sense, Whitney stratifications. Then we use the Isotopy Lemma to show that projections of such sets admit cylindrical decomposition.
We now outline the structure of the article. In section 1 we introduce Whitney stratifications of subsets of $`^k`$, whose strata project with constant rank on each coordinate subspace; we call such stratifications monotonic. In section 2 we study a certain subclass of quantifier free $`\mathrm{}`$-definable subsets from which each quantifier free $`\mathrm{}`$-definable set can be obtained by projection. For each set from this subclass we construct, using \[GrY\] and a construction of Thom-Boardman type, a monotonic Whitney stratification whose strata are again sets from this subclass. This result allows us to construct a cylindrical decomposition for projections of such sets in section 3. As a corollary we obtain model completeness together with o-minimality, proving Theorem A. Finally, in section 4 we comment on the generalization to countably many generic smooth functions, and sketch possible applications.
We have used above, and will be using below, some elementary logic-theoretic notions. For these, we refer the reader to the very readable and concise introductory notes \[Ch\], whose terminology we follow here, or to any other introductory text on the topic. For more details on o-minimal structures see \[vdD2\]. Among articles which could be relevant to the theme discussed here we mention \[KM\], \[W1\], \[W2\], \[RSW\].
1. Compact Whitney stratified subsets of $`^n`$ with monotonic strata.
Below we summarize some facts about Whitney stratifications (for more details and proofs see \[Ma1\], \[Ma2\] or \[GWPL\], Chapters 1 and 2; the exposition here follows \[GWPL\]). The adjective โsmoothโ will mean $`C^{\mathrm{}}`$, though we need only $`C^1`$ or sometimes $`C^2`$ smoothness for the facts stated below. By a submanifold of $`^n`$ we mean an embedded smooth submanifold, not necessarily connected, all of whose components have the same dimension. Let $`X,Y^n`$ be submanifolds; $`X`$ is said to be Whitney regular over $`Y`$ at $`yY`$, if for any two sequences $`(x_i)`$ in $`X`$ and $`(y_i)`$ in $`Y`$, for which $`x_iy_ii`$, $`x_iy`$, $`y_iy`$, such that the sequences of lines $`\overline{x_iy_i}`$ and of tangent spaces $`T_{x_i}X`$ converge, we have $`lim\overline{x_iy_i}limT_{x_i}X`$. $`X`$ is said to be Whitney regular over $`Y`$ if $`yY`$, $`X`$ is Whitney regular over $`Y`$ at $`y`$. We may in fact take instead of $`^n`$ an arbitrary smooth manifold, and define Whitney regularity via local charts. If $`X`$ is Whitney regular over $`Y`$, then either $`\overline{XY}Y=\mathrm{}`$ or $`dim(Y)<dim(X)`$. A stratification $``$ of $`M^n`$ is a locally finite partition of $`M`$ into submanifolds, called strata. It is said to be a Whitney stratification if each stratum is Whitney regular over any other stratum. We say that a stratified set has dimension $`m`$, if $`m`$ is the maximal dimension of its strata. We denote by $`^q`$ the subset of strata of $``$ of dimension smaller or equal to $`q`$, and denote the union of such strata by $`M^q`$.
The Cartesian product of two Whitney stratified subsets is again a Whitney stratified subset. The stratification $``$ is said to satisfy the frontier condition if $`X,Y,\overline{X}Y\mathrm{}`$ implies $`\overline{X}Y`$. A map is said to be transversal to $``$ if it is transversal to each stratum of $``$. In this case, if $``$ is Whitney, the preimages of the strata constitute again a Whitney stratification.
Any ($`\mathrm{}`$-definable) semialgebraic set admits a finite Whitney stratification with ($`\mathrm{}`$-definable) semialgebraic strata.
Our key analytical tool is Thomโs Fisrt Isotopy Lemma. Let $`M`$ be a subset of a smooth manifold $`N`$ with a stratification $``$. Let $`f:NP`$ be a smooth map into another smooth manifold $`P`$. We say that $`(M,)`$ is (topologically) trivial over $`P`$ if there exist a stratified set $`F`$ with a stratification $``$, and a homeomorphism $`h:MP\times F`$, such that the following holds: each stratum of $`M`$ is sent to a stratum of $`P\times `$, and $`\pi h=f`$, where $`\pi :P\times FP`$ is the projection. We say that $`(M,)`$ is (topologically) locally trivial over $`P`$ if each $`pP`$ has a neighbourhood $`V`$, such that $`f^1(V)`$ is again a stratification, and $`(f^1(V)M,f^1(V))`$ is trivial over $`f^1(V)`$.
Thomโs First Isotopy Lemma. (\[GWPL\], Chapter II, Theorem 5.2) Let $`(M,)`$ be a locally closed Whitney stratified subset of the smooth manifold $`N`$, and let $`f:NP`$ be a smooth map such that for each $`S`$, $`f|_S`$ is a submersion and $`f|_{\overline{S}M}`$ is a proper map. Then $`(M,)`$ is locally trivial over $`P`$.
We will sometimes make use of the following fact:
Proposition 1.1 (\[GWPL\], Chapter II, Theorem 5.6, Corollary 5.7) Let $`N`$ be a smooth manifold, and let $`MN`$ be a closed set with a Whitney stratification $``$. Then the components of strata of $``$ form another Whitney stratification, which moreover satisfies the frontier condition.
Let the coordinates on $`^n`$ be denoted by $`x_1,..,x_n`$. By a coordinate plane in $`^n`$ we mean any set of the form $`\{(x_1,..,x_n)^n:x_{i_1}=0,..,x_{i_m}=0\}`$, for some choice of $`mn`$ and $`1i_1<..<i_mn`$. We say that a submanifold $`S^n`$ is monotonic, if the projection of $`S`$ on any coordinate plane in $`^n`$ is a map of constant rank. We call a stratification of a subset of $`^n`$ monotonic if all its strata are monotonic. Let $`M^k`$ be a set with a monotonic stratification $``$, and let $`P`$ be a coordinate plane in $`^k`$. We denote by $`M(q,P)`$ the union of strata of $`M`$ which project on $`P`$ with rank smaller or equal to $`q`$, and by $`(q,P)`$ the corresponding stratification.
Proposition 1.2. Let $`M^k`$ be a closed Whitney stratified set and let $`q0`$ be an integer. Then $`M^q`$ is a closed set. Suppose further that the strata of $``$ are monotonic, and let $`P`$ be a coordinate plane in $`^k`$. Then the set $`M(q,P)`$ is closed as well.
Proof. The set $`M^q`$ is closed since $`M`$ is closed and the closure of a stratum of dimension $`d`$, in a Whitney stratification, can only intersect strata of dimension smaller than $`d`$. The set $`M(q,P)`$ is closed since $`M`$ is closed and because of the following property of Whitney stratifications. Namely, suppose $`y_i`$ is a sequence of points on a stratum $`S`$ converging to a point $`x`$ on a stratum $`S^{}`$, and suppose that the sequence $`T_{y_i}S`$ converges to a subspace $`T`$. Then $`T_xS^{}T`$. $`\mathrm{}`$
Lemma 1.3. Let $`S`$ be a monotonic submanifold of $`^n`$, let $`\pi :^n^k`$ denote the projection on a coordinate plane $`^k^n`$, and let $`x^k`$. Then the set $`\pi ^1(x)S`$ is a monotonic submanifold of $`^n`$.
Proof. Basically, an exercise in linear algebra. $`\mathrm{}`$
Lemma 1.4. Let $`M^n`$ be a compact set with a monotonic Whitney stratification $``$ whose strata are connected. Then the closure of each stratum contains a $`0`$-dimensional stratum.
Proof. Let $`S`$, $`dim(S)>0`$. The stratum $`S`$ cannot be closed, since then it would be compact, and its projection on any coordinate axis would have a critical point. This would imply, since $`S`$ is monotonic, that the rank of each such projection is zero, and consequently $`dim(S)=0`$, which contradicts our assumption. The set $`M`$ is closed, hence $`\overline{S}`$ must intersect another stratum $`S^{}`$ of $``$. Since $``$ is Whitney, $`dim(S^{})<dim(S)`$. Since the strata of $``$ are connected, $``$ satisfies the frontier condition (Proposition 1.1), and thus $`S^{}\overline{S}`$. The submanifold $`S^{}`$ is monotonic and $`\overline{S^{}}\overline{S}`$, so we may repeat the argument and eventually conclude that $`\overline{S}`$ contains a $`0`$-dimensional stratum. $`\mathrm{}`$
The following two lemmas about monotonic Whitney stratifications will be used in section 3.
Lemma 1.5. Let $`M^n`$ be a compact set with a monotonic Whitney stratification $``$ whose strata are connected, and let $`\pi :^n`$ be the projection on a one dimensional coordinate plane of $`^n`$. Let $`^{}`$ be a subset of $``$, and let $`M^{}`$ be the union of strata of $`^{}`$. Then there exists a set $`๐ฏ`$ of $`1`$-dimensional strata of $``$, and a set $`๐ซ`$ of $`0`$-dimensional strata of $``$, such that the following is true:
i) for each $`S๐ฏ`$, $`rank(\pi _S)=1`$,
ii) the boundary of $`\pi (S)`$, for each each $`S๐ฏ`$, is contained in $`\pi (M^0)`$,
iii) the projections of $`_{S๐ฏ}S`$ and $`_{S๐ซ}S`$ are disjoint and their union is equal to $`\pi (M^{})`$.
Proof. Let $`C=\pi (M^0)`$. For each $`pC`$, $`\pi ^1(p)M`$ has the monotonic Whitney stratification $`_p=\{\pi ^1(p)S:S\}`$, whose strata are monotonic by Lemma 1.3. By Lemma 1.4, $`p\pi (_p^0)`$. Therefore for each $`S^{}^{}`$, $`\pi (S^{})C`$ is contained in $`\pi (L)`$ where $`L`$ ranges over $`1`$-dimensional strata of $`\overline{S^{}}`$. Since the strata of $``$ are connected, $``$ satisfies the frontier condition (Proposition 1.1), and thus $`\pi (S^{})C`$ is in fact equal to $`\pi (L)C`$. If we take $`๐ฏ`$ to consist of $`L`$ such that $`dim(L)=1`$, $`L\overline{M^{}}`$, $`rank(\pi |_L)=1`$, and we take $`๐ซ`$ consist of $`P^0`$ such that $`\pi (P)\pi (M^{})_{L๐ฏ}\pi (L)`$, we see that i) and iii) hold. Using Lemma 1.4, one shows that ii) holds as well. $`\mathrm{}`$
Let $`f:NP`$ be a smooth map, and let $`MN`$ have a Whitney stratification $``$. We say that $`pP`$ is a regular value of $`f|_M`$, if $`p`$ is a regular value of each $`f|_S`$, $`S`$.
Lemma 1.6. Let $`\pi :^n^k`$ be the projection of a compact set $`M^n`$ to a coordinate plane of $`^n`$. Suppose that $`M`$ admits a monotonic Whitney stratification. Then the number of components of $`\pi ^1(x)M`$ is uniformly bounded over $`x^k`$.
Proof Choose some monotonic Whitney stratification of $`M`$ and denote it by $``$. By taking, if necessary, the connected components of strata, we may assume that the strata are connected and thus $``$ satisfies the frontier condition (Proposition 1.1). Let $`AM`$. We denote by $`N_x(A)`$, $`x^k`$ the number of components of $`\pi ^1(x)A`$. Observe that
$$N_x(M)\underset{S}{}N_x(\overline{S}).$$
Take a stratum $`S`$ and denote the rank with which it projects to $`^k`$ by $`r(S)`$. By Lemma 1.3, $`\pi ^1(x)S`$ is a monotonic submanifold of $`^n`$, and thus by Lemma 1.4 none of its components can be compact unless $`dim(S)=r(S)`$. If $`dim(S)>r(S)`$, each component of $`\pi ^1(x)\overline{S}`$ must therefore intersect the frontier of $`S`$, defined as $`fr(S)=\overline{S}S`$. We conclude that in the case $`dim(S)>r`$, the number of components of $`\pi ^1(x)\overline{S}`$ is bounded by the number of components of $`\pi ^1(x)fr(S)`$, which, by the frontier condition, is itself bounded by $`_{S^{}fr(S)}N(\overline{S^{}})`$. The dimension of $`S^{}fr(S)`$ must be strictly smaller than $`dim(S)`$ (since the stratification is Whitney). Repeating the argument as many times as needed, we may bound $`N_x(M)`$ by a sum of $`N_x(\overline{S})`$ over strata $`S`$ for which $`dim(S)=r(S)`$ (each term may appear more than once in this sum).
Thus, to prove that the number of components of $`\pi ^1(x)M`$ is uniformly bounded over $`x^k`$, it is sufficient to show, that for each $`S`$ for which $`dim(S)=r(S)`$, the cardinality of $`\pi ^1(x)S`$ is uniformly bounded over $`x`$. Since one may find a coordinate plane in $`^k`$ of dimension $`dim(S)`$, on which $`S`$ projects with rank $`dim(S)`$, we may assume that $`S`$ maps to $`^k`$ by a local diffeomorphism.
Let $`^k=^{k1}\times `$, and consider the projection $`\pi ^{}:^n^{k1}`$. Denote by $`E^{k1}`$ the image of strata of $`\overline{S}`$ which map with rank smaller than $`k1`$ to $`^{k1}`$, and denote by $`\pi ^{\prime \prime }:^k^{k1}`$ the projection from $`^k`$ to $`^{k1}`$. For all $`y^{k1}E`$, $`(\pi ^{})^1(y)\overline{S}`$ is then a one dimensional Whitney stratified set, with monotonic strata, which we denote by $`(\overline{S})_y`$. It looks like a graph, with vertices and edges (with possibly multiple edges connecting a pair of vertices), each edge being a one dimensional submanifold. The number of vertices of the graph is bounded by the number of regular preimages of $`y`$ in $`(\overline{S})^{k1}`$ under $`\pi ^{}`$. It is a consequence of Thomโs First Isotopy Lemma, that the maximal degree $`D`$ of the vertices is bounded uniformly over all $`y^{k1}E`$ (see the proof of Theorem 5.1 in \[IY\]). Moreover, each edge of the graph must land in at least one vertex of the graph, since the edges are monotonic and bounded. Let $`L`$ be an edge which projects with rank 1 to the line $`(\pi ^{\prime \prime })^1(y)`$. There can be only one preimage of $`x(\pi ^{\prime \prime })^1(y)`$ in $`L`$, since otherwise there would be a point on L with tangent which is parallel to the plane $`\pi ^1(x)(\pi ^{})^1(y)`$, contradicting monotonicity of $`L`$.
Denote by $`Z_y(\pi ^{\prime \prime })^1(y)`$ the finite set of projections of vertices of the graph. Thus, for all $`x(\pi ^{\prime \prime })^1(y)Z_y`$, $`y^{k1}E`$, the number of preimages of $`x`$ in $`\overline{S}`$ is bounded by the product of the maximal graph degree $`D`$ and the number of regular preimages of $`y^{k1}`$ in $`(\overline{S})^{k1}`$. Suppose that the latter is uniformly bounded over $`^{k1}E`$. Then there exists a dense set $`R^k`$, such that the number of preimages of $`x`$ in $`\overline{S}`$ is uniformly bounded over $`xR`$, say by $`N>0`$. If there existed $`x_0^k`$ with more than $`N`$ regular preimages, this would be true for every $`x`$ in a sufficiently small neighbourhood of $`x_0`$, thus contradicting the density of $`R`$. Thus, if the number of regular preimages of $`y^{k1}`$ in $`(\overline{S})^{k1}`$ is uniformly bounded, then also the number of regular preimages of $`x^k`$ in $`\overline{S}=(\overline{S})^k`$ is uniformly bounded. Proceeding by induction, we conclude that the number of regular preimages of $`x^k`$ in $`\overline{S}=(\overline{S})^k`$ is indeed uniformly bounded over $`^k`$.
By the reduction made, this proves the lemma. $`\mathrm{}`$
2. Basic $`f`$-sets and their monotonic Whitney stratifications.
In this section we study the quantifier free $`\mathrm{}`$-definable sets of the structure $`_{\widehat{Df}}`$. The class of sets which we now introduce, the $`\mathrm{}`$-definable basic $`f`$-sets, will be shown to form a subclass of the class of $`\mathrm{}`$-definable quantifier free sets, with the property that each $`\mathrm{}`$-definable quantifier free set is a projection of a $`\mathrm{}`$-definable basic $`f`$-set. It will be also shown that each $`\mathrm{}`$-definable basic $`f`$-set has a finite Whitney stratification, whose strata are monotonic and are $`\mathrm{}`$-definable basic $`f`$-sets themselves. We use this in section 3 to establish a cylindrical decomposition result for projections of $`\mathrm{}`$-definable quantifier free sets.
Below, $`J^m(^n,)`$ denotes the space of $`m`$-jets of functions from $`^n`$ to $``$, and $`j^mf(x)`$ denotes the $`m`$-th jet of a smooth function $`f:^n`$ at $`x^n`$. We denote by $`J(r,m,n)`$ the $`rth`$ power of $`J^m(^n,)`$, which we may identify with the space of all tuples $`(j^mf(x_1),..,j^mf(x_r))`$, $`fC^{\mathrm{}}(^n,)`$, $`(x_1,..,x_r)(^n)^r`$. We call it the $`(r,m)`$-multijet space (compare with \[GG\], page 57; our definition is different since we do not require the diagonal in $`(^n)^r`$ to be excluded). We call the map $`j^{r,m}f:(^n)^rJ(r,m,n)`$, defined by $`(x_1,..,x_r)(j^mf(x_1),..,j^mf(x_r))`$, the multijet extension of $`f`$. For our purposes we identify $`J(r,m,n)`$ with a Euclidean space of corresponding dimension.
One encounters multijet preimages of $`\mathrm{}`$-definable semialgebraic sets among the quantifier free $`\mathrm{}`$-definable sets of the structure $`_{\widehat{Df}}`$ (for example, let $`fC^{\mathrm{}}(,)`$, and consider the set $`\{(x_1,x_2)[1,1]^2:f(x_1)>f(x_2)\}`$). Since our aim is to construct special Whitney stratifications for quantifier free $`\mathrm{}`$-definable sets, we at least should be able to produce Whitney stratifications for generic multijet preimages of semialgebraic sets. If we try to use the Multijet Transversality Theorem as stated in \[GG\] (Chapter II, Theorem 4.13), we see that it only implies that for a generic function $`f`$ the multijet preimage of a semialgebraic set in $`(^n)^rDiag`$ is Whitney stratifiable, and not that the preimage in $`(^n)^r`$ itself is Whitney stratifiable (we denote by $`Diag`$ the diagonal). This particular problem was addressed in \[GrY\], the result of which we state below. We need to introduce first the notion of divided differences (for more details see, for instance, \[BZ\]).
Definition 2.1. Let $`fC^{\mathrm{}}(,)`$, and let $`k^0`$. The divided difference of order $`k`$ of $`f`$, which we denote by $`\mathrm{\Delta }^kf`$, is a function from $`^{k+1}`$ to $``$ defined as follows. For $`k=0`$, $`\mathrm{\Delta }^0f=f`$. For $`k>0`$, let $`(x_1,..,x_{k+1})^{k+1}`$ be such that $`x_ix_j`$ $`ij`$. Then $`\mathrm{\Delta }^kf(x_1,..,x_{k+1})`$ is defined by the following recursion relation:
$$\mathrm{\Delta }^kf(x_1,..,x_{k+1})=(\mathrm{\Delta }^{k1}f(x_1,..,x_k)\mathrm{\Delta }^{k1}f(x_2,..,x_{k+1}))/(x_1x_{k+1}).$$
It can be shown that this defines, by continuity, a smooth function from $`^{k+1}`$ to $``$, which is in fact symmetric. There exists an explicit formula, easily proved by induction, to compute the divided difference at points lying on the diagonal, i.e. at points $`(x_1,..,x_{k+1})`$ for which there exist $`i,j`$, $`ij`$, such that $`x_i=x_j`$. For stating this formula, we introduce the following notation. Let $`m`$ be a nonnegative integer; denote by $`\delta ^m:^{m+1}`$ the map which sends $`x`$ to the tuple $`(x,..,x)^{m+1}`$, all of whose entries are equal to $`x`$.
Proposition 2.2. Let $`fC^{\mathrm{}}(,)`$. Then
$$\mathrm{\Delta }^{m_1+..+m_l+l1}f(\delta ^{m_1}(u_1),..,\delta ^{m_l}(u_l))=\frac{1}{m_1!..m_l!}D^{m_1..m_l}(\mathrm{\Delta }^{l1}f(u_1,..,u_l)).$$
Remark. Note that since divided differences are symmetric functions of their arguments, one may indeed use Proposition 2.2 to compute $`\mathrm{\Delta }^kf`$ at any given point of $`^{k+1}`$. Moreover, Proposition 2.2 implies that there exist $`t>k`$ and a semialgebraic function $`\rho :J(k,t,1)`$, so that $`\mathrm{\Delta }^kf=\rho j^{k,t}f`$.
One may introduce divided differences also for functions of several variables. Let $`fC^{\mathrm{}}(^n,)`$, $`f=f(x_1,..,x_n)`$, and fix $`1in`$. We denote by $`\mathrm{\Delta }_{x_i}^kf`$ the function which we obtain by taking the divided difference of $`f`$ of order $`k`$ w.r.t. the variable $`x_i`$, while keeping the other variables fixed. Denote the variables on which $`\mathrm{\Delta }_{x_i}^kf`$ depends by $`x_1,..,x_{i1},x_{i1},..,x_{ik+1},x_{i+1},..,x_n`$. We may now take a divided difference w.r.t. another variable $`x_j`$, $`ji`$, and repeat this operation w.r.t. other variables.
Definition 2.3. Let $`fC^{\mathrm{}}(^n,)`$, $`f=f(x_1,..,x_n)`$, and let $`(\alpha _1,..,\alpha _n)(^0)^n`$. The divided difference of order $`(\alpha _1,..,\alpha _n)`$ of $`f`$, denoted by $`\mathrm{\Delta }^{\alpha _1..\alpha _n}f`$, is the function defined as
$$\mathrm{\Delta }^{\alpha _1..\alpha _n}f=\mathrm{\Delta }_{x_1}^{\alpha _1}..\mathrm{\Delta }_{x_n}^{\alpha _n}f.$$
The function $`\mathrm{\Delta }^{\alpha _1..\alpha _n}f`$ depends on $`\alpha _1+..+\alpha _n+n`$ variables, which we denote by $`x_{11},..,x_{1\alpha _1+1},`$ $`\mathrm{},`$ $`x_{n1},..,x_{n\alpha _n+1}`$. We make the following convention: $`\mathrm{\Delta }^{\alpha _1..\alpha _n}f(X_1,..,X_n)`$, where $`X_1,..,X_n`$ are tuples of real numbers, means that we substitute for $`x_{i1},..,x_{i\alpha _i+1}`$ the first $`\alpha _i+1`$ entries of the tuple $`X_i`$, for each $`1in`$.
Let $`m,r`$ be nonnegative integers, $`r>0`$. Denote by $`diag^m:^r^{(m+1)r}`$ the map which sends $`(x_1,..,x_r)`$ to $`(\delta ^m(x_1),..,\delta ^m(x_r))`$, and by $`\stackrel{~}{๐}^{r,m}f:(^n)^r^{(r(m+1))^n}`$ the map which sends $`(x_{11},..,x_{1r},\mathrm{},x_{n1},..,x_{nr})`$ to the collection of
$$\mathrm{\Delta }^{\alpha _1..\alpha _n}f(diag^m(x_{11},..,x_{1r}),..,diag^m(x_{n1},..,x_{nr})),$$
ordered lexicographically on multiindices, where $`0\alpha _ir(m+1)1`$ for each $`1in`$. Denote by $`๐^{r,m}f:(^n)^r^{nr+(r(m+1))^n}`$ the map which sends $`u(^n)^r`$ to $`(u,\stackrel{~}{๐}^{r,m}f(u))`$, and call it the divided difference extension of $`f`$. Below, we denote by $`\stackrel{~}{D}(r,m,n)=^{(r(m+1))^n}`$ the target space of $`\stackrel{~}{๐}^{r,m}f`$, and by $`D(r,m,n)=^{nr+(r(m+1))^n}`$ the target space of $`๐^{r,m}f`$.
Theorem 2.4 \[GrY\].
i) Let $`MD(r,m,n)`$ be a submanifold. There exists a residual set $`RC^{\mathrm{}}(^n,)`$, such that $`fR`$, $`๐^{r,m}f`$ is transversal to $`M`$.
ii) there exists a polynomial map with rational coefficients $`\beta :D(r,m,n)J(r,m,n)`$, such that $`j^{r,m}f=\pi ๐^{r,m}f`$.
Let us write down a simple example. Take $`fC^{\mathrm{}}(,)`$; then $`(x_{11},x_{12})`$ is mapped by $`j^{2,0}f`$ to $`(x_{11},f(x_{11}),x_{12},f(x_{12}))`$, and by $`๐^{2,0}f`$ to $`(x_{11},x_{12},f(x_{11}),(f(x_{11})f(x_{12}))/(x_{11}x_{12}))`$. In this case the map $`\beta `$ is a birational morphism; this is not true for $`n2`$.
Note that the theorem implies that for any semialgebraic set $`SD(r,m,n)`$ there exists a residual subset $`RC^{\mathrm{}}(^n,)`$, such that for each $`fR`$, $`(j^{r,m}f)^1(S)`$ has a Whitney stratification. (Indeed, choose a Whitney stratification of $`\beta ^1(S)`$. According to the theorem there exists a residual set $`RC^{\mathrm{}}(^n,)`$, such that for each $`fR`$, $`๐^{r,m}f`$ is transversal to each of the strata of this Whitney stratification. Now recall that transversal preimages of Whitney stratified sets have a natural Whitney stratification.) Just having a Whitney stratification, however, is not sufficient, as we would also like the strata to be monotonic. It is this requirement which does not allow us (unless $`n=1`$) to limit ourselves to the subclass of multijet preimages of semialgebraic sets, if we wish the strata to be again sets from this subclass. It turns out that a suitable way to proceed is to consider instead the subclass of preimages of semialgebraic sets under divided difference extensions of $`f`$. More precisely:
Definition 2.5. Let $`fC^{\mathrm{}}(^n,)`$. We call $`M^p\times (^n)^r`$, $`p0`$, $`n,r1`$, a $`\mathrm{}`$-definable basic $`f`$-set, if there exists $`m0`$, and a $`\mathrm{}`$-definable semialgebraic set $`A^p\times D(r,m,n)=^p\times (^n)^r\times \stackrel{~}{D}(r,m,n)`$, such that
i) $`M=(id\times ๐^{r,m}f)^1(A)`$,
ii) $`A`$ is a subset of $`^p\times [1,1]^{nr}\times \stackrel{~}{D}(r,m,n)`$.
Here, the map $`id\times ๐^{r,m}f`$ maps $`(u,v)^p\times (^n)^r`$ to $`(u,๐^{r,m}f(v))`$. The requirement ii) corresponds to restricting $`f`$ to the unit cube; it implies that $`M`$ is a subset of $`^p\times [1,1]^{nr}`$.
Remark. Note that the numbers $`p,n,r`$ are not determined by the dimension of the ambient space of $`M`$. When we speak below of $`\mathrm{}`$-definable basic $`f`$-sets inside some ambient space $`^q`$, we always provide a representation of $`^q`$ in the form $`^p\times (^n)^r`$, from which these numbers can be determined.
Let $`\stackrel{~}{w}^{r,m}f:(^n)^r\stackrel{~}{W}(r,m,n)`$, where $`\stackrel{~}{W}(r,m,n)`$ is a Euclidean space, denote the map whose components are $`D^{\alpha _1..\alpha _n}f(x_{1i_1},..,x_{ni_n})`$, $`0\alpha _1,..,a_nt`$, $`1i_1,..,i_nr`$. Denote by $`w^{r,t}f`$ the map which sends $`u(^n)^r`$ to $`(u,\stackrel{~}{w}^{r,t}f(u))(^n)^r\times \stackrel{~}{W}(r,t,n)`$. Denote $`(^n)^r\times \stackrel{~}{W}(r,t,n)`$ by $`W(r,t,n)`$.
Lemma 2.6. Let $`fC^{\mathrm{}}(^n,)`$, and let $`m0,r>0`$.
i) There exist $`tm`$ and a $`\mathrm{}`$-definable semialgebraic map
$`\rho _{r,t,n}:W(r,t,n)D(r,m,n)`$, such that $`๐^{r,m}f=\rho _{r,t,n}w^{r,t}f`$.
ii) there exists a polynomial map with rational coefficients $`\sigma :D(r,m,n)W(r,m,n)`$, so that $`w^{r,m}f=\sigma ๐^{r,m}f`$.
Remark. Note that although the definition of $`w^{r,m}f`$ may seem similar to that of $`j^{r,m}f`$, it is not the same map, and in fact part ii) of Lemma 2.6 is stronger than part ii) of Theorem 2.4. The semialgebraic map $`\rho _{r,t,n}`$ whose existence is asserted in part i), is not continuous on all of $`W(r,t,n)`$.
Sketch of the proof. Part i) is implied by Proposition 2.2 and by symmetricity of divided differences of functions of one variable. Part ii) follows from the method of proof of Theorem 2.4 ii) in \[GrY\] (see \[GrY\], Remark on pg. 359). $`\mathrm{}`$
Lemma 2.7. Let $`fC^{\mathrm{}}(^n,)`$. Then each $`\mathrm{}`$-definable basic $`f`$-set is a quantifier free $`\mathrm{}`$-definable set of $`_{\widehat{Df}}`$.
Sketch of the proof. Take a $`\mathrm{}`$-definable basic $`f`$-set $`M`$. By definition, there exist nonnegative integers $`r,m,n,p`$ and a $`\mathrm{}`$-definable semialgebraic set $`A^p\times D(r,m,n)`$, such that $`M=(id\times ๐^{r,m}f)^1(A)`$, and such that $`A`$ is a subset of $`^p\times [1,1]^{nr}\times \stackrel{~}{D}(r,m,n)`$. By Lemma 2.6 i) there exist $`t0`$ and a $`\mathrm{}`$-definable semialgebraic map $`\rho _{r,t,n}:W(r,t,n)D(r,m,n)`$, such that $`๐^{r,m}f=\rho _{r,t,n}w^{r,t}f`$. Thus it is also true that $`M=((id\times \rho _{r,t,n})(id\times w^{r,t}f))^1(A)`$, hence $`M=(id\times w^{r,t}f)^1((id\times \rho _{r,t,n})^1(A))`$.
If we show that $`C=(id\times \rho _{r,t,n})^1(A)`$ is a subset of $`^p\times [1,1]^{nr}\times \stackrel{~}{W}(r,t,n)`$, we are done, since this allows us to write a formula in $`_{\widehat{Df}}`$ for the set $`M`$ (indeed, $`C`$ is a $`\mathrm{}`$-definable semialgebraic set, and $`(id\times w^{r,t}f)^1(C)`$ can then be defined by a quantifier free formula involving $`\widehat{D^\alpha f}`$). But since $`๐^{r,m}f=\rho _{r,t,n}w^{r,t}f`$, $`\rho _{r,t,n}`$ is the identity map on the first $`nr`$ coordinates. Hence, the fact that $`A`$ is a subset of $`^p\times [1,1]^{nr}\times \stackrel{~}{D}(r,m,n)`$ implies that $`(\rho _{r,t,n})^1(A)^p\times [1,1]^{nr}\times \stackrel{~}{W}(r,t,n)`$. $`\mathrm{}`$
We wish to show that all quantifier free $`\mathrm{}`$-definable sets of $`_{\widehat{Df}}`$ are in fact projections of $`\mathrm{}`$-definable basic $`f`$-sets (and therefore the class of projections of quantifier free $`\mathrm{}`$-definable sets is identical to the class of projections of $`\mathrm{}`$-definable basic $`f`$-sets). We will need the notion of depth of quantifier free formulas, which we now define for an arbitrary given language $`L`$. We say that a term has depth $`0`$ if it is a variable or a constant of $`L`$. Consider the term $`f(t_1,..,t_s)`$, where $`f`$ is a s-ary function symbol of $`L`$, and $`t_1,..,t_s`$ are also terms. We say that $`f(t_1,..,t_s)`$ has depth $`i`$, if the maximal depth of the terms $`t_1,..,t_n`$ is equal to $`i1`$. The depth of an atomic formula is defined as the maximal depth of the terms on which it depends. We say that a quantifier free formula, which is a boolean combination of atomic formulas, has depth $`i`$, if $`i`$ is the maximal depth of the atomic formulas of which the boolean combination is formed.
Lemma 2.8. Let $`fC^{\mathrm{}}(^n,)`$. Then each quantifier free $`\mathrm{}`$-definable set of $`_{\widehat{Df}}`$ is the projection of a $`\mathrm{}`$-definable basic $`f`$-set on a coordinate plane.
Sketch of the proof. One shows using induction, that for each quantifier free formula $`\varphi (u_1,..,u_s)`$ of any language $`L`$ there exists an equivalent existential formula
$$v_1..v_t(\psi (u_1,..,u_q,v_1,..,v_t),$$
where $`\psi (u_1,..,u_q,v_1,..,v_t)`$ is a quantifier free formula of depth at most $`1`$ (i.e. not involving compositions of functions). Moreover, $`\psi `$ can be so chosen that all its terms of depth $`1`$ do not have constants as arguments, and all depend on pairwise disjoint groups of variables. Applying this observation to the language $`_{\widehat{Df}}`$, one further shows that in fact the formula $`\psi `$ can be chosen in such a way, that it defines the preimage of a $`\mathrm{}`$-definable semialgebraic set $`S^p\times J(r,m,n)`$ under $`id\times j^{r,m}f`$, for some $`r,m,n,p`$. Moreover, if we denote by $`\pi :J(r,m,n)(^n)^r`$ the projection on a subspace of $`J(r,m,n)`$ with the property that $`\pi j^{r,m}f(u)=uu(^n)^r`$, then $`\pi (S)[1,1]^{nr}`$. By Theorem 2.4 ii), there exists a polynomial map $`\beta `$ with rational coefficients from $`D(r,m,n)`$ to $`J(r,m,n)`$, such that $`j^{r,m}f=\beta ๐^{r,m}f`$. Thus
$$(id\times j^{r,m}f)^1(S)=(id\times ๐^{r,m}f)^1((id\times \beta )^1(S)).$$
Since $`\pi (S)[1,1]^{nr}`$, $`(id\times \beta )^1(S)`$ is a subset of $`^p\times [1,1]^{nr}\times \stackrel{~}{D}(r,m,n)`$. Moreover, since $`\beta `$ is a polynomial map with rational coefficients, $`(id\times \beta )^1(S)`$ is a $`\mathrm{}`$-definable semialgebraic set. Therefore, the quantifier free formula $`\psi `$ defines a $`\mathrm{}`$-definable basic $`f`$-set. $`\mathrm{}`$
We illustrate the steps which were not detailed in the proof of Lemma 2.8 in a simple case. Let $`n=1`$, and consider the set defined by the depth $`2`$ formula $`\widehat{f}(\widehat{f}(x_1))>0`$ (we write $`\widehat{f}`$ instead of the $`_{\widehat{Df}}`$ symbol $`\widehat{D^0f}`$). Note that this set can be defined by the following depth $`1`$ formula
$$x_2(\widehat{f}(x_2)>0\widehat{f}(x_1)=x_2),$$
By further transformation, we arrive at the equivalent formula
$$x_2x_3x_4(\widehat{f}(x_3)>0|x_3|1|x_4|1x_2=x_3x_2=0|x_1|>1$$
$$\widehat{f}(x_3)>0|x_3|1|x_4|1x_2=x_3x_2=\widehat{f}(x_4)x_1=x_4),$$
which is the preimage under $`id\times j^{2,0}f`$ of a semialgebraic subset of $`^2\times [1,1]^2\times ^2`$.
We now aim to show that for a generic smooth function $`fC^{\mathrm{}}(^n,)`$, every $`\mathrm{}`$-definable basic $`f`$-set has a Whitney monotonic stratification whose strata are $`\mathrm{}`$-definable basic $`f`$-sets. By โgenericโ we mean that $`f`$ is such that for all $`p,r,m`$, $`id\times ๐^{r,m}f`$ is transversal to any $`\mathrm{}`$-definable semialgebraic submanifold of $`^p\times D(r,m,n)`$. We wish to show that the set of such functions, which we denote by $`R_n`$, is residual.
Proposition 2.9. For each $`n1`$, $`R_n`$ is a residual set.
Proof. Fix the numbers $`p,r,m`$, and let $`N^p\times D(r,m,n)`$ be a submanifold. It is a simple generalization of Theorem 2.4, that the set $`Q(p,r,m,N)`$ of functions $`fC^{\mathrm{}}(^n,R)`$ for which $`id\times ๐^{r,m}f`$ is transversal to $`N`$, is residual. The set $`R_n`$ is the intersection of the sets $`Q(p,r,m,N)`$, as $`p,r,m`$ range over nonnegative integers, and $`N`$ ranges over $`\mathrm{}`$-definable semialgebraic submanifolds of $`^p\times D(r,m,n)`$. Since the family of all $`\mathrm{}`$-definable semialgebraic submanifolds is countable, we conclude that $`R_n`$ is residual. $`\mathrm{}`$
We will also need the following two facts.
Proposition 2.10. Let the set $`M^k`$ have a Whitney stratification $``$ of dimension $`m`$. Let $`๐`$ be a Whitney stratification of $`M^q`$, $`qm`$, which refines $`^q`$. Then $`(^q)๐`$, a stratification of $`M`$ which refines $``$, is Whitney.
Proof. If $`X,Y`$ are submanifolds and $`X`$ is Whitney over $`Y`$, then $`X`$ is necessarily Whitney over any submanifold of $`Y`$. If moreover $`X,Y`$ are disjoint and $`dim(X)dim(Y)`$ then $`\overline{X}Y=\mathrm{}`$, and therefore any submanifold of $`X`$ is Whitney over $`Y`$. The proof follows from these facts. $`\mathrm{}`$
Proposition 2.11. Let $`A_1,..,A_l^k`$ be $`\mathrm{}`$-definable semialgebraic sets. Then there exists a finite Whitney stratification $`๐`$ of $`A_1..A_l`$, whose strata are $`\mathrm{}`$-definable semialgebraic sets, such that for each stratum $`T๐`$ and $`1il`$, either $`TA_i`$ or $`TA_i=\mathrm{}`$.
Proof. This is a well known fact. $`\mathrm{}`$
Lemma 2.12. Let $`fC^{\mathrm{}}(^n,)`$. Fix the numbers $`r,m`$, and consider the map $`๐^{r,m}f:(^n)^rD(r,m,n)`$. There exists $`t>m`$, such that each partial derivative of first order of any of the components of $`๐^{r,m}f`$ is a composition of a $`\mathrm{}`$-definable semialgebraic function on $`D(r,t,n)`$ with $`๐^{r,t}f:(^n)^rD(r,t,n)`$.
Sketch of the proof. Let
$$\mathrm{\Delta }^{\alpha _1..\alpha _n}f(diag^m(x_{11},..,x_{1r}),..,diag^m(x_{n1},..,x_{nr})),$$
($`0\alpha _jr(m+1)1`$, $`j=1,..,n`$), be a given component of the map $`๐^{r,m}f`$. Apply Proposition 2.2 separately to each variable of $`fC^{\mathrm{}}(^n,)`$. This allows us to represent the given component in the form:
$$cD_{x_{n1}..x_{nw_n+1}}^{m..ml_n}\mathrm{\Delta }_{x_n}^{w_n}\mathrm{}D_{x_{11}..x_{1w_1+1}}^{m..ml_i}\mathrm{\Delta }_{x_1}^{w_1}f$$
evaluated on $`(x_{11},..,x_{1w_1+1},\mathrm{},x_{n1},..,x_{nw_n+1})`$. Here $`(m+1)w_i+l_i=\alpha _i`$, $`0l_im`$, $`i=1,..,n`$, and $`c`$ is some rational constant. Apply the partial derivative operator $`_{x_{pq}}`$, $`1pn`$, $`1qw_p+1`$; it commutes with other partial derivative operators and with $`\mathrm{\Delta }_{x_n}^{w_n},..,\mathrm{\Delta }_{x_{p+1}}^{w_{p+1}}`$, so one may write the result as
$$cD_{x_{n1}..x_{nw_n+1}}^{m..ml_n}\mathrm{\Delta }_{x_n}^{w_n}\mathrm{}D_{x_{p1}..x_{pq}..x_{1w_1+1}}^{m..(m+1)..ml_i}\mathrm{\Delta }_{x_p}^{w_p}\mathrm{}D_{x_{11}..x_{1w_1+1}}^{m..ml_i}\mathrm{\Delta }_{x_1}^{w_1}f.$$
Note that this expression will not in general be in the form of a component of $`\mathrm{\Delta }^{r,t}f`$ for some $`t>m`$. However, up to a multiplication by a rational constant, it is a divided difference of $`f`$ (we apply again Proposition 2.2, this time in the opposite direction). Consequently, it can be shown (by the same argument which proves Lemma 2.6 i)) to be the composition of a $`\mathrm{}`$-definable semialgebraic function $`\rho `$, defined on $`W(r,t,n)`$, and of $`w^{r,t}f`$, for some $`t>m`$.
By Lemma 2.6 ii), there exists a polynomial map with rational coefficents $`\sigma :D(r,t,n)W(r,t,n)`$, so that $`\sigma ๐^{r,t}f=w^{r,t}f`$. Therefore, $`\rho w^{r,t}f=\rho \sigma ๐^{r,t}f`$. Since $`\rho \sigma `$ is a $`\mathrm{}`$-definable semialgebraic map, this proves the lemma. $`\mathrm{}`$
Remark. Note that in the proof sketch we wrote the partial derivative of a divided difference as a composition of a semialgebraic function and of $`w^{r,t}f`$. We do not know a way to write it as a composition of a semialgebraic function and of $`j^{r,t}f`$ (for $`fC^{\mathrm{}}(^n,)`$ with $`n>1`$). This is the reason why we introduce additional complexity by taking as basic $`f`$-sets the preimages of semialgebraic sets under divided differences extensions, rather than taking the preimages of semialgebraic sets under multijet extensions, which are simpler.
Below, $`\pi _{t,m}`$, $`tm`$, denotes the natural projection from $`D(r,t,n)`$ to $`D(r,m,n)`$ (for which $`\pi _{t,m}๐^{r,t}f=๐^{r,m}f`$).
Lemma 2.13. Let $`fR_nC^{\mathrm{}}(^n,)`$. Let $`S^p\times D(r,m,n)`$ be a $`\mathrm{}`$-definable semialgebraic set, with a finite $`\mathrm{}`$-definable semialgebraic partition $`๐ฎ`$. There exist $`t>m`$ and a finite $`\mathrm{}`$-definable semialgebraic Whitney stratification $`๐ซ`$ of $`(id\times \pi _{t,m})^1(S)^p\times D(r,t,n)`$, which refines $`(id\times \pi _{t,m})^1(๐ฎ)`$, such that for each $`fR_n`$:
i) the partition $`(id\times ๐^{t,m}f)^1(๐ซ)`$ is a finite Whitney stratification which refines the partition $`(id\times ๐^{r,m}f)^1(๐ฎ)`$,
ii) the upper dimensional strata of $`(id\times ๐^{t,m}f)^1(๐ซ)`$ are monotonic.
Proof. Let $`fR_n`$. By Proposition 2.11, we may refine $`๐ฎ`$ to a finite Whitney stratification $`๐ฎ^{}`$ whose strata are $`\mathrm{}`$-definable semialgebraic sets. Since $`fR_n`$, $`(id\times ๐^{r,m}f)^1(๐ฎ^{})`$ is a finite Whitney stratification. By Lemma 2.12, the Jacobian of $`id\times ๐^{r,m}f`$ is a composition of a $`\mathrm{}`$-definable semialgebraic map on $`^p\times D(r,t,n)`$ and of $`id\times ๐^{r,t}f`$ for some $`t>m`$. This can be seen to imply that for each coordinate plane $`V`$ of $`^p\times (^n)^r`$ there exists a $`\mathrm{}`$-definable semialgebraic map $`\gamma _V:^p\times D(r,t,n)`$, such that the rank of the projection to $`V`$ of the tangent plane to the preimage of $`๐ฎ`$ at point $`(v,u)^p\times (^n)^r`$, is given by $`\gamma _V(id\times ๐^{r,t}f)`$ (if $`(v,u)`$ is not in the preimage of $`S`$, we take the rank to be equal to $`1`$). Since there are only finitely many coordinate planes in $`^p\times (^n)^r`$, we may refine $`\pi _{t,m}^1(๐ฎ^{})`$ into a $`\mathrm{}`$-definable semialgebraic partition $`๐ฌ`$, such that on each element of $`๐ฌ`$ the value of $`\gamma _V`$, for each coordinate plane $`V`$, is constant. By Proposition 2.11, there exists a finite Whitney stratification $`๐ซ`$ which refines $`๐ฌ`$, whose strata are $`\mathrm{}`$-definable semialgebraic sets. Since $`fR_n`$, $`๐=(id\times ๐^{r,t}f)^1(๐ซ)`$ is a Whitney stratification.
In general, we cannot expect the strata of $`๐`$ to be monotonic. Indeed, let $`T๐ซ`$ and let $`Y\pi _{t,m}^1(๐ฎ^{})`$ be the stratum of $`๐ฎ^{}`$ which contains $`T`$. Take any coordinate plane $`V`$ in $`^p\times (^n)^r`$. Let $`A=(id\times ๐^{r,t}f)^1(T)`$ and let $`X=(id\times ๐^{r,t}f)^1(Y)`$. By construction, the rank of the projection of $`X`$ to $`V`$ is constant on $`A`$, which however does not imply that that the projection of $`A`$ itself to $`V`$ is a constant rank map. Nevertheless, this is true if $`dim(X)=dim(A)`$. In particular, this means that the upper dimensional strata of $`๐`$ are monotonic. $`\mathrm{}`$
We get to the main point of this section.
Theorem 2.14. Let $`fR_n`$. Let $`S^p\times D(r,m,n)`$ be a $`\mathrm{}`$-definable semialgebraic set, with a finite $`\mathrm{}`$-definable semialgebraic partition $`๐ฎ`$. Then the set $`(id\times ๐^{r,m}f)^1(S)`$ has a finite Whitney stratification which refines $`(id\times ๐^{r,m}f)^1(๐ฎ)`$, and whose strata are monotonic and are $`\mathrm{}`$-definable basic $`f`$-sets in $`^p\times (^n)^r`$.
Proof. Let $`M=(id\times ๐^{r,m}f)^1(S)`$. We make the following induction assumption. For each positive integer $`i`$ there exist integers $`t_j`$, $`\mathrm{}`$-definable semialgebraic sets $`P_j^p\times D(r,t_j,n)`$ with finite Whitney stratifications $`๐ซ_j`$, $`j=1,..,i`$, and a semialgebraic set $`S_i^p\times D(r,t_i,n)`$ with a finite Whitney stratification $`๐ฎ_i`$, for which the following holds:
i) the strata of $`๐ซ_j`$ are of equal dimension; denoting their codimension by $`c_j`$, we have $`c_1<c_2<..<c_i<codim(๐ฎ_i)`$,
ii) the strata of the stratifications $`๐ซ_1,..,๐ซ_i,๐ฎ_i`$ are $`\mathrm{}`$-definable semialgebraic sets,
iii) the strata (recall that $`fR_n`$) of $`(id\times ๐^{r,t_j}f)^1(๐ซ_j)`$ are monotonic,
iv) the union of $`(id\times ๐^{r,t_i}f)^1(๐ฎ_i)`$ and of $`(id\times ๐^{r,t_j}f)^1(๐ซ_j)`$, $`j=1,..,i`$ is a Whitney stratification of $`M`$.
Note that the induction claim is true for $`i=1`$ by Lemma 2.13. Assume that the induction claim is true for $`i=N1`$. By Lemma 2.13, there exists $`t_N>t_{N1}`$ and a finite Whitney stratification $`๐ซ`$ of $`(id\times \pi ^{t_N,t_{N1}})^1(S_{N1})`$, which refines $`(id\times \pi ^{t_N,t_{N1}})^1(๐ฎ_{N1})`$ and whose strata are $`\mathrm{}`$-definable semialgebraic sets, with the additional property that the upper dimensional strata of $`๐=(id\times ๐^{r,t_N}f)^1(๐ซ)`$ are monotonic. Take $`๐ซ_N`$ to be the collection of the upper dimensional strata of $`๐ซ`$ and take $`๐ฎ_N`$ to be the collection of the rest of strata of $`๐ซ`$. Note that this choice satisfies i),ii),iii) for $`i=N`$. The union $`๐`$ of $`(id\times ๐^{r,t_N}f)^1(๐ซ_N)`$ and $`(id\times ๐^{r,t_N}f)^1(๐ฎ_N)`$ is a Whitney stratification which refines $`(id\times ๐^{r,t_{N1}}f)^1(๐ฎ_{N1})`$. Since iv) holds for $`i=N1`$ and $`dim(S_{N1})<dim(P_j)`$, $`j=1,..,N1`$, Proposition 2.10 implies that iv) holds for $`i=N`$ as well.
Thus the induction claim is true. Since the codimension of $`S_i`$ grows with $`i`$, for some $`i=I`$ its codimension will become larger than $`dim(^p\times (^n)^r)`$. Since $`fR_n`$, this means that $`(id\times ๐^{r,t_N}f)^1(๐ฎ_I)`$ is empty. Thus the collection of strata of $`(id\times ๐^{r,t_N}f)^1(๐ซ_i)`$, $`i=1,..,I`$, forms, according to iv) and iii), a Whitney stratification $``$ of $`M`$ whose strata are monotonic. According to ii), each stratum of $``$ is a $`\mathrm{}`$-definable basic $`f`$-set. $`\mathrm{}`$
Theorem 2.14 has the following immediate corollary.
Corollary 2.15. Let $`fR_n`$. Each $`\mathrm{}`$-definable basic $`f`$-set has a finite Whitney stratification whose strata are monotonic and are $`\mathrm{}`$-definable basic $`f`$-sets. $`\mathrm{}`$
3. Cylindrical decomposition of projections of basic $`f`$-sets.
Let $`fR_nC^{\mathrm{}}(^n,)`$ (for definition of $`R_n`$, see section 2). In this section we intend to show that projections of basic $`f`$-sets on coordinate planes admit cylindrical decomposition whose cells are again projections of basic $`f`$-sets. This then allows to prove Theorem A stated in Introduction. We first establish some auxiliary facts. Denote by $`\pi _{^k,P}`$ the projection from $`^k`$ to a coordinate plane $`P`$ in $`^k`$.
Definition 3.1. We call $`X^k`$ a $`\mathrm{}`$-definable $`f`$-set, if there exists a $`\mathrm{}`$-definable basic $`f`$-set $`S^p\times (^n)^r`$, for some $`p,r,n0`$, such that $`X`$ is the projection of $`S`$ on a coordinate plane of $`^p\times (^n)^r`$.
Lemma 3.2. Let $`X,Y^k`$ be $`\mathrm{}`$-definable $`f`$-sets. Then their intersection and union are $`\mathrm{}`$-definable $`f`$-sets as well.
Proof. The sets $`X,Y`$, being $`\mathrm{}`$-definable $`f`$-sets, can be defined by existential formulas of the language $`_{\widehat{Df}}`$. Therefore, their union and intersection can be also defined by existential formulas of $`_{\widehat{Df}}`$. By Lemma 2.8, both sets are then the projections on coordinate planes of $`\mathrm{}`$-definable basic $`f`$-sets. $`\mathrm{}`$
Lemma 3.3. Let $`X[1,1]^k`$ be a $`\mathrm{}`$-definable $`f`$-set. Then there exists a $`\mathrm{}`$-definable basic $`f`$-set $`S^p\times (^n)^r`$ contained in $`[1,1]^(p+nr)`$, for some $`p,r,n0`$, such that $`X`$ is the projection of $`S`$ on a coordinate plane of $`^p\times (^n)^r`$.
Proof. Since $`X`$ is a $`\mathrm{}`$-definable $`f`$-set, there exist a $`\mathrm{}`$-definable basic $`f`$-set $`S_0^p^{}\times (^n)^r`$, for some nonnegative integers $`p^{},r`$, and a $`k`$-dimensional coordinate plane $`P`$ in $`^p^{}\times (^n)^r`$, such that $`X`$ can be obtained as the projection of $`S_0`$ to $`P`$. By definition, $`S_0`$ is the preimage of a semialgebraic set $`A^p^{}\times D(r,m,n)`$ under $`id\times ๐^{r,m}f:^p^{}\times (^n)^r^p^{}\times D(r,m,n)`$, for some $`m0`$, such that $`A^p^{}\times [1,1]^{nr}\times \stackrel{~}{D}(r,m,n)`$. If $`S_0[1,1]^{p^{}+nr}`$, we may take $`S=S_0`$. If not, choose a coordinate $`x_i`$ from $`^p^{}\times (^n)^r^{p^{}+nr}`$, such that the projection of $`S_0`$ on the $`x_i`$ axis is not contained in $`[1,1]^{p^{}+nr}`$. Note that necessarily $`1ip^{}`$, since $`S_0`$ is a subset of $`^p^{}\times [1,1]^{nr}`$ (because $`A^p^{}\times [1,1]^{nr}\times \stackrel{~}{D}(r,m,n,)`$). The projection of $`^{p^{}+nr}`$ to $`P`$ must be along the coordinate $`x_i`$, since $`X[1,1]^k`$. Denote by $`V_1`$ the coordinate plane in $`^p^{}\times (^n)^r`$, obtained by setting $`x_i=0`$. It is not difficult to check (since $`1ip^{}`$) that the projection of $`S_0`$ on $`V_1\times (^n)^r`$ is given by the preimage of the projection of $`A`$ on $`P\times D(r,m,n)`$, under $`id\times ๐^{r,m}:V_1\times (^n)^rV_1\times D(r,m,n)`$. Since the projection of a $`\mathrm{}`$-definable semialgebraic set is a $`\mathrm{}`$-definable semialgebraic set, we conclude that there exists a $`\mathrm{}`$-definable basic $`f`$-set $`S_1^{p^{}1}\times (^n)^r`$, such that $`X`$ is the projection of $`S_1`$ on the coordinate plane $`P`$.
We may repeat this argument, getting a sequence of $`\mathrm{}`$-definable basic $`f`$-sets $`S_jV_j\times (^s)^r`$, $`dim(V_j)=p^{}j`$, such that for each $`j`$, $`X`$ is the projection of $`S_j`$ on $`P`$. We may do so until $`S_j`$ becomes a subset of $`[1,1]^{p^{}j+nr}`$, which occurs after $`lp^{}`$ steps. Now take $`S=S_l`$. $`\mathrm{}`$
Lemma 3.4. Let $`S^p\times (^n)^r`$ be a $`\mathrm{}`$-definable basic $`f`$-set, contained in $`[1,1]^{p+nr}`$. Then there exists a finite monotonic Whitney stratification $``$ of $`[1,1]^{p+nr}`$, whose strata are $`\mathrm{}`$-definable basic $`f`$-sets, such that $`S`$ is stratified by a subset of $``$.
Proof. By definition, $`S`$ is the preimage of a $`\mathrm{}`$-definable semialgebraic set $`A^p\times D(r,m,n)`$ under $`id\times ๐^{r,m}f`$, for some $`m0`$. Since $`S[1,1]^{p+nr}`$, we may assume that $`A`$ is contained in $`[1,1]^{p+nr}\times \stackrel{~}{D}(r,m,n)`$. Consider the partition $`๐ซ`$ of $`[1,1]^{p+nr}\times \stackrel{~}{D}(r,m,n)`$ whose elements are $`A`$ and its complement, and note that the preimage of $`[1,1]^{p+nr}\times \stackrel{~}{D}(r,m,n)`$, under $`id\times ๐^{r,m}f`$, is $`[1,1]^{p+nr}`$. By Theorem 2.14, there exists a finite Whitney stratification $``$ which refines $`(id\times ๐^{r,m}f)^1(๐ซ)`$, and whose strata are monotonic and are $`\mathrm{}`$-definable basic $`f`$-sets. Since $``$ refines $`(id\times ๐^{r,m}f)^1(๐ซ)`$, there is a subset of $``$ which stratifies $`S`$. $`\mathrm{}`$
Let $`X`$ be a $`\mathrm{}`$-definable $`f`$-set. Then there exists a $`\mathrm{}`$-definable basic $`f`$-set $`S`$, such that $`X`$ is the projection of $`S`$ on some coordinate plane $`P`$. By Corollary 2.15 $`S`$ admits a finite stratification with monotonic strata. We define the dimension of $`X`$ to be equal to the maximal rank with which the strata of this stratification project on $`P`$ (one may check that this definition does not depend on the choice of $`S`$ and its stratification).
We now state the cylindrical decomposition result.
Theorem 3.5. Let $`A^s=^{s1}\times `$ be a $`\mathrm{}`$-definable $`f`$-set, with the coordinates on $`^{s1}\times `$ being denoted by $`(x,t)`$. Then there exists a partition $``$ of $`^{s1}`$ into finitely many connected $`\mathrm{}`$-definable $`f`$-sets, such that $`B`$ there is a finite family of continuous functions $`g_i:B`$, $`i=0,1,..,l_B+1`$
$$g_0(x)\mathrm{}<g_1(x)<\mathrm{}<g_{s_B}(x)<g_{l_B+1}(x)+\mathrm{},$$
with the property that the family of sets of the form
$$\{(x,t):xB,g_i(x)<t<g_{i+1}(x)\}$$
(โstripe setsโ), and sets of the form
$$\{(x,t):xB,t=g_i(x)\}$$
(โgraph setsโ), constitutes a partition $`๐ฅ`$ of $`^s`$ into finitely many connected $`\mathrm{}`$-definable $`f`$\- sets, and there exists a subset $`๐ฅ^{}`$ of $`๐ฅ`$ which constitutes a partition of $`A`$.
Proof. The proof of the theorem will be by induction on the dimension $`s`$. In the case $`s=1`$, the theorem just says that a $`\mathrm{}`$-definable $`f`$-set $`A`$ consists of finitely many components. If $`A[1,1]`$, then by Lemma 3.3 there exists a $`\mathrm{}`$-definable basic $`f`$-set $`Q[1,1]^q^p\times (^n)^r`$, $`q=p+nr`$, such that $`A`$ is the projection of $`Q`$ on a coordinate plane of $`^p\times (^n)^r`$. By Lemma 3.4, there exists a finite Whitney stratification of $`[1,1]^q`$, a subset of which stratifies $`Q`$. The connected components of the strata form again a finite Whitney stratification. Therefore $`Q`$, and thus also its projection, consist of finitely many components. In the case that $`A[1,1]`$, $`A=(A[1,1])(A[1,1])`$. Since by Lemma 3.2 $`A[1,1]`$ is a $`\mathrm{}`$-definable $`f`$-set, we only have to show that $`A[1,1]`$ has finitely many components. Since $`A[1,1]=A([1,1])`$, by Lemma 3.2 it is a $`\mathrm{}`$-definable $`f`$-set. It is not difficult to check that the map $`x1/x`$ maps $`A[1,1]`$ into another $`\mathrm{}`$-definable $`f`$-set, which is now contained in $`[1,1]`$. Therefore $`A[1,1]`$, and thus $`A`$, have finitely many components.
We make the induction assumption that the theorem is true in all dimensions smaller than $`s`$. Suppose first that $`A[1,1]^s`$. Let us make the following ad hoc definition.
Definition 3.5.1. We say that a $`\mathrm{}`$-definable $`f`$-set $`G[1,1]^{s1}^{s1}`$ projects well on a coordinate plane $`P`$ of $`^{s1}`$ if the following holds:
i) $`G`$ projects injectively on $`P`$, and $`\pi _{^{s1},P}(G)P`$ is open,
ii) there exist $`p,r0`$, a finite monotonic Whitney stratification $``$ of the unit cube $`[1,1]^q^p\times (^s)^r`$, $`q=p+sr`$, whose strata are $`\mathrm{}`$-definable basic $`f`$-sets, and a set $`J[1,1]^q`$, stratified by a subset of $``$, such that $`xG`$,
$$\pi _{^s,^{s1}}^1(x)A=\pi _{^q,^s}\left(\pi _{^q,P}^1\left(\pi _{^{s1},P}(x)\right)J\right),$$
iii) each $`y\pi _{^{s1},P}(G)`$ is a regular value of $`\pi _{^q,P}|_{}`$.
We now make two claims, whose proof we defer until later.
Claim 3.5.2. Suppose $`A[1,1]^s`$. If the theorem holds in all dimensions smaller than $`s`$, then there exists a partition of $`\pi _{^s,^{s1}}(A)`$ into finitely many $`\mathrm{}`$-definable $`f`$-sets $`G_1,..,G_N`$, such that for each $`G_i`$, $`i=1,..,N`$, there exists a coordinate plane $`P_i`$ of $`^{s1}`$ on which $`G_i`$ projects well.
Claim 3.5.3. Suppose that the theorem holds in all dimensions smaller than $`s`$, and let $`G^k`$, $`k<s`$, be a $`\mathrm{}`$-definable $`f`$-set. Then $`G`$ has a finite number of components, and each of them is a $`\mathrm{}`$-definable $`f`$-set. Moreover, if $`F_1,..,F_N^k`$ are $`\mathrm{}`$-definable $`f`$-sets, then $`GF_1..F_N`$ is a $`\mathrm{}`$-definable $`f`$-set.
We intend to show, assuming Claims 3.5.2 and 3.5.3 that the induction assumption holds also in dimension $`s`$. Fix $`i`$, $`1iN`$. Since $`G_i`$ projects well on $`P_i`$, $`U=\pi _{^{s1},P_i}(G_i)P_i`$ is open, and there exist $`p,r0`$, a finite Whitney stratification $``$ of the unit cube $`[1,1]^q^p\times (^n)^r`$, $`q=p+nr`$, whose strata are monotonic $`\mathrm{}`$-definable basic $`f`$-sets, and a subset $`๐ฅ`$ of this stratification, such that, writing $`J=_{S๐ฅ}S`$,
$$\pi _{^s,^{s1}}^1(x)A=\pi _{^q,^s}\left(\pi _{^q,P}^1\left(\pi _{^{s1},P}(x)\right)J\right)$$
holds for each $`xG_i`$. Moreover, each $`yU`$ is a regular value of $`\pi _{^q,P}|_{}`$. Note that $`_y^{dimP_i}=\pi _{^q,P_i}^1(y)^{dimP_i}`$ consists of isolated points and is compact. Thus $`_y^{dimP_i}`$ consists of a finite number of points $`yU`$. Fix nonnegative integers $`j,k_1,..,k_j`$. Let $`U_j^{k_1,..,k_j}`$ be the subset of points $`yU`$, for which the points of $`\pi _{^q,^s}\left(_y^{dimP_i}\right)`$ project to precisely $`j`$ distinct points on $`^{s1}`$, $`x_1,..,x_j`$, ordered, say, lexicographically, and the cardinality of $`\pi _{^q,^s}\left(_y^{dimP_i}\right)\pi _{^s,^{s1}}^1(x_l)`$ is equal to $`k_l`$ for each $`l=1,..,j`$. These sets form a partition of $`U`$, which we show now to be finite.
Indeed, observe that $`\pi _{^q,P_i}^1(U)^{dimP_i}`$ is a closed Whitney stratified subset of the manifold $`\pi _{^q,P_i}^1(U)`$, and that $`\pi _{^q,P_i}`$ is a submersion on each stratum $`T`$ of $`\pi _{^q,P_i}^1(U)^{dimP_i}`$. Moreover, the map $`\pi _{^q,P_i}|_{\overline{T}}:\overline{T}U`$, where $`\overline{T}`$ denotes the closure of $`T`$ in $`\pi _{^q,P_i}^1(U)`$, is proper. By the Isotopy Lemma, the fibers $`_y^{dimP_i}`$ are homeomorphic over connected components of $`U`$. Since the fibers are compact and consist of isolated points, there exists $`K>0`$ such that each fiber consists of not more than $`K`$ points. Thus the number of nonempty $`U_j^{k_1,..,k_j}`$ sets is finite. The set $`U`$ is a $`\mathrm{}`$-definable $`f`$-set, so it can be defined by an existential $`_{\widehat{Df}}`$ formula. Since the stratification $``$ is finite and consists of $`\mathrm{}`$-definable basic $`f`$-sets, one may write a suitable $`_{\widehat{Df}}`$ formula for each $`U_j^{k_1,..,k_j}`$ and conclude, by Claim 3.5.3, that each $`U_j^{k_1,..,k_j}P_i`$ is a $`\mathrm{}`$-definable $`f`$-set.
By Claim 3.5.3 again, $`U_j^{k_1,..,k_j}`$ has finitely many components each of which is a $`\mathrm{}`$-definable $`f`$-set. Take any such component and denote it by $`H`$. Denote by $`G_{i,H}^{s1}`$ the set $`\pi _{^{s1},P_i}^1(H)G_i`$. We take $`g_l:G_{i,H}`$, $`l=1,..,h`$ ($`h`$ is equal to one of $`k_1,..,k_j`$) to be the functions which send $`xG_{i,H}`$ to the projections on the $`t`$-axis of the points of
$$\pi _{^q,^s}\left(_y^{dimP_i}\right)\pi _{^s,^{s1}}^1(x),$$
$`y=\pi _{^{s1},P_i}(x)`$, ordered by magnitude. Observe that the corresponding โstripeโ and โgraphโ sets are again $`\mathrm{}`$-definable $`f`$-sets.
Since each $`yH`$ is a regular value of $`\pi _{^q,P_i}|_{}`$, the set $`G_{i,H}`$ has the following property, implied by the Isotopy Lemma. Namely, the points of $`\pi _{^q,^s}\left(_y^{dimP_i}\right)`$, $`y=\pi _{^{s1},P_i}(x)`$, vary continuously as $`x`$ varies in $`G_{i,H}`$. Let $`a(x),b(x)`$ be two such points, such that $`\pi _{^q,^{s1}}(a(x_0))=\pi _{^q,^{s1}}(b(x_0))=x_0`$ for some $`x_0G_{i,H}`$. Then it follows from the definition of the set $`U_j^{k_1,..,k_j}`$ and the fact that $`H`$ is connected, that $`\pi _{^q,^s}(a(x))=\pi _{^q,^s}(b(x))=x`$ for all $`xG_{i,H}`$. In fact, the projections of $`a(x)`$ and $`b(x)`$ on the line $`\pi _{^s,^{s1}}^1(x)`$ will be either equal for all $`xG_{i,H}`$, or distinct for all $`xG_{i,H}`$. This implies in particular that the functions $`g_l`$, $`l=1,..,h`$ are continuous.
Denote by $`_y^{}`$, $`y=\pi _{^{s1},P_i}(x)`$, the stratification obtained by taking the components of strata of $`_y`$. Note that the set $`J_y=\pi _{^q,^s}^1(y)J`$ projects under $`\pi _{^q,^s}`$ on $`\pi _{^s,^{s1}}^1(x)`$, and is stratified by $`\pi _{^q,^s}^1(y)๐ฅ_y`$. Since the frontier condition is satisfied for $`_y^{}`$, $`\overline{J_y}`$ is stratified by a subset of $`_y^{}`$, and has therefore a monotonic Whitney stratification with connected strata which we denote by $`\overline{๐ฅ_y}`$. Since $`J_y`$ is compact and $`\pi _{^q,^s}(\overline{J_y})\pi _{^s,^{s1}}^1(x)`$, there exist, by Lemma 1.5, a subset $`๐ฏ_x\overline{๐ฅ_y}`$ of $`1`$-dimensional strata, which project with rank $`1`$ to $`\pi _{^s,^{s1}}^1(x)`$, and a subset $`๐ซ_x\overline{๐ฅ_y}`$ of $`0`$-dimensional strata, with the following property. Namely, the sets $`T_x=_{S๐ฏ_x}\pi _{^q,^s}(S)`$ and $`P_x=_{S๐ซ_x}\pi _{^q,^s}(S)`$ are disjoint and their union is equal to $`\pi _{^s,^{s1}}^1(x)A`$. Moreover, the boundary points of the projections of strata of $`๐ฏ_x`$ are projections of points from $`_y^{dimP_i}`$. The Isotopy Lemma implies that the projections of the sets $`T_x`$ and $`P_x`$ vary continuously (in the Hausdorff metric) as $`x`$ varies in $`G_{i,H}`$. Together with the fact that each two continuously varying points from $`_y^{dimP_i}`$, $`y=\pi _{^{s1},P_i}(x)`$, have projections which are either always equal or always distinct, as $`x`$ varies over $`G_{i,H}`$, this implies the following: if for some $`x_0G_{i,H}`$, $`(x_0,g_l(x_0))\pi _{^s,^{s1}}^1(x_0)A`$, then $`(x,g_l(x))\pi _{^s,^{s1}}^1(x)A`$ is true $`xG_{i,H}`$. Similarly, if for some $`x_0G_{i,H}`$, $`\{x_0\}\times (g_l(x_0),g_{l+1}(x_0))\pi _{^s,^{s1}}^1(x_0)A`$, then for each $`xG_{i,H}`$ one has $`\{x\}\times (g_l(x),g_{l+1}(x))\pi _{^s,^{s1}}^1(x)A`$. This shows that the partition of $`\pi _{^s,^{s1}}^1(G_{i,H})`$ into โstripeโ and โgraphโ sets, generated by the functions $`g_1,..,g_l`$, is such that the set $`\pi _{^s,^{s1}}^1(G_{i,H})A`$ is a union of elements from this partition.
Thus in the case $`A[1,1]^s`$, we may take $``$ to consist of components of $`G_{i,H}`$, where $`H`$ ranges over the components of $`U_j^{k_1,..,k_j}`$, $`U=\pi _{^{s1},P_i}(G_i)`$, $`i,j,k_1,..,k_j^+`$, and of components of the complement $`^{s1}G`$. By Claim 3.5.3, these are $`\mathrm{}`$-definable $`f`$-sets. This verifies the induction step in the case $`A[1,1]^s`$.
Suppose now that $`A`$ is not a subset of $`[1,1]^s`$. Let $`A_{i_1,..,i_l}`$, $`ln`$, $`1i_1..i_ln`$, denote the subset of points of $`A`$ whose $`i_1,..,i_l`$ coordinates have modulus greater than $`1`$, and the rest of their coordinates have modulus equal or less than $`1`$. These sets form a partition of $`A`$. Fix $`ln`$ and $`i_1,..,i_l`$, $`1i_1..i_ln`$. Let $`\alpha _{i_1,..,i_l}`$ denote the mapping which sends $`x_{i_j}`$ to $`1/x_{i_j}`$ for each $`j=1,..,l`$, and keeps the rest of coordinates unchanged. Note that it maps $`A_{i_1,..,i_l}`$ to its homeomorphic image inside $`[1,1]^s`$. It is not difficult to see that this image is again a $`\mathrm{}`$-definable $`f`$-set. Applying to these homeomorphic image the result which we obtained for $`A[1,1]^s`$, it is possible, via $`\alpha _{i_1,..,i_l}^1`$, to verify that the theorem is true also for the set $`A_{i_1,..,i_l}`$ itself. This verifies the induction step in the case that $`A[1,1]^s`$, and proves the theorem.
It remains to prove Claims 3.5.2 and 3.5.3 on which we relied in the course of the proof of the theorem.
Proof of Claim 3.5.3. If the theorem holds in dimensions smaller than $`s`$, there exists a partition of $`^k`$ into finitely many connected $`\mathrm{}`$-definable $`f`$-sets, such that $`G`$ is a union of elements from a subset of this partition. Each component of $`G`$ must be a union of sets from this partition. Hence, by Lemma 3.2, each component is a $`\mathrm{}`$-definable $`f`$-set. Further, by Lemma 3.2 $`F=F_1..F_N`$ is a $`\mathrm{}`$-definable $`f`$-set, and $`GF=(^kF)G`$. By the induction assumption and Lemma 3.2, $`^kF`$, being a union of $`\mathrm{}`$-definable $`f`$-sets, is itself a $`\mathrm{}`$-definable $`f`$-set. $`\mathrm{}`$
To prove Claim 3.5.2, we first prove an auxiliary statement.
Claim 3.5.4. Suppose that $`A[1,1]^s`$. Let $`G\pi _{^s,^{s1}}(A)`$ be a $`\mathrm{}`$-definable $`f`$-set, and let $`P`$ be a coordinate plane in $`^{s1}`$, $`dim(P)=dim(G)`$. If the theorem holds in dimensions smaller than $`s`$, there exists a partition of $`G`$ into finitely many $`\mathrm{}`$-definable $`f`$-sets $`E,G_1,..,G_N`$, such that $`dim(\pi _{^{s1},P}(E))<dim(P)`$, and each $`G_i`$, $`i=1,..,N`$, projects well on $`P`$.
Proof of Claim 3.5.4. If $`dim(\pi _{^{s1},P}(G))<dim(P)`$, then we just take $`E=G`$. If $`dim(\pi _{^{s1},P}(G))=dim(P)`$, then the induction assumption and Lemma 3.2 imply that there exists $`K>0`$, such that $`\pi _{^{s1},P}^1(y)G`$ consists of at most $`K`$ points $`yPE^{}`$, where $`E^{}`$ is a $`\mathrm{}`$-definable set of dimension smaller than $`dim(P)`$. Order the points in $`^{s1}`$ by, say, the lexicographical order relation. For each $`yPE^{}`$ denote by $`x_i(y)`$ the $`ith`$ largest point of $`\pi _{^{s1},P}^1(y)G`$. Denote by $`F_i`$, $`i=1,..,K`$, the set
$$\{x_i(y):yPE^{}forwhich|\pi _{^{s1},P}^1(y)G|i\}.$$
The sets $`F_i`$ partition $`G\pi _{^{s1},P}^1(PE^{})`$. It can be seen, applying Claim 3.5.3, that the sets $`F_i`$ are $`\mathrm{}`$-definable $`f`$-sets. Fix $`i`$, $`1iK`$. Since $`F_i`$ is a $`\mathrm{}`$-definable $`f`$-set, there exists a $`\mathrm{}`$-definable basic $`f`$-set $`S^p\times (^n)^r`$ and a finite monotonic Whitney stratification $``$ of $`[1,1]^q`$, $`q=p+nr`$, such that $`S`$ is stratified by a subset of $``$, and $`\pi _{^s,^{s1}}^1(F_i)A=\pi _{^q,^s}(S)`$. Let $`HP`$ be the union of projections to $`P`$ of strata of $`S`$ which project to $`P`$ with rank $`dim(P)`$. By Proposition 1.2 the union of strata of $`(dim(P)1,P)`$ is a compact set, and therefore its projection to $`P`$, which we denote by $`Z`$, is compact. By Lemma 3.2 $`Z`$ is a $`\mathrm{}`$-definable $`f`$-set. By the induction assumption and Lemma 3.2, $`HZ`$ is a $`\mathrm{}`$-definable $`f`$-set. Put $`G_i=G\pi _{^{s1},P}^1(HZ)`$. The set $`G_i`$ projects well to $`P`$, and by Lemma 3.2 is a $`\mathrm{}`$-definable $`f`$-set. Note that $`\pi _{^{s1},P}(G)_iG_i`$ has dimension smaller than $`dim(P)`$. Denote the preimage of this set in $`G`$ by $`E`$. The sets $`G_i`$, $`i=1,..,K`$, and $`E`$ form a partition of $`G`$. By Claim 3.5.3 $`E`$ is a $`\mathrm{}`$-definable $`f`$-set as well. $`\mathrm{}`$
Proof of Claim 3.5.2. We suppose that the dimension of $`G`$ is $`d_1n1`$. Put $`E_{11}=G`$. Let us enumerate all coordinate planes of $`^{s1}`$ of dimension $`d_1`$: $`P_1^{d_1},..,P_{k_1}^{d_1}`$. There exists a partition of $`E_{11}`$ into the $`\mathrm{}`$-definable $`f`$-sets $`G_{11}^{d_1},..,G_{1N_1}^{d_1}`$, which project well on $`P_1^{d_1}`$, and a $`\mathrm{}`$-definable $`f`$-set $`E_{12}`$ which projects on $`P_1^{d_1}`$ with dimension smaller than $`d_1`$. Repeating the same step with $`E_{12}`$ and $`P_2^{d_1}`$, we obtain $`G_{21}^{d_1},..,G_{1N_2}^{d_1}`$ and $`E_{13}`$. Repeating this step $`k_1`$ times, we obtain a partition of $`G`$ into the $`\mathrm{}`$-definable $`f`$-sets $`G_{ij}^{d_1}`$ which project well to $`P_i^{d_1}`$, $`1ik_1`$, $`1jN_i`$, and the $`\mathrm{}`$-definable $`f`$-set $`E_{1k_1}`$, which we also denote by $`E_{21}`$. Note that the dimension of $`E_{21}`$, which we denote by $`d_2`$ must be smaller than $`d_1`$. Enumerate now all coordinate planes of $`^{s1}`$ of dimension $`d_2`$, and repeat for dimension $`d_2`$ what we have done earlier for dimension $`d_1`$. We continue in this fashion until for some $`l1`$ we obtain $`E_{l1}=\mathrm{}`$. The $`\mathrm{}`$-definable $`f`$-sets $`G_{ij}^{d_i}`$, which project well on $`P_j^{d_i}`$, $`1jN_{i1}`$, $`i=1,..,l1`$, form a partition of $`G`$. $`\mathrm{}`$
This finishes the proof of Theorem 3.5. $`\mathrm{}`$
We now state our main theorem.
Theorem 3.6. The theory $`T_{\widehat{Df}}`$ is model complete and o-minimal.
Proof. To show that $`T_{\widehat{Df}}`$ is model complete it is sufficient to show that in the structure $`_{\widehat{Df}}`$, the complement of any set which is a projection of a $`\mathrm{}`$-definable quantifier free set, is itself a projection of a $`\mathrm{}`$-definable quantifier free set. By Lemmas 2.7 and 2.8 this is equivalent to being the complement of any $`\mathrm{}`$-definable $`f`$-set again a $`\mathrm{}`$-definable $`f`$-set. The latter is a corollary of Theorem 3.5 and Lemma 3.2.
Further, to show o-minimality, we have to show that any definable (with parameters) set has finitely many components. Let such set be denoted by $`S`$. There is a formula $`\varphi (y_1,..,y_l)`$ and $`y_{i_10},..,y_{i_m0}`$, such that $`S`$ is defined by $`\varphi `$ with $`y_{i_j}=y_{i_j0}`$, $`j=1,..,m`$. Note that $`S`$ can be identified with the intersection of the set defined by $`\varphi (y_1,..,y_l)`$, and the plane given by $`y_{i_j}=y_{i_j0}`$, $`j=1,..,m`$. By model completeness $`\varphi (y)`$ is equivalent to an existential formula $`x\psi (x,y)`$, where $`\psi (x,y)`$ is a quantifier free formula. Thus $`S`$ is the intersection of the set $`M`$ defined by $`x\psi (x,y)`$, with the plane given by $`y_{i_j}=y_{i_j0}`$, $`j=1,..,m`$. Since by Lemma 2.8 $`M`$ is a $`\mathrm{}`$-definable $`f`$-set, Theorem 3.5 applies, and can be seen to imply the finiteness of the number of components of $`S`$. $`\mathrm{}`$
4. A generalization and possible applications.
We comment on how the results extend to the case of at most countably many generic smooth functions. This corresponds to taking instead of the map $`id\times j_r^mf`$ the map $`id\times j_r^mf_1\times ..\times j_r^mf_k`$ and instead of $`id\times ๐^{r,m}f`$ the map $`id\times ๐^{r,m}f_1\times ..\times ๐^{r,m}f_k`$. The transversality arguments go through since the functions (and the divided differences) depend on disjoint sets of arguments. Also, the product of $`C^{\mathrm{}}(^{n_i},)`$, $`i=1,2,..`$ is a Baire space. With these remarks, the proof of Theorem B is almost identical with the proof of Theorem A.
It seems that Theorem B allows to simplify, at least conceptually, some arguments which appear in Matherโs proof of the topological stability of proper generic smooth maps (\[Ma1\], \[Ma2\], \[Ma3\]; we refer to the version given in \[GWPL\]). It seems that one of the main difficulties in this proof is to show that a generic smooth map admits a so called Thom stratification. A related simpler problem is to stratify the range of a generic smooth map. Our result gives the following short proof of the existence of such stratification.
Proposition 4.1. Let $`A[1,1]^s^s`$ be a semialgebraic set, and let $`f:^s^m`$ be a generic smooth map. Then $`f(A)`$ admits a $`C^k`$ Whitney stratification for any $`k1`$.
Proof. We identify $`C^{\mathrm{}}(^n,^m)`$ with $`(C^{\mathrm{}}(^n,))^m`$, and let $`f=(f_1,..,f_m)(C^{\mathrm{}}(^n,))^m`$. The set $`f(A)`$ is defined by
$$x(xAy=f(x)),$$
which, since $`A[1,1]^s`$, can be easily rewritten as a formula of $`_{\widehat{Df_1},..,\widehat{Df_m}}`$. By the results of the theory of o-minimal structures (\[vdDM\], \[L\]), definable sets in o-minimal structures are $`C^k`$ Whitney stratifiable, and thus, since $`f`$ is generic and the conclusion of Theorem B holds, $`f(A)`$ is a $`C^k`$ Whitney stratifiable set. $`\mathrm{}`$
We remark that although the proof is short, it in fact relies on the Isotopy Lemma and on the existence of $`C^k`$ Whitney stratifications of definable sets in o-minimal structures.
A key result about semialgebraic sets is that given two semialgebraic submanifolds $`X,Y`$, the set of points of $`Y`$ at which $`X`$ is not regular over Y, denoted $`B(Y,X)`$, is a semialgebraic sets of dimension smaller than $`dim(Y)`$. One can show that in fact
Proposition 4.2. Let $`X,Y`$ be $`C^k`$ submanifolds, $`k1`$, definable in $`_{\widehat{Df_1},..,\widehat{Df_m}}`$, where $`f_1,..,f_m`$ are generic smooth functions. Then $`B(Y,X)`$ is a definable set as well, of dimension smaller than $`Y`$.
(Note that by Theorem B the dimension is well defined). In \[GWPL\], Chapter I, section 3, there is a proof that a generic polynomial map admits a Thom stratification. Proposition 4.2 allows us to repeat this proof (with relatively minor modifications), and to conclude, assuming the fact that a generic smooth function restricted to its critical set is finite to one, that:
Proposition 4.3. Let $`A[1,1]^s^s`$ be an open semialgebraic set. Then for a generic smooth $`f:^s^p`$, the map $`f|_A:A^p`$ admits a Thom stratification.
The author got initially interested in the problems discussed in this article during an attempt to generalize the results of \[AgGa\] to the generic smooth setting. Related questions were raised before in \[Suss\]. The results presented here are not sufficient to answer most of such questions, since one also needs to consider (suitably restricted) flows of generic vector fields.
Acknowledgements. This text is a revised and corrected version of the draft which the author wrote while staying at SISSA, Trieste. The author thanks A. Agrachev for stimulating discussions, and SISSA and Max Planck Institute for Mathematics for their financial support.
REFERENCES
\[AgGa\]. Agrachev, A.; Gauthier, J.-P. On the subanalyticity of Carnot-Caratheodory distances, Ann. Inst. H. Poincare Anal. Non Lineaire 18 (2001), no. 3, 359โ382.
\[BZ\]. Berezin I.S., Zhidkov N.P., Computing Methods, vol. 1, Pergamon Press, Oxfprd e.a., 1965.
\[Ch\]. Chatzidakis, Z., Introduction to Model Theory, lecture notes, Luminy 2001. Available from http://www.logique.jussieu.fr/www.zoe/.
\[Gab\]. Gabrielov, A. M. Projections of semianalytic sets. (Russian) Funkcional. Anal. i Prilozen. 2 1968 no. 4, 18โ30.
\[GG\]. Golubitsky, M.; Guillemin, V., Stable mappings and their singularities, Graduate Texts in Mathematics, Vol. 14. Springer-Verlag, New York-Heidelberg, 1973.
\[GWPL\]. Gibson, C. G.; Wirthmuller, K.; du Plessis, A. A.; Looijenga, E. J. N., Topological stability of smooth mappings, Lecture Notes in Mathematics, Vol. 552. Springer-Verlag, Berlin-New York, 1976.
\[GrY\]. Grigoriev, A.; Yakovenko, S., Topology of generic multijet preimages and blow-up via Newton interpolation, J. Differential Equations 150 (1998), no. 2, 349โ362.
\[vdD1\]. van den Dries, L., A generalization of the Tarski-Seidenberg theorem, and some nondefinability results, Bull. AMS 15 (1986), 189-193.
\[vdD2\]. van den Dries, L., O-minimal structures, Logic: from foundations to applications (Staffordshire, 1993), 137โ185, Oxford Sci. Publ., Oxford Univ. Press, New York, 1996.
\[vdDM\]. van den Dries, L.; Miller, C., Geometric categories and o-minimal structures, Duke Math. J. 84 (1996), no. 2, 497โ540.
\[IY\]. Ilyashenko, Yu.; Yakovenko, S. Finite cyclicity of elementary polycycles in generic families, Concerning the Hilbert 16th problem, 21โ95, Amer. Math. Soc. Transl. Ser. 2, 165, Amer. Math. Soc., Providence, RI, 1995.
\[L\]. Ta Le Loi, Verdier and strict Thom stratifications in o-minimal structures, Illinois J. Math. 42 (1998), no. 2, 347โ356.
\[KM\]. Karpinski, M.; Macintyre, A., A generalization of Wilkieโs theorem of the complement, and an application to Pfaffian closure, Selecta Math. (N.S.) 5 (1999), no. 4, 507โ516.
\[Ma1\]. Mather, J. N., Stratifications and mappings, Dynamical systems (Proc. Sympos., Univ. Bahia, Salvador, 1971), pp. 195โ232. Academic Press, New York, 1973.
\[Ma2\]. Mather, J. N., How to stratify mappings and jet spaces, Singularites dโapplications differentiables (Sem., Plans-sur-Bex, 1975), pp. 128โ176. Lecture Notes in Math., Vol. 535, Springer, Berlin, 1976.
\[Ma3\]. Mather, J. N., mimeographed notes.
\[RSW\]. Rolin, J.-P.; Speissegger, P.; Wilkie, A. J., Quasianalytic Denjoy-Carleman classes and o-minimality, J. Amer. Math. Soc. 16 (2003), no. 4, 751โ777.
\[Suss\]. Sussmann, H., Some optimal control applications of real analytic stratification and desingularization, Singularities Symposium โ Lojasiewicz 70, B. Jakubczyk, W. Pawlucki, and J. Stasica Eds., Banach Center Publications Vol. 44, Polish Academy of Sciences, Warsaw, Poland, 1998, pp. 211-232.
\[W1\]. Wilkie, A. J., A theorem of the complement and some new o-minimal structures, Selecta Math. (N.S.) 5 (1999), no. 4, 397โ421.
\[W2\]. Wilkie, A. J., Model completeness results for expansions of the ordered field of real numbers by restricted Pfaffian functions and the exponential function. J. Amer. Math. Soc. 9 (1996), no. 4, 1051โ1094.
Present address for correspondence:
alexg@mpim-bonn.mpg.de
|
warning/0506/astro-ph0506201.html
|
ar5iv
|
text
|
# RASS-SDSS Galaxy Cluster Survey.
## 1 Introduction
The galaxy Luminosity Function (LF) is a fundamental tool for understanding galaxy evolution and faint galaxy populations. The shape of the cluster LF provides information on the initial formation and subsequent evolution of galaxies in clusters while the slope of the faint-end indicates how steeply the dwarf number counts rise as a function of magnitude.
Much work has been done on the cluster LF, with various groups finding differences in its shape and the faint-end slope. Different techniques have been used to measure LFs of individual clusters or to make a composite LF from individual clusters LFs (e.g. Dressler 1978; Lugger 1986, 1989; Colless 1989; Biviano et al. 1995; Lumsden et al. 1997; Valotto et al. 1997; Rauzy et al. 1998; Garilli et al. 1999; Paolillo et al. 2001; Goto et al. 2002; Yagi et al. 2002; Popesso et al. 2004a). Whether the LF of cluster galaxies is universal or not, and whether it is different from the LF of field galaxies are still debated issues. Several authors (Dressler 1978; Lumdsen et al. 1997; Valotto et al. 1997; Garilli et al. 1999; Goto et al. 2002; Christlein & Zabludoff 2003) have found significant differences between the LFs of different clusters as well as between the LFs of cluster and field galaxies, while others (Lugger 1986, 1989; Colless 1989; Rauzy et al. 1998; Trentham 1998; Paolillo et al. 2001; Andreon 2004) have concluded that the galaxy LF is universal in all environments. Another debated issue is the slope of the faint end of the LF of cluster galaxies (see, e.g., Driver et al. 1994; Lobo et al. 1997; Smith et al. 1997; Phillipps et al. 1998; Boyce et al. 2001; Beijersbergen et al. 2001; Trentham et al. 2001; Sabatini et al. 2003; Cortese et al. 2003). The LF of cluster galaxies is typically observed to steepen faint-ward of $`M_g18`$, with power-law slopes $`\alpha 1.8\pm 0.4`$. This corresponds to the debated upturn of the cluster LF due to an excess of dwarf galaxies relative to the field LF. The effect may be real, and due to cluster environmental effects, but it could also be generated by systematics in the detection techniques of faint, low surface-brightness galaxies.
In Popesso et al. (2004a, hereafter paper II) we have recently analyzed the LF of clusters from the RASS-SDSS (ROSAT All Sky Survey โ Sloan Digital Sky Survey) galaxy clusters survey down to $`14`$ mag. We concluded that the composite cluster LF is characterized by an upturn and a clear steepening at faint magnitudes, in all SDSS photometric bands. Different methods of background subtraction were shown to lead to the same LF. The observed upturn of the LF at faint magnitudes was shown in particular not to be due to background contamination by large scale structures or multiple clusters along the same line of sight. We concluded that the observed steepening of the cluster LF is due to the presence of a real population of faint cluster galaxies.
The composite LF was well fitted by the sum of two Schechter (1976) functions. The LF at its bright-end was shown to be characterized by the classical slope of $`1.25`$ in all photometric bands, and a decreasing $`M^{}`$ from the $`z`$ to the $`g`$ band. The LF at its faint-end was found to be much steeper than the LF at its bright-end, and characterized by a power-law slope $`2.5\alpha 1.6`$. The observed upturn of the LF was found to occur at $`16`$ in the $`g`$ band, and at $`18.5`$ in the $`z`$ band.
A steep mass function of galactic halos is a robust prediction of currently popular hierarchical clustering theories for the formation and evolution of cosmic structure (e.g. Kauffmann et al. 1993; Cole et al. 1994). This prediction conflicts with the flat galaxy LF measured in the field and in local groups, but is in agreement with the steep LF measured in the RASS-SDSS clusters. Two models have been proposed to explain the observed environmental dependence of the LF. According to Menci et al. (2002), merging processes are responsible for the flattening of the LF; the environmental dependence arises because mergers are more common in the field (or group) environment than in clusters, where they are inhibited by the high velocity dispersion of galaxies. According to Tully et al. (2002), instead, the LF flattening is due to inhibited star formation in dark matter halos that form late, i.e. after photoionization of the intergalactic medium has taken place. Since dark matter halos form earlier in higher density environments, a dependence of the observed LF slope on the environment is predicted. On the other hand, if reionization happens very early in the Universe, this scenario may not work (Davies et al. 2005). Other physical processes are however at work in the cluster environment, such as ram-pressure stripping (Gunn & Gott 1972) and galaxy harassment (e.g. Moore et al. 1996, 1998), which are able to fade cluster galaxies, particularly the less massive ones. Whether the outcome of these processes should be a steepening or a flattening of the LF faint-end is still unclear.
In paper II it was also shown that the bright-end of the LF is independent from the cluster environment, and the same in all clusters. On the other hand, the LF faint-end was found to vary from cluster to cluster. In the present paper (IV in the series of the RASS-SDSS galaxy cluster survey) we show that the previously found variations of the faint end of the cluster LF are due to aperture effects. In other words, when measured within the physical size of the system, given by either $`r_{200}`$ or $`r_{500}`$, the LF is invariant for all clusters, both at the bright and at the faint end. We also analyze how the number ratio of dwarf to giant galaxies in galaxy clusters depends on global cluster properties such as the velocity dispersion, the mass, and the X-ray and optical luminosities. Finally, we investigate the nature of the dwarf galaxies in clusters by studying their color distribution and suggest a possible formation scenario for this population.
The paper is organized as follows. In ยง 2 of the paper we describe our dataset. In ยง 3 we summarize the methods used to calculate the individual and the composite cluster LFs. In ยง 4 we summarize our methods for measuring the clusters characteristic radii. In ยง 5 we analyze the resulting composite and individual LFs. In ยง 6 we determine the cluster composite LF per galaxy type. In ยง 7 we analyse the environmental dependence of the LF, and compare the cluster and field LFs. In ยง 8 we provide our discussion, suggesting a possible formation scenario for the faint galaxy population in clusters. Finally, in ยง 9 we draw our conclusions.
For consistency with paper II and with previous works, we use $`\mathrm{H}_0=100\mathrm{h}\mathrm{km}\mathrm{s}^1\mathrm{Mpc}^1`$, $`\mathrm{\Omega }_m=0.3`$ and $`\mathrm{\Omega }_\mathrm{\Lambda }=0.7`$ throughout the paper.
## 2 The data
In order to study the variation of the cluster LF from system to system, the analysis has to be applied to a large statistical sample of clusters, covering the whole spectrum of properties (in mass, richness, X-ray luminosity and optical luminosity) of the systems considered. Since the X-ray observations provide a robust method of identification of galaxy clusters and the X-ray luminosity is a good estimator of the system total mass and optical luminosity (see paper I and Popesso et al. 2004c, hereafter paper III), we have used for our purpose the RASS-SDSS galaxy cluster sample, which is an X-ray selected sample of objects in a wide range of X-ray luminosity. The updated version of the RASS-SDSS galaxy cluster catalog comprises 130 systems detected in the RASS and in the SDSS sky region (16 clusters more than in the first version of the catalog released in paper I due to the larger sky area available in the SDSS DR2). The X-ray cluster properties and the cluster redshifts have been taken from a variety of X-ray catalogs, that allow to cover the whole $`L_X`$ spectrum. The X-ray intermediate and bright clusters have been selected from three ROSAT based cluster samples: the ROSAT-ESO flux limited X-ray cluster sample (REFLEX, Bรถhringer et al. 2002), the Northern ROSAT All-sky cluster sample (NORAS, Bรถhringer et al. 2000), the NORAS 2 cluster sample (Retzlaff 2001). The X-ray faint clusters and the groups have been selected from two catalogs of X-ray detected objects: the ASCA Cluster Catalog (ACC) from Horner (2001) and the Group Sample (GS) of Mulchaey et al. (2003). The RASS-SDSS galaxy cluster sample comprises only nearby systems at the mean redshift of 0.1. The sample covers the entire range of masses and X-ray luminosities, from very low-mass and X-ray faint groups ($`10^{13}M`$ and $`10^{42}ergs^1`$) to very massive and X-ray bright clusters ($`5\times 10^{15}M`$ and $`5\times 10^{44}ergs^1`$).
The optical photometric data are taken from the 2<sup>nd</sup> data release of the SDSS (Fukugita et al. 1996, Gunn et al. 1998, Lupton et al. 1999, York et al. 2000, Hogg et al. 2001, Eisenstein et al. 2001, Smith et al. 2002, Strauss et al. 2002, Stoughton et al. 2002, Blanton et al. 2003 and Abazajian et al. 2003). The SDSS consists of an imaging survey of $`\pi `$ steradians of the northern sky in the five passbands $`u,g,r,i,z,`$. The imaging data are processed with a photometric pipeline (PHOTO) specially written for the SDSS data. For each cluster we defined a photometric galaxy catalog as described in ยง 3 of Popesso et al. (2004b, paper I). For the analysis in this paper we only use SDSS Model magnitudes (see paper II for details).
In this paper we consider a subsample of 69 clusters of the RASS-SDSS sample for which the masses, velocity dispersion, $`r_{200}`$ and $`r_{500}`$ (see ยง 4) were derived through the virial analysis (see paper III) applied to the spectroscopic galaxy members of each systems.
Since throughout the paper the results obtained with the current analysis of the cluster LF are often compared with the results obtained in paper II, it is important to notice that the cluster sample used here is a subsample of the dataset used in paper II.
## 3 Determination of the individual and composite Luminosity Functions
We here summarize the methods by which we measure the individual and composite cluster LFs. Full details can be found in papers I and II.
We consider two different approaches to the statistical subtraction of the galaxy background. As a first approach, we calculate a local background in an annulus centered on the X-ray cluster center with an inner radius of 3 $`\mathrm{h}^1`$ Mpc and a width of 0.5 deg.
As a second approach we derive a global background correction. We define as $`N_{bg}^g(m)dm`$ the mean of the galaxy number counts determined in five different SDSS sky regions, randomly chosen, each with an area of 30 $`\mathrm{deg}^2`$. A detailed comparison of the local and global background estimates can be found in paper I. The results shown in this paper are obtained using a global background subtraction.
We derive the LFs of each cluster by subtracting from the galaxy counts measured in the cluster region, the field counts rescaled to the cluster area. Following previous literature suggestions, we exclude the brightest cluster galaxies from the clusters LFs.
In order to convert from apparent to absolute magnitudes we use the cluster luminosity distance, correct the magnitudes for the Galactic extinction (obtained from the maps of Schlegel et al. 1998), and apply the K-correction of Fukugita et al. (1995) for elliptical galaxies, which are likely to constitute the main cluster galaxy population.
The composite LF is obtained following Colless (1989) prescriptions. A detailed description of the method can be found in paper II.
### 3.1 Low surface brightness selection effect
It is well known that magnitude-limited surveys may be biased against low-surface brightness galaxies (e.g. Phillips & Driver 1995). An assessment of this bias for the SDSS-EDR and SDSS-DR1 has been done by Cross et al. (2004), who compared these catalogs with the Millennium Galaxy Catalog (Liske, Lemon, Driver et al. 2003), a deep survey limited in surface brightness to 26 mag arcsec<sup>-2</sup>. Cross et al. (2004) concluded that the incompleteness of SDSS-EDR is less than 5% in the range of effective surface-brightness $`21\mu _e25`$ mag arcsec<sup>-2</sup>, and it is around 10% in the range $`25\mu _e26`$ mag arcsec<sup>-2</sup>. In this paper, galaxies contributing to the faint-end of the cluster LFs have magnitudes $`18r21`$. In this magnitude range, 65% of the objects have $`\mu _e23`$ mag arcsec<sup>-2</sup>, 30% have $`23<\mu _e24`$ mag arcsec<sup>-2</sup>, and 5% have $`mu_e25`$ mag arcsec<sup>-2</sup>. Hence, from the results of Cross et al. (2004), we do not expect that the bias against low surface-brightness galaxies results in an incompleteness above $`5`$%. The faint-end of the cluster LFs derived in this paper should thus be quite unaffected by this selection effect.
## 4 The characteristic radii of galaxy clusters
We here describe the methods by which we measure the characteristic radii $`r_{500}`$ and $`r_{200}`$. $`r_{200}`$ is the radius where the mass density of the system is 200 times the critical density of the Universe and it is considered as a robust measure of the virial radius of the cluster. Similarly, $`r_{500}`$ is defined setting 500 instead of 200 in the previous definition and it samples the central region of the cluster. Full details can be found in paper III.
We estimate a cluster characteristic radius through the virial analysis applied on the redshifts of its member galaxies. We use the redshifts provided in the SDSS spectroscopic catalog to define the galaxy membership of each considered system. The SDSS spectroscopic sample comprises all the objects observed in the Sloan r band with pretrosian magnitude $`r_P17.77`$ mag and half-light surface brightness $`\mu _{50}24.5`$ mag$`\mathrm{arcsec}^2`$. The SDSS DR2 spectrocsopic sample used for this analysis counts more tha 250.000 galaxies.
Cluster members are selected following the method of Girardi et al. (1993). First, among the galaxies contained in a circle of radius equal to the Abell radius, those with redshift $`czcz_{cluster}>4000`$ km s<sup>-1</sup> are removed, where $`z_{cluster}`$ is the mean cluster redshift. Then, the gapper procedure (see also Beers et al. 1990) is used to define the cluster limits in velocity space. Galaxies outside these limits are removed. Finally, on the remaining galaxies we apply the interloper-removal method of Katgert et al. (2004; see Appendix A in that paper for more details).
The virial analysis (see, e.g., Girardi et al. 1998) is then performed on the clusters with at least 10 member galaxies. The velocity dispersion is computed using the biweight estimator (Beers et al. 1990). The virial masses are corrected for the surface-pressure term (The & White 1986), using a Navarro et al. (1996, 1997) mass density profile, with concentration parameter $`c=4`$. This profile provides a good fit to the observationally determined average mass profile of rich clusters (see Katgert et al. 2004).
Our clusters span a wide range in mass; since clusters of different masses have different concentrations (see, e.g. Dolag et al. 2004) we should in principle compute the cluster masses, $`M`$โs, using a different concentration parameter $`c`$ for each cluster. According to Dolag et al. (2004), $`cM^{0.102}`$. Taking $`c=4`$ for clusters as massive as those analysed by Katgert et al. (2004), $`M2\times 10^{15}M_{}`$, Dolag et al.โs scaling implies our clusters span a range $`c3`$โ6. Using $`c=6`$ instead of $`c=4`$ makes the mass estimates 4% and 10% higher at, respectively, $`r_{200}`$ and $`r_{500}`$, while using $`c=3`$ makes the mass estimates lower by the same factors. This effect being clearly much smaller than the observational uncertainties, we assume the same $`c=4`$ in the analysis for all clusters.
If $`M_{vir}`$ is the virial mass (corrected for the surface-pressure term) contained in a volume of radius equal to the clustercentric distance of the most distant cluster member in the sample, i.e. the aperture radius $`r_{ap}`$, then, the radius $`r_{200}`$ is then given by:
$$r_{200}r_{ap}[\rho _{vir}/(200\rho _c)]^{1/2.4}$$
(1)
where $`\rho _{vir}3M_{vir}/(4\pi r_{ap}^3)`$ and $`\rho _c(z)`$ is the critical density at redshift $`z`$ in the adopted cosmology. The exponent in eq.(1) is the one that describes the average cluster mass density profile near $`r_{200}`$, as estimated by Katgert et al. (2004) for an ensemble of 59 rich clusters. Similarly, $`r_{500}`$ is estimated by setting 500 instead of 200 in eq.(1).
## 5 Analysis of the individual and composite LFs
In order to analyze the behavior of the composite LF faint-end as a function of waveband and clustercentric distance, we define the number ratio of dwarf to giant galaxies, DGR, as the ratio between the number of faint ($`18M16.5`$) and bright ($`M<20`$) galaxies in the cluster LF. The DGR is found to vary from cluster to cluster, more than expected from statistical errors. These variations are not random however. As shown in Fig. 1, when the DGRs are computed within a fixed metric radius, they are significantly anti-correlated with several cluster global properties, i.e. the cluster velocity dispersions, masses, and X-ray and optical luminosities (velocity dispersions, virial masses, and X-ray luminosities for our cluster sample were derived in paper III). All the correlations are very significant (1โ$`5\times 10^5`$, according to a Spearman correlation test). The more massive a cluster, the lower its fraction of dwarf galaxies.
The correlation between cluster DGRs and cluster masses is most likely due to the choice of a fixed metric aperture for all the clusters. In fact, a fixed metric aperture samples larger (smaller) fractions of the virialized regions of clusters of smaller (respectively, larger) masses, and DGR is known to increase with clustercentric distance (paper II).
Because of this effect, the different cluster physical sizes must be taken into account before comparing different cluster LFs. We then determine the individual and composite LFs within $`r_{500}`$ and $`r_{200}`$ for the subsample of 69 clusters of the RASS-SDSS galaxy cluster sample for which these parameters are known (see paper III).
The composite LF calculated within $`r_{200}`$ is shown in Fig. 2 for four SDSS photometric bands. The $`u`$-band LF is not shown; in this band, there is no evidence for an upturn at faint magnitude levels (see paper II). For all the other bands LFs, a single Schechter function does not provide acceptable fits, and a composite of two Schechter functions is needed:
$$\varphi (L)=\varphi ^{}[(\frac{L}{L_b^{}})^{\alpha _b}exp(\frac{L}{L_b^{}})+(\frac{L}{L_f^{}})^{\alpha _f}exp(\frac{L}{L_f^{}})]$$
(2)
where $`b`$ and $`f`$ label the Schechter parameters of the bright and faint end respectively. From the reduced-$`\chi ^2`$ values given in Table 1 we conclude that a double-Schechter function does provide adequate fits to the 4-bands composite LFs. Alternatively, we fit the composite LFs with a function of this form:
$$\varphi (L)=\varphi ^{}(\frac{L}{L^{}})^\alpha exp(\frac{L}{L^{}})[1+(\frac{L}{L_t})^\beta ].$$
(3)
In this function, $`\varphi ^{}`$ , $`L^{}`$ and $`\alpha `$ are the standard Schechter parameters, $`L_t`$ is a transition luminosity between the two power laws and $`\beta `$ is the power law slope of the very faint end (Loveday 1997). Both functions require the same number of fit parameters. However, the double Schechter component function provides slightly better fits than the Schechter$`+`$power-law function in all the Sloan bands (see Table 1).
The Double Schechter function has been used for the first time by Driver et al. (1994), while Thompson & Gregory (1993) and Biviano et al. (1995) suggested a Gaussian$`+`$Schechter function, to fit respectively the bright and the faint end of the LF. More recently, Hilker et al. (2003) used a double Schechter Function to fit the LF of the Fornax cluster.
The confidence-level contours of the best-fit parameters of the bright- and faint-end Schechter components are shown in Figs. 3 and 4, respectively. Both results for the composite LF within $`r_{500}`$ (dotted contours) and $`r_{200}`$ (solid contours) are shown. Clearly, the best-fit Schechter function to the LF bright-end does not change significantly from $`r_{500}`$ to $`r_{200}`$ (see Fig. 3) confirming the findings of paper II. However, the faint-end LF steepens significantly (by 0.1โ0.15 dex) from $`r_{500}`$ to $`r_{200}`$, and the characteristic magnitude correspondingly brightens by 0.3โ0.4 magnitudes (see Fig. 4), thereby indicating an increasing DGR with radius. Our result is in agreement with the findings of paper II, and several other works in the literature, which were however mostly based on single cluster studies (e.g. Lobo et al. 1997; Durret et al. 2002; Mercurio et al. 2003; Pracy et al. 2004; see however Trentham et al. 2001, for a discordant result).
While our conclusions on the composite LF agree with those of paper II, we find here different results concerning the individual cluster LFs. While in paper II we claimed significant LF variations from cluster to cluster, we discover that such variations disappear when the individual cluster LFs are computed within the physical sizes of each cluster (defined by $`r_{500}`$ or $`r_{200}`$). This can be seen in Fig. 5a, where we plot the individual LFs of 15 clusters (those with the faintest absolute magnitude limits) and, superposed, the composite LF, all measured within $`r_{200}`$ and in the $`r`$-band. The agreement between the composite and individual LFs is very good. Fitting the composite LF to the individual cluster LFs result in the reduced-$`\chi ^2`$ distribution shown in Fig. 5b. For 90% of the clusters the probability that the composite and individual LFs are drawn from the same parent distribution is larger than 95%.
In Fig. 5c we also show the $`z`$-band DGR-distribution. When compared to the DGR distribution found in paper II, the new DGR distribution is much narrower. In this paper we considered the DGR within $`r_{200}`$ of 29 clusters, those with known mass, $`r_{200}`$ and $`r_{500}`$, out of the 35 systems considered in paper II. The mean value of the DGR is 3.5 and its dispersion is indeed very close to the mean DGR statistical error of 1.4, as expected if the individual cluster LFs are indeed all rather similar when computed within a cluster-related physical radius.
Finally, in Fig. 6 we show DGR within $`r_{200}`$ as a function of the cluster mass $`M_{200}`$ (panel $`a`$) and the velocity dispersion (panel $`b`$). There is no hint of the relation previously found (compare with Fig. 1a): the Spearman correlation coefficient is $`0.08`$, which is not statistically significant. Similar results are found also for the $`DGRL_X`$ and $`DGRL_{op}`$ relations.
Hence we conclude that the cluster to cluster LF variation seen in paper II are entirely due to the use of a fixed metric aperture for all clusters, rather than an aperture sampling the same fraction of the virialized region of each cluster.
## 6 The cluster LF per galaxy type
In order to better understand the nature of the cluster galaxies responsible for the LF upturn at low luminosities, we examine their color distribution. In particular, we use the $`ur`$ color, since the $`ur`$ distribution of Sa and earlier-type galaxies is well separated from the $`ur`$ distribution of Sb and later-type galaxies (Strateva et al. 2001), thereby allowing to distinguish the two morphological samples down to very faint magnitudes.
To define the color distribution of the cluster galaxies we statistically subtract the contribution of field galaxies (Boyce et al. 2001), using the same method applied for the statistical subtraction of the background from the magnitude number counts. We determine the background color distribution of field galaxies in an annulus around the cluster with inner radius larger than $`r_{200}`$; significantly under- or over-dense regions (e.g. voids and background clusters) are excluded. By subtracting the background color distribution from the color distribution of galaxies in the cluster region, we obtain the $`ur`$ distribution of cluster galaxies. The validity of the method is confirmed by its application to the spectroscopic subsample, for which cluster membership can be established from the galaxy redshifts.
Fig. 7 shows the (background-subtracted) $`ur`$ distribution of cluster galaxies in the range $`18r16.5`$ (panel $`a`$) and $`16.5r15`$ (panel $`b`$) for the subsample of 15 clusters with the faintest absolute magnitude limit in the $`r`$ band ($`M_{r,lim}15`$). The error bars shown in the figure take into account the galaxy counts Poisson statistics as well as the error due to the background subtraction.
At the redshifts of the 15 clusters considered ($`0.02z0.05`$) early-type galaxies have $`ur`$ colors in the range 2.6โ2.9 (Fukugita et al. 1995), and galaxies redder than $`ur=3`$ are probably in the background. Hence, we can see from Fig. 7a that the residual background contamination after the statistical background subtraction, is generally small ($`10`$ %) and in fact not significant in the bright magnitude range. The contamination is higher for the two clusters RO313 and RX 288, and probably due to the presence of another cluster along the same line-of-sight. In the fainter magnitude range, the average background contamination increases to 25โ35%, but is still not significant (see Fig. 7b).
If we exclude galaxies with $`ur3`$ from our cluster samples, and recalculate the cluster LFs as before (see ยง 3), the modifications are marginal (compare filled points and empty squares in Fig. 8). If anything, a better agreement is now found between the composite LF and the individual LF of the cluster R0313, for which the background contamination is more severe, clearly suggesting that the $`ur3`$ color cut helps in cleaning the cluster sample from background contamination.
We therefore adopt the $`ur<3`$ color cut to select cluster members, and, following Strateva et al. (2001) we distinguish between cluster early- and late-type galaxies using a color-cut $`ur=2.22`$. We restrict our analysis to the very nearby clusters ($`z0.1`$) to minimize the effects of an uncertain K-correction on the derived colors. The composite LFs of the early- and late-type galaxies (defined on the basis of their $`ur`$ colors) are shown in Fig. 9 for four SDSS photometric bands. The late-type galaxy LF is well fitted by a single Schechter function and does not show any evidence of an upturn at the faint end. On the other hand, the early-type LF looks quite different from the late-type LF. It shows a marked bimodal behavior with a pronounced upturn in the faint magnitude region. The best fit parameters are listed in Table 2. Such an upturn is then reflected in the complete (early$`+`$late) LF, with the late-type dwarf galaxies contributing to make the faint-end of the complete LF even steeper. This result is in agreement with Yagi et al. (2002). They determine the total LF of 10 clusters within 1 h<sup>-1</sup> Mpc radius circle. They find that the early-type LF cannot be fitted by a single Schechter function in the magnitude range from $`23`$ to $`15`$, because it flattens at $`M_R=18`$ and then rises again.
## 7 The environmental dependence of the LFs
In order to gain insight into the processes responsible for the shaping of the LF in clusters, we here examine the dependence of the LF on the environmental conditions. In particular we analyze how the LF shape, and the relative fraction of red and blue dwarf galaxies, vary as a function of the clustercentric distance. Fig. 10 shows the behavior of the cluster LF calculated within different clustercentric apertures, separately for the early-type (panel $`a`$) and late-type (panel $`b`$) galaxy populations. Distances are in units of $`r_{200}`$. For simplicity we only plot the best fitting functions and not the data-points. The early-type LF is close to a Schechter function at the center of the cluster (within $`0.2r_{200}`$) and shows a marked upturn afterwards. The location of the upturn varies from $`16.2\pm 0.3`$ mag at distances $`0.3r_{200}`$ to $`17.4\pm 0.4`$ at distances $`r_{200}`$. The late-type LF is well fitted by a single Schechter function at any clustercentric distance. We do not observe blue galaxies within $`0.1r_{200}`$. Moreover, the central late-type LF at $`0.2r_{200}`$ is flatter than the LFs in the outer regions and shows a fainter $`M^{}`$. Since red galaxies are mostly high surface-brightness objects (Blanton et al. 2004), the surface brightness selection effect should be more important for the late-type LF, which, once corrected, would become steeper at the faint-end. If anything, the difference in slope between the faint-ends of the early- and late-type LFs should thus be even larger than observed.
These results are confirmed by the analysis of the early-type LFs in independent clustercentric rings. We consider the region at distances $`r0.3r_{200}`$ (the central ring), $`0.3r/r_{200}0.7`$ (the intermediate ring) and $`0.7r/r_{200}1`$ (the outer ring). The best fitting functions of the cluster early-type LFs within these regions are shown in Fig. 11. In order to emphasize the shape variation of the LF, all three LFs are renormalized to the same value. The upturn at the faint end is brighter in the outer ring than in the central one, confirming the previous analysis. Moreover, the shape of the bright end of the cluster LF seems to be absolutely independent from the faint end. The values of $`M^{}`$ and the slope of the bright end are consistent within the errors in the three regions (as found in paper II). This suggests that the process of formation of the bright cluster galaxies (with magnitude brighter that $`M^{}2`$ mag) is the same in all the cluster environments. Therefore, it seems unlikely that the lack of dwarf systems observed at the center of the cluster is due to a hierarchical process of formation of the bright central galaxies. Indeed, in that case we should observe also a lack of bright galaxies in the outer ring in favor of large amount of dwarf systems, which is not observed.
The analysis so far provides only results about the LF shape. In order to quantify the relative contribution of the early- and late-type dwarf galaxy populations to the faint end of the LF, and its dependence on the environment, we analyse the radial (cumulative and differential) profile of the dwarf systems in the clusters. For this, we consider the galaxies with $`18M_r15`$, and to improve the statistics, we stack the clusters with $`M_{r,lim}15`$ mag, by rescaling the clustercentric distances in units of $`r_{200}`$. The cumulative profiles of the fractions of dwarf galaxies of both the early- and the late-type are shown in Fig. 12$`a`$. The center ($`0.4r_{200}`$) contains less than 30% of dwarf galaxies (half of them are red systems), in the selected magnitude range. Dwarf galaxies are more abundant in the cluster outskirts; the high-density environment in the cluster cores is hostile to dwarf galaxies. This phenomenology has already been observed in several individual clusters (see e.g. Lobo et al. 1997; Boyce et al. 2001; Mercurio et al. 2003; Dahlen et al. 2004);
The early-type dwarf galaxies represents 35% of the whole dwarf population within $`r_{200}`$, i.e. most of the dwarf galaxies are of late-type. However, the dwarf early-type galaxies are the dominant dwarf population region within $`0.4r_{200}`$, their relative fraction reaching a plateau at $`0.6r_{200}`$, while the late-type dwarf galaxies are more abundant in clusters outskirts. This is confirmed also by the ratio between early- and late-type dwarf galaxies calculated in contiguous clustercentric rings (differential profile, see Fig. 12$`b`$). The number of early-type dwarf galaxies is twice the number of late-type dwarf galaxies within $`0.2r_{200}`$ and then decreases to 1/2 at larger distances.
The relation between dwarf morphology and clustercentric distance translates into a morphology-density relation. In Fig. 12$`d`$ we show the ratio between early- and late-type dwarf galaxies as a function of the number density of galaxies brighter than $`M_r18`$ (the bright galaxies number density profile is shown in panel $`c`$ of the same figure). As expected, the early-type dwarf galaxies dominate in high density regions, while the late-type dwarf galaxies are frequent in low density regions. Clearly, the well known morphology density relation for cluster galaxies (Dressler 1980) has an extension into the dwarf regime.
### 7.1 Comparison with the field
In order to extend the morphology-density relation for dwarf cluster galaxies outside clusters, we extract a subsample of galaxies from the SDSS spectroscopic sample. We select a fairly complete sample of galaxies in the redshift range $`z0.02`$ and in the magnitude range $`18M_r16`$. The late-type galaxies ($`ur2.22`$) represent the 93% of the galactic population in that range of magnitude, in agreement with the results of Blanton et al. (2004). We then calculate for each galaxy in the sample the local density of galaxy neighbors, by counting the number of systems with $`M_r18`$ mag, within 2.5 Mpc projected radius and $`\pm 500`$ km/s of the galaxy position and redshift. We divide the subsample in late and early-type galaxies using the color cut of Strateva et al. (2001). Fig. 13 shows the number of galaxies per bin of local density for the two galaxy types. It is clear that late-type galaxies (dashed dotted histogram) populate the very low density regions, while the early-type galaxies distribution (solid histogram) has a much larger spread, with 50% of the systems located in regions with more than 10 galaxy neighbors.
It is also interesting to compare our composite cluster LFs with the LF of field galaxies. Blanton et al. (2004) have recently derived the LF of field SDSS galaxies down to $`12`$ mag. Their LF have a very weak upturn, much shallower and at a fainter carachteristic magnitude than in our cluster LF. The faint-end slope of their LF is $`1.3`$, but could be steeper ($`1.5`$) if a correction is applied to account for low surface-brightness selection effects. The LF of blue field galaxies is even steeper, but the authors do not report the value of the faint-end slope. A similar faint-end slope ($`1.5`$) has also been found by Madgwick et al. (2002) for the LF of field galaxies from the 2dF survey. They also noticed an upturn in the LF, due to an overabundance of early-type galaxies, making it impossible to fit the LF adequately with a single Schechter function. A previous determination of the SDSS field LF was obtained by Nakamura et al. (2003). They found a slope of $`1.9`$ for dIrr, consistent with the value found by Marzke et al. (1994) for the CfA survey.
The faint-end slope of our late-type cluster galaxies LF is steeper than most field LFs for the same galaxy type (see Table 3 in Paper II) but consistent with those of Nakamura et al. (2003) and Marzke et al. (1994). Given the large variance of results for the field LFs, possibly due to the different magnitude limits adopted, or to poor statistics in the fainter bins of the LF (see de Lapparent 2003 for a thorough discussion on this topic), we conclude there is no significant difference between the late-type LF in clusters and the field.
## 8 Discussion
There are many observations and theoretical models in the literature that try to explain the formation and evolution of cluster galaxies, red dwarf galaxies in particular. According to the hierarchical picture for structure formation, small dark matter haloes form before large ones. If one identifies the dwarf galaxies with the small dark matter haloes, they are predicted to origin soon after the structure formation began. Dwarf ellipticals would then be old, passively evolved galaxies. This scenario seems to be inconsistent with the observations of a large spread in age and metallicity in the clusters dwarf early-type galaxies (Conselice et al. 2001,2003; Rakos et al. 2001). Hence, dwarf ellipticals must have had a delayed star formation epoch. The delay could be originated by the intense ultraviolet background intensity at high redshift, keeping the gas of the dwarf galaxies photoionized until $`z1`$, or, perhaps by the intra-cluster medium confinement. The intra-cluster medium pressure could avoid dwarf galaxies losing their gas content by SN ejecta. However, this possibility would require a much more centrally concentrated distribution of dwarf ellipticals in clusters than is observed.
In alternative, the excess of dwarf early-type galaxies in clusters could origin from the evolution of field dIrr when they are accreted by the clusters. The evolution of dIrr into dwarf early-type galaxies is supported by the result of van Zee et al. (2004), namely that there is significant similarity in the scaling relations and properties of dIrr and dEs. A scenario where all dwarf early-type galaxies evolve from dIrr via disk fading does not however seem possible, because many dEs in the Virgo and Fornax clusters are brighter than the dIrr (Conselice et al. 2001).
Perhaps, some dwarf early-type galaxies evolve from dIrr and some evolve from spirals. The evolution of spirals into dwarf spheroidals can occur via the process of โgalaxy harassmentโ proposed by Moore et al. (1996,1998). In this scenario, close, rapid encounters between galaxies can lead to a radical transformation of a galaxy morphology. Gas and stars are progressively stripped out of the disk systems, eventually leaving a spheroidal remnant, that resembles an S0 galaxy or a dwarf spheroidal, depending on the size of the progenitor. Direct support for the harassment scenario comes from the discoveries of disks or even spiral arms in dwarf early-type cluster galaxies (Jerjen et al. 2000; Barazza et al. 2002; Graham et al. 2003). Indirect support comes from the similar velocity distribution of dwarf cluster galaxies (Drinkwater et al. 2001) and gas-rich spirals and irregulars (Biviano et al. 1997), both suggesting infalling orbits.
Is the harassment scenario still viable in view of our results? We can draw the following conclusions from our observational results. First, the universality of the cluster LF suggests that whatever shapes the cluster LF is not strictly dependent on the cluster properties. Second, the difference between the cluster and field LF seems to be related to an excess of dwarf early-type galaxies in clusters. Hence, there is a cluster-related process that leads to the formation of dwarf early-type galaxies, regardless of the cluster intrinsic properties. The process cannot be related, e.g., to the intra-cluster gas density, or the cluster velocity dispersion, or the cluster mass, hence, a process like ram-pressure would seem to be ruled out.
The density dependence of the relative number of early- and late-type dwarfs suggests that the shaping of the cluster LF is related to the excess mean density relative to the field, which is the same for all clusters if, as we have done, the cluster regions are defined within a fixed overdensity radius ($`r_{200}`$ in our case). In other words, the transformation of spirals, and perhaps, dIrr, into dwarf spheroidals or dEs, seems to be a threshold process that occurs when the local density exceeds a given threshold. Judging from Fig. 12, this threshold seems to occur at a clustercentric distance of $`0.6`$โ0.7 $`r_{200}`$.
We have also found that the relative number of dwarf early- and late-type galaxies increases with decreasing clustercentric distance (and increasing density). Galaxies near the cluster center are probably an older cluster population, accreted when the cluster was smaller, according to the hierarchical picture of cluster formation and evolution. Hence, these centrally located galaxies have had more time to accomplish the morphology transformation than galaxies located in the cluster outskirts, which are more recent arrivals.
On the other hand, very near the cluster center, an additional process must be at work to explain our observed fading of the upturn of the cluster early-type LF, and the decrease of both the early- and the late-type dwarf-to-giant galaxy ratio with decreasing clustercentric distance. High-velocity dispersions in clusters inhibit merging processes (e.g. Mihos 2004), hence it is unlikely that dwarf galaxies merge to produce bigger galaxies at the cluster centers. Consistently, we find that the shape of the bright-end of the early-type LF does not depend on the environment, which suggests that bright early-type galaxies are not a recent product of the cluster environment. In fact, the luminosity density profile of bright early-type galaxies has not evolved significantly since redshift $`z0.5`$ (Ellingson 2003).
The most likely explanation for the lack of dwarf galaxies near the cluster center is tidal or collisional disruption of the dwarf galaxies. The fate of the disrupted dwarfs is probably to contribute to the intra-cluster diffuse light (e.g. Feldmeier et al. 2004; Murante et al. 2004; Willman et al. 2004).
The difference between the cluster and field LF could thus be explained as a difference in morphological mix, plus a density-dependent dwarf early-type galaxies LF, that, added to an invariant bright early-type LF, produces a more or less important and bright upturn, depending on the density of the environment.
## 9 Conclusion
We have presented a detailed analysis of the cluster individual and composite luminosity functions down to $`14`$ mag in all the Sloan photometric bands. All the luminosity functions are calculated within the physical size of the systems given by $`r_{500}`$ and $`r_{200}`$. The main conclusions of our analysis are as follows:
* We confirm that the composite LF shows a bimodal behavior with a marked upturn at the faint magnitude range. A double Schechter component function is the best fit for the cluster LF. We show that calculating the individual and the composite LF within a fixed aperture for all the systems introduces selection effects. These selection effects justify the differences observed in the faint end of the individual cluster LFs studied in paper II and the anti-correlations between DGR and the global cluster properties (mass, velocity dispersion, optical and X-ray luminosities) observed in this work. If the cluster LF is calculated within the physical size of the system ($`r_{500}`$ or $`r_{200}`$), the differences due to aperture effects disappear and the individual cluster LF is well represented by the composite LF. Therefore, we conclude that the shape of the cluster LF is universal in all the magnitude ranges.
* We use the $`ur`$ color to study the color distribution of the faint cluster galaxies. The color distribution confirms that the contamination due to background galaxies is due to field-to-field variance of the background. We apply the color cut at $`ur=2.22`$ suggested by Strateva et al. (2001) to separate early-type from late-type galaxies and study the composite LF by morphological type. We observe that the upturn at the faint magnitudes shown by the complete LF is due to early-type galaxies while the late-type LF is well represented by a single Schechter function.
* We study the cumulative and the differential radial profile of the faint early- and late-type galaxies in clusters. The faint early-type galaxies are concentrated in the central regions while the faint late-type galaxies dominate the outskirts of the systems. The analysis of the color-density relation in a reference sample of nearby galaxies selected from the SDSS spectroscopic sample suggests that red galaxies could be a typical cluster galaxy population. Our analysis show that the bright red population seems to have a luminosity distribution absolutely independent from the behavior of the faint red galaxies in different environments. We observe a fading of the LF upturn toward the cluster core.
* We propose to interpret our results in term of a combination of two processes, transformation of spirals and dIrr into dwarf early-type galaxies via harassment, and disruption of dwarf galaxies near the cluster center by collisions and/or tidal effects.
Whether galaxies evolve from one type to another, in response to the local density, to create the morphology-density relation, or whether the relation is established when the galaxies form, is still an open issue (see, e.g., Dressler 2004). Photometric data alone cannot provide conclusive indications about the nature and the origin of the dwarf population in cluster. In this respect, it would be very useful to sample the velocity distributions of a large set of dwarf galaxies in clusters, in order to constrain their orbital characteristics as it has recently been done for bright cluster galaxies (Biviano & Katgert 2004). If the dwarf early-type galaxies evolve from spirals, radially elongated orbits are expected, while if dwarf early-type galaxies are a more pristine cluster population, their orbits should resemble the isotropic orbits of ellipticals. Additional insights may come from higher accuracy spectroscopy of the dwarf galaxies, allowing to deduce information about their internal velocity dispersion and metallicity, which could be used to put constraints on their age (see, e.g., Kauffmann et al. 2004; Carretero et al. 2004).
###### Acknowledgements.
We would like to thank the anonymous referee for the usefull comments which significantly improved the paper.
Funding for the creation and distribution of the SDSS Archive has been provided by the Alfred P. Sloan Foundation, the Participating Institutions, the National Aeronautics and Space Administration, the National Science Foundation, the U.S. Department of Energy, the Japanese Monbukagakusho, and the Max Planck Society. The SDSS Web site is http://www.sdss.org/. The SDSS is managed by the Astrophysical Research Consortium (ARC) for the Participating Institutions. The Participating Institutions are The University of Chicago, Fermilab, the Institute for Advanced Study, the Japan Participation Group, The Johns Hopkins University, Los Alamos National Laboratory, the Max-Planck-Institute for Astronomy (MPIA), the Max-Planck-Institute for Astrophysics (MPA), New Mexico State University, University of Pittsburgh, Princeton University, the United States Naval Observatory, and the University of Washington.
|
warning/0506/cond-mat0506388.html
|
ar5iv
|
text
|
# Efficiency in nanostructured thermionic and thermoelectric devices
## I Introduction
Traditional vacuum thermionic power generators Schilichter (1915); Wilson (1959); Houston (1959) with macroscopic gaps between emitter and collector plates are limited to very high temperature applications ($`T_H>`$ 1000 K). Refrigeration using such devices, as first suggested by Mahan Mahan (1994), is also limited to high temperatures due to a lack of suitable materials with work-functions below $``$0.3 eV.
Nanostructures are currently being investigated in an attempt to develop thermionic devices that can refrigerate or generate power at lower temperatures. The potential for achieving lower barrier heights via the use of semiconductor heterostructures was pointed out by Shakouri and Bowers Shakouri and Bowers (1997a, b), with Mahan et al. Mahan and Woods (1998); Mahan et al. (1998) suggesting multilayers as a means of reducing the phonon heat leaks inherent in the use of solid-state rather than vacuum devices. Successful solid-state thermionic cooling of up to a few degrees has been reported Shakouri et al. (1999); Labounty et al. (2000); Fan et al. (2001a, b).
Another promising direction is the use of nanometer gaps between the emitter and the collector to lower the work function via quantum tunnelling Hishinuma et al. (2001, 2002), with Hishinuma et al. reporting cooling of about a mK in such a system. Cooling by field emission from carbon nanotubes and other nanostructures has also been proposed Miskovsky and Cutler (1999); Fisher and Walker (2002).
In thermoelectrics, nanostructured devices may offer the possibility of substantially increasing the thermoelectric figure of merit, $`ZT`$, over that of traditional bulk bismuth telluride based devices ($`ZT1`$) due to enhanced electron transport and phonon blocking properties. Hicks and Dresselhaus have predicted that $`ZT`$ can be enhanced using quantum-well superlattices Hicks and Dresselhaus (1993a) and quantum wires Hicks and Dresselhaus (1993b). Venkatasubramanian et al. have reported the highest $`ZT`$ to date, $`ZT2.4`$ using a p-type Bi<sub>2</sub>Te<sub>3</sub>/Sb<sub>2</sub>Te<sub>3</sub> superlattice Venkatasubramanian et al. (2001). Other methods used and suggested for the enhancement of the figure of merit include the use of quantum-dot superlattices Harman et al. (2002); Balandin (2003), superlatices with a non-conservation of lateral momentum Vashaee and Shakouri (2004a, b), inhomogeneous doping Humphrey and Linke (2005) and nanotubes Zhao et al. (2005).
Many of these approaches offer the possibility of engineering the energy spectrum (the number of electrons transmitted through the device as a function of energy) in a way that was not possible in traditional vacuum thermionics or bulk thermoelectrics. In light of the new design freedom offered by nanostructures, it is useful to re-examine the impact of the electron energy spectrum upon what has been called the โelectronic efficiencyโ of thermionic devices Hatsopoulos and Syftopoulos (1973), defined as the efficiency associated with strictly electronic processes under ideal conditions of particle transport.
Improvements in electronic efficiency due to better device design which can be achieved without lowering the power will translate into an improvement in the operating efficiency of practical thermionic devices where non-ideal effects, such as phonon and radiative leaks, as well as contact and lead resistances, are important. To achieve high overall efficiency in practical devices it is important to design devices that not only achieve low thermal conductivity, but high electronic efficiency at finite power as well.
In this paper we analyse in detail the dependence of the electronic efficiency of thermionic power generators and refrigerators upon the details of the energy spectrum of electrons transmitted ballistically between the emitter and collector. The term energy filtering is often used to indicate a restriction of electron flux to all those electrons above a certain energy. Here the term energy filtering will be used in a more general sense to indicate any arbitrary restriction on the energy spectrum of transmitted electrons. We examine two idealised models of thermionic nanodevices. In the first, energy (or more precisely, momentum) filtering of electrons occurs in the direction of transport only. This model, which we will denote as a โ$`k_x`$-filtered thermionic deviceโ, is applicable to single-barrier or multibarrier (superlattice) solid-state devices. In the second model, which we will denote as a โ$`k_r`$-filtered thermionic deviceโ, energy filtering of the total energy of electrons is assumed to be possible. This model is applicable to vacuum emission from nanostructures such as carbon nanotubes and solid-state devices in which there is periodic modulation of the potential in all three dimensions (such as quantum dot superlattices), or superlattices in which there is non-conservation of electron momentum in directions perpendicular to transport Vashaee and Shakouri (2004a, b). Fig. 1 shows geometrically the range of electrons transmitted in idealised $`k_x`$ and $`k_r`$ type devices in momentum space.
With thermoelectric devices, the presence of a bandgap in different crystallographic directions ensures electrons contributing to the current have a certain minimum value of momentum in all three dimensions. The close relationship between thermionic and thermoelectric devices has been analysed by a number of authors Mahan et al. (1998); Nolas and Goldsmid (1999); Vining and Mahan (1999); Humphrey et al. (In Pressa). In this paper, this comparison is extended to similarities between the dependence of the electronic efficiencies of these devices on the details of the electron energy spectrum.
Three main results are presented in this paper. Firstly, it is shown that while $`k_r`$ devices may achieve electronic efficiency equal to the Carnot value, conventional $`k_x`$ devices are fundamentally limited to efficiencies less than this. Secondly, the details of the electron energy spectrum are shown to have a significant impact on the electronic efficiency of the device. Narrower electron energy filters and, more significantly, a sharp rise in the transmission probability from zero to complete transmission give dramatic improvements in electronic efficiency. Using a numerical model of ballistic transport in semiconductor hetrostructures, we show that sharply-rising transmission probabilities yielding high electronic efficiency and power may be achieved with wide single barrier and multibarrier devices. Finally, the equivalence of the ballistic and diffusive formalisms for devices with length the order of the electron mean free path Humphrey et al. (In Pressa) means that in this regime, the electronic efficiency of thermionic and thermoelectric devices will have the same dependence upon the details of the electron energy spectrum.
## II Transport Theory
### II.1 Ballistic Transport Theory
A thermionic device consists of two electron reservoirs at different temperatures and electrochemical potentials, separated by a barrier, or series of barriers, which limit the flow of electrons between them to a certain energy range. Whether the device operates as a power generator, pumping high-energy electrons from the hot to the cold against the electrical potential difference, or as a refrigerator, removing high-energy electrons from the cold reservoir, depends upon the relative magnitudes of the opposing temperature and electrochemical potentials.
In a $`k_r`$-filtered thermionic device, where the transmission probability, $`\zeta `$, is a function of the total electron energy, $`E=\mathrm{}^2\text{k}^2/2m^{}`$, the net electrical current density flowing from the cold to the hot reservoir is
$$J_r=q_0^{\mathrm{}}\left[n_r^Cn_r^H\right]\zeta (E)๐E$$
(1)
where $`q=1.602\times 10^{19}`$ C is the charge of an electron and
$$n_r^{C/H}=\frac{m^{}E}{2\pi ^2\mathrm{}^3}f(E,\mu _{C/H},T_{C/H})$$
(2)
is the number of electrons with total energy $`E`$ arriving at the three-dimensional reservoir interface per unit area per unit time and
$$f(E,\mu _{C/H},T_{C/H})=\left[1+\mathrm{exp}\left(\frac{E\mu _{C/H}}{k_BT_{C/H}}\right)\right]^1$$
(3)
is the Fermi-Dirac distribution function in the cold/hot reservoir with chemical potential $`\mu _{C/H}`$ and temperature $`T_{C/H}`$. We have assumed that electron velocity is determined by the reservoirs. A more detailed theory would be required to account for any velocity changes due to the device structure.
One may calculate the heat flux out of the hot and cold reservoirs by noting that an electron leaving or entering the cold/hot reservoir will remove or add respectively an amount of heat equal to the difference between the energy of the electron and the average energy of electrons in the reservoir, that is $`E\mu _{C/H}`$. Introducing this factor inside the integral for number current we may obtain expressions for the net heat flux out of the cold/hot reservoir as
$$\dot{Q}_r^{C/H}=_0^{\mathrm{}}(E\mu _{C/H})\left[n_r^Cn_r^H\right]\zeta (E)๐E.$$
(4)
In a $`k_x`$-filtered device the transmission probability is a function of what may be loosely defined as the โkinetic energy of electrons in the $`x`$ directionโ, $`E_x=\mathrm{}^2k_x^2/2m^{}`$. It is therefore convenient in this case to write the electrical and heat currents in terms of $`E_x`$ Davies (1998)
$$J_x=q_0^{\mathrm{}}\left[n_x^Cn_x^H\right]\zeta (E_x)๐E$$
(5)
where
$$n_x^{C/H}=\frac{m^{}k_BT_{C/H}}{2\pi ^2\mathrm{}^3}\mathrm{log}\left[1+\mathrm{exp}\left(\frac{E_x\mu _{C/H}}{k_BT_{C/H}}\right)\right]$$
(6)
is the number of electrons with kinetic energy in the $`x`$ direction $`E_x`$ arriving at the reservoir interface per unit area per unit time.
In a $`k_x`$-filtered device, the average heat removed from the cold/hot reservoir when an electron with energy in the $`x`$ direction $`E_x`$ leaves, $`E_x+k_BT_{C/H}\mu _{C/H}`$ (assuming Maxwell-Boltzmann statistics), is not the same as that added when an electron with energy in the $`x`$ direction $`E_x`$ arrives, $`E_x+k_BT_{H/C}\mu _{C/H}`$. This difference is due to the fact that, while the barrier system filters electrons according to their momentum in the direction of transport, their momenta in the other two dimensions may take any value, contributing on average an extra $`k_BT_{C/H}`$ to the energy of electrons emitted from the cold/hot reservoir. The heat flux out of the cold/hot reservoir in a $`k_x`$ filtered device is therefore given by
$`\dot{Q}_x^{C/H}={\displaystyle _0^{\mathrm{}}}[(E_x+k_BT_{C/H}\mu _{C/H})n_x^C`$
$`(E_x+k_BT_{H/C}\mu _{C/H})n_x^H]\zeta (E_x)dE.`$ (7)
It may be noted that many cryogenic ballistic refrigerators such as normal-insulating-semiconductor (NIS) junction devices Nahum et al. (1994); Leivo et al. (1996); Manninen et al. (1997) and quantum dot refrigerators Edwards et al. (1995) utilise either two- or one-dimensional reservoirs where the difference between $`k_x`$ and $`k_r`$ filtered devices are less dramatic or non-existent.
The electronic efficiency as a power generator and coefficient of performance (COP) as a refrigerator for both $`k_x`$ and $`k_r`$ filtered devices are given by
$$\eta ^{PG}=VJ/\dot{Q}^H$$
(8)
and
$$\eta ^R=\dot{Q}^C/VJ$$
(9)
respectively, where $`V=(\mu _C\mu _H)/q`$.
### II.2 Diffusive Transport Theory
Thermoelectric devices are generally differentiated from thermionic devices according to whether electron transport is diffusive or ballistic Mahan et al. (1998). There is, however, little to distinguish the underlying thermodynamics of the two types of device, with both achieving reversibility under the same conditions Humphrey et al. (2002); Humphrey and Linke (2005) and both being governed by the same โmaterials parameterโ Vining and Mahan (1999); Ulrich et al. (2001); Humphrey et al. (In Pressa).
Under the relaxation-time approximation the electric current in a thermoelectric device may be calculated using the Boltzmann transport equation as
$$J^d=qD_l\left[v_x^l\right]^2\tau \frac{df}{dx}๐\text{k}.$$
(10)
where $`D_l`$ is the local density of states (DOS), $`\tau =\tau _0E^r`$ is the relaxation time, and $`v_x^l=(1/\mathrm{})[E(k_x)/k_x]`$ is the velocity in the direction of transport. The electron energy spectrum in a diffusive device is thus determined by $`D_l\left[v_x^l\right]^2\tau [df/dx]`$.
The transport equation for ballistic devices, where the mean free path of an electron between collisions is greater than the width of the barrier, or system of barriers, may be written similarly as
$$J^b=qD_r\zeta v_x^r\mathrm{\Delta }f๐\text{k}$$
(11)
where $`D_r=1/(2\pi )^d`$ is the DOS in $`k`$-space in the reservoirs where $`d`$ is the dimensionality of the reservoirs, and $`\mathrm{\Delta }f=f_Cf_H`$ is the difference between the distribution functions in the cold and hot reservoirs. The electron energy spectrum in a thermionic device is therefore determined by $`D_rv_x^r\zeta \mathrm{\Delta }f`$.
We expect that Eqs. 11 and 10 should yield the same results for devices of width close to the electron mean free path. If we take the energy dependence of the relaxation time to be $`r=1/2`$, which is appropriate when scattering is dominated by acoustic phonons, the mean free path in the direction of transport will be independent of energy and given by $`\lambda v_x\tau `$ Humphrey et al. (In Pressa). For a small piece of thermoelectric material of length approximately equal to the electron mean free path $`df/dx\mathrm{\Delta }f/\lambda `$. Eq. 10 then reduces to Humphrey et al. (In Pressa)
$$J^d=qD_lv_x^l\mathrm{\Delta }f๐\text{k}$$
(12)
and is of the same form as that of the ballistic transport equation, Eq. 11. Thus, the term $`D_lv_x^l`$ in the diffusive formalism plays the same role as $`D_rv_x^r\zeta `$ in the ballistic formalism. We therefore expect the dependencies of the electron energy spectrum in both thermionic and thermoelectric device to be similar. Since $`v_x^r`$ and $`D_r`$ are fixed by the reservoirs, at fixed temperature/chemical potential the electron energy spectrum in a ballistic device is determined by the transmission probability as device structure varies. In a diffusive device, both $`D_l`$ and $`v_x^l`$ may change when the device structure is altered and affect the energy spectrum.
The heat-current density out of the cold/hot reservoir is given by
$$\dot{Q}_{C/H}=(E\mu _{C/H})D_lv_x^l\tau \frac{df}{dx}๐\text{k}.$$
(13)
Due to the equivalence of the diffusive and ballistic formalisms in this regime, the intensive efficiency across a small section of thermoelectric material Snyder and Ursell (2003); Vining (1997) and the electronic efficiency/COP of a ballistic device are given by Eqs. 8 and 9 respectively.
## III Reversible Electron Transport
To achieve reversibility in a thermionic or thermoelectric device, electrons must flow only at energies where the Fermi occupation of states, Eq. 3, is constant Humphrey et al. (2002); Humphrey and Linke (2005). Assuming a finite temperature difference at each end of the device, there are two different quasi-static limits in which this requirement is satisfied.
The first way is to restrict the flow of electrons to those with energies approaching infinity where the occupation of states tends to zero. This may be achieved, for instance, with an intrinsic semiconductor where the band gap approaches infinity in a thermoelectric device or an infinitely high barrier system in a thermionic device. In vacuum thermionic devices operating at very high temperatures ($`T_H>1500`$ K), optimim efficiency is approached when the barrier height is almost 20 times larger than $`k_BT_H`$ Hatsopoulos and Syftopoulos (1973), meaning that electronic efficiencies close to the Carnot limit may be obtained. However, at the more moderate emitter temperatures, 300 K $`<T_H<800`$ K, of interest in most applications, achieving finite power production or refrigeration via a thermionic device requires a much lower barrier height, of the order of a few $`k_BT_H`$. It is therefore more practical to utilize the other quasi-static limit to achieve high electronic efficiencies.
The second way to achieve reversibility in a thermionic or thermoelectric device, which we refer to as energy-specific equilibrium, is to allow electrons to flow only at a single energy where the Fermi occupation of states throughout the device is the same Humphrey et al. (2002),
$$E_0=\frac{\mu (T+\delta T)(\mu \delta \mu )T}{\delta T}$$
(14)
where $`\delta T`$ and $`\delta \mu `$ are the temperature and chemical potential changes respectively over a distance $`\delta x`$ in the device. At this energy, the effect, or, in the language of irreversible thermodynamics, the โaffinityโ Callen (1960), of the opposing temperature and electrochemical potential gradients upon electrons exactly cancels and transport occurs reversibly. This is also the energy at which the energy-resolved current changes sign, that is, for electrons with energies less than $`E_0`$ the net current flows from the hot to cold reservoir and for energies greater than $`E_0`$ net current flows from the cold to hot reservoir. Transport of electrons of a single energy only might be achieved using resonant tunneling in a superlattice or quantum dot superlattice. For a thermionic device, the ballistic transmission energy is determined substituting the cold and hot reservoir temperatures and chemical potentials into Eq. 14. Here we shall denote a filtering system which transmits only a single energy of electrons between the reservoirs, be that the single total energy for a $`k_r`$ device or a single $`x`$ energy for a $`k_x`$ device, as an โideal filterโ. For a ballistic device this may be expressed as a transmission probability function as
$$\zeta (E)=\{\begin{array}{cc}1& E=E^{}\\ 0& \text{elsewhere}\end{array}$$
(15)
where $`E_x`$ would be substituted for $`E`$ in a $`k_x`$ system.
In a thermoelectric device, inhomogeneous doping or a graded band structure is required so that Eq. 14 may be satisfied at every point in the device for a particular temperature gradient such that the energy gap between the chemical potential and the transmission energy is given by Humphrey and Linke (2005)
$$E_0\mu _0(x)=\left[\frac{eV_{OC}}{T_HT_C}\right]T(x)$$
(16)
where $`V_{OC}`$ is the open circuit voltage.
## IV Electronic Efficiency With Ideal Filtering
Under ideal filtering, as defined in the previous section, Eqs. 8 and 9 for the $`k_r`$ device reduce to
$$\eta _r^{PG}=\frac{\mu _C\mu _H}{E_0\mu _C}$$
(17)
for power generation ($`E^{}>E_0`$) and
$$\eta _r^R=\frac{E_0\mu _C}{\mu _C\mu _H}$$
(18)
for refrigeration ($`E^{}<E_0`$). The efficiency and COP of ideally filtered $`k_x`$ and $`k_r`$ systems are plotted in Fig. 2 relative to the Carnot values. When the filtering energy is $`E_0`$, reversibility and the Carnot efficiency are achieved for the $`k_r`$ device as shown in Fig. 2. The energy axis for the $`k_x`$ device shown in Fig. 2 is the average total cold reservoir energy, $`E_x+k_BT_C`$.
For all values of total energy shown in Fig. 2, the $`k_r`$ device outperforms the $`k_x`$ device. Importantly, unlike the $`k_r`$ device, the $`k_x`$ filtered thermionic device does not reach the Carnot efficiency for arbitrary electrochemical potentials and finite barrier heights. The reason for this is that although momentum in the $`x`$ direction is restricted to a single value, the momentum in the $`y`$ and $`z`$ directions may take any value, meaning that the energy spectrum has a finite width and reversibility is not achieved. The distributed nature of total electron energies for a $`k_x`$ device, even with a narrow filter, is shown in Fig. 3. For $`k_x`$-filtered power generators, this upper bound upon the electronic efficiency can be obtained analytically in the limit that ($`\mu _C\mu _H)/k_B(T_HT_C)1`$, in which case maximum efficiency is obtained when $`E^{}=E_0`$, where
$$\eta _x^{PG}=\eta _C\left[1+\eta _C(k_BT_H+k_BT_C)/qV\right]^1$$
(19)
where $`\eta _C`$ is the Carnot efficiency. Given that we have taken $`\mu _C>\mu _H`$, so that $`qV`$ is positive, it can be seen by inspection that Eq. 19 always yields an efficiency less than $`\eta _C`$. For the system analysed in Fig. 2, Eq. 19 gives a maximum electronic efficiency for the $`k_x`$ power generator of 80% of the Carnot limit, in agreement with numerical results. This constitutes the first main result of the paper, that for finite barrier heights and electrochemical potential differences, $`k_x`$ filtered thermionic devices are limited to a maximum electronic efficiency less than the Carnot limit. This means that from the point of view of maximising electronic efficiency, $`k_r`$ devices are inherently superior to $`k_x`$ devices.
## V Electronic Efficiency With Non-Ideal Filtering
The filters considered in Sect. IV represent an idealized theoretical limit. We now extend our analysis to non-ideal filters. Firstly, we consider filters of finite width where all electrons over a certain range of total or $`x`$ energies are transmitted. Secondly, we show that a gradual, rather than sharp switch from zero to full transmission has a significant impact on the electronic efficiency of thermionic devices. This will constitute the second main result of the paper.
### V.1 Effect of Finite Filter Width
A filter of finite width corresponds to a transmission probability of
$$\zeta (E)=\{\begin{array}{cc}1& E^{}<E<E^{}+\mathrm{\Delta }E\\ 0& \text{elsewhere}\end{array}$$
(20)
for a $`k_r`$ system, and where $`E_x`$ would be substituted for $`E`$ for a $`k_x`$ system. Such a filter might be used, for example, to approximate a transmission miniband in a superlattice device. For each filter width examined numerically, the starting energy of the filter, $`E^{}`$, was tuned to find the maximum electronic efficiency/COP for that width. The results for filters of width 0.01$`k_BT_C`$ to 100$`k_BT_C`$ are plotted in Fig. 4. The filter of 0.1$`k_BT_C`$ is narrow enough to effectively perform ideal filtering, reflected in the fact that the $`k_r`$ electronic efficiency/COP approaches the the Carnot value for this width and the $`k_x`$ values reach the maximum values obtained in Fig. 2. Fig. 4 shows however that we do not require an ideal filter to achieve an efficiency/COP very close to the maximum value, as seen in the plateau in all curves. The $`k_r`$ and $`k_x`$ systems may achieve an efficiency/COP approximately equal to the maximum value for filter widths of less than about 0.1$`k_BT_C`$ and $`k_BT_C`$ respectively. Filter widths of around these sizes are achievable using practical semiconductor devices as will be discussed later. As the filter widths increase beyond these values the efficiency/COP drops and then plateaus again at a final value. Large filter widths effectively correspond to the situation where all electrons above $`E^{}`$ are being transmitted. As the distribution function rapidly converges to zero at high energies, this means that further increasing filter width has a minimal effect upon the electronic efficiency.
Fig. 5 shows the energy spectrum of the net electric current transmitted from the hot to cold reservoir for a 0.3 eV wide filter. Results are normalized by the net number of electrons with total energy greater than the Fermi energy available to flow between reservoirs such that the number of electrons in $`i`$th energy band are given by
$$N_i=\frac{_{E_i}^{E_i+\delta i}n_{r/x}๐E}{|_{\mu _C}^{\mathrm{}}n_r๐E|}.$$
(21)
It should be noted that the $`k_x`$ energy spectrum in Fig. 5 does not show the total energy spread due to the unfiltered degrees of freedom, as was shown in Fig. 3, since in this case we are considering the spectrum with regard to filtered $`x`$ component of energy only. This illustrates the energy range of the filter for each system when tuned for maximum electronic efficiency/COP.
Fig. 5 shows that there are more electrons being transmitted for the $`k_r`$ system than with the $`k_x`$ system, an effect previously pointed out by Vashaee and Shakouri Vashaee and Shakouri (2004a, b), which results in greater power in a $`k_r`$ device. The calculations presented here show that the difference in the energy spread of electrons in $`k_x`$ and $`k_r`$ filtered devices also gives an increase in the electronic efficiency for $`k_r`$ devices due to a greater concentration of electrons with energies around $`E_0`$. For both refrigeration and power generation, the filters will be positioned such that electrons with energy $`E_0`$ are included. Since when $`E>E_0`$ the net energy-resolved current produces power, the lower edge of the $`k_r`$ power generator filter will always be at $`E_0`$. For the $`k_x`$ system this lower edge is shifted to lower $`x`$ energy due to the additional energy contribution in the two other spatial degrees of freedom. Energy-resolved current in the energy range $`\mu _C<E<E_0`$ refrigerates the cold reservoir and the lower edge of the filter is therefore shifted to this region in Fig. 5(a). Since there are more electrons at higher energies, however, current flow in the region $`E>E_0`$ generates power and a trade off occurs when positioning the filter for maximum COP. Again, the $`x`$ energy of the $`k_x`$ filter is lower than the total energy of the $`k_r`$ filter due to the unfiltered energy contributions.
### V.2 Transmission Probabilities With Finite Slopes
Thus far we have considered only the case where there is a sharp transition from zero to full transmission of electrons. In this section we consider the effect upon the electronic efficiency of a gradual transition, which more closely resembles the shape of the transmission probability in practical devices. We begin by using two convenient โartificialโ transmission probabilities, the slope of which can be easily varied. The first, a Gaussian peak which might approximate the transmission probability of a resonance, is given by
$$\zeta (E)=\mathrm{exp}\left(\frac{(EE_c)^2}{w}\right)$$
(22)
where $`E_c`$ defines the center energy of the peak and $`w`$ is a width parameter which is used to vary the sharpness of the slope. $`E_x`$ would be substituted for $`E`$ for a $`k_x`$ device.
The second artificial transmission probability considered is a โhalf-Gaussianโ intended to approximate the transmission probability of a single barrier of finite width. This is given by Eq. 22 for $`EE_C`$ and is equal to one for $`E>E_C`$. The sharpness of the Gaussian and half-Gaussian transmission probabilities were varied between $`w=10^5`$, corresponding to an ideal filter or perfectly sharp single barrier transmission probability, and $`w=0.1`$. The transmission probabilities associated with these extreme values are shown in 6(a) and (b). The system bias was tuned for each transmission probability for maximum electronic efficiency/COP. Fig. 7(a) shows the COPs associated with a room temperature refrigerator and Fig. 7(b) shows the electronic efficiencies of a heat engine operating at higher temperature.
Since all electrons of energy other than $`E_0`$ reduce the electronic efficiency we expect the sharpest peak in Figs. 7(a) and (b) to yield the highest efficiency/COP, and this is confirmed by the numerical results. The most interesting result, however, is that the electronic efficiency of the half-Gaussian transmission probability is very strongly dependent upon how sharply the transmission rises from zero to unity. A smooth rise in the transmission probability lowers the electronic efficiency for the same physical reason that a $`k_x`$ filtered device has a lower electronic efficiency than a $`k_r`$ filtered device. Net current flow is from the hot to cold reservoir for electron energies just above $`E_0`$, generating power with efficiency approaching the Carnot limit. Conversely, net current flow is from the cold to the hot reservior for electron energies below $`E_0`$ and above $`\mu _C`$, absorbing power and refrigerating the cold reservoir, while the net current transmitted at energies below $`\mu _C`$ both absorb power and heat the cold reservoir. This means that whenever the energy spectrum of transmitted electrons rises slowly to its peak value there is an efficiency lowering trade off which occurs between transmitting the maximum number of electrons with energies near $`E_0`$, which refrigerate or generate power with Carnot efficiency, and minimising the number of electrons transmitted in the range $`E<E_0`$ for power generation or in the range $`E<\mu _C`$ and $`E>E_0`$ for refrigeration. As the transmission probabilities become less sharp, the performance difference between the $`k_x`$ and $`k_r`$ and Gaussian and half-Gaussian systems becomes less significant.
So far we have established the two criteria for maximum electronic efficiency in thermionic power generators and refrigerators. Firstly, we have shown that the narrower the energy spectrum the higher the electronic efficiency. However, in general a gain in electronic efficiency via this mechanism is obtained at the expense of the power of the device. The second criterion is that the sharper the transition from zero to peak value in the energy spectrum, the higher the electronic efficiency. This second method offers the significant advantage of improving the electronic efficiency without sacrificing power through the use of a narrow filter. The maximum power achievable is also greater with a sharply-rising transmission probability if the barrier height is optimised Humphrey et al. (In Pressb). In the next section we analyse design considerations for thermionic devices considering both electronic efficiency and power.
## VI Design Considerations For Achieving High Electronic Efficiency In Practical Devices
### VI.1 Ballistic Devices
Semiconductor-based devices, including superlattices, may be specifically designed to achieve the desired energy spectrum features in $`k_x`$ devices. Filter widths around those required for achieve near-maximum electronic efficiency, as discussed in Sect.V.1, may be achieved using a variably-spaced superlattice energy filter (VSSEF) as proposed by Summers, Brennan and Gaylord Summers and Brennan (1986); Gaylord and Brennan (1988). Such devices consist of alternating semiconductor layers with barrier and well widths chosen such that energy levels in the wells are closely aligned under bias. Tunneling through a simpler multibarrier structure may also suffice. Similarly, a miniband in the transmission probability for a superlattice might be used as a narrower filter compared with complete transmission above the barrier energy. Quantum dot structures Bryllert et al. (2003) or normal-insulating-superconductor junction (NIS) devices Edwards et al. (1995) can also achieve narrow electron transmission bands and may be used for refrigeration at cryogenic temperatures. Relatively narrow energy electron emission peaks from carbon nanotubes have been reported which may be of use in a vacuum based device Fransen et al. (1999). Since the DOS in the reservoirs fixes the number of electrons available for transport in a certain energy range in ballistic devices, the reduction of power in narrow transmission probability devices, with only modest gains in electronic efficiency, is expected to be undesirable in the presence phonon heat leaks.
It is likely that the best way to simultaneously achieve high electronic efficiency and high power in a ballistic device is to design the structure such that the transmission probability rises sharply from zero to one and remains close to unity beyond this. Whilst the most obvious way to achieve a transmission probability of this nature is to utilise a single barrier with a width as large possible (but less than the mean free path of electrons) here we show that an array of thin barriers can also be used to engineer a transmission probability that rises sharply from zero to unity.
The transmission probabilities and associated efficiencies/COPs for single rounded barriers of various widths have been calculated. The transmission probabilities were calculated by obtaining a numerical solution to the time-independent Schrรถdinger equation based on Airy functions Davies (1998); Brennan and Summers (1987). Fig. 8 shows the significant difference in sharpness between a 10-nm and 100-nm barrier. We see in Fig. 9, as expected, the wider barriers with sharper transmission probabilities give the highest efficiencies/COPs approaching the maximum value, in this case, at a width of around 35 nm. Beyond this width, there is little to be gained in terms of electronic efficiency, although phonon mediated heat leaks continue to be reduced with increasing barrier width.
From another point of view, wide barriers might be undesirable. Devices with greater interface density may reduce thermal conductivity as a result of interface scattering and phonon miniband formation Cahill et al. (2003). Here we consider a device where multiple barriers are traversed in an electron mean free path. This allows quantum mechanical effects to be utilised to achieve high electronic efficiency using narrow barriers which give low electronic efficiency when used individually.
Multiple narrow barriers over a distance of the order of the electron mean free path may be used to give a transmission probability that is as sharp as if the electrons were traversing a single wide barrier. Figs. 10(a) and (b) show the transmission probabilities calculated for two-barrier and eight-barrier systems respectively as well as the very smoothly rising transmission probability for a single 5-nm barrier for comparison. The efficiencies/COPs achieved are within 3% of those of a wide single barrier. Thus, high electronic efficiencies may be achieved, whilst allowing the flexible use of narrower barriers.
The โturn-onโ transmission energy for a device with many thin barriers may be shifted to lower energy as shown in Fig. 11(a) and (b). If the chemical potential remains constant, this lowering of the turn-on energy may result in a decrease in efficiency and increase in power compared to a wider single barrier due to the decrease in the work function, $`\varphi =E_{turnon}\mu `$. With the same work functions, which may be achieved by altering the chemical potentials as detailed in Figs. 11(a) and (b), the wide single barrier and many thin barrier systems achieve approximately the same electronic efficiency.
We do not expect a dramatic change in dependence of device behaviour on the electron energy spectrum as the total length of the device increases beyond an electron mean free path. Since the probability of an electron traveling distance $`L`$ without suffering a collision is given by Ashcroft and Mermin (1976)
$$P=\mathrm{exp}(\lambda /L)$$
(23)
the ballistic and diffusive formalisms, Eqs. 11 and 10, may be combined to show that the electrical current will be given by Humphrey et al. (In Pressa)
$$J=q\mathrm{\Delta }f\left(D_rv_x^r\zeta P+\frac{\lambda }{L}D_lv_x^l[1P]\right)๐\text{k}$$
(24)
if $`L\lambda `$. As the number of barriers increases and the overall length of the device becomes significantly greater than the electron mean free path, the ballistic term becomes small so that transport is accurately described using the diffusive formalism, as discussed in the next section.
### VI.2 Diffusive Devices
As was discussed earlier, the energy spectrum in a diffusive device is determined by the product of the local DOS, the velocity squared and the relaxation time at constant temperatures/chemical potentials. The local DOS of an infinite superlattice may be determined using the Kronig-Penney model Davies (1998); Lin and Dresselhaus (2003). Fig. 12 shows the DOS calculated for a many-barrier system and the calculated transmission probability showing the clear relationship between the two. In a pure ballistic device, since the DOS and velocity are fixed by the reservoirs, a sharp electron energy spectrum is achieved via a sharp transmission probability. In diffusive devices, both the DOS and electron velocity may change as the device structure changes and it may be a difficult optimisation problem to design a structure where their product changes sharply from its minimum to maximum value as a function of energy Shakouri (2005).
Mahan and Sofo have shown that the ideal transport distribution function for a thermoelectric device may be achieved with a delta function DOS Mahan and Sofo (1996). Humphrey and Linke showed that the Carnot efficiency may be achieved in thermoelectric devices utilising a delta function DOS and a graded device structure or inhomogeneous doping Humphrey and Linke (2005). Their results are analogous to the results presented earlier in this paper where it was shown that the ideal transmission probability for a ballistic device was one which allowed the transmission of only a very narrow energy range of electrons. The results presented in this paper suggest that not only is the width of the energy spectrum important, but also whether it rises rapidly from zero to its maximum value. In practical devices with loss mechanisms such as phonon heat leaks, the magnitude of the energy spectrum also becomes important to the efficiency as the conductivity is given by the integral of the energy spectrum and occupation of states. Hicks and Dresselhaus have pointed out that the magnitude of the DOS can be increased by using structures of lower dimensionality, potentially increasing the power factor Hicks and Dresselhaus (1993a, b). We also note that the DOS is sharper for lower dimensional systems compared to bulk materials, which may result in an improved energy spectrum.
## VII Experimentally Observable Properties Related To Electronic Efficiency
The presence of phonon heat leaks and contact resistance in solid-state thermionic devices makes direct measurements of the electronic efficiency difficult. Here we discuss experimental properties which may be measured to provide an indication of the shape of the electron energy spectrum and though this, electronic efficiency.
The I-V characteristics of a thermionic device are dependent on the voltage across the barrier system, $`V_B`$, as shown on Fig. 13. $`V_B`$ may be determined from the supplied voltage, $`V_S`$, and measured current, $`I`$, as
$$V^B=V^S+I(R_A+R_B)$$
(25)
where, $`R_A`$ and $`R_B`$ are contact resistances, when the device is generating power. As the bias is increased, the net electrical current decreases and reaches zero at the open circuit voltage $`V_{OC}^B`$, from which the effective Seebeck coefficient may be calculated as
$$S=\frac{V_{OC}^B}{T_HT_C}.$$
(26)
The energy-specific equilibrium energy may be calculated at open circuit voltage as
$$E_0^{V_{OC}}=\mu _C+V_{OC}^B\frac{T_C}{T_HT_H}=\mu _C+ST_C$$
(27)
and is linearly related to the Seebeck coefficient. $`E_0^{V_{OC}}`$ is the energy where energy-resolved currents above and below it are equal, giving zero net current. For a sharply-rising transmission probability, $`E_0^{V_{OC}}`$ would be be positioned as shown in Fig. 14(a) above the barrier energy. If we have another system where electrons with energies lower than the barrier energy are being transmitted without significant change to the high-energy details, for example through decreasing the barrier width, $`E_0^{V_{OC}}`$ is shifted to lower energy as shown in Fig. 14(b). Measuring this relative to a convenient energy, say the barrier energy, $`E_B`$, provides a convenient sharpness indication for the transmission probability,
$$\psi =ST_C\varphi .$$
(28)
The โturn-onโ energy for a multibarrier system may be shifted to lower energy, in which case, the sharpness indicator should be measured relative to this โeffectiveโ barrier height, which might be calculated using the Kronig-Penney model, as discussed previously. A higher sharpness indicator is desirable, indicating a sharper transmission probability and therefore higher expected electronic efficiency/COP. The sharpness indicator has the advantage over the Seebeck coefficient of being less dependent on the chemical potential/barrier height and more so on the sharpness of the energy spectrum, as shown in Fig. 15. Here, the chemical potentials for a number of single barrier transmission probabilities have been varied to give a constant Seebeck coefficient as the barrier width and transmission probability sharpness change. Fig. 15 shows the electronic efficiency varies significantly in this example. Whilst the Seebeck coefficient remains constant, the sharpness indicator, $`\psi `$, increases as the electronic efficiency increases.
## VIII Conclusions
It has been shown the the nature of the electron-energy spectrum has a significant impact on the performance of thermionic and thermoelectric devices. The limiting efficiency of finite barrier height devices was achieved when electrons of a single energy only are transmitted. Whilst $`k_r`$ devices achieve reversibility when the transmission energy was equal to $`E_0`$, $`k_x`$ devices do not due to the finite energy spread associated with the two unfiltered degrees of freedom. For systems with finite-width rectangular transmission probabilities, electronic efficiency was close to the maximum value for filter widths less than $`k_BT`$, but decreases as the range of transmitted electron energies increases, reaching a steady value as the filter width increases beyond a few $`k_BT`$. Our most important result was that an increase in the sharpness of the rise in electron energy spectrum significantly increases electronic efficiency. Improving the electronic efficiency by increasing the sharpness of the transmission probability may also increase the maximum power. We have shown that sharp transmission probabilities may be achieved using wide single barriers or carefully arranged multiple barriers.
Since, in the diffusive formalism, used to describe thermoelectric devices, the product of the local electron group velocity and the local DOS plays the same role as the product of the reservoir DOS, reservoir velocity and the transmission probability in the ballistic formalism for mean free path length devices, the results presented here showing the benefit of sharply-rising energy spectra on electronic efficiency and power are expected to be relevant to thermoelectric devices.
Finally, the sharpness indicator, $`\psi `$, was suggested as an experimental measure providing an indication of the sharpness of the rise in the energy spectrum of a ballistic device and its electronic efficiency and was shown to be superior for this purpose to the Seebeck coefficient alone.
## IX Acknowledgements
MOโD is supported by the Australian Research Council. TH is supported by the Australian Research Council and funding from ONR MURI. The authors acknowledge helpful discussions with Ali Shakouri and Heiner Linke.
|
warning/0506/nucl-th0506046.html
|
ar5iv
|
text
|
# Neutrino nucleus cross sections for low energy neutrinos at SNS facilities
## I Introduction
The study of neutrino nucleus cross sections at low energies is important in understanding various physical and astrophysical processes. The low energy neutrino beams $`(E_\nu 52.8MeV)`$ are generally obtained from the muons decaying at rest. The neutrino energy spectrum obtained from the muons decaying at rest is given by
$$\varphi (E_{\nu _e})=\frac{12}{E_0^4}E_{\nu _e}^2(E_0E_{\nu _e}),E_0=52.8MeV$$
(1)
This is known as Michel spectrum and is shown in Fig.1.
This neutrino energy spectrum and its energy range is similar to the energy spectrum and energy range of neutrinos coming from the core collapse supernovamclaughlin . This similarity in the energy range and spectrum of the supernova neutrinos with the muons decaying at rest opens up the possibility of connecting the ground based neutrino nuclear experiments with the study of neutrino nuclear cross sections in supernova. Such a study will also be useful in understanding the r-process nucleosynthesis leading to the formation of heavy elements in the interstellar mediumraffelt .
The low energy neutrino beams can also be obtained at the Spallation Neutron Source(SNS) facilities where low energy neutrino nucleus reaction experiments can be performedconf . At these facilities the accelerated protons hit a nuclear target and produce $`\pi ^{}`$ and $`\pi ^+`$ particles which give rise to muon and neutrinos as their decay product. There is a small flux of $`\overline{\nu }_e`$ as a result of $`\mu ^{}`$ decay as most of $`\mu ^{}`$ are captured by nuclei. On the other hand $`\mu ^+`$ are not captured but are stopped in nuclei and decay at rest giving rise to substantial flux of $`\nu _e`$. These neutrinos can be used to study neutrino nuclear cross sections. Experiments to study the feasibility of making such neutrino-nucleus cross section measurements using many nuclear targets are proposed at Oak Ridge National Laboratory(ORNL) using stopped muon neutrino facility at the Spallation Neutron Source(SNS). Such a facility is expected to produce $`10^{15}`$ neutrinos per second of each flavor of $`\nu _e`$, $`\nu _\mu `$ and $`\overline{\nu }_\mu `$ with mono-energetic spectra for $`\nu _\mu `$ and continuous spectra for $`\nu _e`$ and $`\overline{\nu }_\mu `$. Out of these flavors of neutrinos, $`\nu _e`$ produces electrons which are observed to study the neutrino nuclear cross sections for charged current reactions.
The electron production from nuclei induced by low energy $`\nu _e`$ beams has been studied in past by many authors. A review of theoretical calculations and experimental results have been given by Donnelly and Pecceidonnelly , Kubodera et al.kubodera and more recently by Kolbe et al.kolbe1 . It is proposed that these calculations are done for a large range of medium and heavy nuclei and attempts be made to study them experimentally. Such a list of nuclei where the study of neutrino nuclear cross section is of some interest has been recently prepared by Avignone and Efremenkoavignone .
In this paper we have studied neutrino nuclear cross sections induced by low energy $`\nu _e`$ obtained from muons decaying at rest for the nuclei proposed by Avignone and Efremenkoavignone . The calculations have been done in local density approximation taking into account Pauli blocking, Fermi motion effects and renormalization of weak transition strengths in the nuclear medium. The method has been successfully applied to study the various electromagnetic and weak processes in nuclei at low and intermediate energiessingh -athar . The effect of Coulomb distortion of the lepton produced in charge current reactions is taken into account by using the Fermi function $`F(Z,E_e)`$, where Z is the atomic number and $`E_e`$ is the outgoing electron energy as well as in a Modified Momentum Approximation(MEMA), where the effect of Coulomb distortion is incorporated by modifying the momentum and energy of charged lepton in the Coulomb potential of the final nucleuskolbe1 ,engel .
In section-II, we describe the formalism. In section-III, we present the numerical results for the total cross section $`\sigma (E)vs`$ the neutrino energy E and the flux averaged cross section and discuss the nuclear medium effects in some detail. A brief summary and conclusion of our results is given in section-IV.
## II Formalism
In this section we derive an expression for the total scattering cross section $`\sigma (E_\nu )`$ for the charged current(CC) reaction
$$\nu _e+{}_{Z_i}{}^{A}Xe^{}+{}_{Z_f}{}^{A}Y,$$
(2)
where $`Z_i(Z_f)`$ is the charge of initial(final) nucleus.
The basic $`\nu _e`$-neutron reaction taking place in $`{}_{Z_i}{}^{A}X`$ nucleus is
$$\nu _e(k)+n(p)e^{}(k^{})+p(p^{}),$$
(3)
where k and p are the four momenta of the incoming neutrino and neutron and $`k^{}`$ and $`p^{}`$ are the four momenta of the outgoing electron and proton respectively.
The matrix element for the basic neutrino process on free nucleon (Eqn.3) is written as
$$T=\frac{G_F}{\sqrt{2}}\mathrm{cos}\theta _cl_\mu J^\mu ,$$
(4)
where
$`l_\mu `$ $`=`$ $`\overline{u}(k^{})\gamma _\mu (1\gamma _5)u(k)`$
$`J^\mu `$ $`=`$ $`\overline{u}(p^{})[F_1^V(q^2)\gamma ^\mu `$ (5)
$`+F_2^V(q^2)i\sigma ^{\mu \nu }{\displaystyle \frac{q_\nu }{2M}}+F_A^V(q^2)\gamma ^\mu \gamma _5]u(p).`$
The form factors $`F_1^V(q^2)`$, $`F_2^V(q^2)`$ and $`F_A^V(q^2)`$, where $`q^2`$ is the four momentum transfer square i.e. $`(kk^{})^2`$, are isovector form factors and are written as
$$F_1^V(q^2)=F_1^p(q^2)F_1^n(q^2),F_2^V(q^2)=F_2^p(q^2)F_2^n(q^2),$$
$$F_A^V(q^2)=F_A(q^2)$$
where
$`F_1^{p,n}(q^2)`$ $`=`$ $`\left[1{\displaystyle \frac{q^2}{4M^2}}\right]^1\left[G_E^{p,n}(q^2){\displaystyle \frac{q^2}{4M^2}}G_M^{p,n}(q^2)\right]`$
$`F_2^{p,n}(q^2)`$ $`=`$ $`\left[1{\displaystyle \frac{q^2}{4M^2}}\right]^1\left[G_M^{p,n}(q^2)G_E^{p,n}(q^2)\right]`$
$$G_E^p(q^2)=\frac{1}{(1\frac{q^2}{M_v^2})^2}$$
$$G_M^p(q^2)=(1+\mu _p)G_E^p(q^2),G_M^n(q^2)=\mu _nG_E^p(q^2);$$
$$G_E^n(q^2)=(\frac{q^2}{4M^2})\mu _nG_E^p(q^2)\xi _n;\xi _n=\frac{1}{1\lambda _n\frac{q^2}{4M^2}}$$
$$\mu _p=1.792847,\mu _n=1.913043,M_v=0.84GeV,\text{and}\lambda _n=5.6.$$
The isovector axial vector form factor $`F_A(Q^2)`$ is given by
$$F_A(Q^2)=\frac{F_A(0)}{(1\frac{q^2}{M_A^2})^2}$$
where $`M_A=1.05GeV`$; $`F_A(0)`$=-1.26
The double differential cross section $`\sigma _0(E_e,|\stackrel{}{k}^{}|)`$ for the basic reaction described in Eqn.(3) is then written as
$$\sigma _0(E_e,|\stackrel{}{k}^{}|)=G_{F}^{}{}_{}{}^{2}\mathrm{cos}^2\theta _c\frac{|\stackrel{}{k}^{}|^2}{8\pi E_{\nu _e}E_e}\frac{M_nM_p}{E_nE_p}\overline{\mathrm{\Sigma }}\mathrm{\Sigma }|T|^2\delta [q_0+E_nE_p]$$
(7)
where $`\overline{\mathrm{\Sigma }}\mathrm{\Sigma }|T|^2`$ is the square modulus of the transition amplitude given by
$`\overline{\mathrm{\Sigma }}\mathrm{\Sigma }|T|^2`$ $`=`$ $`L_{\mu \nu }J^{\mu \nu }with`$
$`L_{\mu \nu }`$ $`=`$ $`8[k_\mu k_{}^{}{}_{\nu }{}^{}+k_\nu k_{}^{}{}_{\mu }{}^{}k.k^{}g_{\mu \nu }+iฯต_{\mu \nu \lambda \sigma }k^\lambda k_{}^{}{}_{}{}^{\sigma }]and`$
$`J^{\mu \nu }`$ $`=`$ $`{\displaystyle J_{}^{\mu }{}_{}{}^{}J^\nu }`$ (8)
$`q_0(q_0=E_{\nu _e}E_e)`$ is the energy transferred to the nucleon.
In a nucleus, the neutrino scatters from a neutron moving in the finite nucleus of neutron density $`\rho _n(r)`$, with a local occupation number $`n_n(๐ฉ,๐ซ)`$. In the local density approximation the scattering cross section is written as
$$\sigma (E_e,|\stackrel{}{k}^{}|)=2๐๐ซ๐๐ฉ\frac{1}{(2\pi )^3}n_n(๐ฉ,๐ซ)\sigma _0(E_e,k^{})$$
(9)
where $`\sigma _0(E_e,|\stackrel{}{k}^{}|)`$ is given by Eqn.(7). The neutron energy $`E_n`$ and proton energy $`E_p`$ are replaced by $`E_n(|\stackrel{}{p}|)`$ and $`E_p(|\stackrel{}{p}+\stackrel{}{q}|)`$ where $`๐ฉ`$ is now the momentum of the target neutron inside the nucleus. In the nucleus the neutrons and protons are not free and their momenta are constrained to satisfy the Pauli principle, i.e., $`p_n<p_{F_n}`$ and $`p_{}^{}{}_{p}{}^{}(=|๐ฉ_n+๐ช|)>p_{F_p}`$, where $`p_{F_n}`$ and $`p_{F_p}`$ are the local Fermi momenta of neutrons and protons at the interaction point in the nucleus and are given by $`p_{F_n}=\left[3\pi ^2\rho _n(r)\right]^{\frac{1}{3}}`$ and $`p_{F_p}=\left[3\pi ^2\rho _p(r)\right]^{\frac{1}{3}}`$, $`\rho _n(r)`$ and $`\rho _p(r)`$ are the neutron and proton nuclear densities. Furthermore, in nuclei the threshold value of the reaction i.e. the Q-value of the reaction has to be taken into account.
To incorporate these modifications, the $`\delta `$ function in Eqn.(7) i.e. $`\delta [q_0+E_nE_p]`$ is modified to $`\delta [q_0+E_n(\stackrel{}{p})E_p(\stackrel{}{p}+\stackrel{}{q})Q]`$ and the factor
$$\frac{d๐ฉ}{(2\pi )^3}n_n(๐ฉ,๐ซ)\frac{M_nM_p}{E_nE_p}\delta [q_0+E_nE_p]$$
(10)
occurring in Eqn.(9) is replaced by $`(1/\pi )`$Im$`U_N(q_0,\stackrel{}{q})`$, where $`U_N(q_0,\stackrel{}{q})`$ is the Lindhard function corresponding to the particle hole(ph) excitation shown in Fig.(2) and is given by
$$U_N(q_0,\stackrel{}{q})=\frac{d๐ฉ}{(2\pi )^3}\frac{M_nM_p}{E_nE_p}\frac{n_n(p)\left[1n_p(\stackrel{}{p}+\stackrel{}{q})\right]}{q_0+E_n(p)E_p(\stackrel{}{p}+\stackrel{}{q})+iฯต}$$
(11)
where $`q_0`$=$`E_{\nu _e}E_eQ`$.
The imaginary part of the Lindhard function is then derived to be:
$$ImU_N(q_0,\stackrel{}{q})=\frac{1}{2\pi }\frac{M_pM_n}{|\stackrel{}{q}|}\left[E_{F_1}A\right]\text{with}$$
(12)
$`q^2<0`$ ,$`E_{F_2}q_0<E_{F_1}`$ and $`\frac{q_0+|\stackrel{}{q}|\sqrt{1\frac{4M^2}{q^2}}}{2}<E_{F_1}`$, where $`E_{F_1}=\sqrt{p{}_{F_n}{}^{}{}_{}{}^{2}+M_{n}^{}{}_{}{}^{2}}`$, $`E_{F_2}=\sqrt{p_{F_p}^{}{}_{}{}^{2}+M_{p}^{}{}_{}{}^{2}}`$ and
A = $`Max[M_n,E_{F_2}q_0,\frac{q_0+|\stackrel{}{q}|\sqrt{1\frac{4M^2}{q^2}}}{2}]`$.
The threshold value, Q, for the neutrino reaction is generally taken to be the Q value corresponding to the lowest allowed Fermi or Gamow Teller transitions. However, in some cases the Q value corresponding to the ground state to ground state(gs-gs) transition is also takenhaxton .
With inclusion of these nuclear effects the cross section $`\sigma (E_\nu )`$ is written as
$`\sigma (E_\nu )={\displaystyle \frac{2G_{F}^{}{}_{}{}^{2}\mathrm{cos}^2\theta _c}{\pi }}{\displaystyle _{r_{min}}^{r_{max}}}r^2๐r{\displaystyle _{p_{e}^{}{}_{}{}^{min}}^{p_{e}^{}{}_{}{}^{max}}}p_{e}^{}{}_{}{}^{2}๐p_e{\displaystyle _1^1}d(cos\theta )`$
$`\times {\displaystyle \frac{1}{E_{\nu _e}E_e}}L_{\mu \nu }J^{\mu \nu }ImU_N[E_{\nu _e}E_eQ,\stackrel{}{q}].`$ (13)
In the nucleus the strength of the electroweak coupling may change from their free nucleon values due to the presence of strongly interacting nucleons. Conservation of Vector Current (CVC) forbids any change in the charge coupling while magnetic and axial vector couplings are likely to change from their free nucleon values. These changes are calculated by considering the interaction of ph excitations in the nuclear medium in Random Phase Approximation (RPA) as shown in Fig.3. The diagram shown in Fig.3 simulates the effects of the strongly interacting nuclear medium at the weak vertex. The ph-ph interaction is shown by the wavy line in Fig.3 and is described by the $`\pi `$ and $`\rho `$ exchanges modulated by the effect of short range correlations.
The weak nucleon current described by Eq.(5) gives, in nonrelativistic limit, terms like $`F_A\stackrel{}{\sigma }\tau _+`$ and $`iF_2\frac{\stackrel{}{\sigma }\times \stackrel{}{q}}{2M}\tau _+`$ which generate spin-isospin transitions in nuclei. While the term $`iF_2\frac{\stackrel{}{\sigma }\times \stackrel{}{q}}{2M}\tau _+`$ couples to the transverse excitations, the term $`F_A\stackrel{}{\sigma }\tau _+`$ couples to the transverse as well as longitudinal channels. These channels produce different RPA responses in the longitudinal and transverse channels when the diagrams of Fig.3 are summed over. This is illustrated by considering a term like $`F_A\sigma ^i`$ in Eq.(5). One of the contributions of this term to the hadronic tensor $`J^{ij}`$ in the medium is proportional to $`F_A^2\delta _{ij}ImU_N`$ which is now split between the longitudinal and transverse components as
$$F_A^2\delta _{ij}ImU_NF_A^2\left[\widehat{๐ช_๐ข}\widehat{๐ช_๐ฃ}+(\delta _{ij}\widehat{๐ช_๐ข}\widehat{๐ช_๐ฃ})\right]ImU_N$$
(14)
The RPA response of this term after summing the higher order diagrams like Fig.3 is modified and is given by $`J_{RPA}^{ij}`$:
$$J^{ij}J_{RPA}^{ij}=F_A^2ImU_N\left[\frac{\widehat{๐ช_๐ข}\widehat{๐ช_๐ฃ}}{1U_NV_l}+\frac{\delta _{ij}\widehat{๐ช_๐ข}\widehat{๐ช_๐ฃ}}{1U_NV_t}\right]$$
(15)
where $`V_l`$ and $`V_t`$ are the longitudinal and transverse part of the nucleon-nucleon potential calculated with $`\pi `$ and $`\rho `$ exchanges and are given by
$`V_l(q)={\displaystyle \frac{f^2}{m_\pi ^2}}\left[{\displaystyle \frac{q^2}{q^2+m_\pi ^2}}\left({\displaystyle \frac{\mathrm{\Lambda }_\pi ^2m_\pi ^2}{\mathrm{\Lambda }_\pi ^2q^2}}\right)^2+g^{}\right],`$
$`V_t(q)={\displaystyle \frac{f^2}{m_\pi ^2}}\left[{\displaystyle \frac{q^2}{q^2+m_\rho ^2}}C_\rho \left({\displaystyle \frac{\mathrm{\Lambda }_{\rho }^{}{}_{}{}^{2}m_\rho ^2}{\mathrm{\Lambda }_{\rho }^{}{}_{}{}^{2}q^2}}\right)^2+g^{}\right]`$ (16)
$`\mathrm{\Lambda }_\pi =1.3GeV`$, $`C_\rho =2`$, $`\mathrm{\Lambda }_\rho =2.5GeV`$, $`m_\pi `$ and $`m_\rho `$ are the pion and $`\rho `$ masses, and $`g^{}`$ is the Landau-Migdal parameter taken to be $`0.7`$ which has been used quite successfully to explain many electromagnetic and weak processes in nucleimukh ,gil .
This modified tensor $`J_{RPA}^{ij}`$ when contracted with the leptonic tensor $`L_{ij}`$ gives the contribution of the $`F_A^2`$ term to the RPA response. The contribution of the time component of the hadronic tensors like $`J^{0i}`$ and $`J^{00}`$ are of higher order in $`\left(\frac{q}{M}\right)^2`$ and are not important at low energies considered in this paper.
The effect of the $`\mathrm{\Delta }`$ degrees of freedom in the nuclear medium is included in the calculation of the RPA response by considering the effect of ph-$`\mathrm{\Delta }`$h and $`\mathrm{\Delta }`$h-$`\mathrm{\Delta }`$h excitations as shown in Fig.3(b). This is done by replacing $`U_N`$ by $`U_N=U_N+U_\mathrm{\Delta }`$, where $`U_\mathrm{\Delta }`$ is the Lindhard function for $`\mathrm{\Delta }`$h excitation in the medium and the expressions for $`U_N`$ and $`U_\mathrm{\Delta }`$ are taken fromoset1 . The different couplings of $`N`$ and $`\mathrm{\Delta }`$ are incorporated in $`U_N`$ and $`U_\mathrm{\Delta }`$ and then the same interaction strengths $`V_l`$ and $`V_t`$ are used to calculate the RPA response. This is discussed in some detail in Refsingh and more recently in Ref.nieves by Nieves et al.
One of the important aspects of charge current neutrino interactions is the treatment of Coulomb distortion of the produced lepton in the Coulomb field of the final nucleus. At low energies of the electron relevant to $`\beta `$ decays in nuclei the Coulomb distortion of electron in the nuclear field is taken into account by multiplying the momentum distribution of the electron in Eqn.(13) by a Fermi function $`F(Z,E_e)`$, where $`F(Z,E_e)`$ is given bybehrens :
$$F(Z,E_e)=\left[1\frac{2}{3}(1\gamma _0)\right]^1f(Z,E_e),$$
where
$$f(Z,E_e)=2(1+\gamma _0))(2p_eR)^{2(1\gamma _0)}\frac{|\mathrm{\Gamma }(\gamma _0+i\eta )|^2}{(\mathrm{\Gamma }(2\gamma _0+1))^2}.$$
Here R is the nuclear radius and $`\gamma _0=\sqrt{1(\alpha Z)^2}`$, $`\eta =\frac{\alpha Zc}{v}`$. This approximation works quite well at low energies, but it is not appropriate at higher energies, specially for high Z nucleikolbe1 ,engel . Therefore, at higher electron energies a different approach is needed to describe the Coulomb distortion effect of the electron. For this purpose, we apply the methods of electron scattering where various approximations have been used to take into account the Coulomb distortion effects of the initial and final electrongco -guisti . One of them is the Modified Effective Momentum Approximation(MEMA) in which the electron momentum and energy are modified by taking into account the Coulomb energy. We have used this approach in the case of charged current quasielastic neutrino scattering and the energy and momentum of the electron present in the final state is modified in the Coulomb field of the final nucleusmukh ,argon -athar . In the local density approximation, the effective energy of the electron in the Coulomb field of the final nucleus is given by:
$$E_{eff}=E_e+V_c(r),$$
where
$$V_c(r)=Z_f\alpha 4\pi (\frac{1}{r}_0^r\frac{\rho _p(r^{})}{Z_f}r_{}^{}{}_{}{}^{2}๐r^{}+_r^{\mathrm{}}\frac{\rho _p(r^{})}{Z_f}r^{}๐r^{})$$
(17)
Thus, in presence of nuclear medium effects the total cross section $`\sigma (E_\nu )`$, with the inclusion of Coulomb distortion effects taken into account by Fermi function(MEMA), is written as
$`\sigma ^{FF(MEMA)}(E_\nu )={\displaystyle \frac{2G_{F}^{}{}_{}{}^{2}\mathrm{cos}^2\theta _c}{\pi }}{\displaystyle _{r_{min}}^{r_{max}}}r^2๐r{\displaystyle _{p_{e}^{}{}_{}{}^{min}}^{p_{e}^{}{}_{}{}^{max}}}p_{e}^{}{}_{}{}^{2}๐p_e{\displaystyle _1^1}d(cos\theta )`$
$`\times {\displaystyle \frac{1}{E_{\nu _e}E_e}}L_{\mu \nu }J_{RPA}^{\mu \nu }ImU_{N}^{}{}_{}{}^{FF(MEMA)}.`$ (18)
where
$`ImU_N^{FF}`$ $`=`$ $`F(Z,E_e)ImU_N[E_{\nu _e}E_eQ,\stackrel{}{q}]and`$
$`ImU_N^{MEMA}`$ $`=`$ $`ImU_N[E_{\nu _e}E_eQV_c(r),\stackrel{}{q}]`$ (19)
## III Results
In this section we present the results for the total cross section $`\sigma (E)`$ as a function of energy and the flux averaged cross section $`<\sigma >`$ for various nuclei which have been presently proposed to be studied at SNS facilities using neutrinos from stopped muon decaysavignone . For the numerical calculation of the cross section, we have classified the nuclei in three groups according to their nuclear densities used in this calculation and have been presented in Tables I-III.
In Table-I, we present the nuclear density parameters for $`{}_{}{}^{12}C`$, $`{}_{}{}^{14}N`$ and $`{}_{}{}^{16}O`$ nuclei using a 2-parameter harmonic oscillator(H.O.) density given by
$$\rho (r)=\rho _0(1+\alpha (\frac{r}{a})^2)exp((\frac{r}{a})^2),$$
(20)
In Table-II, we present the nuclear density parameters for $`{}_{}{}^{19}F`$, $`{}_{}{}^{23}Na`$, $`{}_{}{}^{27}Al`$, $`{}_{}{}^{28}Si`$, $`{}_{}{}^{31}P`$, $`{}_{}{}^{37}Cl`$, $`{}_{}{}^{40}Ar`$, $`{}_{}{}^{51}V`$, $`{}_{}{}^{52}Cr`$, $`{}_{}{}^{55}Mn`$, $`{}_{}{}^{56}Fe`$, $`{}_{}{}^{59}Co`$, $`{}_{}{}^{71}Ga`$, $`{}_{}{}^{89}Y`$, $`{}_{}{}^{93}Nb`$, $`{}_{}{}^{98}Mo`$, $`{}_{}{}^{115}In`$, $`{}_{}{}^{127}I`$, $`{}_{}{}^{139}La`$, $`{}_{}{}^{181}Ta`$, $`{}_{}{}^{208}Pb`$ and $`{}_{}{}^{209}Bi`$ nuclei using a 2-parameter Fermi density (2pF) given by
$$\rho (r)=\frac{\rho _0}{(1+exp((r\alpha )/a))}$$
(21)
In Table-III, we present the nuclear density parameters for $`{}_{}{}^{32}S`$, $`{}_{}{}^{39}K`$ and $`{}_{}{}^{40}Ca`$ nuclei using a three parameter Fermi(3pF) density given by
$$\rho (r)=\frac{\rho _0(1+w\frac{r^2}{\alpha ^2})}{(1+exp((r\alpha )/a))},$$
(22)
The parameters have been taken from de Vries et al.vries except for $`{}_{}{}^{115}In`$ and $`{}_{}{}^{127}I`$ which have been taken from Ref.kosmas . The Q values presented in these tables correspond to the lowest allowed Fermi or Gamow-Teller transitions for the above mentioned nuclei except for the case of $`{}_{}{}^{40}Ca`$ and $`{}_{}{}^{98}Mo`$ for which the Q value corresponding to the ground state to ground state transitions have been takenhaxton ,table .
### III.1 Nuclear Medium Effects
When the reaction $`\nu _e+ne^{}+p`$ takes place in the nucleus, the first consideration is the Q value which inhibits the reaction in the nucleus. This inhibition is quite substantial in the low energy region considered here for the nuclei like $`{}_{}{}^{12}C`$, $`{}_{}{}^{16}O`$, $`{}_{}{}^{18}Si`$, $`{}_{}{}^{32}S`$ and $`{}_{}{}^{40}Ca`$ for which the Q values are rather large (Q$``$13-18MeV). In addition to this, the effect of Pauli blocking which is taken into account through the imaginary part of the Lindhard function is to further reduce the cross section. Finally the renormalisation of weak coupling constants which is generated in our model through RPA correlations and is taken into account by calculating the cross section with the modified hadronic tensor $`J_{RPA}^{\mu \nu }`$ also reduces the cross sections. In Fig.4, we have shown the reduction due to these effects separately for some representative nuclei like $`{}_{}{}^{12}C`$, $`{}_{}{}^{56}Fe`$, $`{}_{}{}^{127}I`$ and $`{}_{}{}^{208}Pb`$ in various mass ranges.
We see that at low energies the major suppression in the cross section comes due to the consideration of Q-values and Pauli blocking in the nuclear medium. The reduction in the cross section $`\sigma (E)`$ due to these effects decreases with the increase of energy. For example at $`E_\nu =50MeV`$, this suppression is $``$ $`93\%`$ for $`{}_{}{}^{12}C`$ and $``$ $`7577\%`$ for other nuclei like $`{}_{}{}^{56}Fe`$, $`{}_{}{}^{127}I`$ and $`{}_{}{}^{208}Pb`$ respectively (compare the solid lines with the dashed line in Fig.4). This suppression reduces to $`4045\%`$ in all these nuclei at $`E_\nu =200MeV`$(not shown in Fig.4).
In addition to the Pauli blocking, the consideration of RPA correlation in the nuclear medium gives rise to further reduction which increases with the mass number and decreases with the increase in energy (compare the dashed line with the dotted line in Fig.4). For example at $`E_\nu =50MeV`$ the RPA correlations give a further reduction of $`50\%`$ for $`{}_{}{}^{12}C`$, $`60\%`$ for $`{}_{}{}^{56}Fe`$ and around $`70\%`$ for $`{}_{}{}^{127}I`$ and $`{}_{}{}^{208}Pb`$. As the energy increases it becomes smaller and at $`E_\nu =200MeV`$ the reduction is $`35\%`$ for $`{}_{}{}^{12}C`$, $`40\%`$ for $`{}_{}{}^{56}Fe`$ and around $`50\%`$ for $`{}_{}{}^{127}I`$ and $`{}_{}{}^{208}Pb`$(not shown here). It should be noted that $`4060\%`$ reduction due to the medium polarisation effects calculated through the RPA correlations in our model is similar to using $`\frac{g_{eff}}{g_A}=0.7`$ in some shell model calculationskolbe2 -suzuki .
### III.2 Effects of Coulomb distortion
The effect of Coulomb distortion is calculated using Fermi function $`F(Z,E_e)`$ as well as with the modified effective momentum approximation(MEMA). The results for some representative nuclei like $`{}_{}{}^{12}C`$, $`{}_{}{}^{56}Fe`$, $`{}_{}{}^{127}I`$ and $`{}_{}{}^{208}Pb`$ in various mass range are shown in Fig.5. The general effect of the Coulomb distortion of the electron is to increase the cross section which depends upon the incident energy and the charge of the final nucleus.
For a fixed Z, this increase in the cross section decreases with the increase in energy while for a fixed energy the inclusion of Coulomb distortion increases with the charge Z. For example for $`{}_{}{}^{12}C`$ this is $`15\%`$ at $`E_\nu =50MeV`$ which becomes $`10\%`$ at $`E_\nu =200MeV`$. For high Z nuclei the Coulomb effect is very large and results in manifold increase in the cross sections. This can be seen by comparing the cross section without Coulomb effect shown by dotted lines and the cross sections with Coulomb effects using the Fermi function $`F(Z,E_e)`$ shown by dashed lines. For example in the case of $`{}_{}{}^{56}Fe`$ nucleus the increase due to Coulomb distortion is $`83\%`$ at $`E_\nu =50MeV`$ which becomes $`75\%`$ at $`E_\nu =200MeV`$. However, as discussed in Section-II, the use of Fermi function to calculate the Coulomb distortion effects overestimates the cross sections and is not appropriate at higher electron energies. Therefore, we use the modified effective momentum approximation(MEMA) and present the results for $`\sigma (E)`$ in Fig.5 with solid lines. It is seen that for low Z nuclei, like $`{}_{}{}^{12}C`$. the results for the cross sections in the two approximations are qualititative similar but the MEMA gives slightly higher cross sections for $`E_\nu <60MeV`$. As the Z increases the cross sections with MEMA is higher than Fermi function at lower energies and becomes smaller than the cross sections obtained with the Fermi function as the energy becomes large. This energy dependence of the cross section is explicitly shown for higher Z nuclei like $`{}_{}{}^{56}Fe`$, $`{}_{}{}^{127}I`$ and $`{}_{}{}^{208}Pb`$ in Fig.5.
We show the results of the energy dependence of the cross sections for all nuclei listed in tables I-III, in Figs.6-9, where dashed lines show the cross sections without RPA correlations and dotted lines show the cross sections with RPA correlations using Fermi function for the Coulomb distortion. The solid lines show the cross section with RPA correlations where the Coulomb distortion effects are calculated with MEMA. In these figures a comparison of dashed lines and dotted lines shows the effect of RPA correlations while a comparison of dotted lines and solid lines shows the effect of Coulomb distortion calculated using Fermi function and MEMA.
### III.3 Flux averaged cross sections
We calculate the flux averaged cross section $`<\sigma >`$ defined as
$$<\sigma >=\varphi (E_\nu )\sigma (E_\nu )๐E_\nu $$
(23)
where $`\varphi _(E_\nu )`$ is given by Eqn.1. The results for $`<\sigma >`$ are presented in Table-IV, where we show by $`<\sigma >_{NC}^{RPA}`$ the flux averaged cross sections with nuclear medium effects without any Coulomb distortion effects. When Coulomb distortion effects are taken into account the cross sections without RPA correlations are shown by $`<\sigma >_C^N`$ and the results with RPA correlations are shown by $`<\sigma >_C^{RPA}`$. We evaluate $`<\sigma >_C^N`$ and $`<\sigma >_C^{RPA}`$ in a hybrid model where at lower energies $`\sigma (E_\nu )`$ calculated with the Fermi function and at higher energies $`\sigma (E_\nu )`$ calculated with MEMA is used to perform the flux averaging in Eqn.23. For low mass nuclei like $`{}_{}{}^{12}C`$, $`{}_{}{}^{16}O`$, etc. the flux averaged cross section has been evaluated with the cross section $`\sigma (E)`$ calculated with the Fermi function. Thus, in hybrid model, it is the lower value of the cross section which is used for calculating $`<\sigma >`$. We see from this table that the effect of the Coulomb distortion is to increase the cross section and this increase is quite large for high $`Z`$ nuclei like $`{}_{}{}^{56}Fe`$, $`{}_{}{}^{208}Pb`$, etc. In case of $`{}_{}{}^{12}C`$, it is small but plays an important role in explaining the experimental result(Compare column 1 and column 3 in table IV). A comparison of column 2 and column 3 in this table shows the strong reduction due to RPA correlations which increases with mass number.
In Table-V, we compare our results with the results of some other calculations. In this energy region of the neutrinos, there are many theoretical calculations done for the inclusive neutrino reactions in $`{}_{}{}^{12}C`$mukh -nieves , kolbe -carbon , while there are few calculations for $`{}_{}{}^{16}O`$auer , $`{}_{}{}^{56}Fe`$kolbe2 ,mintz1 and $`{}_{}{}^{208}Pb`$kolbe2 ,suzuki -volpe1 and some other nucleikosmas . Some calculations are similar to the calculations presented in this paperkosmas -nieves while others make use of Shell Modelvolpe -suzuki , random phase approximation(RPA) with pairing correlationskolbe2 -volpe ,kolbe -volpe1 and elementary particle approachmintz -mintz1 . We see that for $`{}_{}{}^{12}C`$ and $`{}_{}{}^{56}Fe`$ our results are in fair agreement with the experimental results and other theoretical calculations. For $`{}_{}{}^{208}Pb`$ nucleus our results for $`<\sigma >`$ is comparatively smaller than the results of Refs.kolbe2 , suzuki and volpe1 . This is mainly due to the different approaches of taking into account the nuclear effects. However, among the different calculations of the inclusive cross section $`<\sigma >`$ in $`{}_{}{}^{208}Pb`$, the results do not agree among themselveskolbe2 ,suzuki -volpe1 . Therefore, more work is needed for calculating the cross section in $`{}_{}{}^{208}Pb`$ at low energies.
## IV Summary and Conclusions
We have studied the charged current $`\nu _e`$ reactions on various nuclei which are of present interest. The cross section calculations are performed in a local density approximation taking into account the Pauli blocking, Fermi motion and the renormalization of weak transition strengths in the nuclear medium. The effect of Coulomb distortion for the charged lepton while coming out of the nucleus is taken into account by using the Fermi function as well as the modified momentum approximation(MEMA). The cross sections are then averaged over the $`\nu _e`$ spectra obtained from the muons decay at rest where the maximum energy of neutrinos is 52.8MeV.
We find that
1. The role of nuclear effects like Q value, Pauli blocking and Fermi motion is to reduce the cross sections. For a given Z, this reduction becomes smaller with the increase in energy. For example, at $`E_\nu =50MeV`$, this suppression is $``$ $`93\%`$ for $`{}_{}{}^{12}C`$ and $``$ $`7577\%`$ for other nuclei like $`{}_{}{}^{56}Fe`$, $`{}_{}{}^{127}I`$, etc. This suppression reduces to $`4045\%`$ in all these nuclei at $`E_\nu =200MeV`$.
2. There is a further reduction of the cross section due to the renormalization of weak transition strengths in the nuclear medium. For a given Z, this reduction becomes smaller with the increase in neutrino energy, while for a given neutrino energy $`E_\nu `$, this reduction increases with Z. For example at $`E_\nu =50MeV`$ the RPA correlations give a further reduction of $`50\%`$ for $`{}_{}{}^{12}C`$, $`60\%`$ for $`{}_{}{}^{56}Fe`$ and around $`70\%`$ for $`{}_{}{}^{127}I`$ and $`{}_{}{}^{208}Pb`$ and at $`E_\nu =200MeV`$ this reduction is $`35\%`$ for $`{}_{}{}^{12}C`$, $`40\%`$ for $`{}_{}{}^{56}Fe`$ and around $`50\%`$ for $`{}_{}{}^{127}I`$ and $`{}_{}{}^{208}Pb`$.
3. The two methods of treating the Coulomb distortion give similar results for low energy neutrinos in the case of low mass nuclei. For intermediate and heavy mass nuclei the cross sections with Fermi function are smaller than the cross sections with MEMA upto certain energy $`E_{\nu _e}`$ after which the the cross sections calculated with Fermi function become larger. At the energy $`E_{\nu _e}`$ where this cross over takes place changes with nuclei. For example it is around 40MeV for nuclei like $`{}_{}{}^{56}Fe`$ in the intermediate mass range and around 18MeV for nuclei in the heavier mass range like $`{}_{}{}^{208}Pb`$.
4. The total cross sections averaged over the neutrino spectrum obtained from the muons decaying at rest is presented for all nuclei considered here. The results for $`{}_{}{}^{12}C`$, $`{}_{}{}^{16}O`$, $`{}_{}{}^{56}Fe`$ and $`{}_{}{}^{208}Pb`$ nuclei are compared with the available experimental results as well as different theoretical calculations. New results have been presented for many other nuclei.
The numerical results presented in this paper can be a very useful benchmark for neutrino nucleus cross section measurements being proposed at SNS facilities using various nuclei as nuclear targets.
###### Acknowledgements.
This work was financially supported by the Department of Science and Technology, Govt. of India under grant number DST SP/S2/K-07/2000. One of the authors(SA) would like to thank CSIR for the financial support.
|
warning/0506/hep-th0506111.html
|
ar5iv
|
text
|
# 1 The shape of the potential Vฬโข(ฯ) for different values of a.
On spontaneous symmetry breaking in hot QCD<sup>1</sup><sup>1</sup>1Talk at the conference โFizika-2005โ dedicated to the 60th anniversary of National Academy of Sciences of Azerbaijan; Institute of Physics, Baku, 7-9 June 2005.
Fuad M. Saradzhev<sup>2</sup><sup>2</sup>2e-mail: fuad\_saradzhev@hotmail.com
Institute of Physics, National Academy of Sciences of Azerbaijan,
H.Javid pr. 33, AZ-1143 Baku, Azerbaijan
Abstract
We prove that nontrivial vacuum states which can arise in hot QCD are associated with the tachyonic regime of hadronic matter fluctuations. This allows us to improve the condition for such states to appear.
1. It is known that at phase transitions from hadronic to quark and gluon degrees of freedom nontrivial local vacuum states can appear in the hadronic phase . These states are metastable and of particular interest since they have experimental signatures such as an enhanced production of $`\eta `$ and $`\eta ^{}`$ mesons . They can decay via CP violating processes such as $`\eta \pi ^0\pi ^0`$ and because of global parity odd asymmetries for charged pions. The decay rate of CP-odd metastable states was estimated in .
In we used the mean-field approximation to develop the kinetic approach to the decay of the CP-odd phase in hot QCD and to derive a non-Markovian kinetic equation describing the production of $`\eta ^{}`$-mesons. A different kinetic equation was derived for the production of tachyonic modes .
In the present Talk, we aim to show that in addition to these metastable states nontrivial vacua can appear according to the standard spontaneous symmetry breaking picture provided the hadronic matter fluctuations enter a tachyonic regime.
2. We start from the singlet Witten-DiVecchia-Veneziano effective Lagrangian density
$$=\frac{1}{2}\left(_\mu \eta \right)\left(^\mu \eta \right)+f^2\mu ^2\mathrm{cos}\left(\frac{\eta }{f}\right)\frac{a_0}{2}\eta ^2,$$
(1)
where $`f=\sqrt{\frac{3}{2}}f_\pi `$ and $`f_\pi =92MeV`$ is the semileptonic pion decay constant; $`\mu ^2=\frac{1}{3}(m_\pi ^2+2m_K^2)`$ is a parameter depending on $`\pi `$\- and $`K`$-meson masses. The parameter $`a_0`$ represents the topological susceptibility. For zero temperature $`T=0`$, $`a_0=m_\eta ^2+m_\eta ^{}^22m_K^20.726GeV^2`$, $`\mu ^20.171GeV^2`$ and $`f_\pi 93MeV`$. In response to non-zero temperature mesons change their effective masses, $`\mu `$ and $`a_0`$ becoming functions of $`T`$. The model is defined in a finite volume: $`L/2x_iL/2`$, $`i=1,2,3`$. The continuum limit is $`\frac{1}{V}_\stackrel{}{k}\frac{d^3\stackrel{}{k}}{(2\pi )^3}`$.
The meson field $`\eta (\stackrel{}{x},t)`$ obeys the Klein-Gordon type equation
$$\left(\mathrm{}+m_0^2\right)\eta =J_s,$$
(2)
where $`m_0^2a_0+\mu ^2`$ and the current
$$J_sf\mu ^2\left[\mathrm{sin}\left(\frac{\eta }{f}\right)\left(\frac{\eta }{f}\right)\right]$$
(3)
is non-linear in $`\eta `$, i.e. contains orders $`\eta ^3`$ and higher and is therefore completely determined by the self-interaction of the field $`\eta `$.
Following the mean-field approximation we decompose $`\eta (\stackrel{}{x},t)`$ into its space-homogeneous vacuum mean value $`\varphi (t)=\eta (\stackrel{}{x},t)`$ and fluctuations $`\chi `$
$$\eta (\stackrel{}{x},t)=\varphi (t)+\chi (\stackrel{}{x},t),$$
(4)
with $`\chi (\stackrel{}{x},t)=0`$. The vacuum mean field is treated as a classical, self-interacting background field. It is defined with respect to the in-vacuum $`|0`$ as
$$\varphi (t)\eta (\stackrel{}{x},t)\frac{1}{L^3}d^3x0|\eta (\stackrel{}{x},t)|0,$$
(5)
so in the limit $`t\mathrm{}`$ $`\varphi (t)0`$, while quantum fluctuations take place at all times.
Substituting Eq.(4) into Eq.(3) yields the following decomposition for the current
$$J_s=J_s^{(1)}+\overline{J}_s,$$
(6)
where
$$J_s^{(1)}J_s^{(0)}+\mu ^2\left[1\mathrm{cos}\left(\frac{\varphi }{f}\right)\right]\chi $$
(7)
is the current in the first order in $`\chi `$ with the background field-fluctuations interaction term added, while the zero order of the current
$$J_s^{(0)}f\mu ^2\left[\mathrm{sin}\left(\frac{\varphi }{f}\right)\left(\frac{\varphi }{f}\right)\right]$$
(8)
represents only the self-interaction of the background field. The second current in the right-hand side of Eq.(6) includes terms of second and higher orders in $`\chi `$
$$\overline{J}_s=f\mu ^2\mathrm{sin}\left(\frac{\varphi }{f}\right)\left[\mathrm{cos}\left(\frac{\chi }{f}\right)1\right]f\mu ^2\mathrm{cos}\left(\frac{\varphi }{f}\right)\left[\mathrm{sin}\left(\frac{\chi }{f}\right)\left(\frac{\chi }{f}\right)\right].$$
(9)
Substituting Eq.(4) also into Eq.(2) and taking the mean value $`\mathrm{}`$ yields the vacuum mean field equation
$$\ddot{\varphi }+a_0\varphi +f\mu ^2\mathrm{sin}\left(\frac{\varphi }{f}\right)=\overline{J}_s.$$
(10)
Eq.(10) is a generalization of the vacuum mean field equation used in for non-vanishing values of $`\overline{J}_s`$ (see also ). In the Hartree-type approximation,
$$\overline{J}_s=f\mu ^2\mathrm{sin}\left(\frac{\varphi }{f}\right)\mathrm{cos}\left(\frac{\chi }{f}\right)1.$$
(11)
The equation of motion for the quantum fluctuations reads
$$\left(\mathrm{}+m_{eff}^2\right)\chi =\overline{J}_s\overline{J}_s$$
(12)
with
$$m_{eff}^2a_0+\mu ^2\mathrm{cos}\left(\frac{\varphi }{f}\right).$$
(13)
For $`a(a_0/\mu ^2)<1`$, $`m_{eff}^2`$ can be negative for some values of the background field indicating a tachyonic regime. Eqs.(10) and (12) are self-consistently coupled and include back-reactions. The vacuum mean field modifies the equation for fluctuations via a time-dependent frequency, while the fluctuations themselves react back on the vacuum mean field via the source term $`\overline{J}_s`$.
3. With the decomposition (4), we deduce from (1) the effective Lagrangian density governing the dynamics of fluctuations
$$_\chi =\frac{1}{2}\left(_\mu \chi \right)\left(^\mu \chi \right)+f^2\mu ^2\mathrm{cos}\left(\frac{\varphi }{f}\right)\left[\mathrm{cos}\left(\frac{\chi }{f}\right)1\right]$$
$$f^2\mu ^2\mathrm{sin}\left(\frac{\varphi }{f}\right)\left[\mathrm{sin}\left(\frac{\chi }{f}\right)\left(\frac{\chi }{f}\right)\right]\frac{a_0}{2}\chi ^2\overline{J}_s\chi ,.$$
(14)
Expanding (14) in power series in $`\chi `$, yields in the second order
$$_\chi ^{(2)}=\frac{1}{2}\left(_\mu \chi \right)\left(^\mu \chi \right)\frac{1}{2}m_{eff}^2\chi ^2.$$
(15)
For $`a>1`$, the second order effective potential of fluctuations
$$V_\chi ^{(2)}=\frac{1}{2}m_{eff}^2\chi ^2$$
(16)
is $`\chi ^2`$-type potential with oscillating walls. During the time evolution of the background field, the potential (16) fluctuates around $`\frac{1}{2}a_0\chi ^2`$ in tune with the time dependence of $`\varphi `$. For $`a<1`$, for some values of the background field the potential (16) becomes upside down without any stable, particle states.
Let us consider now the exact form of the effective potential,
$$\overline{V}_\chi \frac{1}{f^2\mu ^2}V_\chi =\frac{a}{2}\left(\frac{\chi }{f}\right)^2+\frac{1}{f\mu ^2}\overline{J}_s\left(\frac{\chi }{f}\right)$$
$$\mathrm{cos}\left(\frac{\varphi }{f}\right)\left[\mathrm{cos}\left(\frac{\chi }{f}\right)1\right]+\mathrm{sin}\left(\frac{\varphi }{f}\right)\left[\mathrm{sin}\left(\frac{\chi }{f}\right)\left(\frac{\chi }{f}\right)\right].$$
(17)
It also changes during the time evolution of $`\varphi `$. First of all, the term $`\frac{1}{f\mu ^2}\overline{J}_s\left(\frac{\chi }{f}\right)`$ shifts the minimum of $`\chi ^2`$-potential from $`\chi =0`$ to $`\chi =\frac{\overline{J}_s}{a_0}`$,
$$\frac{a}{2}\left(\frac{\chi }{f}\right)^2+\frac{1}{f\mu ^2}\overline{J}_s\left(\frac{\chi }{f}\right)=\frac{a}{2f^2}\left(\chi +\frac{\overline{J}_s}{a_0}\right)^2+\mathrm{},$$
(18)
the coordinate of the minimum oscillating in tune with the background field.
In addition, in the tachyonic regime the effective potential exhibits the spontaneous symmetry breaking. Let us compare the form of (17) for two different values of the background field, $`\varphi =2\pi `$ and $`\varphi =\pi `$. For both values, $`\overline{J}_s=0`$ in the Hartree-type approximation. For $`\varphi =2\pi `$, the effective potential takes the form
$$\overline{V}_\chi =\frac{a}{2}\left(\frac{\chi }{f}\right)^2\mathrm{cos}\left(\frac{\chi }{f}\right)+1.$$
(19)
It is positive for all values of $`\chi `$ and its minimum is at $`\chi =0`$.
For $`\varphi =\pi `$, the effective potential becomes
$$\overline{V}_\chi =\frac{a}{2}\left(\frac{\chi }{f}\right)^2+\mathrm{cos}\left(\frac{\chi }{f}\right)1.$$
(20)
It is minimized for
$$a\left(\frac{\chi }{f}\right)=\mathrm{sin}\left(\frac{\chi }{f}\right).$$
(21)
For $`a1`$, Eq.(21) has only trivial solution $`\chi =0`$. However, for $`a<1`$ nontrivial solutions appear.
Fig.(1) shows the shape of the potential $`\overline{V}_\chi `$ for different values of $`a`$. The nontrivial local minima appear for $`a<1`$. For $`a>1`$, the spontaneous symmetry breaking does not occur. The special value $`a_{sp}=0.217`$ was found in . For $`a<a_{sp}`$, the number of nontrivial local minima is increasing with decreasing values of $`a`$.The nontrivial minima are of different energy; the ones of higher energy are metastable and can decay by a tunneling.
4. The tachyonic regime can be characterized as a regime of spontaneous symmetry breaking. Although $`a_{sp}=0.217`$ is specified in as a special value defining the first local minima, we have shown that nontrivial minima appear even for $`a_{sp}<a<1`$.
Whether the system evolves in the tachyonic or non-tachyonic regime is fixed by the value of the background field. During its time evolution, energy is transferred from $`\varphi `$ to $`\chi `$. As a result, $`\varphi `$ is damped, while the number of particles in quantum fluctuations increases. If, for example, $`\varphi (t=0)=2\pi `$, then the quantum fluctuations first evolve in the standard, non-tachyonic regime. As soon as $`\varphi (t)`$ reaches $`\pi `$, the tachyonic regime starts (for $`a<1`$), and the intensive production of tachyonic modes results in a rapid damping of $`\varphi `$.
|
warning/0506/nucl-th0506069.html
|
ar5iv
|
text
|
# Bounding the ground-state energy of a many-body system with the differential method
## 1 Introduction
There is little need to stress the great importance of the ground-state in many domains of quantum physics. Nevertheless, computing the lowest energy of most systems cannot be done analytically and approximations are required. Some techniques, being very general, have been well-known for several decades. For instance the variational methods (Rayleigh-Ritz) or the perturbative series (Rayleigh-Schrรถdinger) have their roots in the pre-quantum era; much later, numerical algorithms (numerical diagonalizations as well as Monte-Carlo computations), supported by the increasing power of computers, have been able to provide a tremendous precision on the ground-state of a large variety of very complex systems. However, it is a much more difficult task to rigorously estimate the discrepancies between the exact ground-state energy $`E_0`$ and the approximated one. In particular, though variational methods naturally provide upper bounds on $`E_0`$, obtaining lower estimates requires more sophisticated techniques (for instance the Temple-like methods \[19, ยง XIII.2\]), some of them being very system-dependent (e.g. the moment method proposed in for rational-fraction potentials or the Riccati-Padรฉ method proposed in for one-dimensional Schrรถdinger equations).
The optimized variational methods. For a many-body system governed by pairwise interactions, an interesting strategy is to approximate $`E_0`$ from below in terms of the ground-states of the two-body subsystems . Such an approach has been successfully applied to Coulombian (bosonic and fermionic) systems of charged particles or self-gravitating bosons . Clever refinements have been proposed that provide some very accurate lower bounds of $`E_0`$ for the three-body and the four-body systems . Though not easily generalizable to an arbitrary number of particles, these last optimized variational methods can be applied to interactions that are not necessarily Coulombian and may be relevant for quarks models where some inequalities between baryon and meson masses represent theoretical, numerical and experimental substantial information \[17, 20, and references therein\]. In practice, the optimized variational methods allow to efficiently treat some models that have simple scaling properties, for instance when the two-body interaction can be described by a purely radial potential of the form $`v(r)\mathrm{sign}(\beta )r^\beta `$. The main reason relies in the fact that, except the Coulombian ($`\beta =1`$) and the harmonic ($`\beta =2`$) interactions, the exact form of the ground-state energy of the two-body problem is not known. Yet, one can still take advantage of the power-law behavior of $`v`$ to obtain worthwhile lower bounds for the *ratio* between the $`N`$-body and the $`2`$-body ground-state energies.
The differential method. Besides the variational and perturbative techniques that are mentioned above, there exists a third very general method for approximating the ground-state energy of a quantum system, namely the differential method (see \[15, 16, and references therein for a historical track\]) whose starting point is recalled in section 2. for the sake of completeness. As for the variational methods, the differential method call on a family of trial functions that supposedly mimic the ground-state and that allow for the construction of a function (the average of the Hamiltonian in the former case, the so-called local energy in the latter case) whose absolute extrema within the chosen trial family provide bounds on the exact ground-state energy $`E_0`$. One of the main advantages of the differential method over the other ones is that no integral is required and, then it allows to work, even analytically, with rather complicated trial functions by encapsulating some rich structure of the potential. It is also worth mentioning that the same test function leads to both upper and lower bounds on $`E_0`$ and then the estimates comes with a rigorous window. Though applicable to many models (to systems involving a magnetic field, to discrete systems, to non-Schrรถdinger equations, etc.) the major inconvenient of the differential method is that it requires, as a crucial hypothesis, the exact eigenfunction to remain non-negative in the configuration space (it must have no node). Therefore it excludes any ground-state whose spatial wave-function is antisymmetric under some permutations of its arguments. As far as fermionic systems are involved, the differential method will concern only those whose ground-state eigenfunction remains symmetric under permutations of the spatial positions of the identical particles.
The aim of this paper is to apply the differential method specifically to a system made of $`N`$ non-relativistic particles of masses $`m_i`$ in a d-dimensional space whose Hamiltonian has the form
$$\stackrel{~}{H}=\underset{i=0}{\overset{N1}{}}\frac{๐ฉ_i^2}{2m_i}+V(๐ซ_0,\mathrm{},๐ซ_{N1}).$$
(1)
When the $`N`$ particles located at $`\{๐ซ_i\}_{i=0,\mathrm{},N1}`$ interact only through pairwise potentials $`v_{ij}=v_{ji}`$ , $`V`$ is given by
$$V=\underset{\begin{array}{c}i,j=0\\ i<j\end{array}}{\overset{N1}{}}v_{ij}(๐ซ_{ij})$$
(2)
where $`๐ซ_{ij}\stackrel{\mathrm{def}}{=}๐ซ_j๐ซ_i`$. The spin-dependent interactions, if any, are assumed to be included somehow in the scalar potential $`V`$ and $`\stackrel{~}{H}`$ will be supposed to act on spatial wave-functions only; in other words the possible spin configuration have been factorized out in one way or another. When $`N=3`$ and $`N=4`$ and for power-law potentials, this is the same kind of systems to which the optimized variational method applies also. We will consider the Coulombian case in section 3 and systematically compare the estimates given by the variational methods and the differential method. In section 4, general interactions are considered (not necessarily power-law $`v_{ij}`$โs). This is of course relevant for estimating the ground-state energy of a system where the spin-independent strong interactions are dominant; for heavy enough quarks for instance, it is known that the non-relativistic form (1) may be pertinent<sup>1</sup><sup>1</sup>1Possible relativistic corrections may be included (for instance by considering the spinless Salpeter equation) since the differential method does not require a quadratic kinetic energy.. At atomic scales, the method could be applied to clouds made of neutral atoms where short-range interactions govern the dynamical properties.
## 2 The differential method
### 2.1 The general strategy
The necessary but sufficient condition for the differential method to work is the following: the Hamiltonian $`H`$ has one bound state $`|\mathrm{\Phi }_0`$, associated with energy $`E_0`$, such that $`\mathrm{\Phi }_0(q)\stackrel{\text{def}}{=}q|\mathrm{\Phi }_0`$ remains non-negative in an appropriate $`q`$-representation, say of spatial positions. For a $`N`$-body system governed by the Hamiltonian (1), the dynamics in the center-of-mass frame corresponds to a reduced Hamiltonian $`H`$ whose ground-state<sup>2</sup><sup>2</sup>2We will only consider the cases where at least one bound state exists. Physically, this can be achieved with a confining external potential (a โtrapโ is currently used in experiments involving cold atoms). Formally, this can be obtained in the limit of one mass, say $`m_0`$, being much larger than the others. The external potential appears to be the $`v_{0i}`$โs, created by such an infinitely massive motionless device. It will trap the remaining $`N1`$ particles in some bound states if the $`v_{0i}`$โs increase sufficiently rapidly with the $`r_{0i}`$โs. $`\mathrm{\Phi }_0`$ has precisely this positivity property in the whole configuration space $`๐ฌ_N`$ of the $`(N1)\text{d}`$ relative coordinates $`q_N\stackrel{\text{def}}{=}(๐ซ_1๐ซ_0,\mathrm{},๐ซ_{N1}๐ซ_0)`$. This is the Krein-Rutman theorem (see \[19, ยงXIII.12\]). For each state $`|\phi `$, the hermiticity of $`H`$ implies the identity $`\mathrm{\Phi }_0|(HE_0)|\phi =0`$. If we choose $`|\phi `$ such that its representation $`\phi (q_N)`$ is a smooth normalizable real wave-function, we obtain
$$_{๐ฌ_N}\mathrm{\Phi }_0^{}(q_N)(HE_0)\phi (q_N)๐q_N=0.$$
(3)
Taking into account the positivity of $`\mathrm{\Phi }_0`$ on $`๐ฌ_N`$, there necessarily exists some $`q_N`$ such that $`(HE_0)\phi (q_N)0`$ and some other configurations for which $`(HE_0)\phi (q_N)0`$. Choosing $`\phi >0`$ on $`๐ฌ_N`$, we get both an upper and a lower bound on $`E_0`$:
$$\underset{๐ฌ_N}{inf}\left(E_{\mathrm{loc}}^{[\phi ]}(q_N)\right)E_0\underset{๐ฌ_N}{sup}\left(E_{\mathrm{loc}}^{[\phi ]}(q_N)\right),$$
(4)
where the local energy is defined by
$$E_{\mathrm{loc}}^{[\phi ]}(q_N)\stackrel{\mathrm{def}}{=}\frac{H\phi (q_N)}{\phi (q_N)}.$$
(5)
In other words, the differential method provides an estimate
$$E_0^{(\mathrm{d}.\mathrm{m}.)}\stackrel{\mathrm{def}}{=}\frac{1}{2}\left[\underset{๐ฌ_N}{sup}\left(E_{\mathrm{loc}}^{[\phi ]}(q_N)\right)+\underset{๐ฌ_N}{inf}\left(E_{\mathrm{loc}}^{[\phi ]}(q_N)\right)\right]$$
(6)
that comes with a rigorous windows $`\pm \mathrm{\Delta }E_0^{(\mathrm{d}.\mathrm{m}.)}`$ where
$$\mathrm{\Delta }E_0^{(\mathrm{d}.\mathrm{m}.)}\stackrel{\mathrm{def}}{=}\frac{1}{2}\left[\underset{๐ฌ_N}{sup}\left(E_{\mathrm{loc}}^{[\phi ]}(q_N)\right)\underset{๐ฌ_N}{inf}\left(E_{\mathrm{loc}}^{[\phi ]}(q_N)\right)\right].$$
(7)
Unlike for the variational method, the determination of the absolute extrema of the local energy does not require the computation of any integral. Even the norm of the test function $`\phi `$ is not required provided it remains finite. The two inequalities (4) become equalities (the local energy becomes a flat function) when $`\phi =\mathrm{\Phi }_0`$ and therefore we will try to construct a test function that mimics $`\mathrm{\Phi }_0`$ at best. We will choose $`\phi `$ that respects the a priori known properties of $`\mathrm{\Phi }_0`$: its positivity, its boundary conditions and its symmetries if there are any. Since for each test function the error on $`E_0`$ is controlled by inequalities (4), the strategy for obtaining decent approximations is clear: First, we must choose or construct $`\phi `$ to eliminate all the singularities of the local energy in order to work with a bounded function. For instance, when the Hamiltonian has the form $`p^2+V`$ with $`V`$ being unbounded at some finite or infinite distances, the first kinetic term of the local energy $`E_{\mathrm{loc}}^{[\phi ]}=\mathrm{\Delta }\phi /\phi +V`$ must compensate the singular behavior of $`V`$ for the corresponding configurations (we will work systematically with units such that $`\mathrm{}=1`$). Once a bounded local energy, say $`E_{\mathrm{loc}}^{[\phi _0]}`$, is obtained, we can proceed to a second step: perturb the test function, $`\phi _0\phi =\phi _0+\delta \phi `$, in the neighborhood of the absolute minimum (resp. maximum) of $`E_{\mathrm{loc}}^{[\phi _0]}`$ in order to increase $`\mathrm{min}E_{\mathrm{loc}}^{[\phi _0]}`$ (resp. decrease $`\mathrm{max}E_{\mathrm{loc}}^{[\phi _0]}`$). Up to the end of this article, we will focus on the first step: we will show how obtaining a bounded local energy furnishes some sufficiently constrained guidelines for obtaining reasonable bounds on $`E_0`$<sup>3</sup><sup>3</sup>3One can understand it from the extreme sensitivity of the local energy to any local perturbation of the test function: while, in the variational methods, the quantity $`\phi |H|\phi /\phi |\phi `$ is quite robust to local perturbations because it represents precisely an average on the configuration space, the local energy may become unbounded quite easily by canceling locally $`\phi `$ faster than $`H\phi `$.. We will keep for future work the systematic local improvements of the absolute extrema of the local energy. In \[15, ยง 6\], I have shown on a simple example how this can be done.
### 2.2 Illustration in the two-body case
Before coping with much complex systems, let us first consider the case of the two-body problem that can be reducible to a one single non-relativistic particle of unit mass in an external potential $`V`$. The local energy is
$$E_{\mathrm{loc}}^{[\phi ]}=\frac{\mathrm{\Delta }\phi }{2\phi }+V=\frac{1}{2}\mathrm{\Delta }S\frac{1}{2}(S)^2+V$$
(8)
where $`S\stackrel{\mathrm{def}}{=}\mathrm{ln}(\phi )`$ is a well-defined function when $`\phi >0`$.
For the d Coulombian potential $`V(r)=\kappa /r`$ ($`\text{d}>1`$), the singularity at $`r=0`$ controls the local behavior of the test function if one wants a bounded local energy. If $`lim_{r0}S(r)`$ is finite, by possibly subtracting an irrelevant constant term, we can suppose that this limit vanishes. Therefore, without too much loss of generality, we assume that $`S`$, can be asymptotically expanded near $`r=0`$ on a family of power functions (for $`\text{d}=1`$, the logarithmic functions should be considered) whose dominant term can be written like $`S(r)_{r0}s_0r^{\sigma +1}/(\sigma +1)`$ with $`\sigma 1`$ and $`s_00`$. Balancing the dominant terms in the local energy, it is therefore straightforward, to check that the only choice for the parameters $`s_0`$ and $`\sigma `$ to get rid of the Coulombian singularity is too take $`\sigma =0`$ and $`s_0=2\kappa /(\text{d}1)`$. It happens that for $`S`$ exactly equal to $`2\kappa r/(\text{d}1)`$, we obtain a global constant local energy, namely $`2\kappa ^2/(\text{d}1)^2`$. Therefore we have obtained the exact wave-function of the ground-state provided we eventually check that the wave-function is square-integrable which is true for $`\kappa <0`$.
For the harmonic oscillator $`V(r)=\omega ^2r^2/2`$, $`V`$ is unbounded as $`r`$ increases ($`r=+\mathrm{}`$ is a singular point for $`V`$). If we tentatively look for an $`S`$ whose asymptotic expansion at $`r+\mathrm{}`$ has a leading term of the form $`s_0r^{\sigma +1}/(\sigma +1)`$, we necessarily get $`\sigma =1`$ and $`s_0=\omega `$ (the cases where $`\sigma 1`$ are ruled out by the square-integrable property). The local energy $`\text{d}\omega /2`$ is actually constant for all $`r`$โs and indeed represents the exact ground-state energy. More generally, with the help of standard linear algebra arguments, for $`V`$ being any definite positive quadratic form, we can always find a quadratic form $`S`$ for which the local energy is globally constant.
### 2.3 Formulation for the many-body problem
When the potential $`V`$ has the form (2), a natural choice of trial function is to take (for variational techniques in a few nuclear body context, such a choice has been used by for instance)
$$\phi (q_N)=\underset{\begin{array}{c}i,j=0\\ i<j\end{array}}{\overset{N1}{}}\varphi _{ij}(๐ซ_{ij})$$
(9)
where each of the $`N(N1)/2`$ functions $`\varphi _{ij}(๐ซ)=\varphi _{ji}(๐ซ)`$ depends on d coordinates<sup>4</sup><sup>4</sup>4The present paper wants mainly to stress the simplicity of the differential method. It does not seek for a real performance at the moment and we will not try to improve the choice of coordinates. Working with Jacobi coordinates, for instance, or constructing optimized coordinates as done in may lead to better results. Anyway, we will see that the numerical results of section 3 are satisfactory enough for validating the approach by the differential method..
It is straightforward to check that this choice describes a state with a fixed center-of-mass: indeed we have $`(_{i=0}^{N1}๐ฉ_i)|\phi =\mathrm{๐}`$. Hence, $`\stackrel{~}{H}\phi =H\phi `$ and the local energy is given by
$$\begin{array}{c}\hfill E_{\mathrm{loc}}^{[\phi ]}(q_N)=\underset{\begin{array}{c}i,j=0\\ i<j\end{array}}{\overset{N1}{}}\left(\frac{1}{2m_{ij}}\frac{\mathrm{\Delta }\varphi _{ij}(๐ซ_{ij})}{\varphi _{ij}(๐ซ_{ij})}+v_{ij}(๐ซ_{ij})\right)\\ \hfill \underset{\widehat{j,i,k}}{}\frac{1}{m_i}\frac{\mathbf{}\varphi _{ij}(๐ซ_{ij})}{\varphi _{ij}(๐ซ_{ij})}\frac{\mathbf{}\varphi _{ik}(๐ซ_{ik})}{\varphi _{ik}(๐ซ_{ik})}\end{array}$$
(10)
where $`m_{ij}`$ stands for the reduced masses $`m_im_j/(m_i+m_j)`$. The last sum involves all the $`N(N1)(N2)/2`$ angles $`(\widehat{j,i,k})`$ between $`๐ซ_{ij}`$ and $`๐ซ_{ik}`$ that can be formed with all the triangles made of three particles having three distinct labels ($`ij`$, $`ik`$, $`jk`$). Let us now take $`\varphi _{ij}`$ to be a positive solution of the two-body spectral equation
$$\frac{1}{2m_{ij}}\mathrm{\Delta }\varphi _{ij}+v_{ij}\varphi _{ij}=ฯต_{ij}\varphi _{ij}.$$
(11)
The local energy becomes
$$E_{\mathrm{loc}}^{[\phi ]}(q_N)=\underset{\begin{array}{c}i,j=0\\ i<j\end{array}}{\overset{N1}{}}ฯต_{ij}\underset{\widehat{j,i,k}}{}\frac{1}{m_i}\mathbf{}S_{ij}(๐ซ_{ij})\mathbf{}S_{ik}(๐ซ_{ik})$$
(12)
where $`S_{ij}\stackrel{\mathrm{def}}{=}\mathrm{ln}(\varphi _{ij})`$. The trial wave-function (9) of the global system must be kept square-integrable but it is not necessary for *all* two-body subsystems to have a bound state when isolated<sup>5</sup><sup>5</sup>5But for some pairing, (11) must have a normalizable solution. The cases of Borromean states where no two-body binding is possible cannot be described by the form (9) if we keep (11).. For instance, for two electric charges having the same sign a positive but non-normalizable solution of (11) can be found. When $`v_{ij}`$ admits at least one bound state (see also footnote 2), thanks to the Krein-Rutman theorem, we are certain to get a positive $`\varphi _{ij}`$ when taking the ground-state of the two-body system and $`ฯต_{ij}`$ its corresponding energy. At finite distances, if the possible singularities of $`v_{ij}`$ are not too strong, we expect that $`\varphi _{ij}`$ and then $`S_{ij}`$ to be smooth enough for $`E_{\mathrm{loc}}`$ to remain bounded. At infinite distances, $`E_{\mathrm{loc}}`$ is expected to become infinite if $`v_{ij}`$ does not tend to a constant sufficiently quickly. To see that, one can take purely radial potentials. i.e $`v_{ij}(๐ซ)=v_{ij}(r)`$ where $`r\stackrel{\text{def}}{=}๐ซ`$, and consider the asymptotic behavior of $`S_{ij}`$ given by the semiclassical (jwkb) theory (see for instance ). Its derivative is given by $`S_{ij}^{}(r)_r\mathrm{}\sqrt{2m_{ij}[v_{ij}(r)ฯต_{ij}]}`$ and is not bounded if $`v_{ij}`$ is not (at infinite distances). Strictly speaking, it is only for short-distant potentials that we can hopefully obtain rigorous non-trivial inequalities (4) while keeping the choice (9) with (11). However, as will be discussed in section (4), the ground-state energy may be generally not be very sensitive to the potential at large distances (far away where $`\mathrm{\Phi }_0`$ is localized) and this physical assumption may be implemented by introducing a cut-off length from the beginning.
For purely radial potentials (12) simplifies in
$$E_{\mathrm{loc}}^{[\phi ]}(q_N)=\underset{\begin{array}{c}i,j=0\\ i<j\end{array}}{\overset{N1}{}}ฯต_{ij}\underset{\widehat{j,i,k}}{}\frac{1}{m_i}S_{ij}^{}(r_{ij})S_{ik}^{}(r_{ik})\mathrm{cos}(\widehat{j,i,k}).$$
(13)
Yet, for a multidimensional, non-separable, Schrรถdinger equation like (11), a jwkb-like asymptotic expression is generally not available \[14, Introduction\]. Nevertheless, the differential method is less demanding than the semiclassical approximations: we will try to keep the local energy, like the one given by (12), bounded at infinity but we will not necessarily require it to tend to the same limit in all directions.
## 3 The Coulombian problem
The purely Coulombian problem in $`\text{d}>1`$ dimensions corresponds to the situation where all $`v_{ij}`$โs are radial potentials and have the form
$$v_{ij}(r)=\frac{e_{ij}}{r}$$
(14)
for $`N(N1)/2`$ coupling constants $`e_{ij}`$ that may be or may be not constructed from individual quantities like charges. Provided a $`N`$-body ground-state exists, we can solve exactly (11) making use of $`\mathrm{\Delta }\varphi (r)=\varphi ^{\prime \prime }(r)+(\text{d}1)\varphi ^{}(r)/r`$. We obtain a bounded local energy given by
$$E_{\mathrm{loc}}^{[\phi ]}(q_N)=\underset{\begin{array}{c}i,j=0\\ i<j\end{array}}{\overset{N1}{}}\frac{2m_{ij}e_{ij}^2}{(\text{d}1)^2}\frac{4}{(\text{d}1)^2}\underset{(\widehat{j,i,k})}{}\frac{m_{ij}m_{ik}e_{ij}e_{ik}}{m_i}\mathrm{cos}(\widehat{j,i,k}).$$
(15)
For obtaining upper and lower bounds on $`E_0`$, one has just to calculate the absolute extrema of such a function. It can be done by standard optimization routines up to quite large $`N`$ and even analytically in some cases (see below). The recipe is therefore simple and systematic: as far as only Coulombian interactions are involved, we can work with generic masses and coupling constants for which (9) is normalizable. The remaining of this section will concern the quality of these bounds and then we will accord our attention to cases that have been treated by other methods, mainly those treated in the references cited in the second paragraph of the introduction. More specifically, in order to leave aside the problem of the existence of a ground-state we will consider the case of attractive interactions only<sup>6</sup><sup>6</sup>6For an immediate application in the case of charged electric particles see where the Helium atom is discussed. While the differential method provides an analytical non-trivial upper bound for the ground-state of the Helium-like atoms, $`E_0(Z1/2)^2`$ in the simplest model ($`\text{d}=3`$, non-relativistic, spinless and with an infinitely massive nucleus of charge $`Z`$ in atomic units), in the case of more delicate systems like the positronium ion ($`e^+,e^{},e^{}`$), a systematic improvement is clearly required but is beyond the scope of this paper as explained at the end of subsection 2.1. Indeed, for ($`e^+,e^{},e^{}`$) if we content ourselves with eliminating the singularities, we obtain, for $`\text{d}=3`$, $`9m\alpha ^2/8E_00`$ ($`\alpha `$ being the fine structure constant). The upper bound is trivial while the lower bound is even worse compared to $`3m\alpha ^2/4`$ obtained in \[2, eq. (6.4)\] or to $`3m\alpha ^2/4`$ obtained by a simple crude argument \[2, eq. (6.6)\] (the exact result is $`E_0m\alpha ^2/4`$). (all $`e_{ij}`$โs being negative).
### 3.1 Arbitrary number of identical attractive particles
In this section we consider one species of particles only: for all $`i`$ and $`j`$ we denote $`m_i=m`$ and $`e_{ij}=g^2`$. The local energy (15) becomes
$$E_\phi (q_N)=\frac{ฯต_0}{(\text{d}1)^2}\left(\frac{1}{2}N(N1)+F_N(q_N)\right)$$
(16)
where $`ฯต_0\stackrel{\mathrm{def}}{=}mg^2`$. The function
$$F_N(q_N)\stackrel{\mathrm{def}}{=}\underset{(\widehat{j,i,k})}{}\mathrm{cos}(\widehat{j,i,k})$$
(17)
is invariant under translations and rotations but also under dilations of the particle configuration. For $`N=3`$, the appendix Appendix: Extrema for the three-body Coulombian problem proofs that $`sup_{๐ฌ_3}F_3=3/2`$ is reached when the three particles make an equilateral triangle and $`inf_{๐ฌ_3}F_3=1`$ is obtained when they are aligned. From this last result we are able to provide the lower bounds for $`F_N`$ for any $`N`$: by a decomposition of $`F_N`$ into a sum on $`N(N1)(N2)/6`$ triangle contributions,
$$F_N(q_N)=\underset{\begin{array}{c}\{i_1,i_2,i_3\}\\ 1i_1<i_2<i_3N\end{array}}{}\underset{=F_3(๐ซ_{i_1},๐ซ_{i_2},๐ซ_{i_3})}{\underset{}{\mathrm{cos}(\widehat{i_3,i_1,i_2})+\mathrm{cos}(\widehat{i_1,i_2,i_3})+\mathrm{cos}(\widehat{i_2,i_3,i_1})}},$$
(18)
all the $`F_3`$โs in the sum reach their minimum simultaneously when all the particles are aligned, and for this configuration we have $`inf_{๐ฌ_N}F_N=N(N1)(N2)/6`$. From (16), we deduce that for each $`N`$
$$E_0\frac{ฯต_0}{6(\text{d}1)^2}N(N1)(N+1).$$
(19)
For $`\text{d}=3`$, the same exponential test-functions lead to the better variational estimate \[13, eq. (17)\]
$$E_0ฯต_0\frac{25}{512}N(N1)^2$$
(20)
($`25/5120.04881/240.0417`$). This was expected from the general identity valid for any normalized function $`\phi `$,
$$_๐ฌ\phi ^{}(q)H\phi (q)๐q=_๐ฌ|\phi (q)|^2E_{\mathrm{loc}}^{[\phi ]}(q)๐q\underset{๐ฌ}{sup}\left(E_{\mathrm{loc}}^{[\phi ]}(q)\right).$$
(21)
The differential method always gives worse upper bounds than the variational method with the same test-functions but, in the last case, one still has to be able to compute the integrals and one cannot generally estimate how far from the exact value the average Hamiltonian is. For a different choice of test functions, a better variational upper-bound has been obtained \[1, eq. (16)\],
$$E_0<.0542N(N1)^2.$$
(22)
As far as lower estimates are concerned, bounding $`F_N`$ from above will allow us to improve the existing results, namely (for $`\text{d}=3`$)
$$E_0\frac{1}{16}N^2(N1)$$
(23)
obtained in \[1, eq. (12)\].
First, when $`N`$ is not too large for the numerical computation to remain tractable, the direct calculation of $`sup_{๐ฌ_N}F_N`$ shows (see figure 1) that it gives better lower estimates than (23). For very large $`N`$, we can nevertheless benefit from the maximum of $`F_M`$ for smaller $`M`$. Indeed, for $`MN`$ we can decompose $`F_N`$ into contributions of $`M`$-clusters as follows:
$$F_N(q_N)=\underset{M\text{subclusters}}{}\frac{(M3)!(NM)!}{(N3)!}F_M(q_M)$$
(24)
where the sum is taken on all the $`M`$-subclusters, labeled by the coordinates $`q_M`$, that can be formed with the given configuration $`q_N`$. This sum involves exactly $`N!/M!/(NM)!`$ terms and we have
$$\underset{๐ฌ_N}{sup}F_N\frac{N(N1)(N2)}{M(M1)(M2)}\underset{๐ฌ_M}{sup}F_M.$$
(25)
This leads to define
$$\alpha _M\stackrel{\mathrm{def}}{=}\frac{sup_{๐ฌ_M}F_M}{M(M1)(M2)}$$
(26)
and from (16) we find
$$E_0\frac{ฯต_0}{(\text{d}1)^2}N(N1)\left(\frac{1}{2}+\alpha _M(N2)\right).$$
(27)
Since, from (25), $`\alpha _M`$ is decreasing when $`M`$ increases, the larger $`M`$ the better the lower estimate of $`E_0`$.
For $`M=3`$ we have already seen that $`\alpha _3=1/4`$; for $`\text{d}=3`$, (27) reproduces exactly (23). No better estimate is obtained when considering $`M=4`$. Indeed, the configuration of particles that maximizes $`F_4`$ corresponds to the regular tetrahedron because its faces, that are equilateral triangles, maximize the contributions of all the 3-subclusters simultaneously. We obtain $`sup_{๐ฌ_4}F_4=6`$ and hence $`\alpha _4=\alpha _3`$.
For $`M=5,6,7,8`$ and $`\text{d}=3`$, the configurations that maximize $`F_M`$ can be seen in figure 2. Crossed numerics and analytical studies lead to very plausible conjectures on the geometrical description of the configuration for $`M=5`$ and $`M=8`$ for which explicit analytical value of $`\alpha _M`$ can be proposed . The lower bound for (27), for $`NM`$, is strictly improved when increasing $`M`$ from 5 and in particular is better than (23).
However the sequence of improvements obtained this way seems to saturate up to $`\alpha _{\mathrm{}}=2/9`$:
$$\begin{array}{c}\alpha _3=\alpha _4=\frac{1}{4}\alpha _5.2432\alpha _6.2414\alpha _7.2382\alpha _8.2366\mathrm{}\hfill \\ \hfill \mathrm{}\alpha _{30}.2266\mathrm{}\alpha _{\mathrm{}}=\frac{2}{9}.2222.\end{array}$$
(28)
From the optimized configuration obtained with $`N`$ about several tens (see figure 3a), we can hopefully guess that the limit $`N\mathrm{}`$ leads to a continuous and uniform distribution of the particles on the same sphere. The continuous limit of $`sup_{๐ฌ_N}F_N`$ varies as $`N^3`$ with $`N`$. If, on the unit sphere $`๐ฎ`$, the $`N`$ particles get distributed uniformly with density $`\sigma =N/4\pi `$, the continuous limit of $`sup_{๐ฌ_N}F_N`$ is given by ($`\mathrm{d}S_i`$ is an infinitesimal portion of the sphere near the point $`P_i`$, $`i=0,1,2`$; see figure 3b)
$`\underset{๐ฌ_N}{sup}F_N`$ $``$ $`{\displaystyle _๐ฎ}\sigma dS_0{\displaystyle \frac{1}{2}}{\displaystyle _๐ฎ}\sigma dS_1{\displaystyle _๐ฎ}\sigma dS_2\mathrm{cos}(\widehat{P_1,P_0,P_2});`$
$``$ $`{\displaystyle \frac{1}{2}}\sigma ^3\mathrm{\hspace{0.17em}8}\pi ^2{\displaystyle _0^\pi }d\alpha _1{\displaystyle _0^\pi }d\alpha _2{\displaystyle _0^{2\pi }}d\psi _2\mathrm{sin}(\alpha _1)\mathrm{sin}(\alpha _2)\mathrm{cos}(\widehat{P_1,P_0,P_2}).`$
When $`\mathrm{cos}(\widehat{P_1,P_0,P_2})=\frac{๐_\mathrm{๐}๐_\mathrm{๐}}{๐_\mathrm{๐}๐_\mathrm{๐}}\frac{๐_\mathrm{๐}๐_\mathrm{๐}}{๐_\mathrm{๐}๐_\mathrm{๐}}`$ is expressed as a function of $`\alpha _1`$, $`\alpha _2`$ and $`\psi _2`$, we straightforwardly get $`sup_{๐ฌ_N}F_N/N^3_N\mathrm{}2/9=\alpha _{\mathrm{}}`$. For infinite $`N`$, this last result supplement the upper bound given by (22) and we have
$$N\mathrm{};\frac{1}{18}.0556\frac{E_0}{N^3}.0542.$$
(29)
### 3.2 Three particles with one different from the two others
Without loss of generality, we can choose units where $`e_{12}=e_{13}=e_{23}=1`$, $`m_1=m_2=1`$, $`m_0=m`$. For $`\text{d}=3`$ the local energy (15) simplifies into
$$E_{\mathrm{loc}}^{[\phi ]}(q_3)=\frac{5m+1}{4(m+1)}\frac{m}{2(m+1)}\left(\mathrm{cos}\theta _1+\mathrm{cos}\theta _2+\frac{2}{m+1}\mathrm{cos}\theta _3\right).$$
(30)
The general study of appendix Appendix: Extrema for the three-body Coulombian problem applied for $`(a_1,a_2,a_3)=(1,1,\frac{2}{m+1})`$ allows to get the following analytic bounds on $`E_0`$:
$`0m`$ $`1;`$ $`{\displaystyle \frac{1}{4}}{\displaystyle \frac{m}{8}}{\displaystyle \frac{m(m+2)}{(m+1)^2}}`$ $`E_0{\displaystyle \frac{1}{4}}{\displaystyle \frac{m(2m+1)}{(m+1)^2}};`$ (31a)
$`1m`$ $`3;`$ $`{\displaystyle \frac{1}{4}}{\displaystyle \frac{m}{8}}{\displaystyle \frac{m(m+2)}{(m+1)^2}}`$ $`E_0{\displaystyle \frac{1}{4}}{\displaystyle \frac{m(m+2)}{(m+1)^2}};`$ (31b)
$`3m`$ ; $`{\displaystyle \frac{1}{4}}{\displaystyle \frac{m(2m+1)}{(m+1)^2}}`$ $`E_0{\displaystyle \frac{1}{4}}{\displaystyle \frac{m(m+2)}{(m+1)^2}}.`$ (31c)
For $`m3`$, the lower bound on $`E_0`$ corresponds to a configuration where the particles make a non-degenerate isosceles triangle whose three angles are given by $`\mathrm{cos}\theta _1=\mathrm{cos}\theta _2=(m+1)/4`$ and $`\mathrm{cos}\theta _3=1(m+1)^2/8`$. The other bounds correspond to configurations where the particles are aligned. For $`m1`$, the bounds saturates to $`9/4E_05/4`$ which is quite rough compared to the numerical value $`E_0>1.8`$ obtained for $`m=20`$ with the optimized variational method; yet it is better than the results given by the improved (Hall-Post) variational method \[3, Table 2\] with which it coincides for $`m1`$. For small $`m`$ both upper and lower bounds tend to the 2-body exact energy and provide acceptable bounds: For instance, when $`m=0.05`$, the differential method gives $`.34922E_0.29989`$ while the other ones \[3, Table 2\] give $`.59525E_0`$ (naive variational method) , $`.34922E_0`$ (improved variational method), $`.34666E_0`$ (optimized variational method) and $`E_0.3375`$ (variational with hyperspherical expansion up to $`L=8`$).
As already mentioned, the differential method, though being less precise for $`N=3`$ than the improved or hyperspherical variational approaches, has several advantages: it is much simpler, it provides analytic upper and lower bounds that furnish an explicit estimation of the errors and, at last but not least, can be easily extended to larger $`N`$ (see below) ; though possible in principle, the generalization of the improved variational method has not been done beyond $`N=4`$.
### 3.3 Several examples of four-body systems
The optimized variational method has been successfully proposed for $`N=4`$ in for potentials with scaling-law behavior. For Coulombian interactions with a common coupling constant set to $`1`$, tables and figures 5 and 6 compare the variational results to those obtained from the differential method when $`\text{d}=3`$. The same conclusion as in the previous section can be drawn and here are some examples of explicit analytic bounds that are obtained by partitioning $`F_4`$ in subclusters made of 3 particles:
Let us take $`m_1=m_2=m_3=1`$ and $`m_0=m`$. We have
$$0m1;\frac{9}{8}\frac{3m(m^2+6m+13)}{8(m+1)^2}E_0;$$
(32a)
$$\begin{array}{c}1m2\sqrt{3}1;\hfill \\ \hfill \frac{9}{8}\frac{3m(m^2+6m+13)}{8(m+1)^2}E_0\frac{5m^2+13m+2}{2(m+1)^2};\end{array}$$
(32b)
$$\begin{array}{c}2\sqrt{3}1m;\hfill \\ \hfill \frac{9}{8}\frac{3m(m^2+6m+13)}{8(m+1)^2}+\frac{3m(m+12\sqrt{3})^2}{8(m+1)^2}E_0\frac{5m^2+13m+1}{2(m+1)^2}.\end{array}$$
(32c)
The configuration that minimizes $`E_{\mathrm{loc}}^{[\phi ]}(q_4)`$ given by (15) corresponds to a tetrahedron with an equilateral basis made by particles 1, 2 and 3. The three other faces, with particle $`0`$ at one vertex, are identical isosceles triangles, namely those which maximize (30) when $`m3`$. Such a tetrahedron can indeed be constructed provided the angles at particle $`0`$ are lower than $`2\pi /3`$ which requires $`m2\sqrt{3}12.464`$. For $`m2\sqrt{3}1`$, the configuration that minimizes $`E_{\mathrm{loc}}^{[\phi ]}(q_4)`$ seems numerically to correspond to an equilateral triangle made by particles 1, 2 and 3 with particle $`0`$ at its center (the flat tetrahedron obtained when $`m=2\sqrt{3}1`$). This simple configuration allows to conjecture the analytic lower bound in (32c). For $`m1`$, the upper bound is obtained when the four particles are aligned with the $`0^{\mathrm{th}}`$ at one extremity. For $`m1`$, I am not able to propose an analytic expression for the upper bound.
For $`m_2=m_3=1`$ and $`m_0=m_1=m`$, a tetrahedron that maximizes all the contributions of its faces simultaneously can be constructed for $`m_{}^1mm_{}\stackrel{\mathrm{def}}{=}(1+\sqrt{17}+\sqrt{142\sqrt{17}})/2`$. Two identical faces (see figure 7) corresponding to particles with masses $`(1,1,m)`$ have their three angles given by $`\mathrm{cos}\theta _1=\mathrm{cos}\theta _2=(m+1)/4`$ and $`\mathrm{cos}\theta _3=1(m+1)^2/8`$. The angles $`(\theta _1^{}=\theta _2^{},\theta _3^{})`$ of the two other faces corresponding to particles with masses $`(m,m,1)`$ are obtained replacing $`m`$ by $`m^1`$ in the previous expressions. For $`1/3m3`$ such faces can indeed be constructed but the pairs of identical faces can be put together to construct one tetrahedron provided only that $`\theta _32\theta _1^{}`$. This last condition leads to $`m^4+2m^314m^2+2m+10`$. $`m_{}`$ and $`m_{}^1`$ are the two positive roots of the four-degree-polynomial, the two others being negative. We get
$$m_{}^1.3622mm_{}2.7609;\frac{m^2+10m+1}{2(m+1)}E_0.$$
(33)
### 3.4 Arbitrary number of identical particles plus one different from the others
As we have already seen for identical particles, the possibility of partitioning the local energy in contributions involving more than two particles allow to get some bounds on the ground-state energy of systems made of an arbitrary number $`N`$ of particles. For $`N5`$, this is beyond the scope of the existing optimized variational methods. To see one more last example, let us generalize both cases of sections 3.1 and 3.2 and consider a system made of one particle with mass $`m_0=m`$ and $`N1`$ identical particles of mass $`m_1=\mathrm{}=m_{N1}=1`$. All the identical particles interact with the same coupling constants $`e_{ij}=1`$ and the $`0^{\mathrm{th}}`$ interact with $`e_{0,i}=g^2`$. The local energy is given by (15) as
$$\begin{array}{c}E_{\mathrm{loc}}^{[\phi ]}(q_N)=\frac{1}{(\text{d}1)^2}\{\frac{2mg^4}{m+1}(N1)+\frac{1}{2}(N1)(N2)+F_{N1}(q_{N1})\hfill \\ \hfill +a(m^1)\underset{\begin{array}{c}i,j=1\\ i<j\end{array}}{\overset{N1}{}}[\mathrm{cos}(\widehat{0ij})+\mathrm{cos}(\widehat{0ji})+a(m)\mathrm{cos}(\widehat{i0j})]\}\end{array}$$
(34)
where now $`q_{N1}`$ stands for the configuration of the $`N1`$ identical particles. $`a(m)\stackrel{\mathrm{def}}{=}2g^2/(m+1)`$. By simultaneously bounding the contributions of the triangles that include the $`0^{\mathrm{th}}`$ particle and the contribution of the remaining cluster of the identical particles, we immediately have analytic expressions for bounds on $`E_0`$. For instance the lower bound is given by
$$\begin{array}{c}E_0\frac{1}{(\text{d}1)^2}\{\frac{2mg^4}{m+1}(N1)+\frac{1}{2}(N1)(N2)+\underset{๐ฌ_{N1}}{sup}F_{N1}\hfill \\ \hfill +\frac{1}{2}a(m^1)(N1)(N2)F_3^{\mathrm{max}}(1,1,a(m))\}\end{array}$$
(35)
where $`F_3^{\mathrm{max}}`$ is given by (62) and $`sup_{๐ฌ_{N1}}F_{N1}`$ has been explicitly estimated in section 3.1.
## 4 Arbitrary two-body interaction
### 4.1 Behavior at large distances
Considering attractive Coulombian interactions is relevant for heavy quarks models at short distances but, of course, other kinds of effective potentials are required in most models. Since in general no analytic expressions are known for the two-body ground-state energies $`ฯต_{ij}`$, no method is expected to provide explicit non-trivial bounds on $`E_0`$. However, if one has some experimental clues about $`ฯต_{ij}`$ (by measuring 2-body masses or dissociation energies) or numerical estimates as well, it is always interesting to obtain some relations between the $`ฯต`$โs and the ground-state energies of larger systems. As mentioned in the introduction, this have been achieved in for $`N=3`$ and in for $`N=4`$ when the interactions are of the form $`v_{ij}(r)\mathrm{sign}(\beta )r^\beta `$.
For $`\beta >0`$, the semiclassical argument given at the end of section 2 shows that (13) is expected to be unbounded; then (4) gives no information. If we want to take the advantage of the simple form (13) (that is, to keep the choice (9) with (11) for the test functions), we have to work with finite range potentials. When at large distances, the potential is still confining ($`\beta =2`$ for harmonic forces or $`\beta =1`$ for interquark force in quantum chromodynamics and \[5, for an up-to-date review\]), some different ansatz for $`\phi `$ must be constructed in order to eliminate the singular behavior of the $`v_{ij}`$โs at infinite distances. Actually, the Coulombian case considered in the previous section can be seen as an example of a problem where simple poles at finite distances can be eliminated. Anyway, in many situations, the 2-body ground-state is expected to depend on the behavior of the potential at large distances by exponentially small terms only. If, in the integral (3), we decide to keep only those configurations $`q_N`$ whose size remains in a physical domain bounded by a cut-off length $`\mathrm{\Lambda }`$, then we expect to make an exponentially small error on the estimates of $`E_0`$; this is due to the exponential decay of $`\mathrm{\Phi }_0`$ when two or more particles separate off. Like the $`ฯต_{ij}`$โs, $`\mathrm{\Lambda }`$ is typically obtained from a 2-body dynamics but its precise value is irrelevant if the extremal values of the local energy do not depend on it. It is precisely the case of the Coulombian interactions (more exactly, interactions that can be modeled by Coulombian potential at the energy scale where the ground-state exists) for which the local energy (15) is invariant under dilations.
Since, in the present section, we just want to sketch some main guidelines without working through the details neither being exhaustive, we will consider only the cases where
$$v(r)\underset{r+\mathrm{}}{}0.$$
(36)
### 4.2 Fitting the 2-body ground-state wavefunction
What is new, here, is that the differential method allows us to choose directly the 2-body ground-state wave-functions, or rather their logarithms $`S_{ij}`$. Once some numerical estimate of $`ฯต_{ij}`$ is obtained in one way or another, we can completely bypass the problem of modeling the 2-body potential uniformely. Being free of any integration, the differential method can deal with rather complicated, and therefore rather realistic two-body test functions. An explicit choice of $`S_{ij}`$โs provides an explicit form for the local energy (13).
It frequently happens that we know from experiments the behavior of the two-body potential in some specific regimes (most generally at short and large distances) but not uniformly. We can therefore, in each of these regimes, tentatively obtain, with the help of the differential equation (11), the local functional form of the two-body ground-state wave-function. Matching these local solutions together, and then dealing with quite complicated global expression for $`S`$ and $`v`$, do not represent a serious obstacle for the computation of (13).
To be a little less speculative, let us consider $`N`$ identical particles with unit mass, interacting with a two-body radial potential $`v(r)`$ such that $`v(r)0`$ when $`r\mathrm{}`$ and
$$v(r)\underset{r0^+}{}v_0r^\beta $$
(37)
for some $`v_0`$ and with considering only one case, say $`\beta >0`$. The two-body stationary Schrรถdinger equation (11) becomes
$$\left(\frac{\mathrm{d}^2}{\mathrm{d}r^2}+\frac{\text{d}1}{r}\frac{\mathrm{d}}{\mathrm{d}r}\right)\varphi +v\varphi =ฯต_0\varphi $$
(38)
where $`ฯต_0<0`$ will denote an estimate of the two-body ground-state energy; it can be considered as another parameter that should fit the experiments involving two bodies. Let us guess the behavior of $`S(r)\stackrel{\mathrm{def}}{=}\mathrm{ln}\varphi (r)`$ at short distances by writing for $`\sigma 1`$:
$$S(r)\underset{r0^+}{}\frac{s_0}{\sigma +1}r^{\sigma +1}.$$
(39)
Identifying the leading orders after having reported (39) in (38), we necessarily get (for $`\text{d}>1`$) $`\sigma =1`$ and $`s_0=ฯต_0/\text{d}`$. The next term in the development of $`S`$ can also be determined. For $`0<\beta <2`$, it depends only on the leading term (37) and we have
$$S(r)\underset{r0^+}{}\frac{ฯต_0}{2d}r^2+\frac{v_0}{(\text{d}+\beta )(2+\beta )}r^{2+\beta }+o(r^{2+\beta }).$$
(40)
This local asymptotic series must be matched with the semiclassical behavior at large $`r`$
$$S(r)\underset{r+\mathrm{}}{}\sqrt{ฯต_0}r$$
(41)
since we have supposed (36). The additive constant in $`S`$ is irrelevant since the local energy does not depend on the normalization of $`\varphi `$. A simple choice that ensures the local energy to remain uniformly bounded is to take for $`S^{}`$ a fraction like
$$S^{}(r)=\frac{\frac{ฯต_0}{\text{d}}r+\frac{v_0}{\text{d}+\beta }r^{1+\beta }\sqrt{ฯต_0}r^{1+2\beta }}{1+r^{1+2\beta }}.$$
(42)
Figure 8 show the corresponding $`\varphi `$ for an arbitrary choice of parameters together with the corresponding $`v`$ whose complicated analytic expression on $`[0,+\mathrm{}[`$ is not needed.
### 4.3 Crude bounds
From equation (13), an immediate upper bound on $`E_0`$ is given by:
$$\begin{array}{c}\frac{1}{2}N(N1)ฯต_0\frac{1}{2}N(N1)(N2)\sigma ^2E_0\hfill \\ \hfill \frac{1}{2}N(N1)ฯต_0+\frac{1}{2}N(N1)(N2)\sigma ^2\end{array}$$
(43)
where
$$\sigma \stackrel{\mathrm{def}}{=}\underset{[0,+\mathrm{}[}{sup}|S^{}|.$$
(44)
Because the constraints between the angles $`\widehat{j,i,k}`$ are not taken into account, these inequalities are expected to be rather rough and their quality deteriorate for large $`N`$: when positive, the upper bound becomes irrelevant since we already know that $`E_00`$. Indeed, the decreasing of the $`\sigma ^2`$ term with $`N`$ does not guarantee that the lower bound is better than the Hall-Post bound $`N(N1)\stackrel{~}{ฯต}_0/2`$ or even than the naive one $`N(N1)\stackrel{~}{\stackrel{~}{ฯต}}_0/2`$ \[4, ยง 2.1 and 2.2\] where $`\stackrel{~}{\stackrel{~}{ฯต}}_0`$ (resp. $`\stackrel{~}{ฯต}_0`$ and $`ฯต_0`$) is the ground-state energy for a particle of mass $`(N1)/2`$ (resp. $`N/4`$ and 1/2) in the central potential $`v`$ (recall $`\stackrel{~}{\stackrel{~}{ฯต}}_0<\stackrel{~}{ฯต}_0<ฯต_0`$).
### 4.4 Reduction to a finite number of Coulombian cases
In fact, we can find some bounds of (13) by reducing the problem to a finite number of Coulombian-like cases, that is, where the function to be bound involves constant factors in front of the cosine (compare (15) to (13)). To see this, split the coordinates $`q_N`$ into a scaling factor $`\lambda 0`$ and some angle variables $`\theta `$ among which $`(N1)\text{d}1`$ are independent. Each distance writes $`r_{ij}=\lambda \rho _{ij}(\theta )`$ where the $`\rho _{ij}`$โs are functions that do not depend on the global size of the configuration but on its shape only. Now, from (13), we define (recall $`m_i=1`$)
$$G_N(q_N)\stackrel{\mathrm{def}}{=}\underset{(\widehat{j,i,k})}{}S^{}(r_{ij})S^{}(r_{ik})\mathrm{cos}(\widehat{j,i,k})$$
(45)
and we have
$$\underset{q_N}{inf}\left(G_N(q_N)\right)=\underset{\theta }{inf}\left(\stackrel{~}{G}_N(\theta )\right)$$
(46)
where
$$\stackrel{~}{G}_N(\theta )=\underset{\lambda }{inf}\underset{(\widehat{j,i,k})}{}S^{}[\lambda \rho _{ij}(\theta )]S^{}[\lambda \rho _{ik}(\theta )]\mathrm{cos}(\widehat{j,i,k}).$$
(47)
Analogous relations are obtained for the maxima. For fixed $`\theta `$, when $`\lambda `$ varies from $`0`$ to $`+\mathrm{}`$, the map $`\lambda \left(S^{}[\lambda \rho _{ij}(\theta )]\right)_{0i<jN1}`$ defines a curve $`๐_\theta `$ in a $`n`$-dimensional space with $`n=N(N1)/2`$. $`๐_\theta `$ is bounded if $`S^{}`$ is bounded. More precisely, $`๐_\theta `$ is inside the $`n`$-dimensional hypercube $`\stackrel{\mathrm{def}}{=}[\sigma _{\mathrm{min}},\sigma _{\mathrm{max}}]^n`$ where
$$\sigma _{\mathrm{min}}\stackrel{\mathrm{def}}{=}\underset{[0,+\mathrm{}[}{inf}S^{}$$
(48a)
and
$$\sigma _{\mathrm{max}}\stackrel{\mathrm{def}}{=}\underset{[0,+\mathrm{}[}{sup}S^{}.$$
(48b)
If $`S^{}`$ has the form shown in figure 8, $`๐_\theta `$ starts at the origin ($`S^{}(0)=0`$) and ends at the point $`(1,\mathrm{},1)`$. Taking all the points in $``$ rather than the points of $`๐_\theta `$ leads to a lower bound of $`\stackrel{~}{G}_N`$:
$$\underset{\stackrel{}{s}}{inf}\underset{(\widehat{j,i,k})}{}s_{ij}s_{ik}\mathrm{cos}(\widehat{j,i,k})\stackrel{~}{G}_N(\theta )$$
(49)
where $`\stackrel{}{s}=(s_{ij})_{0i<jN1}`$. Now, whatever the values of the cosines may be, the quadratic function in $`\stackrel{}{s}`$ appearing in the left-hand side of (49) reaches its minimum at a vertex of $``$<sup>7</sup><sup>7</sup>7For any constant $`A`$, any $`n`$-dimensional vector $`\stackrel{}{B}`$ and any symmetric $`n\times n`$-matrix $`C`$ with vanishing diagonal coefficients, the critical points of $`f_n(\stackrel{}{s})=\stackrel{}{s}C\stackrel{}{s}+\stackrel{}{B}\stackrel{}{s}+A`$ are always saddle points: the direction $`s_2=\pm \mathrm{sign}(C_{12})s_1`$ and $`s_i=0`$ for $`i>2`$ makes $`f`$ increase/decrease as $`\pm |C_{12}|s_1^2`$ from its critical value. Therefore the extrema of $`f`$ are reached on the boundary of the domain of $`\stackrel{}{s}`$. For $`\stackrel{}{s}`$ restricted to a $`n`$-dimensional squared box whose faces are given by fixing one $`s_i`$, the restriction of $`f_n`$ to one face, i.e, to a $`(n1)`$-dimensional box, leads to a function $`f_{n1}`$ to which the above argument may be applied again. By repetition down to $`n=1`$, we see that the maximum and the minimum of $`f`$ is necessarily reached at one of the vertices of the original $`n`$-box. . Let us denote by $`๐ฑ`$, the finite set of the $`2^n`$ vertices $`\stackrel{}{s}`$ of $``$ (i.e. for all $`\stackrel{}{s}`$ in $`๐ฑ`$, each $`s_{ij}`$ is either $`\sigma _{\mathrm{min}}`$ or $`\sigma _{\mathrm{max}}`$). We have
$$\underset{\stackrel{}{s}๐ฑ}{inf}\left[\underset{\theta }{inf}\left(F_N^{(\stackrel{}{s})}(\theta )\right)\right]\underset{q_N}{inf}\left(G_N(q_N)\right).$$
(50)
where
$$F_N^{(\stackrel{}{s})}(\theta )\stackrel{\mathrm{def}}{=}\underset{(\widehat{j,i,k})}{}s_{ij}s_{ik}\mathrm{cos}(\widehat{j,i,k}).$$
(51)
In fact, what we have done by obtaining the left-hand side of (50) is to make the values of $`S^{}(r_{ij})`$ independent from those of $`\theta `$. It follows that the inequality (50) will be strict if the value of $`\stackrel{}{s}`$ at the minimizing vertex are incompatible with the geometrical constraints on the configuration of the $`N`$ points. We will illustrate this point in the next subsection. We have obtained
$$E_0\frac{1}{2}N(N1)ฯต_0\underset{\stackrel{}{s}๐ฑ}{inf}\left[\underset{\theta }{inf}\left(F_N^{(\stackrel{}{s})}(\theta )\right)\right].$$
(52a)
and similarly
$$\frac{1}{2}N(N1)ฯต_0\underset{\stackrel{}{s}๐ฑ}{sup}\left[\underset{\theta }{sup}\left(F_N^{(\stackrel{}{s})}(\theta )\right)\right]E_0.$$
(52b)
The function $`F_N^{(\stackrel{}{s})}(\theta )`$ has the same form as the second sum of the right-hand of $`(\text{15})`$. Therefore, the computation of the bounds in (52) is equivalent to a finite number of Coulombian problems (with not necessarily attractive interactions since the sign of $`s_{ij}`$ may change) where we must consider all the possible $`\stackrel{}{s}`$ whose components are either $`\sigma _{\mathrm{min}}`$ or $`\sigma _{\mathrm{max}}`$.
### 4.5 Three bodies
As we have seen in section 3, even in the purely attractive Coulombian cases, an analytic expression of the extrema of $`F_N`$ is not known in general. Anyway, one can always group in clusters the terms involved in (51) like in (24), then use inequalities like (25) and reduce the number of particles. Let us then consider $`N=3`$. It can be shown<sup>8</sup><sup>8</sup>8 The extrema of $`F_3^{(\stackrel{}{s})}(\theta )`$ when $`\theta `$ varies can be calculated with the help of the appendix with $`a_3=s_{01}s_{02}`$, $`a_2=s_{02}s_{12}`$ and $`a_1=s_{01}s_{12}`$. From (61), we get $`f(a_1,a_2,a_3)=\frac{1}{2}(s_{01}^2+s_{02}^2+s_{12}^2)`$ which is always positive and larger than $`a_1+a_2a_3`$, $`a_1+a_2+a_3`$ and $`a_1a_2+a_3`$ corresponding to the three aligned configurations for different ordering of the particles. From (62), the minimum of $`F_3^{(\stackrel{}{s})}(\theta )`$ must therefore correspond to an aligned configuration. Its maximum is reached for the configuration described just after equation (57). that
$$\underset{\stackrel{}{s}๐ฑ}{inf}\left[\underset{\theta }{inf}\left(F_3^{(\stackrel{}{s})}(\theta )\right)\right]=2\sigma _{\mathrm{min}}\sigma _{\mathrm{max}}\sigma ^2$$
(53)
with definitions (44) and (48). $`inf_\theta F_3^{(\stackrel{}{s})}(\theta )`$ is obtained for an aligned configuration which is generically incompatible with $`\stackrel{}{s}`$ being a vertex of the cube $`[\sigma _{\mathrm{min}},\sigma _{\mathrm{max}}]^3`$. For instance, suppose that $`S^{}`$ has the shape depicted in figure 8 where $`\sigma =\sigma _{\mathrm{min}}>\sigma _{\mathrm{max}}>0`$; the value $`2\sigma _{\mathrm{min}}\sigma _{\mathrm{max}}\sigma _{\mathrm{min}}^2`$ is obtained for $`\stackrel{}{s}=(s_{01},s_{02},s_{12})=(\sigma _{\mathrm{min}},\sigma _{\mathrm{min}},\sigma _{\mathrm{max}})`$ and should be realized for $`r_{01}1`$, $`r_{02}1`$ and $`r_{12}r_{\mathrm{max}}`$ (the unique finite distance at which $`S^{\prime \prime }`$ vanishes); but this is incompatible with the alignment condition $`\mathrm{cos}(\widehat{1,0,2})=1`$ where particle $`0`$ is in between the two others which implies $`r_{12}=r_{01}+r_{02}`$. The inequality (50) is therefore strongly strict. It can be improved by reducing the size of the cube $``$ to make its minimizing vertices compatible with the aligned configuration. It can be shown that for $`S^{}`$ of the form shown in figure 8, we have
$$2\sigma _{\mathrm{max}}S^{}(2r_0)\sigma _{\mathrm{max}}^2\underset{q_3}{inf}\left(G_3(q_3)\right)$$
(54)
where $`r_0`$ is the unique strictly positive distance where $`S^{}`$ vanishes<sup>9</sup><sup>9</sup>9Any aligned configuration with $`r_{01}r_{02}r_0`$ and $`r_{12}=r_{01}+r_{02}>r_0`$ corresponds to a negative $`G_3`$. Therefore, as far as its minimum is concerned, the configurations leading to a positive $`inf_\theta F_3^{(\stackrel{}{s})}`$ can be forgotten (see note 8). It is straightforward to check that all the possible relative positions of $`r_{ij}`$ with respect to $`r_{\mathrm{max}}`$ and $`r_0`$ that are compatible with $`r_{12}=r_{01}+r_{02}`$ provide a $`G_3`$ such that (54). . Even though the inequality is still strict because $`r_{12}=2r_0r_{01}+r_{02}=2r_{\mathrm{max}}`$ in general, the bound is much better than (53). For instance, if we take the value of the parameters corresponding to figure 8 we have
$$2\sigma _{\mathrm{min}}\sigma _{\mathrm{max}}\sigma _{\mathrm{min}}^21.2705<2\sigma _{\mathrm{max}}S^{}(2r_0)\sigma _{\mathrm{max}}^2.205.$$
(55)
to be compared with the result of the numerical minimization of $`G_3`$
$$\underset{q_3}{inf}\left(G_3(q_3)\right).1150.$$
(56)
obtained for the aligned configuration where $`r_{01}=r_{02}=r_{12}/2.6107`$.
The other bound
$$\underset{q_3}{sup}\left(G_3(q_3)\right)=\underset{\stackrel{}{s}๐ฑ}{sup}\left[\underset{\theta }{sup}\left(F_3^{(\stackrel{}{s})}(\theta )\right)\right]=\frac{3}{2}\sigma ^2$$
(57)
is actually obtained at the vertex $`\stackrel{}{s}=(\sigma ,\sigma ,\sigma )`$ for an equilateral configuration where the common distance $`r_{01}=r_{02}=r_{12}`$ is where $`|S^{}|`$ reaches its maximum. For $`S^{}`$ like in figure 8, it corresponds to a very large triangle ($`r_{01}1`$) where $`\sigma 1`$. For $`\text{d}=3`$, $`ฯต_0=1`$, $`v_0=1`$, $`\beta =1.5`$, from (50) and results (56), (57), the inequalities (52) give
$$33/2=4.5E_03+.1150=2.885$$
(58)
corresponding to a relative error of $`\mathrm{\Delta }E_0^{(\mathrm{d}.\mathrm{m}.)}/E_0^{(\mathrm{d}.\mathrm{m}.)}22\%`$. This not really impressive (again we emphasize that we are not looking for numerical performance at this stage of development of the differential method) but can be seen as an encouraging starting point since the interactions involved so far in the three-body system are highly not trivial. It would have required much more numerical work to obtain a rigorous window for $`E_0`$ with variational methods (specially a lower bound since the potential considered here does not follow a power-law behavior).
## 5 Conclusion
The differential method appears to offer a completely new strategy for estimating a ground-state energy. For many-body systems, we have seen on several examples how this approach can be fruitful. For attractive Coulombian particles, it can compete with existing others methods (that are based on the variational principle) on several levels: it provides upper *and* lower bounds with comparable numerical precision, its simplicity renders the analytic calculations tractable even for large $`N`$ and/or allows a low cost of numerical computation. Beyond purely Coulombian systems, the differential method, being so general, offers a remarkable flexibility. As have been sketched in the previous section, one can deal with systems where interactions can be very rich (possibly short-ranged with an a priori cut-off); several regimes which are valid at different scales can be implemented at once. There is some hope that future works successfully apply the differential methods for proper realistic potentials.
Unfortunately, I have not been able to generalize the differential method to fermionic systems where the ground-state spatial wave-function is antisymmetric. In such cases, the presence of non-trivial nodal lines breaks down the proof of inequalities (4).
There is a lot of work to be done regarding a systematic improvement of the bounds, once some finite ones have been found with a $`\phi `$ given at first attempt. In this paper, we have not considered some free parameters on which a (family) of test functions, say $`\phi _\zeta `$, may depend. As shown for a one-dimensional system , the locality of the differential method may require a very few number of $`\lambda `$โs at each optimization step (unlike for the variational test functions) for obtaining substantial improvements of the bounds by calculating, say $`sup_\zeta \left[inf_q\left(E_{\mathrm{loc}}^{[\phi _\zeta ]}(q)\right)\right]`$. A precise proof that this approach is efficient for several dimensions remains an open interesting problem.
I thank Jean-Marc Richard for a critical reading of the first proof of this manuscript and acknowledge the generous hospitality of Dominique Delande and Benoรฎt Grรฉmaud of the group โDynamique des systรจmes coulombiensโ at the Laboratoire Kastler-Brossel.
## Appendix: Extrema for the three-body Coulombian problem
For the three-body Coulombian problem, as can be seen from the second sum in the right-hand side of (15), we must find
$$F_3^{\mathrm{max}}(a_1,a_2,a_3)\stackrel{\mathrm{def}}{=}\underset{\mathrm{triangles}}{sup}[a_1\mathrm{cos}\theta _1+a_2\mathrm{cos}\theta _2+a_3\mathrm{cos}\theta _3]$$
(59)
and
$$F_3^{\mathrm{min}}(a_1,a_2,a_3)\stackrel{\mathrm{def}}{=}\underset{\mathrm{triangles}}{inf}[a_1\mathrm{cos}\theta _1+a_2\mathrm{cos}\theta _2+a_3\mathrm{cos}\theta _3]$$
(60)
where the $`\theta `$โs are the angles at the three vertices of the triangle made of the three particles. The $`a`$โs are some real parameters that depend on the masses and the coupling constants. Let us define
$$f(a_1,a_2,a_3)\stackrel{\mathrm{def}}{=}\frac{1}{2}\left(\frac{a_1a_2}{a_3}+\frac{a_1a_3}{a_2}+\frac{a_2a_3}{a_1}\right),$$
(61)
then we have
$$F_3^{\begin{array}{c}\mathrm{max}\\ \mathrm{min}\end{array}}(a_1,a_2,a_3)=\begin{array}{c}\mathrm{max}\\ \mathrm{min}\end{array}\{a_1+a_2a_3,a_1a_2+a_3,a_1+a_2+a_3,f(a_1,a_2,a_3)\}$$
(62)
$`f`$ is considered in the list (62) if and only if the following three conditions are satisfied simultaneously
$$\frac{1}{2}\left|\frac{a_2}{a_3}+\frac{a_3}{a_2}\frac{a_2a_3}{a_1^2}\right|1;\frac{1}{2}\left|\frac{a_1}{a_2}+\frac{a_2}{a_1}\frac{a_1a_2}{a_3^2}\right|1;\frac{1}{2}\left|\frac{a_3}{a_1}+\frac{a_1}{a_3}\frac{a_1a_3}{a_2^2}\right|1.$$
(63)
Here is the proof: We will restrict the values of the $`\theta `$โs to $`[0,\pi ]`$ and the constraint $`\theta _1+\theta _2+\theta _3=\pi `$ is implemented by the Lagrange multiplier method. We are led to extremalise the function $`G_3(\theta _1,\theta _2,\theta _3)\stackrel{\mathrm{def}}{=}a_1\mathrm{cos}\theta _1+a_2\mathrm{cos}\theta _2+a_3\mathrm{cos}\theta _3+\mathrm{}(\theta _1+\theta _2+\theta _3\pi )`$ for unconstrained $`(\theta _1,\theta _2,\theta _3)[0,\pi ]^3`$, $`\mathrm{}`$ being the Lagrange multiplier. The three conditions $`_{\theta _i}G_3=0`$ for $`i=1,2,3`$ lead to $`\mathrm{}=a_1\mathrm{sin}\theta _1=a_2\mathrm{sin}\theta _2=a_3\mathrm{sin}\theta _3`$. The case $`\mathrm{}=0`$ corresponds to the alignment of the three particles and gives the three first values in the list (62) corresponding to $`(\theta _1,\theta _2,\theta _3)=(0,0,\pi )`$ and its circular permutations.
Taking into account the constraint on the angles, we have $`\mathrm{}=a_3\mathrm{sin}\theta _3=a_3\mathrm{sin}\theta _1\mathrm{cos}\theta _2+a_3\mathrm{sin}\theta _2\mathrm{cos}\theta _1=\mathrm{}(a_3\mathrm{cos}\theta _2/a_1+a_3\mathrm{cos}\theta _1/a_2)`$. Therefore when $`\mathrm{}0`$, we find $`a_1\mathrm{cos}\theta _1+a_2\mathrm{cos}\theta _2=a_1a_2/a_3`$ as well as the other relations that are obtained by circular permutations of the indices. From the decomposition $`F_3=\frac{1}{2}(a_1\mathrm{cos}\theta _1+a_2\mathrm{cos}\theta _2)+\frac{1}{2}(a_2\mathrm{cos}\theta _2+a_3\mathrm{cos}\theta _3)+\frac{1}{2}(a_1\mathrm{cos}\theta _1+a_3\mathrm{cos}\theta _3)`$, we obtain the value (61) that must be considered in (62) if and only if there exists some $`\theta `$โs such that
$$a_1\mathrm{sin}\theta _1=a_2\mathrm{sin}\theta _2=a_3\mathrm{sin}\theta _3\mathrm{and}\theta _1+\theta _2+\theta _3=\pi .$$
(64)
Solving these three equations leads to the values for the three $`|\mathrm{cos}\theta _i|`$ that are precisely given by the left-hand sides of the inequalities (63).
|
warning/0506/hep-ph0506293.html
|
ar5iv
|
text
|
# Leading twist contribution to color singlet ๐_{๐โข0,2}โ๐โข๐ decays
## I Introduction
Recently the discrepancy between theoretical prediction for $`\psi =J/\psi `$ and $`\eta _c`$ production in $`e^+e^{}`$ annihilation at $`\sqrt{s}=10.6\mathrm{GeV}`$ Braaten and Lee (2003) and experimental result Abe et al. (2002) has found a surprisingly simple explanation. In the works Ma and Si (2004); Bondar and Chernyak (2004) it was shown that taking into account the intrinsic motion of quarks inside the charmonium mesons in the hard part of the amplitude one can significantly increase the theoretical predictions for the cross section of this reaction and reach the agreement with the experiment.
In the recent work Braguta et al. (2005) we have confirmed this result using a slightly different model. In that paper we have also studied the influence of internal quark motion on scalar and tensor mesons decays into two vector ones (that is $`\chi _{0,2}VV`$. Specifically the decays $`\chi _{c0,2}\rho \rho ,\varphi \varphi `$ and $`\chi _{b0,2}\psi \psi `$ were considered) and shown that the branching fractions of these decays also increase when one takes into consideration in intrinsic quark motion. For example, the agreement between theoretical predictions and experimental result for the $`\chi _{c0}\varphi \varphi `$ branching fraction can be reached. We have also studied the possibility of using $`\chi _{b0,2}\psi \psi `$ decay for $`\chi _{b0,2}`$ mesons observation at Tevatron and LHC colliders.
In the recent paper the BES (2005) $`\chi _{c0,2}`$ mesons were observed in the $`\omega \omega `$ mode and the results for the branching fractions $`\mathrm{Br}(\chi _{c0,2}\omega \omega )`$ were presented. The aim of this short note is to use the formulae presented in Braguta et al. (2005) for these decays and to compare our results with the experimental ones.
## II Helicity matrix elements and distribution functions
Analytical formulae for the width of the decay $`\chi _0VV`$ were presented in the work Anselmino and Murgia (1993) and we will use this results in our paper. The nonzero helicity amplitudes $`๐_{\lambda _1,\lambda _2}^{(0)}`$ of the decay of scalar meson $`\chi _0`$ into vector mesons $`V_1`$ and $`V_2`$ with the helicities $`\lambda _1`$ and $`\lambda _2`$ are given by the expressions
$`๐_{1,1}^{(0)}`$ $`=`$ $`๐_{1,1}^{(0)}=i{\displaystyle \frac{2^{13}}{9\sqrt{3}}}\pi ^3\alpha _s^2ฯต^2{\displaystyle \frac{|R^{}(0)|}{M^4}}f_{}^2I_{1,1}^{(0)}(ฯต),`$
$`๐_{0,0}^{(0)}`$ $`=`$ $`i{\displaystyle \frac{2^{12}}{9\sqrt{3}}}\pi ^3\alpha _s^2{\displaystyle \frac{|R^{}(0)|}{M}}f^2I_{0,0}^{(0)}(ฯต),`$
where
$`ฯต`$ $`=`$ $`m/M,`$
$`m`$ and $`M`$ are masses of vector and scalar mesons respectively, $`R(r)`$ is the radial part of the scalar mesonโs wave function, $`f`$ and $`f_{}`$ are longitudinal and transverse leptonic constants of the vector meson and coefficients $`I_{\lambda _1,\lambda _2}^{(0)}`$ are equal to
$`I_{1,1}^{(0)}`$ $`=`$ $`{\displaystyle \frac{1}{32}}{\displaystyle \underset{0}{\overset{1}{}}}dxdy\varphi _{}(x)\varphi _{}(y){\displaystyle \frac{1}{xy+(xy)^2ฯต^2}}{\displaystyle \frac{1}{(1x)(1y)+(xy)^2ฯต^2}}\times `$
$`\times `$ $`{\displaystyle \frac{1}{2xyxy+2(xy)^2ฯต^2}}\left[1+{\displaystyle \frac{1}{2}}{\displaystyle \frac{(xy)^2(14ฯต^2)}{2xyxy+2(xy)^2ฯต^2}}\right],`$
$`I_{0,0}^{(0)}`$ $`=`$ $`{\displaystyle \frac{1}{32}}{\displaystyle \underset{0}{\overset{1}{}}}dxdy\varphi _{}(x)\varphi _{}(y){\displaystyle \frac{1}{xy+(xy)^2ฯต^2}}{\displaystyle \frac{1}{1x)(1y)+(xy)^2ฯต^2}}\times `$
$`\times `$ $`{\displaystyle \frac{1}{2xyxy+2(xy)^2ฯต^2}}\{1{\displaystyle \frac{1}{2}}{\displaystyle \frac{(xy)^2(14ฯต^2)}{2xyxy+2(xy)^2ฯต^2}}`$
$``$ $`2ฯต^2[1+{\displaystyle \frac{1}{2}}{\displaystyle \frac{(xy)^2(14ฯต^2)}{2xyxy+2(xy)^2ฯต^2}}]\}.`$
In the above equations $`x`$ and $`y`$ are the momentum fractions of the final mesons, carried by quarks and $`\varphi _,(x)`$ are longitudinal and transverse distribution functions of these quarks in mesons.
In Anselmino and Murgia (1993) the similar formulae for nonzero helicity amplitudes of tensor meson decay are also presented:
$`๐_{\lambda _1\lambda _2;\mu }^{(2)}`$ $`=`$ $`\stackrel{~}{๐}_{\lambda _1\lambda _2}e^{i\mu \phi }d_{\mu ,\lambda _1\lambda _1}^{(2)}(\theta ),`$
where $`\mu `$ is the meson spin projection on fixed axe, $`\theta `$ and $`\phi `$ are polar and azimuthal angles of one of the final mesons in $`\chi _2`$ rest frame and reduced amplitudes $`\stackrel{~}{๐}_{\lambda _1\lambda _2}`$ are given by the expressions
$`\stackrel{~}{๐}_{1,1}`$ $`=`$ $`\stackrel{~}{๐}_{1,1}=i{\displaystyle \frac{2^{13}\sqrt{2}}{9\sqrt{3}}}\pi ^3\alpha _s^2ฯต^2{\displaystyle \frac{|R^{}(0)|}{M^4}}f_{}^2I_{1,1}^{(2)},`$
$`\stackrel{~}{๐}_{1,0}`$ $`=`$ $`\stackrel{~}{๐}_{0,1}=\stackrel{~}{๐}_{1,0}=\stackrel{~}{๐}_{0,1}=i{\displaystyle \frac{2^{12}\sqrt{2}}{9}}\pi ^3\alpha _s^2ฯต{\displaystyle \frac{|R^{}(0)|}{M^4}}f_{}fI_{1,0}^{(2)},`$
$`\stackrel{~}{๐}_{1,1}`$ $`=`$ $`\stackrel{~}{๐}_{1,1}=i{\displaystyle \frac{2^{12}}{9}}\pi ^3\alpha _s^2{\displaystyle \frac{|R^{}(0)|}{M^4}}f_{}^2I_{1,1}^{(2)},`$
$`\stackrel{~}{๐}_{0,0}`$ $`=`$ $`i{\displaystyle \frac{2^{11}\sqrt{2}}{9\sqrt{3}}}\pi ^3\alpha _s^2{\displaystyle \frac{|R^{}(0)|}{M^4}}f^2I_{0,0}^{(2)},`$
$`I_{1,1}^{(2)}`$ $`=`$ $`I_{1,1}^{(0)},`$
$`I_{1,0}^{(2)}`$ $`=`$ $`{\displaystyle \frac{1}{32}}{\displaystyle \underset{0}{\overset{1}{}}}{\displaystyle \underset{0}{\overset{1}{}}}dxdy\varphi _{}(x)\varphi _{}(y){\displaystyle \frac{1}{xy+(xy)^2ฯต^2}}{\displaystyle \frac{1}{(1x)(1y)+(xy)^2ฯต^2}}\times `$
$`\times `$ $`{\displaystyle \frac{1}{2xyxy+2(xy)^2ฯต^2}}\left[1+{\displaystyle \frac{1}{2}}{\displaystyle \frac{(xy)^2(14ฯต^2)}{2xyxy+2(xy)^2ฯต^2}}\right],`$
$`I_{1,1}^{(2)}`$ $`=`$ $`{\displaystyle \frac{1}{32}}{\displaystyle \underset{0}{\overset{1}{}}}{\displaystyle \underset{0}{\overset{1}{}}}dxdy\varphi _{}(x)\varphi _{}(y){\displaystyle \frac{1}{xy+(xy)^2ฯต^2}}{\displaystyle \frac{1}{(1x)(1y)+(xy)^2ฯต^2}}\times `$
$`\times `$ $`{\displaystyle \frac{1}{2xyxy+2(xy)^2ฯต^2}},`$
$`I_{0,0}^{(2)}`$ $`=`$ $`{\displaystyle \frac{1}{32}}{\displaystyle \underset{0}{\overset{1}{}}}{\displaystyle \underset{0}{\overset{1}{}}}dxdy\varphi _{}(x)\varphi _{}(y){\displaystyle \frac{1}{xy+(xy)^2ฯต^2}}{\displaystyle \frac{1}{(1x)(1y)+(xy)^2ฯต^2}}\times `$
$`\times `$ $`{\displaystyle \frac{1}{2xyxy+2(xy)^2ฯต^2}}\{1+{\displaystyle \frac{(xy)^2(14ฯต^2)}{2xyxy+2(xy)^2ฯต^2}}+`$
$`+`$ $`4ฯต^2[1+{\displaystyle \frac{1}{2}}{\displaystyle \frac{(xy)^2(14ฯต^2)}{2xyxy+2(xy)^2ฯต^2}}]\}.`$
The leading twist structure functions $`\varphi _{}(x)`$ and $`\varphi _{}(x)`$ can be expressed through the Gegenbauer polynomials (see Lepage and Brodsky (1980) and refrences therein)
$`\varphi _,(x)`$ $`=`$ $`6x(1x)\left[1+{\displaystyle \underset{n=2,4,\mathrm{}}{}}a_n^,C_n^{3/2}(2x1)\right].`$ (1)
In what follows we will restrict ourself to first two terms of this expansion. The derivative of the $`\chi _c`$-meson wave function in the origin can be expressed through the decay widths of these mesons:
$`\mathrm{\Gamma }_{\chi _{c0}}`$ $``$ $`\mathrm{\Gamma }(\chi _{c0}gg)=96{\displaystyle \frac{\alpha _s^2}{M_{\chi _{b0}}^4}}|R^{}(0)|^2,`$ (2)
$`\mathrm{\Gamma }_{\chi _{c2}}`$ $``$ $`\mathrm{\Gamma }(\chi _{c2}gg)={\displaystyle \frac{128}{5}}{\displaystyle \frac{\alpha _s^2}{M_{\chi _{b2}}^4}}|R^{}(0)|^2`$ (3)
and the longitudinal $`\omega `$-meson leptonic constant can be expressed through the $`\omega e^+e^{}`$ decay width using the relation
$`\mathrm{\Gamma }(\omega e^+e^{})`$ $`=`$ $`{\displaystyle \frac{4\pi }{3}}\left({\displaystyle \frac{e_u+e_d}{\sqrt{2}}}\right)^2\alpha ^2{\displaystyle \frac{f^2}{M_\omega }}.`$ (4)
On the other hand, the derivation of the transverse leptonic constant $`f_{}`$ and the structure function momenta $`a_2^,`$ is not so simple. The values of these parameters can be obtained using QCD sum rules Chernyak and Zhitnitsky (1984); Ball and Braun (1996) and will be discussed in the next section.
## III Numerical results and conclusion
With the help of equations (2),(3),(4) we get the following values:
$`|R^{}(0)|^2`$ $`=`$ $`0.16\mathrm{GeV}^5,f=196\mathrm{MeV}.`$
The values of transverse leptonic constant $`f`$ and structure function momenta $`a_2^,`$ were obtained in Chernyak and Zhitnitsky (1984) and reanalyzed in Ball and Braun (1996). According to the last paper, the values of these constants at the renormalization scale $`\mu _0=1\mathrm{GeV}`$ are equal to
$`f_{}(\mu _0)=(160\pm 10)\mathrm{MeV},a_2^{}(\mu _0)`$ $`=`$ $`0.2\pm 0.1,a_2^{}(\mu _0)=0.18\pm 0.10`$
The evaluation to other renormalization scale can be done using the equations
$`a_2^{}(\mu )=a_2^{}(\mu _0)\left({\displaystyle \frac{\alpha _s(\mu )}{\alpha _s(\mu _0)}}\right)^{2/3},a_2^{}(\mu )=a_2^{}(\mu _0)\left({\displaystyle \frac{\alpha _s(\mu )}{\alpha _s(\mu _0)}}\right)^{8/15},f_{}(\mu )=f_{}(\mu _0)\left({\displaystyle \frac{\alpha _s(\mu )}{\alpha _s(\mu _0)}}\right)^{4/27}.`$
Using these values of the parameters we obtain the following branching fractions:
$`\mathrm{Br}(\chi _{c0}\omega \omega )`$ $`=`$ $`(2.3\pm 1.1)10^3,\mathrm{Br}(\chi _{c2}\omega \omega )=(6\pm 3)10^3,`$
where the errors are caused by the errors in distribution functions momenta. In the massless quark approximation (that is using $`f_{}=f`$, $`a_2^{}=a_2^{}=0`$) we get
$`\mathrm{Br}(\chi _{c0}\omega \omega )`$ $`=`$ $`1.110^3,\mathrm{Br}(\chi _{c2}\omega \omega )=5.710^3.`$
These results should be compared with the experimental values
$`\mathrm{Br}(\chi _{c0}\omega \omega )`$ $`=`$ $`(2.29\pm 0.58\pm 0.41)10^3,\mathrm{Br}(\chi _{c2}\omega \omega )=(1.77\pm 0.47\pm 0.36)10^3.`$
As it can be easily seen, the branching fractions of the $`\chi _{c0,2}\omega \omega `$ decays strongly depend on the choice of the meson structure functions and more precise values of their parameters is important. For example, the 50% error in $`\chi _{c2}\omega \omega `$ branching fraction is mainly caused by the same errors in $`a_2^,`$.
Our prediction for the $`\chi _{c0}\omega \omega `$ branching fraction is in excellent agreement with the experimental value. The $`\mathrm{Br}(\chi _{c2}\omega \omega )`$ result, on the contrary, differs significantly from the experiment. The reasons for such discrepantly could be poor knowledge of meson distribution functions or the contribution of the color-octet states that were neglected in this note.
###### Acknowledgements.
The author thanks A.K. Likhoded and V.V. Braguta for useful discussions. This work was partially supported by Russian Foundation of Basic Research under grant 04-02-17530, Russian Education Ministry grant E02-31-96, CRDF grant MO-011-0, Scientific School grant SS-1303.2003.2.
|
warning/0506/quant-ph0506106.html
|
ar5iv
|
text
|
# Measurement of the electric polarizability of lithium by atom interferometry
## Abstract
We have built an atom interferometer and, by applying an electric field on one of the two interfering beams, we have measured the static electric polarizability of lithium $`\alpha =(24.33\pm 0.16)\times 10^{30}`$ m<sup>3</sup> with a $`0.66`$% uncertainty. Our experiment is similar to an experiment done on sodium in 1995 by D. Pritchard and co-workers, with several improvements: the electric field can be calculated analytically and our phase measurements are very accurate. This experiment illustrates the extreme sensitivity of atom interferometry: when the atom enters the electric field, its velocity increases and the fractional change, equal to $`4\times 10^9`$ for our largest field, is measured with a $`10^3`$ accuracy.
An atom interferometer is the ideal tool to measure any weak modification of the atom propagation due to electromagnetic or inertial fields. The application of a static electric field is particularly interesting because it gives access to the electric polarizability $`\alpha `$ and this quantity cannot be measured by spectroscopy which is sensitive only to polarizability differences (for a review on polarizability measurements, see reference bonin97 ).
Several experiments with atom interferometers have exhibited a sensitivity to the electric electric field shimizu92 ; nowak98 ; nowak99 without aiming at a polarizability measurement and interferometers using an inelastic diffraction process have been used to measure the difference of polarizability between the ground state and an excited state rieger93 ; morinaga96 . A very accurate measurement of the atom polarizability $`\alpha `$ requires that a well-defined electric field is applied on only one interfering beam and, up-to-now, such an experiment has been made only by D. Pritchard et al. ekstrom95 ; roberts04 by inserting a thin electrode, a septum, between the two atomic paths. We have made a similar experiment with our lithium atom interferometer, represented in figure 1 and we are going to describe its first results. With respect to the experiment of D. Pritchard et al., we have made several improvements: we have designed a capacitor with an analytically calculable electric field; we have a better phase sensitivity; finally our interferometer based on laser diffraction is species selective. Our experimental accuracy is presently limited by the knowledge of the mean atom velocity.
When an electric field $`E`$ is applied, the ground state energy decreases by the polarizability term $`U=2\pi ฯต_0\alpha E^2`$. Therefore, when an atom enters the electric field, its kinetic energy increases and its wave vector $`k`$ becomes $`k+\mathrm{\Delta }k`$, with $`\mathrm{\Delta }k=2\pi ฯต_0\alpha E^2m/(\mathrm{}k)`$. The resulting phase shift $`\varphi `$ of the atomic wave is given by:
$$\varphi =\frac{2\pi ฯต_0\alpha }{\mathrm{}v}E^2(z)๐z$$
(1)
$`v=\mathrm{}k/m`$ is the atom velocity and the spatial dependence of the electric field along the atomic path is taken into account.
To know precisely the electric field along the atomic path, guard electrodes are needed, as discussed in ekstrom95 . We have developed a capacitor where guard electrodes are in the plane of the high voltage electrode, as shown in figure 2, which defines the notations. In this case, the field can be expressed analytically from the potential distribution $`V(z,x=h)`$ in the plane of the high-voltage electrode. We give here only the results of the calculation which will be published elsewhere miffre05a .
The integral of $`E^2`$ along the septum surface can be written :
$$E(z,0)^2๐z=\left[\frac{V_0}{h}\right]^2L_{eff}$$
(2)
$`V_0/h`$ is the electric field of an infinitely long capacitor and the capacitor effective length $`L_{eff}`$ is given by:
$$L_{eff}2a(2h/\pi )$$
(3)
where exponentially small corrections of the order of $`\mathrm{exp}(2\pi a/h)`$ are neglected. The atoms do not sample the electric field on the septum surface but at a small distance $`x`$ from the septum and we should add to the effective length a small correction proportional to $`x^2`$. In our experiment, with $`x50`$ $`\mu `$m and $`h2`$ mm, this correction is below $`10^4L_{eff}`$ and negligible.
The capacitor external electrodes are made of thick glass plates covered by an aluminium layer. The guard electrodes are insulated from the high voltage electrode by $`100`$ $`\mu `$m wide gaps which have been made by laser evaporation and, under vacuum, we can operate the capacitor up to $`V=450`$ V. The glass spacers are glued on the external electrodes and the septum, made of a $`6`$ $`\mu `$m thick mylar foil aluminized on both faces, is stretched and glued on the electrode-spacer assemblies. In our calculation, we assume that the potential on the high-voltage electrode is known everywhere but we ignore the potential inside the $`100`$ $`\mu `$m wide dielectric gaps which may get charged. A superiority of our design is that these gaps are very narrow, thus minimizing the uncertainty on the capacitor effective length. Another defect is that the spacer thicknesses are not perfectly constant. We use equation (2) by replacing $`h`$ by its mean value $`h`$, thus making a relative error of the order of $`(hh)^2/h^2`$ which is fully negligible.
We have previously described our three-grating Mach-Zehnder atom interferometer delhuille02a ; miffre05 . The lithium atomic beam is a supersonic beam seeded in argon and we use Bragg diffraction on laser standing waves at $`\lambda =671`$ nm. By choosing a laser detuned by about $`3`$ GHz on the blue side of the $`{}_{}{}^{2}S_{1/2}^{}`$ \- $`{}_{}{}^{2}P_{3/2}^{}`$ transition of the $`{}_{}{}^{7}Li`$ isotope, the signal is almost purely due to this isotope (natural abundance $`92.4`$%) and not to the other isotope $`{}_{}{}^{6}Li`$. Any other species present in the beam, for instance lithium dimers or heavier alkali atoms, is not diffracted and does not contribute to the signal. In three-grating interferometers, the phase of the interference fringes depends on the $`x`$-position of the gratings depending themselves on the position $`x_i`$ of the mirrors $`M_i`$ forming the three laser standing waves and this phase is given by $`\psi =2pk_L(x_1+x_32x_2)`$, where $`k_L`$ is the laser wavevector and $`p`$ is the diffraction order. By scanning the position $`x_3`$ of mirror $`M_3`$, we have observed interference fringes with an excellent visibility $`๐ฑ`$, up to $`84.5`$%.
The capacitor is placed just before the second laser standing wave, with the septum between the two atomic beams. In the present work, we have used only the diffraction order $`p=1`$ so that the center of the two beams are separated by about $`90`$ $`\mu `$m in the capacitor. When the septum is inserted between the two atomic paths, the atom propagation is almost not affected and we observe interference fringes with a visibility $`๐ฑ=84`$ % and a negligible reduction of the atomic flux. To optimize the phase sensitivity, we have opened the collimation slit $`S_1`$ and the detection slit $`S_D`$ (see reference miffre05 ) with widths $`e_1=18`$ $`\mu `$m and $`e_D=50`$ $`\mu `$m, thus increasing the mean flux up to $`10^5`$ counts/s and slightly reducing the fringe visibility down to $`๐ฑ_0=62`$% (see figure 3). We have made a series of recordings, labelled by an index $`i`$ from $`1`$ to $`44`$, with $`V_0=0`$ when $`i`$ is odd and with $`V_00`$ when $`i`$ is even with $`V_010\times i`$ Volts. For each recording, we apply a linear ramp on the piezo-drive of mirror $`M_3`$ in order to observe interference fringes and $`471`$ data points are recorded with a counting time per channel equal to $`0.36`$ s. Figure 3 presents a pair of consecutive recordings. The high voltage power supply has stability close to $`10^4`$ and the applied voltage is measured by a HP model 34401A voltmeter with a relative accuracy better than $`10^5`$.
The data points $`I_i(n)`$ have been fitted by a function $`I_i(n)=I_{0i}\left[1+๐ฑ_i\mathrm{cos}\psi _i(n)\right]`$, with $`\psi (n)=a_i+b_in+c_in^2`$ where $`n`$ labels the channel number, $`a_i`$ represents the initial phase of the pattern, $`b_i`$ an ideal linear ramp and $`c_i`$ the non-linearity of the piezo-drive. For the $`V=0`$ recordings, $`a_i`$, $`b_i`$ and $`c_i`$ have been adjusted as well as the mean intensity $`I_{0i}`$ and the visibility $`๐ฑ_i`$, while, for the $`V0`$ recording, we have fitted only $`a_i`$, $`I_{0i}`$ and $`๐ฑ_i`$, while fixing $`b_i`$ and $`c_i`$ to their value $`b_{i1}`$ and $`c_{i1}`$ from the previous $`V=0`$ recording. Our best phase measurements are given by the mean phase $`\psi _i`$ obtained by averaging $`\psi _i(n)`$ over the $`471`$ channels. The $`1\sigma `$ error bar of these mean phases are of the order of $`23`$ mrad, increasing with the applied voltage up to $`23`$ mrad because of the reduced visibility.
The mean phase values $`\psi _i`$ values of the $`V_0=0`$ recordings present a drift equal to $`7.5\pm 0.2`$ mrad/minute and some scatter around this regular drift. The drift is explained by the differential thermal expansion of the structure supporting the three mirrors: its temperature was steadily drifting at $`1.17\times 10^3`$ K/minute during the experiment. We have no explanation of the phase scatter, which presents a quasi-periodic structure: its rms value is equal to $`33`$ milliradian and unfortunately this error dominates our phase determination.
The phase shift $`\varphi (V_0)`$ due to the polarizability effect is taken equal to $`\varphi (V_0)=\psi _i\left(\psi _{i1}+\psi _{i+1}\right)/2`$ where the recording $`i`$ corresponds to the applied voltage $`V_0`$: the average of the mean phase of the two $`V_0=0`$ recordings done just before and after is our best estimator of the mean phase of the interference signal in zero field. In figures 4 and 5, we have plotted the phase shift $`\varphi (V_0)`$ and the fringe visibility $`๐ฑ`$ as a function of the applied voltage $`V_0`$.
To interpret these results, we must take into account the velocity distribution of the lithium atoms, as the phase shift is proportional to $`v^1`$. We assume that the velocity distribution is given by:
$$P(v)=\frac{S_{}}{u\sqrt{\pi }}\mathrm{exp}\left[\left((vu)S_{}/u\right)^2\right]$$
(4)
with the most probable velocity $`u`$ and $`S_{}`$ is the parallel speed ratio. The traditional $`v^3`$ pre-factor haberland85 , which has minor effects, is omitted when $`S_{}`$ is large. The interference signals $`I`$ can be written:
$$I=I_0๐vP(v)\left[1+๐ฑ_0\mathrm{cos}\left(\psi +\varphi _m\frac{u}{v}\right)\right]$$
(5)
with $`\varphi _m=\varphi (u)`$. If we expand $`u/v`$ in powers of $`(vu)/u`$ up to second order, the integral can be calculated exactly. This approximation is very good miffre05a but not accurate enough and we have calculated the integral (5) numerically. We have thus made a single fit for the phase and visiblility results, with two adjustable parameters: $`\varphi _m/V_0^2`$ and $`S_{}`$. As shown in figures 4 and 5, the agreement is very good, in particular for $`\varphi `$, and we deduce a very accurate value $`\varphi _m/V_0^2`$:
$$\frac{\varphi _m}{V_0^2}=\frac{2\pi ฯต_0\alpha L_{eff}}{\mathrm{}uh^2}=(1.3870\pm 0.0010)\times 10^4\text{ rad/V}^2$$
(6)
The relative uncertainty $`0.07`$% proves the coherence of our measurements. The parallel speed ratio $`S_{}=8.00\pm 0.06`$ is slightly larger than expected for our lithium beam, because Bragg diffraction is velocity selective.
From measurements made on our capacitor, we get $`2a=50.00\pm 0.10`$ mm and $`h=2.056\pm 0.003`$ mm. We have measured the mean velocity $`u`$ using Doppler effect, by recording atom deflection due to photon recoil with a laser beam almost counterpropagating with the atoms. The uncertainty on the cosine of the angle is negligible ($`0.12`$%) and we get $`u=1066.4\pm 8.0`$ m/s. We have also recorded the diffraction probability as a function of the Bragg angle, by tilting the mirror forming a standing wave. Using an independent calibration of the mirror rotation as a function of the applied voltage on the piezo-drive, we get a measurement of the Bragg angle $`\theta _B=h/(mu\lambda _L)=79.62\pm 0.63`$ $`\mu `$rad corresponding to $`u=1065.0\pm 8.4`$ m/s. These two values are perfectly coherent and we take the mean velocity as their weighted average $`u=1065.7\pm 5.8`$ m/s. The theory of supersonic expansion can be used to check this result: the velocity of a pure argon beam given by $`u=\sqrt{5k_BT_0/m}`$ (where $`T_0=1073\pm 11`$ K is the nozzle temperature and $`m`$ the argon atomic mass) must be corrected: the dominant correction is the velocity slip effect estimated to be $`1`$% skovorodko04a and we get $`u=1068.4\pm 5.5`$ m/s in very good agreement with our measurements.
Finally, we get the lithium electric polarizability of <sup>7</sup>Li $`\alpha =(24.33\pm 0.16)\times 10^{30}`$ m<sup>3</sup>, in excellent agreement with the previous measurements, $`\alpha =(22.\pm 2.)\times 10^{30}`$ m<sup>3</sup> by Chamberlain and Zorn chamberlain63 in 1963 and $`\alpha =(24.3\pm 0.5)\times 10^{30}`$ m<sup>3</sup>, by Bederson and co-workers molof74 in 1974. Our result compares also very well with ab initio calculations of $`\alpha `$: most calculations predict $`\alpha `$ values in the range $`(24.3224.45)\times 10^{30}`$ m<sup>3</sup> (see reference kobayashi97 and references therein).
With respect to the experiment done on sodium by D. Pritchard and co-workers ekstrom95 , we have made several important improvements:
Our capacitor design provides an analytical calculation of the $`E^2`$ integral along the atomic path. This property is helpful in minimizing the uncertainty on this quantity, through a better understanding of the influence of small defects. With an improved construction, we expect to reduce the uncertainty on this integral below $`0.1`$ %, the main limitation being due to the unknown potential in the dielectric gaps.
Thanks to a large signal and an excellent fringe visibility, the phase sensitivity of our interferometer is considerably larger than previously achieved. The accuracy on phase measurement is presently limited by the lack of reproducibility of the mean phase of the recordings. We hope to improve this reproducibility by stabilizing the temperature of the rail supporting the three mirrors. The consistency and accuracy of our phase measurements is proved by the quality of the fit of figure 4 and by the $`0.07`$% uncertainty obtained for the measurement of $`\varphi _m/V_0^2`$. We have deduced the value of the electric polarizability $`\alpha `$ with a $`0.66`$% relative uncertainty dominated by the uncertainty on the mean atom velocity $`u`$.
Our interferometer is species selective thanks to laser diffraction and this is also a very favorable circumstance. In his thesis, T. D. Roberts reanalyzes the measurement of sodium atom electric polarizability made by C. R. Ekstrom et al. ekstrom95 : he estimates that a weak contribution of sodium dimer to the interference signals might have introduced a non negligible systematic error in the result.
T. D. Roberts et al. roberts04 have devised a very clever technique to correct for the velocity dependence of the phase shift $`\mathrm{\Delta }\varphi `$, so that they can observe fringes with a good visibility up to very large $`\varphi `$ values. The present result proves that a very accurate measurement can be also made in the presence of an important velocity dispersion without any compensation of the associated phase dispersion, provided that the velocity distribution is taken into account in the analysis.
Finally, we would like to emphasize two very striking properties of atom interferometry. Our phase measurement consists in measuring the increase $`\mathrm{\Delta }v`$ of the atom velocity $`v`$ when entering the field:
$$\frac{\mathrm{\Delta }v}{v}=\frac{\lambda _{dB}}{L_{eff}}\times \frac{\varphi }{2\pi }$$
(7)
$`\mathrm{\Delta }v/v`$ is extremely small reaching only $`\mathrm{\Delta }v/v4\times 10^9`$ for our largest field. Our ultimate sensitivity, close to a $`3`$ mrad phase shift, corresponds to $`\mathrm{\Delta }v/v6\times 10^{13}`$!
In the capacitor, the atom wavefunction samples two regions of space separated by $`100`$ $`\mu `$m with a macroscopic object lying in between and this situation extends over $`10^4`$ second, without any loss of coherence. This consequence of quantum mechanics remains surprising!
We thank J. C. Lehmann and J. F. Arribart from Saint-Gobain, J. F. Bobo and M. Nardone from LPMC and the staff of the AIME for their help in the construction of the capacitor. We thank CNRS SPM and Rรฉgion Midi Pyrรฉnรฉes for financial support.
|
warning/0506/cs0506044.html
|
ar5iv
|
text
|
# Minimal Network Coding for Multicast
## I Introduction
In their seminal work, Ahlswede et al showed that if the nodes in the network are allowed to perform network coding rather than just routing then the max flow min cut bound on the multicast capacity is achievable. Li et al showed that linear codes are sufficient to achieve the multicast capacity. Since then several techniques have been proposed to design codes that achieve the multicast capacity. Among them, the idea of random network coding seems very promising. Ho et al propose a scheme in which data is collected in the form of packets of, say, length $`n`$. These packets are then treated as elements of a finite field of size $`q=2^n`$ (assuming that the data is in bits) and they show that if the messages on the outgoing edges of every node are set to be a random linear combination of the messages received along the incoming edges over a finite field of size $`q`$ then the probability that the resulting code is not a valid multicast code is $`O(1/q)`$. (We call a multicast valid if the destination nodes can decode the data.) Therefore a valid multicast code can be designed with very high probability by random coding over a large field.
Random network coding by itself could be inefficient in terms of network resources. Since the scheme is completely distributed and there is no communication between the nodes, each node sends messages on all its outgoing edges in the process using up all the available bandwidth. But, this problem can be solved. In Lun et al proposed a distributed algorithm which can be used, for example, to find a sub network that minimizes link usage costs while having the same multicast capacity as the given network. Random network coding can be employed on this sub network to achieve the multicast capacity.
In general, minimal cost network coding solutions are of practical interest. The cost to be minimized may depend on the network and application at hand. For example, if a router that employs network coding is expensive we will want to minimize the number of nodes that perform network coding. In optical networks the operation of computing linear combination of inputs may require conversion from optical signals to electrical signals which is expensive and hence we may want to minimize the number of packets that undergo network coding. Random network coding as such would result in schemes where every node performs network coding. In this paper, we will address the problem of minimal cost network coding where the cost is the number of packets that need to be network coded. We also consider the problem of finding minimum cost solutions when some of the nodes are restricted to perform only routing. The multicast capacity for a special case of this problem when all the nodes only route has been studied in . We will refer to nodes employing network coding by network coding nodes and nodes restricted to routing by routing nodes.
In , the authors consider costs such as bandwidth and delay and investigate minimum cost multicast. However, the results in cannot be directly used to solve the problems considered in this paper because the fluid model used to represent flows to individual receivers cannot be used when some of the nodes are restricted to routing. It is also not possible to differentiate between the operations of network coding and routing at a node by only looking at the input and output flows of that node. The main contribution of this paper is to give a new information flow based interpretation for the multicast flow and use this model to set up optimization problems that can be solved in a distributed manner.
The optimization problem formulated in this paper has a complexity that grows exponentially with the number of receivers but in many applications like video conferencing the number of receivers is quite small and hence these algorithms can be of practical use.
In section II we give the notation used in the paper. We present the new information flow model in section III. In section IV we set up the optimization problems and finally conclude in section V.
## II Notation
We represent a network by a directed graph $`๐ข=(๐ฑ,)`$, where $`๐ฑ`$ is the set of vertices (nodes) and $``$ is the set of edges (links). The capacity of edge $`e`$ is given by $`C(e)`$. For each node $`v๐ฑ`$ we define sets $`E_{in}(v)`$ and $`E_{out}(v)`$ as the set of all edges that come into $`v`$ and that go out of $`v`$ respectively.
We consider a multicast problem with one source $`S๐ฑ`$ and $`K`$ receivers in the set $`D๐ฑ`$. We assume that $`D=\{1,2,\mathrm{},K\}`$. For convenience we define two sets $`๐ซ`$ and $`๐ฌ`$ where $`๐ซ`$ is the power set of $`D`$ (neglecting the empty set) and $`๐ฌ`$ is a set containing all collections of two or more disjoint sets in $`๐ซ`$. For example, when $`K=3`$, $`๐ซ=\{\{1\},`$ $`\{2\},\{3\},\{1,2\},\{1,3\},\{2,3\},\{1,2,3\}\}`$ and $`๐ฌ=\{\{\{1\},\{2\}\},\{\{1\},\{3\}\},\{\{2\},\{3\}\},\{\{1\},\{2\},\{3\}\},\{\{1\},\{2,3\}\},\{\{2\},\{1,3\}\},\{\{3\},\{1,2\}\}\}`$. We fix the ordering in $`๐ซ`$ and $`๐ฌ`$ and represent the $`i`$-th element in $`๐ซ`$ and the $`j`$-th element in $`๐ฌ`$ by $`P_i`$ and $`Q_j`$ respectively.
With each edge $`e`$ and a coding scheme we associate a $`2^K1`$ length information flow vector $`X_e`$ where the $`i`$-th element denoted by $`x_e(P_i)`$ represents the amount of information common to and only common to receivers in the set $`P_i`$ that flows through the edge $`e`$. We define $`I_k(X_e)=_{i:kP_i}x_e(P_i)`$ as the amount of flow along edge $`e`$ in the flow decomposition of receiver $`k`$. The definitions will be made precise in Section III.
It is sometimes convenient to assume that the edges capacities and the flow vectors are integers. This assumption is justified since we can always consider the network over multiple time instances.
## III Information Flow
In a multicast setup, any multicast solution can be decomposed into flows to individual receivers , . The flows to different receivers could overlap. Overlapping flows indicates that the data sent along the overlapping part of the flow has to be eventually conveyed to all the receivers whose flows overlap.
The main idea here is to partition the flows to the individual receivers as components of the form $`x_e(P_i)`$. We formally do this in the rest of the section. We define $`x_e(\{k_1,k_2,\mathrm{},k_j\})`$ for an edge $`e`$ as the amount of overlap in the flows along $`e`$ from the source node to the receiver nodes $`k_1,k_2,\mathrm{},k_j`$ and that does not overlap with any other flow for any other receiver. To identify the overlapping flows, consider a network obtained by expanding the original network by replacing each edge $`e`$ by parallel edges, $`e_1^{},\mathrm{},e_{C(e)}^{}`$, of unit capacity (assuming edge capacities are integers). The expanded network also supports the same rate (h, also assumed to be an integer) as the original network and hence $`h`$ edge disjoint paths from source to receiver $`k`$ for each $`k`$ can be found , . The paths to the different receivers could have overlapping edges. For an edge $`e^{}`$ in the expanded network, let $`P_i`$ be the set of all receivers that have edge $`e^{}`$ in one of their paths. The element $`x_e^{}(P)`$ in the information flow vector for $`e^{}`$ is then $`1`$ for $`P=P_i`$ and $`0`$ otherwise. If no paths pass through $`e^{}`$ its information flow vector is zero. The information flow vector for the edge $`e`$ in the original network is the sum of the information flow vectors of the parallel edges $`e_1^{},\mathrm{},e_{C(e)}^{}`$.
To keep the notation brief we also use $`x_e(k_1,k_2,\mathrm{},k_j)`$ with $`k_1<k_2<\mathrm{}<k_j`$ to represent $`x_e(\{k_1,k_2,\mathrm{},k_j\})`$. It is east to see that the flow to receiver $`k`$ along edge $`e`$ is given by $`_{i:kP_i}x_e(P_i)`$. Since this is a function of $`X_e`$ for each $`k`$, we represent it by $`I_k(X_e)`$. We show the flow vector and information flow vector for some multicast networks in Example 1.
Example 1: Consider the network shown in Fig. 1a. A code that achieves the multicast capacity is shown in Fig. 1a. The flows to the two receivers are shown in Fig. 1b. In Fig. 1c the information flow vector for each edge is shown. The information flow vector is (1,0,0) when the edge carries data at unit rate only for receiver 1, is (0,1,0) when the edge carries data at unit rate for receiver 2 and (0,0,1) when the edge carries data at unit rate meant for both the receivers. In Fig. 1d we show a code over two time instances that achieves the routing capacity of the network and in Fig. 1e we show the corresponding flows. We note that the edge between node 4 and node 3 has flows for both the receivers but they are not overlapping flows. It is easy to verify in Fig. 1b, 1c, 1e, and 1f that $`I_1(X_e)`$ and $`I_2(X_e)`$ gives the amount of flow along edge e to receiver 1 and 2 respectively.
The amount of data flowing along an edge $`e`$ is the sum of the elements of $`X_e`$. Since each edge has a capacity constraint, we have the following constraint on $`X_e`$.
$$\text{sum}(X_e)=๐^๐X_eC(e)e$$
(1)
where $`๐`$ is the all one vector of length $`2^K1`$. We denote the constraint in (1) as the edge constraint.
###### Theorem 1
In any multicast network that supports a rate $`h`$, we can find $`X_e`$ for every edge $`e`$ satisfying the edge constraint such that
$`{\displaystyle \underset{eE_{in}(v)}{}}I_k(X_e)`$ $`=`$ $`{\displaystyle \underset{eE_{out}(v)}{}}I_k(X_e)k,v๐ฑ\{S,k\}`$
$`{\displaystyle \underset{eE_{out}(S)}{}}I_k(X_e)`$ $`=`$ $`{\displaystyle \underset{eE_{in}(k)}{}}I_k(X_e)=hk`$ (2)
###### Proof:
It is possible to decompose any multicast code into $`h`$ flows to individual receivers . Consider the $`X_e`$โs and $`I_k(X_e)`$โs corresponding to one such flow decomposition. $`I_k(X_e)`$ is the amount of flow along edge $`e`$ in the flow decomposition of receiver $`k`$. The equations in (2) claim that in the flow decomposition of receiver $`k`$, the flow coming into any intermediate node is equal to the flow coming out of the node, the flow coming out of the source node is $`h`$, and, the flow into receiver $`k`$ is $`h`$. These are well known properties of the flow decomposition . โ
It is convenient to define $`X_{in}^v`$ and $`X_{out}^v`$ for node $`v๐ฑ`$ as $`_{eE_{in}(v)}X_e`$ and $`_{eE_{out}(v)}X_e`$ respectively. To keep the notation brief we will drop the superscript $`v`$ in discussions involving just one node. Since $`I`$ is a linear function of $`X`$, the conditions in (2) reduce to
$`I_k(X_{in}^v)`$ $`=`$ $`I_k(X_{out}^v)k,v๐ฑ\{S,k\}`$
$`I_k(X_{out}^S)`$ $`=`$ $`I_k(X_{in}^k)=hk`$ (3)
We will call the necessary conditions in (III) as flow constraints. In Fig. 1 we can easily see that the edge and the flow constraints are satisfied.
### III-A Routing, Replicating and Network Coding
Let us take a closer look at the different operations that occur in a node in a multicast network. In Fig. 1 we see that there are three different operations that happen at a node. The first and simplest operation is when a packet is routed to one of the output edges. The second type of operation is replication in which multiple copies of the packet are sent along different edges. The third operation, network coding, refers to the case when two or more packets are combined into one packet. We will see that these three operations are sufficient to represent any necessary processing being done at the node but before that we need to understand what the different operations represent.
We first look at routing and replication. Each packet that comes into a node has an associated set of receivers $`QD`$. The packet has to eventually reach each node in $`Q`$. When it gets routed onto one of the output edges, the packet on the output edge still has to reach all nodes in $`Q`$. In terms of the information flows to the various receivers, this corresponds to the case when overlapping flows or a simple flow passes through a node and continues unaffected.
When a packet gets replicated, then each copy of the packet on the output edge has to reach nodes in $`P_i`$ a subset of $`Q`$. ($`P_i`$ has to be a subset of $`Q`$ since the packet has to reach only nodes in $`Q`$.) Since the packet has to reach all nodes in $`Q`$ we have $`P_i=Q`$. Moreover, the same packet does not need to reach the same destination along two different paths. Therefore the $`P_i^{}s`$ are disjoint $`P_{i_1}P_{i_2}=\varphi i_1i_2`$. In the flow decomposition replication corresponds to the point where two or more overlapping flows diverge.
For example, consider the node 3 in Fig. 1. The incoming packet has to be sent to both node 1 and node 2. $`X_{in}^3=[0,0,1]`$. The node replicates it and forwards it to two edges. Along one of the edges that packet reaches node 1 $`(X_{e(3,1)}=[1,0,0])`$ and the packet sent on the other edge is meant for node 2 $`(X_{e(3,2)}=[0,1,0])`$. At the output $`X_{out}^3=[1,1,0]`$.
This concept becomes clearer when we look at the relationship between $`X_{in}`$ and $`X_{out}`$. We consider the case for two and three destinations and then generalize the results. When there are two destinations replication occurs only when a packet meant for both destinations is replicated and sent on two different paths, one path for each receiver node. $`x_{in}(1,2)`$ represents the average number of packets coming in per unit time that need to go to both $`1`$ and $`2`$. If $`r(r0)`$ of these packets are duplicated and transmitted per unit time we have
$`x_{out}(1)`$ $`=`$ $`x_{in}(1)+r`$
$`x_{out}(2)`$ $`=`$ $`x_{in}(2)+r`$ (4)
$`x_{out}(1,2)`$ $`=`$ $`x_{in}(1,2)r`$
Now consider the case with three destinations. Similar to the two receiver case, a packet meant for two destinations can get replicated to produce two packets for the two destinations. Let $`r_1`$, $`r_2`$ and $`r_3`$ represent the amount of replication corresponding to flows to receiver sets $`\{\{1\},\{2\}\}`$, $`\{\{1\},\{3\}\}`$, and $`\{\{2\},\{3\}\}`$ respectively. When packets meant for all three receivers replicate, they split the flow in four possible ways {{1},{2},{3}}, {{1},{2,3}}, {{2},{1,3}} and {{3},{1,2}}. Let $`r_4`$, $`r_5`$, $`r_6`$ and $`r_7`$ represent the number of packets replicated per unit time corresponding to the four cases. The relation between the $`X_{in}`$ and $`X_{out}`$ is therefore given by
$`x_{out}(1)`$ $`=`$ $`x_{in}(1)+r_1+r_2+r_4+r_5`$
$`x_{out}(2)`$ $`=`$ $`x_{in}(2)+r_1+r_3+r_4+r_6`$
$`x_{out}(3)`$ $`=`$ $`x_{in}(3)+r_2+r_3+r_4+r_7`$
$`x_{out}(1,2)`$ $`=`$ $`x_{in}(1,2)r_1+r_7`$ (5)
$`x_{out}(1,3)`$ $`=`$ $`x_{in}(1,3)r_2+r_6`$
$`x_{out}(2,3)`$ $`=`$ $`x_{in}(2,3)r_3+r_5`$
$`x_{out}(1,2,3)`$ $`=`$ $`x_{in}(1,2,3)r_4r_5r_6r_7`$
We note that each of the $`r`$โs are $`0`$. Moreover, if all the $`r`$โs equal 0 then only routing is performed at a node. In the general case we will have a routing variable $`r_j`$ associated with every set $`Q_j`$ corresponding to flow for receivers in the set $`_{QQ_j}Q`$ being replicated with each copy meant for a set in $`Q_j`$. We denote the set of routing variables $`r_j`$โs by $`R`$. The general equation is
$$x_{out}(P_i)=x_{in}(P_i)+\underset{j:P_iQ_j}{}r_j\underset{j:_{QQ_j}Q=P_i}{}r_j$$
(6)
Any node that is restricted to routing/replicating has to satisfy (6). We will call this constraint on $`X_{in}`$ and $`X_{out}`$ as routing constraint. Note that although we call the variable $`r_j`$โs as routing variables they actually correspond to replication. Also when we say a node is a routing node we allow for replication at that node.
The third type of operation is network coding. This happens at nodes where two or more flows merge. Similar to the routing variables we define a set of network coding variables $`N`$ where element $`n_j`$ represents the amount of flow meant for each set of receivers $`QQ_j`$ that merges to form one $`n_j`$ flow that has to reach all receivers in the set $`_{QQ_j}Q`$. $`n_j|Q_j|`$ packets are network coded to form $`n_j`$ packets. It is easy to see that for a network coding node the relationship between $`X_{in}`$ and $`X_{out}`$ has to be of the form
$`x_{out}(P_i)=x_{in}(P_i)+{\displaystyle \underset{j:P_iQ_j}{}}r_j{\displaystyle \underset{j:_{QQ_j}Q=P_i}{}}r_j`$
$`{\displaystyle \underset{j:P_iQ_j}{}}n_j+{\displaystyle \underset{j:_{QQ_j}Q=P_i}{}}n_j`$ (7)
which reduces to
$$x_{out}(P_i)=x_{in}(P_i)+\underset{j:P_iQ_j}{}(r_jn_j)\underset{j:_{QQ_j}Q=P_i}{}(r_jn_j)$$
(8)
We note that it is sufficient to consider variables $`r_jn_j`$ but we retain both for now. We will refer to the conditions in (8) as node constraints.
In the following theorem, for any pair of $`X_{in}`$ and $`X_{out}`$ that satisfy the flow constraints, we show that the operations at the node can be decomposed into routing, replicating and network coding operations and hence these operations are sufficient to represent any processing done at the node.
###### Theorem 2
The relationship between $`X_{in}`$ and $`X_{out}`$ for any valid operation at the node can be expressed in terms of routing variables $`R`$ and network coding variables $`N`$ such that each element of $`R`$ and $`N`$ is $`0`$ .
###### Proof:
We will give a particular solution satisfying all the conditions. The main idea used in constructing the particular solution is that all packets meant for more than one receiver can be replicated to produce packets such that each packet is meant for one receiver. They can then be suitably network coded to get the desired output information flow vector.
Consider a set of receivers $`P_i`$ and corresponding set $`Q(P_i)`$ the set of all singleton subsets of $`P_i`$. For every set $`P_i๐ซ`$ containing two or more elements set $`r_j=x_{in}(P_i)`$ and $`n_j=x_{out}(P_i)`$ where $`Q_j=Q(P_i)`$. Set all other routing and network coding variables to 0. We will show that this solution satisfies the constraints in (8). On substituting for the routing and network coding variables that have been set to 0, for all non singleton sets $`P_i`$ the constraints in (8) reduce to $`x_{out}(P_i)=x_{in}(P_i)r_j+n_j`$, $`Q_j=Q(P_i)`$ which is satisfied by choice of $`r_j`$ and $`n_j`$. For singleton sets $`P=\{k\}`$ we have
$`x_{out}(k)`$ $`=`$ $`x_{in}(k)+{\displaystyle \underset{j:\{k\}Q_j}{}}(r_jn_j)`$
$`=`$ $`x_{in}(k)+{\displaystyle \underset{i:\{k\}Q_j=Q(P_i),|P_i|>1}{}}(r_jn_j)`$
$`=`$ $`x_{in}(k)+{\displaystyle \underset{i:kP_i,|P_i|>1}{}}x_{in}(P_i)x_{out}(P_i)`$
which is exactly the flow constraint on information flow to the receiver $`k`$ (Eq. III) and hence is satisfied. โ
###### Theorem 3
Given a network $`๐ข=(๐ฑ,)`$, flow vectors $`X_e`$ for each edge $`e`$ and routing and network coding variables $`R^v`$ and $`N^v`$ for each node $`v๐ฑ`$ such that the edge, flow and node constraints are satisfied, we can construct a valid multicast code that performs routing and network coding as specified by $`R^v`$ and $`N^v`$.
###### Proof:
We prove the theorem by replacing each node in the network by a network that has routing and network coding nodes corresponding to the variables $`R^v`$ and $`N^v`$ such that there is no loss in the multicast rate.
With every node $`v๐ฑ`$ associate a set of $`(2^k1)`$ nodes where each new node, $`v(P_i)`$, corresponds to one set of receivers $`P_i๐ซ`$. For every set $`P_i๐ซ`$ connect all the $`x_{in}^v(P_i)`$ incoming edges and the $`x_{out}^v(P_i)`$ outgoing edges of node $`v`$ carrying data for receivers in and only in set $`P_i`$ as input and output edges to the node $`v(P_i)`$.
Corresponding to each non zero routing variable $`r_j^v`$ construct $`r_j^v`$ nodes, each node having exactly one incoming edge coming from node $`v(_{P_iQ_j}P_i)`$ and $`|Q_j|`$ outgoing edges that are connected as inputs to nodes in $`\{v(P_i):P_i๐ฌ_๐ฟ\}`$. Corresponding to each non zero network coding variable $`n_j^v`$ construct $`n_j^v`$ nodes with each node having one input edge from every node in $`\{v(P_i):P_iQ_j\}`$ and one output edge that is connected as input to node $`v(_{P_iQ_j}P_i)`$.
Now the number of incoming edges to node $`v(P_i)`$ is $`x_{in}^v(P_i)+_{j:P_iQ_j}r_j^v+_{j:_{QQ_j}Q=P_i}n_j^v`$ and the number of outgoing edges is $`x_{out}^v(P_i)+_{j:P_iQ_j}n_j^v+_{j:_{QQ_j}Q=P_i}r_j^v`$. From (8) the number of incoming edges is equal to the number of outgoing edges. Randomly connect the set of input edges and the set of output edges of node $`v(P_i)`$ in a one to one manner and delete node $`v(P_i)`$.
It is easy to see that this construction procedure replaces each node by a network that maintains the same flows and hence there is no loss in rate. โ
In the construction procedure provided in the proof for Theorem 3, the network that replaces each node could have cycles. These cycles are formed when a packet meant for a set of receivers $`P_i`$ goes through a series of network coding and routing operations to get back a packet meant for $`P_i`$ itself. Clearly the involvement of this packet in those operations is unnecessary. All cycles correspond to unnecessary operations and hence can be removed. We note that cycles within a node will be absent in solutions that minimizes the number of network coding operations. The construction procedure provided can be used along with ideas of random network coding to construct multicast codes corresponding to the given information flow vectors.
## IV Optimization
Since any solution to the set of linear equations specified by (1), (III) and (8) corresponds to a network coding solution, we can use the set of equations to obtain a network coding solution in order to minimize a โcostโ associated with the network code. The problem can be stated as follows:
minimize Cost
subject to
$`x_e(P_i)0P_iP,e,`$
$`r_j^v0,n_j^v0jv๐ฑ`$
Edge Constraints:
$`{\displaystyle \underset{P_iP}{}}x_e(P_i)C(e)e`$
Node Constraints: (9)
$`x_{out}^v(P_i)=x_{in}^v(P_i)+{\displaystyle \underset{j:P_iQ_j}{}}(r_j^vn_j^v)`$
$`{\displaystyle \underset{j:_{QQ_j}Q=P_i}{}}(r_j^vn_j^v)P_i๐ซv๐ฑ`$
$`I_k(X_{out}^S)=hk`$
$`I_k(X_{in}^k)=hk`$
where
$`x_{in}^v(P_i)={\displaystyle \underset{eE_{in}(v)}{}}x_e(P_i),x_{out}^v(P_i)={\displaystyle \underset{eE_{out}(v)}{}}x_e(P_i)`$
$`\text{and }I_k(X_e)={\displaystyle \underset{j:kP_j}{}}x_e(P_j)`$
Note that we have dropped some of the flow constraints in (III) as they are satisfied automatically if the node constraints in (8) are satisfied.
In the remainder of the section, we list a few natural cost criteria.
1. Number of Network Coding nodes. Since additional coding capabilities are required at a node in order to perform network coding, it is potentially of interest to minimize the number of nodes performing network coding. Using Theorem 3, it follows that network coding needs to be performed at a node $`v`$ only if $`n_i^v>0`$ for some $`i`$. Since $`n_i^v0`$, this condition is equivalent to $`_in_i^v>0`$. Thus, the number of nodes in the network performing network coding is $`_{v๐ฑ}I(_in_i^v>0)`$, which we choose as the cost function.
However, note that for $`n_i^v0`$, the function $`_{v๐ฑ}I(_in_i^v>0)`$ is a concave function and the problem becomes one of minimizing a concave function over a convex set. This solution might admit local minima and standard convex minimization techniques cannot be used to solve this problem. We relax this problem and investigate minimizing the number of network coding operations and minimizing the number of packets involved in network coding in the following problems.
2. Number of network coding operations. In this problem we investigate minimizing the number of network coding operations at a node $`v`$. From Theorem 3 it follows that network coding operations (linear encoding of packets) need to be performed corresponding to each $`n_i^v`$. Thus the number of network coding operations at node $`v`$ is $`_in_i^v`$. We define this quantity as the amount of network coding. Thus, the cost function in this problem is given by $`_{v๐ฑ}_in_i^v`$.
3. Number of packets involved in network coding. In this problem we investigate minimizing the number of packets over which network coding is performed at a node $`v`$. This is particularly relevant in optical networks when a conversion from optical signals to electrical signals is involved in order to encode the packets. We conjecture that the cost function is given by $`_{v๐ฑ}_i\mathrm{max}(A_i^v,0)`$ where $`A_i^v=_{j:P_iQ_j}(n_j^v\lambda _{i,j}^v)`$ $`_{j:_{QQ_j}Q=P_i}(n_j^v\mathrm{max}_{i1}\lambda _{i1,j}^v)`$. $`\lambda _{i,j}`$ represents the number of packets meant for receivers $`P_i`$ that participate in network coding and that are obtained by routing packets meant for $`_{QQ_j}Q`$ ($`P_iQ_j`$). From the definition it follows that $`0\lambda _{i,j}r_j`$.
4. Minimum resource cost. In the setup considered in , each edge $`e`$ is associated with a cost function $`f_e(z_e)`$ when the data rate on $`e`$ is $`z_e`$. The net cost associated with the network is then given by $`_ef_e(z_e)`$. This cost was minimized over the set of equations specified by equations (1) and (2) in . The same approach can be applied in the setting where only certain nodes are allowed to perform network coding. The restriction that a node $`v`$ can perform only routing can be imposed by further constraining the equations in (9) by $`n_i^v=0`$ for all $`i`$.
5. Maximum rate. The problem conidered here is one of maximizing $`h`$ constrained to (9) and additionally the set of equations $`n_i^v=0`$ for all $`i`$ and nodes $`v`$ which are restricted to routing.
Note that the problems 2, 3, 4 (if the cost function $`f_e()`$ is linear) and 5 are linear problems and can be solved by standard linear programming approaches. It remains to be investigated if the decentralized subgradient optimization suggested in can be applied to these problems. To the end of providing decentralized solutions to these linear problem, we consider the approach suggested by in which a linear function $`ax`$ is approximated by a stricly convex function $`(ax)^{1+\alpha }`$ where $`\alpha >0`$ is chosen small enough for a valid approximation. This makes the problem a convex optimization problem which can be solved in a decentralized manner by a modified version of the primal-dual algorithm used in . We do not prove this due to lack of space. The main idea in the proof is to show that the edge and node constraints involved are local in the sense of involving variables of the neighbouring edges or nodes and then follow the same steps as used in .
Problem 4 is a convex optimization problem if the function $`f_e()`$ is convex. If we further assume that the function $`f_e()`$ is strictly convex, it follows that problem 4 admits a unique solution. Further, it can be shown that the primal-dual algorithm used in can be modified to solve problem 4 in a decentralized manner. Again we do not prove this due to lack of space.
## V Conclusion
In this paper, we presented a new Information flow model to represent multicast flows. Using this model we set up optimization problems and presented distributed algorithms to minimize costs like number of packets undergoing network coding and amount of network coding. We also showed that this approach can be used to minimize network costs like link usage when some nodes are restricted to routing.
|
warning/0506/cond-mat0506334.html
|
ar5iv
|
text
|
# Pressure-induced phase transition and bi-polaronic sliding in a hole-doped Cu2O3 ladder system
## I introduction
A material of considerable recent interest is the compound Sr<sub>14-x</sub>Ca<sub>x</sub>Cu<sub>24</sub>O<sub>41</sub> (SCCO), which shows superconductivity under pressure uehara ; nagata . The SCCO structure includes quasi-one dimensional (Q1D) two-leg ladders Cu<sub>2</sub>O<sub>3</sub> and one-dimensional (1D) CuO<sub>2</sub> chains, while other high-$`T_c`$ cuprate superconductors found so far contain two-dimensional (2D) CuO<sub>2</sub> planes. The Cu<sub>2</sub>O<sub>3</sub> ladders and CuO<sub>2</sub> chains in this system are intrinsically hole-doped even at $`x=0`$ with a total of 6 holes per formula unit. (The total hole concentration can be decreased by La and Y substitution for Sr, with e.g. La<sub>6</sub>Cu<sub>8</sub>O<sub>41</sub> containing no holes.) Of these, only approximately one hole goes into the ladder component of the formula unit, (Cu<sub>2</sub>O<sub>3</sub>)<sub>7</sub>, which results in an effective doping of about 7% per Cu site in the ladder nucker ; osafune . Ca substitution, $`x`$, does not change the total number of carriers, but transfers the holes from the chains to the ladders osafune ; mizuno ; ruzicka . The conductivity increases with increasing $`x`$ yamane .
The doped holes can create a polaronic or charge-density-wave (CDW) state, and a charge sliding mode could be expected as a collective excitation. The existence of such states and modes is supported by some experiments in the material with $`x=0`$. Resonant X-ray scattering has revealed a five-site periodic hole structure in the ladder abbamonte ; rusydi . Microwave measurements show a relatively small $`c`$-axis conductivity with a narrow peak in a very low-energy region ($``$ 0.2 meV) kitano1 ; kitano2 . This low-energy resonance is observed up to a temperature ($``$ 100 K) too high to be attributed to single particle excitations, which would be completely broadened and no longer observed above 10K ($``$1meV) due to thermal fluctuations. Similarity has been noted between the nonlinear behavior of the conductivity in SCCO at $`x=0`$ and that of the sliding mode in materials supporting CDW states. Blumberg et al. have reported blumberg that the low-frequency dielectric constant ($`ฯต_04\times 10^6`$) obtained by transport measurements is consistent with estimates from the pinning energy ($``$ 0.2 meV) suggested by microwave measurements.
There are several possible polaronic configurations. One is to randomly distribute the polarons. Another is that some of them bind to compose various multi-polaron configurations: In a 2D case, the doped holes can arrange into stripes, which is one kind of multi-polaron state. The groundstate configuration depends on the doping level, pressure, etc. Modeling the pressure-dependence of the groundstate configuration is one aim of this study.
In a previous work martin ; kane , we studied the electronic and phononic excitations in a 2D CuO<sub>2</sub> plane with inhomogeneous charge-lattice-spin structures (stripes and other polaron patterns). We identified local โedgeโ or โinterfaceโ modes in the phononic and electronic (spin and charge) excitations induced by the inhomogeneity. In the Cu<sub>2</sub>O<sub>3</sub> two-leg ladder in SCCO, we can similarly anticipate the existence of inhomogeneous structure and associated local excitation modes in spin, charge, and lattice degrees of freedom.
Here, we investigate the ground states and the electronic and phononic excitations in the two-leg ladder system by applying an unrestricted Hartree-Fock and a direct-space Random Phase approximations (RPA) to a multi-band Peierls-Hubbard Hamiltonian. We consider single-polaron (SP) and bi-polaron (BP) states: The latter comprise bound polarons extending over rungs. Compared to the 2D cases, the SP state that includes isolated polarons, is found to possess similar phonon excitations as the diagonal stripe state or the periodic polaron state martin ; kane . On the other hand, the BP state shows the same type of local phonon modes as the vertical stripe state in the 2D system. This is reasonable, since the vertical stripe state is a form of multi-polaron state, which includes several polarons coupled by shared O ions. To model the effect of pressure on the groundstate configuration, we compare the energies of the SP state and the BP state, while varying the Cu-O hopping integral. We find that as a function of increasing โpressureโ (modeled by increasing hopping strength) a transition from the SP state to the BP state is induced, together with interesting intermediate states. Most strikingly, we identify a transition between site- and bond-centered BPs accompanied by phonon softening indicative of the onset of sliding or other instabilities.
## II formulation
### II.1 Hamiltonian
To study a Cu<sub>2</sub>O<sub>3</sub> ladder, we use the following three-band extended Peierls-Hubbard Hamiltonian, which includes both electron-electron and electron-lattice interactions yone ; yi :
$`H_0={\displaystyle \underset{<ij>\sigma }{}}t_{pd}(u_{ij})(c_{i\sigma }^{}c_{j\sigma }+H.c.)+{\displaystyle \underset{i,\sigma }{}}ฯต_i(u_{ij})c_{i\sigma }^{}c_{i\sigma }`$
$`+{\displaystyle \underset{<ij>}{}}{\displaystyle \frac{1}{2}}K_{ij}u_{ij}^2+{\displaystyle \underset{i,j,\sigma ,\sigma ^{}}{}}U_{ij}n_{i\sigma }n_{j\sigma ^{}}.`$ (1)
We impose periodic (open) boundary condition in the $`x`$ ($`y`$) direction, i.e., there are two periodic Cu-O chains along the $`x`$-direction (we term this oxygen O<sub>x</sub> subsequently), connected together through the other oxygens (O<sub>y</sub>). In this Hamiltonian, $`c_{i\sigma }^{}`$ creates a hole with spin $`\sigma `$ on site $`i`$, and each site has one orbital (d$`_{x^2y^2}`$ on Cu, and p<sub>x</sub> or p<sub>y</sub> on O). The Cu (O) site electronic energy is $`ฯต_d`$ ($`ฯต_p`$). $`U_{ij}`$ represents the on-site Cu (O) Coulomb, $`U_d`$ ($`U_p`$), or the nearest-neighbor repulsion, $`U_{pd}`$. The electron-lattice interaction modifies the Cu-O hopping strength linearly through the oxygen displacement $`u_{ij}`$: $`t_{pd}(u_{ij})=t_{pd}(1\pm \alpha u_{ij})`$, where $`+()`$ applies if the Cu-O bond shrinks (stretches) for a positive $`u_{ij}`$; it also affects the Cu on-site energies $`ฯต_d(u_{ij})=ฯต_d+\beta _j(\pm u_{ij})`$, where the sum is over the three neighboring O ions. Other oxygen modes (buckling, bending, etc) are assumed to couple to electron charge more weakly and are neglected here for simplicity, but can be included as necessary within the same approach. We use variations around the following set of model parameters used in 2D CuO<sub>2</sub> models martin ; kane : $`ฯต_pฯต_d=4`$ eV, $`U_d=8`$ eV, $`U_p=3`$ eV, $`U_{pd}=1`$ eV, and $`K=32`$ eV/ร
<sup>2</sup>, $`\alpha =2.0`$ eV/ร
, $`\beta =1`$ eV/ร
; we vary $`t_{pd}=15`$eV to simulate the pressure effect. This is clearly an oversimplified representation of pressure effects, but serves to demonstrate the ground state phases and transitions. Effects of varying the coupling strength are also considered below with similar results. To approximately solve the above model, we use unrestricted Hartree-Fock combined with an inhomogeneous generalized RPA to study linear fluctuations of lattice, spin or charge yone in a supercell of size $`N_x\times 2`$ (we take $`N_x=8`$ here).
### II.2 Phonon spectral function
The output of the calculation is the Hartree-Fock ground state and the linearized fluctuation eigen-frequencies and eigen-vectors with respect to it. From the phonon eigen-modes, we calculate the corresponding neutron scattering cross section:
$`S(๐ค,\omega )={\displaystyle ๐te^{i\omega t}\underset{ll^{}}{}e^{i\mathrm{๐ค๐ซ}_l^ฯต(0)}e^{i\mathrm{๐ค๐ซ}_l^{}^ฯต(t)}},`$ (2)
where $`ฯต`$ labels the five ions in the unit cell of the ladder: (1) O<sub>x</sub> ions in the lower leg, (2) O<sub>y</sub> ions in rungs, (3) O<sub>x</sub> ions in the upper leg, (4) Cu ions in the lower leg, and (5) Cu ions in the upper leg; Here the position of each ion is expressed by $`๐ซ_l^ฯต(t)=๐ฑ_l+๐^ฯต+๐ฎ_l^ฯต(t)`$, where each of the terms represents the location of the $`l`$-th unit cell origin $`๐ฑ_l`$($`=x_l\widehat{1}_x`$), time-dependent vibrational component $`๐ฎ_l^ฯต(t)`$, and position within the unit cell $`๐^ฯต`$:
$`๐^{(1)}={\displaystyle \frac{a}{2}}\widehat{1}_x,๐^{(2)}={\displaystyle \frac{a}{2}}\widehat{1}_y,๐^{(3)}={\displaystyle \frac{a}{2}}\widehat{1}_x+a\widehat{1}_y,`$
$`๐^{(4)}=\widehat{0},๐^{(5)}=a\widehat{1}_y.`$ (3)
As noted above, for simplicity we consider Cu ions as fixed, and the motion of O ions oriented along the corresponding Cu-O bond: $`๐ฎ_l^ฯต=u_l^ฯต\widehat{\mathrm{e}}_ฯต`$ with $`\widehat{\mathrm{e}}_1=\widehat{\mathrm{e}}_3=\widehat{1}_x`$, $`\widehat{\mathrm{e}}_2=\widehat{1}_y`$, $`\widehat{\mathrm{e}}_4=\widehat{\mathrm{e}}_5=\widehat{0}`$. The scalar displacements can now be expressed in terms of the normal modes $`z_n`$ as $`u_l^ฯต(t)=_n\mathrm{\Phi }_{x_l,n}^ฯตz_n(t)`$. Making a second-order expansion in the oxygen displacements, we obtain
$`S(๐ค,\omega )=`$
$`{\displaystyle \underset{n}{}}\{[k_x^2|\mathrm{\Phi }_{k_x,n}^{(1)}|^2+k_y^2|\mathrm{\Phi }_{k_x,n}^{(2)}|^2+k_x^2|\mathrm{\Phi }_{k_x,n}^{(3)}|^2]`$
$`+[k_xk_y\mathrm{e}^{\mathrm{i}(k_xk_y)\frac{a}{2}}\mathrm{\Phi }_{k_x,n}^{(1)}(\mathrm{\Phi }_{k_x,n}^{(2)})^{}+c.c.]`$
$`+[k_xk_y\mathrm{e}^{\mathrm{i}(k_x\frac{a}{2}+k_y\frac{a}{2})}\mathrm{\Phi }_{k_x,n}^{(2)}(\mathrm{\Phi }_{k_x,n}^{(3)})^{}+c.c.]`$
$`+[k_x^2\mathrm{e}^{\mathrm{i}k_ya}\mathrm{\Phi }_{k_x,n}^{(3)}(\mathrm{\Phi }_{k_x,n}^{(1)})^{}+c.c.]\}`$
$`\times {\displaystyle \frac{\mathrm{}}{2m\omega _n}}[(1+n_B)\delta (\omega \omega _n)+n_B\delta (\omega +\omega _n)].`$ (4)
Here, $`\mathrm{\Phi }_{k_x,n}^ฯต=_le^{\mathrm{i}k_xx_l}\mathrm{\Phi }_{x_l,n}^ฯต`$, and $`n_B=(e^{\omega _n/T}1)^1`$ is the thermal population of the phonon mode $`n`$. This is a generalization of the usual neutron scattering intensity expression lovesay for the case of phonons with a larger real space unit cell. We plot $`S(๐ค,\omega )/|๐ค|^2`$ for k-directions sampling longitudinal modes, consistent with the common experimental convention.
### II.3 Electron spectral function
To investigate the neutral electronic excitations, we calculate the spectral function fetter :
$`{\displaystyle \underset{n}{}}|\mathrm{\Psi }_0|๐ช(๐ค)|\mathrm{\Psi }_n|^2\delta (\omega E_nE_0),`$ (5)
where $`|\mathrm{\Psi }_0`$ ($`|\mathrm{\Psi }_n`$) is the RPA ground (excited) state whose energy is represented by $`E_0`$ ($`E_n`$), and $`๐ช(๐ค)`$ is an operator, e.g. spin $`๐(๐ค)`$ or charge $`n(๐ค)`$, summed over Cu- and O-sites:
$`๐ช(๐ค)={\displaystyle \underset{ฯต=\mathrm{\hspace{0.17em}1}}{\overset{5}{}}}๐ช^ฯต(๐ค)e^{\mathrm{i}\mathrm{๐ค๐}^ฯต}.`$ (6)
The effect of an infinitesimal external field corresponding to the excitation $`\mathrm{\Psi }_n`$ can be represented by the change of an observable $`๐ช`$ in the state $`\mathrm{\Psi }=\mathrm{\Psi }_0+\eta \mathrm{\Psi }_n`$ ($`|\eta |1`$):
$`๐ช๐ช_0+\delta ๐ช_n,`$ (7)
$`\delta ๐ช_n\mathrm{\Psi }_0|๐ช|\mathrm{\Psi }_n,`$ (8)
where $`๐ช_0`$ is the expectation value with respect to the ground state.
As noted earlier, we simulate the effect of pressure by varying $`t_{pd}`$. We identify a transition from the SP to the BP state with distinct electronic and phononic signatures. Note that the BP states here are polaron bound states on different legs of the ladder (in contrast to same-chain BPs).
## III results
### III.1 Polaronic ground states and phase transitions
We first show in Fig. 1 the configurations of the ground states obtained by the Hartree-Fock calculation for several values of $`t_{pd}`$. For $`1.0<t_{pd}<5.0`$, there are six types of groundstate configurations. (A) in Fig. 1 is a single-polaron (SP) state, which has $`n`$ polarons in a staggered arrangement for an $`n`$-hole doped system ($`n=4`$ here). (B) and (C) are diagonal bi-polaron (DBP) states, which have diagonally-bound polarons; and (D) and (E) are vertical bi-polaron (VBP) states, which have the same structure as a short segment of a vertical stripe in the 2D system. (D) consists of site-centered VBPs, and (E) of bond-centered VBPs. (F) is the uniform (UNI) state, which does not have any local spin or charge modulation, or lattice displacement.
Note that the undoped system shows the AF configuration without any lattice displacement (not shown here). One can expect that a singlet solid is the more likely exact undoped ground state, however the Hartree-Fock calculation favors an AF.
With $`E_{(\mathrm{A})}`$, $`E_{(\mathrm{B})}`$, โฆ, $`E_{(\mathrm{F})}`$ the energies of (A), (B), โฆ, (F), respectively, we compare these energies, and determine the regions of $`t_{pd}`$ corresponding to the (A)-(F) phases for $`1.0<t_{pd}<5.0`$. The $`t_{pd}`$-dependence of these groundstate configurations is shown in Fig. 2.
There are five groundstate transitions for $`1.0<t_{pd}<5.0`$. These are first-order transitions except for the one between (E) and (F),which is of second order. The main feature here is that the larger $`t_{pd}`$, the more delocalized the ground state becomes. This follows from the fact that the transition with increasing $`t_{pd}`$ is SP $``$ BP $``$ UNI. By studying the polaron eigen-functions, we find that the transitions occur when the SP-SP or BP-BP overlap achieves sufficient levels, resulting in the polaron melting ,i.e., the sequence of the transitions SP $``$ BP $``$ UNI.
The transition DBP $``$ VBP with increasing $`t_{pd}`$ is similar to the transition from diagonal to vertical stripe with increasing hole-concentration in some of the high-$`T_c`$ cuprates with CuO<sub>2</sub> planes machida ; matsuda . The energy difference between (D) and (E) near the transition point is smaller than for the other transitions. $`E_{(\mathrm{D})}E_{(\mathrm{E})}`$ and $`E_{(\mathrm{E})}E_{(\mathrm{F})}`$ both asymptotically vanish, as $`t_{pd}`$ approaches the (E)-(F) transition point from smaller $`t_{pd}`$. This transition is associated with recovery of the broken symmetry. Below, we show the existence of the VBP sliding mode, which recovers the translational symmetry.
### III.2 Sliding mode in bi-polaronic states
The $`t_{pd}`$-dependence of the phonon eigen-frequencies obtained by a direct-space RPA calculation is shown in Fig. 3. A main branch lies in the range of 80-85 meV and is consistent with the main phonon branch observed experimentally. This frequency range is insensitive to the doping level. However, some characteristic local modes are induced below this frequency range by hole doping, similar to the 2D cases martin ; kane . Additionally, an extremely soft mode is found in the (D) and (E) phases. The frequency of the soft phonon mode in the (E) configuration is purely imaginary in the (D) phase region, and vice versa. To further understand this soft mode, we calculate the corresponding electronic excitations \[Fig. 4\]. We find one soft charge excitation \[Fig. 4 (a)\], which shows a sliding mode \[Fig. 4 (b)\] and whose excitation energy shows the same behavior as that of the soft phonon mode. We identify the soft phonon mode as one coupling with a sliding mode of VBPs along the ladder. We identify the frequency of this mode as a pinning energy of the VBP sliding, corresponding to a Peierls-Nabarro barrier from the lattice discreteness. In the (D) phase \[(E) phase\], VBPs are pinned Cu site-centered \[bond-centered\] by a potential energy; the potential energy is minimal at a Cu site \[an O<sub>x</sub> site\] and maximal at an O<sub>x</sub> site \[a Cu site\]. The minimum and maximum points of the potential energy exchange at the (D)-(E) transition point. We will discuss this further below.
From the softening of the sliding mode and the second-order-like behavior of the energy differences, $`E_{(\mathrm{D})}E_{(\mathrm{E})}`$ and $`E_{(\mathrm{E})}E_{(\mathrm{F})}`$ near the (E)-(F) transition point, we conclude that the pinning potential becomes flat at that transition point.
### III.3 Effects of electron-lattice coupling
We next consider effects of the electron-lattice coupling on the groundstate configuration. For this purpose, changing the coupling strength $`\alpha `$ (=1.0, 2.0, 3.0, 4.5) we calculate the critical values of $`t_{pd}`$ in the same manner as above. In this way, the $`t_{pd}\alpha `$ phase diagram is found as in Fig. 1. Increasing the coupling strength tends to raise the critical values of $`t_{pd}`$ except for that between (E) and (F): The boundary between (E) and (F) is insensitive to the change of the coupling strength. The difference in the $`\alpha `$-dependence shows that the transition between (E) and (F) has a different character than other transitions: As seen above, the transition between (E) and (F) is most likely of second-order.
## IV discussion
We now discuss the details of the results obtained above. First, we analyze the polaron size by using Gaussian fitting, and describe the overall picture of the transitions as the delocalization of polarons induced by the pressure. Next, we discuss the transition between site- and bond-centered VBPs and the pinning frequencies of a CDW in terms of Ginzburg-Landau theory.
### IV.1 Size of polaron
Fig. 6 shows hole-density profiles for several cases. The data shown by circles in Fig. 6 includes averages over nearest O sites:
$`\overline{n}_i=n_{\mathrm{Cu},i}+{\displaystyle \frac{1}{2}}{\displaystyle \underset{\genfrac{}{}{0pt}{}{j:\mathrm{nearest}}{\mathrm{neighbors}}}{}}n_{\mathrm{O},j},`$ (9)
where $`n_{\mathrm{Cu},i}`$ and $`n_{\mathrm{O},j}`$ are hole densities, respectively, at Cu and O sites obtained by Hartree-Fock calculation. We fit the data by a least-squares method with the following function:
$`\rho (x,y)={\displaystyle \underset{i}{}}{\displaystyle \frac{1}{2\pi \sigma ^2}}\mathrm{exp}\left\{{\displaystyle \frac{(x\stackrel{~}{x}_i)^2+(y\stackrel{~}{y}_i)^2}{2\sigma ^2}}\right\}+h.`$
(10)
Here, the center of the $`i`$-th polaron is represented by $`(\stackrel{~}{x}_i,\stackrel{~}{y}_i)`$, and $`h`$ takes a value of about 1.1. In the SP and VBP cases (upper and lower in Fig. 6), the data is well-fitted by the function in Eq. (10). In the DBP case, on the other hand, it is not as well-fitted, especially for large $`\alpha `$ (middle right in Fig. 6).
We define the size of a polaron as $`2\sigma /a`$. The $`t_{pd}`$ dependence of this polaron size is shown in Fig. 7. In the (A) case, we find $`2\sigma /a`$ is between 1.1 and 1.2 within the $`t_{pd}`$ range between 1.0 and 2.5. There is little change with increasing $`t_{pd}`$ in this case. The (B) state shows the same behavior as the (C) state. The polaron size of the DBP states increases as $`t_{pd}`$ increases. This polaron size growth in the DBP states is greater than in the SP state. However, the polaron sizes in the VBP are larger than those of SP or DBP.
Comparing with Fig. 2, we find that the phase transitions occur from a state of smaller polarons to another state of larger ones. Especially, we can understand the aspects of the phase transitions (D)$``$(E)$``$(F) by considering the pinning potential. Delocalization of the polaron induced by changing $`t_{pd}`$ causes less distortion of the lattice, as shown in Fig. 1. Therefore, the potential energies for site- and bond-centered states should change: For small $`t_{pd}`$, the pinning at Cu is stronger than that at O. As is discussed in Appendix A, if the change of $`t_{pd}`$ varies the ratio of the pinning potential at Cu and O sites, then phase transitions are induced. The phase transitions are also well-described by Ginzburg-Landau theory. This is discussed in the following section, where, within the Ginzburg-Landau picture, we explain the behavior of the pinning frequencies.
### IV.2 Pinning frequency
We now discuss the transition between site- and bond-centered VBPs and the behavior of the pinning frequencies in terms of a Ginzburg-Landau theory. We can describe the aspects of the transitions between VBP and UNI, and between site- and bond-centered VBPs by introducing the Landau function (See Appendix B for details):
$`F={\displaystyle ๐xf[\psi (x),m(x)]}.`$ (11)
Here $`\psi (x)`$ and $`m(x)`$ are the charge and spin order parameters. $`m(x)`$ is the staggered spin density, and $`\psi (x)`$ is defined by the deviation of the charge density from the uniform state:
$`m(x)`$ $`=`$ $`(1)^{\frac{x}{a}}S(x)`$ (12)
$`\psi (x)`$ $`=`$ $`\rho (x)\rho _0.`$ (13)
In general, $`f`$ can be written in the following form for an $`L`$-site-periodic commensurate CDW McMillan :
$`f[\psi (x),m(x)]`$ $`=`$ $`f_0[\psi (x),m(x)]`$ (14)
$`+p(x)\psi (x)^L+q(x)\psi (x)^{2L}.`$
Here, $`f_0`$ is concerned with the lattice-independent spin and charge ordering, and the remaining terms lead to the lattice pinning effect of the CDW. We assume $`f_0`$ is of the form
$`f_0[\psi (x),m(x)]`$ $`=`$ $`r_0m(x)^2+u_0m(x)^4`$ (15)
$`+s_0m(x)^2\psi (x)+v_0\psi (x)^2`$
with $`v_0>0`$, that is, the charge order is induced by the magnetic order pryadko . If we write the charge order parameter in sinusoidal form, the amplitude $`\rho _1`$ is found to be (see Appendix B)
$`\rho _1t_0t_{pd},`$ (16)
where $`t_0`$ is the BP-UNI transition point.
Considering the small oscillations around the equilibrium state, the pinning frequencies are derived in Appendix B as
$`\mathrm{\Omega }\{\begin{array}{cc}|t_{pd}t_c|^{1/2}\hfill & \text{for }t_{pd}t_c\hfill \\ |t_0t_{pd}|^{(L2)/2}\hfill & \text{for }t_{pd}\stackrel{<}{}t_0\hfill \end{array},`$ (19)
where $`t_c`$ is the transition point between site- and bond-centered VBPs. Both expressions are plotted in Fig. 8 (b), and show a good agreement with the RPA data. If $`t_c`$ is not very far from $`t_0`$, the following form well-describes the behavior of the phonon frequencies as a function of $`t_{pd}`$ over the whole region around the transition points \[see Fig. 8 (a)\]:
$`\mathrm{\Omega }|t_{pd}t_c|^{1/2}|t_0t_{pd}|^{(L2)/2}.`$ (20)
As shown in Appendix B, the $`t_{pd}`$ dependence of the energy difference between site- and bond-centered VBPs is given by
$`|\mathrm{\Delta }F|\{\begin{array}{cc}|t_{pd}t_c|\hfill & \text{for }t_{pd}t_c\hfill \\ |t_0t_{pd}|^L\hfill & \text{for }t_{pd}t_0\hfill \end{array}.`$ (23)
These functions are plotted in Fig. 8 (c), where $`|E_{(D)}E_{(E)}|`$ from the Hartree-Fock calculation are also plotted for comparison.
Eq. (19) well reproduces the features of the pinned CDW in the $`L=4`$ case. Here we investigate energies and the pinning frequencies for different dopings (resulting in different-period CDWs), and further show the validity of Eq. (19) for other $`L`$. Figs. 9 and 10 show the results for the $`\frac{1}{3}`$-hole doping ($`L=3`$), and the $`\frac{1}{5}`$-hole doping ($`L=5`$) cases. In the both cases, the results suggest that Eq. (19) agrees with the RPA calculation: The Hartree-Fock and RPA calculations were performed in the systems whose sizes were $`6\times 2`$ ($`L=3`$) and $`10\times 2`$ ($`L=5`$). The $`t_{pd}`$ dependence of the energy difference between site- and bond-centered VBPs is also well-described by the function obtained from Ginzburg-Landau theory in both cases.
Similar to the $`\frac{1}{4}`$-hole doping, for $`\frac{1}{3}`$-hole doping. Eq. (20) is also a good approximation. For $`\frac{1}{5}`$-hole doping, Eq. (20) does not give a good agreement with RPA data. This is not surprising, since $`t_c`$ is far from $`t_0`$ in this case.
## V summary
In summary, we have modeled a pressure effect in a Cu<sub>2</sub>O<sub>3</sub> ladder system by using a multi-band Peierls-Hubbard model and simulating the effect of pressure through the hopping strength $`t_{pd}`$. With increasing $`t_{pd}`$, we find a sequence of transitions from SP/BP charge localization to sliding to delocalization, all occurring within a magnetically ordered background. The ground state has the same number of SPs as doped holes in the case of small $`t_{pd}`$ ($`t_{pd}<1.6`$ with the parameters used here). The ground state configuration changes as SP $``$ BP $``$ UNI states, as $`t_{pd}`$ increases. While SPs are localized and isolated, BPs are partially delocalized. This means the pressure produces a more delocalized ground state. In the BP phase, there is also a phase transition between DBP and VBP states. A similar transition has been found in some other cuprates: namely, the transition between diagonal- and vertical-stripe states induced by hole doping.
In the VBP phase, we also find a soft mode transition between site- and bond-centered VBP states, although the energies are very close. Calculations of the phonon eigen-frequency and electronic excitation in the VBP phase yields a sliding mode of VBPs with weak pinning. The pinning energy in the bond-centered VBP phase is around 15meV at most for $`t_{pd}3.5`$ eV. Increasing $`t_{pd}`$ up to $`t_{pd}4.6`$ eV makes the pinning zero, and a transition from the VBP to UNI state occurs. These results suggest experimentally exploring pressure dependence of the low-energy modes found by Kitano et al. kitano1 ; kitano2 and Blumberg et al. blumberg If these modes correspond to those we have identified, then their pressure-dependence will follow the interesting pattern in Fig. 3. Studying IR, Raman and optical signatures would further clarify the mode assignments. The resonant soft X-ray scattering technique of Refs. abbamonte ; rusydi could also be used to probe our predicted charge ordering structures as a function of pressure.
The sequence of phases (SP-BP-UNI) is reminiscent of the insulator-metal transition with doping observed in other low-dimensional broken-symmetry groundstate materials, including conjugated polymers heeger and layered cuprates. It is tempting to associate the mode softening with the onset of a sliding CDW in the spirit of Frรถhlich. However, as the phonon-fluctuations soften, additional degrees of freedom (quantum lattice and spin fluctuations) become relevant and need to be considered โ in particular to identify the superconductivity mechanism. The superconductivity is observed experimentally in a finite range of pressure. Whether this can be associated with the finite range of $`t_{pd}`$ with low pinning frequencies (Fig. 3) requires comparison with more detailed experiments, but is a tempting scenario. We have also explored other values of $`\alpha `$ and found the same general phase sequence as a function of $`t_{pd}`$ shown in Fig. 5.
In this study, we mainly considered only the $`\frac{1}{4}`$-hole-doped case and related commensurate dopings. Other cases including incommensurate fillings with discommensurations will be reported elsewhere. We can expect related transitions with doping as with $`t_{pd}`$, since they should both be controlled by SP or BP wave-function overlaps.
## ACKNOWLEDGEMENT
This work was supported by the U.S. DOE.
## Appendix A Pinning potential
Here we attempt to describe the pinning potential of VBP states, and show the details of the discussion regarding the transition between site- and bond-centered VBPs in Sec. IV.1. Since we are interested only in VBP states with different phases here, the system we consider can be reduced to one dimension. Therefore, the pinning potential is a function of $`x`$, and has a minimum at $`x=na`$ for the site-centered VBP and at $`x=(n+\frac{1}{2})a`$ for the bond-centered VBP. Approximately, the pinning potential may be attributed to the potential at Cu and O sites:
$`E(x)={\displaystyle \underset{i}{}}\left(_{\mathrm{Cu},i}(x)+_{\mathrm{O},i}(x)\right),`$ (24)
where the phase is chosen such that the state becomes site-centered when $`x=0`$. We continue our discussion with the following two assumptions: (1) both $`_{\mathrm{Cu}}`$ and $`_\mathrm{O}`$ are composed of Gaussians,
$`_{\mathrm{Cu},i}(x)={\displaystyle \frac{C_1}{\sqrt{2\pi }\sigma _1}}\mathrm{exp}\left\{{\displaystyle \frac{(xai)^2}{2\sigma _1^2}}\right\}`$ (25)
$`_{\mathrm{O},i}(x)={\displaystyle \frac{C_2}{\sqrt{2\pi }\sigma _2}}\mathrm{exp}\left\{{\displaystyle \frac{(xai+\frac{a}{2})^2}{2\sigma _2^2}}\right\},`$ (26)
and (2) both $`\sigma _1`$ and $`\sigma _2`$ are comparable to $`a`$. The case we consider here can satisfy these conditions. Since the density profile of VBP states is well-fitted by Gaussians (shown in Sec. IV), we expect the contribution of the partial free energy of the Cu and O sites to the total is also formed of Gaussians, and that the size of the Gaussians should also be similar to the VBP size.
We investigate the potential energy in Eq. (24) with assumption (1). The pinning potential can be expanded as a Fourier series:
$`E(x)={\displaystyle \underset{n}{}}E_n\mathrm{cos}(nGx),`$ (27)
where $`G={\displaystyle \frac{2\pi }{a}}`$. The Fourier coefficients are given by the following form:
$`E_n`$ $`=`$ $`{\displaystyle \frac{1}{a}}{\displaystyle _{\frac{a}{2}}^{\frac{a}{2}}}๐x{\displaystyle \underset{i}{}}\left[{\displaystyle \frac{C_1}{\sqrt{2\pi }\sigma _1}}\mathrm{exp}\left\{{\displaystyle \frac{(xai)^2}{2\sigma _1^2}}\right\}+{\displaystyle \frac{C_2}{\sqrt{2\pi }\sigma _2}}\mathrm{exp}\left\{{\displaystyle \frac{(xai+\frac{a}{2})^2}{2\sigma _2^2}}\right\}\right]\mathrm{cos}(nGx)`$ (28)
$`=`$ $`{\displaystyle \frac{1}{a}}\left[{\displaystyle \frac{C_1}{\sqrt{2\pi }\sigma _1}}{\displaystyle _{\mathrm{}}^{\mathrm{}}}๐x\mathrm{exp}\left\{{\displaystyle \frac{x^2}{2\sigma _1^2}}\right\}\mathrm{cos}(nGx)+(1)^n{\displaystyle \frac{C_2}{\sqrt{2\pi }\sigma _2}}{\displaystyle _{\mathrm{}}^{\mathrm{}}}๐x\mathrm{exp}\left\{{\displaystyle \frac{x^2}{2\sigma _2^2}}\right\}\mathrm{cos}(nGx)\right]`$ (29)
$`=`$ $`\left[{\displaystyle \frac{C_1}{a}}\mathrm{exp}\left\{{\displaystyle \frac{n^2G^2\sigma _1^2}{2}}\right\}+(1)^n{\displaystyle \frac{C_2}{a}}\mathrm{exp}\left\{{\displaystyle \frac{n^2G^2\sigma _2^2}{2}}\right\}\right].`$ (30)
From Eq. (30), it follows that $`E_n`$ for large $`n`$ vanishes. By considering the fact that the Gaussian almost vanishes at three-fold half maximum full-width, we evaluate the condition to neglect the components as:
$`{\displaystyle \frac{n^2G^2\sigma ^2}{2}}={\displaystyle \frac{2\pi ^2n^2\sigma ^2}{a^2}}{\displaystyle \frac{3^2}{2}}.`$ (31)
Here all variables are positive, so that the condition is given by
$`{\displaystyle \frac{2\sigma }{a}}`$ $``$ $`{\displaystyle \frac{1}{n}}.`$ (32)
If we consider the case that both $`\sigma _1`$ and $`\sigma _2`$ are comparable to $`a`$ (assumption (2)), $`E(x)`$ is approximately represented by the cosine curve or a slightly modified one:
$`E(x)E_0+E_1\mathrm{cos}(Gx)+E_2\mathrm{cos}(2Gx).`$ (33)
Here $`E_2`$ is small and it does not change the shape of $`\mathrm{cos}(Gx)`$ very much unless $`E_1`$ is small as well as $`E_2`$. In such a situation, only one of the site- or bond-centered VBP states is stable, and is determined by the relation between the $`C_1`$ and $`C_2`$ magnitudes. However, if $`E_1`$ is smaller than $`4E_2`$, $`E(x)`$ has minima at both Cu and O sites. This can happen when $`C_1C_2`$ (see Fig. 11 (a)).
In Fig. 11 (b), the scheme of phase transition which follows from the Hartree-Fock calculation (Fig. 2) is shown. From the phonon frequency calculation (Fig. 3), the bi-stability interval, $`t_1t_2`$, is very narrow and difficult to identify numerically. Comparing this scheme and the behavior of $`dE(x)/dx`$, we understand the phase transition (D)$``$(E)$``$(F) as follows. (1) Since the site-centered state (D) is found at small $`t_{pd}`$ ($`<t_2`$), $`C_2/C_1`$ has to be less than 1 in this case. (2) Since the double-minimum region is very narrow ($`t_1t_2`$), the zig-zag boundary near $`C_2/C_11`$ in Fig. 11 is almost flat when $`t_{pd}`$ is close to the transition point. (3) Since the bond-centered state (E) is found above the transition point ($`>t_1`$), $`C_2/C_1`$ should be found in the upper region ($`>1`$) across the coexistence point $`t_c`$. (4) Increasing $`t_{pd}`$ far above $`t_1`$, $`E(x)`$ becomes close to a constant $`F_0`$, which implies that the transition from (E) to (F) occurs.
## Appendix B Ginzburg-Landau theory of the site-centered to bond-centered VBP transition
Here we give details of the discussed transition between site- and bond-centered VBPs in Sec. IV.2. First, we describe the statics of the transition by considering the free energy. Then, by considering small long-wavelength oscillations around the ground state, we estimate the pinning frequencies.
We start with the Landau function:
$`F={\displaystyle ๐x}`$ $`[`$ $`r_0m(x)^2+u_0m(x)^4`$ (34)
$`+s_0m(x)^2\psi (x)+v_0\psi (x)^2`$
$`+p(x)\psi (x)^L+q(x)\psi (x)^{2L}]`$
We write $`p(x)`$ and $`q(x)`$ in an expanded form, for example
$`p(x)={\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}p_n\mathrm{cos}(nGx).`$ (35)
As seen in Sec. IV, the density profile of the VBP is well fitted by large-size Gaussians, so we can neglect higher harmonic terms of the order parameters and describe them by sinusoidal waves:
$`\psi (x,\varphi )\rho _1\mathrm{cos}(\frac{G}{L}x+\frac{\varphi }{L}),`$ (36)
$`m(x,\varphi )m_1\mathrm{sin}(\frac{G}{2L}x+\frac{\varphi }{2L}).`$ (37)
We may constrain $`\rho _10`$.
We evaluate the free energy using the following integral ($`l0`$):
$`{\displaystyle _V}{\displaystyle \frac{dx}{V}}\mathrm{cos}^n(\frac{G}{L}x+\frac{\varphi }{L})\mathrm{cos}(lGx)`$
$`={\displaystyle \underset{h=0}{\overset{(n1\stackrel{~}{\delta }_n)/2}{}}}{\displaystyle \frac{1}{2^n}}\left({\displaystyle \genfrac{}{}{0pt}{}{n}{h}}\right)\delta _{n2h,lL}\mathrm{cos}(l\varphi ),`$ (38)
where $`\left(\genfrac{}{}{0pt}{}{n}{l}\right)=\frac{n!}{(nl)!l!}`$, $`\delta `$ is Kroneckerโs delta, and $`\stackrel{~}{\delta }_n`$ is unity for even $`n`$ and zero for odd $`n`$. Integrating over volume $`V`$, the free energy per unit volume is
$`F_\varphi (t_{pd},\rho _1,m_1)`$ $`=`$ $`F^{(0)}(t_{pd},\rho _1,m_1)`$ (39)
$`+\left[\stackrel{~}{p}_1\rho _1^L+\stackrel{~}{q}_1\rho _1^{2L}\right]\mathrm{cos}(\varphi )`$
$`+\stackrel{~}{q}_2\rho _1^{2L}\mathrm{cos}(2\varphi ),`$
$`F^{(0)}(t_{pd},\rho _1,m_1)`$ $`=`$ $`\stackrel{~}{r}_0m_1^2+\stackrel{~}{u}_0m_1^4`$ (40)
$`\stackrel{~}{s}_0m_1^2\rho _1+\stackrel{~}{v}_0\rho _1^2,`$
where
$`\stackrel{~}{p}_1={\displaystyle \frac{1}{2^L}}p_1,\stackrel{~}{q}_1=\stackrel{~}{\delta }_L\left({\displaystyle \genfrac{}{}{0pt}{}{2L}{\frac{L}{2}}}\right){\displaystyle \frac{1}{2^{2L}}}q_1,\stackrel{~}{q}_2={\displaystyle \frac{1}{2^{2L}}}q_2,`$ (41)
$`\stackrel{~}{u}_0={\displaystyle \frac{3}{8}}u_0,\stackrel{~}{v}_0={\displaystyle \frac{1}{2}}v_0,\stackrel{~}{r}_0={\displaystyle \frac{1}{2}}r_0,\stackrel{~}{s}_0={\displaystyle \frac{1}{4}}s_0,`$ (42)
Note that the signs of these variables are the same with or without tilde. The $`r_0`$ and $`u_0`$ terms govern the BP-UNI continuous transition, and the rest characterize the transition between site- and bond-centered VBPs. Subsequently, we neglect the $`\stackrel{~}{q}_1\rho _1^{2L}`$ term in Eq.(39), since this term would be smaller than the $`\stackrel{~}{p}_1\rho _1^L`$ term.
First, we consider the BP-UNI transition. Phenomenologically assuming $`r_0t_{pd}t_0`$ ($`t_0`$ is the BP-UNI transition point, where $`p_1=0`$) and $`u_0>0`$, we find $`m_1(t_0t_{pd})^{1/2}`$ for $`t_{pd}<t_0`$. From the last two terms of Eq. (40), it follows that
$`\rho _1m_1^2t_0t_{pd}`$ (43)
for $`t_{pd}<t_0`$. Next, we consider the transition between site- and bond-centered VBPs. In the case $`t_{pd}t_c`$ ($`t_c`$ is the coexistence point), the minimum point of this free energy is controlled by the $`p_1`$ term. At $`t_{pd}=t_c`$, the $`p_1`$ term vanishes, and the minimum of the free energy is determined by the $`q_2`$ term.
The minimum point is given by
$`\varphi =\{\begin{array}{cc}\hfill \pi & \text{for }p_1>0\hfill \\ \hfill 0& \text{for }p_1<0\hfill \end{array}.`$ (46)
$`\varphi =\pi `$ corresponds to the bond-centered case at $`t_{pd}>t_c`$, and $`\varphi =0`$ to the site-centered case at $`t_{pd}<t_c`$. At the critical point ($`t_{pd}=t_c`$), the $`p_1`$ term vanishes; and for
$`q_2\rho _1^{2L}<0`$ (47)
there are minima at both $`\varphi =0`$ and $`\pi `$.
Next we consider the small oscillations around the equilibrium state: $`\varphi =\varphi _0+\delta \varphi `$ ($`\varphi _0`$ takes either 0 or $`\pi `$ for the ground state). Using $`p_1\mathrm{cos}(\varphi _0)=|p_1|`$ and $`\mathrm{cos}(2\varphi _0)=1`$, the free energy is expanded for $`\delta \varphi `$ as
$`F_{\varphi _0+\delta \varphi }(t_{pd},\psi ,m)`$ (49)
$`=`$ $`F^0(t_{pd},\rho _1,m_1)+\stackrel{~}{p}_1\rho _1^L\mathrm{cos}(\varphi _0)\left[1{\displaystyle \frac{(\delta \varphi )^2}{2}}\right]`$
$`+\stackrel{~}{q}_2\rho _1^{2L}\mathrm{cos}(2\varphi _0)\left[1{\displaystyle \frac{(2\delta \varphi )^2}{2}}\right]`$
$`=`$ $`\left[F^0(t_{pd},\rho _1,m_1)|\stackrel{~}{p}_1|\rho _1^L+\stackrel{~}{q}_2\rho _1^{2L}\right]`$
$`+{\displaystyle \frac{1}{2}}\left(|\stackrel{~}{p}_1|\rho _1^L4\stackrel{~}{q}_2\rho _1^{2L}\right)(\delta \varphi )^2.`$
The second line of Eq. (49) is used to find mode frequencies. The Lagrangian for the oscillation of VBPs is given by
$`={\displaystyle \frac{M}{2}}{\displaystyle \frac{d(\delta \varphi )}{dt}}{\displaystyle \frac{M\mathrm{\Omega }^2}{2}}(\delta \varphi )^2+\mathrm{const}.`$ (50)
Here, $`M`$ is the effective mass of the CDW, and
$`\mathrm{\Omega }^2={\displaystyle \frac{1}{M}}\left(|\stackrel{~}{p}_1|\rho _1^L4\stackrel{~}{q}_2\rho _1^{2L}\right).`$ (51)
Using $`p_1t_{pd}t_c`$ and $`\rho _1t_0t_{pd}`$, and supposing $`M\rho _1^2`$, for $`t_{pd}<t_0`$, we find
$`\mathrm{\Omega }\{\begin{array}{cc}|t_{pd}t_c|^{1/2}\hfill & \text{for }t_{pd}t_c\hfill \\ |t_0t_{pd}|^{(L2)/2}\hfill & \text{for }t_{pd}\stackrel{<}{}t_0\hfill \end{array}.`$ (54)
It also follows from Eq. (39) that the $`t_{pd}`$ dependence of the energy difference between site- and bond-centered VBPs, $`|\mathrm{\Delta }F|`$, is given by
$`|F_{\varphi =0}F_{\varphi =\pi }|\{\begin{array}{cc}|t_{pd}t_c|\hfill & \text{for }t_{pd}t_c\hfill \\ |t_0t_{pd}|^L\hfill & \text{for }t_{pd}t_0\hfill \end{array}.`$ (57)
|
warning/0506/hep-ph0506274.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
One major task of a future $`e^+e^{}`$ linear collider will be the exploration of the Higgs sector, in the Standard Model and beyond it. Despite its success, the Standard Model suffers from the appearance of quadratically divergent contributions to the Higgs boson mass. However, this problem is solved by Supersymmetry. The theoretical framework of our study is the minimal supersymmetric extension of the Standard Model (MSSM), where the Higgs spectrum consists of three unphysical Goldstone modes ($`G^+`$, $`G^{}`$ and $`G^0`$) as well as five physical states. Two of these are charged ($`H^+`$ and $`H^{}`$) and, among the three neutral Higgs bosons, two are CP-even states, $`h^0`$ and $`H^0`$, and one is CP-odd, $`A^0`$. Several other aspects of the MSSM Higgs boson phenomenology are reviewed in .
The processes $`e^+e^{}H^+H^{},H^0A^0,h^0A^0`$ that will be observable at future $`e^+e^{}`$ linear colliders, such as ILC and/or CLIC, are among the best places where one can accurately check the Higgs structure, see references to for details. At tree level, $`e^+`$ and $`e^{}`$ annihilate through a photon and a $`Z`$ boson in the case of $`H^+H^{}`$ production, and through only a $`Z`$ boson in the case of $`H^0A^0`$ and $`h^0A^0`$ production. At this level, the amplitudes depend on the masses of the Higgs bosons and on the mixing angle $`\alpha `$. At the one loop level, most of the MSSM parameter space is involved through self-energies, triangle and box diagrams. In a previous paper it was shown that, at high energy, at the leading and sub-leading (Sudakov) logarithmic orders, a great simplification occurs. The gauge and the SUSY structures of these processes reflect directly in the coefficients of the quadratic and linear logarithmic terms. In this high energy range, they depend only on a few parameters (the Standard Model inputs, the angles $`\alpha `$ and $`\beta `$, as well as the SUSY scale $`M_{SUSY}`$). The next step is to study more deeply the SUSY structure by looking at sub-sub-leading effects. First, one should determine the energy range in which the above Sudakov limit is an acceptable approximation and can be accurately tested. Then, one can study the effects of the successive sub-sub-leading terms (constants, $`m^2/s`$, etc) and classify the various parameters which control each of them. We should then estimate the accuracy at which these parameters can be measured. For these purposes, we have developed a code allowing to compute numerically the complete electroweak one loop contributions to the pair production cross section of MSSM charged and neutral Higgs bosons in $`e^+e^{}`$ collisions. The purpose of the present paper is to write in an explicit fashion all details of the electroweak one loop contributions that are computed by this numerical code.
In Section 2, we review the tree level MSSM Higgs sector and we calculate the production cross section for $`H^+H^{}`$, $`H^0A^0`$ and $`h^0A^0`$ pairs in $`e^+e^{}`$ collisions at tree level. In the rest of the paper, we focus on the various one loop terms. The contributions of the initial vertices and of $`e^\pm `$ self-energy are given in Section 3, the intermediate gauge boson self-energies are discussed in Section 4, the contributions of final vertices and of Higgs self-energies are calculated in Section 5, and the effect of box diagrams are presented in Section 6. Finally, a summary and some outlooks (in particular a more detailed description of our numerical code) are given in Section 7.
## 2 Tree level calculations
### 2.1 Tree level structure of the MSSM Higgs sector
In the MSSM, two complex scalar Higgs doublets are responsible for the breaking of the electroweak symmetry:
$$H_1=\left(\begin{array}{c}(v_1+\varphi _1^0i\chi _1^0)/\sqrt{2}\\ \varphi _1^{}\end{array}\right),H_2=\left(\begin{array}{c}\varphi _2^+\\ (v_2+\varphi _2^0+i\chi _2^0)/\sqrt{2}\end{array}\right).$$
(1)
They have opposite hypercharge ($`Y_1=1`$ and $`Y_2=+1`$) and their vacuum expectation values are respectively $`v_1`$ and $`v_2`$. After diagonalization, one obtains the following states:
$$\left(\begin{array}{c}H^0\\ h^0\end{array}\right)=\left(\begin{array}{cc}\mathrm{cos}\alpha & \mathrm{sin}\alpha \\ \mathrm{sin}\alpha & \mathrm{cos}\alpha \end{array}\right)\left(\begin{array}{c}\varphi _1^0\\ \varphi _2^0\end{array}\right),$$
(2)
$$\left(\begin{array}{c}G^0\\ A^0\end{array}\right)=\left(\begin{array}{cc}\mathrm{cos}\beta & \mathrm{sin}\beta \\ \mathrm{sin}\beta & \mathrm{cos}\beta \end{array}\right)\left(\begin{array}{c}\chi _1^0\\ \chi _2^0\end{array}\right),$$
(3)
$$\left(\begin{array}{c}G^\pm \\ H^\pm \end{array}\right)=\left(\begin{array}{cc}\mathrm{cos}\beta & \mathrm{sin}\beta \\ \mathrm{sin}\beta & \mathrm{cos}\beta \end{array}\right)\left(\begin{array}{c}\varphi _1^\pm \\ \varphi _2^\pm \end{array}\right).$$
(4)
Here, $`G^+`$, $`G^{}`$ and $`G^0`$ describe three unphysical Goldstone modes. The five physical states are two charged bosons ($`H^+`$ and $`H^{}`$), two neutral scalar bosons with CP = $`+1`$ ($`h^0`$ and $`H^0`$) and one pseudoscalar neutral boson with CP = $`1`$ ($`A^0`$).
The quadratic part of the Higgs potential, which contains the soft breaking masses and the gauge couplings, depends on two independent parameters, which are usually chosen as the mass $`M_A`$ of the $`A^0`$ boson and the ratio between the vacuum expectation values $`\mathrm{tan}\beta =v_2/v_1`$. The masses of the other physical states are expressed as follows:
$$M_H^2=M_A^2+M_W^2,$$
(5)
$$M_{H^0,h^0}^2=\frac{1}{2}\left(M_A^2+M_Z^2\pm \sqrt{(M_A^2+M_Z^2)^24M_A^2M_Z^2\mathrm{cos}^22\beta }\right).$$
(6)
As for the mixing angle between $`H^0`$ and $`h^0`$, it is given by:
$$\mathrm{tan}2\alpha =\mathrm{tan}2\beta \times \frac{M_A^2+M_Z^2}{M_A^2M_Z^2},\frac{\pi }{2}\alpha 0.$$
(7)
Note that these results are only valid at tree level and they become slightly different when one includes radiative corrections.
### 2.2 Production cross section at tree level
In $`e^+e^{}`$ collisions, charged Higgs bosons are pair produced through virtual photon and $`Z`$ boson exchange (and in top decays if $`M_H`$ is small enough). As for the neutral Higgs bosons, they can be produced through several mechanisms: $`WW`$ and $`ZZ`$ fusion processes, Higgsstrahlung or pair production. In this paper, we only focus on this latter process (note that CP conservation forbids virtual photon exchange). The Feynman diagrams of interest are shown in Figure 1. More details about the various production mechanisms and decay modes of MSSM Higgs bosons can be found in . Here, we only focus on the total pair production cross sections and we ignore the different contributions of the decay channels.
The tree level production cross section can be easily derived using the Feynman rules. If $`s_W\mathrm{sin}\theta _W`$, $`c_W\mathrm{cos}\theta _W`$ and $`\eta q^2/(q^2M_Z^2)`$, then the Born amplitudes $`a_{L,R}^{Born}`$ are:
* for charged Higgs bosons:
$$a_{L,R}^{Born}(H^+H^{})=1\frac{(12s_W^2)}{4s_W^2c_W^2}\eta g_{L,R}$$
(8)
* for neutral Higgs bosons:
$$a_{L,R}^{Born}(H^0A^0/h^0A^0)=\frac{i}{4s_W^2c_W^2}\eta g_{L,R}\left[Z_{ab}\right]$$
(9)
where $`g_L=2s_W^21`$, $`g_R=2s_W^2`$ and $`\left[Z_{ab}\right]=[\mathrm{sin}(\beta \alpha );\mathrm{cos}(\beta \alpha )]`$ for $`H^0A^0`$ and $`h^0A^0`$ final states, respectively.
In this paper, we use the following renormalization for the amplitude:
$$A=\frac{2e^2}{q^2}\overline{v}(e^+)(p/)(a_LP_L+a_RP_R)u(e^{}),P_{L,R}=\frac{1\gamma _5}{2}.$$
(10)
The differential tree level cross sections are then given by:
$$\frac{d\sigma _{L,R}^{Born}}{d\mathrm{cos}\theta }=\frac{\pi \alpha _{em}^2\beta _H^3}{8q^2}\times (1\mathrm{cos}^2\theta )\times |a_{L,R}^{Born}|^2.$$
(11)
Here, $`\beta _H`$ is the velocity of the outgoing Higgs bosons. If $`M_1`$ and $`M_2`$ are the masses of the two outgoing Higgs bosons, then $`\beta _H(M_1,M_2)`$ is defined by:
$`\beta _H`$ $`=`$ $`{\displaystyle \frac{2|p|}{\sqrt{s}}}={\displaystyle \frac{1}{s}}\times \sqrt{\left(s+M_1^2M_2^2\right)^24sM_1^2}`$ (12)
$`=`$ $`\sqrt{\left(1+{\displaystyle \frac{M_2+M_1}{\sqrt{s}}}\right)\left(1{\displaystyle \frac{M_2+M_1}{\sqrt{s}}}\right)\left(1+{\displaystyle \frac{M_2M_1}{\sqrt{s}}}\right)\left(1{\displaystyle \frac{M_2M_1}{\sqrt{s}}}\right)}.`$
After integration over $`\mathrm{cos}\theta `$, one gets:
* for charged Higgs bosons:
$$\sigma _{H^+H^{}}^{Born}=\frac{e^4}{48\pi s}\left(1\frac{4M_H^2}{s}\right)^{3/2}\times \left(1+\frac{2c_V^{}c_V}{1M_Z^2/s}+\frac{c_{V}^{}{}_{}{}^{2}(c_V^2+c_A^2)}{(1M_Z^2/s)^2}\right)$$
(13)
with $`c_V={\displaystyle \frac{1+4s_W^2}{4s_Wc_W}},c_A={\displaystyle \frac{1}{4s_Wc_W}},c_V^{}={\displaystyle \frac{1+2s_W^2}{2s_Wc_W}}`$.
* for neutral Higgs bosons:
$$\sigma _{H^0A^0/h^0A^0}^{Born}=\frac{e^4}{48\pi s}\times \left(\frac{8s_W^44s_W^2+1}{32s_W^4c_W^4}\right)\times \left[Z_{ab}\right]^2\times \frac{\beta _H^3(M_{H^0/h^0},M_A)}{(1M_Z^2/s)^2}.$$
(14)
In the decoupling limit ($`M_AM_Z`$ and $`M_AM_{H^0}`$), $`\mathrm{cos}(\beta \alpha )0`$ and $`e^+e^{}h^0A^0`$ is strongly suppressed, i.e. only the $`H^0A^0`$ pairs can be produced in $`e^+e^{}`$ collisions, with a tree level cross section given by:
$$\sigma _{H^0A^0}^{Born}\frac{e^4}{48\pi s}\left(1\frac{4M_A^2}{s}\right)^{3/2}\times \left(\frac{8s_W^44s_W^2+1}{32s_W^4c_W^4}\right)\times \frac{1}{(1M_Z^2/s)^2}.$$
(15)
Figure 2 shows the pair production cross section for the MSSM charged and neutral Higgs bosons in $`e^+e^{}`$ collisions, at tree level, as a function of $`M_A`$ and for various values of the centre-of-mass energy.
### 2.3 Complete amplitude for calculations at the one loop level
The analytical expressions of all electroweak one loop contributions to the MSSM Higgs bosons pair production cross sections are given in the following: $`e^\pm `$ self-energy and initial vertices in Section 3, $`\gamma `$ and $`Z`$ self-energies with counter terms in Section 4, final vertices and Higgs self-energy in Section 5, and box diagrams in Section 6. The complete renormalized amplitude used to calculate the cross section is the sum of Born and one loop terms:
$`A(e^+e^{}\text{Higgs pair})`$ $`=`$ $`A^{Born}(e^+e^{}\text{Higgs pair})`$ (16)
$`+`$ $`A^{in}(e^+e^{}\text{Higgs pair})`$
$`+`$ $`A^{RG}(e^+e^{}\text{Higgs pair})+A^{ct}(e^+e^{}\text{Higgs pair})`$
$`+`$ $`A^{fin}(e^+e^{}\text{Higgs pair})`$
$`+`$ $`A^{box}(e^+e^{}\text{Higgs pair}).`$
The notations used in our calculations, in particular when writing vertices in terms of real coupling constants, are described in Appendix A. In the following, we use a formalism which involves Passarino-Veltman functions, see Appendix B for details.
## 3 Contribution of initial vertices and $`๐^\mathbf{\pm }`$ self-energy
The amplitudes $`a_{L,R}^{in}`$ corresponding to initial triangles and $`e^\pm `$ self-energy are:
* for charged Higgs bosons:
$$a_{L,R}^{in}(H^+H^{})=\frac{\mathrm{\Gamma }^{in,\gamma }}{e}+\frac{(12s_W^2)\eta }{2s_Wc_W}\times \frac{\mathrm{\Gamma }^{in,Z}}{e}$$
(17)
* for neutral Higgs bosons:
$$a_{L,R}^{in}(H^0A^0/h^0A^0)=\frac{i\eta }{2s_Wc_W}\times \frac{\mathrm{\Gamma }^{in,Z}}{e}\times \left[Z_{ab}\right]$$
(18)
where we write $`\mathrm{\Gamma }^{in,V}=\mathrm{\Gamma }_L^{in,V}P_L+\mathrm{\Gamma }_R^{in,V}P_R`$ for $`V=\gamma \text{or}Z`$.
$`\mathrm{\Gamma }^{in,\gamma }`$ and $`\mathrm{\Gamma }^{in,Z}`$ are the same in the charged and neutral sectors, since they only depend on the initial state. They are obtained by summing various contributions:
$`\mathrm{\Gamma }^{in,V}`$ $`=`$ $`\mathrm{\Gamma }_{e^+e^{}}^V(W\nu W)\mathrm{\Gamma }_{e^+e^{}}^V(pinch)`$ (19)
$`+`$ $`\mathrm{\Gamma }_{e^+e^{}}^V(\nu W\nu )+\mathrm{\Gamma }_{e^+e^{}}^V(eZe)+\mathrm{\Gamma }_{e^+e^{}}^V(e\gamma e)`$
$`+`$ $`\mathrm{\Gamma }_{e^+e^{}}^V(\stackrel{~}{\chi }_j\stackrel{~}{\nu }_L\stackrel{~}{\chi }_i)+\mathrm{\Gamma }_{e^+e^{}}^V(\stackrel{~}{\chi }_j^0\stackrel{~}{e}\stackrel{~}{\chi }_i^0)`$
$`+`$ $`\mathrm{\Gamma }_{e^+e^{}}^V(\stackrel{~}{\nu }_L\stackrel{~}{\chi }_i\stackrel{~}{\nu }_L)+\mathrm{\Gamma }_{e^+e^{}}^V(\stackrel{~}{e}\stackrel{~}{\chi }_i^0\stackrel{~}{e})`$
$`+`$ $`\mathrm{\Gamma }_{e^+e^{}}^V(e.s.e).`$
The particles inside the initial triangle have internal masses $`m_1`$, $`m_2`$ and $`m_3`$. They are ordered clockwise, $`m_1`$ being the mass of the particle just after the junction involving the momentum $`q`$.
1) The contribution of the $`W\nu W`$ triangle is:
$$\mathrm{\Gamma }_{e^+e^{}}^V(W\nu W)=\frac{e\alpha _{em}}{8\pi s_W^2}f^V\stackrel{~}{C}_{WW}P_L,f^V=\{\begin{array}{c}1\text{for}V=\gamma \hfill \\ c_W/s_W\text{for}V=Z\hfill \end{array}$$
(20)
where $`\stackrel{~}{C}_{WW}=12C_{24}(W\nu W)+22q^2\left[C_0(W\nu W)+C_{11}(W\nu W)+C_{23}(W\nu W)\right]`$.
2) In $`e^+e^{}`$ annihilations, $`WW`$ contributions arise in the photon and $`Z`$ self-energies, as well as in the triangles connecting the photon and the $`Z`$ boson to the initial $`e^+e^{}`$ pair or to the final Higgs pair. Therefore, it is convenient to extract a certain part (so-called pinch) from such a triangle with two $`W`$ lines (in our case the $`W\nu W`$ triangle) and then to put it inside the photon and $`Z`$ self-energies contributions, in order to have universal charge renormalization :
$$\mathrm{\Gamma }^V(pinch)=\frac{e\alpha _{em}}{4\pi s_W^2}f^VB_0(WW,q^2)P_L,f^V=\{\begin{array}{c}1\text{for}V=\gamma \hfill \\ c_W/s_W\text{for}V=Z\hfill \end{array}.$$
(21)
3) As for the $`\nu W\nu `$ triangle, since neutrinos do not couple to photons, one has:
$$\mathrm{\Gamma }_{e^+e^{}}^\gamma (\nu W\nu )=0$$
(22)
while, for the $`Z`$ boson, one obtains:
$$\mathrm{\Gamma }_{e^+e^{}}^Z(\nu W\nu )=\frac{e\alpha _{em}}{16\pi s_W^3c_W}\stackrel{~}{C}_WP_L$$
(23)
where $`\stackrel{~}{C}_W=4C_{24}(\nu W\nu )2+2q^2\left[C_{11}(\nu W\nu )+C_{23}(\nu W\nu )\right]`$.
4) The contribution of the $`eZe`$ triangle is:
$$\mathrm{\Gamma }_{e^+e^{}}^\gamma (eZe)=\frac{e\alpha _{em}}{16\pi s_W^2c_W^2}\stackrel{~}{C}_Z\left[g_L^2P_L+g_R^2P_R\right]$$
(24)
or
$$\mathrm{\Gamma }_{e^+e^{}}^Z(eZe)=\frac{e\alpha _{em}}{32\pi s_W^3c_W^3}\stackrel{~}{C}_Z\left[g_L^3P_L+g_R^3P_R\right]$$
(25)
where $`\stackrel{~}{C}_Z=4C_{24}(eZe)2+2q^2\left[C_{11}(eZe)+C_{23}(eZe)\right]`$.
5) The contribution of the $`e\gamma e`$ triangle is:
$$\mathrm{\Gamma }_{e^+e^{}}^\gamma (e\gamma e)=\frac{e\alpha _{em}}{4\pi }\stackrel{~}{C}_\gamma \left[P_L+P_R\right]$$
(26)
or
$$\mathrm{\Gamma }_{e^+e^{}}^Z(e\gamma e)=\frac{e\alpha _{em}}{8\pi s_Wc_W}\stackrel{~}{C}_\gamma \left[g_LP_L+g_RP_R\right]$$
(27)
where $`\stackrel{~}{C}_\gamma =4C_{24}(e\gamma e)2+2q^2\left[C_{11}(e\gamma e)+C_{23}(e\gamma e)\right]`$.
6) The contribution of the $`\stackrel{~}{\chi }_j\stackrel{~}{\nu }_L\stackrel{~}{\chi }_i`$ triangles is:
$`\mathrm{\Gamma }_{e^+e^{}}^\gamma (\stackrel{~}{\chi }_i\stackrel{~}{\nu }_L\stackrel{~}{\chi }_i)`$ $`=`$ $`{\displaystyle \underset{i}{}}{\displaystyle \frac{e\alpha _{em}}{4\pi s_W^2}}|Z_{1i}^+|^2(2\stackrel{~}{C}_{24}^{ii}|M_{\stackrel{~}{\chi }_i}|^2\stackrel{~}{C}_0^{ii})P_L`$ (28)
or
$`\mathrm{\Gamma }_{e^+e^{}}^Z(\stackrel{~}{\chi }_j\stackrel{~}{\nu }_L\stackrel{~}{\chi }_i)`$ $`=`$ $`{\displaystyle \underset{ij}{}}{\displaystyle \frac{e\alpha _{em}}{8\pi s_W^3c_W}}Z_{1i}^+Z_{1j}^+\times `$ (29)
$`\{[Z_{1i}^+Z_{1j}^++\delta _{ij}(c_W^2s_W^2)]2\stackrel{~}{C}_{24}^{ij}`$
$`[Z_{1i}^{}Z_{1j}^{}+\delta _{ij}(c_W^2s_W^2)]M_{\stackrel{~}{\chi }_i}M_{\stackrel{~}{\chi }_j}\stackrel{~}{C}_0^{ij}\}P_L`$
where $`\stackrel{~}{C}_{24}^{ij}=C_{24}(\stackrel{~}{\chi }_j\stackrel{~}{\nu }_L\stackrel{~}{\chi }_i){\displaystyle \frac{1}{4}}+{\displaystyle \frac{q^2}{2}}\left[C_{12}(\stackrel{~}{\chi }_j\stackrel{~}{\nu }_L\stackrel{~}{\chi }_i)+C_{23}(\stackrel{~}{\chi }_j\stackrel{~}{\nu }_L\stackrel{~}{\chi }_i)\right]`$ and $`\stackrel{~}{C}_0^{ij}=C_0(\stackrel{~}{\chi }_j\stackrel{~}{\nu }_L\stackrel{~}{\chi }_i)`$.
7) As for the $`\stackrel{~}{\chi }_j^0\stackrel{~}{e}\stackrel{~}{\chi }_i^0`$ triangles, since neutralinos do not couple to photons, one has:
$$\mathrm{\Gamma }_{e^+e^{}}^\gamma (\stackrel{~}{\chi }_j^0\stackrel{~}{e}\stackrel{~}{\chi }_i^0)=0$$
(30)
while, for the $`Z`$ boson, one obtains:
$$\mathrm{\Gamma }_{e^+e^{}}^Z(\stackrel{~}{\chi }_j^0\stackrel{~}{e}\stackrel{~}{\chi }_i^0)=\underset{ij}{}\frac{e\alpha _{em}}{8\pi s_W^3c_W^3}\times \left[K_L^{ij}P_L+K_R^{ij}P_R\right]$$
(31)
by defining $`K_L^{ij}`$ and $`K_R^{ij}`$ as follows:
$`K_L^{ij}`$ $`=`$ $`{\displaystyle \frac{(Z_{1j}^Ns_W+Z_{2j}^Nc_W)(Z_{1i}^Ns_W+Z_{2i}^Nc_W)}{2}}\times `$ (32)
$`\left[2(Z_{3j}^NZ_{3i}^NZ_{4j}^NZ_{4i}^N)\stackrel{~}{C}_{24}^{ij}+M_{\stackrel{~}{\chi }_i^0}M_{\stackrel{~}{\chi }_j^0}(Z_{3j}^NZ_{3i}^NZ_{4j}^NZ_{4i}^N)\stackrel{~}{C}_0^{ij}\right]`$
$`K_R^{ij}`$ $`=`$ $`2(Z_{1j}^NZ_{1i}^N)s_W^2\times `$ (33)
$`\left[2(Z_{4j}^NZ_{4i}^NZ_{3j}^NZ_{3i}^N)\stackrel{~}{C}_{24}^{ij}+M_{\stackrel{~}{\chi }_i^0}M_{\stackrel{~}{\chi }_j^0}(Z_{4j}^NZ_{4i}^NZ_{3j}^NZ_{3i}^N)\stackrel{~}{C}_0^{ij}\right]`$
where $`\stackrel{~}{C}_{24}^{ij}=C_{24}(\stackrel{~}{\chi }_j^0\stackrel{~}{e}\stackrel{~}{\chi }_i^0){\displaystyle \frac{1}{4}}+{\displaystyle \frac{q^2}{2}}\left[C_{12}(\stackrel{~}{\chi }_j^0\stackrel{~}{e}\stackrel{~}{\chi }_i^0)+C_{23}(\stackrel{~}{\chi }_j^0\stackrel{~}{e}\stackrel{~}{\chi }_i^0)\right]`$ and $`\stackrel{~}{C}_0^{ij}=C_0(\stackrel{~}{\chi }_j^0\stackrel{~}{e}\stackrel{~}{\chi }_i^0)`$.
8) As for the $`\stackrel{~}{\nu }_L\stackrel{~}{\chi }_i\stackrel{~}{\nu }_L`$ triangles, since sneutrinos do not couple to photons, one has:
$$\mathrm{\Gamma }_{e^+e^{}}^\gamma (\stackrel{~}{\nu }_L\stackrel{~}{\chi }_i\stackrel{~}{\nu }_L)=0$$
(34)
while, for the $`Z`$ boson, one obtains:
$$\mathrm{\Gamma }_{e^+e^{}}^Z(\stackrel{~}{\nu }_L\stackrel{~}{\chi }_i\stackrel{~}{\nu }_L)=\underset{i}{}\frac{e\alpha _{em}}{4\pi s_W^3c_W}|Z_{1i}^+|^2\stackrel{~}{C}_{24}^iP_L$$
(35)
where $`\stackrel{~}{C}_{24}^i=C_{24}(\stackrel{~}{\nu }_L\stackrel{~}{\chi }_i\stackrel{~}{\nu }_L)`$.
9) The contribution of the $`\stackrel{~}{e}\stackrel{~}{\chi }_i^0\stackrel{~}{e}`$ triangles is:
$`\mathrm{\Gamma }_{e^+e^{}}^\gamma (\stackrel{~}{e}\stackrel{~}{\chi }_i^0\stackrel{~}{e})`$ $`=`$ $`{\displaystyle \underset{i}{}}{\displaystyle \frac{e\alpha _{em}}{4\pi s_W^2c_W^2}}\times `$ (36)
$`\left[|Z_{1i}^Ns_W+Z_{2i}^Nc_W|^2\stackrel{~}{C}_{24}^iP_L+4s_W^2|Z_{1i}^N|^2\stackrel{~}{C}_{24}^iP_R\right]`$
or
$`\mathrm{\Gamma }_{e^+e^{}}^Z(\stackrel{~}{e}\stackrel{~}{\chi }_i^0\stackrel{~}{e})`$ $`=`$ $`{\displaystyle \underset{i}{}}{\displaystyle \frac{e\alpha _{em}}{4\pi s_W^3c_W^3}}\times `$ (37)
$`\left[(s_W^2{\displaystyle \frac{1}{2}})|Z_{1i}^Ns_W+Z_{2i}^Nc_W|^2\stackrel{~}{C}_{24}^iP_L+4s_W^4|Z_{1i}^N|^2\stackrel{~}{C}_{24}^iP_R\right]`$
where $`\stackrel{~}{C}_{24}^i=C_{24}(\stackrel{~}{e}\stackrel{~}{\chi }_i^0\stackrel{~}{e})`$.
10) The electron self-energy ($`e.s.e`$) contributions are obtained as follows:
$$\mathrm{\Gamma }_{e^+e^{}}^\gamma (e.s.e)=e[\delta _LP_L+\delta _RP_R]$$
(38)
or
$$\mathrm{\Gamma }_{e^+e^{}}^Z(e.s.e)=\frac{e}{2s_Wc_W}[\delta _Lg_LP_L+\delta _Rg_RP_R]$$
(39)
where the following loops are taken into account: $`(W\nu )`$, $`(Ze)`$, $`(\gamma e)`$, $`(\stackrel{~}{\chi }\stackrel{~}{\nu })`$, $`(\stackrel{~}{\chi }^0\stackrel{~}{e})`$.
For the $`(W\nu )`$ loop, one has:
$$\delta _L(W\nu )=\frac{\alpha _{em}}{4\pi s_W^2}\left(B_1(W\nu ,0)+\frac{1}{2}\right),$$
(40)
$$\delta _R(W\nu )=0.$$
(41)
For the $`(Ze)`$ loop, one has:
$$\delta _L(Ze)=\frac{\alpha _{em}g_L^2}{8\pi s_W^2c_W^2}\left(B_1(Ze,0)+\frac{1}{2}\right),$$
(42)
$$\delta _R(Ze)=\frac{\alpha _{em}g_R^2}{8\pi s_W^2c_W^2}\left(B_1(Ze,0)+\frac{1}{2}\right).$$
(43)
For the $`(\gamma e)`$ loop, one has:
$$\delta _L(\gamma e)=\delta _R(\gamma e)=\frac{\alpha _{em}}{2\pi }\left(B_1(\gamma e,0)+\frac{1}{2}\right).$$
(44)
For each $`(\stackrel{~}{\chi }_i\stackrel{~}{\nu })`$ loop, one has:
$$\delta _L(\stackrel{~}{\chi }_i\stackrel{~}{\nu })=\frac{\alpha _{em}}{4\pi s_W^2}|Z_{1i}^+|^2B_1(\stackrel{~}{\chi }_i\stackrel{~}{\nu }_L,0),$$
(45)
$$\delta _R(\stackrel{~}{\chi }_i\stackrel{~}{\nu })=0.$$
(46)
For each $`(\stackrel{~}{\chi }_i^0\stackrel{~}{e})`$ loop, one has:
$$\delta _L(\stackrel{~}{\chi }_i^0\stackrel{~}{e})=\frac{\alpha _{em}}{8\pi s_W^2c_W^2}|Z_{1i}^Ns_W+Z_{2i}^Nc_W|^2B_1(\stackrel{~}{\chi }_i^0\stackrel{~}{e}_L,0),$$
(47)
$$\delta _R(\stackrel{~}{\chi }_i^0\stackrel{~}{e})=\frac{\alpha _{em}}{2\pi c_W^2}|Z_{1i}^N|^2B_1(\stackrel{~}{\chi }_i^0\stackrel{~}{e}_R,0).$$
(48)
## 4 Contribution of $`๐ธ`$ and $`๐`$ self-energies
### 4.1 Definition of gauge self-energy functions
The on-shell renormalization procedure that allows full determination of the MSSM Higgs sector at one loop, as well as of the corresponding counter terms, makes use of several gauge self-energy functions, which are detailed in the following of this section . Let us first define several useful expressions:
$`PV1(XY,q^2)`$ $`=`$ $`{\displaystyle \frac{M_X^2+M_Y^2}{2}}2B_{22}(XY,q^2){\displaystyle \frac{q^2}{6}}q^2\left[B_1(XY,q^2)+B_{21}(XY,q^2)\right],`$ (49)
$`PV2(XY,q^2)`$ $`=`$ $`10B_{22}(XY,q^2)+(4q^2+M_X^2+M_Y^2)B_0(XY,q^2)`$ (50)
$`+`$ $`A(M_X^2)+A(M_Y^2)2\left(M_X^2+M_Y^2{\displaystyle \frac{q^2}{3}}\right),`$
$`PV3(XY,q^2)`$ $`=`$ $`2B_{22}(XY,q^2){\displaystyle \frac{A(M_X^2)+A(M_Y^2)}{2}}+{\displaystyle \frac{(q^2M_X^2M_Y^2)}{2}}B_0(XY,q^2).`$ (51)
a) Photon self-energies:
The photon self-energy is defined as:
$$A_{\gamma \gamma }(q^2)=\mathrm{\Sigma }_{\gamma \gamma }(q^2)+A_{\gamma \gamma }(pinch).$$
(52)
The pinch term is given by:
$$A_{\gamma \gamma }(pinch)=\frac{\alpha _{em}}{\pi }q^2B_0(WW,q^2).$$
(53)
The self-energy term without pinch $`\mathrm{\Sigma }_{\gamma \gamma }(q^2)`$ is the sum of various contributions:
$`\mathrm{\Sigma }_{\gamma \gamma }(q^2)`$ $`=`$ $`\mathrm{\Sigma }_{\gamma \gamma }(\text{g+H})+\mathrm{\Sigma }_{\gamma \gamma }(ff)`$ (54)
$`+`$ $`\mathrm{\Sigma }_{\gamma \gamma }(\stackrel{~}{\chi }\stackrel{~}{\chi })+\mathrm{\Sigma }_{\gamma \gamma }(\stackrel{~}{f}\stackrel{~}{f}).`$
The contribution of the gauge and Higgs sectors is:
$`\mathrm{\Sigma }_{\gamma \gamma }(\text{g+H})`$ $`={\displaystyle \frac{\alpha _{em}}{2\pi }}`$ $`\{2B_{22}(HH,q^2)A(M_H^2)+6B_{22}(WW,q^2)`$ (55)
$`3A(M_W^2)+2q^2B_0(WW,q^2)+{\displaystyle \frac{q^2}{3}}\}.`$
The contribution of the fermion pairs is:
$`\mathrm{\Sigma }_{\gamma \gamma }(f\overline{f})`$ $`=`$ $`{\displaystyle \underset{f}{}}{\displaystyle \frac{\alpha _{em}N_c^fQ_f^2}{\pi }}\left\{PV1(ff,q^2)+M_f^2B_0(ff,q^2)\right\}.`$ (56)
The contribution of the chargino pairs is:
$`\mathrm{\Sigma }_{\gamma \gamma }(\stackrel{~}{\chi }\stackrel{~}{\chi })`$ $`=`$ $`{\displaystyle \underset{i}{}}{\displaystyle \frac{\alpha _{em}}{\pi }}\left\{PV1(\stackrel{~}{\chi }_i\stackrel{~}{\chi }_i,q^2)+M_{\stackrel{~}{\chi }_i}^2B_0(\stackrel{~}{\chi }_i\stackrel{~}{\chi }_i,q^2)\right\}.`$ (57)
The contribution of the sfermion pairs is:
$`\mathrm{\Sigma }_{\gamma \gamma }(\stackrel{~}{f}\stackrel{~}{f})`$ $`=`$ $`{\displaystyle \underset{\stackrel{~}{f}}{}}{\displaystyle \frac{\alpha _{em}N_c^fQ_f^2}{2\pi }}{\displaystyle \underset{i=1,2}{}}\left\{2B_{22}(\stackrel{~}{f}_i\stackrel{~}{f}_i,q^2)A(M_{\stackrel{~}{f}_i}^2)\right\}.`$ (58)
Here, $`\stackrel{~}{f}_1,\stackrel{~}{f}_2`$ account for $`\stackrel{~}{f}_L,\stackrel{~}{f}_R`$ in the case of unmixed sfermions, or for the physical states obtained after mixing (i.e. $`\stackrel{~}{t}_1`$, $`\stackrel{~}{t}_2`$, $`\stackrel{~}{b}_1`$, $`\stackrel{~}{b}_2`$ in the case of third generation squarks). The coupling between a photon and a sfermion pair is the same with and without mixing.
b) $`Z`$ self-energies:
The $`Z`$ boson self-energy is defined as:
$$A_{ZZ}(q^2)=\mathrm{\Sigma }_{ZZ}(q^2)+A_{ZZ}(pinch).$$
(59)
The pinch term is given by:
$$A_{ZZ}(pinch)=\frac{\alpha _{em}c_W^2}{\pi s_W^2}(q^2M_Z^2)B_0(WW,q^2).$$
(60)
The self-energy term without pinch $`\mathrm{\Sigma }_{ZZ}(q^2)`$ is the sum of various contributions:
$`\mathrm{\Sigma }_{ZZ}(q^2)`$ $`=`$ $`\mathrm{\Sigma }_{ZZ}(\text{g+H})+\mathrm{\Sigma }_{ZZ}(ff)`$ (61)
$`+`$ $`\mathrm{\Sigma }_{ZZ}(\stackrel{~}{\chi }\stackrel{~}{\chi })+\mathrm{\Sigma }_{ZZ}(\stackrel{~}{\chi }^0\stackrel{~}{\chi }^0)+\mathrm{\Sigma }_{ZZ}(\stackrel{~}{f}\stackrel{~}{f}).`$
The contribution of the gauge and Higgs sectors is:
$`\mathrm{\Sigma }_{ZZ}(\text{g+H})`$ $`={\displaystyle \frac{\alpha _{em}}{4\pi s_W^2c_W^2}}`$ $`\{{\displaystyle \frac{1}{4}}[A(M_{h^0}^2)+A(M_{H^0}^2)+A(M_A^2)+A(M_Z^2)]`$ (62)
$`+\mathrm{sin}^2(\beta \alpha )\left[M_Z^2B_0(Zh^0,q^2)B_{22}(Zh^0,q^2)B_{22}(A^0H^0,q^2)\right]`$
$`+\mathrm{cos}^2(\beta \alpha )\left[M_Z^2B_0(ZH^0,q^2)B_{22}(ZH^0,q^2)B_{22}(A^0h^0,q^2)\right]`$
$`{\displaystyle \frac{1}{2}}\mathrm{cos}^2(2\theta _W)\left[2B_{22}(HH,q^2)A(M_H^2)\right]`$
$`\left[8c_W^4+\mathrm{cos}^2(2\theta _W)\right]B_{22}(WW,q^2)`$
$`\left[4c_W^4q^2+2M_W^2\mathrm{cos}(2\theta _W)\right]B_0(WW,q^2)`$
$`+{\displaystyle \frac{1}{2}}[12c_W^44c_W^2+1]A(M_W^2){\displaystyle \frac{2}{3}}c_W^4q^2\}.`$
The contribution of the fermion pairs is:
$`\mathrm{\Sigma }_{ZZ}(f\overline{f})`$ $`=`$ $`{\displaystyle \underset{f}{}}{\displaystyle \frac{\alpha _{em}N_c^f}{4\pi s_W^2c_W^2}}\left\{(g_{Vf}^2+g_{Af}^2)PV1(M_f^2,q^2)+(g_{Vf}^2g_{Af}^2)M_f^2B_0(f\overline{f},q^2)\right\}`$ (63)
where $`g_{Vf}=T_f^3(14|Q_f|s_W^2)`$ and $`g_{Af}=T_f^3`$.
The contribution of the chargino pairs is:
$`\mathrm{\Sigma }_{ZZ}(\stackrel{~}{\chi }\stackrel{~}{\chi })`$ $`={\displaystyle \frac{2}{16\pi ^2}}{\displaystyle \underset{ij}{}}`$ $`\{PV1(\stackrel{~}{\chi }_i\stackrel{~}{\chi }_j,q^2)[๐ช_{ij}^{ZL}๐ช_{ij}^{ZL}+๐ช_{ij}^{ZR}๐ช_{ij}^{ZR}]`$ (64)
$`+M_{\stackrel{~}{\chi }_i}M_{\stackrel{~}{\chi }_j}[๐ช_{ij}^{ZL}๐ช_{ij}^{ZR}+๐ช_{ij}^{ZL}๐ช_{ij}^{ZR}]B_0(\stackrel{~}{\chi }_i\stackrel{~}{\chi }_j,q^2)\}.`$
The contribution of the neutralino pairs is:
$`\mathrm{\Sigma }_{ZZ}(\stackrel{~}{\chi }^0\stackrel{~}{\chi }^0)`$ $`={\displaystyle \frac{1}{16\pi ^2}}{\displaystyle \underset{ij}{}}`$ $`\{PV1(\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j^0,q^2)[๐ช_{ij}^{0L}๐ช_{ij}^{0L}+๐ช_{ij}^{0R}๐ช_{ij}^{0R}]`$ (65)
$`+M_{\stackrel{~}{\chi }_i^0}M_{\stackrel{~}{\chi }_j^0}[๐ช_{ij}^{0L}๐ช_{ij}^{0R}+๐ช_{ij}^{0L}๐ช_{ij}^{0R}]B_0(\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j^0,q^2)\}.`$
The contribution of sfermion pairs is:
* for unmixed sfermions:
$$\mathrm{\Sigma }_{ZZ}^{light}(\stackrel{~}{f}\stackrel{~}{f})=\underset{\stackrel{~}{f}_{L,R}}{}\frac{\alpha _{em}N_c^f}{2\pi }\left(\frac{g_{Z\stackrel{~}{f}\stackrel{~}{f}}^0}{e}\right)^2\left\{2B_{22}(\stackrel{~}{f}\stackrel{~}{f},q^2)A(M_{\stackrel{~}{f}}^2)\right\},$$
(66)
* with sfermion mixing (third generation squarks):
$`\mathrm{\Sigma }_{ZZ}^{heavy}(\stackrel{~}{f}\stackrel{~}{f})`$ $`={\displaystyle \frac{3\alpha _{em}}{2\pi s_W^2c_W^2}}{\displaystyle \underset{\stackrel{~}{f}=\stackrel{~}{t},\stackrel{~}{b}}{}}`$ $`\{{\displaystyle \frac{c_{\stackrel{~}{f}}^2s_{\stackrel{~}{f}}^2}{2}}[B_{22}(\stackrel{~}{f}_1\stackrel{~}{f}_2,q^2)+B_{22}(\stackrel{~}{f}_2\stackrel{~}{f}_1,q^2)]`$ (67)
$`+2(T_{f_L}^3c_{\stackrel{~}{f}}^2s_W^2Q_f)^2B_{22}(\stackrel{~}{f}_1\stackrel{~}{f}_1,q^2)`$
$`\left[c_{\stackrel{~}{f}}^2(T_{f_L}^3s_W^2Q_f)^2+s_{\stackrel{~}{f}}^2Q_f^2s_W^4\right]A(M_{\stackrel{~}{f}_1}^2)`$
$`+2(T_{f_L}^3s_{\stackrel{~}{f}}^2s_W^2Q_f)^2B_{22}(\stackrel{~}{f}_2\stackrel{~}{f}_2,q^2)`$
$`[s_{\stackrel{~}{f}}^2(T_{f_L}^3s_W^2Q_f)^2+c_{\stackrel{~}{f}}^2Q_f^2s_W^4]A(M_{\stackrel{~}{f}_2}^2)\}.`$
c) Mixed $`\gamma Z`$ self-energies:
The mixed $`\gamma Z`$ self-energy is defined as:
$$A_{\gamma Z}(q^2)=\mathrm{\Sigma }_{\gamma Z}(q^2)+A_{\gamma Z}(pinch).$$
(68)
The pinch term is given by:
$$A_{\gamma Z}(pinch)=\frac{\alpha _{em}c_W}{\pi s_W}(q^2\frac{M_Z^2}{2})B_0(WW,q^2).$$
(69)
The self-energy term without pinch $`\mathrm{\Sigma }_{\gamma Z}(q^2)`$ is the sum of various contributions:
$`\mathrm{\Sigma }_{\gamma Z}(q^2)`$ $`=`$ $`\mathrm{\Sigma }_{\gamma Z}(\text{g+H})+\mathrm{\Sigma }_{\gamma Z}(ff)`$ (70)
$`+`$ $`\mathrm{\Sigma }_{\gamma Z}(\stackrel{~}{\chi }\stackrel{~}{\chi })+\mathrm{\Sigma }_{\gamma Z}(\stackrel{~}{f}\stackrel{~}{f}).`$
The contribution of the gauge and Higgs sectors can be expressed in several ways. Here, we choose the definition given in :
$`\mathrm{\Sigma }_{\gamma Z}(\text{g+H})`$ $`={\displaystyle \frac{\alpha _{em}}{4\pi }}`$ $`\{2{\displaystyle \frac{c_W^2s_W^2}{s_Wc_W}}[B_{22}(HH,q^2)+B_{22}(WW,q^2)]`$ (71)
$`+{\displaystyle \frac{c_W^2s_W^2}{s_Wc_W}}\left[A(M_H^2)+A(M_W^2)\right]+{\displaystyle \frac{c_W}{s_W}}\left[6A(M_W^2)4M_W^2\right]`$
$`+2{\displaystyle \frac{c_W}{s_W}}B_{22}(WW,q^2)2s_Wc_WM_Z^2B_0(WW,q^2)`$
$`{\displaystyle \frac{c_W}{s_W}}PV2(WW,q^2)\}.`$
The contribution of the fermion pairs is:
$`\mathrm{\Sigma }_{\gamma Z}(f\overline{f})`$ $`=`$ $`{\displaystyle \underset{f}{}}{\displaystyle \frac{\alpha _{em}N_c^fQ_fg_{Vf}}{2\pi s_Wc_W}}\left\{PV1(ff,q^2)+M_f^2B_0(ff,q^2)\right\}.`$ (72)
The contribution of the chargino pairs is:
$`\mathrm{\Sigma }_{\gamma Z}(\stackrel{~}{\chi }\stackrel{~}{\chi })`$ $`{\displaystyle \frac{2}{16\pi ^2}}{\displaystyle \underset{ij}{}}`$ $`\{PV1(\stackrel{~}{\chi }_i\stackrel{~}{\chi }_j,q^2)[๐ช_{ij}^{\gamma L}๐ช_{ij}^{ZL}+๐ช_{ij}^{\gamma R}๐ช_{ij}^{ZR}]`$ (73)
$`+M_{\stackrel{~}{\chi }_i}M_{\stackrel{~}{\chi }_j}[๐ช_{ij}^{\gamma L}๐ช_{ij}^{ZR}+๐ช_{ij}^{ZL}๐ช_{ij}^{\gamma R}]B_0(\stackrel{~}{\chi }_i\stackrel{~}{\chi }_j,q^2)\}.`$
The contribution of the sfermion pairs is:
* for unmixed sfermions:
$$\mathrm{\Sigma }_{\gamma Z}^{light}(\stackrel{~}{f}\stackrel{~}{f})=\frac{\alpha _{em}}{2\pi }\underset{\stackrel{~}{f}_{L,R}}{}N_c^fQ_f\left(\frac{g_{Z\stackrel{~}{f}\stackrel{~}{f}}^0}{e}\right)\times \left\{2B_{22}(\stackrel{~}{f}\stackrel{~}{f},q^2)A(M_{\stackrel{~}{f}}^2)\right\},$$
(74)
* with sfermion mixing (third generation squarks):
$`\mathrm{\Sigma }_{\gamma Z}^{heavy}(\stackrel{~}{f}\stackrel{~}{f})`$ $`={\displaystyle \frac{3\alpha _{em}}{2\pi s_Wc_W}}{\displaystyle \underset{\stackrel{~}{f}=\stackrel{~}{t},\stackrel{~}{b}}{}}Q_f`$ $`\{(T_{f_L}^3c_{\stackrel{~}{f}}^2s_W^2Q_f)[2B_{22}(\stackrel{~}{f}_1\stackrel{~}{f}_1,q^2)A(M_{\stackrel{~}{f}_1}^2)]`$ (75)
$`+(T_{f_L}^3s_{\stackrel{~}{f}}^2s_W^2Q_f)[2B_{22}(\stackrel{~}{f}_2\stackrel{~}{f}_2,q^2)A(M_{\stackrel{~}{f}_2}^2)]\}.`$
d) $`W`$ self-energies:
The self-energy term without pinch $`\mathrm{\Sigma }_{WW}(q^2)`$ is the sum of various contributions:
$`\mathrm{\Sigma }_{WW}(q^2)`$ $`=`$ $`\mathrm{\Sigma }_{WW}(\text{g+H})+\mathrm{\Sigma }_{WW}(ff^{})`$ (76)
$`+`$ $`\mathrm{\Sigma }_{WW}(\stackrel{~}{\chi }\stackrel{~}{\chi }^0)+\mathrm{\Sigma }_{WW}(\stackrel{~}{f}\stackrel{~}{f}^{}).`$
The contribution of the gauge and Higgs sectors can be expressed in several ways. Here, we choose the definition given in :
$`\mathrm{\Sigma }_{WW}(\text{g+H})`$ $`={\displaystyle \frac{\alpha _{em}}{4\pi s_W^2}}`$ $`\{\mathrm{sin}^2(\beta \alpha )[B_{22}(HH^0,q^2)+B_{22}(Wh^0,q^2)]`$ (77)
$`\mathrm{cos}^2(\beta \alpha )\left[B_{22}(Hh^0,q^2)+B_{22}(WH^0,q^2)\right]`$
$`B_{22}(WZ,q^2)B_{22}(HA^0,q^2)`$
$`+2s_W^2B_{22}(\gamma W,q^2)+2c_W^2B_{22}(WZ,q^2)`$
$`+{\displaystyle \frac{1}{4}}\left[A(M_{H^0}^2)+A(M_{h^0}^2)+A(M_Z^2)+A(M_A^2)\right]`$
$`+{\displaystyle \frac{1}{2}}\left[A(M_W^2)+A(M_H^2)\right]`$
$`+M_W^2\left[\mathrm{sin}^2(\beta \alpha )B_0(h^0W,q^2)+\mathrm{cos}^2(\beta \alpha )B_0(H^0W,q^2)\right]`$
$`+M_W^2\left[s_W^2B_0(W\gamma ,q^2)+{\displaystyle \frac{s_W^4}{c_W^2}}B_0(WZ,q^2)\right]`$
$`+\left[3A(M_W^2)2M_W^2\right]+c_W^2\left[3A(M_Z^2)2M_Z^2\right]`$
$`c_W^2PV2(ZW,q^2)s_W^2PV2(\gamma W,q^2)\}.`$
The contribution of the fermion pairs is:
$$\mathrm{\Sigma }_{WW}(ff^{})=\underset{(ff^{})}{}\frac{\alpha _{em}N_c^f}{4\pi s_W^2}PV_3(ff^{},q^2).$$
(78)
The contribution of the gaugino pairs is:
$`\mathrm{\Sigma }_{WW}(\stackrel{~}{\chi }\stackrel{~}{\chi }^0)`$ $`={\displaystyle \frac{\alpha _{em}}{2\pi s_W^2}}{\displaystyle \underset{ij}{}}`$ $`\{(๐ช_{ij}^{WL}๐ช_{ij}^{WL}+๐ช_{ij}^{WR}๐ช_{ij}^{WR})PV_3(\stackrel{~}{\chi }_i\stackrel{~}{\chi }_j^0,q^2)`$ (79)
$`+(๐ช_{ij}^{WL}๐ช_{ij}^{WR}+๐ช_{ij}^{WL}๐ช_{ij}^{WR})M_{\stackrel{~}{\chi }_i}M_{\stackrel{~}{\chi }_j^0}B_0(\stackrel{~}{\chi }_i\stackrel{~}{\chi }_j^0,q^2)\}.`$
The contribution of the sfermion pairs is:
* for unmixed sfermions:
$$\mathrm{\Sigma }_{WW}^{light}(\stackrel{~}{f}\stackrel{~}{f}^{})=\frac{\alpha _{em}}{2\pi s_W^2}\underset{(ff^{})}{}N_c^f\left[B_{22}(\stackrel{~}{f}\stackrel{~}{f}^{},q^2)\frac{A(M_{\stackrel{~}{f}}^2)+A(M_{\stackrel{~}{f}^{}}^2)}{4}\right],$$
(80)
* with sfermion mixing (third generation squarks):
$`\mathrm{\Sigma }_{WW}^{heavy}(\stackrel{~}{f}\stackrel{~}{f}^{})`$ $`={\displaystyle \frac{3\alpha _{em}}{2\pi s_W^2}}`$ $`\{c_{\stackrel{~}{t}}^2c_{\stackrel{~}{b}}^2B_{22}(\stackrel{~}{t}_1\stackrel{~}{b}_1,q^2)+c_{\stackrel{~}{t}}^2s_{\stackrel{~}{b}}^2B_{22}(\stackrel{~}{t}_1\stackrel{~}{b}_2,q^2)`$ (81)
$`+s_{\stackrel{~}{t}}^2c_{\stackrel{~}{b}}^2B_{22}(\stackrel{~}{t}_2\stackrel{~}{b}_1,q^2)+s_{\stackrel{~}{t}}^2s_{\stackrel{~}{b}}^2B_{22}(\stackrel{~}{t}_2\stackrel{~}{b}_2,q^2)`$
$`{\displaystyle \frac{1}{4}}[c_{\stackrel{~}{t}}^2A(M_{\stackrel{~}{t}_1}^2)+s_{\stackrel{~}{t}}^2A(M_{\stackrel{~}{t}_2}^2)+c_{\stackrel{~}{b}}^2A(M_{\stackrel{~}{b}_1}^2)+s_{\stackrel{~}{b}}^2A(M_{\stackrel{~}{b}_2}^2)]\}.`$
### 4.2 Charged Higgs sector
For $`e^+e^{}H^+H^{}`$, the on-shell renormalization procedure leads to the following $`RG`$ terms:
$$a_{L,R}^{RG}(H^+H^{})=\left[\frac{\eta ^2(12s_W^2)g_{L,R}}{4s_W^2c_W^2}\right]\frac{A_{ZZ}(q^2)}{q^2}\left[\frac{\eta (12s_W^2g_{L,R})}{2s_Wc_W}\right]\frac{A_{\gamma Z}(q^2)}{q^2}\frac{A_{\gamma \gamma }(q^2)}{q^2}$$
(82)
with the corresponding counter terms:
$`a_{L,R}^{ct}(H^+H^{})`$ $`=`$ $`\left[\mathrm{\Pi }_{\gamma \gamma }(0)+{\displaystyle \frac{2s_W\mathrm{\Sigma }_{\gamma Z}(0)}{c_WM_Z^2}}\right]\times \left[1+\eta g_{L,R}{\displaystyle \frac{2s_W^21}{4s_W^2c_W^2}}\right]`$ (83)
$`+`$ $`{\displaystyle \frac{\eta g_{L,R}(2s_W^21)}{4s_W^2c_W^2}}\times \left[{\displaystyle \frac{\mathrm{\Sigma }_{ZZ}(M_Z^2)}{q^2M_Z^2}}\right]`$
$`+`$ $`{\displaystyle \frac{\eta }{4s_W^2c_W^2}}\times [{\displaystyle \frac{\mathrm{\Sigma }_{ZZ}(M_Z^2)}{M_Z^2}}{\displaystyle \frac{\mathrm{\Sigma }_{WW}(M_W^2)}{M_W^2}})]\times [{\displaystyle \frac{g_L}{s_W^2}}P_L+g_RP_R].`$
Here, $`\mathrm{\Pi }_{\gamma \gamma }(q^2){\displaystyle \frac{\mathrm{\Sigma }_{\gamma \gamma }(q^2)}{q^2}}`$ (no pinch term) and $`\mathrm{\Pi }_{\gamma \gamma }(0)`$ is thus simply obtained as follows:
$$\mathrm{\Pi }_{\gamma \gamma }(0)=\left(\frac{d\mathrm{\Sigma }_{\gamma \gamma }}{dq^2}\right)_{q^2=0}.$$
(84)
### 4.3 Neutral Higgs sector
For $`e^+e^{}H^0A^0/h^0A^0`$, the on-shell renormalization procedure leads to the following $`RG`$ terms:
$$a_{L,R}^{RG}(H^0A^0/h^0A^0)=\frac{i}{q^2}\left[Z_{ab}\right]\times \left(\frac{\eta ^2g_{L,R}}{4s_W^2c_W^2}A_{ZZ}(q^2)\frac{\eta }{2s_Wc_W}A_{\gamma Z}(q^2)\right).$$
(85)
As for the counter terms, we only consider those corresponding to electroweak couplings and gauge boson masses here (the counter terms corresponding to $`H^0A^0`$ or $`h^0A^0`$ final states will be calculated in Section 5.3.9):
$`a_{L,R}^{ct}(H^0A^0/h^0A^0)`$ $`=`$ $`i\eta \left[Z_{ab}\right]\left[{\displaystyle \frac{12s_W^2+2s_W^4}{4s_W^4c_W^2}}P_L+{\displaystyle \frac{1}{2c_W^2}}P_R\right]\left[{\displaystyle \frac{\mathrm{\Sigma }_{WW}(M_W^2)}{M_W^2}}{\displaystyle \frac{\mathrm{\Sigma }_{ZZ}(M_Z^2)}{M_Z^2}}\right]`$ (86)
$``$ $`i\eta \left[Z_{ab}\right]{\displaystyle \frac{g_{L,R}}{4s_W^2c_W^2}}\left[\mathrm{\Pi }_{\gamma \gamma }(0)+{\displaystyle \frac{\mathrm{\Sigma }_{ZZ}(M_Z^2)}{q^2M_Z^2}}+{\displaystyle \frac{2s_W\mathrm{\Sigma }_{\gamma Z}(0)}{c_WM_Z^2}}\right].`$
## 5 Contribution of final vertices and Higgs self-energies
### 5.1 Diagram structures for final triangles
Several useful expressions are needed when estimating the contributions of the final vertices. The particles inside the final triangle have internal masses $`m_1`$, $`m_2`$ and $`m_3`$. They are ordered clockwise, $`m_1`$ being the mass of the particle just after the junction involving the momentum $`q`$.
Let $`P_{f1}`$ and $`P_{f2}`$ (respectively $`M_1`$ and $`M_2`$) be the momenta (respectively the masses) of the two outgoing Higgs bosons (i.e. $`H^+H^{}`$ or $`H^0A^0`$ or $`h^0A^0`$), then one has:
$`P_{f1}^2`$ $`=`$ $`M_1^2,`$ (87)
$`P_{f2}^2`$ $`=`$ $`M_2^2,`$ (88)
$`P_{f1}P_{f2}`$ $`=`$ $`{\displaystyle \frac{q^2(M_1^2+M_2^2)}{2}}.`$ (89)
a) Tri1-type triangles:
$`๐_1`$ $`=`$ $`{\displaystyle \frac{1}{6}}+6(C_{001}C_{002})+P_{f1}^2C_{111}P_{f2}^2C_{222}`$ (90)
$`+`$ $`(2P_{f1}P_{f2}P_{f1}^2)C_{112}+(P_{f2}^22P_{f1}P_{f2})C_{122}`$
$`+`$ $`2\left[P_{f1}P_{f2}C_{21}P_{f2}^2C_{22}+(P_{f2}^2P_{f1}P_{f2})C_{23}C_{24}\right]`$
$``$ $`(2P_{f1}P_{f2}+P_{f1}^2)(C_{11}C_{12}).`$
b) Tri2-type triangles:
$`๐_2`$ $`=`$ $`(8P_{f1}P_{f2}+6P_{f1}^2+2P_{f2}^2)C_0+(8P_{f1}P_{f2}+7P_{f1}^2+P_{f2}^2)C_{11}`$ (91)
$`+`$ $`(P_{f1}^2P_{f2}^2)C_{12}+(2P_{f1}P_{f2}+2P_{f1}^2)C_{21}+(2P_{f1}P_{f2}+2P_{f2}^2)C_{22}`$
$`+`$ $`(4P_{f1}P_{f2}+2P_{f1}^2+2P_{f2}^2)C_{23}+12C_{24}2.`$
c) Tri3-type triangles:
$`๐_3`$ $`=`$ $`{\displaystyle \frac{1}{6}}+6(C_{001}C_{002})+P_{f1}^2C_{111}P_{f2}^2C_{222}+(2P_{f1}P_{f2}P_{f1}^2)C_{112}`$ (92)
$`+`$ $`(P_{f2}^22P_{f1}P_{f2})C_{122}+P_{f1}^2C_{21}(2P_{f1}P_{f2}+P_{f2}^2)C_{22}`$
$``$ $`2P_{f1}^2C_{23}q^2C_{12}2C_{24}+{\displaystyle \frac{1}{2}},`$
$`๐_3^{}`$ $`=`$ $`C_{11}C_{12},`$ (93)
$`๐_3^{\prime \prime }`$ $`=`$ $`C_0+C_{11}C_{12}.`$ (94)
d) Tri4-type triangles:
$`๐_4`$ $`=`$ $`C_{12}C_{11}2C_0.`$ (95)
e) Tri5-type triangles:
$`๐_5`$ $`=`$ $`C_{11}C_{12}C_0.`$ (96)
f) Tri6-type triangles:
$`๐_6`$ $`=`$ $`C_{11}C_{12}.`$ (97)
### 5.2 Charged Higgs sector
The amplitudes $`a_{L,R}^{fin}`$ corresponding to final vertices with $`H^+H^{}`$ states are:
$$a_{L,R}^{fin}(H^+H^{})=\frac{\mathrm{\Gamma }^{fin,\gamma }(H^+H^{})}{2e}\frac{\eta g_{L,R}}{2s_Wc_W}\times \frac{\mathrm{\Gamma }^{fin,Z}(H^+H^{})}{2e}.$$
(98)
Here, $`\mathrm{\Gamma }^{fin,\gamma }(H^+H^{})`$ and $`\mathrm{\Gamma }^{fin,Z}(H^+H^{})`$ are obtained by summing the contributions of various triangles and of charged Higgs self-energy terms, as detailed in the following.
For the photon, one has:
$`\mathrm{\Gamma }^{fin,\gamma }(H^+H^{})`$ $`=`$ $`\mathrm{\Gamma }_{H^+H^{}}^\gamma (\text{1ch})+\mathrm{\Gamma }_{H^+H^{}}^\gamma (2)\mathrm{\Gamma }_{H^+H^{}}^\gamma (2,pinch)`$ (99)
$`+`$ $`\mathrm{\Gamma }_{H^+H^{}}^\gamma (3f)+\mathrm{\Gamma }_{H^+H^{}}^\gamma (\stackrel{~}{\chi }\stackrel{~}{\chi }^0\stackrel{~}{\chi })+\mathrm{\Gamma }_{H^+H^{}}^\gamma (\text{6ch})+\mathrm{\Gamma }_{H^+H^{}}^\gamma (6\stackrel{~}{f})`$
$`+`$ $`\mathrm{\Gamma }_{H^+H^{}}^\gamma (\text{4-leg})+\mathrm{\Gamma }_{H^+H^{}}^\gamma (H.s.e).`$
For the $`Z`$ boson, one has:
$`\mathrm{\Gamma }^{fin,Z}(H^+H^{})`$ $`=`$ $`\mathrm{\Gamma }_{H^+H^{}}^Z(\text{1ch})+\mathrm{\Gamma }_{H^+H^{}}^Z(\text{1n})+\mathrm{\Gamma }_{H^+H^{}}^Z(2)\mathrm{\Gamma }_{H^+H^{}}^Z(2,pinch)`$ (100)
$`+`$ $`\mathrm{\Gamma }_{H^+H^{}}^Z(3f)+\mathrm{\Gamma }_{H^+H^{}}^Z(\stackrel{~}{\chi }\stackrel{~}{\chi }^0\stackrel{~}{\chi })+\mathrm{\Gamma }_{H^+H^{}}^Z(\stackrel{~}{\chi }^0\stackrel{~}{\chi }\stackrel{~}{\chi }^0)+\mathrm{\Gamma }_{H^+H^{}}^Z(4)`$
$`+`$ $`\mathrm{\Gamma }_{H^+H^{}}^Z(\text{6ch})+\mathrm{\Gamma }_{H^+H^{}}^Z(\text{6n})+\mathrm{\Gamma }_{H^+H^{}}^Z(6\stackrel{~}{f})`$
$`+`$ $`\mathrm{\Gamma }_{H^+H^{}}^Z(\text{4-leg})+\mathrm{\Gamma }_{H^+H^{}}^Z(H.s.e).`$
#### 5.2.1 Tri1-type triangles
The Tri1-type triangles contribute to $`\mathrm{\Gamma }^{fin,\gamma }(H^+H^{})`$ with:
$`\mathrm{\Gamma }_{H^+H^{}}^\gamma (\text{1ch})`$ $`=`$ $`{\displaystyle \frac{e^3}{8\pi ^2}}\left[๐_1(H\gamma H)+\left({\displaystyle \frac{12s_W^2}{2s_Wc_W}}\right)^2๐_1(HZH)\right].`$ (101)
The Tri1-type triangles contribute to $`\mathrm{\Gamma }^{fin,Z}(H^+H^{})`$ with:
$`\mathrm{\Gamma }_{H^+H^{}}^Z(\text{1ch})`$ $`=`$ $`{\displaystyle \frac{e^3}{8\pi ^2}}\left({\displaystyle \frac{12s_W^2}{2s_Wc_W}}\right)\left[๐_1(H\gamma H)+\left({\displaystyle \frac{12s_W^2}{2s_Wc_W}}\right)^2๐_1(HZH)\right],`$ (102)
$`\mathrm{\Gamma }_{H^+H^{}}^Z(\text{1n})`$ $`={\displaystyle \frac{e^3}{64\pi ^2s_W^3c_W}}\times `$ $`\{\mathrm{sin}^2(\beta \alpha )[๐_1(H^0WA^0)+๐_1(A^0WH^0)]`$ (103)
$`+\mathrm{cos}^2(\beta \alpha )[๐_1(h^0WA^0)+๐_1(A^0Wh^0)]\}.`$
#### 5.2.2 Tri2-type triangles
With $`f^V=\{\begin{array}{c}1\text{for}V=\gamma \hfill \\ c_W/s_W\text{for}V=Z\hfill \end{array}`$ , the contribution of the Tri2-type triangles is:
$`\mathrm{\Gamma }_{H^+H^{}}^V(2)`$ $`={\displaystyle \frac{e^3f^V}{64\pi ^2s_W^2}}\times `$ $`\{\mathrm{sin}^2(\beta \alpha )๐_2(WH^0W)`$ (104)
$`+\mathrm{cos}^2(\beta \alpha )๐_2(Wh^0W)`$
$`+๐_2(WA^0W)\}.`$
However, one must also take into account the pinch term:
$`\mathrm{\Gamma }_{H^+H^{}}^V(2,pinch)`$ $`=`$ $`{\displaystyle \frac{e^3f^V}{8\pi ^2s_W^2}}\times B_0(WW,q^2).`$ (105)
#### 5.2.3 Tri3-type triangles
The Tri3-type triangles contribute to both $`\mathrm{\Gamma }^{fin,\gamma }(H^+H^{})`$ and $`\mathrm{\Gamma }^{fin,Z}(H^+H^{})`$ with:
$`\mathrm{\Gamma }_{H^+H^{}}^V(3f)`$ $`=`$ $`{\displaystyle \frac{1}{8\pi ^2}}{\displaystyle \underset{(ff^{})}{}}{\displaystyle \frac{N_c^fe^3}{2s_W^2M_W^2}}\times `$ (106)
$`\{[g_{VRf}M_f^2\text{cot}^2\beta +g_{VLf}M_f^{}^2\mathrm{tan}^2\beta ]๐_3(ff^{}f)`$
$`\left[g_{VRf^{}}M_f^{}^2\mathrm{tan}^2\beta +g_{VLf^{}}M_f^2\text{cot}^2\beta \right]๐_3(f^{}ff^{})`$
$`+2M_f^2M_f^{}^2(g_{VRf}+g_{VLf})๐_3^{}(ff^{}f)`$
$`2M_f^2M_f^{}^2(g_{VRf^{}}+g_{VLf^{}})๐_3^{}(f^{}ff^{})`$
$`+M_f^2\left[g_{VLf}M_f^2\text{cot}^2\beta +g_{VRf}M_f^{}^2\mathrm{tan}^2\beta \right]๐_3^{\prime \prime }(ff^{}f)`$
$`M_f^{}^2[g_{VLf^{}}M_f^{}^2\mathrm{tan}^2\beta +g_{VRf^{}}M_f^2\text{cot}^2\beta ]๐_3^{\prime \prime }(f^{}ff^{})\}`$
$`\mathrm{\Gamma }_{H^+H^{}}^V(\stackrel{~}{\chi }\stackrel{~}{\chi }^0\stackrel{~}{\chi })`$ $`={\displaystyle \frac{1}{8\pi ^2}}{\displaystyle \underset{ijk}{}}`$ $`\{[๐ช_{kj}^{VL}c_{Hji}^Rc_{Hki}^R+๐ช_{kj}^{VR}c_{Hji}^Lc_{Hki}^L]๐_3(\stackrel{~}{\chi }_k\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j)`$ (107)
$`+M_{\stackrel{~}{\chi }_j}M_{\stackrel{~}{\chi }_i^0}\left[๐ช_{kj}^{VL}c_{Hji}^Lc_{Hki}^R+๐ช_{kj}^{VR}c_{Hji}^Rc_{Hki}^L\right]๐_3^{}(\stackrel{~}{\chi }_k\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j)`$
$`+M_{\stackrel{~}{\chi }_i^0}M_{\stackrel{~}{\chi }_k}\left[๐ช_{kj}^{VL}c_{Hji}^Rc_{Hki}^L+๐ช_{kj}^{VR}c_{Hji}^Lc_{Hki}^R\right]๐_3^{}(\stackrel{~}{\chi }_k\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j)`$
$`+M_{\stackrel{~}{\chi }_j}M_{\stackrel{~}{\chi }_k}[๐ช_{kj}^{VL}c_{Hji}^Lc_{Hki}^L+๐ช_{kj}^{VR}c_{Hji}^Rc_{Hki}^R]๐_3^{\prime \prime }(\stackrel{~}{\chi }_k\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j)\},`$
where $`V=\gamma \text{or}Z`$, and where $`(f,f^{})`$ is either $`(q_u,q_d)`$ or $`(\nu _{\mathrm{}},\mathrm{})`$ for each fermion generation.
In addition, Tri3-type triangles contribute to $`\mathrm{\Gamma }^{fin,Z}(H^+H^{})`$ with the following term:
$`\mathrm{\Gamma }_{H^+H^{}}^Z(\stackrel{~}{\chi }^0\stackrel{~}{\chi }\stackrel{~}{\chi }^0)`$ $`={\displaystyle \frac{1}{8\pi ^2}}{\displaystyle \underset{ijk}{}}`$ $`\{[๐ช_{jk}^{0L}c_{Hij}^Lc_{Hik}^L+๐ช_{jk}^{0R}c_{Hij}^Rc_{Hik}^R]๐_3(\stackrel{~}{\chi }_j^0\stackrel{~}{\chi }_i\stackrel{~}{\chi }_k^0)`$ (108)
$`+M_{\stackrel{~}{\chi }_k^0}M_{\stackrel{~}{\chi }_i}\left[๐ช_{jk}^{0L}c_{Hij}^Lc_{Hik}^R+๐ช_{jk}^{0R}c_{Hij}^Rc_{Hik}^L\right]๐_3^{}(\stackrel{~}{\chi }_j^0\stackrel{~}{\chi }_i\stackrel{~}{\chi }_k^0)`$
$`+M_{\stackrel{~}{\chi }_i}M_{\stackrel{~}{\chi }_j^0}\left[๐ช_{jk}^{0L}c_{Hij}^Rc_{Hik}^L+๐ช_{jk}^{0R}c_{Hij}^Lc_{Hik}^R\right]๐_3^{}(\stackrel{~}{\chi }_j^0\stackrel{~}{\chi }_i\stackrel{~}{\chi }_k^0)`$
$`+M_{\stackrel{~}{\chi }_k^0}M_{\stackrel{~}{\chi }_j^0}[๐ช_{jk}^{0L}c_{Hij}^Rc_{Hik}^R+๐ช_{jk}^{0R}c_{Hij}^Lc_{Hik}^L]๐_3^{\prime \prime }(\stackrel{~}{\chi }_j^0\stackrel{~}{\chi }_i\stackrel{~}{\chi }_k^0)\}.`$
#### 5.2.4 Tri4-type triangles
The Tri4-type triangles contribute to $`\mathrm{\Gamma }^{fin,Z}(H^+H^{})`$ with the following term:
$`\mathrm{\Gamma }_{H^+H^{}}^Z(4)`$ $`=`$ $`{\displaystyle \frac{1}{16\pi ^2}}\left({\displaystyle \frac{e^2M_W(12s_W^2)}{2s_W^2c_W^3}}\right)\times `$ (109)
$`\{g_{H^0HH}\mathrm{cos}(\beta \alpha )[๐_4(H^0HZ)+๐_4(ZHH^0)]`$
$`+g_{h^0HH}\mathrm{sin}(\beta \alpha )[๐_4(h^0HZ)+๐_4(ZHh^0)]\}.`$
#### 5.2.5 Tri5-type triangles
There is no contribution from Tri5-type triangles in the production of $`H^+H^{}`$ pairs.
#### 5.2.6 Tri6-type triangles
The Tri6-type triangles contribute to both $`\mathrm{\Gamma }^{fin,\gamma }(H^+H^{})`$ and $`\mathrm{\Gamma }^{fin,Z}(H^+H^{})`$ with:
$`\mathrm{\Gamma }_{H^+H^{}}^V(\text{6ch})`$ $`={\displaystyle \frac{1}{8\pi ^2}}`$ $`\{g_{VGG}g_{A^0GH}^2๐_6(GA^0G)`$ (110)
$`+g_{VGG}g_{H^0GH}^2๐_6(GH^0G)+g_{VGG}g_{h^0GH}^2๐_6(Gh^0G)`$
$`+g_{VHH}g_{H^0HH}^2๐_6(HH^0H)+g_{VHH}g_{h^0HH}^2๐_6(Hh^0H)\}`$
and
$$\mathrm{\Gamma }_{H^+H^{}}^V(6\stackrel{~}{f})=\mathrm{\Gamma }_{H^+H^{}}^{V,heavy}(6\stackrel{~}{f})+\mathrm{\Gamma }_{H^+H^{}}^{V,light}(6\stackrel{~}{f}).$$
(111)
The third generation squark contribution, with sfermion mixing, is:
$`\mathrm{\Gamma }_{H^+H^{}}^{V,heavy}(6\stackrel{~}{f})`$ $`={\displaystyle \frac{3}{8\pi ^2}}{\displaystyle \underset{ijk=1,2}{}}`$ $`\{g_{V\stackrel{~}{b}_i\stackrel{~}{b}_k}g_{H\stackrel{~}{t}_j\stackrel{~}{b}_i}g_{H\stackrel{~}{t}_j\stackrel{~}{b}_k}๐_6(\stackrel{~}{b}_i\stackrel{~}{t}_j\stackrel{~}{b}_k)`$ (112)
$`g_{V\stackrel{~}{t}_i\stackrel{~}{t}_k}g_{H\stackrel{~}{t}_i\stackrel{~}{b}_j}g_{H\stackrel{~}{t}_k\stackrel{~}{b}_j}๐_6(\stackrel{~}{t}_i\stackrel{~}{b}_j\stackrel{~}{t}_k)\}.`$
The coupling of L-sfermions to the charged Higgs boson does not vanish like the fermion mass, so they also contribute to $`\mathrm{\Gamma }^V(6\stackrel{~}{f})`$. With $`(f,f^{})=(u,d),(c,s)\text{or}3\times (\nu _{\mathrm{}},\mathrm{})`$, one has:
$`\mathrm{\Gamma }_{H^+H^{}}^{V,light}(6\stackrel{~}{f})`$ $`=`$ $`{\displaystyle \frac{1}{8\pi ^2}}{\displaystyle \underset{(\stackrel{~}{f}\stackrel{~}{f}^{})}{}}N_c^fg_{H\stackrel{~}{f}_L\stackrel{~}{f}_L^{}}^2\left\{g_{V\stackrel{~}{f}_L^{}\stackrel{~}{f}_L^{}}^0๐_6(\stackrel{~}{f}_L^{}\stackrel{~}{f}_L\stackrel{~}{f}_L^{})g_{V\stackrel{~}{f}_L\stackrel{~}{f}_L}^0๐_6(\stackrel{~}{f}_L\stackrel{~}{f}_L^{}\stackrel{~}{f}_L)\right\}.`$ (113)
In addition, Tri6-type triangles contribute to $`\mathrm{\Gamma }^{fin,Z}(H^+H^{})`$ with the following term:
$`\mathrm{\Gamma }_{H^+H^{}}^Z(\text{6n})`$ $`=`$ $`{\displaystyle \frac{1}{8\pi ^2}}\left({\displaystyle \frac{e^2M_W}{4s_W^2c_W}}\right)\times `$ (114)
$`\{g_{H^0GH}\mathrm{sin}(\beta \alpha )[๐_6(H^0GA^0)+๐_6(A^0GH^0)]`$
$`g_{h^0GH}\mathrm{cos}(\beta \alpha )[๐_6(h^0GA^0)+๐_6(A^0Gh^0)]\}.`$
#### 5.2.7 4-leg diagrams
The 4-leg diagrams contribute to $`\mathrm{\Gamma }^{fin,\gamma }(H^+H^{})`$ with the following term:
$`\mathrm{\Gamma }_{H^+H^{}}^\gamma (\text{4-leg})`$ $`={\displaystyle \frac{1}{8\pi ^2}}`$ $`\{2e^3[B_0(H\gamma ,M_H^2)B_1(H\gamma ,M_H^2)]`$ (115)
$`+{\displaystyle \frac{e^3(12s_W^2)^2}{2s_W^2c_W^2}}\left[B_0(HZ,M_H^2)B_1(HZ,M_H^2)\right]`$
$`+{\displaystyle \frac{e^3\mathrm{sin}^2(\beta \alpha )}{4s_W^2}}\left[B_0(H^0W,M_H^2)B_1(H^0W,M_H^2)\right]`$
$`+{\displaystyle \frac{e^3\mathrm{cos}^2(\beta \alpha )}{4s_W^2}}\left[B_0(h^0W,M_H^2)B_1(h^0W,M_H^2)\right]`$
$`+{\displaystyle \frac{e^3}{4s_W^2}}[B_0(A^0W,M_H^2)B_1(A^0W,M_H^2)]\}.`$
The 4-leg diagrams contribute to $`\mathrm{\Gamma }^{fin,Z}(H^+H^{})`$ with the following term:
$`\mathrm{\Gamma }_{H^+H^{}}^Z(\text{4-leg})`$ $`={\displaystyle \frac{1}{8\pi ^2}}`$ $`\{{\displaystyle \frac{2e^3(12s_W^2)}{2s_Wc_W}}[B_0(H\gamma ,M_H^2)B_1(H\gamma ,M_H^2)]`$ (116)
$`+{\displaystyle \frac{e^3(12s_W^2)^3}{4s_W^3c_W^3}}\left[B_0(HZ,M_H^2)B_1(HZ,M_H^2)\right]`$
$`{\displaystyle \frac{e^3\mathrm{sin}^2(\beta \alpha )}{4s_Wc_W}}\left[B_0(H^0W,M_H^2)B_1(H^0W,M_H^2)\right]`$
$`{\displaystyle \frac{e^3\mathrm{cos}^2(\beta \alpha )}{4s_Wc_W}}\left[B_0(h^0W,M_H^2)B_1(h^0W,M_H^2)\right]`$
$`{\displaystyle \frac{e^3}{4s_Wc_W}}[B_0(A^0W,M_H^2)B_1(A^0W,M_H^2)]\}.`$
#### 5.2.8 Charged Higgs self-energies
The charged Higgs self-energies contribute to $`\mathrm{\Gamma }^{fin,\gamma }(H^+H^{})`$ and $`\mathrm{\Gamma }^{fin,Z}(H^+H^{})`$ with:
$$\mathrm{\Gamma }_{H^+H^{}}^\gamma (H.s.e)=2e\times \left(\frac{d\mathrm{\Sigma }_{H^+H^{}}}{dq^2}\right)_{q^2=M_H^2}$$
(117)
and
$$\mathrm{\Gamma }_{H^+H^{}}^Z(H.s.e)=\frac{2e(12s_W^2)}{2s_Wc_W}\times \left(\frac{d\mathrm{\Sigma }_{H^+H^{}}}{dq^2}\right)_{q^2=M_H^2}$$
(118)
where $`\mathrm{\Sigma }_{H^+H^{}}(q^2)`$ is the sum of various bubble terms.
These terms contain some combinations of Passarino-Veltman functions, such as:
$`SE_1^\pm (XY,q^2)`$ $`=`$ $`4B_{22}(XY,q^2)+q^2\left[B_0(XY,q^2)+B_{21}(XY,q^2)2B_1(XY,q^2)\right],`$ (119)
$`SE_2^\pm (XY,q^2)`$ $`=`$ $`4B_{22}(XY,q^2)+q^2\left[B_1(XY,q^2)+B_{21}(XY,q^2)\right].`$ (120)
Here, we consider only the contributions which depend on $`q^2`$, because $`\mathrm{\Gamma }^{fin,\gamma }(H^+H^{})`$ and $`\mathrm{\Gamma }^{fin,Z}(H^+H^{})`$ depend on the derivate of $`\mathrm{\Sigma }_{H^+H^{}}(q^2)`$. Four types of bubbles are taken into account when calculating this โreducedโ self-energy, which we refer to as $`\stackrel{~}{\mathrm{\Sigma }}_{H^+H^{}}(q^2)`$:
$$\stackrel{~}{\mathrm{\Sigma }}_{H^+H^{}}(q^2)=\stackrel{~}{\mathrm{\Sigma }}_{H^+H^{}}(VS,q^2)+\stackrel{~}{\mathrm{\Sigma }}_{H^+H^{}}(SS^{},q^2)+\stackrel{~}{\mathrm{\Sigma }}_{H^+H^{}}(ff^{},q^2)+\stackrel{~}{\mathrm{\Sigma }}_{H^+H^{}}(\stackrel{~}{\chi }\stackrel{~}{\chi }^0,q^2).$$
(121)
The $`VS`$ bubbles contribute to $`\stackrel{~}{\mathrm{\Sigma }}_{H^+H^{}}(q^2)`$ with:
$`\stackrel{~}{\mathrm{\Sigma }}_{H^+H^{}}(VS,q^2)`$ $`={\displaystyle \frac{1}{16\pi ^2}}`$ $`\{e^2SE_1^\pm (H\gamma ,q^2)+{\displaystyle \frac{e^2(12s_W^2)^2}{4s_W^2c_W^2}}SE_1^\pm (HZ,q^2)`$ (122)
$`+{\displaystyle \frac{e^2\mathrm{sin}^2(\beta \alpha )}{4s_W^2}}SE_1^\pm (H^0W,q^2)+{\displaystyle \frac{e^2\mathrm{cos}^2(\beta \alpha )}{4s_W^2}}SE_1^\pm (h^0W,q^2)`$
$`+{\displaystyle \frac{e^2}{4s_W^2}}SE_1^\pm (A^0W,q^2)\}.`$
The $`SS^{}`$ bubbles contribute to $`\stackrel{~}{\mathrm{\Sigma }}_{H^+H^{}}(q^2)`$ with:
$`\stackrel{~}{\mathrm{\Sigma }}_{H^+H^{}}(SS^{},q^2)`$ $`={\displaystyle \frac{1}{16\pi ^2}}`$ $`\{{\displaystyle \underset{light(\stackrel{~}{f}\stackrel{~}{f}^{})}{}}N_c^fg_{H\stackrel{~}{f}_L\stackrel{~}{f}_L^{}}^2B_0(\stackrel{~}{f}_L\stackrel{~}{f}_L^{},q^2)+{\displaystyle \underset{ij=1,2}{}}3g_{H\stackrel{~}{t}_i\stackrel{~}{b}_j}^2B_0(\stackrel{~}{t}_i\stackrel{~}{b}_j,q^2)`$ (123)
$`+g_{H^0HH}^2B_0(HH^0,q^2)+g_{h^0HH}^2B_0(Hh^0,q^2)`$
$`+g_{H^0GH}^2B_0(GH^0,q^2)+g_{h^0GH}^2B_0(Gh^0,q^2)`$
$`+g_{A^0GH}^2B_0(GA^0,q^2)\}.`$
The fermion and gaugino bubbles contribute to $`\stackrel{~}{\mathrm{\Sigma }}_{H^+H^{}}(q^2)`$ with respectively:
$`\stackrel{~}{\mathrm{\Sigma }}_{H^+H^{}}(ff^{},q^2)`$ $`={\displaystyle \frac{e^2}{16\pi ^2s_W^2M_W^2}}{\displaystyle \underset{(ff^{})}{}}N_c^f`$ $`\{(M_f^{}^2\mathrm{tan}^2\beta +M_f^2\text{cot}^2\beta )SE_2^\pm (ff^{},q^2)`$ (124)
$`+2M_f^2M_f^{}^2B_0(ff^{},q^2)\}`$
and
$`\stackrel{~}{\mathrm{\Sigma }}_{H^+H^{}}(\stackrel{~}{\chi }\stackrel{~}{\chi }^0,q^2)`$ $`={\displaystyle \frac{1}{8\pi ^2}}{\displaystyle \underset{ij}{}}`$ $`\{[c_{Hij}^Rc_{Hij}^R+c_{Hij}^Lc_{Hij}^L]SE_2^\pm (\stackrel{~}{\chi }_i\stackrel{~}{\chi }_j^0,q^2)`$ (125)
$`+M_{\stackrel{~}{\chi }_i}M_{\stackrel{~}{\chi }_j^0}[c_{Hij}^Rc_{Hij}^L+c_{Hij}^Lc_{Hij}^R]B_0(\stackrel{~}{\chi }_i\stackrel{~}{\chi }_j^0,q^2)\}.`$
After having computed the full expression for $`\stackrel{~}{\mathrm{\Sigma }}_{H^+H^{}}(q^2)`$, one simply needs to calculate its derivative at $`q^2=M_H^2`$ and then insert it into equations (117) and (118).
### 5.3 Neutral Higgs sector
The amplitudes $`a_{L,R}^{fin}`$ corresponding to final vertices with $`H^0A^0`$ or $`h^0A^0`$ are:
$$a_{L,R}^{fin}(H^0A^0/h^0A^0)=\frac{i\mathrm{\Gamma }^{fin,\gamma }(H^0A^0/h^0A^0)}{2e}\frac{\eta g_{L,R}}{2s_Wc_W}\times \frac{i\mathrm{\Gamma }^{fin,Z}(H^0A^0/h^0A^0)}{2e}.$$
(126)
$`\mathrm{\Gamma }^{fin,\gamma }(H^0A^0/h^0A^0)`$ and $`\mathrm{\Gamma }^{fin,Z}(H^0A^0/h^0A^0)`$ depend on the final state. They are obtained by summing the contributions of various triangles and of neutral Higgs self-energy terms.
For the photon, one has:
$`\mathrm{\Gamma }^{fin,\gamma }(H^0A^0/h^0A^0)`$ $`=`$ $`\mathrm{\Gamma }_{H^0A^0/h^0A^0}^\gamma (\text{1ch})+\mathrm{\Gamma }_{H^0A^0/h^0A^0}^\gamma (2)\mathrm{\Gamma }_{H^0A^0/h^0A^0}^\gamma (2,pinch)`$ (127)
$`+`$ $`\mathrm{\Gamma }_{H^0A^0/h^0A^0}^\gamma (3f)+\mathrm{\Gamma }_{H^0A^0/h^0A^0}^\gamma (\stackrel{~}{\chi }\stackrel{~}{\chi }\stackrel{~}{\chi })+\mathrm{\Gamma }_{H^0A^0/h^0A^0}^\gamma (\text{4ch})`$
$`+`$ $`\mathrm{\Gamma }_{H^0A^0/h^0A^0}^\gamma (6\stackrel{~}{f})+\mathrm{\Gamma }_{H^0A^0/h^0A^0}^\gamma (\text{6ch})+\mathrm{\Gamma }_{H^0A^0/h^0A^0}^\gamma (\text{4-leg}).`$
For the $`Z`$ boson, one has:
$`\mathrm{\Gamma }^{fin,Z}(H^0A^0/h^0A^0)`$ $`=`$ $`\mathrm{\Gamma }_{H^0A^0/h^0A^0}^Z(\text{1ch})+\mathrm{\Gamma }_{H^0A^0/h^0A^0}^Z(\text{1n})`$ (128)
$`+`$ $`\mathrm{\Gamma }_{H^0A^0/h^0A^0}^Z(2)\mathrm{\Gamma }_{H^0A^0/h^0A^0}^Z(2,pinch)+\mathrm{\Gamma }_{H^0A^0/h^0A^0}^Z(3f)+`$
$`+`$ $`\mathrm{\Gamma }_{H^0A^0/h^0A^0}^Z(\stackrel{~}{\chi }\stackrel{~}{\chi }\stackrel{~}{\chi })+\mathrm{\Gamma }_{H^0A^0/h^0A^0}^Z(\stackrel{~}{\chi }^0\stackrel{~}{\chi }^0\stackrel{~}{\chi }^0)+\mathrm{\Gamma }_{H^0A^0/h^0A^0}^Z(\text{4n})`$
$`+`$ $`\mathrm{\Gamma }_{H^0A^0/h^0A^0}^Z(\text{4ch})+\mathrm{\Gamma }_{H^0A^0/h^0A^0}^Z(5)+\mathrm{\Gamma }_{H^0A^0/h^0A^0}^Z(6\stackrel{~}{f})`$
$`+`$ $`\mathrm{\Gamma }_{H^0A^0/h^0A^0}^Z(\text{6ch})+\mathrm{\Gamma }_{H^0A^0/h^0A^0}^Z(\text{6n})+\mathrm{\Gamma }_{H^0A^0/h^0A^0}^Z(\text{4-leg})+`$
$`+`$ $`\mathrm{\Gamma }_{H^0A^0/h^0A^0}^Z(H.s.e)+\mathrm{\Gamma }_{H^0A^0/h^0A^0}^Z(H.c.t).`$
#### 5.3.1 Tri1-type triangles
The Tri1-type triangles contribute to $`\mathrm{\Gamma }^{fin,\gamma }(H^0A^0/h^0A^0)`$ with charged terms only:
$`\mathrm{\Gamma }_{H^0A^0/h^0A^0}^\gamma (\text{1ch})`$ $`=`$ $`{\displaystyle \frac{1}{8\pi ^2}}\times {\displaystyle \frac{e^3}{2s_W^2}}\left[Z_{ab}\right]\times ๐_1(HWH).`$ (129)
The Tri1-type triangles contribute to $`\mathrm{\Gamma }^{fin,Z}(H^0A^0/h^0A^0)`$ with the following charged terms:
$`\mathrm{\Gamma }_{H^0A^0/h^0A^0}^Z(\text{1ch})`$ $`=`$ $`{\displaystyle \frac{1}{8\pi ^2}}\times {\displaystyle \frac{e^3(12s_W^2)}{4s_W^3c_W}}\left[Z_{ab}\right]\times ๐_1(HWH)`$ (130)
and with the following neutral terms:
$`\mathrm{\Gamma }_{H^0A^0}^Z(\text{1n})`$ $`={\displaystyle \frac{e^3\mathrm{sin}(\beta \alpha )}{64\pi ^2s_W^3c_W^3}}`$ $`\{\mathrm{sin}^2(\beta \alpha )๐_1(A^0ZH^0)+\mathrm{cos}^2(\beta \alpha )๐_1(G^0ZH^0)`$ (131)
$`+\mathrm{cos}^2(\beta \alpha )๐_1(A^0Zh^0)\mathrm{cos}^2(\beta \alpha )๐_1(G^0Zh^0)\},`$
$`\mathrm{\Gamma }_{h^0A^0}^Z(\text{1n})`$ $`={\displaystyle \frac{e^3\mathrm{cos}(\beta \alpha )}{64\pi ^2s_W^3c_W^3}}`$ $`\{\mathrm{sin}^2(\beta \alpha )๐_1(A^0ZH^0)\mathrm{sin}^2(\beta \alpha )๐_1(G^0ZH^0)`$ (132)
$`+\mathrm{cos}^2(\beta \alpha )๐_1(A^0Zh^0)+\mathrm{sin}^2(\beta \alpha )๐_1(G^0Zh^0)\}.`$
#### 5.3.2 Tri2-type triangles
With $`f^V=\{\begin{array}{c}1\text{for}V=\gamma \hfill \\ c_W/s_W\text{for}V=Z\hfill \end{array}`$ , the contribution of the Tri2-type triangles is:
$`\mathrm{\Gamma }_{H^0A^0/h^0A^0}^V(2)`$ $`=`$ $`{\displaystyle \frac{e^3f^V}{32\pi ^2s_W^2}}\left[Z_{ab}\right]\times ๐_2(WHW).`$ (133)
However, one must also take into account the pinch term:
$`\mathrm{\Gamma }_{H^0A^0/h^0A^0}^V(2,pinch)`$ $`=`$ $`{\displaystyle \frac{e^3f^V}{8\pi ^2s_W^2}}\left[Z_{ab}\right]\times B_0(WW,q^2).`$ (134)
#### 5.3.3 Tri3-type triangles
Let $`V`$ be either the photon or the $`Z`$ boson. The Tri3-type triangles contribute to both $`\mathrm{\Gamma }^{fin,\gamma }(H^0A^0/h^0A^0)`$ and $`\mathrm{\Gamma }^{fin,Z}(H^0A^0/h^0A^0)`$ with two different terms.
The first term, with fermion triangles at the final vertex, is given by:
$$\mathrm{\Gamma }_{H^0A^0/h^0A^0}^V(3f)=\frac{e^3}{16\pi ^2s_W^2M_W^2}\underset{f}{}N_c^fM_f^2y_f(g_{VLf}g_{VRf})\left[๐_3(fff)M_f^2๐_3^{\prime \prime }(fff)\right].$$
(135)
In the previous equation, the term $`y_f`$ depends both on the fermion in the triangle and on the final state. More explicitely, for $`(q_u,q_d)`$ or $`(\nu _{\mathrm{}},\mathrm{})`$ doublets, one writes $`y_f`$ as follows:
$`y_f`$ $`=`$ $`({\displaystyle \frac{\mathrm{sin}\alpha \text{cot}\beta }{\mathrm{sin}\beta }},{\displaystyle \frac{\mathrm{cos}\alpha \mathrm{tan}\beta }{\mathrm{cos}\beta }})\text{for}H^0A^0,`$
$`y_f`$ $`=`$ $`({\displaystyle \frac{\mathrm{cos}\alpha \text{cot}\beta }{\mathrm{sin}\beta }},{\displaystyle \frac{\mathrm{sin}\alpha \mathrm{tan}\beta }{\mathrm{cos}\beta }})\text{for}h^0A^0.`$ (137)
The second term corresponds to chargino triangles at the final vertex.
For $`H^0A^0`$ final states:
$`\mathrm{\Gamma }_{H^0A^0}^V(\stackrel{~}{\chi }\stackrel{~}{\chi }\stackrel{~}{\chi })`$ $`={\displaystyle \frac{1}{8\pi ^2}}{\displaystyle \underset{ijk}{}}`$ $`\{[๐ช_{ik}^{VL}c_{A^0kj}^Rc_{H^0ji}^L+๐ช_{ik}^{VR}c_{A^0kj}^Lc_{H^0ji}^R]๐_3(\stackrel{~}{\chi }_i\stackrel{~}{\chi }_j\stackrel{~}{\chi }_k)`$ (138)
$`+M_{\stackrel{~}{\chi }_j}M_{\stackrel{~}{\chi }_k}\left[๐ช_{ik}^{VL}c_{A^0kj}^Lc_{H^0ji}^L+๐ช_{ik}^{VR}c_{A^0kj}^Rc_{H^0ji}^R\right]๐_3^{}(\stackrel{~}{\chi }_i\stackrel{~}{\chi }_j\stackrel{~}{\chi }_k)`$
$`+M_{\stackrel{~}{\chi }_i}M_{\stackrel{~}{\chi }_j}\left[๐ช_{ik}^{VL}c_{A^0kj}^Rc_{H^0ji}^R+๐ช_{ik}^{VR}c_{A^0kj}^Lc_{H^0ji}^L\right]๐_3^{}(\stackrel{~}{\chi }_i\stackrel{~}{\chi }_j\stackrel{~}{\chi }_k)`$
$`+M_{\stackrel{~}{\chi }_i}M_{\stackrel{~}{\chi }_k}[๐ช_{ik}^{VL}c_{A^0kj}^Lc_{H^0ji}^R+๐ช_{ik}^{VR}c_{A^0kj}^Rc_{H^0ji}^L]๐_3^{\prime \prime }(\stackrel{~}{\chi }_i\stackrel{~}{\chi }_j\stackrel{~}{\chi }_k)\}`$
$`{\displaystyle \frac{1}{8\pi ^2}}{\displaystyle \underset{ijk}{}}`$ $`\{[๐ช_{ki}^{VR}c_{A^0jk}^Rc_{H^0ij}^L+๐ช_{ki}^{VL}c_{A^0jk}^Lc_{H^0ij}^R]๐_3(\stackrel{~}{\chi }_i\stackrel{~}{\chi }_j\stackrel{~}{\chi }_k)`$
$`+M_{\stackrel{~}{\chi }_j}M_{\stackrel{~}{\chi }_k}\left[๐ช_{ki}^{VR}c_{A^0jk}^Lc_{H^0ij}^L+๐ช_{ki}^{VL}c_{A^0jk}^Rc_{H^0ij}^R\right]๐_3^{}(\stackrel{~}{\chi }_i\stackrel{~}{\chi }_j\stackrel{~}{\chi }_k)`$
$`+M_{\stackrel{~}{\chi }_i}M_{\stackrel{~}{\chi }_j}\left[๐ช_{ki}^{VR}c_{A^0jk}^Rc_{H^0ij}^R+๐ช_{ki}^{VL}c_{A^0jk}^Lc_{H^0ij}^L\right]๐_3^{}(\stackrel{~}{\chi }_i\stackrel{~}{\chi }_j\stackrel{~}{\chi }_k)`$
$`+M_{\stackrel{~}{\chi }_i}M_{\stackrel{~}{\chi }_k}[๐ช_{ki}^{VR}c_{A^0jk}^Lc_{H^0ij}^R+๐ช_{ki}^{VL}c_{A^0jk}^Rc_{H^0ij}^L]๐_3^{\prime \prime }(\stackrel{~}{\chi }_i\stackrel{~}{\chi }_j\stackrel{~}{\chi }_k)\},`$
For $`h^0A^0`$ final states:
$`\mathrm{\Gamma }_{h^0A^0}^V(\stackrel{~}{\chi }\stackrel{~}{\chi }\stackrel{~}{\chi })`$ $`={\displaystyle \frac{1}{8\pi ^2}}{\displaystyle \underset{ijk}{}}`$ $`\{[๐ช_{ik}^{VL}c_{A^0kj}^Rc_{h^0ji}^L+๐ช_{ik}^{VR}c_{A^0kj}^Lc_{h^0ji}^R]๐_3(\stackrel{~}{\chi }_i\stackrel{~}{\chi }_j\stackrel{~}{\chi }_k)`$ (139)
$`+M_{\stackrel{~}{\chi }_j}M_{\stackrel{~}{\chi }_k}\left[๐ช_{ik}^{VL}c_{A^0kj}^Lc_{h^0ji}^L+๐ช_{ik}^{VR}c_{A^0kj}^Rc_{h^0ji}^R\right]๐_3^{}(\stackrel{~}{\chi }_i\stackrel{~}{\chi }_j\stackrel{~}{\chi }_k)`$
$`+M_{\stackrel{~}{\chi }_i}M_{\stackrel{~}{\chi }_j}\left[๐ช_{ik}^{VL}c_{A^0kj}^Rc_{h^0ji}^R+๐ช_{ik}^{VR}c_{A^0kj}^Lc_{h^0ji}^L\right]๐_3^{}(\stackrel{~}{\chi }_i\stackrel{~}{\chi }_j\stackrel{~}{\chi }_k)`$
$`+M_{\stackrel{~}{\chi }_i}M_{\stackrel{~}{\chi }_k}[๐ช_{ik}^{VL}c_{A^0kj}^Lc_{h^0ji}^R+๐ช_{ik}^{VR}c_{A^0kj}^Rc_{h^0ji}^L]๐_3^{\prime \prime }(\stackrel{~}{\chi }_i\stackrel{~}{\chi }_j\stackrel{~}{\chi }_k)\}`$
$`{\displaystyle \frac{1}{8\pi ^2}}{\displaystyle \underset{ijk}{}}`$ $`\{[๐ช_{ki}^{VR}c_{A^0jk}^Rc_{h^0ij}^L+๐ช_{ki}^{VL}c_{A^0jk}^Lc_{h^0ij}^R]๐_3(\stackrel{~}{\chi }_i\stackrel{~}{\chi }_j\stackrel{~}{\chi }_k)`$
$`+M_{\stackrel{~}{\chi }_j}M_{\stackrel{~}{\chi }_k}\left[๐ช_{ki}^{VR}c_{A^0jk}^Lc_{h^0ij}^L+๐ช_{ki}^{VL}c_{A^0jk}^Rc_{h^0ij}^R\right]๐_3^{}(\stackrel{~}{\chi }_i\stackrel{~}{\chi }_j\stackrel{~}{\chi }_k)`$
$`+M_{\stackrel{~}{\chi }_i}M_{\stackrel{~}{\chi }_j}\left[๐ช_{ki}^{VR}c_{A^0jk}^Rc_{h^0ij}^R+๐ช_{ki}^{VL}c_{A^0jk}^Lc_{h^0ij}^L\right]๐_3^{}(\stackrel{~}{\chi }_i\stackrel{~}{\chi }_j\stackrel{~}{\chi }_k)`$
$`+M_{\stackrel{~}{\chi }_i}M_{\stackrel{~}{\chi }_k}[๐ช_{ki}^{VR}c_{A^0jk}^Lc_{h^0ij}^R+๐ช_{ki}^{VL}c_{A^0jk}^Rc_{h^0ij}^L]๐_3^{\prime \prime }(\stackrel{~}{\chi }_i\stackrel{~}{\chi }_j\stackrel{~}{\chi }_k)\}.`$
In addition, neutralino triangles at the final vertex give no contribution to $`\mathrm{\Gamma }^{fin,\gamma }(H^0A^0/h^0A^0)`$ but they enter into the expression of $`\mathrm{\Gamma }^{fin,Z}(H^0A^0/h^0A^0)`$.
For $`H^0A^0`$ final states:
$`\mathrm{\Gamma }_{H^0A^0}^V(\stackrel{~}{\chi }^0\stackrel{~}{\chi }^0\stackrel{~}{\chi }^0)`$ $`={\displaystyle \frac{1}{16\pi ^2}}{\displaystyle \underset{ijk}{}}`$ $`\{[๐ช_{ik}^{0L}n_{A^0kj}^Rn_{H^0ji}^L+๐ช_{ik}^{0R}n_{A^0kj}^Ln_{H^0ji}^R]๐_3(\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j^0\stackrel{~}{\chi }_k^0)`$ (140)
$`+M_{\stackrel{~}{\chi }_j^0}M_{\stackrel{~}{\chi }_k^0}\left[๐ช_{ik}^{0L}n_{A^0kj}^Ln_{H^0ji}^L+๐ช_{ik}^{0R}n_{A^0kj}^Rn_{H^0ji}^R\right]๐_3^{}(\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j^0\stackrel{~}{\chi }_k^0)`$
$`+M_{\stackrel{~}{\chi }_i^0}M_{\stackrel{~}{\chi }_j^0}\left[๐ช_{ik}^{0L}n_{A^0kj}^Rn_{H^0ji}^R+๐ช_{ik}^{0R}n_{A^0kj}^Ln_{H^0ji}^L\right]๐_3^{}(\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j^0\stackrel{~}{\chi }_k^0)`$
$`+M_{\stackrel{~}{\chi }_i^0}M_{\stackrel{~}{\chi }_k^0}[๐ช_{ik}^{0L}n_{A^0kj}^Ln_{H^0ji}^R+๐ช_{ik}^{0R}n_{A^0kj}^Rn_{H^0ji}^L]๐_3^{\prime \prime }(\chi _i^0\chi _j^0\chi _k^0)\}`$
$`{\displaystyle \frac{1}{16\pi ^2}}{\displaystyle \underset{ijk}{}}`$ $`\{[๐ช_{ki}^{0R}n_{A^0jk}^Rn_{H^0ij}^L+๐ช_{ki}^{0L}n_{A^0jk}^Ln_{H^0ij}^R]๐_3(\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j^0\stackrel{~}{\chi }_k^0)`$
$`+M_{\stackrel{~}{\chi }_j^0}M_{\stackrel{~}{\chi }_k^0}\left[๐ช_{ki}^{0R}n_{A^0jk}^Ln_{H^0ij}^L+๐ช_{ki}^{0L}n_{A^0jk}^Rn_{H^0ij}^R\right]๐_3^{}(\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j^0\stackrel{~}{\chi }_k^0)`$
$`+M_{\stackrel{~}{\chi }_i^0}M_{\stackrel{~}{\chi }_j^0}\left[๐ช_{ki}^{0R}n_{A^0jk}^Rn_{H^0ij}^R+๐ช_{ki}^{0L}n_{A^0jk}^Ln_{H^0ij}^L\right]๐_3^{}(\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j^0\stackrel{~}{\chi }_k^0)`$
$`+M_{\stackrel{~}{\chi }_i^0}M_{\stackrel{~}{\chi }_k^0}[๐ช_{ki}^{0R}n_{A^0jk}^Ln_{H^0ij}^R+๐ช_{ki}^{0L}n_{A^0jk}^Rn_{H^0ij}^L]๐_3^{\prime \prime }(\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j^0\stackrel{~}{\chi }_k^0)\}.`$
For the $`h^0A^0`$ final states:
$`\mathrm{\Gamma }_{h^0A^0}^V(\stackrel{~}{\chi }^0\stackrel{~}{\chi }^0\stackrel{~}{\chi }^0)`$ $`={\displaystyle \frac{1}{16\pi ^2}}{\displaystyle \underset{ijk}{}}`$ $`\{[๐ช_{ik}^{0L}n_{A^0kj}^Rn_{h^0ji}^L+๐ช_{ik}^{0R}n_{A^0kj}^Ln_{h^0ji}^R]๐_3(\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j^0\stackrel{~}{\chi }_k^0)`$ (141)
$`+M_{\stackrel{~}{\chi }_j^0}M_{\stackrel{~}{\chi }_k^0}\left[๐ช_{ik}^{0L}n_{A^0kj}^Ln_{h^0ji}^L+๐ช_{ik}^{0R}n_{A^0kj}^Rn_{h^0ji}^R\right]๐_3^{}(\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j^0\stackrel{~}{\chi }_k^0)`$
$`+M_{\stackrel{~}{\chi }_i^0}M_{\stackrel{~}{\chi }_j^0}\left[๐ช_{ik}^{0L}n_{A^0kj}^Rn_{h^0ji}^R+๐ช_{ik}^{0R}n_{A^0kj}^Ln_{h^0ji}^L\right]๐_3^{}(\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j^0\stackrel{~}{\chi }_k^0)`$
$`+M_{\stackrel{~}{\chi }_i^0}M_{\stackrel{~}{\chi }_k^0}[๐ช_{ik}^{0L}n_{A^0kj}^Ln_{h^0ji}^R+๐ช_{ik}^{0R}n_{A^0kj}^Rn_{h^0ji}^L]๐_3^{\prime \prime }(\chi _i^0\chi _j^0\chi _k^0)\}`$
$`{\displaystyle \frac{1}{16\pi ^2}}{\displaystyle \underset{ijk}{}}`$ $`\{[๐ช_{ki}^{0R}n_{A^0jk}^Rn_{h^0ij}^L+๐ช_{ki}^{0L}n_{A^0jk}^Ln_{h^0ij}^R]๐_3(\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j^0\stackrel{~}{\chi }_k^0)`$
$`+M_{\stackrel{~}{\chi }_j^0}M_{\stackrel{~}{\chi }_k^0}\left[๐ช_{ki}^{0R}n_{A^0jk}^Ln_{h^0ij}^L+๐ช_{ki}^{0L}n_{A^0jk}^Rn_{h^0ij}^R\right]๐_3^{}(\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j^0\stackrel{~}{\chi }_k^0)`$
$`+M_{\stackrel{~}{\chi }_i^0}M_{\stackrel{~}{\chi }_j^0}\left[๐ช_{ki}^{0R}n_{A^0jk}^Rn_{h^0ij}^R+๐ช_{ki}^{0L}n_{A^0jk}^Ln_{h^0ij}^L\right]๐_3^{}(\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j^0\stackrel{~}{\chi }_k^0)`$
$`+M_{\stackrel{~}{\chi }_i^0}M_{\stackrel{~}{\chi }_k^0}[๐ช_{ki}^{0R}n_{A^0jk}^Ln_{h^0ij}^R+๐ช_{ki}^{0L}n_{A^0jk}^Rn_{h^0ij}^L]๐_3^{\prime \prime }(\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j^0\stackrel{~}{\chi }_k^0)\}.`$
#### 5.3.4 Tri4-type triangles
Let $`V`$ denote either the photon or the $`Z`$ boson. The Tri4-type triangles contribute to both $`\mathrm{\Gamma }^{fin,\gamma }(H^0A^0/h^0A^0)`$ and $`\mathrm{\Gamma }^{fin,Z}(H^0A^0/h^0A^0)`$ with charged terms, which are given by:
$$\mathrm{\Gamma }_{H^0A^0}^V(\text{4ch})=\frac{1}{8\pi ^2}g_{VWG}\times \left\{g_{WHA^0}g_{H^0GH}๐_4(GHW)+g_{A^0GH}g_{WHH^0}๐_4(WHG)\right\}$$
(142)
and
$$\mathrm{\Gamma }_{h^0A^0}^V(\text{4ch})=\frac{1}{8\pi ^2}g_{VWG}\times \left\{g_{WHA^0}g_{h^0GH}๐_4(GHW)+g_{A^0GH}g_{WHh^0}๐_4(WHG)\right\}.$$
(143)
The coupling constants $`g_{WHH^0}`$, $`g_{WHh^0}`$ and $`g_{A^0GH}`$ depend on the charge of the particles at the vertex. Note that, in the triangles considered here, $`W`$ and $`G`$ carry the same charge.
Tri4-type triangles also contribute to $`\mathrm{\Gamma }^{fin,Z}(H^0A^0/h^0A^0)`$ with neutral terms:
$`\mathrm{\Gamma }_{H^0A^0}^Z(\text{4n})`$ $`={\displaystyle \frac{1}{16\pi ^2}}`$ $`\{g_{ZZH^0}[g_{H^0H^0H^0}g_{ZH^0A^0}๐_4(H^0H^0Z)+g_{h^0H^0H^0}g_{Zh^0A^0}๐_4(H^0h^0Z)]`$ (144)
$`+g_{ZZh^0}\left[g_{h^0H^0H^0}g_{ZH^0A^0}๐_4(h^0H^0Z)+g_{H^0h^0h^0}g_{Zh^0A^0}๐_4(h^0h^0Z)\right]`$
$`+g_{ZZH^0}\left[g_{H^0A^0A^0}g_{ZH^0A^0}๐_4(H^0A^0Z)+g_{H^0A^0G^0}g_{ZH^0G^0}๐_4(H^0G^0Z)\right]`$
$`+g_{ZZh^0}[g_{h^0A^0A^0}g_{ZH^0A^0}๐_4(h^0A^0Z)+g_{h^0A^0G^0}g_{ZH^0G^0}๐_4(h^0G^0Z)]\},`$
$`\mathrm{\Gamma }_{h^0A^0}^Z(\text{4n})`$ $`={\displaystyle \frac{1}{16\pi ^2}}`$ $`\{g_{ZZH^0}[g_{h^0H^0H^0}g_{ZH^0A^0}๐_4(H^0H^0Z)+g_{H^0h^0h^0}g_{Zh^0A^0}๐_4(H^0h^0Z)]`$ (145)
$`+g_{ZZh^0}\left[g_{H^0h^0h^0}g_{ZH^0A^0}๐_4(h^0H^0Z)+g_{h^0h^0h^0}g_{Zh^0A^0}๐_4(h^0h^0Z)\right]`$
$`+g_{ZZH^0}\left[g_{H^0A^0A^0}g_{Zh^0A^0}๐_4(H^0A^0Z)+g_{H^0A^0G^0}g_{Zh^0G^0}๐_4(H^0G^0Z)\right]`$
$`+g_{ZZh^0}[g_{h^0A^0A^0}g_{Zh^0A^0}๐_4(h^0A^0Z)+g_{h^0A^0G^0}g_{Zh^0G^0}๐_4(h^0G^0Z)]\}.`$
#### 5.3.5 Tri5-type triangles
The Tri5-type triangles contribute only to $`\mathrm{\Gamma }^{fin,Z}(H^0A^0/h^0A^0)`$.
$$\mathrm{\Gamma }_{H^0A^0}^Z(5)=\frac{1}{16\pi ^2}\left[g_{ZZH^0}^2g_{ZH^0A^0}๐_5(ZZH^0)+g_{ZZh^0}g_{ZZH^0}g_{Zh^0A^0}๐_5(ZZh^0)\right],$$
(146)
$$\mathrm{\Gamma }_{h^0A^0}^Z(5)=\frac{1}{16\pi ^2}\left[g_{ZZH^0}g_{ZZh^0}g_{ZH^0A^0}๐_5(ZZH^0)+g_{ZZh^0}^2g_{Zh^0A^0}๐_5(ZZh^0)\right].$$
(147)
#### 5.3.6 Tri6-type triangles
Two Tri6-type triangles contribute to both $`\mathrm{\Gamma }^{fin,\gamma }(H^0A^0/h^0A^0)`$ and $`\mathrm{\Gamma }^{fin,Z}(H^0A^0/h^0A^0)`$.
Let $`V`$ denote either the photon or the $`Z`$ boson, we focus on $`\mathrm{\Gamma }_{H^0A^0/h^0A^0}^V(6\stackrel{~}{f})`$ first. The coupling of $`A^0`$ to sfermions is proportional to the corresponding fermion mass and it is thus negligible, except in the case of third generation squarks, with sfermion mixing.
For $`H^0A^0`$ final states, one has:
$`\mathrm{\Gamma }_{H^0A^0}^V(6\stackrel{~}{f})`$ $`={\displaystyle \frac{3}{8\pi ^2}}{\displaystyle \underset{ijk=1,2}{}}`$ $`\left\{g_{V\stackrel{~}{b}_i\stackrel{~}{b}_k}g_{H^0\stackrel{~}{b}_i\stackrel{~}{b}_j}g_{A^0\stackrel{~}{b}_j\stackrel{~}{b}_k}๐_6(\stackrel{~}{b}_i\stackrel{~}{b}_j\stackrel{~}{b}_k)+g_{V\stackrel{~}{b}_i\stackrel{~}{b}_k}g_{A^0\stackrel{~}{b}_i\stackrel{~}{b}_j}g_{H^0\stackrel{~}{b}_j\stackrel{~}{b}_k}๐_6(\stackrel{~}{b}_i\stackrel{~}{b}_j\stackrel{~}{b}_k)\right\}`$ (148)
$`+{\displaystyle \frac{3}{8\pi ^2}}{\displaystyle \underset{ijk=1,2}{}}`$ $`\left\{g_{V\stackrel{~}{t}_i\stackrel{~}{t}_k}g_{H^0\stackrel{~}{t}_i\stackrel{~}{t}_j}g_{A^0\stackrel{~}{t}_j\stackrel{~}{t}_k}๐_6(\stackrel{~}{t}_i\stackrel{~}{t}_j\stackrel{~}{t}_k)+g_{V\stackrel{~}{t}_i\stackrel{~}{t}_k}g_{A^0\stackrel{~}{t}_i\stackrel{~}{t}_j}g_{H^0\stackrel{~}{t}_j\stackrel{~}{t}_k}๐_6(\stackrel{~}{t}_i\stackrel{~}{t}_j\stackrel{~}{t}_k)\right\}.`$
For $`h^0A^0`$ final states, one has:
$`\mathrm{\Gamma }_{h^0A^0}^V(6\stackrel{~}{f})`$ $`={\displaystyle \frac{3}{8\pi ^2}}{\displaystyle \underset{ijk=1,2}{}}`$ $`\left\{g_{V\stackrel{~}{b}_i\stackrel{~}{b}_k}g_{h^0\stackrel{~}{b}_i\stackrel{~}{b}_j}g_{A^0\stackrel{~}{b}_j\stackrel{~}{b}_k}๐_6(\stackrel{~}{b}_i\stackrel{~}{b}_j\stackrel{~}{b}_k)+g_{V\stackrel{~}{b}_i\stackrel{~}{b}_k}g_{A^0\stackrel{~}{b}_i\stackrel{~}{b}_j}g_{h^0\stackrel{~}{b}_j\stackrel{~}{b}_k}๐_6(\stackrel{~}{b}_i\stackrel{~}{b}_j\stackrel{~}{b}_k)\right\}`$ (149)
$`+{\displaystyle \frac{3}{8\pi ^2}}{\displaystyle \underset{ijk=1,2}{}}`$ $`\left\{g_{V\stackrel{~}{t}_i\stackrel{~}{t}_k}g_{h^0\stackrel{~}{t}_i\stackrel{~}{t}_j}g_{A^0\stackrel{~}{t}_j\stackrel{~}{t}_k}๐_6(\stackrel{~}{t}_i\stackrel{~}{t}_j\stackrel{~}{t}_k)+g_{V\stackrel{~}{t}_i\stackrel{~}{t}_k}g_{A^0\stackrel{~}{t}_i\stackrel{~}{t}_j}g_{h^0\stackrel{~}{t}_j\stackrel{~}{t}_k}๐_6(\stackrel{~}{t}_i\stackrel{~}{t}_j\stackrel{~}{t}_k)\right\}.`$
The 6ch triangles also contribute to both $`\mathrm{\Gamma }^{fin,\gamma }(H^0A^0/h^0A^0)`$ and $`\mathrm{\Gamma }^{fin,Z}(H^0A^0/h^0A^0)`$:
$`\mathrm{\Gamma }_{H^0A^0}^V(\text{6ch})`$ $`=`$ $`{\displaystyle \frac{eM_W}{8\pi ^2s_W}}g_{H^0GH}\left[g_{VHH}๐_6(HGH)g_{VGG}๐_6(GHG)\right],`$ (150)
$`\mathrm{\Gamma }_{h^0A^0}^V(\text{6ch})`$ $`=`$ $`{\displaystyle \frac{eM_W}{8\pi ^2s_W}}g_{h^0GH}\left[g_{VHH}๐_6(HGH)g_{VGG}๐_6(GHG)\right].`$ (151)
Since neutral Higgs or Goldstone bosons do not couple to a photon, the 6n triangles contribute to $`\mathrm{\Gamma }^{fin,Z}(H^0A^0/h^0A^0)`$ only.
For $`H^0A^0`$ final states, one has:
$`\mathrm{\Gamma }_{H^0A^0}^Z(\text{6n})`$ $`={\displaystyle \frac{1}{8\pi ^2}}`$ $`\{g_{Zh^0G^0}[g_{h^0H^0H^0}g_{H^0A^0G^0}๐_6(h^0H^0G^0)+g_{H^0h^0h^0}g_{h^0A^0G^0}๐_6(h^0h^0G^0)]`$ (152)
$`+g_{ZH^0G^0}\left[g_{H^0H^0H^0}g_{H^0A^0G^0}๐_6(H^0H^0G^0)+g_{h^0H^0H^0}g_{h^0A^0G^0}๐_6(H^0h^0G^0)\right]`$
$`+g_{Zh^0A^0}\left[g_{h^0H^0H^0}g_{H^0A^0A^0}๐_6(h^0H^0A^0)+g_{H^0h^0h^0}g_{h^0A^0A^0}๐_6(h^0h^0A^0)\right]`$
$`+g_{ZH^0A^0}[g_{H^0H^0H^0}g_{H^0A^0A^0}๐_6(H^0H^0A^0)+g_{h^0H^0H^0}g_{h^0A^0A^0}๐_6(H^0h^0A^0)]\}`$
$`{\displaystyle \frac{1}{8\pi ^2}}`$ $`\{g_{Zh^0G^0}[g_{H^0G^0G^0}g_{h^0A^0G^0}๐_6(G^0G^0h^0)+g_{H^0A^0G^0}g_{h^0A^0A^0}๐_6(G^0A^0h^0)]`$
$`+g_{ZH^0G^0}\left[g_{H^0G^0G^0}g_{H^0A^0G^0}๐_6(G^0G^0H^0)+g_{H^0A^0G^0}g_{H^0A^0A^0}๐_6(G^0A^0H^0)\right]`$
$`+g_{Zh^0A^0}\left[g_{H^0A^0A^0}g_{h^0A^0A^0}๐_6(A^0A^0h^0)+g_{H^0A^0G^0}g_{h^0A^0G^0}๐_6(A^0G^0h^0)\right]`$
$`+g_{ZH^0A^0}[g_{H^0A^0A^0}^2๐_6(A^0A^0H^0)+g_{H^0A^0G^0}^2๐_6(A^0G^0H^0)]\}.`$
For $`h^0A^0`$ final states, one has:
$`\mathrm{\Gamma }_{h^0A^0}^Z(\text{6n})`$ $`={\displaystyle \frac{1}{8\pi ^2}}`$ $`\{g_{ZH^0G^0}[g_{h^0H^0H^0}g_{H^0A^0G^0}๐_6(H^0H^0G^0)+g_{H^0h^0h^0}g_{h^0A^0G^0}๐_6(H^0h^0G^0)]`$ (153)
$`+g_{Zh^0G^0}\left[g_{H^0h^0h^0}g_{H^0A^0G^0}๐_6(h^0H^0G^0)+g_{h^0h^0h^0}g_{h^0A^0G^0}๐_6(h^0h^0G^0)\right]`$
$`+g_{ZH^0A^0}\left[g_{h^0H^0H^0}g_{H^0A^0A^0}๐_6(H^0H^0A^0)+g_{H^0h^0h^0}g_{h^0A^0A^0}๐_6(H^0h^0A^0)\right]`$
$`+g_{Zh^0A^0}[g_{H^0h^0h^0}g_{H^0A^0A^0}๐_6(h^0H^0A^0)+g_{h^0h^0h^0}g_{h^0A^0A^0}๐_6(h^0h^0A^0)]\}`$
$`{\displaystyle \frac{1}{8\pi ^2}}`$ $`\{g_{ZH^0G^0}[g_{h^0G^0G^0}g_{H^0A^0G^0}๐_6(G^0G^0H^0)+g_{h^0A^0G^0}g_{H^0A^0A^0}๐_6(G^0A^0H^0)]`$
$`+g_{Zh^0G^0}\left[g_{h^0G^0G^0}g_{h^0A^0G^0}๐_6(G^0G^0h^0)+g_{h^0A^0G^0}g_{h^0A^0A^0}๐_6(G^0A^0h^0)\right]`$
$`+g_{ZH^0A^0}\left[g_{h^0A^0A^0}g_{H^0A^0A^0}๐_6(A^0A^0H^0)+g_{H^0A^0G^0}g_{h^0A^0G^0}๐_6(A^0G^0H^0)\right]`$
$`+g_{Zh^0A^0}[g_{h^0A^0A^0}^2๐_6(A^0A^0h^0)+g_{h^0A^0G^0}^2๐_6(A^0G^0h^0)]\}.`$
#### 5.3.7 4-leg diagrams
For $`H^0A^0`$ final states, the 4-leg diagrams give the following contributions:
$`\mathrm{\Gamma }_{H^0A^0}^\gamma (\text{4-leg})`$ $`={\displaystyle \frac{\mathrm{sin}(\beta \alpha )}{16\pi ^2}}\left({\displaystyle \frac{e^3}{2s_W^2}}\right)\times `$ $`\{[B_0(HW,M_{H^0}^2)B_1(HW,M_{H^0}^2)]`$ (154)
$`+[B_0(HW,M_A^2)B_1(HW,M_A^2)]\}`$
$`\mathrm{\Gamma }_{H^0A^0}^Z(\text{4-leg})`$ $`={\displaystyle \frac{s_W}{c_W}}\mathrm{\Gamma }^\gamma (\text{4-leg})`$ $`+{\displaystyle \frac{\mathrm{sin}(\beta \alpha )}{16\pi ^2}}\left({\displaystyle \frac{e^3}{4s_W^3c_W^3}}\right)\times `$ (155)
$`\{[B_0(A^0Z,M_{H^0}^2)B_1(A^0Z,M_{H^0}^2)]`$
$`+[B_0(H^0Z,M_A^2)B_1(H^0Z,M_A^2)]\}.`$
For $`h^0A^0`$ final states, the 4-leg diagrams give the following contributions:
$`\mathrm{\Gamma }_{h^0A^0}^\gamma (\text{4-leg})`$ $`={\displaystyle \frac{\mathrm{cos}(\beta \alpha )}{16\pi ^2}}\left({\displaystyle \frac{e^3}{2s_W^2}}\right)\times `$ $`\{[B_0(HW,M_{h^0}^2)B_1(HW,M_{h^0}^2)]`$ (156)
$`+[B_0(HW,M_A^2)B_1(HW,M_A^2)]\}`$
$`\mathrm{\Gamma }_{h^0A^0}^Z(\text{4-leg})`$ $`={\displaystyle \frac{s_W}{c_W}}\mathrm{\Gamma }^\gamma (\text{4-leg})`$ $`{\displaystyle \frac{\mathrm{cos}(\beta \alpha )}{16\pi ^2}}\left({\displaystyle \frac{e^3}{4s_W^3c_W^3}}\right)\times `$ (157)
$`\{[B_0(A^0Z,M_{h^0}^2)B_1(A^0Z,M_{h^0}^2)]`$
$`+[B_0(H^0Z,M_A^2)B_1(H^0Z,M_A^2)]\}.`$
#### 5.3.8 Neutral Higgs self-energies
Before estimating the self-energy terms, let us first define several useful expressions:
$`v_1=\mathrm{sin}2\alpha \mathrm{sin}2\beta {\displaystyle \frac{s_W^2}{c_W^2}}\mathrm{cos}2\alpha \mathrm{cos}2\beta .`$ (158)
$`v_2=\mathrm{cos}2\alpha \mathrm{sin}2\beta +{\displaystyle \frac{s_W^2}{c_W^2}}\mathrm{sin}2\alpha \mathrm{cos}2\beta .`$ (159)
$`SE_1^0(XY,q^2)`$ $`=`$ $`2q^2B_1(XY,q^2)A(M_Y^2)(q^2+M_X^2)B_0(XY,q^2),`$ (160)
$`SE_2^0(ff,q^2)`$ $`=`$ $`2M_f^2B_0(ff,q^2)+A(M_f^2)+q^2B_1(ff,q^2),`$ (161)
$`SE_3^0(XY,q^2,a,b,c,d)`$ $`=`$ $`8\{(ad+bc)[q^2B_1(XY,q^2)+A(M_X^2)+M_Y^2B_0(XY,q^2)]`$ (162)
$`+(ac+bd)M_XM_YB_0(XY,q^2)\},`$
$`SE_4^0(XY,q^2)`$ $`=`$ $`B_1(XY,q^2)B_0(XY,q^2),`$ (163)
$`SE_5^0(XY,q^2)`$ $`=`$ $`2B_1(XY,q^2)+B_0(XY,q^2),`$ (164)
$`SE_6^0(XY,q^2,a,b,c,d)`$ $`=`$ $`8\{(ad+bc)M_XB_1(XY,q^2)`$ (165)
$`+(ac+bd)M_Y[B_0(XY,q^2)+B_1(XY,q^2)]\},`$
In the following, all neutral Higgs self-energies $`\mathrm{\Sigma }(q^2)`$ and the Higgs tadpoles $`T_{H^0/h^0}`$ are written as the sum of various contributions coming from the gauge and Higgs sectors, fermion pairs, gaugino pairs and sfermion pairs (where we consider separately the unmixed case and the third generation squarks with mixing):
$`\mathrm{\Sigma }(q^2)`$ $`=`$ $`\mathrm{\Sigma }(\text{g+H})+\mathrm{\Sigma }(ff)+\mathrm{\Sigma }(\stackrel{~}{\chi }\stackrel{~}{\chi })+\mathrm{\Sigma }(\stackrel{~}{\chi }^0\stackrel{~}{\chi }^0)+\mathrm{\Sigma }(\stackrel{~}{f}\stackrel{~}{f}),`$ (166)
$`T_{H^0/h^0}`$ $`=`$ $`T_{H^0/h^0}(\text{g+H})+T_{H^0/h^0}(ff)+T_{H^0/h^0}(\stackrel{~}{\chi }\stackrel{~}{\chi })+T_{H^0/h^0}(\stackrel{~}{\chi }^0\stackrel{~}{\chi }^0)+T_{H^0/h^0}(\stackrel{~}{f}\stackrel{~}{f}).`$ (167)
a) $`H^0`$ self-energies:
The contribution of the gauge and Higgs sectors is:
$`\mathrm{\Sigma }_{H^0H^0}(\text{g+H})`$ $`={\displaystyle \frac{1}{16\pi ^2}}`$ $`\{g_{ZH^0A^0}^2SE_1^0(A^0Z,q^2)+g_{ZH^0G^0}^2SE_1^0(ZZ,q^2)`$ (168)
$`+2g_{WHH^0}^2SE_1^0(HW,q^2)+2g_{WGH^0}^2SE_1^0(WW,q^2)`$
$`+2g_{WWH^0}^2\left[2B_0(WW,q^2)1\right]+g_{ZZH^0}^2\left[2B_0(ZZ,q^2)1\right]`$
$`+g_{H^0HH}^2B_0(HH,q^2)+g_{H^0GG}^2B_0(WW,q^2)+2g_{H^0GH}^2B_0(WH,q^2)`$
$`+{\displaystyle \frac{1}{2}}\left[g_{H^0h^0h^0}^2B_0(h^0h^0,q^2)+g_{H^0H^0H^0}^2B_0(H^0H^0,q^2)\right]`$
$`+{\displaystyle \frac{1}{2}}\left[g_{H^0A^0A^0}^2B_0(A^0A^0,q^2)+g_{H^0G^0G^0}^2B_0(ZZ,q^2)\right]`$
$`+g_{h^0H^0H^0}^2B_0(H^0h^0,q^2)+g_{H^0A^0G^0}^2B_0(A^0Z,q^2)`$
$`{\displaystyle \frac{e^2M_W^2\mathrm{cos}^2(\beta \alpha )}{2s_W^2}}\left[B_0(WW,q^2)+{\displaystyle \frac{1}{2c_W^4}}B_0(ZZ,q^2)\right]`$
$`+{\displaystyle \frac{e^2}{s_W^2}}\left(\left[2A(M_W^2)M_W^2\right]+{\displaystyle \frac{1}{2c_W^2}}\left[2A(M_Z^2)M_Z^2\right]\right)`$
$`+{\displaystyle \frac{e^2}{8s_W^2c_W^2}}\left[(3\mathrm{sin}^22\alpha 1)A(M_{h^0}^2)+3\mathrm{cos}^22\alpha A(M_{H^0}^2)\right]`$
$`{\displaystyle \frac{e^2}{8s_W^2c_W^2}}\mathrm{cos}2\beta \mathrm{cos}2\alpha \left[A(M_A^2)A(M_Z^2)\right]`$
$`+{\displaystyle \frac{e^2}{4s_W^2}}[(1v_1)A(M_W^2)+(1+v_1)A(M_H^2)]\}.`$
The contribution of the fermion pairs is:
$`\mathrm{\Sigma }_{H^0H^0}(ff)`$ $`=`$ $`{\displaystyle \frac{1}{4\pi ^2}}{\displaystyle \underset{f}{}}N_c^f\times (c_{H^0f}^L)^2\times SE_2^0(ff,q^2).`$ (169)
The contributions of the gaugino pairs are:
$`\mathrm{\Sigma }_{H^0H^0}(\stackrel{~}{\chi }\stackrel{~}{\chi })`$ $`=`$ $`{\displaystyle \frac{1}{64\pi ^2}}{\displaystyle \underset{ij}{}}SE_3^0(\stackrel{~}{\chi }_i\stackrel{~}{\chi }_j,q^2,c_{H^0ji}^L,c_{H^0ij}^L,c_{H^0ij}^L,c_{H^0ji}^L),`$ (170)
$`\mathrm{\Sigma }_{H^0H^0}(\stackrel{~}{\chi }^0\stackrel{~}{\chi }^0)`$ $`=`$ $`{\displaystyle \frac{1}{128\pi ^2}}{\displaystyle \underset{ij}{}}SE_3^0(\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j^0,q^2,n_{H^0ji}^L,n_{H^0ij}^L,n_{H^0ij}^L,n_{H^0ji}^L).`$ (171)
The contribution of sfermion pairs consists of two terms:
$`\mathrm{\Sigma }_{H^0H^0}^{light}(\stackrel{~}{f}\stackrel{~}{f})`$ $`={\displaystyle \frac{1}{16\pi ^2}}{\displaystyle \underset{\stackrel{~}{f}}{}}N_c^f`$ $`\{[g_{H^0\stackrel{~}{f}_L\stackrel{~}{f}_L}^2+g_{H^0\stackrel{~}{f}_R\stackrel{~}{f}_R}^2]B_0(\stackrel{~}{f}\stackrel{~}{f},q^2)`$ (172)
$`[g_{H^0H^0\stackrel{~}{f}_L\stackrel{~}{f}_L}+g_{H^0H^0\stackrel{~}{f}_R\stackrel{~}{f}_R}]A(M_{\stackrel{~}{f}}^2)\},`$
$`\mathrm{\Sigma }_{H^0H^0}^{heavy}(\stackrel{~}{f}\stackrel{~}{f})`$ $`={\displaystyle \frac{3}{16\pi ^2}}`$ $`{\displaystyle \underset{ij=1,2}{}}\left\{g_{H^0\stackrel{~}{t}_i\stackrel{~}{t}_j}^2B_0(\stackrel{~}{t}_i\stackrel{~}{t}_j,q^2)+g_{H^0\stackrel{~}{b}_i\stackrel{~}{b}_j}^2B_0(\stackrel{~}{b}_i\stackrel{~}{b}_j,q^2)\right\}`$ (173)
$`{\displaystyle \frac{3}{16\pi ^2}}`$ $`{\displaystyle \underset{\stackrel{~}{f}=\stackrel{~}{t},\stackrel{~}{b}}{}}\left\{c_{\stackrel{~}{f}}^2\left[g_{H^0H^0\stackrel{~}{f}_L\stackrel{~}{f}_L}A(M_{\stackrel{~}{f}_1}^2)+g_{H^0H^0\stackrel{~}{f}_R\stackrel{~}{f}_R}A(M_{\stackrel{~}{f}_2}^2)\right]\right\}`$
$`{\displaystyle \frac{3}{16\pi ^2}}`$ $`{\displaystyle \underset{\stackrel{~}{f}=\stackrel{~}{t},\stackrel{~}{b}}{}}\left\{s_{\stackrel{~}{f}}^2\left[g_{H^0H^0\stackrel{~}{f}_R\stackrel{~}{f}_R}A(M_{\stackrel{~}{f}_1}^2)+g_{H^0H^0\stackrel{~}{f}_L\stackrel{~}{f}_L}A(M_{\stackrel{~}{f}_2}^2)\right]\right\}.`$
b) $`h^0`$ self-energies:
The contribution of the gauge and Higgs sectors is:
$`\mathrm{\Sigma }_{h^0h^0}(\text{g+H})`$ $`={\displaystyle \frac{1}{16\pi ^2}}`$ $`\{g_{Zh^0A^0}^2SE_1^0(A^0Z,q^2)+g_{Zh^0G^0}^2SE_1^0(ZZ,q^2)`$ (174)
$`+2g_{WHh^0}^2SE_1^0(HW,q^2)+2g_{WGh^0}^2SE_1^0(WW,q^2)`$
$`+2g_{WWh^0}^2\left[2B_0(WW,q^2)1\right]+g_{ZZh^0}^2\left[2B_0(ZZ,q^2)1\right]`$
$`+g_{h^0HH}^2B_0(HH,q^2)+g_{h^0GG}^2B_0(WW,q^2)+2g_{h^0GH}^2B_0(WH,q^2)`$
$`+{\displaystyle \frac{1}{2}}\left[g_{h^0H^0H^0}^2B_0(H^0H^0,q^2)+g_{h^0h^0h^0}^2B_0(h^0h^0,q^2)\right]`$
$`+{\displaystyle \frac{1}{2}}\left[g_{h^0A^0A^0}^2B_0(A^0A^0,q^2)+g_{h^0G^0G^0}^2B_0(ZZ,q^2)\right]`$
$`+g_{H^0h^0h^0}^2B_0(H^0h^0,q^2)+g_{h^0A^0G^0}^2B_0(A^0Z,q^2)`$
$`{\displaystyle \frac{e^2M_W^2\mathrm{sin}^2(\beta \alpha )}{2s_W^2}}\left[B_0(WW,q^2)+{\displaystyle \frac{1}{2c_W^4}}B_0(ZZ,q^2)\right]`$
$`+{\displaystyle \frac{e^2}{s_W^2}}\left(\left[2A(M_W^2)M_W^2\right]+{\displaystyle \frac{1}{2c_W^2}}\left[2A(M_Z^2)M_Z^2\right]\right)`$
$`+{\displaystyle \frac{e^2}{8s_W^2c_W^2}}\left[(3\mathrm{sin}^22\alpha 1)A(M_{H^0}^2)+3\mathrm{cos}^22\alpha A(M_{h^0}^2)\right]`$
$`+{\displaystyle \frac{e^2}{8s_W^2c_W^2}}\mathrm{cos}2\beta \mathrm{cos}2\alpha \left[A(M_A^2)A(M_Z^2)\right]`$
$`+{\displaystyle \frac{e^2}{4s_W^2}}[(1+v_1)A(M_W^2)+(1v_1)A(M_H^2)]\}.`$
The contribution of the fermion pairs is:
$`\mathrm{\Sigma }_{h^0h^0}(ff)`$ $`=`$ $`{\displaystyle \frac{1}{4\pi ^2}}{\displaystyle \underset{f}{}}N_c^f\times (c_{h^0f}^L)^2\times SE_2^0(ff,q^2).`$ (175)
The contributions of the gaugino pairs are:
$`\mathrm{\Sigma }_{h^0h^0}(\stackrel{~}{\chi }\stackrel{~}{\chi })`$ $`=`$ $`{\displaystyle \frac{1}{64\pi ^2}}{\displaystyle \underset{ij}{}}SE_3^0(\stackrel{~}{\chi }_i\stackrel{~}{\chi }_j,q^2,c_{h^0ji}^L,c_{h^0ij}^L,c_{h^0ij}^L,c_{h^0ji}^L),`$ (176)
$`\mathrm{\Sigma }_{h^0h^0}(\stackrel{~}{\chi }^0\stackrel{~}{\chi }^0)`$ $`=`$ $`{\displaystyle \frac{1}{128\pi ^2}}{\displaystyle \underset{ij}{}}SE_3^0(\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j^0,q^2,n_{h^0ji}^L,n_{h^0ij}^L,n_{h^0ij}^L,n_{h^0ji}^L).`$ (177)
The contribution of sfermion pairs consists of two terms:
$`\mathrm{\Sigma }_{h^0h^0}^{light}(\stackrel{~}{f}\stackrel{~}{f})_{\text{light}}`$ $`={\displaystyle \frac{1}{16\pi ^2}}{\displaystyle \underset{\stackrel{~}{f}}{}}N_c^f`$ $`\{[g_{h^0\stackrel{~}{f}_L\stackrel{~}{f}_L}^2+g_{h^0\stackrel{~}{f}_R\stackrel{~}{f}_R}^2]B_0(\stackrel{~}{f}\stackrel{~}{f},q^2)`$ (178)
$`[g_{h^0h^0\stackrel{~}{f}_L\stackrel{~}{f}_L}+g_{h^0h^0\stackrel{~}{f}_R\stackrel{~}{f}_R}]A(M_{\stackrel{~}{f}}^2)\},`$
$`\mathrm{\Sigma }_{h^0h^0}^{heavy}(\stackrel{~}{f}\stackrel{~}{f})`$ $`={\displaystyle \frac{3}{16\pi ^2}}`$ $`{\displaystyle \underset{ij=1,2}{}}\left\{g_{h^0\stackrel{~}{t}_i\stackrel{~}{t}_j}^2B_0(\stackrel{~}{t}_i\stackrel{~}{t}_j,q^2)+g_{h^0\stackrel{~}{b}_i\stackrel{~}{b}_j}^2B_0(\stackrel{~}{b}_i\stackrel{~}{b}_j,q^2)\right\}`$ (179)
$`{\displaystyle \frac{3}{16\pi ^2}}`$ $`{\displaystyle \underset{\stackrel{~}{f}=\stackrel{~}{t},\stackrel{~}{b}}{}}\left\{c_{\stackrel{~}{f}}^2\left[g_{h^0h^0\stackrel{~}{f}_L\stackrel{~}{f}_L}A(M_{\stackrel{~}{f}_1}^2)+g_{h^0h^0\stackrel{~}{f}_R\stackrel{~}{f}_R}A(M_{\stackrel{~}{f}_2}^2)\right]\right\}`$
$`{\displaystyle \frac{3}{16\pi ^2}}`$ $`{\displaystyle \underset{\stackrel{~}{f}=\stackrel{~}{t},\stackrel{~}{b}}{}}\left\{s_{\stackrel{~}{f}}^2\left[g_{h^0h^0\stackrel{~}{f}_R\stackrel{~}{f}_R}A(M_{\stackrel{~}{f}_1}^2)+g_{h^0h^0\stackrel{~}{f}_L\stackrel{~}{f}_L}A(M_{\stackrel{~}{f}_2}^2)\right]\right\}.`$
c) Mixed $`H^0h^0`$ self-energies:
The contribution of the gauge and Higgs sectors is:
$`\mathrm{\Sigma }_{H^0h^0}(\text{g+H})`$ $`={\displaystyle \frac{1}{16\pi ^2}}`$ $`\{g_{ZH^0A^0}g_{Zh^0A^0}SE_1^0(A^0Z,q^2)+g_{ZH^0G^0}g_{Zh^0G^0}SE_1^0(ZZ,q^2)`$ (180)
$`+2g_{WHH^0}g_{WHh^0}SE_1^0(HW,q^2)+2g_{WGH^0}g_{WGh^0}SE_1^0(WW,q^2)`$
$`+2g_{WWH^0}g_{WWh^0}\left[2B_0(WW,q^2)1\right]`$
$`+g_{ZZH^0}g_{ZZh^0}\left[2B_0(ZZ,q^2)1\right]`$
$`+g_{H^0HH}g_{h^0HH}B_0(HH,q^2)+g_{H^0GG}g_{h^0GG}B_0(WW,q^2)`$
$`+2g_{H^0GH}g_{h^0GH}B_0(WH,q^2)`$
$`+{\displaystyle \frac{1}{2}}\left[g_{H^0H^0H^0}g_{h^0H^0H^0}B_0(H^0H^0,q^2)+g_{H^0h^0h^0}g_{h^0h^0h^0}B_0(h^0h^0,q^2)\right]`$
$`+{\displaystyle \frac{1}{2}}\left[g_{H^0A^0A^0}g_{h^0A^0A^0}B_0(A^0A^0,q^2)+g_{H^0G^0G^0}g_{h^0G^0G^0}B_0(ZZ,q^2)\right]`$
$`+g_{h^0H^0H^0}g_{H^0h^0h^0}B_0(H^0h^0,q^2)+g_{H^0A^0G^0}g_{h^0A^0G^0}B_0(A^0Z,q^2)`$
$`{\displaystyle \frac{e^2M_W^2\mathrm{sin}(\beta \alpha )\mathrm{cos}(\beta \alpha )}{2s_W^2}}\left[B_0(WW,q^2)+{\displaystyle \frac{1}{2c_W^4}}B_0(ZZ,q^2)\right]`$
$`+{\displaystyle \frac{3e^2\mathrm{sin}2\alpha \mathrm{cos}2\alpha }{8s_W^2c_W^2}}\left[A(M_{h^0}^2)A(M_{H^0}^2)\right]`$
$`+{\displaystyle \frac{e^2}{8s_W^2c_W^2}}\mathrm{cos}2\beta \mathrm{sin}2\alpha \left[A(M_A^2)A(M_Z^2)\right]`$
$`+{\displaystyle \frac{e^2v_2}{4s_W^2}}[A(M_H^2)A(M_W^2)]\}.`$
The contribution of the fermion pairs is:
$`\mathrm{\Sigma }_{H^0h^0}(ff)`$ $`=`$ $`{\displaystyle \frac{1}{4\pi ^2}}{\displaystyle \underset{f}{}}N_c^f\times c_{H^0f}^Lc_{h^0f}^L\times SE_2^0(ff,q^2).`$ (181)
The contributions of the gaugino pairs are:
$`\mathrm{\Sigma }_{H^0h^0}(\stackrel{~}{\chi }\stackrel{~}{\chi })`$ $`=`$ $`{\displaystyle \frac{1}{64\pi ^2}}{\displaystyle \underset{ij}{}}SE_3^0(\stackrel{~}{\chi }_i\stackrel{~}{\chi }_j,q^2,c_{h^0ji}^L,c_{h^0ij}^L,c_{H^0ij}^L,c_{H^0ji}^L),`$ (182)
$`\mathrm{\Sigma }_{H^0h^0}(\stackrel{~}{\chi }^0\stackrel{~}{\chi }^0)`$ $`=`$ $`{\displaystyle \frac{1}{128\pi ^2}}{\displaystyle \underset{ij}{}}SE_3^0(\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j^0,q^2,n_{h^0ji}^L,n_{h^0ij}^L,n_{H^0ij}^L,n_{H^0ji}^L).`$ (183)
The contribution of the sfermion pairs consists of two terms:
$`\mathrm{\Sigma }_{H^0h^0}^{light}(\stackrel{~}{f}\stackrel{~}{f})`$ $`={\displaystyle \frac{1}{16\pi ^2}}{\displaystyle \underset{\stackrel{~}{f}}{}}N_c^f`$ $`\{[g_{H^0\stackrel{~}{f}_L\stackrel{~}{f}_L}g_{h^0\stackrel{~}{f}_L\stackrel{~}{f}_L}+g_{H^0\stackrel{~}{f}_R\stackrel{~}{f}_R}g_{h^0\stackrel{~}{f}_R\stackrel{~}{f}_R}]B_0(\stackrel{~}{f}\stackrel{~}{f},q^2)`$ (184)
$`[g_{H^0h^0\stackrel{~}{f}_L\stackrel{~}{f}_L}+g_{H^0h^0\stackrel{~}{f}_R\stackrel{~}{f}_R}]A(M_{\stackrel{~}{f}}^2)\}.`$
$`\mathrm{\Sigma }_{H^0h^0}^{heavy}(\stackrel{~}{f}\stackrel{~}{f})`$ $`={\displaystyle \frac{3}{16\pi ^2}}`$ $`{\displaystyle \underset{ij=1,2}{}}\left\{g_{H^0\stackrel{~}{t}_i\stackrel{~}{t}_j}g_{h^0\stackrel{~}{t}_i\stackrel{~}{t}_j}B_0(\stackrel{~}{t}_i\stackrel{~}{t}_j,q^2)+g_{H^0\stackrel{~}{b}_i\stackrel{~}{b}_j}g_{h^0\stackrel{~}{b}_i\stackrel{~}{b}_j}B_0(\stackrel{~}{b}_i\stackrel{~}{b}_j,q^2)\right\}`$ (185)
$`{\displaystyle \frac{3}{16\pi ^2}}`$ $`{\displaystyle \underset{\stackrel{~}{f}=\stackrel{~}{t},\stackrel{~}{b}}{}}\left\{c_{\stackrel{~}{f}}^2\left[g_{H^0h^0\stackrel{~}{f}_L\stackrel{~}{f}_L}A(M_{\stackrel{~}{f}_1}^2)+g_{H^0h^0\stackrel{~}{f}_R\stackrel{~}{f}_R}A(M_{\stackrel{~}{f}_2}^2)\right]\right\}`$
$`{\displaystyle \frac{3}{16\pi ^2}}`$ $`{\displaystyle \underset{\stackrel{~}{f}=\stackrel{~}{t},\stackrel{~}{b}}{}}\left\{s_{\stackrel{~}{f}}^2\left[g_{H^0h^0\stackrel{~}{f}_R\stackrel{~}{f}_R}A(M_{\stackrel{~}{f}_1}^2)+g_{H^0h^0\stackrel{~}{f}_L\stackrel{~}{f}_L}A(M_{\stackrel{~}{f}_2}^2)\right]\right\}.`$
d) $`A^0`$ self-energies:
The contribution of the gauge and Higgs sectors is:
$`\mathrm{\Sigma }_{A^0A^0}(\text{g+H})`$ $`={\displaystyle \frac{1}{16\pi ^2}}`$ $`\{g_{Zh^0A^0}^2SE_1^0(h^0Z,q^2)+g_{ZH^0A^0}^2SE_1^0(H^0Z,q^2)`$ (186)
$`+2g_{WHA^0}^2SE_1^0(HW,q^2)+2g_{A^0GH}^2B_0(WH,q^2)`$
$`+g_{H^0A^0A^0}^2B_0(A^0H^0,q^2)+g_{h^0A^0A^0}^2B_0(A^0h^0,q^2)`$
$`+g_{H^0A^0G^0}^2B_0(ZH^0,q^2)+g_{h^0A^0G^0}^2B_0(Zh^0,q^2)`$
$`+{\displaystyle \frac{e^2}{s_W^2}}\left(\left[2A(M_W^2)M_W^2\right]+{\displaystyle \frac{1}{2c_W^2}}\left[2A(M_Z^2)M_Z^2\right]\right)`$
$`+{\displaystyle \frac{e^2\mathrm{cos}^22\beta }{4s_W^2c_W^2}}A(M_H^2)+{\displaystyle \frac{e^2}{4s_W^2}}\left[1+\mathrm{sin}^22\beta {\displaystyle \frac{s_W^2}{c_W^2}}\mathrm{cos}^22\beta \right]A(M_W^2)`$
$`+{\displaystyle \frac{e^2}{8s_W^2c_W^2}}\left[(3\mathrm{sin}^22\beta 1)A(M_Z^2)+3\mathrm{cos}^22\beta A(M_A^2)\right]`$
$`+{\displaystyle \frac{e^2}{8s_W^2c_W^2}}\mathrm{cos}2\beta \mathrm{cos}2\alpha [A(M_{h^0}^2)A(M_{H^0}^2)]\}.`$
The contribution of the fermion pairs is:
$`\mathrm{\Sigma }_{A^0A^0}(ff)`$ $`={\displaystyle \frac{1}{64\pi ^2}}`$ $`{\displaystyle \underset{f}{}}N_c^fSE_3^0(ff,q^2,c_{A^0f}^L,c_{A^0f}^L,c_{A^0f}^L,c_{A^0f}^L).`$ (187)
The contributions of the gaugino pairs are:
$`\mathrm{\Sigma }_{A^0A^0}(\stackrel{~}{\chi }\stackrel{~}{\chi })`$ $`=`$ $`{\displaystyle \frac{1}{64\pi ^2}}{\displaystyle \underset{ij}{}}SE_3^0(\stackrel{~}{\chi }_i\stackrel{~}{\chi }_j,q^2,c_{A^0ji}^L,c_{A^0ij}^L,c_{A^0ij}^L,c_{A^0ji}^L),`$ (188)
$`\mathrm{\Sigma }_{A^0A^0}(\stackrel{~}{\chi }^0\stackrel{~}{\chi }^0)`$ $`=`$ $`{\displaystyle \frac{1}{128\pi ^2}}{\displaystyle \underset{ij}{}}SE_3^0(\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j^0,q^2,n_{A^0ji}^L,n_{A^0ij}^L,n_{A^0ij}^L,n_{A^0ji}^L).`$ (189)
The contribution of the sfermion pairs consists of two terms:
$`\mathrm{\Sigma }_{A^0A^0}^{light}(\stackrel{~}{f}\stackrel{~}{f})`$ $`=`$ $`{\displaystyle \frac{1}{16\pi ^2}}{\displaystyle \underset{\stackrel{~}{f}}{}}N_c^f\left\{\left[g_{A^0A^0\stackrel{~}{f}_L\stackrel{~}{f}_L}+g_{A^0A^0\stackrel{~}{f}_R\stackrel{~}{f}_R}\right]A(M_{\stackrel{~}{f}}^2)\right\},`$ (190)
$`\mathrm{\Sigma }_{A^0A^0}^{heavy}(\stackrel{~}{f}\stackrel{~}{f})`$ $`={\displaystyle \frac{3}{16\pi ^2}}`$ $`{\displaystyle \underset{ij=1,2}{}}\left\{g_{A^0\stackrel{~}{t}_i\stackrel{~}{t}_j}^2B_0(\stackrel{~}{t}_i\stackrel{~}{t}_j,q^2)+g_{A^0\stackrel{~}{b}_i\stackrel{~}{b}_j}^2B_0(\stackrel{~}{b}_i\stackrel{~}{b}_j,q^2)\right\}`$ (191)
$`{\displaystyle \frac{3}{16\pi ^2}}`$ $`{\displaystyle \underset{\stackrel{~}{f}=\stackrel{~}{t},\stackrel{~}{b}}{}}\left\{c_{\stackrel{~}{f}}^2\left[g_{A^0A^0\stackrel{~}{f}_L\stackrel{~}{f}_L}A(M_{\stackrel{~}{f}_1}^2)+g_{A^0A^0\stackrel{~}{f}_R\stackrel{~}{f}_R}A(M_{\stackrel{~}{f}_2}^2)\right]\right\}`$
$`{\displaystyle \frac{3}{16\pi ^2}}`$ $`{\displaystyle \underset{\stackrel{~}{f}=\stackrel{~}{t},\stackrel{~}{b}}{}}\left\{s_{\stackrel{~}{f}}^2\left[g_{A^0A^0\stackrel{~}{f}_R\stackrel{~}{f}_R}A(M_{\stackrel{~}{f}_1}^2)+g_{A^0A^0\stackrel{~}{f}_L\stackrel{~}{f}_L}A(M_{\stackrel{~}{f}_2}^2)\right]\right\}.`$
e) Mixed $`A^0Z`$ self-energies:
The contribution of the gauge and Higgs sectors is:
$`\mathrm{\Sigma }_{A^0Z}(\text{g+H})`$ $`={\displaystyle \frac{e^2}{32\pi ^2s_W^2c_W^2}}`$ $`\{M_Z\mathrm{cos}(\beta \alpha )\mathrm{sin}(\beta \alpha )SE_4^0(H^0Z,q^2)`$ (192)
$`M_Z\mathrm{cos}(\beta \alpha )\mathrm{sin}(\beta \alpha )SE_4^0(h^0Z,q^2)`$
$`+{\displaystyle \frac{1}{2}}M_Z\mathrm{cos}2\beta \mathrm{cos}(\beta +\alpha )\mathrm{sin}(\beta \alpha )SE_5^0(A^0H^0,q^2)`$
$`{\displaystyle \frac{1}{2}}M_Z\mathrm{cos}2\beta \mathrm{sin}(\beta +\alpha )\mathrm{cos}(\beta \alpha )SE_5^0(A^0h^0,q^2)`$
$`{\displaystyle \frac{1}{2}}M_Z\mathrm{sin}2\beta \mathrm{cos}(\beta +\alpha )\mathrm{cos}(\beta \alpha )SE_5^0(ZH^0,q^2)`$
$`+{\displaystyle \frac{1}{2}}M_Z\mathrm{sin}2\beta \mathrm{sin}(\beta +\alpha )\mathrm{sin}(\beta \alpha )SE_5^0(Zh^0,q^2)\}.`$
The contribution of the fermion pairs is:
$`\mathrm{\Sigma }_{A^0Z}(ff)`$ $`={\displaystyle \frac{1}{64\pi ^2}}`$ $`{\displaystyle \underset{f}{}}N_c^fSE_6^0(ff,q^2,v_f+a_f,v_fa_f,c_{A^0f}^L,c_{A^0f}^L).`$ (193)
Here, $`v_f`$ and $`a_f`$ are defined as follows:
$`v_f={\displaystyle \frac{1}{2s_Wc_W}}\left[+{\displaystyle \frac{1}{2}}{\displaystyle \frac{4}{3}}s_W^2\right],`$ $`a_f=+{\displaystyle \frac{1}{4s_Wc_W}}`$ $`\text{if}f=q_u\text{or}\nu _{\mathrm{}},`$ (194)
$`v_f={\displaystyle \frac{1}{2s_Wc_W}}\left[{\displaystyle \frac{1}{2}}+{\displaystyle \frac{2}{3}}s_W^2\right],`$ $`a_f={\displaystyle \frac{1}{4s_Wc_W}}`$ $`\text{if}f=q_d\text{or}\mathrm{}.`$ (195)
The contributions of the gaugino pairs are:
$`\mathrm{\Sigma }_{A^0Z}(\stackrel{~}{\chi }\stackrel{~}{\chi })`$ $`={\displaystyle \frac{1}{8\pi ^2}}{\displaystyle \underset{ij}{}}`$ $`\{M_{\stackrel{~}{\chi }_i}(c_{A^0ji}^L๐ช_{ij}^{ZR}+c_{A^0ji}^R๐ช_{ij}^{ZL})[B_0(\stackrel{~}{\chi }_i\stackrel{~}{\chi }_j,q^2)+B_1(\stackrel{~}{\chi }_i\stackrel{~}{\chi }_j,q^2)]`$ (196)
$`+M_{\stackrel{~}{\chi }_j}[c_{A^0ji}^L๐ช_{ij}^{ZL}+c_{A^0ji}^R๐ช_{ij}^{ZR}]B_1(\stackrel{~}{\chi }_i\stackrel{~}{\chi }_j,q^2)\},`$
$`\mathrm{\Sigma }_{A^0Z}(\stackrel{~}{\chi }^0\stackrel{~}{\chi }^0)`$ $`={\displaystyle \frac{1}{16\pi ^2}}{\displaystyle \underset{ij}{}}`$ $`\{M_{\stackrel{~}{\chi }_i^0}(n_{A^0ji}^L๐ช_{ij}^{ZR}+n_{A^0ji}^R๐ช_{ij}^{ZL})[B_0(\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j^0,q^2)+B_1(\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j^0,q^2)]`$ (197)
$`+M_{\stackrel{~}{\chi }_j^0}[n_{A^0ji}^L๐ช_{ij}^{ZL}+n_{A^0ji}^R๐ช_{ij}^{ZR}]B_1(\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j^0,q^2)\}.`$
Note that there is no contribution from sfermion pairs to $`\mathrm{\Sigma }_{A^0Z}(q^2)`$.
f) $`H^0`$ tadpole:
The contribution of the gauge and Higgs sectors is:
$`T_{H^0}(\text{g+H})`$ $`={\displaystyle \frac{1}{16\pi ^2}}`$ $`\{g_{H^0HH}A(M_H^2)+g_{H^0GG}A(M_W^2)`$ (198)
$`+{\displaystyle \frac{1}{2}}g_{H^0H^0H^0}A(M_{H^0}^2)+{\displaystyle \frac{1}{2}}g_{H^0h^0h^0}A(M_{h^0}^2)`$
$`+{\displaystyle \frac{1}{2}}g_{H^0A^0A^0}A(M_{A^0}^2)+{\displaystyle \frac{1}{2}}g_{H^0G^0G^0}A(M_Z^2)`$
$`g_{WWH^0}\left[4A(M_W^2)2M_W^2\right]{\displaystyle \frac{1}{2}}g_{ZZH^0}\left[4A(M_Z^2)2M_Z^2\right]`$
$`+{\displaystyle \frac{eM_W\mathrm{cos}(\beta \alpha )}{s_W}}[A(M_W^2)+{\displaystyle \frac{1}{2c_W^2}}A(M_Z^2)]\}.`$
The contribution of the fermion pairs is:
$`T_{H^0}(ff)`$ $`=`$ $`{\displaystyle \frac{1}{4\pi ^2}}{\displaystyle \underset{f}{}}N_c^fc_{H^0f}^LM_fA(M_f^2).`$ (199)
The contributions of the gaugino pairs are:
$`T_{H^0}(\stackrel{~}{\chi }\stackrel{~}{\chi })`$ $`=`$ $`{\displaystyle \frac{1}{4\pi ^2}}{\displaystyle \underset{i}{}}c_{H^0ii}^LM_{\stackrel{~}{\chi }_i}A(M_{\stackrel{~}{\chi }_i}^2),`$ (200)
$`T_{H^0}(\stackrel{~}{\chi }^0\stackrel{~}{\chi }^0)`$ $`=`$ $`{\displaystyle \frac{1}{8\pi ^2}}{\displaystyle \underset{i}{}}n_{H^0ii}^LM_{\stackrel{~}{\chi }_i^0}A(M_{\stackrel{~}{\chi }_i^0}^2).`$ (201)
The contribution of the sfermion pairs is the sum of two terms:
$`T_{H^0}^{light}(\stackrel{~}{f}\stackrel{~}{f})`$ $`=`$ $`{\displaystyle \frac{1}{16\pi ^2}}{\displaystyle \underset{f}{}}N_c^f\left\{\left[g_{H^0\stackrel{~}{f}_L\stackrel{~}{f}_L}+g_{H^0\stackrel{~}{f}_R\stackrel{~}{f}_R}\right]A(M_{\stackrel{~}{f}}^2)\right\},`$ (202)
$`T_{H^0}^{heavy}(\stackrel{~}{f}\stackrel{~}{f})`$ $`=`$ $`{\displaystyle \frac{3}{16\pi ^2}}{\displaystyle \underset{i=1,2}{}}\left\{g_{H^0\stackrel{~}{t}_i\stackrel{~}{t}_i}A(M_{\stackrel{~}{t}_i}^2)+g_{H^0\stackrel{~}{b}_i\stackrel{~}{b}_i}A(M_{\stackrel{~}{b}_i}^2)\right\}.`$ (203)
g) $`h^0`$ tadpole:
The contribution of the gauge and Higgs sectors is:
$`T_{h^0}(\text{g+H})`$ $`={\displaystyle \frac{1}{16\pi ^2}}`$ $`\{g_{h^0HH}A(M_H^2)+g_{h^0GG}A(M_W^2)`$ (204)
$`+{\displaystyle \frac{1}{2}}g_{h^0H^0H^0}A(M_{H^0}^2)+{\displaystyle \frac{1}{2}}g_{h^0h^0h^0}A(M_{h^0}^2)`$
$`+{\displaystyle \frac{1}{2}}g_{h^0A^0A^0}A(M_{A^0}^2)+{\displaystyle \frac{1}{2}}g_{h^0G^0G^0}A(M_Z^2)`$
$`g_{WWh^0}\left[4A(M_W^2)2M_W^2\right]{\displaystyle \frac{1}{2}}g_{ZZh^0}\left[4A(M_Z^2)2M_Z^2\right]`$
$`+{\displaystyle \frac{eM_W\mathrm{sin}(\beta \alpha )}{s_W}}[A(M_W^2)+{\displaystyle \frac{1}{2c_W^2}}A(M_Z^2)]\}.`$
The contribution of the fermion pairs is:
$`T_{h^0}(ff)`$ $`=`$ $`{\displaystyle \frac{1}{4\pi ^2}}{\displaystyle \underset{f}{}}N_c^fc_{h^0f}^LM_fA(M_f^2).`$ (205)
The contributions of the gaugino pairs are:
$`T_{h^0}(\stackrel{~}{\chi }\stackrel{~}{\chi })`$ $`=`$ $`{\displaystyle \frac{1}{4\pi ^2}}{\displaystyle \underset{i}{}}c_{h^0ii}^LM_{\stackrel{~}{\chi }_i}A(M_{\stackrel{~}{\chi }_i}^2),`$ (206)
$`T_{h^0}(\stackrel{~}{\chi }^0\stackrel{~}{\chi }^0)`$ $`=`$ $`{\displaystyle \frac{1}{8\pi ^2}}{\displaystyle \underset{i}{}}n_{h^0ii}^LM_{\stackrel{~}{\chi }_i^0}A(M_{\stackrel{~}{\chi }_i^0}^2).`$ (207)
The contribution of the sfermion pairs is the sum of two terms:
$`T_{h^0}^{light}(\stackrel{~}{f}\stackrel{~}{f})`$ $`=`$ $`{\displaystyle \frac{1}{16\pi ^2}}{\displaystyle \underset{f}{}}N_c^f\left\{\left[g_{h^0\stackrel{~}{f}_L\stackrel{~}{f}_L}+g_{h^0\stackrel{~}{f}_R\stackrel{~}{f}_R}\right]A(M_{\stackrel{~}{f}}^2)\right\},`$ (208)
$`T_{h^0}^{heavy}(\stackrel{~}{f}\stackrel{~}{f})`$ $`=`$ $`{\displaystyle \frac{3}{16\pi ^2}}{\displaystyle \underset{i=1,2}{}}\left\{g_{h^0\stackrel{~}{t}_i\stackrel{~}{t}_i}A(M_{\stackrel{~}{t}_i}^2)+g_{h^0\stackrel{~}{b}_i\stackrel{~}{b}_i}A(M_{\stackrel{~}{b}_i}^2)\right\}.`$ (209)
h) Expressions of the renormalized self-energies:
When calculating the effective contribution of the neutral Higgs self-energies to the pair production cross section at the one loop level, we must consider the renormalized self-energy terms $`\widehat{\mathrm{\Sigma }}`$, obtained by adding various counter terms to the unnormalized self-energies $`\mathrm{\Sigma }`$.
Let us first define the Higgs field renormalization constants:
$`\delta Z_{H_1}`$ $`=`$ $`\left({\displaystyle \frac{d\mathrm{\Sigma }_{A^0A^0}}{dq^2}}\right)_{q^2=M_A^2}{\displaystyle \frac{\text{cot}\beta }{M_Z}}\mathrm{\Sigma }_{A^0Z}(M_A^2),`$ (210)
$`\delta Z_{H_2}`$ $`=`$ $`\left({\displaystyle \frac{d\mathrm{\Sigma }_{A^0A^0}}{dq^2}}\right)_{q^2=M_A^2}+{\displaystyle \frac{\mathrm{tan}\beta }{M_Z}}\mathrm{\Sigma }_{A^0Z}(M_A^2).`$ (211)
They are used in the calculation of the various mass counter terms, together with the Higgs tadpoles and the parameter $`\delta M_{A^0A^0}^2`$ defined as:
$$\delta M_{A^0A^0}^2=\mathrm{\Sigma }_{A^0A^0}(M_A^2)M_A^2\left(\frac{d\mathrm{\Sigma }_{A^0A^0}}{dq^2}\right)_{q^2=M_A^2}.$$
(212)
Indeed, one has the following expressions for the mass counter terms:
$`\delta M_{H^0H^0}^2`$ $`=`$ $`\mathrm{sin}^2(\beta \alpha )\delta M_{A^0A^0}^2+\mathrm{cos}^2(\beta +\alpha )\mathrm{\Sigma }_{ZZ}(M_Z^2)`$ (213)
$``$ $`{\displaystyle \frac{e\mathrm{cos}(\beta \alpha )}{2s_WM_W}}\left\{\left[1+\mathrm{sin}^2(\beta \alpha )\right]T_{H^0}\mathrm{cos}(\beta \alpha )\mathrm{sin}(\beta \alpha )T_{h^0}\right\}`$
$`+`$ $`M_Z^2\mathrm{cos}(\beta +\alpha )\mathrm{cos}(\beta \alpha )\left[\delta Z_{H_1}\delta Z_{H_2}\right]`$
$`+`$ $`M_Z^2\mathrm{cos}^2(\beta +\alpha )\left[\delta Z_{H_1}\mathrm{sin}^2\beta +\delta Z_{H_2}\mathrm{cos}^2\beta \right],`$
$`\delta M_{h^0h^0}^2`$ $`=`$ $`\mathrm{cos}^2(\beta \alpha )\delta M_{A^0A^0}^2+\mathrm{sin}^2(\beta +\alpha )\mathrm{\Sigma }_{ZZ}(M_Z^2)`$ (214)
$`+`$ $`{\displaystyle \frac{e\mathrm{sin}(\beta \alpha )}{2s_WM_W}}\left\{\mathrm{sin}(\beta \alpha )\mathrm{cos}(\beta \alpha )T_{H^0}\left[1+\mathrm{cos}^2(\beta \alpha )\right]T_{h^0}\right\}`$
$``$ $`M_Z^2\mathrm{sin}(\beta +\alpha )\mathrm{sin}(\beta \alpha )\left[\delta Z_{H_1}\delta Z_{H_2}\right]`$
$`+`$ $`M_Z^2\mathrm{sin}^2(\beta +\alpha )\left[\delta Z_{H_1}\mathrm{sin}^2\beta +\delta Z_{H_2}\mathrm{cos}^2\beta \right],`$
$`\delta M_{H^0h^0}^2`$ $`=`$ $`\mathrm{sin}(\beta \alpha )\mathrm{cos}(\beta \alpha )\delta M_{A^0A^0}^2\mathrm{cos}(\beta +\alpha )\mathrm{sin}(\beta +\alpha )\mathrm{\Sigma }_{ZZ}(M_Z^2)`$ (215)
$``$ $`{\displaystyle \frac{e}{2s_WM_W}}\left\{\mathrm{sin}^3(\beta \alpha )T_{H^0}+\mathrm{cos}^3(\beta \alpha )T_{h^0}\right\}`$
$``$ $`M_Z^2\mathrm{sin}\alpha \mathrm{cos}\alpha \left[\delta Z_{H_1}\delta Z_{H_2}\right]`$
$``$ $`M_Z^2\mathrm{cos}(\beta +\alpha )\mathrm{sin}(\beta +\alpha )\left[\delta Z_{H_1}\mathrm{sin}^2\beta +\delta Z_{H_2}\mathrm{cos}^2\beta \right].`$
There is no contribution from the neutral Higgs self-energies to $`\mathrm{\Gamma }^{fin,\gamma }(H^0A^0/h^0A^0)`$. As for their contributions to $`\mathrm{\Gamma }^{fin,Z}(H^0A^0/h^0A^0)`$, they can be derived using the renormalized self-energies, which are expressed as follows:
$`\widehat{\mathrm{\Sigma }}_{H^0H^0}(q^2)`$ $`=`$ $`\mathrm{\Sigma }_{H^0H^0}(q^2)+q^2\left[\delta Z_{H_1}\mathrm{cos}^2\alpha +\delta Z_{H_2}\mathrm{sin}^2\alpha \right]\delta M_{H^0H^0}^2,`$ (216)
$`\widehat{\mathrm{\Sigma }}_{h^0h^0}(q^2)`$ $`=`$ $`\mathrm{\Sigma }_{h^0h^0}(q^2)+q^2\left[\delta Z_{H_1}\mathrm{sin}^2\alpha +\delta Z_{H_2}\mathrm{cos}^2\alpha \right]\delta M_{h^0h^0}^2,`$ (217)
$`\widehat{\mathrm{\Sigma }}_{H^0h^0}(q^2)`$ $`=`$ $`\mathrm{\Sigma }_{H^0h^0}(q^2)+q^2\mathrm{sin}\alpha \mathrm{cos}\alpha \left[\delta Z_{H_2}\delta Z_{H_1}\right]\delta M_{H^0h^0}^2.`$ (218)
For $`H^0A^0`$ final states, one has:
$$\mathrm{\Gamma }_{H^0A^0}^Z(H.s.e)=\frac{e\mathrm{sin}(\beta \alpha )}{s_Wc_W}[\text{cot}(\beta \alpha )\frac{\widehat{\mathrm{\Sigma }}_{H^0h^0}(M_{H^0}^2)}{M_{H^0}^2M_{h^0}^2}\frac{1}{2}\left(\frac{d\widehat{\mathrm{\Sigma }}_{H^0H^0}}{dq^2}\right)_{q^2=M_{H^0}^2}].$$
(219)
For $`h^0A^0`$ final states, one has:
$$\mathrm{\Gamma }_{h^0A^0}^Z(H.s.e)=+\frac{e\mathrm{cos}(\beta \alpha )}{s_Wc_W}[\mathrm{tan}(\beta \alpha )\frac{\widehat{\mathrm{\Sigma }}_{H^0h^0}(M_{h^0}^2)}{M_{h^0}^2M_{H^0}^2}\frac{1}{2}\left(\frac{d\widehat{\mathrm{\Sigma }}_{h^0h^0}}{dq^2}\right)_{q^2=M_{h^0}^2}].$$
(220)
#### 5.3.9 Neutral Higgs counter terms
For $`H^0A^0`$ final states, the neutral Higgs counter terms give the following contribution:
$`\mathrm{\Gamma }_{H^0A^0}^Z(H.c.t)`$ $`=`$ $`{\displaystyle \frac{e\mathrm{cos}(\beta \alpha )}{2s_Wc_W}}\left[\mathrm{cos}\beta \mathrm{sin}\beta +\mathrm{cos}\alpha \mathrm{sin}\alpha \right](\delta Z_{H_2}\delta Z_{H_1})`$ (221)
$``$ $`{\displaystyle \frac{e\mathrm{sin}(\beta \alpha )}{2s_Wc_W}}\left[(\mathrm{cos}^2\alpha +\mathrm{sin}^2\beta )\delta Z_{H_1}+(\mathrm{sin}^2\alpha +\mathrm{cos}^2\beta )\delta Z_{H_2}\right].`$
For $`h^0A^0`$ final states, the neutral Higgs counter terms give the following contribution:
$`\mathrm{\Gamma }_{h^0A^0}^Z(H.c.t)`$ $`=`$ $`{\displaystyle \frac{e\mathrm{sin}(\beta \alpha )}{2s_Wc_W}}\left[\mathrm{cos}\beta \mathrm{sin}\beta \mathrm{cos}\alpha \mathrm{sin}\alpha \right](\delta Z_{H_2}\delta Z_{H_1})`$ (222)
$`+`$ $`{\displaystyle \frac{e\mathrm{cos}(\beta \alpha )}{2s_Wc_W}}\left[(\mathrm{sin}^2\alpha +\mathrm{sin}^2\beta )\delta Z_{H_1}+(\mathrm{cos}^2\alpha +\mathrm{cos}^2\beta )\delta Z_{H_2}\right].`$
## 6 Contribution of box diagrams
### 6.1 Diagram structures for boxes
Several useful expressions are needed when estimating the contributions of the box diagrams. The particles inside the box are ordered clockwise and have internal masses $`m_1`$, $`m_2`$, $`m_3`$, $`m_4`$ starting with $`m_1`$ between the $`e^{}`$ and $`e^+`$ junctions. In the following, we will use the Mandelstam variables $`s`$, $`t`$ and $`u`$, which can be expressed as a function of $`q^2`$ and of the masses $`M_1`$ and $`M_2`$ of the two outgoing Higgs bosons (i.e. $`H^+H^{}`$ or $`H^0A^0`$ or $`h^0A^0`$):
$`s`$ $`=`$ $`q^2,`$ (223)
$`t`$ $`=`$ $`{\displaystyle \frac{1}{2}}(M_2^2+M_1^2s)`$ (224)
$`+`$ $`{\displaystyle \frac{s\mathrm{cos}\theta }{2}}\sqrt{\left(1+{\displaystyle \frac{M_2+M_1}{\sqrt{s}}}\right)\left(1{\displaystyle \frac{M_2+M_1}{\sqrt{s}}}\right)\left(1+{\displaystyle \frac{M_2M_1}{\sqrt{s}}}\right)\left(1{\displaystyle \frac{M_2M_1}{\sqrt{s}}}\right)},`$
$`u`$ $`=`$ $`{\displaystyle \frac{1}{2}}(M_2^2+M_1^2s)`$ (225)
$``$ $`{\displaystyle \frac{s\mathrm{cos}\theta }{2}}\sqrt{\left(1+{\displaystyle \frac{M_2+M_1}{\sqrt{s}}}\right)\left(1{\displaystyle \frac{M_2+M_1}{\sqrt{s}}}\right)\left(1+{\displaystyle \frac{M_2M_1}{\sqrt{s}}}\right)\left(1{\displaystyle \frac{M_2M_1}{\sqrt{s}}}\right)}.`$
Note that $`u`$ and $`t`$ have an angular dependence and, as a result, so does the contribution of the diagram boxes to the one loop cross sections.
Let $`\mathrm{}`$ and $`\mathrm{}^{}`$ be the momenta of the incoming electron and positron, respectively. Let $`P_{f1}`$ and $`P_{f2}`$ be the momenta of the two outgoing Higgs bosons. When not otherwise specified, $`\mathrm{}^{}`$, $`P_{f1}`$, $`P_{f2}`$ and $`\mathrm{}`$ are oriented clockwise around the box. Then, one has:
$`P_{f1}^2`$ $`=`$ $`M_1^2,`$ (226)
$`P_{f2}^2`$ $`=`$ $`M_2^2,`$ (227)
$`P_{f1}P_{f2}`$ $`=`$ $`{\displaystyle \frac{s(M_1^2+M_2^2)}{2}},`$ (228)
$`\mathrm{}^{}P_{f1}`$ $`=`$ $`{\displaystyle \frac{tM_1^2}{2}},`$ (229)
$`\mathrm{}^{}P_{f2}`$ $`=`$ $`{\displaystyle \frac{uM_2^2}{2}}.`$ (230)
a) Box7 structures:
$`๐_7`$ $`=`$ $`6(D_{002}D_{003})+P_{f1}^2(D_{222}D_{223})+P_{f2}^2(D_{332}D_{333})`$ (231)
$``$ $`2P_{f1}P_{f2}(D_{323}D_{322})2\mathrm{}^{}P_{f2}(D_{133}D_{132})2\mathrm{}^{}P_{f1}(D_{123}D_{122})`$
$`+`$ $`4\left[P_{f1}^2D_{22}+P_{f2}^2D_{23}+2\mathrm{}^{}P_{f1}D_{24}+2\mathrm{}^{}P_{f2}D_{25}+2P_{f1}P_{f2}D_{26}+4D_{27}\right]`$
$``$ $`4\left[P_{f2}^2D_{13}P_{f1}^2D_{12}2\mathrm{}^{}P_{f1}D_{11}\right].`$
b) Box8 structures:
$`๐_8`$ $`=`$ $`6(D_{002}D_{003})+P_{f1}^2(D_{222}D_{223})+P_{f2}^2(D_{332}D_{333})`$ (232)
$``$ $`2P_{f1}P_{f2}(D_{323}D_{322})2\mathrm{}^{}P_{f2}(D_{133}D_{132})2\mathrm{}^{}P_{f1}(D_{123}D_{122})`$
$`+`$ $`(2\mathrm{}^{}P_{f1}+P_{f1}^2)D_{22}(2\mathrm{}^{}P_{f2}+2P_{f1}P_{f2}+P_{f2}^2)D_{23}(2\mathrm{}^{}P_{f2}+2\mathrm{}^{}P_{f1})D_{25}`$
$`+`$ $`(2\mathrm{}^{}P_{f2}2\mathrm{}^{}P_{f1}2P_{f1}^2)D_{26}2D_{27},`$
$`๐_8^{}`$ $`=`$ $`D_{12}D_{13},`$ (233)
$`D_8^{\prime \prime }`$ $`=`$ $`D_0+D_{12}D_{13}.`$ (234)
c) Box9 structures:
$`๐_9`$ $`=`$ $`D_{12}D_{13}.`$ (235)
d) Box10 structures:
$`๐_{10}`$ $`=`$ $`D_{12}+D_0,`$ (236)
$`D_{10}^{}`$ $`=`$ $`D_{12}.`$ (237)
Finally, the crossed functions $`\overline{๐}_j`$ are obtained from the $`๐_j`$ functions by making the following changes: $`P_{f1}P_{f2}`$, i.e. $`tu`$ and $`M_1^2M_2^2`$.
### 6.2 Charged Higgs sector
For $`e^+e^{}H^+H^{}`$, the box diagrams give the following one loop contribution:
$`a_{L,R}^{box}(H^+H^{})`$ $`={\displaystyle \frac{q^2}{2e^2}}\times `$ $`\{A_{L,R}^{box7}(H^+H^{})+A_{L,R}^{box8}(H^+H^{})`$ (238)
$`+A_{L,R}^{box9}(H^+H^{})+A_{L,R}^{box10}(H^+H^{})\}.`$
#### 6.2.1 Box7 diagrams
The amplitude $`A_{L,R}^{box7}(H^+H^{})`$ is obtained by summing various contributions with gauge and Higgs bosons inside the box:
$`A_{L,R}^{box7}(H^+H^{})`$ $`=`$ $`{\displaystyle \frac{\alpha _{em}^2}{8s_W^4}}\mathrm{sin}^2(\beta \alpha )P_L๐_7(\nu WH^0W)`$ (239)
$`+`$ $`{\displaystyle \frac{\alpha _{em}^2}{8s_W^4}}\mathrm{cos}^2(\beta \alpha )P_L๐_7(\nu Wh^0W)`$
$`+`$ $`{\displaystyle \frac{\alpha _{em}^2}{8s_W^4}}P_L๐_7(\nu WA^0W)`$
$``$ $`\left({\displaystyle \frac{12s_W^2}{2s_Wc_W}}\right)\times \left[{\displaystyle \frac{g_LP_L+g_RP_R}{2s_Wc_W}}\right]\times \alpha _{em}^2\left\{๐_7(e\gamma HZ)\overline{๐}_7(e\gamma HZ)\right\}`$
$``$ $`\left({\displaystyle \frac{12s_W^2}{2s_Wc_W}}\right)\times \left[{\displaystyle \frac{g_LP_L+g_RP_R}{2s_Wc_W}}\right]\times \alpha _{em}^2\left\{๐_7(eZH\gamma )\overline{๐}_7(eZH\gamma )\right\}`$
$`+`$ $`\left({\displaystyle \frac{12s_W^2}{2s_Wc_W}}\right)^2\times \left[{\displaystyle \frac{g_L^2P_L+g_R^2P_R}{4s_W^2c_W^2}}\right]\times \alpha _{em}^2\left\{๐_7(eZHZ)\overline{๐}_7(eZHZ)\right\}`$
$`+`$ $`\left[P_L+P_R\right]\times \alpha _{em}^2\left\{๐_7(e\gamma H\gamma )\overline{๐}_7(e\gamma H\gamma )\right\}.`$
#### 6.2.2 Box8 diagrams
The amplitude $`A_{L,R}^{box8}(H^+H^{})`$ is obtained by summing two types of box diagrams, which have gauginos inside the box, $`\stackrel{~}{\nu }\stackrel{~}{\chi }\stackrel{~}{\chi }^0\stackrel{~}{\chi }`$ and $`\stackrel{~}{e}\stackrel{~}{\chi }^0\stackrel{~}{\chi }\stackrel{~}{\chi }^0`$:
$`A_{L,R}^{box8}(H^+H^{})`$ $`=`$ $`P_{L,R}\left[A_{H^+H^{}}^{box8}(\stackrel{~}{\nu }\stackrel{~}{\chi }\stackrel{~}{\chi }^0\stackrel{~}{\chi })+A_{H^+H^{}}^{box8}(\stackrel{~}{e}\stackrel{~}{\chi }^0\stackrel{~}{\chi }\stackrel{~}{\chi }^0)\right].`$ (240)
For the $`\stackrel{~}{\nu }\stackrel{~}{\chi }\stackrel{~}{\chi }^0\stackrel{~}{\chi }`$ boxes, since there is no right-handed sneutrino in the MSSM, only a left-handed term is considered:
$`A_{H^+H^{}}^{box8}(\stackrel{~}{\nu }_L\stackrel{~}{\chi }\stackrel{~}{\chi }^0\stackrel{~}{\chi })`$ $`={\displaystyle \frac{e^2}{16\pi ^2s_W^2}}{\displaystyle \underset{ijk}{}}Z_{1i}^+Z_{1k}^+\times `$ $`\{M_{\stackrel{~}{\chi }_i}M_{\stackrel{~}{\chi }_k}c_{Hij}^Lc_{Hkj}^L๐_8^{\prime \prime }(\stackrel{~}{\nu }_L\stackrel{~}{\chi }_i\stackrel{~}{\chi }_j^0\stackrel{~}{\chi }_k)`$ (241)
$`+M_{\stackrel{~}{\chi }_i}M_{\stackrel{~}{\chi }_j^0}c_{Hij}^Lc_{Hkj}^R๐_8^{}(\stackrel{~}{\nu }_L\stackrel{~}{\chi }_i\stackrel{~}{\chi }_j^0\stackrel{~}{\chi }_k)`$
$`+M_{\stackrel{~}{\chi }_j^0}M_{\stackrel{~}{\chi }_k}c_{Hij}^Rc_{Hkj}^L๐_8^{}(\stackrel{~}{\nu }_L\stackrel{~}{\chi }_i\stackrel{~}{\chi }_j^0\stackrel{~}{\chi }_k)`$
$`+c_{Hij}^Rc_{Hkj}^R๐_8(\stackrel{~}{\nu }_L\stackrel{~}{\chi }_i\stackrel{~}{\chi }_j^0\stackrel{~}{\chi }_k)\}.`$
For the $`\stackrel{~}{e}\stackrel{~}{\chi }^0\stackrel{~}{\chi }\stackrel{~}{\chi }^0`$ boxes, one has both left-handed and right-handed terms. The left-handed term is given by:
$`A_{H^+H^{}}^{box8}(\stackrel{~}{e}_L\stackrel{~}{\chi }^0\stackrel{~}{\chi }\stackrel{~}{\chi }^0)`$ $`={\displaystyle \frac{e^2}{32\pi ^2s_W^2c_W^2}}`$ $`{\displaystyle \underset{ijk}{}}(Z_{1i}^Ns_W+Z_{2i}^Nc_W)(Z_{1k}^Ns_W+Z_{2k}^Nc_W)\times `$ (242)
$`\{M_{\stackrel{~}{\chi }_i^0}M_{\stackrel{~}{\chi }_k^0}c_{Hji}^Rc_{Hjk}^R\overline{๐}_8^{\prime \prime }(\stackrel{~}{e}_L\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j\stackrel{~}{\chi }_k^0)`$
$`+M_{\stackrel{~}{\chi }_i^0}M_{\stackrel{~}{\chi }_j}c_{Hji}^Rc_{Hjk}^L\overline{๐}_8^{}(\stackrel{~}{e}_L\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j\stackrel{~}{\chi }_k^0)`$
$`+M_{\stackrel{~}{\chi }_j}M_{\stackrel{~}{\chi }_k^0}c_{Hji}^Lc_{Hjk}^R\overline{๐}_8^{}(\stackrel{~}{e}_L\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j\stackrel{~}{\chi }_k^0)`$
$`+c_{Hji}^Lc_{Hjk}^L\overline{๐}_8(\stackrel{~}{e}_L\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j\stackrel{~}{\chi }_k^0)\}`$
$`{\displaystyle \frac{e^2}{32\pi ^2s_W^2c_W^2}}`$ $`{\displaystyle \underset{ijk}{}}(Z_{1i}^Ns_W+Z_{2i}^Nc_W)(Z_{1k}^Ns_W+Z_{2k}^Nc_W)\times `$
$`\{M_{\stackrel{~}{\chi }_i^0}M_{\stackrel{~}{\chi }_k^0}c_{Hji}^Lc_{Hjk}^L๐_8^{\prime \prime }(\stackrel{~}{e}_L\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j\stackrel{~}{\chi }_k^0)`$
$`+M_{\stackrel{~}{\chi }_i^0}M_{\stackrel{~}{\chi }_j}c_{Hji}^Lc_{Hjk}^R๐_8^{}(\stackrel{~}{e}_L\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j\stackrel{~}{\chi }_k^0)`$
$`+M_{\stackrel{~}{\chi }_j}M_{\stackrel{~}{\chi }_k^0}c_{Hji}^Rc_{Hjk}^L๐_8^{}(\stackrel{~}{e}_L\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j\stackrel{~}{\chi }_k^0)`$
$`+c_{Hji}^Rc_{Hjk}^R๐_8(\stackrel{~}{e}_L\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j\stackrel{~}{\chi }_k^0)\}`$
while the right-handed term is:
$`A_{H^+H^{}}^{box8}(\stackrel{~}{e}_R\stackrel{~}{\chi }^0\stackrel{~}{\chi }\stackrel{~}{\chi }^0)`$ $`={\displaystyle \frac{e^2}{8\pi ^2c_W^2}}{\displaystyle \underset{ijk}{}}Z_{1i}^NZ_{1k}^N\times `$ $`\{M_{\stackrel{~}{\chi }_i^0}M_{\stackrel{~}{\chi }_k^0}c_{Hji}^Lc_{Hjk}^L\overline{๐}_8^{\prime \prime }(\stackrel{~}{e}_R\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j\stackrel{~}{\chi }_k^0)`$ (243)
$`M_{\stackrel{~}{\chi }_i^0}M_{\stackrel{~}{\chi }_j}c_{Hji}^Lc_{Hjk}^R\overline{๐}_8^{}(\stackrel{~}{e}_R\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j\stackrel{~}{\chi }_k^0)`$
$`+M_{\stackrel{~}{\chi }_j}M_{\stackrel{~}{\chi }_k^0}c_{Hji}^Rc_{Hjk}^L\overline{๐}_8^{}(\stackrel{~}{e}_R\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j\stackrel{~}{\chi }_k^0)`$
$`+c_{Hji}^Rc_{Hjk}^R\overline{๐}_8(\stackrel{~}{e}_R\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j\stackrel{~}{\chi }_k^0)\}`$
$`{\displaystyle \frac{e^2}{8\pi ^2c_W^2}}{\displaystyle \underset{ijk}{}}Z_{1i}^NZ_{1k}^N\times `$ $`\{M_{\stackrel{~}{\chi }_i^0}M_{\stackrel{~}{\chi }_k^0}c_{Hji}^Rc_{Hjk}^R๐_8^{\prime \prime }(\stackrel{~}{e}_R\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j\stackrel{~}{\chi }_k^0)`$
$`+M_{\stackrel{~}{\chi }_i^0}M_{\stackrel{~}{\chi }_j}c_{Hji}^Rc_{Hjk}^L๐_8^{}(\stackrel{~}{e}_R\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j\stackrel{~}{\chi }_k^0)`$
$`+M_{\stackrel{~}{\chi }_j}M_{\stackrel{~}{\chi }_k^0}c_{Hji}^Lc_{Hjk}^R๐_8^{}(\stackrel{~}{e}_R\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j\stackrel{~}{\chi }_k^0)`$
$`+c_{Hji}^Lc_{Hjk}^L๐_8(\stackrel{~}{e}_R\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j\stackrel{~}{\chi }_k^0)\}.`$
#### 6.2.3 Box9 diagrams
There is no right-handed contribution from box9 diagrams. As for the left-handed amplitude $`A_L^{box9}(H^+H^{})`$, it is obtained as follows:
$`A_L^{box9}(H^+H^{})`$ $`=`$ $`{\displaystyle \frac{e^2}{32\pi ^2s_W^2c_W^2}}{\displaystyle \underset{i}{}}|Z_{1i}^Ns_W+Z_{2i}^Nc_W|^2g_{H\stackrel{~}{e}_L\stackrel{~}{\nu }_L}^2๐_9(\stackrel{~}{\chi }_i^0\stackrel{~}{e}_L\stackrel{~}{\nu }_L\stackrel{~}{e}_L)`$ (244)
$``$ $`{\displaystyle \frac{e^2}{16\pi ^2s_W^2}}{\displaystyle \underset{i}{}}|Z_{1i}^+|^2g_{H\stackrel{~}{e}_L\stackrel{~}{\nu }_L}^2\overline{๐}_9(\stackrel{~}{\chi }_i\stackrel{~}{\nu }_L\stackrel{~}{e}_L\stackrel{~}{\nu }_L).`$
#### 6.2.4 Twisted box10 diagrams
Twisted box10 diagrams contribute only with a left-handed term:
$`A_L^{box10}(H^+H^{})`$ $`={\displaystyle \frac{e^2}{16\sqrt{2}\pi ^2s_W^2c_W}}`$ $`{\displaystyle \underset{ij}{}}Z_{1i}^+(Z_{1j}^Ns_W+Z_{2j}^Nc_W)\times g_{H\stackrel{~}{e}_L\stackrel{~}{\nu }_L}\times `$ (245)
$`\left[M_{\stackrel{~}{\chi }_i}c_{Hij}^L๐_{10}(\stackrel{~}{\nu }_L\stackrel{~}{\chi }_i\stackrel{~}{\chi }_j^0\stackrel{~}{e}_L)+M_{\stackrel{~}{\chi }_j^0}c_{Hij}^R๐_{10}^{}(\stackrel{~}{\nu }_L\stackrel{~}{\chi }_i\stackrel{~}{\chi }_j^0\stackrel{~}{e}_L)\right]`$
$`{\displaystyle \frac{e^2}{16\sqrt{2}\pi ^2s_W^2c_W}}`$ $`{\displaystyle \underset{ij}{}}Z_{1i}^+(Z_{1j}^Ns_W+Z_{2j}^Nc_W)\times g_{H\stackrel{~}{e}_L\stackrel{~}{\nu }_L}\times `$
$`\left[M_{\stackrel{~}{\chi }_j^0}c_{Hij}^R\overline{๐}_{10}(\stackrel{~}{e}_L\stackrel{~}{\chi }_j^0\stackrel{~}{\chi }_i\stackrel{~}{\nu }_L)+M_{\stackrel{~}{\chi }_i}c_{Hij}^L\overline{๐}_{10}^{}(\stackrel{~}{e}_L\stackrel{~}{\chi }_j^0\stackrel{~}{\chi }_i\stackrel{~}{\nu }_L)\right].`$
### 6.3 Neutral Higgs sector
For $`e^+e^{}H^0A^0/h^0A^0`$, there is no box9 diagram, so one has:
$`a_{L,R}^{box}(H^0A^0/h^0A^0)`$ $`={\displaystyle \frac{iq^2}{2e^2}}`$ $`\{A_{L,R}^{box7}(H^0A^0/h^0A^0)+A_{L,R}^{box8}(H^0A^0/h^0A^0)`$ (246)
$`+A_{L,R}^{box10}(H^0A^0/h^0A^0)\}.`$
#### 6.3.1 Box7 diagrams
In the neutral Higgs sector, there is no right-handed contribution from box7 diagrams. The left-handed amplitude $`A_L^{box7}(H^0A^0/h^0A^0)`$ is obtained as follows:
$`A_L^{box7}(H^0A^0/h^0A^0)`$ $`=`$ $`{\displaystyle \frac{e^4}{128\pi ^2s_W^4}}\times \left[Z_{ab}\right]\times \left\{๐_7(\nu WHW)+\overline{๐}_7(\nu WHW)\right\}.`$ (247)
#### 6.3.2 Box8 diagrams
Both $`\stackrel{~}{\nu }\stackrel{~}{\chi }\stackrel{~}{\chi }\stackrel{~}{\chi }`$ and $`\stackrel{~}{e}\stackrel{~}{\chi }^0\stackrel{~}{\chi }^0\stackrel{~}{\chi }^0`$ box diagrams contribute to the amplitude $`A_{L,R}^{box8}(H^0A^0/h^0A^0)`$:
$`A_{L,R}^{box8}(H^0A^0/h^0A^0)`$ $`=`$ $`P_{L,R}\left[A_{H^0A^0/h^0A^0}^{box8}(\stackrel{~}{\nu }\stackrel{~}{\chi }\stackrel{~}{\chi }\stackrel{~}{\chi })+A_{H^0A^0/h^0A^0}^{box8}(\stackrel{~}{e}\stackrel{~}{\chi }^0\stackrel{~}{\chi }^0\stackrel{~}{\chi }^0)\right].`$ (248)
The $`\stackrel{~}{\nu }\stackrel{~}{\chi }\stackrel{~}{\chi }\stackrel{~}{\chi }`$ boxes contribute only with left-handed terms:
$`A_{H^0A^0}^{box8}(\stackrel{~}{\nu }_L\stackrel{~}{\chi }\stackrel{~}{\chi }\stackrel{~}{\chi })`$ $`={\displaystyle \frac{e^2}{16\pi ^2s_W^2}}{\displaystyle \underset{ijk}{}}Z_{1i}^+Z_{1k}^+\times `$ $`\{M_{\stackrel{~}{\chi }_i}M_{\stackrel{~}{\chi }_k}c_{A^0ji}^Rc_{H^0kj}^L๐_8^{\prime \prime }(\stackrel{~}{\nu }_L\stackrel{~}{\chi }_i\stackrel{~}{\chi }_j\stackrel{~}{\chi }_k)`$ (249)
$`+M_{\stackrel{~}{\chi }_i}M_{\stackrel{~}{\chi }_j}c_{A^0ji}^Rc_{H^0kj}^R๐_8^{}(\stackrel{~}{\nu }_L\stackrel{~}{\chi }_i\stackrel{~}{\chi }_j\stackrel{~}{\chi }_k)`$
$`+M_{\stackrel{~}{\chi }_j}M_{\stackrel{~}{\chi }_k}c_{A^0ji}^Lc_{H^0kj}^L๐_8^{}(\stackrel{~}{\nu }_L\stackrel{~}{\chi }_i\stackrel{~}{\chi }_j\stackrel{~}{\chi }_k)`$
$`+c_{A^0ji}^Lc_{H^0kj}^R๐_8(\stackrel{~}{\nu }_L\stackrel{~}{\chi }_i\stackrel{~}{\chi }_j\stackrel{~}{\chi }_k)\}`$
$`{\displaystyle \frac{e^2}{16\pi ^2s_W^2}}{\displaystyle \underset{ijk}{}}Z_{1i}^+Z_{1k}^+\times `$ $`\{M_{\stackrel{~}{\chi }_i}M_{\stackrel{~}{\chi }_k}c_{H^0ji}^Rc_{A^0kj}^L\overline{๐}_8^{\prime \prime }(\stackrel{~}{\nu }_L\stackrel{~}{\chi }_i\stackrel{~}{\chi }_j\stackrel{~}{\chi }_k)`$
$`+M_{\stackrel{~}{\chi }_i}M_{\stackrel{~}{\chi }_j}c_{H^0ji}^Rc_{A^0kj}^R\overline{๐}_8^{}(\stackrel{~}{\nu }_L\stackrel{~}{\chi }_i\stackrel{~}{\chi }_j\stackrel{~}{\chi }_k)`$
$`+M_{\stackrel{~}{\chi }_j}M_{\stackrel{~}{\chi }_k}c_{H^0ji}^Lc_{A^0kj}^L\overline{๐}_8^{}(\stackrel{~}{\nu }_L\stackrel{~}{\chi }_i\stackrel{~}{\chi }_j\stackrel{~}{\chi }_k)`$
$`+c_{H^0ji}^Lc_{A^0kj}^R\overline{๐}_8(\stackrel{~}{\nu }_L\stackrel{~}{\chi }_i\stackrel{~}{\chi }_j\stackrel{~}{\chi }_k)\},`$
$`A_{h^0A^0}^{box8}(\stackrel{~}{\nu }_L\stackrel{~}{\chi }\stackrel{~}{\chi }\stackrel{~}{\chi })`$ $`={\displaystyle \frac{e^2}{16\pi ^2s_W^2}}{\displaystyle \underset{ijk}{}}Z_{1i}^+Z_{1k}^+\times `$ $`\{M_{\stackrel{~}{\chi }_i}M_{\stackrel{~}{\chi }_k}c_{A^0ji}^Rc_{h^0kj}^L๐_8^{\prime \prime }(\stackrel{~}{\nu }_L\stackrel{~}{\chi }_i\stackrel{~}{\chi }_j\stackrel{~}{\chi }_k)`$ (250)
$`+M_{\stackrel{~}{\chi }_i}M_{\stackrel{~}{\chi }_j}c_{A^0ji}^Rc_{h^0kj}^R๐_8^{}(\stackrel{~}{\nu }_L\stackrel{~}{\chi }_i\stackrel{~}{\chi }_j\stackrel{~}{\chi }_k)`$
$`+M_{\stackrel{~}{\chi }_j}M_{\stackrel{~}{\chi }_k}c_{A^0ji}^Lc_{h^0kj}^L๐_8^{}(\stackrel{~}{\nu }_L\stackrel{~}{\chi }_i\stackrel{~}{\chi }_j\stackrel{~}{\chi }_k)`$
$`+c_{A^0ji}^Lc_{h^0kj}^R๐_8(\stackrel{~}{\nu }_L\stackrel{~}{\chi }_i\stackrel{~}{\chi }_j\stackrel{~}{\chi }_k)\}`$
$`{\displaystyle \frac{e^2}{16\pi ^2s_W^2}}{\displaystyle \underset{ijk}{}}Z_{1i}^+Z_{1k}^+\times `$ $`\{M_{\stackrel{~}{\chi }_i}M_{\stackrel{~}{\chi }_k}c_{h^0ji}^Rc_{A^0kj}^L\overline{๐}_8^{\prime \prime }(\stackrel{~}{\nu }_L\stackrel{~}{\chi }_i\stackrel{~}{\chi }_j\stackrel{~}{\chi }_k)`$
$`+M_{\stackrel{~}{\chi }_i}M_{\stackrel{~}{\chi }_j}c_{h^0ji}^Rc_{A^0kj}^R\overline{๐}_8^{}(\stackrel{~}{\nu }_L\stackrel{~}{\chi }_i\stackrel{~}{\chi }_j\stackrel{~}{\chi }_k)`$
$`+M_{\stackrel{~}{\chi }_j}M_{\stackrel{~}{\chi }_k}c_{h^0ji}^Lc_{A^0kj}^L\overline{๐}_8^{}(\stackrel{~}{\nu }_L\stackrel{~}{\chi }_i\stackrel{~}{\chi }_j\stackrel{~}{\chi }_k)`$
$`+c_{h^0ji}^Lc_{A^0kj}^R\overline{๐}_8(\stackrel{~}{\nu }_L\stackrel{~}{\chi }_i\stackrel{~}{\chi }_j\stackrel{~}{\chi }_k)\}.`$
The $`\stackrel{~}{e}\stackrel{~}{\chi }^0\stackrel{~}{\chi }^0\stackrel{~}{\chi }^0`$ boxes contribute with both left-handed and right-handed terms:
$`A_{H^0A^0}^{box8}(\stackrel{~}{e}\stackrel{~}{\chi }^0\stackrel{~}{\chi }^0\stackrel{~}{\chi }^0)`$ $`={\displaystyle \frac{e^2}{32\pi ^2s_W^2c_W^2}}P_L`$ $`{\displaystyle \underset{ijk}{}}(Z_{1i}^Ns_W+Z_{2i}^Nc_W)(Z_{1k}^Ns_W+Z_{2k}^Nc_W)\times `$ (251)
$`\{M_{\stackrel{~}{\chi }_i^0}M_{\stackrel{~}{\chi }_k^0}n_{A^0ij}^Rn_{H^0jk}^L๐_8^{\prime \prime }(\stackrel{~}{e}_L\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j^0\stackrel{~}{\chi }_k^0)`$
$`+M_{\stackrel{~}{\chi }_i^0}M_{\stackrel{~}{\chi }_j^0}n_{A^0ij}^Rn_{H^0jk}^R๐_8^{}(\stackrel{~}{e}_L\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j^0\stackrel{~}{\chi }_k^0)`$
$`+M_{\stackrel{~}{\chi }_j^0}M_{\stackrel{~}{\chi }_k^0}n_{A^0ij}^Ln_{H^0jk}^L๐_8^{}(\stackrel{~}{e}_L\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j^0\stackrel{~}{\chi }_k^0)`$
$`+n_{A^0ij}^Ln_{H^0jk}^R๐_8(\stackrel{~}{e}_L\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j^0\stackrel{~}{\chi }_k^0)\}`$
$`{\displaystyle \frac{e^2}{32\pi ^2s_W^2c_W^2}}P_L`$ $`{\displaystyle \underset{ijk}{}}(Z_{1i}^Ns_W+Z_{2i}^Nc_W)(Z_{1k}^Ns_W+Z_{2k}^Nc_W)\times `$
$`\{M_{\stackrel{~}{\chi }_i^0}M_{\stackrel{~}{\chi }_k^0}n_{H^0ij}^Rn_{A^0jk}^L\overline{๐}_8^{\prime \prime }(\stackrel{~}{e}_L\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j^0\stackrel{~}{\chi }_k^0)`$
$`+M_{\stackrel{~}{\chi }_i^0}M_{\stackrel{~}{\chi }_j^0}n_{H^0ij}^Rn_{A^0jk}^R\overline{๐}_8^{}(\stackrel{~}{e}_L\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j^0\stackrel{~}{\chi }_k^0)`$
$`+M_{\stackrel{~}{\chi }_j^0}M_{\stackrel{~}{\chi }_k^0}n_{H^0ij}^Ln_{A^0jk}^L\overline{๐}_8^{}(\stackrel{~}{e}_L\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j^0\stackrel{~}{\chi }_k^0)`$
$`+n_{H^0ij}^Ln_{A^0jk}^R\overline{๐}_8(\stackrel{~}{e}_L\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j^0\stackrel{~}{\chi }_k^0)\}`$
$`{\displaystyle \frac{e^2}{8\pi ^2c_W^2}}P_R{\displaystyle \underset{ijk}{}}Z_{1i}^NZ_{1k}^N\times `$ $`\{M_{\stackrel{~}{\chi }_i^0}M_{\stackrel{~}{\chi }_k^0}n_{A^0ij}^Ln_{H^0jk}^R๐_8^{\prime \prime }(\stackrel{~}{e}_R\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j^0\stackrel{~}{\chi }_k^0)`$
$`+M_{\stackrel{~}{\chi }_i^0}M_{\stackrel{~}{\chi }_j^0}n_{A^0ij}^Ln_{H^0jk}^L๐_8^{}(\stackrel{~}{e}_R\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j^0\stackrel{~}{\chi }_k^0)`$
$`+M_{\stackrel{~}{\chi }_j^0}M_{\stackrel{~}{\chi }_k^0}n_{A^0ij}^Rn_{H^0jk}^R๐_8^{}(\stackrel{~}{e}_R\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j^0\stackrel{~}{\chi }_k^0)`$
$`+n_{A^0ij}^Rn_{H^0jk}^L๐_8(\stackrel{~}{e}_R\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j^0\stackrel{~}{\chi }_k^0)\}`$
$`+{\displaystyle \frac{e^2}{8\pi ^2c_W^2}}P_R{\displaystyle \underset{ijk}{}}Z_{1i}^NZ_{1k}^N\times `$ $`\{M_{\stackrel{~}{\chi }_i^0}M_{\stackrel{~}{\chi }_k^0}n_{H^0ij}^Ln_{A^0jk}^R\overline{๐}_8^{\prime \prime }(\stackrel{~}{e}_R\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j^0\stackrel{~}{\chi }_k^0)`$
$`+M_{\stackrel{~}{\chi }_i^0}M_{\stackrel{~}{\chi }_j^0}n_{H^0ij}^Ln_{A^0jk}^L\overline{๐}_8^{}(\stackrel{~}{e}_R\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j^0\stackrel{~}{\chi }_k^0)`$
$`+M_{\stackrel{~}{\chi }_j^0}M_{\stackrel{~}{\chi }_k^0}n_{H^0ij}^Rn_{A^0jk}^R\overline{๐}_8^{}(\stackrel{~}{e}_R\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j^0\stackrel{~}{\chi }_k^0)`$
$`+n_{H^0ij}^Rn_{A^0jk}^L\overline{๐}_8(\stackrel{~}{e}_R\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j^0\stackrel{~}{\chi }_k^0)\},`$
$`A_{h^0A^0}^{box8}(\stackrel{~}{e}\stackrel{~}{\chi }^0\stackrel{~}{\chi }^0\stackrel{~}{\chi }^0)`$ $`={\displaystyle \frac{e^2}{32\pi ^2s_W^2c_W^2}}P_L`$ $`{\displaystyle \underset{ijk}{}}(Z_{1i}^Ns_W+Z_{2i}^Nc_W)(Z_{1k}^Ns_W+Z_{2k}^Nc_W)\times `$ (252)
$`\{M_{\stackrel{~}{\chi }_i^0}M_{\stackrel{~}{\chi }_k^0}n_{A^0ij}^Rn_{h^0jk}^L๐_8^{\prime \prime }(\stackrel{~}{e}_L\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j^0\stackrel{~}{\chi }_k^0)`$
$`+M_{\stackrel{~}{\chi }_i^0}M_{\stackrel{~}{\chi }_j^0}n_{A^0ij}^Rn_{h^0jk}^R๐_8^{}(\stackrel{~}{e}_L\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j^0\stackrel{~}{\chi }_k^0)`$
$`+M_{\stackrel{~}{\chi }_j^0}M_{\stackrel{~}{\chi }_k^0}n_{A^0ij}^Ln_{h^0jk}^L๐_8^{}(\stackrel{~}{e}_L\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j^0\stackrel{~}{\chi }_k^0)`$
$`+n_{A^0ij}^Ln_{h^0jk}^R๐_8(\stackrel{~}{e}_L\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j^0\stackrel{~}{\chi }_k^0)\}`$
$`{\displaystyle \frac{e^2}{32\pi ^2s_W^2c_W^2}}P_L`$ $`{\displaystyle \underset{ijk}{}}(Z_{1i}^Ns_W+Z_{2i}^Nc_W)(Z_{1k}^Ns_W+Z_{2k}^Nc_W)\times `$
$`\{M_{\stackrel{~}{\chi }_i^0}M_{\stackrel{~}{\chi }_k^0}n_{h^0ij}^Rn_{A^0jk}^L\overline{๐}_8^{\prime \prime }(\stackrel{~}{e}_L\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j^0\stackrel{~}{\chi }_k^0)`$
$`+M_{\stackrel{~}{\chi }_i^0}M_{\stackrel{~}{\chi }_j^0}n_{h^0ij}^Rn_{A^0jk}^R\overline{๐}_8^{}(\stackrel{~}{e}_L\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j^0\stackrel{~}{\chi }_k^0)`$
$`+M_{\stackrel{~}{\chi }_j^0}M_{\stackrel{~}{\chi }_k^0}n_{h^0ij}^Ln_{A^0jk}^L\overline{๐}_8^{}(\stackrel{~}{e}_L\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j^0\stackrel{~}{\chi }_k^0)`$
$`+n_{h^0ij}^Ln_{A^0jk}^R\overline{๐}_8(\stackrel{~}{e}_L\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j^0\stackrel{~}{\chi }_k^0)\}`$
$`{\displaystyle \frac{e^2}{8\pi ^2c_W^2}}P_R{\displaystyle \underset{ijk}{}}Z_{1i}^NZ_{1k}^N\times `$ $`\{M_{\stackrel{~}{\chi }_i^0}M_{\stackrel{~}{\chi }_k^0}n_{A^0ij}^Ln_{h^0jk}^R๐_8^{\prime \prime }(\stackrel{~}{e}_R\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j^0\stackrel{~}{\chi }_k^0)`$
$`+M_{\stackrel{~}{\chi }_i^0}M_{\stackrel{~}{\chi }_j^0}n_{A^0ij}^Ln_{h^0jk}^L๐_8^{}(\stackrel{~}{e}_R\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j^0\stackrel{~}{\chi }_k^0)`$
$`+M_{\stackrel{~}{\chi }_j^0}M_{\stackrel{~}{\chi }_k^0}n_{A^0ij}^Rn_{h^0jk}^R๐_8^{}(\stackrel{~}{e}_R\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j^0\stackrel{~}{\chi }_k^0)`$
$`+n_{A^0ij}^Rn_{h^0jk}^L๐_8(\stackrel{~}{e}_R\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j^0\stackrel{~}{\chi }_k^0)\}`$
$`+{\displaystyle \frac{e^2}{8\pi ^2c_W^2}}P_R{\displaystyle \underset{ijk}{}}Z_{1i}^NZ_{1k}^N\times `$ $`\{M_{\stackrel{~}{\chi }_i^0}M_{\stackrel{~}{\chi }_k^0}n_{h^0ij}^Ln_{A^0jk}^R\overline{๐}_8^{\prime \prime }(\stackrel{~}{e}_R\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j^0\stackrel{~}{\chi }_k^0)`$
$`+M_{\stackrel{~}{\chi }_i^0}M_{\stackrel{~}{\chi }_j^0}n_{h^0ij}^Ln_{A^0jk}^L\overline{๐}_8^{}(\stackrel{~}{e}_R\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j^0\stackrel{~}{\chi }_k^0)`$
$`+M_{\stackrel{~}{\chi }_j^0}M_{\stackrel{~}{\chi }_k^0}n_{h^0ij}^Rn_{A^0jk}^R\overline{๐}_8^{}(\stackrel{~}{e}_R\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j^0\stackrel{~}{\chi }_k^0)`$
$`+n_{h^0ij}^Rn_{A^0jk}^L\overline{๐}_8(\stackrel{~}{e}_R\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j^0\stackrel{~}{\chi }_k^0)\}.`$
#### 6.3.3 Twisted box10 diagrams
Two types of twisted box10 diagrams must be considered: $`\stackrel{~}{\nu }\stackrel{~}{\chi }\stackrel{~}{\chi }\stackrel{~}{\nu }`$ and $`\stackrel{~}{e}\stackrel{~}{\chi }^0\stackrel{~}{\chi }^0\stackrel{~}{e}`$.
The $`\stackrel{~}{\nu }\stackrel{~}{\chi }\stackrel{~}{\chi }\stackrel{~}{\nu }`$ boxes contribute with left-handed terms only:
$`A_{H^0A^0}^{box10}(\stackrel{~}{\nu }_L\stackrel{~}{\chi }\stackrel{~}{\chi }\stackrel{~}{\nu }_L)`$ $`=`$ $`{\displaystyle \frac{e^2}{16\pi ^2s_W^2}}{\displaystyle \underset{ij}{}}Z_{1i}^+Z_{1j}^+\times g_{H^0\stackrel{~}{\nu }_L\stackrel{~}{\nu }_L}\times `$ (253)
$`\left[M_{\stackrel{~}{\chi }_i}c_{A^0ji}^R๐_{10}(\stackrel{~}{\nu }_L\stackrel{~}{\chi }_i\stackrel{~}{\chi }_j\stackrel{~}{\nu }_L)+M_{\stackrel{~}{\chi }_j}c_{A^0ji}^L๐_{10}^{}(\stackrel{~}{\nu }_L\stackrel{~}{\chi }_i\stackrel{~}{\chi }_j\stackrel{~}{\nu }_L)\right],`$
$`A_{h^0A^0}^{box10}(\stackrel{~}{\nu }_L\stackrel{~}{\chi }\stackrel{~}{\chi }\stackrel{~}{\nu }_L)`$ $`=`$ $`{\displaystyle \frac{e^2}{16\pi ^2s_W^2}}{\displaystyle \underset{ij}{}}Z_{1i}^+Z_{1j}^+\times g_{h^0\stackrel{~}{\nu }_L\stackrel{~}{\nu }_L}\times `$ (254)
$`\left[M_{\stackrel{~}{\chi }_i}c_{A^0ji}^R๐_{10}(\stackrel{~}{\nu }_L\stackrel{~}{\chi }_i\stackrel{~}{\chi }_j\stackrel{~}{\nu }_L)+M_{\stackrel{~}{\chi }_j}c_{A^0ji}^L๐_{10}^{}(\stackrel{~}{\nu }_L\stackrel{~}{\chi }_i\stackrel{~}{\chi }_j\stackrel{~}{\nu }_L)\right].`$
The $`\stackrel{~}{e}\stackrel{~}{\chi }^0\stackrel{~}{\chi }^0\stackrel{~}{e}`$ boxes contribute with both left-handed and right-handed terms. Writing all these terms into one single expression leads to:
$`A_{H^0A^0}^{box10}(\stackrel{~}{e}\stackrel{~}{\chi }^0\stackrel{~}{\chi }^0\stackrel{~}{e})`$ $`={\displaystyle \frac{e^2}{32\pi ^2s_W^2c_W^2}}P_L`$ $`{\displaystyle \underset{ij}{}}(Z_{1i}^Ns_W+Z_{2i}^Nc_W)(Z_{1j}^Ns_W+Z_{2j}^Nc_W)\times g_{H^0\stackrel{~}{e}_L\stackrel{~}{e}_L}\times `$ (255)
$`\left[M_{\stackrel{~}{\chi }_i^0}n_{A^0ji}^R๐_{10}(\stackrel{~}{e}_L\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j^0\stackrel{~}{e}_L)+M_{\stackrel{~}{\chi }_j^0}n_{A^0ji}^L๐_{10}^{}(\stackrel{~}{e}_L\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j^0\stackrel{~}{e}_L)\right]`$
$`+{\displaystyle \frac{e^2}{8\pi ^2c_W^2}}P_R`$ $`{\displaystyle \underset{ij}{}}Z_{1i}^NZ_{1j}^N\times g_{H^0\stackrel{~}{e}_R\stackrel{~}{e}_R}\times `$
$`\left[M_{\stackrel{~}{\chi }_i^0}n_{A^0ji}^L๐_{10}(\stackrel{~}{e}_R\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j^0\stackrel{~}{e}_R)+M_{\stackrel{~}{\chi }_j^0}n_{A^0ji}^R๐_{10}^{}(\stackrel{~}{e}_R\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j^0\stackrel{~}{e}_R)\right],`$
$`A_{h^0A^0}^{box10}(\stackrel{~}{e}\stackrel{~}{\chi }^0\stackrel{~}{\chi }^0\stackrel{~}{e})`$ $`={\displaystyle \frac{e^2}{32\pi ^2s_W^2c_W^2}}P_L`$ $`{\displaystyle \underset{ij}{}}(Z_{1i}^Ns_W+Z_{2i}^Nc_W)(Z_{1j}^Ns_W+Z_{2j}^Nc_W)\times g_{h^0\stackrel{~}{e}_L\stackrel{~}{e}_L}\times `$ (256)
$`\left[M_{\stackrel{~}{\chi }_i^0}n_{A^0ji}^R๐_{10}(\stackrel{~}{e}_L\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j^0\stackrel{~}{e}_L)+M_{\stackrel{~}{\chi }_j^0}n_{A^0ji}^L๐_{10}^{}(\stackrel{~}{e}_L\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j^0\stackrel{~}{e}_L)\right]`$
$`+{\displaystyle \frac{e^2}{8\pi ^2c_W^2}}P_R`$ $`{\displaystyle \underset{ij}{}}Z_{1i}^NZ_{1j}^N\times g_{h^0\stackrel{~}{e}_R\stackrel{~}{e}_R}\times `$
$`\left[M_{\stackrel{~}{\chi }_i^0}n_{A^0ji}^L๐_{10}(\stackrel{~}{e}_R\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j^0\stackrel{~}{e}_R)+M_{\stackrel{~}{\chi }_j^0}n_{A^0ji}^R๐_{10}^{}(\stackrel{~}{e}_R\stackrel{~}{\chi }_i^0\stackrel{~}{\chi }_j^0\stackrel{~}{e}_R)\right].`$
## 7 Conclusion and outlooks
In this paper, we discussed all electroweak one loop contributions to the pair production cross section for charged and neutral Higgs bosons in $`e^+e^{}`$ collisions, in the theoretical framework of the MSSM. The one loop amplitudes of initial vertices and $`e^\pm `$ self-energy, of $`\gamma `$ and $`Z`$ boson self-energies, of the corresponding counter terms, of final vertices and Higgs self-energies and of box diagrams are respectively given by equations (17), (82), (83), (98) and (238) for the charged Higgs sector, and by equations (18), (85), (86), (126) and (246) for the neutral Higgs sector. The left-handed and right-handed amplitudes of all these electroweak one loop contributions are:
* in the charged Higgs sector:
$`a_{L,R}^{1loop}(H^+H^{})`$ $`=`$ $`a_{L,R}^{in}(H^+H^{})`$ (257)
$`+`$ $`a_{L,R}^{RG}(H^+H^{})+a_{L,R}^{ct}(H^+H^{})`$
$`+`$ $`a_{L,R}^{fin}(H^+H^{})`$
$`+`$ $`a_{L,R}^{box}(H^+H^{}),`$
* in the neutral Higgs sector:
$`a_{L,R}^{1loop}(H^0A^0/h^0A^0)`$ $`=`$ $`a_{L,R}^{in}(H^0A^0/h^0A^0)`$ (258)
$`+`$ $`a_{L,R}^{RG}(H^0A^0/h^0A^0)+a_{L,R}^{ct}(H^0A^0/h^0A^0)`$
$`+`$ $`a_{L,R}^{fin}(H^0A^0/h^0A^0)`$
$`+`$ $`a_{L,R}^{box}(H^0A^0/h^0A^0).`$
The differential production cross section for charged or neutral Higgs bosons at the one loop level can then be calculated as follows:
$$\frac{d\sigma }{d\mathrm{cos}\theta }=\frac{\pi \alpha _{em}^2\beta _H^3}{8q^2}(1\mathrm{cos}^2\theta )\times \left[|a_L^{Born}|^2+2|a_L^{Born}a_L^{1loop}|+|a_R^{Born}|^2+2|a_R^{Born}a_R^{1loop}|\right].$$
(259)
In the previous equation, $`a_{L,R}^{Born}`$ is the Born amplitude of equation (8) or (9). As for $`\beta _H`$, it stands for the velocity of the Higgs bosons, see equation (12). After integration over $`\mathrm{cos}\theta `$ (which appears in the contributions of the box diagrams), one obtains the total pair production cross section at the one loop level. Note that, in the case of the tree level cross sections for $`H^+H^{}`$ and $`H^0A^0+h^0A^0`$, there is a direct dependence on $`M_A`$ only and not on the other MSSM parameters. However, after having taken into account all electroweak one loop contributions, this is not true anymore. Indeed, $`a_{L,R}^{1loop}`$ depends on other MSSM parameters than just $`M_A`$.
A C++ numerical code has been developed in order to calculate accurately all one loop electroweak contributions and, in turn, to compare the pair production cross sections for MSSM charged and neutral Higgs bosons at tree level and at the one loop level. The relevant Feynman diagrams are computed by calling the suitable functions in the LoopTools 2.1 library . The input of the code, in standard notation, is the following: $`\mathrm{tan}\beta `$, $`\mu `$, $`M_A`$ (the mass of the $`A^0`$ Higgs boson), the gaugino parameters $`M_1`$ and $`M_2`$, the scalar mass scale $`M_S`$, the sfermion mixing matrix parameters $`A_u`$ and $`A_d`$. A possible reference for these parameters is . This input requires a preliminary pre-processing using the FeynHiggs 2.1 code. A subset of these parameters is then fed into FeynHiggs in order to compute the masses of the Higgs bosons $`h^0`$, $`H^0`$, $`H^\pm `$, as well as the mixing angle in the neutral sector $`\alpha `$. These additional parameters do indeed appear in the analytical expressions described in the text. The output of the code is the cross section for the various processes under consideration. We have successfully checked that the MSSM Higgs bosons pair production cross section computed by our code at the one loop level remains stable against UV divergences, both in the charged and neutral Higgs sectors. Also, we have checked that the variation of the computed one loop cross section with $`q^2`$ agrees with our expectations.
Note that, in this code, we have included all virtual contributions involving particles having electroweak interactions in the MSSM, but we did not treat pure QED effects, such as Initial State Radiation (ISR) and Final State Radiation (FSR). The reason is that these effects may depend on the characteristics of the detectors (for instance, they need specific kinematical cuts) and some specific codes exist in order to treat them. Nevertheless, in order to be able to test electroweak symmetry properties at high energy, in particular those of supersymmetric nature, which is indeed the purpose of this work, we have included the virtual photon effects, but with a photon mass set to $`M_Z`$ in order to keep these effects finite. In order to have the complete (observable) contribution including QED effects, one should compute the following combination: Our contribution + ISR + FSR + virtual soft photon with zero mass $``$ virtual soft photon with $`M_Z`$. This combination should be calculated at the level of the codes that include the ISR and FSR effects.
## Appendix A: Vertices and couplings
Gauge - Fermion - Fermion
Let $`Q_f`$ and $`T_f^3`$ be respectively the charge and the third isospin component of the fermion $`f`$, then the couplings of gauge bosons to left-handed and right-handed fermions are:
$`P_Lg_{\gamma ff}=Q_f`$ and $`P_Rg_{\gamma ff}=Q_f,`$
$`P_Lg_{Zff}={\displaystyle \frac{T_f^3Q_fs_W^2}{s_Wc_W}}`$ and $`P_Rg_{Zff}=Q_f{\displaystyle \frac{s_W}{c_W}},`$
$`P_Lg_{Wff^{}}={\displaystyle \frac{1}{s_W\sqrt{2}}}`$ and $`P_Rg_{Wff^{}}=0.`$
Note that, in this paper, we have also used a simplified notation for the couplings of $`\gamma `$ or $`Z`$ to fermion pairs:
$`g_{VLf}`$ $``$ $`P_Lg_{Vff},`$
$`g_{VRf}`$ $``$ $`P_Rg_{Vff},`$
where $`V`$ stands for either $`\gamma `$ or $`Z`$.
Gauge - Gaugino - Gaugino
$`๐ช_{ij}^{\gamma L}=e\delta _{ij}`$ and $`๐ช_{ij}^{\gamma R}=e\delta _{ij},`$
$`๐ช_{ij}^{ZL}={\displaystyle \frac{e\left[Z_{1i}^+Z_{1j}^++\delta _{ij}(c_W^2s_W^2)\right]}{2s_Wc_W}}`$ and $`๐ช_{ij}^{ZR}={\displaystyle \frac{e\left[Z_{1i}^{}Z_{1j}^{}+\delta _{ij}(c_W^2s_W^2)\right]}{2s_Wc_W}},`$
$`๐ช_{ij}^{0L}={\displaystyle \frac{e(Z_{4i}^NZ_{4j}^NZ_{3i}^NZ_{3j}^N)}{2s_Wc_W}}`$ and $`๐ช_{ij}^{0R}={\displaystyle \frac{e(Z_{4i}^NZ_{4j}^NZ_{3i}^NZ_{3j}^N)}{2s_Wc_W}},`$
$`๐ช_{ij}^{WL}={\displaystyle \frac{e}{s_W}}\left(Z_{2j}^NZ_{1i}^+{\displaystyle \frac{1}{\sqrt{2}}}Z_{4k}^NZ_{2i}^+\right)`$ and $`๐ช_{ij}^{WR}={\displaystyle \frac{e}{s_W}}\left(Z_{2j}^NZ_{1i}^{}+{\displaystyle \frac{1}{\sqrt{2}}}Z_{3j}^NZ_{2i}^{}\right).`$
Here, the $`Z_{ij}`$ terms correspond to the various elements of the unitary mixing matrices of the charginos and neutralinos. They are derived from the diagonalization of the MSSM gaugino mass matrix, see for instance reference for details.
Gauge - Sfermion - Sfermion
The couplings of gauge bosons to unmixed left-handed and right-handed sfermions are:
$`g_{\gamma \stackrel{~}{f}_L\stackrel{~}{f}_L}^0=eQ_f`$ and $`g_{\gamma \stackrel{~}{f}_R\stackrel{~}{f}_R}^0=eQ_f,`$
$`g_{Z\stackrel{~}{f}_L\stackrel{~}{f}_L}^0={\displaystyle \frac{e(T_f^3Q_fs_W^2)}{s_Wc_W}}`$ and $`g_{Z\stackrel{~}{f}_R\stackrel{~}{f}_R}^0=eQ_f{\displaystyle \frac{s_W}{c_W}},`$
$`g_{W\stackrel{~}{f}_L\stackrel{~}{f}_L^{}}^0={\displaystyle \frac{e}{s_W\sqrt{2}}}`$ and $`g_{W\stackrel{~}{f}_R\stackrel{~}{f}_R^{}}^0=0.`$
Let $`\theta _{\stackrel{~}{f}}`$ be the mixing angle of the sfermion $`\stackrel{~}{f}`$ (generally a third generation squark). Let us also define $`c_{\stackrel{~}{f}}\mathrm{cos}\theta _{\stackrel{~}{f}}`$ and $`s_{\stackrel{~}{f}}\mathrm{sin}\theta _{\stackrel{~}{f}}`$. The coupling between a gauge boson and two sfermions with mixing is then given by:
$`g_{\gamma \stackrel{~}{f}_1\stackrel{~}{f}_1}=g_{\gamma \stackrel{~}{f}_2\stackrel{~}{f}_2}=g_{\gamma \stackrel{~}{f}_L\stackrel{~}{f}_L}^0=g_{\gamma \stackrel{~}{f}_R\stackrel{~}{f}_R}^0=eQ_f,`$
$`g_{Z\stackrel{~}{f}_1\stackrel{~}{f}_1}`$ $`=`$ $`c_{\stackrel{~}{f}}^2g_{Z\stackrel{~}{f}_L\stackrel{~}{f}_L}^0+s_{\stackrel{~}{f}}^2g_{Z\stackrel{~}{f}_R\stackrel{~}{f}_R}^0,`$
$`g_{Z\stackrel{~}{f}_2\stackrel{~}{f}_2}`$ $`=`$ $`s_{\stackrel{~}{f}}^2g_{Z\stackrel{~}{f}_L\stackrel{~}{f}_L}^0+c_{\stackrel{~}{f}}^2g_{Z\stackrel{~}{f}_R\stackrel{~}{f}_R}^0,`$
$`g_{Z\stackrel{~}{f}_1\stackrel{~}{f}_2}`$ $`=`$ $`g_{Z\stackrel{~}{f}_2\stackrel{~}{f}_1}=c_{\stackrel{~}{f}}s_{\stackrel{~}{f}}(g_{Z\stackrel{~}{f}_R\stackrel{~}{f}_R}^0g_{Z\stackrel{~}{f}_L\stackrel{~}{f}_L}^0),`$
$`g_{W\stackrel{~}{f}_1\stackrel{~}{f}_1^{}}`$ $`=`$ $`c_{\stackrel{~}{f}}c_{\stackrel{~}{f}^{}}g_{W\stackrel{~}{f}_L\stackrel{~}{f}_L^{}}^0,`$
$`g_{W\stackrel{~}{f}_2\stackrel{~}{f}_2^{}}`$ $`=`$ $`s_{\stackrel{~}{f}}s_{\stackrel{~}{f}^{}}g_{W\stackrel{~}{f}_L\stackrel{~}{f}_L^{}}^0,`$
$`g_{W\stackrel{~}{f}_1\stackrel{~}{f}_2^{}}`$ $`=`$ $`c_{\stackrel{~}{f}}s_{\stackrel{~}{f}^{}}g_{W\stackrel{~}{f}_L\stackrel{~}{f}_L^{}}^0,`$
$`g_{W\stackrel{~}{f}_2\stackrel{~}{f}_1^{}}`$ $`=`$ $`c_{\stackrel{~}{f}^{}}s_{\stackrel{~}{f}}g_{W\stackrel{~}{f}_L\stackrel{~}{f}_L^{}}^0.`$
Gauge - Gauge - Higgs
$`g_{ZZH^0}={\displaystyle \frac{eM_W}{s_Wc_W^2}}\mathrm{cos}(\beta \alpha )`$ and $`g_{ZZh^0}={\displaystyle \frac{eM_W}{s_Wc_W^2}}\mathrm{sin}(\beta \alpha ),`$
$`g_{WWH^0}={\displaystyle \frac{eM_W}{s_W}}\mathrm{cos}(\beta \alpha )`$ and $`g_{WWh^0}={\displaystyle \frac{eM_W}{s_W}}\mathrm{sin}(\beta \alpha ),`$
$`g_{\gamma WG}=eM_W`$ and $`g_{ZWG}=eM_W{\displaystyle \frac{s_W}{c_W}}.`$
Gauge - Higgs - Higgs
$`g_{ZH^0A^0}={\displaystyle \frac{e}{2s_Wc_W}}\mathrm{sin}(\beta \alpha )`$ and $`g_{Zh^0A^0}=+{\displaystyle \frac{e}{2s_Wc_W}}\mathrm{cos}(\beta \alpha ),`$
$`g_{ZH^0G^0}=+{\displaystyle \frac{e}{2s_Wc_W}}\mathrm{cos}(\beta \alpha )`$ and $`g_{Zh^0G^0}=+{\displaystyle \frac{e}{2s_Wc_W}}\mathrm{sin}(\beta \alpha ),`$
$`g_{W^\pm H^\pm H^0}=+Q_W\times {\displaystyle \frac{e}{2s_W}}\mathrm{sin}(\beta \alpha )`$ and $`g_{W^\pm H^\pm h^0}=Q_W\times {\displaystyle \frac{e}{2s_W}}\mathrm{cos}(\beta \alpha ),`$
$`g_{W^\pm G^\pm H^0}=Q_W\times {\displaystyle \frac{e}{2s_W}}\mathrm{cos}(\beta \alpha )`$ and $`g_{W^\pm G^\pm h^0}=Q_W\times {\displaystyle \frac{e}{2s_W}}\mathrm{sin}(\beta \alpha ),`$
$`g_{WHA^0}=+{\displaystyle \frac{e}{2s_W}}`$ and $`g_{WHG^0}=0,`$
$`g_{WGA^0}=0`$ and $`g_{WGG^0}=+{\displaystyle \frac{e}{2s_W}},`$
$`g_{\gamma HH}=e`$ and $`g_{ZHH}=e{\displaystyle \frac{12s_W^2}{2s_Wc_W}},`$
$`g_{\gamma GG}=e`$ and $`g_{ZGG}=e{\displaystyle \frac{12s_W^2}{2s_Wc_W}}.`$
Higgs - Fermion - Fermion
The coupling constant between a Higgs boson and a fermion pair is proportional to the mass of the fermion(s). Thus, only the third generation quarks are usually considered.
In the charged Higgs sector:
* for $`btH^{}`$: $`c_{btH^{}}^L={\displaystyle \frac{e}{\sqrt{2}s_WM_W}}M_t\text{cot}\beta `$ and $`c_{btH^{}}^R={\displaystyle \frac{e}{\sqrt{2}s_WM_W}}M_b\mathrm{tan}\beta `$,
* for $`tbH^+`$: $`c_{tbH^+}^L={\displaystyle \frac{e}{\sqrt{2}s_WM_W}}M_b\mathrm{tan}\beta `$ and $`c_{tbH^+}^R={\displaystyle \frac{e}{\sqrt{2}s_WM_W}}M_t\text{cot}\beta `$.
In the neutral Higgs sector, the left-handed coupling constants are:
* $`c_{H^0t}^L={\displaystyle \frac{eM_t}{2s_WM_W}}\times {\displaystyle \frac{\mathrm{sin}\alpha }{\mathrm{sin}\beta }}`$ and $`c_{H^0b}^L={\displaystyle \frac{eM_b}{2s_WM_W}}\times {\displaystyle \frac{\mathrm{cos}\alpha }{\mathrm{cos}\beta }}`$,
* $`c_{h^0t}^L={\displaystyle \frac{eM_t}{2s_WM_W}}\times {\displaystyle \frac{\mathrm{cos}\alpha }{\mathrm{sin}\beta }}`$ and $`c_{H^0b}^L=+{\displaystyle \frac{eM_b}{2s_WM_W}}\times {\displaystyle \frac{\mathrm{sin}\alpha }{\mathrm{cos}\beta }}`$,
* $`c_{A^0t}^L={\displaystyle \frac{eM_t}{2s_WM_W}}\times \text{cot}\beta `$ and $`c_{A^0b}^L={\displaystyle \frac{eM_b}{2s_WM_W}}\times \mathrm{tan}\beta `$.
As for the right-handed couplings constants, one simply has:
* $`c_{H^0f}^R=c_{H^0f}^L`$,
* $`c_{h^0f}^R=c_{h^0f}^L`$,
* $`c_{A^0f}^R=c_{A^0f}^L`$.
Higgs - Gaugino - Gaugino
In the charged Higgs sector, there are two types of vertex to consider: $`\stackrel{~}{\chi }_j^0\stackrel{~}{\chi }_i^+H^{}`$ and $`\stackrel{~}{\chi }_j^+\stackrel{~}{\chi }_i^0H^+`$. Let us choose the case where $`i`$ and $`j`$ label respectively a chargino and a neutralino, then the corresponding left-handed and right-handed coupling constants are:
* $`c_{Hij}^L={\displaystyle \frac{e\mathrm{sin}\beta }{s_Wc_W}}\left[{\displaystyle \frac{1}{\sqrt{2}}}Z_{2i}^{}(Z_{1j}^Ns_W+Z_{2j}^Nc_W)Z_{1i}^{}Z_{3j}^Nc_W\right]`$,
* $`c_{Hij}^R={\displaystyle \frac{e\mathrm{cos}\beta }{s_Wc_W}}\left[{\displaystyle \frac{1}{\sqrt{2}}}Z_{2i}^+(Z_{1j}^Ns_W+Z_{2j}^Nc_W)+Z_{1i}^+Z_{4j}^Nc_W\right]`$.
For the other vertex (where $`i`$ and $`j`$ label respectively a neutralino and a chargino), one should instead use the left-handed and right-handed coupling constants $`c_{Hji}^R`$ and $`c_{Hji}^L`$, respectively.
In the neutral Higgs sector, one must consider the coupling between a neutral Higgs boson and either two charginos or two neutralinos.
For the coupling between a neutral Higgs boson and two charginos:
$`c_{H^0ij}^L={\displaystyle \frac{e}{\sqrt{2}s_W}}\left[\mathrm{cos}\alpha Z_{2i}^{}Z_{1j}^++\mathrm{sin}\alpha Z_{1i}^{}Z_{2j}^+\right]`$ and $`c_{H^0ij}^R=c_{H^0ji}^L,`$
$`c_{h^0ij}^L={\displaystyle \frac{e}{\sqrt{2}s_W}}\left[\mathrm{sin}\alpha Z_{2i}^{}Z_{1j}^++\mathrm{cos}\alpha Z_{1i}^{}Z_{2j}^+\right]`$ and $`c_{h^0ij}^R=c_{h^0ji}^L,`$
$`c_{A^0ij}^L={\displaystyle \frac{e}{\sqrt{2}s_W}}\left[\mathrm{sin}\beta Z_{2i}^{}Z_{1j}^++\mathrm{cos}\beta Z_{1i}^{}Z_{2j}^+\right]`$ and $`c_{A^0ij}^R=c_{A^0ji}^L.`$
Let us now consider the coupling between a neutral Higgs boson and two neutralinos. For the left-handed components, one has:
$`n_{H^0ij}^L`$ $`={\displaystyle \frac{e}{2s_Wc_W}}\times `$ $`\{(\mathrm{cos}\alpha Z_{3j}^N\mathrm{sin}\alpha Z_{4j}^N)(Z_{1i}^Ns_WZ_{2i}^Nc_W)`$
$`+(\mathrm{cos}\alpha Z_{3i}^N\mathrm{sin}\alpha Z_{4i}^N)(Z_{1j}^Ns_WZ_{2j}^Nc_W)\},`$
$`n_{h^0ij}^L`$ $`={\displaystyle \frac{e}{2s_Wc_W}}\times `$ $`\{(\mathrm{sin}\alpha Z_{3j}^N+\mathrm{cos}\alpha Z_{4j}^N)(Z_{1i}^Ns_WZ_{2i}^Nc_W)`$
$`+(\mathrm{sin}\alpha Z_{3i}^N+\mathrm{cos}\alpha Z_{4i}^N)(Z_{1j}^Ns_WZ_{2j}^Nc_W)\},`$
$`n_{A^0ij}^L`$ $`={\displaystyle \frac{e}{2s_Wc_W}}\times `$ $`\{(\mathrm{sin}\beta Z_{3j}^N\mathrm{cos}\beta Z_{4j}^N)(Z_{1i}^Ns_WZ_{2i}^Nc_W)`$
$`+(\mathrm{sin}\beta Z_{3i}^N\mathrm{cos}\beta Z_{4i}^N)(Z_{1j}^Ns_WZ_{2j}^Nc_W)\}.`$
As for the right-handed components, one simply has:
$`n_{H^0ij}^R`$ $`=`$ $`n_{H^0ji}^L,`$
$`n_{h^0ij}^R`$ $`=`$ $`n_{h^0ji}^L,`$
$`n_{A^0ij}^R`$ $`=`$ $`n_{A^0ji}^L.`$
Higgs - Sfermion - Sfermion
Let us first consider the light unmixed sfermions. Their coupling constant to the charged and neutral Higgs bosons is not proportional to the mass of the corresponding fermion(s) and it can thus not be neglected.
In the charged Higgs sector, if $`\stackrel{~}{f}`$ and $`\stackrel{~}{f}^{}`$ represent respectively up-squarks and down-squarks of the first and second generations, or sneutrinos and charged sleptons, one has:
$`g_{H\stackrel{~}{f}_L\stackrel{~}{f}_L^{}}={\displaystyle \frac{eM_W}{s_W\sqrt{2}}}\mathrm{sin}2\beta `$ and $`g_{H\stackrel{~}{f}_R\stackrel{~}{f}_R^{}}=0.`$
In the neutral Higgs sector, one has:
$`g_{H^0\stackrel{~}{f}_L\stackrel{~}{f}_L}={\displaystyle \frac{eM_W}{s_Wc_W^2}}(T_f^3Q_fs_W^2)\mathrm{cos}(\alpha +\beta )`$ and $`g_{H^0\stackrel{~}{f}_R\stackrel{~}{f}_R}={\displaystyle \frac{eM_W}{s_Wc_W^2}}Q_fs_W^2\mathrm{cos}(\alpha +\beta ),`$
$`g_{h^0\stackrel{~}{f}_L\stackrel{~}{f}_L}={\displaystyle \frac{eM_W}{s_Wc_W^2}}(T_f^3Q_fs_W^2)\mathrm{sin}(\alpha +\beta )`$ and $`g_{h^0\stackrel{~}{f}_R\stackrel{~}{f}_R}={\displaystyle \frac{eM_W}{s_Wc_W^2}}Q_fs_W^2\mathrm{sin}(\alpha +\beta ).`$
Note that the couplings between $`A^0`$ and a pair of light unmixed sfermions are proportional to the fermion mass and are thus negligible.
Let us now consider the heavy sfermions, i.e. the third generation squarks, and, in a first step, let us assume that there is no mixing. The coupling constants between the MSSM Higgs boson and a pair of unmixed heavy sfermions are given below.
For the charged Higgs bosons $`H`$, one has:
$`g_{H\stackrel{~}{t}_L\stackrel{~}{b}_L}^0`$ $`=`$ $`{\displaystyle \frac{eM_W}{s_W\sqrt{2}}}\left[\mathrm{sin}2\beta {\displaystyle \frac{M_b^2\mathrm{tan}\beta +M_t^2\text{cot}\beta }{M_W^2}}\right],`$
$`g_{H\stackrel{~}{t}_R\stackrel{~}{b}_R}^0`$ $`=`$ $`{\displaystyle \frac{eM_tM_b}{s_WM_W\sqrt{2}}}\left[\mathrm{tan}\beta +\text{cot}\beta \right],`$
$`g_{H\stackrel{~}{t}_L\stackrel{~}{b}_R}^0`$ $`=`$ $`{\displaystyle \frac{eM_b}{s_WM_W\sqrt{2}}}\left[\mu A_b\mathrm{tan}\beta \right].`$
$`g_{H\stackrel{~}{t}_R\stackrel{~}{b}_L}^0`$ $`=`$ $`{\displaystyle \frac{eM_t}{s_WM_W\sqrt{2}}}\left[\mu A_t\text{cot}\beta \right].`$
For the neutral Higgs boson $`A^0`$, there are only off-diagonal terms:
$`g_{A^0\stackrel{~}{t}_L\stackrel{~}{t}_L}^0=g_{A^0\stackrel{~}{t}_R\stackrel{~}{t}_R}^0=0\text{and}g_{A^0\stackrel{~}{b}_L\stackrel{~}{b}_L}^0=g_{A^0\stackrel{~}{b}_R\stackrel{~}{b}_R}^0=0,`$
$`g_{A^0\stackrel{~}{t}_L\stackrel{~}{t}_R}^0=g_{A^0\stackrel{~}{t}_R\stackrel{~}{t}_L}^0={\displaystyle \frac{eM_t}{2s_WM_W}}\left[\mu A_t\text{cot}\beta \right],`$
$`g_{A^0\stackrel{~}{b}_L\stackrel{~}{b}_R}^0=g_{A^0\stackrel{~}{b}_R\stackrel{~}{b}_L}^0={\displaystyle \frac{eM_b}{2s_WM_W}}\left[\mu A_b\mathrm{tan}\beta \right].`$
For the neutral Higgs boson $`H^0`$, one has:
$`g_{H^0\stackrel{~}{t}_L\stackrel{~}{t}_L}^0`$ $`=`$ $`{\displaystyle \frac{eM_W}{s_Wc_W^2}}\left({\displaystyle \frac{1}{2}}{\displaystyle \frac{2}{3}}s_W^2\right)\mathrm{cos}(\alpha +\beta ){\displaystyle \frac{eM_t^2}{s_WM_W}}{\displaystyle \frac{\mathrm{sin}\alpha }{\mathrm{sin}\beta }},`$
$`g_{H^0\stackrel{~}{t}_R\stackrel{~}{t}_R}^0`$ $`=`$ $`{\displaystyle \frac{eM_W}{s_Wc_W^2}}\left({\displaystyle \frac{2}{3}}s_W^2\right)\mathrm{cos}(\alpha +\beta ){\displaystyle \frac{eM_t^2}{s_WM_W}}{\displaystyle \frac{\mathrm{sin}\alpha }{\mathrm{sin}\beta }},`$
$`g_{H^0\stackrel{~}{t}_L\stackrel{~}{t}_R}^0`$ $`=`$ $`g_{H^0\stackrel{~}{t}_R\stackrel{~}{t}_L}^0={\displaystyle \frac{eM_t}{2s_WM_W}}\left({\displaystyle \frac{\mu \mathrm{cos}\alpha +A_t\mathrm{sin}\alpha }{\mathrm{sin}\beta }}\right).`$
$`g_{H^0\stackrel{~}{b}_L\stackrel{~}{b}_L}^0`$ $`=`$ $`{\displaystyle \frac{eM_W}{s_Wc_W^2}}\left({\displaystyle \frac{1}{2}}{\displaystyle \frac{1}{3}}s_W^2\right)\mathrm{cos}(\alpha +\beta ){\displaystyle \frac{eM_b^2}{s_WM_W}}{\displaystyle \frac{\mathrm{cos}\alpha }{\mathrm{cos}\beta }},`$
$`g_{H^0\stackrel{~}{b}_R\stackrel{~}{b}_R}^0`$ $`=`$ $`{\displaystyle \frac{eM_W}{s_Wc_W^2}}\left({\displaystyle \frac{1}{3}}s_W^2\right)\mathrm{cos}(\alpha +\beta ){\displaystyle \frac{eM_b^2}{s_WM_W}}{\displaystyle \frac{\mathrm{cos}\alpha }{\mathrm{cos}\beta }},`$
$`g_{H^0\stackrel{~}{b}_L\stackrel{~}{b}_R}^0`$ $`=`$ $`g_{H^0\stackrel{~}{b}_R\stackrel{~}{b}_L}^0={\displaystyle \frac{eM_b}{2s_WM_W}}\left({\displaystyle \frac{\mu \mathrm{sin}\alpha +A_b\mathrm{cos}\alpha }{\mathrm{cos}\beta }}\right).`$
For the neutral Higgs boson $`h^0`$, one has:
$`g_{h^0\stackrel{~}{t}_L\stackrel{~}{t}_L}^0`$ $`=`$ $`{\displaystyle \frac{eM_W}{s_Wc_W^2}}\left({\displaystyle \frac{1}{2}}{\displaystyle \frac{2}{3}}s_W^2\right)\mathrm{sin}(\alpha +\beta ){\displaystyle \frac{eM_t^2}{s_WM_W}}{\displaystyle \frac{\mathrm{cos}\alpha }{\mathrm{sin}\beta }},`$
$`g_{h^0\stackrel{~}{t}_R\stackrel{~}{t}_R}^0`$ $`=`$ $`{\displaystyle \frac{eM_W}{s_Wc_W^2}}\left({\displaystyle \frac{2}{3}}s_W^2\right)\mathrm{sin}(\alpha +\beta ){\displaystyle \frac{eM_t^2}{s_WM_W}}{\displaystyle \frac{\mathrm{cos}\alpha }{\mathrm{sin}\beta }},`$
$`g_{h^0\stackrel{~}{t}_L\stackrel{~}{t}_R}^0`$ $`=`$ $`g_{h^0\stackrel{~}{t}_R\stackrel{~}{t}_L}^0={\displaystyle \frac{eM_t}{2s_WM_W}}\left({\displaystyle \frac{\mu \mathrm{sin}\alpha A_t\mathrm{cos}\alpha }{\mathrm{sin}\beta }}\right).`$
$`g_{h^0\stackrel{~}{b}_L\stackrel{~}{b}_L}^0`$ $`=`$ $`{\displaystyle \frac{eM_W}{s_Wc_W^2}}\left({\displaystyle \frac{1}{2}}{\displaystyle \frac{1}{3}}s_W^2\right)\mathrm{sin}(\alpha +\beta )+{\displaystyle \frac{eM_b^2}{s_WM_W}}{\displaystyle \frac{\mathrm{sin}\alpha }{\mathrm{cos}\beta }},`$
$`g_{h^0\stackrel{~}{b}_R\stackrel{~}{b}_R}^0`$ $`=`$ $`{\displaystyle \frac{eM_W}{s_Wc_W^2}}\left({\displaystyle \frac{1}{3}}s_W^2\right)\mathrm{sin}(\alpha +\beta )+{\displaystyle \frac{eM_b^2}{s_WM_W}}{\displaystyle \frac{\mathrm{sin}\alpha }{\mathrm{cos}\beta }},`$
$`g_{h^0\stackrel{~}{b}_L\stackrel{~}{b}_R}^0`$ $`=`$ $`g_{h^0\stackrel{~}{b}_R\stackrel{~}{b}_L}^0={\displaystyle \frac{eM_b}{2s_WM_W}}\left({\displaystyle \frac{\mu \mathrm{cos}\alpha A_b\mathrm{sin}\alpha }{\mathrm{cos}\beta }}\right).`$
Let us know take the sfermion mixing into account and let $`R^{\stackrel{~}{t}}`$ and $`R^{\stackrel{~}{b}}`$ be the rotation matrices for $`\stackrel{~}{t}`$ and $`\stackrel{~}{b}`$ squarks, respectively. If $`\stackrel{~}{f}_{1,2}^0\stackrel{~}{f}_{L,R}`$, then:
$`\stackrel{~}{f}_i`$ $`=`$ $`R_{ij}^{\stackrel{~}{f}}\stackrel{~}{f}_j^0\text{with}R_{ij}^{\stackrel{~}{f}}=\left(\begin{array}{cc}c_{\stackrel{~}{f}}& s_{\stackrel{~}{f}}\\ s_{\stackrel{~}{f}}& c_{\stackrel{~}{f}}\end{array}\right).`$ (262)
In the charged Higgs sector, one has:
$`g_{H\stackrel{~}{t}_i\stackrel{~}{b}_j}`$ $`=`$ $`{\displaystyle \underset{i^{}j^{}}{}}R_{ii^{}}^{\stackrel{~}{t}}R_{jj^{}}^{\stackrel{~}{b}}g_{H\stackrel{~}{t}_i^{}\stackrel{~}{b}_j^{}}^0.`$
Similarly, in the neutral Higgs sector, for $`\stackrel{~}{f}`$ stands for either $`\stackrel{~}{t}`$ or $`\stackrel{~}{b}`$, then one obtains the following coupling constants:
$`g_{A^0\stackrel{~}{f}_i\stackrel{~}{f}_j}`$ $`=`$ $`{\displaystyle \underset{i^{}j^{}}{}}R_{ii^{}}^{\stackrel{~}{f}}R_{jj^{}}^{\stackrel{~}{f}}g_{A^0\stackrel{~}{f}_i^{}\stackrel{~}{f}_j^{}}^0`$
$`g_{H^0\stackrel{~}{f}_i\stackrel{~}{f}_j}`$ $`=`$ $`{\displaystyle \underset{i^{}j^{}}{}}R_{ii^{}}^{\stackrel{~}{f}}R_{jj^{}}^{\stackrel{~}{f}}g_{H^0\stackrel{~}{f}_i^{}\stackrel{~}{f}_j^{}}^0`$
$`g_{h^0\stackrel{~}{f}_i\stackrel{~}{f}_j}`$ $`=`$ $`{\displaystyle \underset{i^{}j^{}}{}}R_{ii^{}}^{\stackrel{~}{f}}R_{jj^{}}^{\stackrel{~}{f}}g_{h^0\stackrel{~}{f}_i^{}\stackrel{~}{f}_j^{}}^0.`$
Note that, in the case of the neutral boson $`A^0`$, the coupling to a pair of sfermions is the same with or without mixing.
Higgs - Higgs - Higgs
$`g_{H^0HH}={\displaystyle \frac{eM_W}{s_W}}\left[{\displaystyle \frac{\mathrm{cos}2\beta \mathrm{cos}(\beta +\alpha )}{2c_W^2}}\mathrm{cos}(\beta \alpha )\right],`$
$`g_{h^0HH}={\displaystyle \frac{eM_W}{s_W}}\left[{\displaystyle \frac{\mathrm{cos}2\beta \mathrm{sin}(\beta +\alpha )}{2c_W^2}}+\mathrm{sin}(\beta \alpha )\right],`$
$`g_{H^0GG}={\displaystyle \frac{eM_W}{2s_Wc_W^2}}\left[\mathrm{cos}2\beta \mathrm{cos}(\beta +\alpha )\right],`$
$`g_{h^0GG}={\displaystyle \frac{eM_W}{2s_Wc_W^2}}\left[\mathrm{cos}2\beta \mathrm{sin}(\beta +\alpha )\right],`$
$`g_{H^0GH}={\displaystyle \frac{eM_W}{2s_W}}\left[\mathrm{sin}(\beta \alpha ){\displaystyle \frac{\mathrm{sin}2\beta \mathrm{cos}(\alpha +\beta )}{c_W^2}}\right],`$
$`g_{h^0GH}={\displaystyle \frac{eM_W}{2s_W}}\left[\mathrm{cos}(\beta \alpha ){\displaystyle \frac{\mathrm{sin}2\beta \mathrm{sin}(\alpha +\beta )}{c_W^2}}\right],`$
$`g_{A^0G^\pm H^\pm }=Q_G\times {\displaystyle \frac{eM_W}{2s_W}},`$
$`g_{H^0H^0H^0}={\displaystyle \frac{3eM_W}{2s_Wc_W^2}}\mathrm{cos}2\alpha \mathrm{cos}(\beta +\alpha ),`$
$`g_{h^0h^0h^0}={\displaystyle \frac{3eM_W}{2s_Wc_W^2}}\mathrm{cos}2\alpha \mathrm{sin}(\beta +\alpha ),`$
$`g_{h^0H^0H^0}={\displaystyle \frac{eM_W}{2s_Wc_W^2}}\left[2\mathrm{sin}2\alpha \mathrm{cos}(\beta +\alpha )+\mathrm{cos}2\alpha \mathrm{sin}(\beta +\alpha )\right],`$
$`g_{H^0h^0h^0}={\displaystyle \frac{eM_W}{2s_Wc_W^2}}\left[2\mathrm{sin}2\alpha \mathrm{sin}(\beta +\alpha )\mathrm{cos}2\alpha \mathrm{cos}(\beta +\alpha )\right],`$
$`g_{H^0G^0G^0}={\displaystyle \frac{eM_W}{2s_Wc_W^2}}\mathrm{cos}2\beta \mathrm{cos}(\beta +\alpha ),`$
$`g_{h^0G^0G^0}={\displaystyle \frac{eM_W}{2s_Wc_W^2}}\mathrm{cos}2\beta \mathrm{sin}(\beta +\alpha ),`$
$`g_{H^0A^0A^0}={\displaystyle \frac{eM_W}{2s_Wc_W^2}}\mathrm{cos}2\beta \mathrm{cos}(\beta +\alpha ),`$
$`g_{h^0A^0A^0}={\displaystyle \frac{eM_W}{2s_Wc_W^2}}\mathrm{cos}2\beta \mathrm{sin}(\beta +\alpha ),`$
$`g_{H^0A^0G^0}={\displaystyle \frac{eM_W}{2s_Wc_W^2}}\mathrm{sin}2\beta \mathrm{cos}(\beta +\alpha ),`$
$`g_{h^0A^0G^0}={\displaystyle \frac{eM_W}{2s_Wc_W^2}}\mathrm{sin}2\beta \mathrm{sin}(\beta +\alpha ).`$
Higgs - Higgs - Sfermion - Sfermion
For light unmixed sfermions, the coupling constants are not proportional to the mass of the fermion(s) and can not be neglected. The coupling constants for the heavy sfermions are then obtained by adding a term proportional to the mass of the corresponding fermion(s).
If $`\stackrel{~}{f}`$ is a slepton ($`\stackrel{~}{\mathrm{}}`$ or $`\stackrel{~}{\nu }`$) or a squark from the first or second generation ($`\stackrel{~}{q}`$), one has:
$`g_{H^0H^0\stackrel{~}{f}_L\stackrel{~}{f}_L}`$ $`=`$ $`{\displaystyle \frac{e^2}{2s_W^2}}\left[{\displaystyle \frac{\mathrm{cos}2\alpha }{c_W^2}}\left(T_f^3Q_fs_W^2\right)\right],`$
$`g_{H^0H^0\stackrel{~}{f}_R\stackrel{~}{f}_R}`$ $`=`$ $`{\displaystyle \frac{e^2}{2s_W^2}}\left[{\displaystyle \frac{\mathrm{cos}2\alpha }{c_W^2}}\left(Q_fs_W^2\right)\right],`$
$`g_{h^0h^0\stackrel{~}{f}_L\stackrel{~}{f}_L}`$ $`=`$ $`{\displaystyle \frac{e^2}{2s_W^2}}\left[{\displaystyle \frac{\mathrm{cos}2\alpha }{c_W^2}}\left(T_f^3Q_fs_W^2\right)\right],`$
$`g_{h^0h^0\stackrel{~}{f}_R\stackrel{~}{f}_R}`$ $`=`$ $`{\displaystyle \frac{e^2}{2s_W^2}}\left[{\displaystyle \frac{\mathrm{cos}2\alpha }{c_W^2}}\left(Q_fs_W^2\right)\right],`$
$`g_{H^0h^0\stackrel{~}{f}_L\stackrel{~}{f}_L}`$ $`=`$ $`{\displaystyle \frac{e^2}{2s_W^2}}\left[{\displaystyle \frac{\mathrm{sin}2\alpha }{c_W^2}}\left(T_f^3Q_fs_W^2\right)\right],`$
$`g_{H^0h^0\stackrel{~}{f}_R\stackrel{~}{f}_R}`$ $`=`$ $`{\displaystyle \frac{e^2}{2s_W^2}}\left[{\displaystyle \frac{\mathrm{sin}2\alpha }{c_W^2}}\left(Q_fs_W^2\right)\right],`$
$`g_{A^0A^0\stackrel{~}{f}_L\stackrel{~}{f}_L}`$ $`=`$ $`{\displaystyle \frac{e^2}{2s_W^2}}\left[{\displaystyle \frac{\mathrm{cos}2\beta }{c_W^2}}\left(T_f^3Q_fs_W^2\right)\right],`$
$`g_{A^0A^0\stackrel{~}{f}_R\stackrel{~}{f}_R}`$ $`=`$ $`{\displaystyle \frac{e^2}{2s_W^2}}\left[{\displaystyle \frac{\mathrm{cos}2\beta }{c_W^2}}\left(Q_fs_W^2\right)\right].`$
For the third generation squarks (stop and sbottom), one has:
$`g_{H^0H^0\stackrel{~}{t}_{L,R}\stackrel{~}{t}_{L,R}}`$ $`=`$ $`g_{H^0H^0\stackrel{~}{u}_{L,R}\stackrel{~}{u}_{L,R}}{\displaystyle \frac{e^2}{2s_W^2}}\left({\displaystyle \frac{M_t\mathrm{sin}\alpha }{M_W\mathrm{sin}\beta }}\right)^2,`$
$`g_{h^0h^0\stackrel{~}{t}_{L,R}\stackrel{~}{t}_{L,R}}`$ $`=`$ $`g_{h^0h^0\stackrel{~}{u}_{L,R}\stackrel{~}{u}_{L,R}}{\displaystyle \frac{e^2}{2s_W^2}}\left({\displaystyle \frac{M_t\mathrm{cos}\alpha }{M_W\mathrm{sin}\beta }}\right)^2,`$
$`g_{H^0h^0\stackrel{~}{t}_{L,R}\stackrel{~}{t}_{L,R}}`$ $`=`$ $`g_{H^0h^0\stackrel{~}{u}_{L,R}\stackrel{~}{u}_{L,R}}{\displaystyle \frac{e^2\mathrm{sin}2\alpha }{4s_W^2}}\left({\displaystyle \frac{M_t}{M_W\mathrm{sin}\beta }}\right)^2,`$
$`g_{A^0A^0\stackrel{~}{t}_{L,R}\stackrel{~}{t}_{L,R}}`$ $`=`$ $`g_{A^0A^0\stackrel{~}{u}_{L,R}\stackrel{~}{u}_{L,R}}{\displaystyle \frac{e^2}{2s_W^2}}\left({\displaystyle \frac{M_t\mathrm{cos}\beta }{M_W\mathrm{sin}\beta }}\right)^2`$
and
$`g_{H^0H^0\stackrel{~}{b}_{L,R}\stackrel{~}{b}_{L,R}}`$ $`=`$ $`g_{H^0H^0\stackrel{~}{d}_{L,R}\stackrel{~}{d}_{L,R}}{\displaystyle \frac{e^2}{2s_W^2}}\left({\displaystyle \frac{M_b\mathrm{cos}\alpha }{M_W\mathrm{cos}\beta }}\right)^2,`$
$`g_{h^0h^0\stackrel{~}{b}_{L,R}\stackrel{~}{b}_{L,R}}`$ $`=`$ $`g_{h^0h^0\stackrel{~}{d}_{L,R}\stackrel{~}{d}_{L,R}}{\displaystyle \frac{e^2}{2s_W^2}}\left({\displaystyle \frac{M_b\mathrm{sin}\alpha }{M_W\mathrm{cos}\beta }}\right)^2,`$
$`g_{H^0h^0\stackrel{~}{b}_{L,R}\stackrel{~}{b}_{L,R}}`$ $`=`$ $`g_{H^0h^0\stackrel{~}{d}_{L,R}\stackrel{~}{d}_{L,R}}{\displaystyle \frac{e^2\mathrm{sin}2\alpha }{4s_W^2}}\left({\displaystyle \frac{M_b}{M_W\mathrm{cos}\beta }}\right)^2,`$
$`g_{A^0A^0\stackrel{~}{b}_{L,R}\stackrel{~}{b}_{L,R}}`$ $`=`$ $`g_{A^0A^0\stackrel{~}{d}_{L,R}\stackrel{~}{d}_{L,R}}{\displaystyle \frac{e^2}{2s_W^2}}\left({\displaystyle \frac{M_b\mathrm{sin}\beta }{M_W\mathrm{cos}\beta }}\right)^2.`$
## Appendix B: Passarino-Veltman functions
The calculation of one loop Feynman diagrams can be performed by combining propagators using the following formula:
$$\frac{1}{A_1^{\alpha _1}\mathrm{}A_n^{\alpha _n}}=\frac{\mathrm{\Gamma }(\alpha _1+\mathrm{}+\alpha _n)}{\mathrm{\Gamma }(\alpha _1)\mathrm{}\mathrm{\Gamma }(\alpha _n)}_0^1๐x_1\mathrm{}๐x_n\delta (x_1+\mathrm{}+x_n1)\frac{x_1^{\alpha _11}\mathrm{}x_n^{\alpha _n1}}{(A_1x_1+\mathrm{}+A_nx_n)^{\alpha _1+\mathrm{}+\alpha _n}}.$$
However, it is often convenient to reduce each one loop diagram to the sum of standard contributions, the so-called Passarino-Veltman (PV) functions.
### B.1 Standard definitions
Let us define the 1, 2, 3 and 4 point functions according to:
$`A_0(a)`$ $`=`$ $`{\displaystyle \frac{d^Dk}{i\pi ^2}\frac{1}{N_a}},`$
$`\{B_0,B^\mu ,B^{\mu \nu }\}(ab)`$ $`=`$ $`{\displaystyle \frac{d^Dk}{i\pi ^2}\frac{\{1,k^\mu ,k^\mu k^\nu \}}{N_aN_b}},`$
$`\{C_0,C^\mu ,C^{\mu \nu }\}(abc)`$ $`=`$ $`{\displaystyle \frac{d^Dk}{i\pi ^2}\frac{\{1,k^\mu ,k^\mu k^\nu \}}{N_aN_bN_c}},`$
$`\{D_0,D^\mu ,D^{\mu \nu },D^{\mu \nu \rho }\}(abcd)`$ $`=`$ $`{\displaystyle \frac{d^Dk}{i\pi ^2}\frac{\{1,k^\mu ,k^\mu k^\nu ,k^\mu k^\nu k^\rho \}}{N_aN_bN_cN_d}},`$
where the denominators are:
$`N_1`$ $`=`$ $`k^2m_1^2+iฯต,`$
$`N_2`$ $`=`$ $`(k+p_1)^2m_2^2+iฯต,`$
$`N_3`$ $`=`$ $`(k+p_1+p_2)^2m_3^2+iฯต,`$
$`N_4`$ $`=`$ $`(k+p_1+p_2+p_3)^2m_4^2+iฯต.`$
Here, all integrals are kept $`D`$-dimensional. However, the rest of the calculations will be performed in four dimensions.
In one loop diagrams, the following conventions are used:
* the external momenta $`p_{1\mathrm{}N}`$ are oriented clockwise,
* the internal masses $`m_{1\mathrm{}N}`$ are oriented clockwise as well, with $`m_1`$ between $`p_N`$ and $`p_1`$.
Let $`K`$ be a multi-index such that:
$`B_K(12)`$ $`=`$ $`B_K(p_1^2,m_1^2,m_2^2),`$
$`C_K(123)`$ $`=`$ $`C_K(p_1^2,p_2^2,(p_1+p_2)^2,m_1^2,m_2^2,m_3^2),`$
$`D_K(1234)`$ $`=`$ $`D_K(p_1^2,p_2^2,p_3^2,(p_1+p_2+p_3)^2,(p_1+p_2)^2,(p_2+p_3)^2,m_1^2,m_2^2,m_3^2,m_4^2).`$
The reduction of tensorial functions into scalar functions can then be done according to the following standard notations:
$`B^\mu (12)`$ $`=`$ $`p_1^\mu B_1(12),`$
$`B^{\mu \nu }(12)`$ $`=`$ $`p_1^\mu p_1^\nu B_{21}(12)+g^{\mu \nu }B_{22}(12),`$
$`C^\mu (123)`$ $`=`$ $`p_1^\mu C_{11}(123)+p_2^\mu C_{12}(123),`$
$`C^{\mu \nu }(123)`$ $`=`$ $`p_1^\mu p_1^\nu C_{21}(123)+p_2^\mu p_2^\nu C_{22}(123)+p_1^{\{\mu }p_2^{\nu \}}C_{23}(123)+g^{\mu \nu }C_{24}(123),`$
$`C^{\mu \nu \rho }(123)`$ $`=`$ $`(g^{\mu \nu }p_1^\rho +g^{\mu \rho }p_1^\nu +g^{\nu \rho }p_1^\mu )C_{001}(123)`$
$`+`$ $`(g^{\mu \nu }p_2^\rho +g^{\mu \rho }p_2^\nu +g^{\nu \rho }p_2^\mu )C_{002}(123)`$
$`+`$ $`p_1^\mu p_1^\nu p_1^\rho C_{111}(123)+p_2^\mu p_2^\nu p_2^\rho C_{222}(123)`$
$`+`$ $`(p_1^\mu p_1^\nu p_2^\rho +p_1^\mu p_2^\nu p_1^\rho +p_2^\mu p_1^\nu p_1^\rho )C_{112}(123)`$
$`+`$ $`(p_2^\mu p_2^\nu p_1^\rho +p_2^\mu p_1^\nu p_2^\rho +p_1^\mu p_2^\nu p_2^\rho )C_{122}(123),`$
$`D^\mu (1234)`$ $`=`$ $`p_1^\mu D_{11}(1234)+p_2^\mu D_{12}(1234)+p_3^\mu D_{13}(1234),`$
$`D^{\mu \nu }(1234)`$ $`=`$ $`p_1^\mu p_1^\nu D_{21}(1234)+p_2^\mu p_2^\nu D_{22}(1234)+p_3^\mu p_3^\nu D_{23}(1234)`$
$`+`$ $`p_1^{\{\mu }p_2^{\nu \}}D_{24}(1234)+p_1^{\{\mu }p_3^{\nu \}}D_{25}(1234)+p_2^{\{\mu }p_3^{\nu \}}D_{26}(1234)+g^{\mu \nu }D_{27}(1234),`$
$`D^{\mu \nu \rho }(1234)`$ $`=`$ $`(g^{\mu \nu }p_1^\rho +g^{\nu \rho }p_1^\mu +g^{\mu \rho }p_1^\nu )D_{001}(1234)`$
$`+`$ $`(g^{\mu \nu }p_2^\rho +g^{\nu \rho }p_2^\mu +g^{\mu \rho }p_2^\nu )D_{002}(1234)`$
$`+`$ $`(g^{\mu \nu }p_3^\rho +g^{\nu \rho }p_3^\mu +g^{\mu \rho }p_3^\nu )D_{003}(1234)`$
$`+`$ $`{\displaystyle \underset{1ijk3}{}}p_i^\mu p_j^\nu p_k^\rho D_{ijk}(1234).`$
In the reduction of $`D^{\mu \nu \rho }`$, the sum is over all triplets $`(i,j,k)`$ with repetitions, i.e. $`3^3=27`$ terms. By construction, the coefficients $`D_{ijk}`$ are invariant under index permutations.
More details about this approach and about the reduction of the PV tensorial integrals into scalar ones can be found in .
### B.2 LoopTools definitions
Sometimes, as for instance in the LoopTools library , it is convenient to use another notation and to introduce momenta $`k_i`$ given by:
$`k_1`$ $`=`$ $`p_1,`$
$`k_2`$ $`=`$ $`p_1+p_2,`$
$`k_3`$ $`=`$ $`p_1+p_2+p_3\mathrm{}`$
$`k_N`$ $`=`$ $`{\displaystyle \underset{i=1}{\overset{N}{}}}p_i`$
In that case, the tensorial coefficients are:
$`B^\mu `$ $`=`$ $`k_1^\mu B_1^L,`$
$`B^{\mu \nu }`$ $`=`$ $`k_1^\mu k_1^\nu B_{11}^L+g^{\mu \nu }B_{00}^L.`$
$`C^\mu `$ $`=`$ $`k_1^\mu C_1^L+k_2^\mu C_2^L,`$
$`C^{\mu \nu }`$ $`=`$ $`{\displaystyle \underset{ij=1,2}{}}k_i^\mu k_j^\nu C_{ij}^L+g^{\mu \nu }C_{00}^L,`$
$`C^{\mu \nu \rho }`$ $`=`$ $`{\displaystyle \underset{ijl=1,2}{}}k_i^\mu k_j^\nu k_l^\rho C_{ijl}^L+{\displaystyle \underset{i=1,2}{}}(g^{\mu \nu }k_i^\rho +g^{\mu \rho }k_i^\nu +g^{\nu \rho }k_i^\mu )C_{00i}^L,`$
$`D^\mu `$ $`=`$ $`k_1^\mu D_1^L+k_2^\mu D_2^L+k_3^\mu D_3^L,`$
$`D^{\mu \nu }`$ $`=`$ $`{\displaystyle \underset{1ijk3}{}}k_i^\mu k_j^\nu D_{ij}^L+g^{\mu \nu }D_{00}^L,`$
where $`C_{ij}^L`$, $`C_{ijl}^L`$ and $`D_{ij}^L`$ are completely symmetric.
By expanding these equations and by then comparing all their terms to those arising from the reduction of standard tensorial functions, one can find the relations that exist between the standard PV functions and those which are computed in the LoopTools library.
For the 2 point functions, one obtains:
$`B_1`$ $`=`$ $`B_1^L`$
$``$
$`B_{21}`$ $`=`$ $`B_{11}^L`$
$`B_{22}`$ $`=`$ $`B_{00}^L`$
For the 3 point functions, one obtains:
$`C_{11}`$ $`=`$ $`C_1^L+C_2^L`$
$`C_{12}`$ $`=`$ $`C_2^L`$
$``$
$`C_{21}`$ $`=`$ $`C_{11}^L+2C_{12}^L+C_{22}^L`$
$`C_{22}`$ $`=`$ $`C_{22}^L`$
$`C_{23}`$ $`=`$ $`C_{12}^L+C_{22}^L`$
$`C_{24}`$ $`=`$ $`C_{00}^L`$
$``$
$`C_{001}`$ $`=`$ $`C_{001}^L+C_{002}^L`$
$`C_{002}`$ $`=`$ $`C_{002}^L`$
$``$
$`C_{111}`$ $`=`$ $`C_{111}^L+3C_{112}^L+3C_{122}^L+C_{222}^L`$
$`C_{222}`$ $`=`$ $`C_{222}^L`$
$`C_{112}`$ $`=`$ $`C_{112}^L+2C_{122}^L+C_{222}^L`$
$`C_{122}`$ $`=`$ $`C_{122}^L+C_{222}^L`$
For the 4 point functions, one obtains:
$`D_{11}`$ $`=`$ $`D_1^L+D_2^L+D_3^L`$
$`D_{12}`$ $`=`$ $`D_2^L+D_3^L`$
$`D_{13}`$ $`=`$ $`D_3^L`$
$``$
$`D_{21}`$ $`=`$ $`D_{11}^L+D_{22}^L+D_{33}^L+2(D_{12}^L+D_{13}^L+D_{23}^L)`$
$`D_{22}`$ $`=`$ $`D_{22}^L+2D_{23}^L+D_{33}^L`$
$`D_{23}`$ $`=`$ $`D_{33}^L`$
$`D_{24}`$ $`=`$ $`D_{12}^L+D_{13}^L+D_{22}^L+2D_{23}^L+D_{33}^L`$
$`D_{25}`$ $`=`$ $`D_{13}^L+D_{23}^L+D_{33}^L`$
$`D_{26}`$ $`=`$ $`D_{23}^L+D_{33}^L`$
$`D_{27}`$ $`=`$ $`D_{00}^L`$
$``$
$`D_{001}`$ $`=`$ $`D_{001}^L+D_{002}^L+D_{003}^L`$
$`D_{002}`$ $`=`$ $`D_{002}^L+D_{003}^L`$
$`D_{003}`$ $`=`$ $`D_{003}^L`$
$``$
$`D_{111}`$ $`=`$ $`D_{111}^L+3D_{112}^L+3D_{113}^L+3D_{122}^L+6D_{123}^L+3D_{133}^L+D_{222}^L+3D_{223}^L+3D_{233}^L+D_{333}^L`$
$`D_{112}`$ $`=`$ $`D_{112}^L+D_{113}^L+2D_{122}^L+4D_{123}^L+2D_{133}^L+D_{222}^L+3D_{223}^L+3D_{233}^L+D_{333}^L`$
$`D_{113}`$ $`=`$ $`D_{113}^L+2D_{123}^L+2D_{133}^L+D_{223}^L+2D_{233}^L+D_{333}^L`$
$`D_{122}`$ $`=`$ $`D_{122}^L+2D_{123}^L+D_{133}^L+D_{222}^L+3D_{223}^L+3D_{233}^L+D_{333}^L`$
$`D_{133}`$ $`=`$ $`D_{133}^L+D_{233}^L+D_{333}^L`$
$`D_{123}`$ $`=`$ $`D_{123}^L+D_{133}^L+D_{223}^L+2D_{233}^L+D_{333}^L`$
$``$
$`D_{222}`$ $`=`$ $`D_{222}^L+3D_{223}^L+3D_{233}^L+D_{333}^L`$
$`D_{223}`$ $`=`$ $`D_{223}^L+2D_{233}^L+D_{333}^L`$
$`D_{233}`$ $`=`$ $`D_{233}^L+D_{333}^L`$
$``$
$`D_{333}`$ $`=`$ $`D_{333}^L`$
|
warning/0506/astro-ph0506592.html
|
ar5iv
|
text
|
# On metal-deficient barium stars and their link with yellow symbiotic starsBased on observations carried out at the European Southern Observatory (ESO, La Silla, Chile) and with the 1-m Swiss telescope at the Haute-Provence Observatory
## 1 The problem
Our understanding of the link between chemically-peculiar red giants like barium stars or CH stars, and yellow symbiotic stars (YSyS) has made substantial progress in the last decade (see the reviews by Jorissen, 2003a, b), mainly with the realisation that likely all yellow symbiotics (i.e., involving a giant of spectral type G or K as primary component) involve barium stars (Smith et al., 1996, 1997; Pereira & Porto de Mello, 1997; Pereira et al., 1998). The metal-deficient nature of the giant is a key factor, because it implies a rather large luminosity for the giant. Since evolutionary tracks of metal-deficient stars are shifted towards the blue, metal-deficient giants of spectral type K must lie on the upper part of the (asymptotic) giant branch (see Fig. 11 of Smith et al., 1996), where they suffer strong mass loss. If such metal-deficient K giants are in binary systems, their strong wind will interact with the companion and trigger symbiotic activity.
The other facet of this problem, namely whether all metal-deficient barium stars are symbiotic stars, is not yet fully answered. The present paper offers a first step in that direction.
We have collected a list of candidate metal-deficient barium stars and have assembled new observations to check (i) whether these stars are binaries, (ii) whether they are barium stars and (iii) whether they exhibit symbiotic activity.
## 2 The samples
Before discussing metal-deficient barium stars, it is useful to first summarize the properties of YSyS, to which metal-deficient barium stars may be compared.
### 2.1 Yellow symbiotic stars
All known YSyS are listed in Table 1, which shows that all the stars studied so far exhibit the barium syndrome. YSyS with a stellar infrared continuum (s-type, as opposed to the dusty dโ-type; see below) are clearly metal-deficient objects, as revealed by their low metallicities and high space velocities (CD $`43^{}14304`$ may be an exception; however, it is of spectral type K7, and should perhaps not be included in the family of YSyS). The presence of the barium syndrome among a family of binary metal-deficient stars fully supports the commonly accepted hypothesis that the s-process is more efficient at low metallicities (Clayton, 1988; Jorissen, 2003a). s-Type YSyS, with their metallicities lower than classical barium stars, may be expected to be, on average, more luminous than the latter (see Fig. 11 of Smith et al., 1996, comparing the luminosity function of Pop.I and Pop.II K giants). This is a direct consequence of the fact that evolutionary tracks shift towards the blue in the Hertzsprung-Russell (HR) diagram as metallicity decreases, as shown in Fig. 1b. Fig. 1a confirms that the YSyS AG Dra and BD $`21^{}3873`$ are indeed more luminous than classical barium stars. This difference in the average luminosity โ and hence mass-loss rate โ of the two populations thus explains why YSyS, despite hosting a K giant, exhibit symbiotic activity whereas barium stars do not. The larger mass-loss rates for the cool components of s-type YSyS โ as compared to Ba stars โ may be inferred from the comparison of their IRAS $``$ color indices, which reflect the amount of dust present in the system: ( $``$ )$`{}_{\mathrm{Ba}}{}^{}<`$ 0.1, as compared to 0.45 for AG Dra (Smith et al., 1996). Mรผrset et al. (1991) and Drake et al. (1987) provide direct measurements (or upper limits) for the mass loss rates of AG Dra and of Ba stars, respectively, which confirm the above conclusion.
YSyS with a dusty infrared continuum (dโ-type; Allen, 1982; Schmid & Nussbaumer, 1993) differ from their s-type counterparts in several respects (Table 1): they host a complex circumstellar environment (including cool dust, bipolar outflows, extended optical nebulae or emission-line spectra closely resembling those of planetary nebulae), the cool components have early spectral types (F to early K), they are often fast rotators (with the possible exception of M 1-2 =V471 Per; Grauer & Bond, 1981) and, finally, they belong to the galactic disk unlike s-type YSyS which belong to the halo.
All these arguments suggest that the hot component in dโ-type SyS has just evolved from the AGB to the WD stage. The rather cool dust (Schmid & Nussbaumer, 1993) is a relic from the mass lost by the AGB star. The optical nebulae observed in dโ-type SyS are most likely genuine planetary nebulae rather than the nebulae associated with the ionized wind of the cool component (Corradi et al., 1999). This is especially clear for AS 201 which actually hosts two nebulae (Schwarz, 1991): a large fossil planetary nebula detected by direct imaging, and a small nebula formed in the wind of the current cool component. Finally, the rapid rotation of the cool component has likely been caused by spin accretion from the former AGB wind like in WIRRING systems (Jeffries & Stevens, 1996; Jorissen, 2003b). The fact that the cool star has not yet been slowed down by magnetic braking is another indication that the mass transfer occurred fairly recently (Theuns et al., 1996). Corradi & Schwarz (1997) obtained 4000 y for the age of the nebula around AS 201, and 40000 y for V417 Cen.
### 2.2 Metal-deficient barium stars
Metal-deficient barium stars (with metallicities in the range $`1.1`$ to $`1.8`$, comparable to that of YSyS) were identified by Luck & Bond (1991), Mennessier et al. (1997) and Zaฤs et al. (2000), and occupy the same region of the HR diagram as YSyS (Fig. 1b). The question thus arises why metal-deficient barium stars are not SyS. Different answers must be sought, depending upon their absolute visual magnitudes $`M_\mathrm{V}`$. The most luminous systems, with $`M_\mathrm{V}<2`$, are likely located on the thermally-pulsing AGB, so that their Ba syndrome may be explained by internal nucleosynthesis. They thus should not be binaries, and therefore cannot be SyS! HD 104340 (open circle in Fig. 1b), a metal-deficient Ba star studied by Junqueira & Pereira (2001), and BD +$`03^{}2688`$ (Table 2) provide good illustrations of this situation, since they both lie above the TP-AGB threshold and CORAVEL radial-velocity measurements spanning several years do not reveal any clear orbital motion (Figs. 9 and 10, as well as Sect. 4).
The less luminous and warmest among metal-deficient Ba stars, clumping around $`M_\mathrm{V}+1`$ in the HR diagram, are also sometimes classified as CH stars (crosses in Fig. 1b). They are not losing mass at a large enough rate to trigger any symbiotic activity, as revealed by their small $``$ color indices ($`<0.3`$; Smith et al., 1996).
Finally, at intermediate luminosities ($`2M_\mathrm{V}+1`$) where YSyS are located, metal-deficient Ba stars are not luminous enough to be TP-AGB (hence they should be binaries), but yet their mass loss rates must be large enough to trigger symbiotic activity (provided that the orbital separation is not too large, since it is the mass accretion rate by the compact companion which is in fact the key parameter; see Sect. 5 and Jorissen 2003a). It is thus of great interest to check (i) the Ba nature of those metal-deficient stars with intermediate luminosities, (ii) their binary nature, and (iii) their suspected symbiotic activity. The first two issues are addressed in Sects. 3 and 4, respectively.
As far as a possible symbiotic activity is concerned, there is no indication from their photometric $`UB`$ and $`BV`$ indices that the metal-deficient stars in Table 2 have a strong blue continuum which could betray their symbiotic nature. It is thus very likely that none among these stars is a full-fledged symbiotic star. No signature of weak symbiotic activity (of the kind exhibited by some binary S stars; see Fig. 14 and Van Eck & Jorissen, 2002) is observed in the H<sub>ฮฑ</sub> line profile either (Fig. 2).
## 3 Abundances
The classification of the stars in Table 2 as metal-deficient Ba stars is subject to caution, as it does not rely on spectroscopic abundance analyses, but rather on a maximum-likelihood assignment based on kinematic, spatial and luminosity properties (Mennessier et al., 1997) for barium stars from the list of Lรผ (1991). Nevertheless, when a metallicity determination is available, it confirms the metal-deficient nature of the object (see Table 2). HD 139409 is an exception, though, since detailed spectral analyses reveal that it is neither metal-poor nor strongly enriched in s-process elements (see below). It may nevertheless be hoped that the metal-deficient assignment made by Mennessier et al. (1997) is valid in all the other cases. Regarding the Ba nature of these stars, it is known that the Lรผ (1991) catalogue of barium stars, from which the sample of barium stars used by Mennessier et al. (1997) was drawn, is contaminated by many non-barium stars (Griffin & Keenan, 1992; Jorissen et al., 1996), especially among those stars having a Ba index smaller than 1.
It would therefore in principle be necessary to re-evaluate the Ba nature of all the stars listed in Table 2. So far, spectra could be obtained for two of them, HD 139409 and HD 148897, which are discussed in detail in the present section. Of these, only HD 139409 appears to be a mild barium star, thus confirming the suspicion about the Lรผ (1991) catalogue expressed above.
### 3.1 The case of HD 148897 and HD 139409
HD 148897 (= HR 6152) has been tagged as a โlikely marginal barium starโ by Boyle & McClure (1975) and as a marginal CH star by Vilhu et al. (1977). It therefore found its way into the barium-star catalogs of Lรผ et al. (1983) and Lรผ (1991), as well as the catalog of CH and metal-deficient Ba stars of ล leivyte & Bartkeviฤius (1990). The star was classified as G8.5III CN-2 Fe-1 CH-1 by Keenan & McNeil (1989), and this classification as CN- and CH-weak contradicts the earlier assignments.
A detailed abundance analysis has thus been performed to clarify the situation, and its results are compared with previous analyses by Kyrรถlรคinen et al. (1986) and Luck (1991) in Sect. 3.1.2.
HD 139409 (= HIP 76605) has been classified as a marginal barium star by MacConnell et al. (1972), as G5 III Ba1 by Yamashita & Norimoto (1981) and as K0 III/II Ba 0.5 by Lรผ (1991).
#### 3.1.1 Observations
A high-resolution spectrum of HD 148897 was obtained using the Coudรฉ Matrix Echelle Spectrometer (MAESTRO; Musaev et al., 1999) delivering a resolving power of 45 000 and installed on the 2-m Zeiss telescope of the Terskol Observatory (located in Northern Caucasus at an altitude of 3100 m). The spectrometer is equipped with a Wright Instruments CCD detector with 1242 $`\times `$ 1152 pixels (22.5 $`\times `$ 22.5 $`\mu `$m). A total exposure of 1800 s was taken on February 18, 2003. The spectrum covers the range 365 to 1020 nm spread over 85 spectral orders.
A high-resolution spectrum of HD 139409 was obtained on the HARPS spectrograph (Mayor et al., 2003), delivering a resolution of 115 000 and installed on the ESO La Silla 3.6 m telescope. A total exposure of 200 s was obtained by Xavier Bonfils on May 31, 2004. In order to check equivalent widths delivered by HARPS, a spectrum of the standard star Arcturus was obtained as well by Fabien Carrier. The HARPS spectra were reduced by the observers using standard pipeline processing. Equivalent widths for the same set of lines as those studied in HD 139409 have been measured by one of us (L.Z.) in the HARPS spectrum of Arcturus and compared to those from the Arcturus spectral atlas (Hinkle et al., 2000). The agreement between the two sets of equivalent widths is excellent (Fig. 3), thus qualifying HARPS for abundance analyses. Spectra around the $`\lambda 614.172`$ nm Ba II line are shown in Fig. 4 for the two analyzed stars and for Arcturus (Hinkle et al., 2000).
#### 3.1.2 Reduction and analysis
HD 148897
The reduction of the CCD frames (subtraction of bias, dark and scattered light, flat fielding, extraction of echelle orders and wavelength calibration) was performed with the DECH20T software (Galazutdinov, 1992). More than 500 weak to medium-strong atomic lines, free of blends, were identified and their equivalent widths were measured with the DECH routines. The radial velocity was measured using a large number of symmetric absorption lines. The observed velocity has been brought to the heliocentric system by adding +22.7 km s<sup>-1</sup>. The mean heliocentric radial velocity for HD 144897 was found to be +16.7 km s<sup>-1</sup>. The atmospheric parameters for that star cannot be derived from photometry, since the standard temperature calibrations refer to stars of normal chemical composition. To obtain a colour-independent estimate of the temperature, a spectroscopic temperature has been derived from the excitation equilibrium of Fe I lines. The surface gravity $`\mathrm{log}g`$ was determined using the FeI/FeII ionization balance, whereas the microturbulent velocity $`v_t`$ was derived by forcing the abundances of individual Fe I lines to be independent of the reduced equivalent width. The resulting atmospheric parameters for HD 148897 are as follows: $`T_{\mathrm{eff}}=4350`$ K, $`\mathrm{log}g=1.0`$ (cgs), and $`v_t=2.0`$ km s<sup>-1</sup>. An independent determination of the surface gravity, using the above $`T_{\mathrm{eff}}`$ value, Mennessier et al. (1997) absolute visual magnitude ($`M_V=1.8`$), a bolometric correction of $`0.5`$ and a mass of 1 M, yields $`\mathrm{log}g=1.1`$, in agreement with the adopted value (see Luck, 1991, for a discussion of the discrepancy usually observed between the photometric and spectroscopic gravities). Comparison with atmospheric parameters from the literature is presented in Table 3.
The abundance analysis has been performed with the standard LTE line analysis program WIDTH9 developed by Kurucz. The model atmospheres were taken from Gustafsson et al. (1975). The synthetic spectra were generated using the spectral synthesis code STARSP (Tsymbal, 1996). Oscillator strengths have been taken from the VALD database (Piskunov et al., 1995). The resulting abundances (normalized by the solar-system abundances of Grevesse & Sauval, 1998) are listed in Table 3, from which it may be concluded that HD 148897 appears to be a rather typical metal-deficient star (McWilliam, 1997), and should certainly not be considered as a (metal-deficient) barium star.
Interestingly enough, there are no indications whatsoever from the Hipparcos and Tycho data (applying the methods described in Sect. 4) that this star is binary, in agreement with the fact that it is not a barium star.
HD 139409
The effective temperature of HD 139409 has been derived from the excitation equilibrium of Fe i, Ti i and Cr i lines (see Fig. 5 for Fe). The surface gravity $`\mathrm{log}g`$ was determined from the Fe i/Fe ii ionization balance, and the microturbulent velocity by forcing the abundances of individual Fe i, Ti i and Cr i lines to be independent of equivalent width (see Fig. 6 for Fe). Although spectroscopic gravity and temperature determinations in late-type, metal-deficient stars are probably affected by non-LTE effects, these effects remain small when \[Fe/H\] $`1.0`$ (see, for example, Allende Prieto et al., 1999). The stellar parameters of Arcturus (\[Fe/H\]$`=0.6`$, $`\mathrm{log}g=1.3`$) derived by the spectroscopic method (applied on the HARPS spectrum) are in good agreement indeed with those derived by other (non-spectroscopic) methods.
The resulting atmospheric parameters for HD 139409 are as follows: $`T_{\mathrm{eff}}=5000`$ K, $`\mathrm{log}g=2.8`$ (cgs), and $`\xi _t`$ = 2.0 km s<sup>-1</sup>. The spectroscopic gravity, combined with a mass of 1 M, leads to $`M_V+1.5`$. Thus our calculations indicate that HD 139409 is less luminous than predicted by Mennessier et al. (1997). The derived iron abundance is \[Fe/H\] = $`0.42`$ \[adopting $`\mathrm{log}(ฯต(\mathrm{Fe}))=7.50`$\].
The basic conclusion from the set of abundances listed in Table 4 and displayed in Fig. 7 is that HD 139409 appears to be a mild barium star.<sup>1</sup><sup>1</sup>1Interestingly, Pinsonneault et al. (1984) quoting a private communication from E. Luck, remark that Luck โis completing an abundance analysis of HD 139409, which will show that the light s-process elements do have ($`+1.5`$ dex) enhancements, while heavier s-process element enhancements are much smaller ($`+0.3`$ dex)โ. Luckโs study seems to have never been published. The s-process overabundances observed in HD 139409, although quite moderate, are not much smaller than those observed in a yellow symbiotic star like BD -213873, which exhibit overabundances of s-process elements in the range 0.3 โ 0.8 dex (Smith et al., 1997). With \[Fe/H\] = $`0.4`$, HD 139409 is, however, not as metal-deficient as the other stars considered in this paper.
## 4 Statistics of binarity among metal-deficient Ba stars
The list of confirmed or suspected metal-deficient barium stars remaining after the screening process based on abundances (as described in Sect. 3) is given in Table 5. The possible binary nature of those stars may be assessed using three different methods:
* checking for radial-velocity variations;
* checking for astrometric orbital motion, directly from the Hipparcos astrometric data;
* checking for astrometric orbital motion, indirectly from a comparison of the Hipparcos and Tycho-2 proper motions.
These methods are now described in turn.
### 4.1 Radial-velocity variations
Several stars from Table 5 (namely HIP 4347, HIP 29740, HIP 34795, HIP 55852, HIP 58956, HIP 76605 and BD +3$`{}_{}{}^{}2688`$) have been monitored for many years using the CORAVEL spectrovelocimeter (Baranne et al., 1979), as part of a larger program aiming at finding the frequency of spectroscopic binaries among s-process-rich late-type giants (see Jorissen & Mayor, 1988, 1992; Jorissen et al., 1998, for details and other results from this CORAVEL monitoring). Individual radial-velocity measurements for those stars (in the CORAVEL-ELODIE system as defined in Udry et al., 1999) are given in Tables 7 or in Udry et al. (1998). For a few other stars (HIP 43042, CS 22942-019 and CS 22948-027), radial velocities were monitored using other instruments, and their results were taken from the literature (Preston & Sneden, 2001; Zaฤs et al., 2005).
Orbits were already available for HIP 4347 and HIP 29740 (Udry et al., 1998), HIP 43042 (Zaฤs et al., 2005), as well as for CS 22942-019 and CS 22948-027 (Preston & Sneden, 2001). A new orbital solution has been derived for BD +042466 (Table 6 and Fig. 8). The binary nature of those stars is therefore beyond doubt.
The radial-velocity standard deviation of HIP 58596 is larger than expected based on the uncertainty on one measurement (Fig. 9), but no satisfactory orbital solution could be found. A 3-d orbit (with an eccentricity of 0.30) is possible, but this short orbital period is not consistent with the giant nature of HIP 58596, which imposes orbital periods of at least 20 d (see Fig. 4 of Pourbaix et al., 2004). The large standard deviation exhibited by HIP 58596 is therefore very likely another example of the large intrinsic jitter often observed for metal-deficient stars, as discussed by McClure (1984) and Carney et al. (2003). Nevertheless, the proper motion analysis presented in Sect. 4.3 hints at a possible long orbital period for HIP 58596.
A 14 y radial-velocity monitoring for HIP 34795 with the northern and southern CORAVELs is not very conclusive either, mostly because there are difficulties in finding the zero-point offset between the two instruments for such large radial velocities (Udry et al., 1999). When a $`1`$ km s<sup>-1</sup> offset is applied to the northern velocities (with respect to the values listed in Table 7), a long-term trend seems to be present, albeit with some superimposed jitter (Fig. 11). The analysis of the Hipparcos astrometric data presented in Sect. 4.2 suggests that the star might be binary, although the evidence is not very conclusive.
Finally, there is no sign of radial-velocity variations for BD +$`3^{}2688`$ (Fig. 10).
### 4.2 Orbital motion from Hipparcos astrometric data
A tailored reprocessing of the Hipparcos Intermediate Astrometric Data (hereafter IAD; van Leeuwen & Evans, 1998) makes it possible to look for a possible orbital signature in the astrometric motion, following the method outlined by Pourbaix & Jorissen (2000), Pourbaix & Boffin (2003), Pourbaix (2004) and applied to barium stars by Jorissen et al. (2004b). We give here only a brief summary of the method.
The basic idea is to quantify the likelihood of the fit of the Hipparcos astrometric data with an orbital model. For that purpose, Pourbaix & Arenou (2001) (see also Jancart et al., 2005) introduced several statistical indicators which allow us to decide whether to keep or to discard an orbital solution. Those indicators relevant to our purpose are the following:
* The addition of 4 supplementary parameters (the four Thiele-Innes orbital constants) describing the orbital motion should result in a statistically significant decrease of the $`\chi ^2`$ for the fit of the $`N`$ IAD with an orbital model with 9 free parameters ($`\chi _T^2`$), as compared to a fit with a single-star solution with 5 free parameters ($`\chi _S^2`$). This criterion is expressed by an $`F`$-test:
$$Pr_2=Pr[\widehat{F}>F(4,N9)],$$
(1)
where
$$\widehat{F}=\frac{N9}{4}\frac{\chi _S^2\chi _T^2}{\chi _T^2}.$$
(2)
$`Pr_2`$ is thus the first kind risk associated with the rejection of the null hypothesis:โthere is no orbital wobble present in the dataโ.
* Getting a substantial reduction of the $`\chi ^2`$ with the Thiele-Innes model does not necessarily imply that the four Thiele-Innes constants $`A,B,F,G`$ are significantly different from 0. The first kind risk associated with the rejection of the null hypothesis โthe orbital semi-major axis is equal to zeroโ may be expressed as
$$Pr_3=Pr[\chi _{ABFG}^2>\chi ^2(4)],$$
(3)
where
$$\chi _{ABFG}^2=X^tV^1X,$$
(4)
and $`X`$ is the vector of components $`A,B,F,G`$ and $`V`$ is its covariance matrix.<sup>2</sup><sup>2</sup>2Since it may be shown that $`\chi _S^2\chi _T^2=\chi _{ABFG}^2`$, the $`Pr_2`$ and $`Pr_3`$ tests are in fact equivalent provided that $`\chi _T^2N9`$. Thus, if $`Pr_2`$ and $`Pr_3`$ are significantly different, it means either that the Thiele-Innes orbital model does not fit the data very well ($`\chi _T^2>>N9`$), or that it fits much better than could be expected ($`\chi _T^2<<N9`$). We are indebted to L. Lindegren for this clarification (see also Jancart et al., 2005).
* For the orbital solution to be a significant one, its parameters should not be strongly correlated with the other astrometric parameters (e.g., the proper motion). In other words, the covariance matrix of the astrometric solution should be dominated by its diagonal terms, as measured by the efficiency $`ฯต`$ of the matrix being close to 1 (Eichhorn, 1989). The efficiency is simply expressed by
$$ฯต=\sqrt[m]{\frac{\mathrm{\Pi }_{k=1}^m\lambda _k}{\mathrm{\Pi }_{k=1}^mV_{kk}}},$$
(5)
where $`\lambda _k`$ and $`V_{kk}`$ are respectively the eigenvalues and the diagonal terms of the covariance matrix $`V`$.
With the above notations, the requirements for a star to qualify as a binary is then
$$\alpha (Pr_2+Pr_3)/ฯต0.02,$$
(6)
where the threshold value of 0.02 has been chosen to minimize false detections (Jorissen et al., 2004b).
Hipparcos data are, however, seldom precise enough to derive the orbital elements from scratch. Therefore, when a spectroscopic orbit is available beforehand, it is advantageous to import $`e,P,T`$ from the spectroscopic orbit and to derive the remaining astrometric elements (as done by Pourbaix & Jorissen, 2000; Pourbaix & Boffin, 2003). If a spectroscopic orbit is not available, trial $`(e,P,T)`$ triplets scanning a regular grid (with $`10P(\mathrm{d})5000`$ imposed by the Hipparcos scanning law and the mission duration) may be used. The quality factor $`\alpha `$ is then computed for each trial $`(e,P,T)`$ triplet, and if there exist triplets yielding $`\alpha <0.02`$, the star is flagged as a binary. To test its success rate, this method has been applied by Jorissen et al. (2004b) on a sample of barium stars. These authors show that, when $`\varpi >5`$ mas and $`100<P(\mathrm{d})<4000`$, the (astrometric) binary detection rate is close to 100%, i.e., the astrometric method recovers all known spectroscopic binaries (see also Jancart et al., 2005). When the orbit is not known beforehand, the method makes it even possible to find a good estimate for the orbital period, provided, however, that the true period is not an integer fraction, or a multiple, of one year. Here the method is applied to the sample of metal-deficient barium stars listed in Table 5.
The method flags as definite binaries the stars HIP 29740, 34795, 43042, 97874 and 107478 (Figs. 12 and 13). In two cases (HIP 29740 and 43042), the IAD method thus confirms the conclusion from the radial-velocity monitoring, but yields as well three new binaries (HIP 34795, 97874 and 107478). Two spectroscopic binaries (HIP 4347 and HIP 55852) are not detected by the IAD method because of their small parallax or long orbital period. The non-binary nature of HIP 58596, already suspected from the radial-velocity data, is confirmed by the analysis of the IAD (Fig. 13).
### 4.3 Orbital motion from a comparison of Hipparcos and Tycho-2 proper motions
Kaplan & Makarov (2003) suggested that the comparison of Hipparcos and Tycho-2 (Hรธg et al., 2000b) proper motions offers a way to detect binaries with long periods (typically from 2000 to 4000 d). The Hipparcos proper motion, being based on observations spanning only 3 y, may be altered by the orbital motion, especially for systems with periods in the range of 2000 to 4000 d whose orbital motion was not recognized by Hipparcos. On the other hand, this effect should average out in the Tycho-2 proper motion, which is derived from observations covering a much longer time span. This method, already used by Makarov (2004), Pourbaix (2004) and Jancart et al. (2006), works best when applied to stars with parallaxes in excess of about 5 mas.
The method evaluates the quantity
$$\chi _{\mathrm{obs}}^2=(\mu _{\mathrm{HIP}}\mu _{\mathrm{Tyc}})^t๐^1(\mu _{\mathrm{HIP}}\mu _{\mathrm{Tyc}}),$$
(7)
where $`\mu _{\mathrm{HIP}}`$ and $`\mu _{\mathrm{Tyc}}`$ are the vectors of $`\alpha `$ and $`\delta `$ components of the Hipparcos and Tycho-2 proper motions, respectively, and W is the associated $`2\times 2`$ variance-covariance matrix. The covariance between $`\mu _{\alpha ,\mathrm{HIP}}`$ and $`\mu _{\delta ,\mathrm{HIP}}`$, as provided by field H28 of the Hipparcos catalogue (ESA, 1997) and the correlation between Tycho-2 and Hipparcos proper motions, as encapsulated in the quantity $`R`$ of Table 1 of Hรธg et al. (2000a), have both been been considered (see Jancart et al. 2006 for details).
Since the above quantity follows a $`\chi ^2`$ probability distribution function with 2 degrees of freedom, it is then possible to compute the probability Prob$`(\chi ^2>\chi _{\mathrm{obs}}^2)`$, giving the first kind risk of rejecting the null hypothesis $`\mu _{\mathrm{Tycho}}=\mu _{\mathrm{HIP}}`$ while it is actually true. This probability is listed in Table 5, along with $`\chi _{\mathrm{obs}}^2`$, and the star is flagged as binary if Prob$`<0.1`$.
Only HIP 29740, HIP 76605 and HIP 97874 satisfy the test at the 10% threshold. Note, however, that all the other stars have parallaxes smaller than 5 mas, which make the test less efficient.
## 5 Summary of the binary criteria and discussion
The situation may be summarized as follows (see also last column of Table 5):
* Definite binaries with known orbits: HIP 4347, HIP 29740 (passes all three binarity tests), HIP 43042, HIP 55852, CS 22942-019, CS 22948-027;
* Suspected binaries from astrometric data (either IAD or proper motions; no or inconclusive radial-velocity data): HIP 34795, HIP 76605, HIP 97874 (both astrometric tests yield positive results), HIP 107478;
* Data inconclusive (mainly because of too small a parallax): HIP 11595, HIP 25161, HIP 69834;
* Non-binary stars (mainly from radial-velocity data, not contradicted by astrometry): HIP 58596, BD +32688.
The latter two non-binary stars are in fact good candidate thermally-pulsing AGB stars, as revealed by their location in the HR diagram (Table 2 and right panel of Fig. 1). Hence, they must not be binaries.
Finally, we come to the central question of this paper: Why do the metal-deficient barium stars, despite being binaries and occupying the same location of the HR diagram as YSyS, do not exhibit symbiotic activity? Three possible answers have been suggested in this paper:
(i) Some among the stars listed in Table 2 and displayed in Fig. 1 are in fact not barium stars (especially HIP 80843 = HD 148897).
(ii) Among those which are barium stars, some seem to lie on the TP-AGB, and thus need not be binaries (HIP 58596 = HD 104340, BD +32688). They therefore cannot exhibit symbiotic activity.
(iii) Finally, there remain a few genuine metal-deficient barium stars in the sample. Why are they not symbiotic stars? It seems that the answer to that question lies in the different period distributions for YSyS and metal-deficient barium stars: YSyS have shorter orbital period than metal-deficient barium stars, as seen in Fig. 14. This argument seems to apply especially to HIP 29740 (=HD 43389), which has been assigned a very bright absolute visual magnitude of $`3.5`$ by the maximum likelihood method of Mennessier et al. (1997). It is therefore expected to have a rather strong mass loss rate, and be a good candidate YSyS. However, with its orbital period of 1689 d, it lies at the long-period edge of the period distribution of YSyS (Fig. 14). The same difference seems to exist between the period distributions of red symbiotics and binary S stars (Van Eck & Jorissen, 2002, and Fig. 14).
However, a firm conclusion on this issue should await the determination of the orbital periods for the metal-deficient barium stars flagged as binaries by the IAD method (especially HIP 11595, HIP 25161, HIP 69834, HIP 97874, HIP 107478), so as to make the comparison between the orbital period distributions of metal-deficient barium stars and YSyS more meaningful.
###### Acknowledgements.
This work was performed in the framework of the NATO Collaborative Linkage Grant SA (PST.CLG.979128)6774/FP. We thank D. Pourbaix for the processing of the Hipparcos data of the metal-deficient stars. F. Carrier and X. Bonfils are thanked for obtaining the Arcturus and HD 139409 spectra. LZ thanks I. Platais for valuable discussions and support. The Fonds National Suisse de la Recherche Scientifique has funded the operations of the CORAVEL spectrometer and of the Swiss 1-m telescope installed at the Haute-Provence Observatory.
|
warning/0506/math0506232.html
|
ar5iv
|
text
|
# Untitled Document
A New View of Combinatorial Maps
by Smarandacheโs Notion
Linfan MAO
(Academy of Mathematics and Systems of Chinese Academy of Sciences, Beijing 100080, P.R.China)
e-mail: maolinfan@163.com
Abstract: On a geometrical view, the conception of map geometries is introduced, which is a nice model of the Smarandache geometries, also new kind of and more general intrinsic geometry of surfaces. Some open problems related combinatorial maps with the Riemann geometry and Smarandache geometries are presented.
Key Words: map, Smarandache geometry, model, classification.
AMS(2000): 05C15, 20H15, 51D99, 51M05
$`1.`$ What is a combinatorial map
A graph $`\mathrm{\Gamma }`$ is a $`2`$-tuple $`(V,E)`$ consists of a finite non-empty set $`V`$ of vertices together with a set $`E`$ of unordered pairs of vertices, i.e., $`EV\times V`$. Often denoted by $`V(\mathrm{\Gamma })`$, $`E(\mathrm{\Gamma })`$ the vertex set and edge set of the graph $`\mathrm{\Gamma }`$(\[$`9`$\]).
For example, the graph in the Fig.$`1`$ is the complete graph $`K_4`$ with vertex set $`V=\{1,2,3,4\}`$ and edge set $`E=\{12,13,14,23,24,34\}`$.
A map is a connected topological graph cellularly embedded in a surface. In 1973, Tutte gave an algebraic representation for an embedding of a graph on locally orientable surface $`([18])`$, which transfer a geometrical partition of a surface to a kind of permutation in algebra as follows($`[7][8]`$).
A combinatorial map $`M=(๐ณ_{\alpha ,\beta },๐ซ)`$ is defined to be a basic permutation $`๐ซ`$, i.e, for any $`x๐ณ_{\alpha ,\beta }`$, no integer $`k`$ exists such that $`๐ซ^kx=\alpha x`$, acting on $`๐ณ_{\alpha ,\beta }`$, the disjoint union of quadricells $`Kx`$ of $`xX`$ (the base set), where $`K=\{1,\alpha ,\beta ,\alpha \beta \}`$ is the Klein group, satisfying the following two conditions:
($`i`$) $`\alpha ๐ซ=๐ซ^1\alpha `$;
($`ii`$) the group $`\mathrm{\Psi }_J=<\alpha ,\beta ,๐ซ>`$ is transitive on $`๐ณ_{\alpha ,\beta }`$.
For a given map $`M=(๐ณ_{\alpha ,\beta },๐ซ)`$, it can be shown that $`M^{}=(๐ณ_{\beta ,\alpha },๐ซ\alpha \beta )`$ is also a map, call it the dual of the map $`M`$. The vertices of $`M`$ are defined as the pairs of conjugatcy orbits of $`๐ซ`$ action on $`๐ณ_{\alpha ,\beta }`$ by the condition $`(Ci)`$ and edges the orbits of $`K`$ on $`๐ณ_{\alpha ,\beta }`$, for example,$`x๐ณ_{\alpha ,\beta }`$, $`\{x,\alpha x,\beta x,\alpha \beta x\}`$ is an edge of the map $`M`$. Define the faces of $`M`$ to be the vertices in the dual map $`M^{}`$. Then the Euler characteristic $`\chi (M)`$ of the map $`M`$ is
$$\chi (M)=\nu (M)\epsilon (M)+\varphi (M)$$
where,$`\nu (M),\epsilon (M),\varphi (M)`$ are the number of vertices, edges and faces of the map $`M`$, respectively. For each vertex of a map $`M`$, its valency is defined to be the length of the orbits of $`๐ซ`$ action on a quadricell incident with $`u`$.
For example, the graph $`K_4`$ on the tours with one face length $`4`$ and another $`8`$ , can be algebraic represented as follows:
A map $`(๐ณ_{\alpha ,\beta },๐ซ)`$ with $`๐ณ_{\alpha ,\beta }=\{x,y,z,u,v,w,\alpha x,\alpha y,\alpha z,\alpha u,\alpha v,\alpha w,\beta x,\beta y,\beta z,`$ $`\beta u,\beta v,\beta w,\alpha \beta x,\alpha \beta y,\alpha \beta z,\alpha \beta u,\alpha \beta v,\alpha \beta w\}`$ and
$`๐ซ`$ $`=`$ $`(x,y,z)(\alpha \beta x,u,w)(\alpha \beta z,\alpha \beta u,v)(\alpha \beta y,\alpha \beta v,\alpha \beta w)`$
$`\times `$ $`(\alpha x,\alpha z,\alpha y)(\beta x,\alpha w,\alpha u)(\beta z,\alpha v,\beta u)(\beta y,\beta w,\beta v)`$
The four vertices of this map are $`\{(x,y,z),(\alpha x,\alpha z,\alpha y)\}`$, $`\{(\alpha \beta x,u,w),(\beta x,\alpha w,\alpha u)\}`$, $`\{(\alpha \beta z,\alpha \beta u,v),(\beta z,\alpha v,\beta u)\}`$ and $`\{(\alpha \beta y,\alpha \beta v,\alpha \beta w),(\beta y,\beta w,\beta v)\}`$ and six edges are $`\{e,\alpha e,\beta e,\alpha \beta e\}`$, where, $`e\{x,y,z,u,v,w\}`$. The Euler characteristic $`\chi (M)`$ is $`\chi (M)=46+2=0`$.
Geometrically, an embedding $`M`$ of a graph $`\mathrm{\Gamma }`$ on a surface is a map and has an algebraic representation. The graph $`\mathrm{\Gamma }`$ is said the underlying graph of the map $`M`$ and denoted by $`\mathrm{\Gamma }=\mathrm{\Gamma }(M)`$. For determining a given map $`(๐ณ_{\alpha ,\beta },๐ซ)`$ is orientable or not, the following condition is needed.
($`iii`$) If the group $`\mathrm{\Psi }_I=<\alpha \beta ,๐ซ>`$ is transitive on $`๐ณ_{\alpha ,\beta }`$, then $`M`$ is non-orientable. Otherwise, orientable.
It can be shown that the number of orbits of the group $`\mathrm{\Psi }_I=<\alpha \beta ,๐ซ>`$ in the Fig.$`2`$ action on $`๐ณ_{\alpha ,\beta }=\{x,y,z,u,v,w,\alpha x,\alpha y,`$ $`\alpha z,\alpha u,\alpha v,\alpha w,\beta x,\beta y,\beta z,\beta u,`$ $`\beta v,\beta w,\alpha \beta x,\alpha \beta y,\alpha \beta z,\alpha \beta u,\alpha \beta v,\alpha \beta w\}`$ is $`2`$. Whence, it is an orientable map and the genus of the surface is $`1`$. Therefore, the algebraic representation is correspondent with its geometrical mean.
$`2.`$ What are lost in combinatorial maps
As we known, mathematics is a powerful tool of sciences for its unity and neatness, without any shade of mankind. On the other hand, it is also a kind of aesthetics deep down in oneโs mind. There is a famous proverb says that only the beautiful things can be handed down to today, which is also true for the mathematics.
Here, the term unity and neatness is relative and local, also have various conditions. For acquiring the target, many unimportant matters are abandoned in the process. Whether are those matters in this time still unimportant in another time? It is not true. That is why we need to think the question: what are lost in the classical mathematics?
For example, a compact surface is topological equivalent to a polygon with even number of edges by identifying each pairs of edges along a given direction on it($`[17]`$). If label each pair of edges by a letter $`e,e`$, a surface $`S`$ is also identifying to a cyclic permutation such that each edge $`e,e`$ just appears two times in $`S`$, one is $`e`$ and another is $`e^1`$. Let $`a,b,c,\mathrm{}`$ denote the letters in $``$ and $`A,B,C,\mathrm{}`$ the sections of successive letters in linear order on a surface $`S`$ (or a string of letters on $`S`$). Then, a surface can be represented as follows:
$$S=(\mathrm{},A,a,B,a^1,C,\mathrm{}),$$
where $`a`$,$`A,B,C`$ denote a string of letters. Define three elementary transformations as follows:
$`(O_1)(A,a,a^1,B)(A,B);`$
$`(O_2)(i)(A,a,b,B,b^1,a^1)(A,c,B,c^1);`$
$`(ii)(A,a,b,B,a,b)(A,c,B,c);`$
$`(O_3)(i)(A,a,B,C,a^1,D)(B,a,A,D,a^1,C);`$
$`(ii)(A,a,B,C,a,D)(B,a,A,C^1,a,D^1).`$
If a surface $`S_0`$ can be obtained by the elementary transformation $`O_1`$-$`O_3`$ from a surface $`S`$, it is said that $`S`$ elementary equivalent with $`S_0`$, denoted by $`S_{El}S_0`$.
We have known the following formula in $`[8]`$:
$`(i)(A,a,B,b,C,a^1,D,b^1,E)_{El}(A,D,C,B,E,a,b,a^1,b^1);`$
$`(ii)(A,c,B,c)_{El}(A,B^1,C,c,c);`$
$`(iii)(A,c,c,a,b,a^1,b^1)_{El}(A,c,c,a,a,b,b).`$
Then we can get the classification theorem of compact surface as follows($`[14]`$):
Any compact surface is homeomorphic to one of the following standard surfaces:
($`P_0`$) The sphere: $`aa^1`$;
($`P_n`$) The connected sum of $`n,n1`$, tori:
$$a_1b_1a_1^1b_1^1a_2b_2a_2^1b_2^1\mathrm{}a_nb_na_n^1b_n^1;$$
($`Q_n`$) The connected sum of $`n,n1`$, projective planes:
$$a_1a_1a_2a_2\mathrm{}a_na_n.$$
Generally, a combinatorial map is a kind of decomposition of a surface. Notice that all the standard surfaces are just one face map underlying an one vertex graph. By combinatorial view, a combinatorial map is also a surface. But this assertion need more clarifying. For example, see the tetrahedron graph $`\mathrm{\Pi }_4`$ in the $`R^3`$ and a map $`K_4`$ on the sphere. Whether we can say it is the sphere? Certainly NOT. Since any point $`u`$ on a sphere has a neighborhood $`N(u)`$ homeomorphic to the open disc, therefore, all angles incident with the point $`1`$ must all be $`120^{}`$ degree on a sphere. But in $`\mathrm{\Pi }_4`$, they are all $`60^{}`$ degree. For making them topologically same, i.e., homeomorphism, we must blow up the $`\mathrm{\Pi }_4`$ to a sphere, as shown in the Fig.$`3`$. Whence, for getting the classification theorem of compact surfaces, we lose the angle,area, volume,distance,curvature,$`\mathrm{}`$, etc, which are also lost in the combinatorial maps.
Klein Erlanger Program says that any geometry is finding invariant properties under the transformation group of this geometry. This is essentially the group action idea and widely used in mathematics today. In the combinatorial maps, we know the following problems are applications of the Klein Erlanger Program:
($`i`$)to determine isomorphism maps or rooted maps;
($`ii`$)to determine equivalent embeddings of a graph;
($`iii`$)to determine an embedding whether exists;
($`iv`$)to enumerate maps or rooted maps on a surface;
($`v`$)to enumerate embeddings of a graph on a surface;
($`vi`$) $`\mathrm{}`$, etc.
All the problems are extensively investigated by researches in the last century and papers related those problems are still appearing frequently on the journals today. Then, what are their importance to classical mathematics? and what are their contributions to science? Those are the central topics of this paper.
$`3.`$ The Smarandache geometries
The Smarandache geometries is proposed by Smarandache in 1969 ($`[16]`$), which is a generalization of the classical geometries, i.e., the Euclid, Lobachevshy-Bolyai-Gauss and Riemannian geometries may be united altogether in the same space, by some Smarandache geometries. These last geometries can be either partially Euclidean and partially Non-Euclidean, or Non-Euclidean. It seems that the Smarandache geometries are connected with the Relativity Theory (because they include the Riemann geometry in a subspace) and with the Parallel Universes (because they combine separate spaces into one space) too(\[$`5`$\]). For a detail illustration, we need to consider the classical geometries.
The axioms system of Euclid geometry are the following:
(A1)there is a straight line between any two points.
(A2)a finite straight line can produce a infinite straight line continuously.
(A3)any point and a distance can describe a circle.
(A4)all right angles are equal to one another.
(A5)if a straight line falling on two straight lines make the interior angles on the same side less than two right angles, then the two straight lines, if produced indefinitely, meet on that side on which are the angles less than the two right angles.
The axiom (A5) can be also replaced by:
(A5โ)given a line and a point exterior this line, there is one line parallel to this line.
The Lobachevshy-Bolyai-Gauss geometry, also called hyperbolic geometry, is a geometry with axioms $`(A1)(A4)`$ and the following axiom $`(L5)`$:
(L5) there are infinitely many line parallels to a given line passing through an exterior point.
The Riemann geometry, also called elliptic geometry, is a geometry with axioms $`(A1)(A4)`$ and the following axiom $`(R5)`$:
there is no parallel to a given line passing through an exterior point.
By the thought of Anti-Mathematics: not in a nihilistic way, but in a positive one, i.e., banish the old concepts by some new ones: their opposites, Smarandache introduced the paradoxist geometry, non-geometry, counter-projective geometry and anti-geometry in $`[16]`$ by contradicts the axioms $`(A1)(A5)`$ in Euclid geometry, generalize the classical geometries.
Paradoxist geometry
In this geometry, its axioms are $`(A1)(A4)`$ and with one of the following as the axiom $`(P5)`$:
($`i`$)there are at least a straight line and a point exterior to it in this space for which any line that passes through the point intersect the initial line.
($`ii`$)there are at least a straight line and a point exterior to it in this space for which only one line passes through the point and does not intersect the initial line.
($`iii`$)there are at least a straight line and a point exterior to it in this space for which only a finite number of lines $`l_1,l_2,\mathrm{},l_k,k2`$ pass through the point and do not intersect the initial line.
($`iv`$)there are at least a straight line and a point exterior to it in this space for which an infinite number of lines pass through the point (but not all of them) and do not intersect the initial line.
($`v`$)there are at least a straight line and a point exterior to it in this space for which any line that passes through the point and does not intersect the initial line.
Non-Geometry
The non-geometry is a geometry by denial some axioms of $`(A1)(A5)`$, such as:
($`A1^{}`$)It is not always possible to draw a line from an arbitrary point to another arbitrary point.
($`A2^{}`$)It is not always possible to extend by continuity a finite line to an infinite line.
($`A3^{}`$)It is not always possible to draw a circle from an arbitrary point and of an arbitrary interval.
($`A4^{}`$)not all the right angles are congruent.
($`A5^{}`$)if a line, cutting two other lines, forms the interior angles of the same side of it strictly less than two right angle, then not always the two lines extended towards infinite cut each other in the side where the angles are strictly less than two right angle.
Counter-Projective geometry
Denoted by $`P`$ the point set, $`L`$ the line set and $`R`$ a relation included in $`P\times L`$. A counter-projective geometry is a geometry with the following counter-axioms:
($`C1`$)There exist: either at least two lines, or no line, that contains two given distinct points.
($`C2`$)Let $`p_1,p_2,p_3`$ be three non-collinear points, and $`q_1,q_2`$ two distinct points. Suppose that $`\{p_1.q_1,p_3\}`$ and $`\{p_2,q_2,p_3\}`$ are collinear triples. Then the line containing $`p_1,p_2`$ and the line containing $`q_1,q_2`$ do not intersect.
($`C3`$)Every line contains at most two distinct points.
Anti-Geometry
A geometry by denial some axioms of the Hilbertโs $`21`$ axioms of Euclidean geometry. As shown in $`[5]`$, there are at least $`2^{21}1`$ anti-geometries.
The Smarandache geometries are defined as follows.
Definition $`3.1`$ An axiom is said Smarandachely denied if the axiom behaves in at least two different ways within the same space, i.e., validated and invalided, or only invalided but in multiple distinct ways.
A Smarandache geometry is a geometry which has at least one Smarandachely denied axiom($`1969`$).
A nice model for the Smarandache geometries, called $`s`$-manifolds, is found by Iseri in $`[3][4]`$, which is defined as follows:
An $`s`$-manifold is any collection $`๐(T,n)`$ of these equilateral triangular disks $`T_i,1in`$ satisfying the following conditions:
$`(i)`$ Each edge $`e`$ is the identification of at most two edges $`e_i,e_j`$ in two distinct triangular disks $`T_i,T_j,1i,jn`$ and $`ij`$;
$`(ii)`$ Each vertex $`v`$ is the identification of one vertex in each of five, six or seven distinct triangular disks.
The vertices are classified by the number of the disks around them. A vertex around five, six or seven triangular disks is called an elliptic vertex, a Euclid vertex or a hyperbolic vertex, respectively.
An $`s`$-manifold is called closed if each edge is shared by exactly two triangular disks. An elementary classification for closed $`s`$-manifolds by triangulation are made in the reference $`[11]`$. The closed $`s`$-manifolds are classified into $`7`$ classes in $`[11]`$, as follows:
Classical Type:
$`(1)`$ $`\mathrm{\Delta }_1=\{5regulartriangularmaps\}`$ (elliptic);
$`(2)`$ $`\mathrm{\Delta }_2=\{6regulartriangularmaps\}`$(euclidean);
$`(3)`$ $`\mathrm{\Delta }_3=\{7regulartriangularmaps\}`$(hyperbolic).
Smarandache Type:
$`(4)`$ $`\mathrm{\Delta }_4=\{triangularmapswithvertexvalency5and6\}`$ (euclid-elliptic);
$`(5)`$ $`\mathrm{\Delta }_5=\{triangularmapswithvertexvalency5and7\}`$ (elliptic-hyperbolic);
$`(6)`$ $`\mathrm{\Delta }_6=\{triangularmapswithvertexvalency6and7\}`$ (euclid-hyperbolic);
$`(7)`$ $`\mathrm{\Delta }_7=\{triangularmapswithvertexvalency5,6and7\}`$ (mixed).
It is proved in $`[11]`$ that $`|\mathrm{\Delta }_1|=2`$, $`|\mathrm{\Delta }_5|2`$ and $`|\mathrm{\Delta }_i|,i=2,3,4,6,7`$ are infinite. Isier proposed a question in $`[3]`$: Do the other closed $`2`$-manifolds correspond to $`s`$-manifolds with only hyperbolic vertices?. Since there are infinite Hurwitz maps, i.e., $`|\mathrm{\Delta }_3|`$ is infinite, the answer is affirmative.
$`4.`$ The map geometries
Combinatorial maps can be used to construct new geometries, which are nice models for the Smarandache geometries, also a generalization of Isierโs model and Poincarรฉโs model for hyperbolic geometry.
$`4.1`$ Map geometries without boundary
For a given map on a surface, the map geometries without boundary are defined as follows.
Definition $`4.1`$ For a combinatorial map $`M`$ with each vertex valency$`3`$, associates a real number $`\mu (u),0<\mu (u)<\pi `$, to each vertex $`u,uV(M)`$. Call $`(M,\mu )`$ a map geometry with out boundary, $`\mu (u)`$ the angle factor of the vertex $`u`$ and to be orientablle or non-orientable if $`M`$ is orientable or not.
The realization of each vertex $`u,uV(M)`$ in $`R^3`$ space is shown in the Fig.$`1`$ for each case of $`\rho (u)\mu (u)>2\pi `$, $`=2\pi `$ or $`<2\pi `$.
$`\rho (u)\mu (u)<2\pi `$ $`\rho (u)\mu (u)=2\pi `$ $`\rho (u)\mu (u)>2\pi `$
Fig.$`1`$
As pointed out in the Section $`2`$, this kind of realization is not a surface, but it is homeomorphic to a surface. We classify points in a map geometry $`(M,\mu )`$ with out boundary as follows.
Definition $`4.2`$ A point $`u`$ in a map geometry $`(M,\mu )`$ is called elliptic, euclidean or hyperbolic if $`\rho (u)\mu (u)<2\pi `$, $`\rho (u)\mu (u)=2\pi `$ or $`\rho (u)\mu (u)>2\pi `$.
Then we have the following results.
Proposition $`4.1`$ Let $`M`$ be a map with $`uV(M),\rho (u)3`$. Then for $`uV(M)`$, there is a map geometries $`(M,\mu )`$ without boundary such that $`u`$ is elliptic, euclidean or hyperbolic in this geometry.
Proof Since $`\rho (u)3`$, we can choose the angle factor $`\mu (u)`$ such that $`\mu (u)\rho (u)<2\pi `$, $`\mu (u)\rho (u)=2\pi `$ or $`\mu (u)\rho (u)>2\pi `$. Notice that
$$0<\frac{2\pi }{\rho (u)}<\pi .$$
Whence, we can also choose $`\mu (u)`$ satisfying that $`0<\mu (u)<\pi \mathrm{}`$
Proposition $`4.2`$ Let $`M`$ be a map of order$`3`$ and $`uV(M),\rho (u)3`$. Then there exists a map geometry $`(M,\mu )`$ with out boundary, in which all points are one of the elliptic vertices, euclidean vertices and hyperbolic vertices or their mixed.
Proof According to the Proposition $`4.1`$, we can choose an angle factor $`\mu `$ such that a vertex $`u,uV(M)`$ to be elliptic, or euclidean, or hyperbolic. Since $`|V(M)|3`$, we can also choose the angle factor $`\mu `$ such that any two vertices $`v,wV(M)\backslash \{u\}`$ to be elliptic, or euclidean, or hyperbolic according to our wish. Then the map geometry $`(M,\mu )`$ makes the assertion hold. $`\mathrm{}`$
A geodesic in a manifold is a curve as straight as possible. Similarly, in a map geometry, its $`m`$-lines and $`m`$-points are defined as follows.
Definition $`4.3`$ Let $`(M,\mu )`$ be a map geometry without boundary. An $`m`$-line in $`(M,\mu )`$ is a curve with a constant curvature and points in it are called $`m`$-points.
If an $`m`$-line pass through an elliptic point or a hyperbolic point $`u`$, it must has the angle $`\frac{\mu (u)\rho (u)}{2}`$ with the entering line, not $`180^{}`$, which are explained in the Fig.$`2`$.
$`\mathrm{a}=\frac{\mu (u)\rho (u)}{2}<\pi `$ $`\mathrm{a}=\frac{\mu (u)\rho (u)}{2}>\pi `$
Fig.$`2`$
The following proposition asserts that all map geometries without boundary are Smarandache geometries.
Proposition $`4.3`$ For a map $`M`$ on a locally orientable surface with order$`3`$ and vertex valency$`3`$, there is an angle factor $`\mu `$ such that $`(M,\mu )`$ is a Smarandache geometry by denial the axiom (A5) with the axioms (A5),(L5) and (R5).
Proof According to the Proposition $`4.1`$, we know that there exist an angle factor $`\mu `$ such that there are elliptic vertices, euclidean vertices and hyperbolic vertices in $`(M,\mu )`$ simultaneously. The proof is divided into three cases.
Case $`1.`$ $`M`$ is a planar map
Notice that for a given line $`L`$ not pass through the vertices in the map $`M`$ and a point $`u`$ on its left side in $`(M,\mu )`$, if $`u`$ is an euclidean point, then there is one and only one line passes through $`u`$ not intersect with $`L`$, and if $`u`$ is an elliptic point, then there are infinite lines pass through $`u`$ not intersect with $`L`$, but if $`u`$ is a hyperbolic point, then each line passes through $`u`$ will intersect with $`L`$. Therefore, $`(M,\mu )`$ is a Smarandache geometry by denial the axiom (A5) with the axioms (A5), (L5) and (R5).
Case $`2.`$ $`M`$ is an orientable map
According to the classification theorem of compact surfaces, We only need to prove this result for the torus. Notice that on the torus, an $`m`$-line has the following properties (\[$`15`$\]):
If the slope $`\varsigma `$ of $`m`$-line $`L`$ is a rational number, then $`L`$ is a closed line on the torus. Otherwise, $`L`$ is infinite, and moreover $`L`$ passes arbitrarily close to every point of the torus.
Whence, if $`L_1`$ is an $`m`$-line on the torus, not passes through an elliptic or hyperbolic point, then for any point $`u`$ exterior $`L_1`$, we know that if $`u`$ is an euclidean point, then there is only one $`m`$-line passes through $`u`$ not intersect with $`L_1`$, and if $`u`$ is elliptic or hyperbolic, then any $`m`$-line passes through $`u`$ will intersect with $`L_1`$.
Now let $`L_2`$ be an $`m`$-line passes through an elliptic or hyperbolic point, such as the $`m`$-line in the Fig.$`3`$ and $`v`$ an euclidean point.
Fig.$`3`$
Then any $`m`$-line $`L`$ in the shade filed passes through the point $`v`$ will not intersect with $`L_2`$. Therefore, $`(M,\mu )`$ is a Smarandache geometry by denial the axiom (A5) with the axioms (A5),(L5) and (R5).
Case $`3.`$ $`M`$ is a non-orientable map
Similar to the Case $`2`$, by the classification theorem of the compact surfaces, we only need to prove this result for the projective plane. Now let the $`m`$-line passes through the center in the circle. Then if $`u`$ is an euclidean point, there is only one $`m`$-line passes through $`u`$, see (a) in the Fig.$`4`$. If $`v`$ is an elliptic point and there is an $`m`$-line passes through it and intersect with $`L`$, see (b) in the Fig.$`4`$, assume the point $`1`$ is a point such that the $`m`$-line $`1v`$ passes through $`0`$, then any $`m`$-line in the shade of (b) passes through the point $`v`$ will intersect with $`L`$.
Fig.$`4`$
If $`w`$ is a point and there is an $`m`$-line passes through it and does not intersect with $`L`$, see (c) in the Fig.$`4`$, then any $`m`$-line in the shade of (c) passes through the point $`w`$ will not intersect with $`L`$. Since the position of the vertices of the map $`M`$ on the projective plane can be choose as our wish, the proof is complete. $`\mathrm{}`$.
$`4.2`$ Map geometries with boundary
The Poincarรฉโs model for hyperbolic geometry hints us to introduce the map geometries with boundary, which is defined as follows.
Definition $`4.4`$ For a map geometry $`(M,\mu )`$ without boundary and faces $`f_1,f_2,\mathrm{},f_lF(M),1l\varphi (M)1`$, if $`(M,\mu )\{f_1,f_2,\mathrm{},f_l\}`$ is connected, then call $`(M,\mu )^l=(M,\mu )\{f_1,f_2,\mathrm{},f_l\}`$ a map geometry with boundary $`f_1,f_2,\mathrm{},f_l`$ and orientable or not if $`(M,\mu )`$ is orientable or not.
A connected curve with constant curvature in $`(M,\mu )^l`$ is called an $`m^{}`$-line and points $`m^{}`$-points.
The map geometries with boundary also are Smarandache geometries, which is convince by the following result.
Proposition $`4.4`$ For a map $`M`$ on a locally orientable surface with order$`3`$, vertex valency$`3`$ and a face $`fF(M)`$, there is an angle factor $`\mu `$ such that $`(M,\mu )^1`$ is a Smarandache geometry by denial the axiom (A5) with the axioms (A5),(L5) and (R5).
Proof Similar to the proof of the Proposition $`4.3`$, consider the map $`M`$ being a planar map, an orientable map on a torus or a non-orientable map on a projective plane, respectively. We get the assertion. $`\mathrm{}`$
Notice that for an one face map geometry $`(M,\mu )^1`$ with boundary, if we choose all points being euclidean, then $`(M,\mu )^1`$ is just the Poincarรฉโs model for hyperbolic geometry.
$`4.3`$ Classification of map geometries
For the classification of the map geometries, we introduce the following definition.
Definition $`4.5`$ Two map geometries $`(M_1,\mu _1)`$ and $`(M_2,\mu _2)`$ or $`(M_1,\mu _1)^l`$ and $`(M_2,\mu _2)^l`$ are equivalent if there is a bijection $`\theta :M_1M_2`$ such that for $`uV(M)`$, $`\theta (u)`$ is euclidean, elliptic or hyperbolic iff $`u`$ is euclidean, elliptic or hyperbolic.
The relation of the numbers of unrooted maps with the map geometries is the following.
Proposition $`4.5`$ If $``$ is a set of non-isomorphisc maps with order $`n`$ and $`m`$ faces, then the number of map geometries without boundary is $`3^n||`$ and the number of map geometries with one face being its boundary is $`3^nm||`$.
Proof By the definition, for a map $`M`$, there are $`3^n`$ map geometries without boundary and $`3^nm`$ map geometries with one face being its boundary by the Proposition $`4.3`$. Whence, we get $`3^n||`$ map geometries without boundary and $`3^nm||`$ map geometries with one face being its boundary from $``$. $`\mathrm{}`$.
We have the following enumeration result for the non-equivalent map geometries without boundary.
Proposition $`4.6`$ The numbers $`n^O(\mathrm{\Gamma },g)`$, $`n^N(\mathrm{\Gamma },g)`$ of non-equivalent orientable, non-orientable map geometries without boundary underlying a simple graph $`\mathrm{\Gamma }`$ by denial the axiom (A5) by (A5), (L5) or (R5) are
$$n^O(\mathrm{\Gamma },g)=\frac{3^{|\mathrm{\Gamma }|}\underset{vV(\mathrm{\Gamma })}{}(\rho (v)1)!}{2|\mathrm{Aut}\mathrm{\Gamma }|},$$
and
$$n^N(\mathrm{\Gamma },g)=\frac{(2^{\beta (\mathrm{\Gamma })}1)3^{|\mathrm{\Gamma }|}\underset{vV(\mathrm{\Gamma })}{}(\rho (v)1)!}{2|\mathrm{Aut}\mathrm{\Gamma }|},$$
where $`\beta (\mathrm{\Gamma })=\epsilon (\mathrm{\Gamma })\nu (\mathrm{\Gamma })+1`$ is the Betti number of the graph $`\mathrm{\Gamma }`$.
Proof Denote by $`(\mathrm{\Gamma })`$ the set of non-isomorphic maps underlying the graph $`\mathrm{\Gamma }`$ on locally orientable surfaces and by $`(\mathrm{\Gamma })`$ the set of embeddings of the graph $`\mathrm{\Gamma }`$ on the locally orientable surfaces. For a map $`M,M(\mathrm{\Gamma })`$, there are $`\frac{3^{|M|}}{|\mathrm{Aut}M|}`$ different map geometries without boundary by choosing the angle factor $`\mu `$ on a vertex $`u`$ such that $`u`$ is euclidean, elliptic or hyperbolic. From permutation groups, we know that
$$|\mathrm{Aut}\mathrm{\Gamma }\times <\alpha >|=|(\mathrm{Aut}\mathrm{\Gamma })_M||M^{\mathrm{Aut}\mathrm{\Gamma }\times <\alpha >}|=|\mathrm{Aut}M||M^{\mathrm{Aut}\mathrm{\Gamma }\times <\alpha >}|.$$
Therefore, we get that
$`n^O(\mathrm{\Gamma },g)`$ $`=`$ $`{\displaystyle \underset{M(\mathrm{\Gamma })}{}}{\displaystyle \frac{3^{|M|}}{|\mathrm{Aut}M|}}`$
$`=`$ $`{\displaystyle \frac{3^{|\mathrm{\Gamma }|}}{|\mathrm{Aut}\mathrm{\Gamma }\times <\alpha >|}}{\displaystyle \underset{M(\mathrm{\Gamma })}{}}{\displaystyle \frac{|\mathrm{Aut}\mathrm{\Gamma }\times <\alpha >|}{|\mathrm{Aut}M|}}`$
$`=`$ $`{\displaystyle \frac{3^{|\mathrm{\Gamma }|}}{|\mathrm{Aut}\mathrm{\Gamma }\times <\alpha >|}}{\displaystyle \underset{M(\mathrm{\Gamma })}{}}|M^{\mathrm{Aut}\mathrm{\Gamma }\times <\alpha >}|`$
$`=`$ $`{\displaystyle \frac{3^{|\mathrm{\Gamma }|}}{|\mathrm{Aut}\mathrm{\Gamma }\times <\alpha >|}}|^O(\mathrm{\Gamma })|`$
$`=`$ $`{\displaystyle \frac{3^{|\mathrm{\Gamma }|}\underset{vV(\mathrm{\Gamma })}{}(\rho (v)1)!}{2|\mathrm{Aut}\mathrm{\Gamma }|}}.`$
Similarly, we get that
$`n^N(\mathrm{\Gamma },g)`$ $`=`$ $`{\displaystyle \frac{3^{|\mathrm{\Gamma }|}}{|\mathrm{Aut}\mathrm{\Gamma }\times <\alpha >|}}|^N(\mathrm{\Gamma })|`$
$`=`$ $`{\displaystyle \frac{(2^{\beta (\mathrm{\Gamma })}1)3^{|\mathrm{\Gamma }|}\underset{vV(\mathrm{\Gamma })}{}(\rho (v)1)!}{2|\mathrm{Aut}\mathrm{\Gamma }|}}.`$
This completes the proof. $`\mathrm{}`$
For the classification of map geometries with boundary, we have the following result.
Proposition $`4.7`$ The numbers $`n^O(\mathrm{\Gamma },g)`$, $`n^N(\mathrm{\Gamma },g)`$ of non-equivalent orientable, non-orientable map geometries with one face being its boundary and underlying a simple graph $`\mathrm{\Gamma }`$ by denial the axiom (A5) by (A5), (L5) or (R5) are
$$n^O(\mathrm{\Gamma },g)=\frac{3^{|\mathrm{\Gamma }|}}{2|\mathrm{Aut}\mathrm{\Gamma }|}[(\beta (\mathrm{\Gamma })+1)\underset{vV(\mathrm{\Gamma })}{}(\rho (v)1)!\frac{2d(g[\mathrm{\Gamma }](x))}{dx}|_{x=1}]$$
and
$$n^N(\mathrm{\Gamma },g)=\frac{(2^{\beta (\mathrm{\Gamma })}1)3^{|\mathrm{\Gamma }|}}{2|\mathrm{Aut}\mathrm{\Gamma }|}[(\beta (\mathrm{\Gamma })+1)\underset{vV(\mathrm{\Gamma })}{}(\rho (v)1)!\frac{2d(g[\mathrm{\Gamma }](x))}{dx}|_{x=1}],$$
where $`g[\mathrm{\Gamma }](x)`$ is the genus polynomial of the graph $`\mathrm{\Gamma }`$ ( see \[$`12`$\]), i.e., $`g[\mathrm{\Gamma }](x)=\underset{k=\gamma (\mathrm{\Gamma })}{\overset{\gamma _m(\mathrm{\Gamma })}{}}g_k[\mathrm{\Gamma }]x^k`$ with $`g_k[\mathrm{\Gamma }]`$ being the number of embeddings of $`\mathrm{\Gamma }`$ on the orientable surface of genus $`k`$.
Proof Notice that $`\nu (M)\epsilon (M)+\varphi (M)=22g(M)`$ for an orientable map $`M`$ by the Euler characteristic. Similar to the proof of the Proposition $`4.6`$ with the notation $`(\mathrm{\Gamma })`$, by the Proposition $`4.5`$ we know that
$`n^O(\mathrm{\Gamma },g)`$ $`=`$ $`{\displaystyle \underset{M(\mathrm{\Gamma })}{}}{\displaystyle \frac{\varphi (M)3^{|M|}}{|\mathrm{Aut}M|}}`$
$`=`$ $`{\displaystyle \underset{M(\mathrm{\Gamma })}{}}{\displaystyle \frac{(2+\epsilon (\mathrm{\Gamma })\nu (\mathrm{\Gamma })2g(M))3^{|M|}}{|\mathrm{Aut}M|}}`$
$`=`$ $`{\displaystyle \underset{M(\mathrm{\Gamma })}{}}{\displaystyle \frac{(2+\epsilon (\mathrm{\Gamma })\nu (\mathrm{\Gamma }))3^{|M|}}{|\mathrm{Aut}M|}}{\displaystyle \underset{M(\mathrm{\Gamma })}{}}{\displaystyle \frac{2g(M)3^{|M|}}{|\mathrm{Aut}M|}}`$
$`=`$ $`{\displaystyle \frac{(2+\epsilon (\mathrm{\Gamma })\nu (\mathrm{\Gamma }))3^{|M|}}{|\mathrm{Aut}\mathrm{\Gamma }\times <\alpha >|}}{\displaystyle \underset{M(\mathrm{\Gamma })}{}}{\displaystyle \frac{|\mathrm{Aut}\mathrm{\Gamma }\times <\alpha >|}{|\mathrm{Aut}M|}}`$
$``$ $`{\displaystyle \frac{2\times 3^{|\mathrm{\Gamma }|}}{|\mathrm{Aut}\mathrm{\Gamma }\times <\alpha >|}}{\displaystyle \underset{M(\mathrm{\Gamma })}{}}{\displaystyle \frac{g(M)|\mathrm{Aut}\mathrm{\Gamma }\times <\alpha >|}{|\mathrm{Aut}M|}}`$
$`=`$ $`{\displaystyle \frac{(\beta (\mathrm{\Gamma })+1)3^{|M|}}{|\mathrm{Aut}\mathrm{\Gamma }\times <\alpha >|}}{\displaystyle \underset{M}{}}(\mathrm{\Gamma })|M^{\mathrm{Aut}\mathrm{\Gamma }\times <\alpha >}|`$
$``$ $`{\displaystyle \frac{3^{|\mathrm{\Gamma }|}}{|\mathrm{Aut}\mathrm{\Gamma }|}}{\displaystyle \underset{M(\mathrm{\Gamma })}{}}g(M)|M^{\mathrm{Aut}\mathrm{\Gamma }\times <\alpha >}|`$
$`=`$ $`{\displaystyle \frac{(\beta (\mathrm{\Gamma })+1)3^{|\mathrm{\Gamma }|}}{2|\mathrm{Aut}\mathrm{\Gamma }|}}{\displaystyle \underset{vV(\mathrm{\Gamma })}{}}(\rho (v)1)!{\displaystyle \frac{3^{|\mathrm{\Gamma }|}}{|\mathrm{Aut}\mathrm{\Gamma }|}}{\displaystyle \underset{k=\gamma (\mathrm{\Gamma })}{\overset{\gamma _m(\mathrm{\Gamma })}{}}}kg_k[\mathrm{\Gamma }]`$
$`=`$ $`{\displaystyle \frac{3^{|\mathrm{\Gamma }|}}{2|\mathrm{Aut}\mathrm{\Gamma }|}}[(\beta (\mathrm{\Gamma })+1){\displaystyle \underset{vV(\mathrm{\Gamma })}{}}(\rho (v)1)!{\displaystyle \frac{2d(g[\mathrm{\Gamma }](x))}{dx}}|_{x=1}].`$
Notice that $`n^L(\mathrm{\Gamma },g)=n^O(\mathrm{\Gamma },g)+n^N(\mathrm{\Gamma },g)`$ and the number of re-embeddings of an orientable map $`M`$ on surfaces is $`2^{\beta (M)}`$ (see also \[$`13`$\]). We have that
$`n^L(\mathrm{\Gamma },g)`$ $`=`$ $`{\displaystyle \underset{M(\mathrm{\Gamma })}{}}{\displaystyle \frac{2^{\beta (M)}\times 3^{|M|}\varphi (M)}{|\mathrm{Aut}M|}}`$
$`=`$ $`2^{\beta (M)}n^O(\mathrm{\Gamma },g).`$
Whence, we get that
$`n^N(\mathrm{\Gamma },g)`$ $`=`$ $`(2^{\beta (M)}1)n^O(\mathrm{\Gamma },g)`$
$`=`$ $`{\displaystyle \frac{(2^{\beta (M)}1)3^{|\mathrm{\Gamma }|}}{2|\mathrm{Aut}\mathrm{\Gamma }|}}[(\beta (\mathrm{\Gamma })+1){\displaystyle \underset{vV(\mathrm{\Gamma })}{}}(\rho (v)1)!{\displaystyle \frac{2d(g[\mathrm{\Gamma }](x))}{dx}}|_{x=1}].`$
This completes the proof. $`\mathrm{}`$
$`4.4`$ Polygons in a map geometry
A $`k`$-polygon in a map geometry is a $`k`$-polygon with each line segment being $`m`$-lines or $`m^{}`$-lines. For the sum of the internal angles in a $`k`$-polygon, we have the following result.
Proposition $`4.8`$ Let $`P`$ be a $`k`$-polygon in a map geometry with each line segment passes through at most one elliptic or hyperbolic point. If $`H`$ is the set of elliptic points and hyperbolic points on the line segment of $`P`$, then the sum of the internal angles in $`P`$ is
$$(k+|H|2)\pi \frac{1}{2}\underset{uH}{}\rho (u)\mu (u).$$
Proof Denote by $`U,V`$ the sets of elliptic points and hyperbolic points in $`H`$ and $`|U|=p,|V|=q`$. If an $`m`$-line segment passes through an elliptic point $`u`$, add a straight line segment in the plane as the Fig.$`6`$(1). Then we get that
$$\mathrm{angle}\mathrm{a}=\mathrm{angle1}+\mathrm{angle2}=\pi \frac{\rho (u)\mu (u)}{2}.$$
If an $`m`$-line passes through an hyperbolic point $`v`$, also add a straight line segment in the plane as the Fig.$`6`$(2). Then we get that
$$\mathrm{angle}b=\mathrm{angle3}+\mathrm{angle4}=\frac{\rho (v)\mu (v)}{2}\pi .$$
Fig.$`5`$
Since the sum of the internal angles of a $`k`$-polygon in the plane is $`(k2)\pi `$, we know that the sum of the internal angles in $`P`$ is
$`(k`$ $`2)\pi +{\displaystyle }_{uU}(\pi {\displaystyle \frac{\rho (u)\mu (u)}{2}}){\displaystyle }_{vV}({\displaystyle \frac{\rho (u)\mu (u)}{2}}\pi )`$
$`=`$ $`(k+p+q2)\pi {\displaystyle \frac{1}{2}}{\displaystyle \underset{uH}{}}\rho (u)\mu (u)`$
$`=`$ $`(k+|H|2)\pi {\displaystyle \frac{1}{2}}{\displaystyle \underset{uH}{}}\rho (u)\mu (u).`$
This completes the proof. $`\mathrm{}`$
As a corollary, we get the sum of the internal angles of a triangle in a map geometry as follows, which is consistent with the classical results.
Corollary $`4.1`$ Let $`\mathrm{}`$ be a triangle in a map geometry. Then
($`i`$) if $`\mathrm{}`$ is euclidean, then then the sum of its internal angles is equal to $`\pi `$;
($`ii`$) if $`\mathrm{}`$ is elliptic, then the sum of its internal angles is less than $`\pi `$;
($`iii`$) if $`\mathrm{}`$ is hyperbolic, then the sum of its internal angles is more than $`\pi `$.
$`5.`$ Open problems for applying maps to classical geometries
Here is a collection of open problems concerned combinatorial maps with the Riemann geometry and Smarandache geometries. Although they are called open problems, in fact, any solution for one of these problems needs to establish a new mathematical system first.
$`5.1`$ The uniformization theorem for simple connected Riemann surfaces
The uniformization theorem for simple connected Riemann surfaces is one of those beautiful results in the Riemann surface theory, which is stated as follows(\[$`2`$\]).
If $`๐ฎ`$ is a simple connected Riemann surface, then $`๐ฎ`$ is conformally equivalent to one and only one of the following three:
$`(a)`$$`๐\mathrm{}`$;
$`(b)`$$`๐`$;
$`(c)`$$`\mathrm{}=\{z๐||z|<1\}.`$
We have proved in $`[11]`$ that any automorphism of a map is conformal. Therefore, we can also introduced the conformal mapping between maps. Then, how can we define the conformal equivalence for maps enabling us to get the uniformization theorem of maps? What is the correspondence class maps with the three type $`(a)(c)`$ Riemann surfaces?
$`5.2`$ Combinatorial construction of an algebraic curve of genus
A complex plane algebraic curve $`๐_l`$ is a homogeneous equation $`f(x,y,z)=0`$ in $`P_2๐=(C^2(0,0,0))/`$, where $`f(x,y,z)`$ is a polynomial in $`x,y`$ and $`z`$ with coefficients in $`๐`$. The degree of $`f(x,y,z)`$ is said the degree of the curve $`๐_l`$. For a Riemann surface $`S`$, a well-known result is ($`[2]`$)there is a holomorphic mapping $`\phi :SP_2๐`$ such that $`\phi (S)`$ is a complex plane algebraic curve and
$$g(S)=\frac{(d(\phi (S))1)(d(\phi (S))2)}{2}.$$
By map theory, we know a combinatorial map also is on a surface with genus. Then whether we can get an algebraic curve by all edges in a map or by make operations on the vertices or edges of the map to get plane algebraic curve with given $`k`$-multiple points? and how do we find the equation $`f(x,y,z)=0`$?
$`5.3`$ Classification of $`s`$-manifolds by maps
We present an elementary classification for the closed $`s`$-manifolds in the Section $`3`$. For the general $`s`$-manifolds, their correspondence combinatorial model is the maps on surfaces with boundary, founded by Bryant and Singerman in $`1985`$ (). The later is also related to the modular groups of spaces and need to investigate further itself. The questions are
$`(i)`$ how can we combinatorially classify the general $`s`$-manifolds by maps with boundary?
$`(ii)`$ how can we find the automorphism group of an $`s`$-manifold?
$`(iii)`$ how can we know the numbers of non-isomorphic $`s`$-manifolds, with or without root?
$`(iv)`$ find rulers for drawing an $`s`$-manifold on a surface, such as, the torus, the projective plane or Klein bottle, not the plane.
The $`s`$-manifolds only using the triangulations of surfaces with vertex valency in $`\{5,6,7\}`$. Then what are the geometrical mean of the other maps, such as, the $`4`$-regular maps on surfaces. It is already known that the later is related to the Gauss cross problem of curves($`[9]`$).
$`5.4`$ Map geometries
As we have seen in the previous section, map geometries are the nice model of the Smarandache geometries. More works should be dong for them.
($`i`$) For a given graph, determine properties of the map geometries underlying this graph.
($`ii`$) For a given locally orientable surface, determine the properties of map geometries on this surface.
($`iii`$) Classify the map geometries on a locally orientable surface.
($`iv`$) Enumerate non-equivalent map geometries underlying a graph or on a locally orientable surface.
($`v`$) Establish the surface geometry by map geometries.
$`5.5`$ Gauss mapping among surfaces
In the classical differential geometry, a Gauss mapping among surfaces is defined as follows():
Let $`๐ฎR^3`$ be a surface with an orientation $`๐`$. The mapping $`N:๐ฎR^3`$ takes its value in the unit sphere
$$S^2=\{(x,y,z)R^3|x^2+y^2+z^2=1\}$$
along the orientation $`๐`$. The map $`N:๐ฎS^2`$, thus defined, is called the Gauss mapping.
we know that for a point $`P๐ฎ`$ such that the Gaussian curvature $`K(P)0`$ and $`V`$ a connected neighborhood of $`P`$ with $`K`$ does not change sign,
$$K(P)=\underset{A0}{lim}\frac{N(A)}{A},$$
where $`A`$ is the area of a region $`BV`$ and $`N(A)`$ is the area of the image of $`B`$ by the Gauss mapping $`N:๐ฎS^2`$. The questions are
$`(i)`$ what is its combinatorial meaning of the Gauss mapping? How to realizes it by maps?
$`(ii)`$ how we can define various curvatures for maps and rebuilt the results in the classical differential geometry?
$`5.6`$ The Gauss-Bonnet theorem
Let $`๐ฎ`$ be a compact orientable surface. Then
$$_๐ฎK๐\sigma =2\pi \chi (๐ฎ),$$
where $`K`$ is Gaussian curvature on $`๐ฎ`$.
This is the famous Gauss-Bonnet theorem for compact surface ($`[2],[6])`$. The questions are
$`(i)`$ what is its combinatorial mean of the Gauss curvature?
$`(ii)`$ how can we define the angle, area, volume, curvature, $`\mathrm{}`$, of a map?
($`iii`$) can we rebuilt the Gauss-Bonnet theorem by maps? or can we get a generalization of the classical Gauss-Bonnet theorem by maps?
$`5.7`$ Riemann manifolds
A Riemann surface is just a Riemann $`2`$-manifold, which has become a source of the mathematical creative power. A Riemann $`n`$-manifold $`(M,g)`$ is a $`n`$-manifold $`M`$ with a Riemann metric $`g`$. Many important results in Riemann surfaces are generalized to Riemann manifolds with higher dimension ($`[6]`$). For example, let $``$ be a complete, simple-connected Riemann $`n`$-manifold with constant sectional curvature $`c`$, then we know that $``$ is isometric to one of the model spaces $`^n,S_^n`$ or $`H_^n`$. Whether can we systematically rebuilt the Riemann manifold theory by combinatorial maps? or can we make a combinatorial generalization of results in the Riemann geometry, for example, the Chern-Gauss-Bonnet theorem ($`[6]`$)?
References
R.P.Bryant and D.Singerman, Foundations of the theory of maps on surfaces with boundary,Quart.J.Math.Oxford(2),36(1985), 17-41.
H. M.Farkas and I. Kra, Riemann Surfaces, Springer-Verlag New York inc(1980).
H.Iseri, Smarandache manifolds, American Research Press, Rehoboth, NM,2002.
H.Iseri, Partially Paradoxist Smarandache Geometries, http://www.gallup.unm. edu/sฬmarandache/Howard-Iseri-paper.htm.
L.Kuciuk and M.Antholy, An Introduction to Smarandache Geometries, Mathematics Magazine, Aurora, Canada, Vol.12(2003)
J.M.Lee, Riemann Manifolds, Springer-Verlag New York,Inc(1997).
Y.P.Liu, Advances in Combinatorial Maps, Northern Jiaotong University Publisher, Beijing (2003).
Yanpei Liu, Enumerative Theory of Maps, Kluwer Academic Publisher, Dordrecht / Boston / London (1999).
Y.P.Liu, Embeddability in Graphs, Kluwer Academic Publisher, Dordrecht / Boston / London (1995).
Mantredo P.de Carmao, Differential Geometry of Curves and Surfaces, Pearson Education asia Ltd (2004).
L.F.Mao, Automorphism groups of maps, surfaces and Smarandache geometries, American Research Press, Rehoboth, NM,2005.
L.F.Mao and Y.P.Liu, A new approach for enumerating maps on orientable surfaces, Australasian J. Combinatorics, vol.30(2004), 247-259.
L.F.Mao and Y.P.Liu, Group action approach for enumerating maps on surfaces,J.Applied Math. & Computing, vol.13, No.1-2,201-215.
W.S.Massey, Algebraic topology: an introduction, Springer-Verlag,New York, etc.(1977).
V.V.Nikulin and I.R.Shafarevlch, Geometries and Groups, Springer-Verlag Berlin Heidelberg (1987)
F. Smarandache, Mixed noneuclidean geometries, eprint arXiv: math/0010119, 10/2000.
J.Stillwell, Classical topology and combinatorial group theory, Springer-Verlag New York Inc., (1980).
W.T.Tutte, What is a maps? in New Directions in the Theory of Graphs (ed.by F.Harary), Academic Press (1973), 309 325.
|
warning/0506/astro-ph0506578.html
|
ar5iv
|
text
|
# The NII spectrum of the Orion Nebula
## 1 Introduction
Recent deep spectroscopic surveys of the Orion nebula (Esteban et al. 1998; Baldwin et al. 2000, hereafter BVV; Esteban et al. 2004, hereafter EPG) show that N<sup>+2</sup>+e recombination cannot explain the intensities of the N ii permitted lines with a nitrogen abundance that is consistent with the N ii forbidden line intensities. Seaton (1968) first suggested the possibility that permitted lines of C and O ions in nebulae may be excited by continuum fluorescence of starlight, and Grandi (1976) noted that this also must be an important mechanism to excite the N ii lines in Orion. Grandi suggested that additional absorption of photons of the He i $`1\mathrm{s}^2{}_{}{}^{1}\mathrm{S}_{0}^{}`$$`1\mathrm{s}8\mathrm{p}^1\mathrm{P}_1^\mathrm{o}`$ line at $`\lambda `$508.643 ร
from the diffuse field of the nebulae by the N ii $`2\mathrm{p}^2{}_{}{}^{3}\mathrm{P}_{0}^{}`$$`2\mathrm{p}4\mathrm{s}^3\mathrm{P}_1^\mathrm{o}`$ transition at $`\lambda `$508.668 ร
followed by decay to 3p terms would enhance the observed intensity of some lines. There is few direct evidence of the plausibility of this mechanism. BVV and EPG observed lines at $`\lambda `$3829.92, $`\lambda `$3838.47 and $`\lambda `$3856.27 that could be produced by the $`3\mathrm{p}^3\mathrm{P}_1`$$`4\mathrm{s}^3\mathrm{P}_2^\mathrm{o}`$, $`3\mathrm{p}^3\mathrm{P}_2`$$`4\mathrm{s}^3\mathrm{P}_2^\mathrm{o}`$ and $`3\mathrm{p}^3\mathrm{P}_2`$$`4\mathrm{s}^3\mathrm{P}_1^\mathrm{o}`$ transitions respectively. Unfortunately those lines are probably blended with lines from other elements, and their identification and intensity are uncertain. Sharpee et al. (2003) have detected those and other lines of the $`3\mathrm{p}^3\mathrm{P}`$$`4\mathrm{s}^3\mathrm{P}^\mathrm{o}`$ multiplet in a planetary nebula (IC 418), where the N ii spectrum is probably excited by fluorescence. The $`4\mathrm{s}^3\mathrm{P}_2^\mathrm{o}`$ level requires pumping of the the N ii $`2\mathrm{p}^2{}_{}{}^{3}\mathrm{P}_{2}^{}`$$`2\mathrm{p}4\mathrm{s}^3\mathrm{P}_2^\mathrm{o}`$$`\lambda `$508.697 ร
transition, which lies 32$`\mathrm{km}\mathrm{s}^1`$ from the He i line. Other lines from the $`4\mathrm{s}^3\mathrm{P}^\mathrm{o}`$ term like those of the $`3\mathrm{p}^3\mathrm{D}`$$`4\mathrm{s}^3\mathrm{P}^\mathrm{o}`$ multiplet in the $`\lambda \lambda `$3311.42โ3331.31 interval, which should be as intense as the $`3\mathrm{p}^3\mathrm{P}`$$`4\mathrm{s}^3\mathrm{P}^\mathrm{o}`$ multiplet, were not detected by EPG. The efficiency of this Bowenโtype line fluorescence depends heavily on uncertain nebular parameters that are needed in the theory of line radiative transfer, and modeling can become quite arbitrary. Therefore we will not consider it in this work (for further discussion see Escalante, 2002; Liu et al., 2001).
The critical parameters that determine the intensities of the lines and the relative importance of the fluorescence mechanism over the recombination process are the N<sup>+</sup> and N<sup>+2</sup> column densities in the gas and the stellar UV radiation field. Absorption of a UV photon by a resonant transition between a ground configuration state and an excited state has a higher probability of subsequent reemission in the same transition. Decay to an intermediate state will be favoured when the optical depth of the resonant transition is large, and the resonant photon is scattered a few times until it is converted into a lower energy photon producing a subordinate line. However the optical depth of the resonant transition must not become too large in order to allow enough resonant photons to penetrate into the N<sup>+</sup> zone. The efficiency of the continuum fluorescence excitation in N<sup>+</sup> depends on the transfer of resonant photons between the lowest resonant transition that can produce a subordinate line, $`2\mathrm{p}^2{}_{}{}^{3}\mathrm{P}`$$`3\mathrm{d}^3\mathrm{D}^\mathrm{o}`$$`\lambda \lambda `$533.51-533.88 ร
, and the ionization limit at 419 ร
. We have also included transitions from the $`2\mathrm{p}^2{}_{}{}^{1}\mathrm{D}_{2}^{}`$ and $`{}_{}{}^{1}\mathrm{S}_{0}^{}`$ metastable statesโpopulated mostly by collisionsโto other singlets in order to consider the observation of singlet lines in Orion. In the singlet system the lowest transition that produces a subordinate line is $`2\mathrm{s}^22\mathrm{p}^2{}_{}{}^{1}\mathrm{S}_{0}^{}`$$`2\mathrm{s}2\mathrm{p}^3{}_{}{}^{1}\mathrm{P}_{1}^{\mathrm{o}}`$$`\lambda `$745.84 ร
. Fig. 1 shows the observed transitions in Orion of the singlet and triplet N ii systems. Possible observations of quintet lines and higher excitation states are discussed in section 4.4.
The observed intensities of permitted lines in planetary nebulae are often higher than their intensities predicted by recombination rates with CNO abundances measured from forbidden lines (Liu et al., 1995, 2001; Luo et al., 2001, and references therein). Some of the N ii lines observed in PNe have 4f upper levels, which are excited mainly by recombination. In Orion there exists a similar situation and we will show that fluorescence cannot account for the excess intensity of lines from 4f levels. The accuracy of the recombination theory can be more easily tested by comparing line ratios from 4f decays because all the transitions involved are optically thin and the absolute line intensities have a similar dependence on temperature and density.
The fluorescence theory is more difficult to test because it depends, often nonโlinearly, on several model parameters. We will use the ratio of predicted over observed intensities, averaged over all N ii lines with reasonably accurate identifications and measurements, hereafter referred to as $`R=I_{\mathrm{cal}}/I_{\mathrm{obs}}`$, to test the model parameters. The scatter around $`R`$ will be used as a test of the atomic data and the details of the transferred stellar continuum.
Realistic nebular models and hot star model atmospheres along with available atomic data bases can explain successfully the intensities of a majority of the forbidden lines in Orion as well as general observed trends of ion abundances in Galactic HII regions (see for example Baldwin et al., 1991; Stasiลska & Schaerer, 1997). This paper demonstrates that the intensity of most of the N ii permitted lines in the Orion nebula can be predicted by fluorescence of the starlight continuum and some contribution of recombination by these models with currently accepted physical conditions and abundances of the nebula.
## 2 Calculation of population densities
### 2.1 Atomic processes
The important processes that can produce excited states with low principal quantum numbers in N<sup>+</sup> at nebular temperatures are absorption of UV photons by transitions from ground and metastable states and recombinations of N<sup>+2</sup>. Subsequent decays of these states produce the optical N ii spectrum. UV radiation is provided mostly by the star $`\theta ^1\mathrm{C}\mathrm{Ori}`$. The contribution of the diffuse continuum to the absorption is negligible.
The number density of a N<sup>+</sup> excited state $`j`$ at a certain point in the nebula in steady state, $`n_j`$, is given by
$$A_jn_j=\alpha _jn_en(\mathrm{N}^{+2})+\underset{g}{}\beta _{gj}n_g+\underset{k>j}{}A_{kj}n_k,$$
(1)
where $`A_j=_iA_{ji}`$ is the spontaneous decay rate, $`\alpha _j`$ is the recombination (radiative plus dielectronic) coefficient to state $`j`$, $`n_g`$ is the population density of one of the five N<sup>+</sup> ground and metastable states, $`2\mathrm{p}^2{}_{}{}^{3}\mathrm{P}_{0,1,2}^{},^1\mathrm{D}_2,^1\mathrm{S}_0`$, and $`\beta _{gj}`$ is the radiation absorption rate. If $`\overline{J}_\nu `$ is the local starlight mean intensity averaged over the absorption profile,
$$\beta _{gj}=\sigma _{gj}\left(\frac{4\pi \overline{J}_\nu }{h\nu }\right)\gamma _{gj},$$
(2)
where $`\sigma _{gj}=0.02654f_{gj}\mathrm{cm}^2\mathrm{Hz}`$ is the absorption cross section and $`f_{gj}`$ is the fโvalue of transition $`gj`$. $`\overline{J}_\nu `$ is attenuated by continuum opacity and geometry. The โpumping probabilityโ to account for the attenuation due to the transition $`gj`$ as defined by Ferland (1992) is
$$\gamma _{gj}=e^{\tau _0\varphi (x)}\varphi (x)๐x,$$
(3)
where $`\varphi (x)`$ is the normalized Voigt profile for a displacement $`x=(\nu \nu _0)/\mathrm{\Delta }\nu `$, and $`\tau _0=\sigma _{gj}N_g/\mathrm{\Delta }\nu `$ is the mean optical depth for a column density $`N_g=n_g๐r`$ integrated along a ray from the star.
An actual stellar spectrum shows absorption lines superposed on the continuum and the PโCygni profiles of the wind. If the structure of the stellar absorption spectrum varies in scales comparable to the width of $`\mathrm{\Delta }\lambda 0.01\text{ร
}`$ of the UV resonant lines in the nebular gas, equation (3) should be changed to
$$\gamma _{gj}=\psi (x)e^{\tau _0\varphi (x)}\varphi (x)๐x,$$
(4)
where $`\psi (x)=J_\nu /\overline{J}_\nu `$ is the continuum stellar profile around the resonant line. Grids with that resolution are available in the optical (Murphy & Meiksin, 2004). In the UV Smith et al. (2002) and Sternberg et al. (2003) have published models with a lower resolution.
The populations of the metastable states at nebular temperatures are controlled by electron collisions, and are nearly independent of other excited states. The system of equations (1) is triangular, and can be solved in terms of the cascade matrix, $`C_{kj}`$ (Seaton, 1959). $`C_{kj}`$ is the probability that a state $`j`$ is produced by the excitation of state $`k`$ followed by radiative decays by all possible routes ending in $`j`$. Equation (1) thus becomes
$$A_jn_j=\alpha _j^{\mathrm{eff}}n_en(\mathrm{N}^{+2})+\underset{g}{}\beta _{gj}^{\mathrm{eff}}n_g,$$
(5)
where
$`\alpha _j^{\mathrm{eff}}={\displaystyle \underset{kj}{}}\alpha _kC_{kj},`$ (6)
$`\beta _{gj}^{\mathrm{eff}}={\displaystyle \underset{kj}{}}\beta _{gk}C_{kj},`$ (7)
are the effective recombination coefficient and the effective fluorescence rate respectively. In practice the sums in equations (6) and (7) must be truncated and a correction must be added to equation (6) by extrapolation or by using hydrogenic populations (Escalante & Victor 1990, hereafter EV) because recombination coefficients of individual levels decrease slowly with $`n`$ and the contribution of states with high angular momentum and high $`n`$ must be included. Care must also be taken to avoid roundoff errors in the summations. The effective fluorescence rate in equation (7) is less sensitive to that type of correction because the absorption rate $`\beta _{gk}`$ is nonโzero only for states connected to the ground and metastable states by dipoleโallowed transitions, and decreases more rapidly with the principal quantum number $`n`$ of the upper state. The contribution of levels with $`4n12`$ and orbital angular momentum $`0l2`$ in equation (7) is less than 5% for the N ii lines observed in Orion. The contribution of f states to the fluorescence rate is even less important because they are not connected by resonant transitions to the ground term, and the transitions $`n\mathrm{d}m\mathrm{f}`$ have a low relative probability. By eliminating states with $`n>9`$ and $`l>2`$ in equation (7), the CPU time decreases by a factor of 10. However the f states (and higher $`l`$ states) must be included in the cascade due to recombinations, which tend to favor highโangular momentum states.
Most of the observed N ii transitions in Orion come from decays of the 3p and 3d triplet terms. Practically all the excitations of the 3d terms are produced by direct absorptions in the multiplets $`2\mathrm{s}^22\mathrm{p}^2{}_{}{}^{3}\mathrm{P}`$$`2\mathrm{p}3\mathrm{d}^3\mathrm{P}^\mathrm{o}`$$`\lambda \lambda `$529.36โ529.87 ร
and $`2\mathrm{s}^22\mathrm{p}^2{}_{}{}^{3}\mathrm{P}`$$`2\mathrm{p}3\mathrm{d}^3\mathrm{D}^\mathrm{o}`$$`\lambda \lambda `$533.51โ533.88 ร
. Cascades contribute negligibly to the populations of the 3d terms. The 3p terms, not being connected by direct transitions to the ground term, have more varied excitation channels. Between half and 80% of excitations of 3p terms come from decays of 3d terms. The rest comes mostly from absorptions in the multiplet $`2\mathrm{s}^22\mathrm{p}^2{}_{}{}^{3}\mathrm{P}`$$`2\mathrm{p}4\mathrm{s}^3\mathrm{P}^\mathrm{o}`$$`\lambda `$$`\lambda `$508.48โ509.01 ร
and higher s and d states. Table 1 shows a comparison of $`f`$โvalues for these multiplets, which are critical in our calculations.
The advantage of using the cascade matrix is that it needs to be computed once in the model in either the optically thin case (case A) or the extreme thick case (case B) because it depends solely on the Einstein coefficients, or more precisely, on the branching ratios $`P_{kj}=A_{kj}/_iA_{ki}`$. In the escape probability formalism the equations must be modified by substituting the Einstein coefficients of transitions connected to the states of the ground configuration $`2\mathrm{p}^2{}_{}{}^{3}\mathrm{P},^1\mathrm{D},^1\mathrm{S}`$ by $`P_eA_{kg}`$, where $`P_e`$ is the escape probability to be discussed below. $`P_e`$ is a local quantity that depends on the optical depth of the line as a function of position and thus the cascade matrix must be recomputed at each point in the nebula. However only a small fraction of the matrix elements depend on $`P_e`$. We found that recomputing only this fraction of the cascade matrix reduces the CPU time by a factor of 30.
### 2.2 Escape probabilities
We have used the escape probability theory to handle nearly 200 resonant transitions that produce 2630 transitions by fluorescence. A review of the limitations of this theory has been given by Dumont et al. (2003), but a full line transfer calculation is beyond present computation capabilities. There are many approximations for the escape probability function, which differ by several factors at high optical depths. In an ionized nebula absorption of resonant photons by an overlapping continuum is important (Hummer, 1968). In a uniform slab of total optical thickness $`T`$ the escape probability at depth $`\tau `$ is $`P_e=bF(b)+(1/2)[K_2(\tau ,b)+K_2(T\tau ,b)]`$, where the functions $`F`$ and $`K_2`$ are defined in Hummer & Storey (1992), and $`b=k_c/k_l`$ is the continuumโtoโmean line opacity ratio. The term
$$bF(b)=_{\mathrm{}}^{\mathrm{}}\frac{k_c}{k_l+k_c}\varphi (x)๐x,$$
(8)
is the probability that the resonant photon will be lost by continuum absorption. The effect of $`P_e`$ is to increase considerably the probability that a resonant absorption decays into the subordinate line rather than reemitting the resonant photon. The average ratio, $`R=I_{\mathrm{cal}}/I_{\mathrm{obs}}`$, increases by a factor of 20 between the two limits $`P_e1`$ (case A) and $`P_e0`$ (case B) respectively. A spectrum dominated by fluorescence, however, must be in an intermediate regime in order to allow the penetration of UV stellar photons through a large column density of absorbers at the same time that the resonant photons are scattered repeatedly.
All the resonant photons that produce the N ii optical spectrum have a high probability of being lost by conversion into other lines besides being absorbed by the continuum, and the probability of a large number of scatterings is small. Consequently we used a Doppler core in the calculation of $`K_2(\tau ,b)`$ and $`F(b)`$.
### 2.3 Atomic Data
The main source of $`A`$โ and $`f`$โvalues for this work is the compilation of Wiese et al. (1996), which is based primarily on the Opacity Project data (Seaton et al., 1994) and configurationโinteraction calculations, but also includes intermediate coupling calculations and experimental measurements. That compilation has data for most lines of the series $`2\mathrm{s}^22\mathrm{p}(^2\mathrm{P})nl`$ with $`n5`$ and $`l2`$, and $`2\mathrm{s}2\mathrm{p}^2(^4\mathrm{P})nl`$ with $`n2`$ and $`l1`$, including spinโforbidden transitions between quintet and triplet terms. The model potential calculations from Victor & Escalante (1988) were used for the rest of the transitions needed in the cascade matrix of the effective fluorescence rate in equation (7). Transition probabilities between fine structure levels were obtained by applying LS fractions to the multiplet data (Allen, 1973).
Table 2 shows the effect of different atomic databases on the effective fluorescence rate (7) summed over the ground and metastable states times the branching ratio of the subordinate line. The continuum is a blackbody spectrum at $`T=37\mathrm{kK}`$ with a photon flux of $`9.24\times 10^4\mathrm{cm}^2\mathrm{s}^1\mathrm{Hz}^1`$ at $`\lambda `$533.7 ร
and $`\tau _0=0`$ for all lines, i.e., $`\gamma _{gj}=1`$, $`P_e=1`$. The compilation of Wiese et al. (1996) does not include transitions to levels with principal quantum number greater than $`5`$ and produce lower fluorescence rates of lines with upper p levels because those levels have cascade contributions from higher levels. The model potential data of Victor & Escalante (1988) does not have transitions to doublyโexcited configurations and consequently give higher rates as shown in the fourth column of table 2. Transitions between states with the $`2\mathrm{s}2\mathrm{p}^3`$ configuration and the singly excited configurations $`2\mathrm{s}^22\mathrm{p}3\mathrm{p}`$ can change the branching ratios significantly. The entries in the fifth column have been complemented with transitions to states with the $`2\mathrm{s}2\mathrm{p}^2`$ core configuration and spin-forbidden transitions taken from Wiese et al. (1996). The last column combines both data sets. The cascade matrix elements tend to be similar for different data sets because systematic differences in the atomic parameters between different data sets cancel out in the branching ratios $`P_{ji}=A_{ji}/A_j`$.
A recent calculation of effective recombination coefficients by Kisielius & Storey (2002) shows a general agreement with the model potential calculations of Escalante & Victor (1990) (hereafter EV) and the calculations of Pรฉquignot et al. (1991) for the $`2\mathrm{s}^22\mathrm{p}(^2\mathrm{P}^\mathrm{o})3\mathrm{d}`$ and 4f configurations. The most important differences between the three calculations are in the branching ratios of multiplets from the $`2\mathrm{s}^22\mathrm{p}(^2\mathrm{P}^\mathrm{o})3\mathrm{p}^3\mathrm{D}`$ term. The accuracy of the model potential in this case was limited by the lack of observed energies in the $`2\mathrm{s}^22\mathrm{p}(^2\mathrm{P}^\mathrm{o})\mathrm{np}`$ series. The best agreement between the three data sets is in the 4f terms, where most of the contribution to the recombination comes from levels with small non-hydrogenic effects. The main uncertainty is in the line fractions involving 4f terms, where LSโcoupling is not a good approximation. EV used LK coupling for these terms, and line fractions for other couplings are available (Escalante & Gongora, 1990), but general intermediateโcoupling calculations for those states are clearly needed.
This work uses the recombination coefficients of EV with branching ratios given by the Aโvalues of Victor & Escalante (1988) and Wiese et al. (1996). Effective recombination coefficients for the levels of a term were obtained by assuming that the coefficients are proportional to the statistical weights of the levels.
## 3 Model calculations
### 3.1 Nebular models
In order to determine the electron, N<sup>+2</sup>, and N<sup>+</sup> densities in equation (5), as well as the temperature and opacity at each point in the nebula, we used the codes cloudy (Ferland, 1996, version 90.05) and nebu (Pรฉquignot et al., 2001; Morisset et al., 2002). Models of the Orion nebula support the existence of a main emitting layer at the back of a cavity in the OMCโ1 molecular cloud. The thickness of the layer is highly variable across the nebula (Wen & OโDell, 1995). We approximated the layer by a planeโparallel model at constant pressure with cloudy and constant density with nebu. cloudy allowed us to use the grain composition used by Baldwin et al. (1991) in their Orion model. Predicted line intensities by cloudy show some sensitivity when the radiation is included in the pressure law (Baldwin et al., 1996). Models with a constant gas pressure produce larger N<sup>+</sup> column densities than models with a constant gas plus radiation pressure. We have not tried to find a single model fit to the observed forbidden line intensities in Orion. Instead we have run a series of models to find the most important dependencies of the N ii lines excited by fluorescence on model parameters.
### 3.2 Model atmospheres
The Orion nebula is mostly excited by $`\theta ^1\mathrm{C}\mathrm{Ori}`$. The other Trapezium stars increase the fluorescence of the N ii lines by less than 2% and will not be considered in this calculation. $`\theta ^1\mathrm{C}\mathrm{Ori}`$ is a class V star with variable wind features that produce uncertainties in the determination of its spectral classification and effective temperature. Different authors give values close to $`T_{\mathrm{eff}}=39\mathrm{kK}`$ and $`\mathrm{log}(g)=4`$ for this star (Howarth & Prinja, 1989; Hillenbrand, 1997, and references therein). However comparisons of the intensities of forbidden lines with nebular model predictions suggest temperatures as low as 36 kK (BVV). Metalicity measurements by Cunha & Lambert (1994) show that the Trapezium stars are slightly underabundant with respect to the Sun.
The calculation of the fluorescence of the gas needs a high resolution stellar spectrum. This is important with hot massive stars, which have expanding atmospheres with a dense forest of absorption lines and broad overlapping P Cygni profiles. The emission peaks and absorption troughs of the overlapping P Cygni profiles can increase or decrease the absorption rate in equation (3) by large factors and change the fluorescence excitation rate significantly. On the other hand nebular models usually smooth the spectrum of the exciting star to calculate the ionization structure. In order to take into account the detailed spectral structure of the model atmosphere, we input the unattenuated stellar spectrum into the nebular model and extracted the predicted attenuated local continuum at each point in the nebula. To calculate the pumping rate in equation (3), the full resolution spectrum was read and interpolated at the absorption frequency and was scaled by the ratio of the attenuated to the unattenuated stellar continuum predicted by the nebular model.
The motion of the star with respect to the gas introduces a Doppler shift that can change the intensity of the continuum at the absorbing frequencies. The proper motion velocity of $`\theta ^1\mathrm{C}\mathrm{Ori}`$ is uncertain (Wen & OโDell, 1995). Doppler shifts of up to $`\pm 30\mathrm{km}\mathrm{s}^1`$ did not change predicted line intensities by more than a few percent with the resolution of about $`\mathrm{\Delta }\lambda 1\text{ร
}`$ in the far UV of the model atmospheres that we used. Therefore we assumed a static star with respect to the gas.
Recent model atmospheres of O stars include the effects of line blanketing and line blocking of the stellar wind. We used model atmospheres calculated with the wmbasic code (Pauldrach et al., 2001; Sternberg et al., 2003) to account for these effects. We also used the LTE, lineโblanketed atmospheres of Kurucz (1991) for comparison purposes.
## 4 Basic model
All transitions of the observed permitted lines in Orion end in excited states and are optically thin. Their intensities can thus be obtained by integration of the emissivity along the line of sight:
$$I=\frac{h\nu }{4\pi }A_{ji}n_j๐r$$
(9)
We now examine the dependence of the fluorescence excitation on the stellar spectrum and the density and compare them with the observations of BVV and EPG.
We have adopted a set of central values for the parameters of the models that are given in table 3. These values are close to the ones recommended by Baldwin et al. (1991), and we will refer to them as the basic model (BM). We discuss below a few features of this model.
### 4.1 Gas abundances and density.
The N<sup>+2</sup> abundance in the nebula can be estimated from the measurements of lines with upper 4f levels, which are populated mostly by recombination. Table 4 lists the observed intensities of these lines in Orion and their predicted recombination emission rates normalized to the observed intensities and recombination rate of the $`3\mathrm{s}^3\mathrm{P}_2^\mathrm{o}`$$`3\mathrm{p}^3\mathrm{D}_3`$$`\lambda `$5679.56 ร
line. BVV and EPG measured the lines at $`\lambda \lambda `$4236.91, 4237.05 and 4241.784, which account for 92% of the total strength of the $`3\mathrm{d}^3\mathrm{D}^\mathrm{o}`$$`4\mathrm{f}\mathrm{F}`$ multiplet if the 4f levels are described by an LK coupling scheme as shown in the table. We do not include in the abundance estimation the $`\lambda `$4242.49 line because its measured intensity is much higher than the one predicted from the recombination theory. At $`T_e=10^4\mathrm{K}`$ the effective recombination coefficient of multiplet $`3\mathrm{d}^3\mathrm{D}^\mathrm{o}`$$`4\mathrm{f}\mathrm{F}`$ is $`5.6\times 10^{14}\mathrm{cm}^3\mathrm{s}^1`$ (EV), which implies $`\mathrm{N}^{+2}/\mathrm{H}^+=9.5\times 10^5`$ if the lines were produced solely by recombination. This abundance is higher than the one implied by the forbidden lines $`6\times 10^5`$ in Orion and many galactic H ii regions (e.g., Shaver et al., 1983; Afflerbach et al., 1977).
We notice that EPG detected one line from the $`3\mathrm{d}^3\mathrm{F}^\mathrm{o}`$$`4\mathrm{f}\mathrm{G}`$ multiplet ($`\lambda `$4041.31) although other lines of this multiplet theoretically should be more intense than the observed lines of the $`3\mathrm{d}^3\mathrm{D}^\mathrm{o}`$$`4\mathrm{f}\mathrm{F}`$ multiplet. The corresponding $`\mathrm{N}^{+2}/\mathrm{H}^+`$ abundance from this line is $`8.5\times 10^5`$. Liu et al. (2001) has reported a N<sup>+2</sup> abundance of $`4.47\times 10^5`$ in Orion from the $`\lambda \lambda `$4041.31 and 4043.53 lines, which is more consistent with the abundance from the forbidden lines.
Several authors (Liu et al., 2000, 2001; Tsamis et al., 2003; Peimbert et al., 2004) have detected N ii lines in planetary nebula with relative rates that are more compatible with the recombination theory and will be the subject of future work. In a typical H ii region there is not enough N<sup>+2</sup> column density to produce a N<sup>+</sup> recombination spectrum. Fig. 2 shows separately the intensities given by equation (9) (integrated from right to left) due to recombination and fluorescence as a function of depth for a line. While the recombination intensity grows linearly with distance the fluorescence intensity grows more rapidly. The fluorescence excitation is favoured by the more intense starlight continuum and low opacity in the near side of the nebula, while the recombination emissivity is more uniformly distributed. In the far side of the nebula the high optical depth in a resonant line scatters more photons and increases the probability of reabsorption in the line, thus increasing the pumping of the fluorescence emission. The measurements by BVV show that most N ii permitted lines are blueshifted by 2 or 3 $`\mathrm{km}\mathrm{s}^1`$ with respect to the N ii forbidden lines, and support the idea that they are formed in different layers of the nebula.
We have chosen as a starting point the set of gas abundances of Baldwin et al. (1996), which were derived to fit forbidden line intensities with cloudy. The only exception was nitrogen for which we took a value of $`6\times 10^5`$ as an average from the determinations of Baldwin et al. (1996) and EPG. Differences in the abundances of the other elements between the two sets are not important in N ii permitted line intensities.
The N<sup>+</sup> column density remains remarkably constant with varying gas density for a given effective temperature of the star, $`T_{\mathrm{eff}}`$. For the $`T_{\mathrm{eff}}=37\mathrm{kK}`$, $`\mathrm{log}g=4`$ atmosphere, $`N(\mathrm{N}^+)=2.5\times 10^{16}\mathrm{cm}^2`$ for gas densities between 4,000 and 25,000$`\mathrm{cm}^3`$. Therefore we will adopt a fixed density of $`n=10^4\mathrm{cm}^3`$.
### 4.2 Stellar continuum and geometry
The blister model, in which $`\theta ^1\mathrm{C}\mathrm{Ori}`$ is near the wall of the OMCโ1 molecular cloud (Zuckerman, 1973; Balick et al., 1974), gives the most likely geometry for the nebula. The Lyman photon flux of the star at a distance $`r_0`$ from the illuminated face of the nebula is $`\varphi _0=Q_0/4\pi r_0^2`$, and can be constrained by the $`\mathrm{H}\beta `$ observed brightness up to a geometrical factor that depends on the angle of illumination $`\theta `$ of the slab of gas as described by Wen & OโDell (1995). The value $`\varphi =10^{12.9}\mathrm{cm}^2\mathrm{s}^1`$ and $`\theta =0`$, fixes the $`\mathrm{H}\beta `$ intensity around $`4\pi I=2.6\mathrm{erg}\mathrm{cm}^2\mathrm{s}^1`$ observed by Baldwin et al. (1991) if corrections are considered to account for reflected optical and absorbed UV radiations (Ferland, 2001). The $`T_{\mathrm{eff}}`$=37kK is too low for the probable spectral type of $`\theta ^1\mathrm{C}\mathrm{Ori}`$, but was chosen because it reproduces the absolute intensities of the fluorescence lines with respect to the $`\mathrm{H}\beta `$ flux and the N abundance measured from forbidden lines as discussed in section 5.1.
### 4.3 The recombination spectrum.
The 4f levels in N<sup>+</sup> are in an intermediate coupling between LK and jK couplings (Cowan & Andrew, 1965). In both couplings the total angular momentum is $`J=K\pm 1/2`$. The line fractions for either coupling can be obtained from Escalante & Gongora (1990). We employed an LK classification to use the recombination rates of EV.
Table 4 shows that all but two of the observed ratios of intensities of lines from 4f levels with respect to lines from 3d levels tend to be twice as strong and are closer to their predicted values by the recombination theory in PNe than in Orion. The two exceptions are the $`\lambda `$4242.49 line, which seems much stronger than its predicted recombination rate, and the line of the $`3\mathrm{d}^3\mathrm{D}^\mathrm{o}`$$`4\mathrm{f}\mathrm{G}[7/2]`$ multiplet, which violates the $`\mathrm{\Delta }L=0,\pm 1`$ selection rule in LK coupling due to strong mixing with the $`4\mathrm{f}\mathrm{F}[7/2]`$ term (Cowan & Andrew, 1965), and has no predicted recombination rate by EV. Liu et al. (2000, 2001) observed lines from other 4f terms in PNe, which have not been detected in Orion and are not listed in table 4. The higher intensity of the lines from 4f levels with respect to the lines from 3d levels in PNe indicates that fluorescence is less important relative to recombination in these objects because recombination can efficiently populate levels with high angular momentum while fluorescence is limited to resonant levels. One important exception to this trend is PN IC 418, which shows a strong enhancement of lines from 3p and 3d levels with respect to lines from the 4f levels (Sharpee et al., 2003). This nebula has a lower ionization level than most PNe that favors the excitation of N ii lines by fluorescence over recombination as in Orion, and will be the subject of a forthcoming paper.
Fluorescence can only contribute to population of the 4f levels through transitions from higher d levels, but this process is inefficient because the transition probabilities are much smaller than transitions to p levels. For example, fluorescence can populate the $`4\mathrm{f}\mathrm{F}`$ levels through absorptions in the multiplet $`2\mathrm{p}^2{}_{}{}^{3}\mathrm{P}`$$`5\mathrm{d}^3\mathrm{D}^\mathrm{o}`$$`\lambda `$453, but in a pure case B the $`5\mathrm{d}^3\mathrm{D}^\mathrm{o}`$ term decays to the terms $`3\mathrm{p}^3\mathrm{P}`$, $`3\mathrm{p}^3\mathrm{D}`$, $`2\mathrm{s}2\mathrm{p}^2(^4\mathrm{P})3\mathrm{s}^3\mathrm{P}`$, $`4\mathrm{p}^3\mathrm{P}`$ and $`4\mathrm{f}\mathrm{F}`$ among others with branching ratios $`P(5\mathrm{d},3\mathrm{p}^3\mathrm{P})=0.70`$, $`P(5\mathrm{d},3\mathrm{p}^3\mathrm{D})=0.14`$, $`P(5\mathrm{d},3\mathrm{s})=0.09`$, and $`P(5\mathrm{d},4\mathrm{f})=0.04`$. As shown in section 4.1, N<sup>+2</sup> abundances obtained from recombination rates of the 4f lines in Orion are 1.6 times the N abundance derived from collisionally excited lines. Suppose that the intensity of a 3dโ4f line were enhanced with respect to $`\mathrm{H}\beta `$ by fluorescence populating the 5d term with rate $`B(5\mathrm{d})`$:
$$\frac{I_\lambda }{I_{\mathrm{H}\beta }}=\frac{4861}{\lambda }P(\lambda )\frac{E+P(5d,4f)B(5\mathrm{d})}{E(\mathrm{H}\beta )}$$
(10)
where $`E(\mathrm{H}\beta )=\alpha _{\mathrm{H}\beta }^{\mathrm{eff}}n(\mathrm{H}^+)n_e๐\mathrm{}`$, $`E=\alpha _{4\mathrm{f}\mathrm{F}}^{\mathrm{eff}}n(\mathrm{N}^{+2})n_e๐\mathrm{}`$ and $`P(\lambda )`$ is the branching ratio of the 3dโ4f line with $`\lambda `$ given in ร
. To account for the excess abundance of 0.6 obtained from the effective recombination rate of multiplet $`3\mathrm{d}^3\mathrm{D}^\mathrm{o}`$$`4\mathrm{f}\mathrm{F}`$$`\lambda \lambda `$4236.91โ4247.20, we need $`B(5\mathrm{d})=0.6E/P(5\mathrm{d},4\mathrm{f})`$, which means a rate of population of the $`3\mathrm{p}^3\mathrm{P}`$ term of $`B(5\mathrm{d})P(5\mathrm{d},3\mathrm{p}^3\mathrm{P})=10.5E`$. The contribution to the intensity of a multiplet like $`3\mathrm{s}^3\mathrm{P}`$$`3\mathrm{p}^3\mathrm{P}`$$`\lambda \lambda `$4601.48โ4643.08 due to this additional excitation with $`\mathrm{N}^{+2}/\mathrm{H}^+6\times 10^5`$ would be $`B(5\mathrm{d})P(5\mathrm{d},3\mathrm{p}^3\mathrm{P})P(3\mathrm{p}^3\mathrm{P},3\mathrm{s}^3\mathrm{P}^\mathrm{o})/E(\mathrm{H}\beta )=7.7\times 10^4`$ where we took $`P(3\mathrm{p}^3\mathrm{P},3\mathrm{s}^3\mathrm{P}^\mathrm{o})=0.36`$ and an effective recombination coefficient of the $`4\mathrm{f}\mathrm{F}`$ term $`\alpha _{4\mathrm{f}\mathrm{F}}^{\mathrm{eff}}=1.0\times 10^{13}\mathrm{cm}^3\mathrm{s}^1`$ at $`10^4\mathrm{K}`$ (EV). The strongest line of the multiplet, $`3\mathrm{s}^3\mathrm{P}_2^\mathrm{o}`$$`3\mathrm{p}^3\mathrm{P}_2`$$`\lambda `$4630.54, has an LS line fraction of $`11.25/27`$ (Allen, 1973) and the corresponding increase in intensity would be at least 0.034 ($`I(\mathrm{H}\beta )=100`$), which is comparable the observed intensity of 0.048 (EPG) produced by more direct cascade routes following absorption of photons at $`\lambda \lambda `$530 and 534 ร
by 3d states. Thus the excitation of 4f states by fluorescence would produce lines from 3p and 3d states with intensities much higher than the observed values unless the stellar continuum had an unusual shape that selectively excited states above the 4f states. Our calculations show negligible contribution of fluorescence to the excitation of the 4f levels because the absorption rate is much less for higher resonant levels than for the 3d levels, and point to other mechanisms to excite them (Tsamis et al., 2004).
EPG also observed the lines at $`\lambda `$5001.14 and 5001.48 with upper levels $`3\mathrm{d}^3\mathrm{F}_{2,3}^\mathrm{o}`$. The most intense component of the multiplet at $`\lambda `$5005.15 is blended with the \[O iii\] line. As with the lines with upper 4f levels, the $`\lambda `$5005.15 line is mostly excited by recombination because its upper level $`3\mathrm{d}^3\mathrm{F}_4^\mathrm{o}`$ can receive only indirect contributions from the fluorescence excitation of higher levels. The other levels, $`3\mathrm{d}^3\mathrm{F}_{2,3}^\mathrm{o}`$, are connected to the ground state through weak dipoleโallowed transitions (Bell et al., 1995), and have a substantial fluorescence contribution. Our model calculations show that fluorescence contributes less than 5% to the intensity of the lines produced by the $`3\mathrm{d}^3\mathrm{F}_4^\mathrm{o}`$ and 4f levels in Orion, and it is not sufficient to explain the discrepancy between the abundances determined from recombination and collisionally excited lines.
### 4.4 The fluorescence spectrum
Table 5 gives the predicted intensities for the observed upper terms in Orion by BVV and EPG that are excited mostly by fluorescence. Although the two data sets are from different parts of the nebula, it is important to notice that both sets give similar measurements of the N ii permitted line intensities with respect to $`\mathrm{H}\beta `$. Uncertain observed intensities due to blends, low S/N or dubious identifications are marked with โ?โ in the table as indicated by those authors. We also added a question mark to the line at 4987.4 ร
, which is probably blended with the \[Fe III\]$`\lambda `$4987.20 line, and thus has an observed intensity much higher than our prediction. The line at 4994.4 ร
belonging to the same multiplet should be theoretically more intense, contrary to the observations.
The most intense fluorescence lines are triplets connected by resonant transitions to the ground term, $`2\mathrm{p}^2{}_{}{}^{3}\mathrm{P}`$. At temperatures characteristic of H ii regions, the fine structure populations of the ground term are approximately proportional to their statistical weight, and consequently the relative intensities of all the other triplet levels are given by the LS line fractions (Allen, 1973).
The lines at 5001.15 and 5001.48 ร
arising from the $`3\mathrm{d}^3\mathrm{F}^\mathrm{o}`$ term are blended. We split the total intensity according to the LS line fractions: 21.0:31.1, in order to compare them with our predictions in table 5.
Multiplet $`3\mathrm{p}^3\mathrm{P}`$$`4\mathrm{s}^3\mathrm{P}^\mathrm{o}`$ has many strong lines in the $`\lambda `$3838.37โ3856.06 ร
interval that were not detected because of the lower S/N at the blue end of the spectrum and because they are blended with other lines. Therefore they are not listed in table 5.
EPG marginally detected the $`3\mathrm{d}^3\mathrm{P}_2^\mathrm{o}`$$`4\mathrm{p}^3\mathrm{S}_1`$$`\lambda `$6809.99 ร
line. Lines from the 4p levels must be excited by cascades from upper levels in a fluorescence spectrum. We found very low intensities for all the lines from 4p levels. The most intense line of this type is $`2\mathrm{p}^3{}_{}{}^{3}\mathrm{P}_{2}^{\mathrm{o}}`$$`4\mathrm{p}^3\mathrm{S}_1`$$`\lambda `$1060.2 ร
with an intensity of $`4\times 10^5`$ with respect to $`\mathrm{H}\beta `$. Our predicted intensities of lines from 4p levels in the optical are below the instrumental sensitivity.
In table 5 we also list some singlets because BVV marginally detected the $`2\mathrm{p}^3{}_{}{}^{1}\mathrm{D}_{2}^{\mathrm{o}}`$$`3\mathrm{p}^1\mathrm{P}_1`$$`\lambda `$4895.11 ร
line. Singlets can be excited by absorptions from the $`{}_{}{}^{1}\mathrm{D}`$ and $`{}_{}{}^{1}\mathrm{S}`$ terms of the ground configuration, spinโforbidden transitions from the triplets and recombination. The quintet lines $`3\mathrm{p}^5\mathrm{D}_1^\mathrm{o}`$$`3\mathrm{d}^5\mathrm{P}_1`$$`\lambda `$4815.62 ร
, $`3\mathrm{s}^5\mathrm{P}_{1,3}`$$`3\mathrm{p}^5\mathrm{D}_{1,4}`$$`\lambda `$5535.36 ร
and $`3\mathrm{s}^5\mathrm{P}_3`$$`3\mathrm{p}^5\mathrm{D}_3`$$`\lambda `$5551.99 ร
measured by BVV and EPG are not listed. These lines have the $`2\mathrm{s}2\mathrm{p}^2{}_{}{}^{4}\mathrm{P}`$ excited core configuration, and can only be populated with transitions in which the LS coupling breaks down. The $`3\mathrm{d}^5\mathrm{P}`$ term is autoionizing, and the most likely mechanism in this case is dielectronic recombination. EPG reported lines at $`\lambda \lambda `$6744.42, 7535.32 and 9016.42 ร
that would be produced by highly excited d and f states of N<sup>+</sup>. Their identification as N ii, however, is uncertain and their production in our model is very unlikely. Many more lines are predicted by our calculations, but their intensities are lower than the weaker lines reported by BVV or EPG, and their intensities are dominated by recombination. Intensities for other lines up to $`n=8`$ and $`l=1`$ are available from the authors upon request. Most of the observed lines with confident measurements fall within 0.2 dex of the predicted values in the basic model as shown in the upper panels of Fig. 3.
## 5 Variation of parameters
### 5.1 Stellar temperature
Fig. 3 shows a comparison of observed and predicted intensities with two WMbasic atmospheres. The reduction in N<sup>+</sup> column density produced by the hotter star reduces the predicted intensities in general, but the intensities of the lines originating from d states suffer a much greater reduction than the ones from p states by an order of magnitude. We have traced this effect to an important decrease in the model atmosphere flux by a factor of $`2`$ at 529.6 ร
and a factor of $`5`$ at 533.7 ร
for $`T_{\mathrm{eff}}38\mathrm{kK}`$ as shown in Fig. 4. As mentioned in section 2.1, 3d states are pumped almost entirely by absorptions at those two wavelengths. At the same time there is an increase of a factor of $`2`$ in the flux at 508.7ร
for $`T_{\mathrm{eff}}37\mathrm{kK}`$, which is important in the pumping of 3p states, and compensates the decrease in the N<sup>+</sup> column density.
The difference in the pumping rates of d and p states for $`T_{\mathrm{eff}}38\mathrm{kK}`$ depends only on the shape of the spectrum and the contribution of recombination to the population of those states. Thus the disagreement between predicted and observed intensities of lines from d states with $`T_{\mathrm{eff}}=39\mathrm{kK}`$ shown in Fig. 3 persists when the N abundance is increased to $`1\times 10^4`$ or when the recombination contribution to the intensities is increased with the hypothesis of ultracold plasma proposed by Tsamis et al. (2004). A plasma temperature of $`2000\mathrm{K}`$ doubles the predicted intensities of lines from 3p states, but lines from 3d states increase their intensities in lower proportions because fluorescence is more important in the population of those states.
The other critical parameter in the fluorescence line intensities is the column density. The N<sup>+</sup> column density decreases with $`T_{\mathrm{eff}}`$ at a much lower rate for $`T_{\mathrm{eff}}38\mathrm{kK}`$ as reflected in Fig. 5. The NII fluorescence lines and the \[N ii\] lines decrease little for harder spectra due to that persistent N<sup>+</sup> concentration, but their different behaviour at lower $`T_{\mathrm{eff}}`$ can be understood in terms of the N<sup>+</sup> concentration and the escape probability concept. As shown in Fig. 2, the fluorescence N ii lines form 50% of their intensity in the inner layers of the nebula, much closer to the star than the \[N ii\] lines, which are produced in the outer N<sup>+</sup> zone. As $`T_{\mathrm{eff}}`$ decreases, the N<sup>+</sup> concentration and the optical depth of the resonant transitions increase, but the intensities of lines from p and d states behave differently as shown in Fig. 5 for the lines $`3\mathrm{s}^3\mathrm{P}_2^\mathrm{o}`$$`3\mathrm{p}^3\mathrm{P}_2`$$`\lambda `$4630.54 and $`3\mathrm{p}^3\mathrm{P}_2`$$`3\mathrm{d}^3\mathrm{D}_3^\mathrm{o}`$$`\lambda `$5941.65. Absorption transitions that populate the p states have a much lower optical depth than the ones pumping the d states. As a result the escape probability decreases more for d states than for p states with lower $`T_{\mathrm{eff}}`$, and the pumping due to reabsorption of resonant photons for d states increases.
nebu tends to give larger column densities than cloudy, and thus predicts higher intensities. Predicted line intensities by nebu in the UV tend to be 30% more intense than those given by cloudy because nebu does not consider internal dust extinction, but predictions of the two codes are within 20% of each other in the optical.
### 5.2 Kurucz atmospheres
cloudy contains a grid of lowโresolution Kurucz atmospheres (Kurucz, 1991) that can be readily used as continua in our calculations. A comparison of Fig. 3 and 6 shows that the differences between the calculated intensities of p and d states with Kurucz atmospheres are much smaller than the differences with the WMbasic atmospheres because the Kurucz atmospheres do not have the structure of the WMbasic atmospheres that causes the different absorption rates between p and d states.
Modeling of Orion with cloudy (Baldwin et al., 1991, 1996, 2000) has favored stellar temperatures that are lower than current spectroโphotometric measurements. Our results with the WMbasic atmospheres also favor a lower $`T_{\mathrm{eff}}`$, but Kurucz atmospheres give a better agreement with observations because they are softer than other models and give a larger N<sup>+</sup> column.
### 5.3 Stellar flux
Unlike most forbidden and recombination lines, fluorescence line intensities are more sensitive to changes in the stellar flux illuminating the nebula. Their intensities increase with $`\varphi _0`$ up to $`10^{12.5}\mathrm{cm}^2\mathrm{s}^1`$ and remain nearly constant for higher $`\varphi _0`$. This behavior can be understood in similar terms to the curve of growth of the resonant lines. As the intensity of the ionizing flux grows, the N<sup>+</sup> column density and the optical depth increase, and the cores of the resonant lines become saturated. Eqs. 2 and 9 show that the intensity of the fluorescence lines is approximately proportional to the integral along the line of sight of the pumping rate of equation (2) times the density of the absorbing state, $`n_g\beta _{gj}`$. Changing variable from $`r`$ to $`\tau _0`$, eliminating constant quantities and assuming a constant Doppler width, the intensity of a fluorescence line will be
$`I{\displaystyle _0^{\mathrm{}}}\overline{J}_\nu {\displaystyle e^{\tau _0\varphi (x)}\varphi (x)๐x๐\tau _0}`$
$`=J_\nu (0){\displaystyle _0^{\mathrm{}}}e^{\tau _c}{\displaystyle e^{\tau _0\varphi (x)}\varphi (x)๐x๐\tau _0}.`$
where $`J_\nu (0)`$ is the stellar continuum at the illuminated face of the cloud, and $`\tau _c`$ is the continuum opacity. The integration over $`\tau _0`$ can be performed exactly if we assume a mean value for $`e^{\tau _c}`$. For fixed $`T_{\mathrm{eff}}`$ and $`\nu `$, $`J_\nu (0)`$ is proportional to $`\varphi _0`$, which in turn is proportional to the $`\mathrm{H}\beta `$ flux. Therefore the fluorescence line intensity normalized to $`\mathrm{H}\beta `$ must be proportional to
$$e^{\tau _c}(1e^{\tau _0\varphi (x)})๐x.$$
The integral is proportional to the curve of growth $`W(\tau _0)`$. Fig. 7 shows that the intensity of the lines follows closely a fit of the form
$$I/I(\mathrm{H}\beta )e^{2.5\times 10^{22}N(\mathrm{H}^+)}W(\tau _0).$$
(11)
where $`N(\mathrm{H}^+)`$ is the H<sup>+</sup> column density in $`\mathrm{cm}^2`$ and $`W(\tau _0)`$ is the curve of growth of the $`2\mathrm{p}^2{}_{}{}^{3}\mathrm{P}_{2}^{}`$$`4\mathrm{s}^3\mathrm{P}_2^\mathrm{o}`$$`\lambda `$508.697 ร
line, which pumps most of the 4630.54 line.
## 6 Conclusions
The intensity of the lines in the N ii spectrum of the Orion nebula can be explained by fluorescence of the UV radiation of $`\theta ^1\mathrm{C}\mathrm{Ori}`$ in the ionized gas. Recombination of N<sup>+2</sup> contributes a minor part of the observed intensities of lines from 3p and 3d levels connected to the ground state. The effective temperature of the star must be below 38000 K in order to reproduce the observed line intensities with typical ionization models that are consistent with the forbidden line intensities. An increased N abundance does not allow the use of a higher star temperature. The existence of intervening ionized material in the foreground (OโDell et al., 1993) was not considered in our model and may help increase the predicted intensities of the N ii lines. Fluorescence does not increase the intensity of the lines from 4f levels, and other mechanisms must be proposed to explain their strong intensities with respect to the collisionally excited and fluorescence lines in the Orion nebula.
## Acknowlegments
The authors are very grateful to Katia Verner and Gary Ferland for valuable advice in running cloudy.
|
warning/0506/hep-ph0506154.html
|
ar5iv
|
text
|
# Neutral Higgs Bosons in the SU(3){_๐ฟ}โU(1)N model
## I Introduction
The Higgs sector still remains one of the most indefinite part of the standard model wsg , but it still represents a fundamental rule by explaining how the particles gain masses by means of a isodublet scalar field, which is responsible for the spontaneous breakdown of the gauge symmetry, the process by which the spectrum of all particles are generated. This process of mass generation is the so called Higgs mechanism, which plays a central role in gauge theories. In this process there remains a single neutral scalar, manifesting itself as the Higgs particle. In the standard model only one SU(2) Higgs doublet is necessary and enough to break the gauge symmetry and to generate the particles masses. However, the standard model does not predicts the number of scalar multiplets of the theory, for that reason, there are several extensions of the standard model containing neutral and charged Higgs bosons. The standard model is not able to predicts the mass of the Higgs boson. However, indirect experimental limits are obtained from precision measurements of the electroweak parameters. These measurements are now realized in radiative correction levels, which have a logarithmic dependence of standard Higgs boson mass. From several experiments the present value for the standard Higgs boson mass is $`126_{48}^{+73}`$ GeV lep .
Since the standard model leaves many questions open, there are several well motivated extensions of it. For example, if the Grand Unified Theory (GUT) contains the standard model at high energies, then the Higgs bosons associated with GUT symmetry breaking must have masses of order $`M_X๐ช(10^{15})`$ GeV. Supersymmetry supers provides a solution to this hierarchy problem through the cancellation of the quadratic divergences via the contributions of fermionic and bosonic loops cancell . Moreover, the Minimal Supersymmetric extension of the Standard Model (MSSM) can be derived as an effective theory from supersymmetric Grand Unified Theories sgut . Another promissory class of models is the one based on the $`SU(3)_CSU(3)_LU(1)_N`$ (3-3-1 for short) semi-simple symmetry group PT93 .
These models emerge as an alternative solution to the problem of violation of unitarity at high energies in processes such as $`e^{}e^{}W^{}V^{}`$, induced by right-handed currents coupled to a vector boson $`V^{}`$. The usual way to circumvent this problem is to give particular values to model parameters in order to cancel the amplitude of the process PP92 , but in PP92 was proposed an elegant solution assuming the presence of a doubly charged vector boson. The simplest electroweak gauge model that is able to realize naturaly a double charge gauge boson is the one based on the SU(3)$``$U(1) symmetry PP92 . As a consequence of the extended gauge symmetry, the model is compelled to accommodate a much richer Higgs sector.
The main feature of the 3-3-1 model is that it is able to predicts the correct number of fermions families. This is because, contrary to the standard model, the 3-3-1 model is anomalous in each generation. The anomalies are cancelled only if the number of families is a multiple of three. In addition, if we take into account that the asymptotic freedom condition of the QCD is valid only if the number of generations of quarks is to be less than five, we conclude that the number of generations is three LS01 . Another good feature is that the model predicts an upper bound for the Weinberg mixing angle at $`\mathrm{sin}^2\theta _W<1/4`$. Therefore, the evolution of $`\theta _W`$ to high values leads to an upper bound to the new mass scale between 3 TeV and 4 TeV JJ97 .
In this work we are interested in a version of the 3-3-1 model, whose scalar sector has only three Higgs triplets PT93 . The text is organized as follow. In Sec. II we give the relevant features of the model. In Sec. III we compute the total cross sections of the processes and the Sec. IV contains our results and conclusions.
## II Basic facts about the 3-3-1 model
The three Higgs triplets of the model are
$$\eta =\left(\begin{array}{c}\eta ^0\\ \eta _1^{}\\ \eta _2^+\end{array}\right),\rho =\left(\begin{array}{c}\rho ^+\\ \rho ^0\\ \rho ^{++}\end{array}\right),\chi =\left(\begin{array}{c}\chi ^{}\\ \chi ^{}\\ \chi ^0\end{array}\right)$$
(1)
transforming as $`(\mathrm{๐},0)`$, $`(\mathrm{๐},1)`$ and $`(\mathrm{๐},1)`$, respectively.
The neutral scalar fields develop the vacuum expectation values (VEVs) $`\eta ^0v_\eta `$, $`\rho ^0v_\rho `$ and $`\chi ^0v_\chi `$, with $`v_\eta ^2+v_\rho ^2=v_W^2=(246\text{ GeV})^2`$. The pattern of symmetry breaking is $`\text{SU(3)}_L\text{U(1)}_N\stackrel{\chi }{}\text{SU(2)}_L\text{U(1)}_Y\stackrel{\eta ,\rho }{}\text{U(1)}_{\mathrm{em}}`$ and so, we can expect $`v_\chi v_\eta ,v_\rho `$. The $`\eta `$ and $`\rho `$ scalar triplets give masses to the ordinary fermions and gauge bosons, while the $`\chi `$ scalar triplet gives masses to the new fermions and new gauge bosons. The most general, gauge invariant and renormalizable Higgs potential is
$`V(\eta ,\rho ,\chi )`$ $`=`$ $`\mu _1^2\eta ^{}\eta +\mu _2^2\rho ^{}\rho +\mu _3^2\chi ^{}\chi +\lambda _1\left(\eta ^{}\eta \right)^2+\lambda _2\left(\rho ^{}\rho \right)^2+\lambda _3\left(\chi ^{}\chi \right)^2+`$ (5)
$`+\left(\eta ^{}\eta \right)\left[\lambda _4\left(\rho ^{}\rho \right)+\lambda _5\left(\chi ^{}\chi \right)\right]+\lambda _6\left(\rho ^{}\rho \right)\left(\chi ^{}\chi \right)+\lambda _7\left(\rho ^{}\eta \right)\left(\eta ^{}\rho \right)+`$
$`+\lambda _8\left(\chi ^{}\eta \right)\left(\eta ^{}\chi \right)+\lambda _9\left(\rho ^{}\chi \right)\left(\chi ^{}\rho \right)+\lambda _{10}\left(\eta ^{}\rho \right)\left(\eta ^{}\chi \right)+`$
$`{\displaystyle \frac{1}{2}}\left(fฯต^{ijk}\eta _i\rho _j\chi _k+\text{H. c.}\right).`$
Here $`\mu _i`$ $`\left(i=1,2,3\right)`$, $`f`$ are constants with dimension of mass and the $`\lambda _i`$, $`\left(i=1,\mathrm{},10\right)`$ are dimensionalless constants. $`f`$ and $`\lambda _3`$ are negative from the positivity of the scalar masses. The term proportional to $`\lambda _{10}`$ violates lepto-barionic number, therefore it was not considered in the analysis of the Ref. TO96 (another analysis of the 3-3-1 scalar sector are given in Ref. AK00 and references cited therein). We can notice that this term contributes to the mass matrices of the charged scalar fields, but not to the neutral ones. However, it can be checked that in the approximation $`v_\chi v_\eta ,v_\rho `$ we can still work with the masses and eigenstates given in Ref. TO96 . Here this term is important to the decay of the lightest exotic fermion. Therefore, we will keep it in the Higgs potential (5).
As usual, symmetry breaking is implemented by shifting the scalar neutral fieldas $`\phi =v_\phi +\xi _\phi +i\zeta _\phi `$, with $`\phi `$ $`=`$ $`\eta ^0`$, $`\rho ^0`$, $`\chi ^0`$. Thus, the physical neutral scalar eigenstates $`H_1^0`$, $`H_2^0`$, $`H_3^0`$ and $`h^0`$ are related to the shifted fields as
$$\left(\begin{array}{c}\xi _\eta \\ \xi _\rho \end{array}\right)\frac{1}{v_W}\left(\begin{array}{cc}v_\eta & v_\rho \\ v_\rho & v_\eta \end{array}\right)\left(\begin{array}{c}H_1^0\\ H_2^0\end{array}\right),\xi _\chi H_3^0,\zeta _\chi h^0,$$
(6)
and in the charged scalar sector we have
$`\eta _1^+={\displaystyle \frac{v_\rho }{v_W}}H_1^+,\rho ^+={\displaystyle \frac{v_\eta }{v_W}}H_1^+,\eta _2^+={\displaystyle \frac{v_\chi }{\sqrt{v_\eta ^2+v_\chi ^2}}}H_2^+,\chi ^+={\displaystyle \frac{v_\eta }{\sqrt{v_\eta ^2+v_\chi ^2}}}H_2^+,`$ (7)
$`\rho ^{++}={\displaystyle \frac{v_\chi }{\sqrt{v_\rho ^2+v_\chi ^2}}}H^{++},\chi ^{++}={\displaystyle \frac{v_\rho }{v_\chi }}H^{++},`$ (8)
with the condition that $`v_\chi v_\eta ,v_\rho `$ TO96 .
The content of matter fields form the three SU(3)<sub>L</sub> triplets
$$\psi _{aL}=\left(\begin{array}{c}\nu _\mathrm{}a^{}\\ \mathrm{}_a^{}\\ P_a^{}\end{array}\right),Q_{1L}=\left(\begin{array}{c}u_1^{}\\ d_1^{}\\ J_1\end{array}\right),Q_{\alpha L}=\left(\begin{array}{c}J_\alpha ^{}\\ u_\alpha ^{}\\ d_\alpha ^{}\end{array}\right),$$
(9)
transform as $`(\mathrm{๐},0)`$, $`(\mathrm{๐},2/3)`$ and $`(\mathrm{๐}^{},1/3)`$, respectively, where $`\alpha =2,3`$. In Eqs. (9) $`P_a`$ are heavy leptons, $`\mathrm{}_a^{}=e^{},\mu ^{},\tau ^{}`$. The model also predicts the exotic $`J_1`$ quark, which carries $`5/3`$ units of elementary electric charge and $`J_2`$ and $`J_3`$ with $`4/3`$ each. The numbers $`0`$, $`2/3`$ and $`1/3`$ in Eqs. (9) are the U<sub>N</sub> charges. We also have the right-handed counterpart of the left-handed matter fields, $`\mathrm{}_R^{}(\mathrm{๐},1)`$, $`P_R^{}(\mathrm{๐},1)`$, $`U_R^{}(\mathrm{๐},2/3)`$, $`D_R^{}(\mathrm{๐},1/3)`$, $`J_{1R}^{}(\mathrm{๐},5/3)`$ and $`J_{2,3R}^{}(\mathrm{๐},4/3)`$, where $`U=u,c,t`$ and $`D=d,s,b`$ for the ordinary quarks.
The Yukawa Lagrangians that respect the gauge symmetry are
$`_{\mathrm{}}^Y`$ $`=`$ $`G_{ab}\overline{\psi _{aL}}\mathrm{}_{bR}^{}G_{ab}^{}\overline{\psi _{aL}^{}}P^{}\chi +\text{H. c.},`$ (10)
$`_q^Y`$ $`=`$ $`{\displaystyle \underset{a}{}}\left[\overline{Q_1L}\left(G_{1a}U_{aR}^{}\eta +\stackrel{~}{G}_{1a}D_{aR}^{}\rho \right)+{\displaystyle \underset{\alpha }{}}\overline{Q_{\alpha L}}\left(F_{\alpha a}U_{aR}^{}\rho ^{}+\stackrel{~}{F}_{\alpha a}D_{aR}^{}\eta ^{}\right)\right]+`$ (12)
$`+{\displaystyle \underset{\alpha \beta }{}}F_{\alpha \beta }^J\overline{Q_{\alpha J}}J_{\beta R}^{}\chi ^{}+G^J\overline{Q_{1L}}J_{1R}+\text{ H. c.}`$
Here, the $`G`$โs, $`\stackrel{~}{G}`$โs, $`F`$โs and $`\stackrel{~}{F}`$โs are Yukawa coupling constants with $`a,b=1,2,3`$ and $`\alpha =2,3`$.
It should be noticed that the ordinary quarks couple only through $`H_1^0`$ and $`H_2^0`$. This is because these physical scalar states are linear combinations of the interactions eigenstates $`\eta `$ and $`\rho `$, which break the SU(2)<sub>L</sub>$``$U(1)<sub>Y</sub> symmetry to U(1)<sub>em</sub>. On the other hand the heavy-leptons and quarks couple only through $`H_3^0`$ and $`h^0`$ in scalar sector, i. e., throught the Higgs that induces the symmetry breaking of SU(3) <sub>L</sub>$``$U(1)<sub>N</sub> to SU(2)<sub>L</sub>$``$U(1)<sub>Y</sub>. The Higgs particle spectrum consists of seven states: three scalars $`(H_1^0,H_2^0,H_3^0)`$, one neutral pseudoscalar $`h^0`$ and three charged Higgs bosons, $`H_1^+`$, $`H_2^+`$ and $`H^{++}`$.
In this work we study the production of a neutral Higgs boson at $`e^{}e^+`$ colliders because of lower backgrounds and since it is one of the most promising in the search for the Higgs. The Higgs $`H_i`$, where $`i=1,2`$ can be radiated from a $`Z`$ and $`Z^{}`$ boson. The $`Z(Z^{})ZH_1(H_2)`$ process is the dominant mechanism at the $`Z`$ resonance energy. We discuss this process only for on-shell $`Z`$ production. In this work, we will study the production mechanism for Higgs particles in $`e^+e^{}`$ colliders such as the Next Linear Collider (NLC) ($`\sqrt{s}=500`$ GeV) and CERN Linear Collider (CLIC) ($`\sqrt{s}=1000`$ GeV).
## III Cross section production
The main mechanism for the production of Higgs particles in $`e^+e^{}`$ collisions occurs in association with the boson $`Z`$, and $`Z^{}`$ through the Drell-Yan mechanism. The process $`e^+e^{}H_iZ`$ $`(i=1,2)`$ takes place through the exchange of bosons $`Z`$ and $`Z^{}`$ in the $`s`$ channel. Then using the interaction Lagrangian (12) and (5) we obtain the differential cross section
$`{\displaystyle \frac{d\widehat{\sigma }}{d\mathrm{cos}\theta }}`$ $`=`$ $`{\displaystyle \frac{\beta _H\alpha ^2\pi }{32\mathrm{sin}_{\theta _W}^4\mathrm{cos}_{\theta _W}^2s}}\times `$ (15)
$`\times [{\displaystyle \frac{cZH_i^2}{\left(sM_Z^2+iM_Z\mathrm{\Gamma }_Z\right)^2}}(2M_Z^2+{\displaystyle \frac{2tu}{M_Z^2}}2t2u+2s)(g_V^{e^2}+g_A^{e^2})+`$
$`+{\displaystyle \frac{cZ^{}H_i^2}{\left(sM_Z^2+iM_Z^{}\mathrm{\Gamma }_Z^{}\right)^2}}\left(2M_Z^2+{\displaystyle \frac{2tu}{M_Z^2}}2t2u+2s\right)\left(g_V^{}^{e^2}+g_A^{}^{e^2}\right)+`$
$`+{\displaystyle \frac{2cZH_icZ^{}H_i}{\left(sM_Z^2+iM_Z\mathrm{\Gamma }_Z\right)\left(sM_Z^{}^2+iM_Z^{}\mathrm{\Gamma }_Z^{}\right)}}\times `$
$`\times (2M_Z^2+{\displaystyle \frac{2tu}{M_Z^2}}2t2u+2s)(g_V^eg_V^{}^e+g_A^eg_A^{}^e)].`$
The primes $`({}_{}{}^{})`$ are for the case when we take a $`Z^{}`$ boson, $`\mathrm{\Gamma }_Z`$ and $`\mathrm{\Gamma }_Z^{}`$ cieto , are the total width of the $`Z`$ and $`Z^{}`$ boson, $`g_{V,A}^e`$ are the standard lepton coupling constants, $`g_{V^{},A^{}}^e`$ are the $`331`$ lepton coupling constants, $`\sqrt{s}`$ is the center of mass energy of the $`e^{}e^+`$ system. For the $`Z^{}`$ boson we take $`M_Z^{}=\left(0.53\right)`$ TeV, since $`M_Z^{}`$ is proportional to the VEV $`v_\chi `$ PP92 ; FR92 . For the standard model parameters we assume PDG values, i. e., $`M_Z=91.19`$ GeV, $`\mathrm{sin}^2\theta _W=0.2315`$, and $`M_W=80.33`$ GeV Cea98 , the velocity of the Higgs in the center of mmass (CM) of the process is denoted through $`\beta _H`$, t and u are the kinemetic invariants, the $`cZZH_i^0(cZZ^{}H_i^0)`$ are the coupling constants of the $`Z`$ boson to $`Z(Z^{})`$ bosons and Higgs $`H_i^0`$ where i stands for $`H_1^0,H_2^0`$, the cHi0VPM are the coupling constants of the $`H_i^0`$, where i= 1,2, to $`V^{}V^+`$, the cHi0UPP are the coupling constants of the $`H_i^0`$, where i=1,2, to $`U^{}U^{++}`$, and the cHi0HPP are the coupling constants of the $`H_i^0`$, where i=1,2, to $`H^{}H^{++}`$. We then have that
$`t`$ $`=`$ $`M_Z^2{\displaystyle \frac{s}{2}}\{1+{\displaystyle \frac{M_Z^2M_H^2}{s}}+`$ (17)
$`\mathrm{cos}\theta \left[(1{\displaystyle \frac{\left(M_Z+M_H\right)^2}{s}})(1{\displaystyle \frac{\left(M_ZM_H\right)^2}{s}})\right]^{1/2}\},`$
$`u`$ $`=`$ $`M_H^2{\displaystyle \frac{s}{2}}\{1{\displaystyle \frac{M_Z^2M_H^2}{s}}+`$ (19)
$`+\mathrm{cos}\theta \left[(1{\displaystyle \frac{\left(M_Z+M_H\right)^2}{s}})(1{\displaystyle \frac{\left(M_ZM_H\right)^2}{s}})\right]^{1/2}\},`$
$`cZZ^{}H_1^0`$ $`=`$ $`i{\displaystyle \frac{g^2}{2\sqrt{3}v_W}}{\displaystyle \frac{M_Z}{M_W}}{\displaystyle \frac{\left[v_\eta ^2\left(6t_W^2+1\right)v_\rho ^2\right]}{\sqrt{1+3t_W^2}}},`$ (20)
$`cZZ^{}H_2^0`$ $`=`$ $`i{\displaystyle \frac{g^2}{\sqrt{3}}}{\displaystyle \frac{v_\eta v_\rho }{v_W}}\sqrt{1+4t_W^2},cH10VPM=ig^2{\displaystyle \frac{v_\rho ^2}{v_W}},`$ (21)
$`cH20VPM`$ $`=`$ $`ig^2{\displaystyle \frac{v_\eta v_\rho }{v_W}},cH10UPP=ig^2{\displaystyle \frac{v_\eta ^2}{v_W}},cH20UPP=ig^2{\displaystyle \frac{v_\eta v_\rho }{v_W}},`$ (22)
$`cH10HPP`$ $`=`$ $`i{\displaystyle \frac{2\left[\left(\lambda _6+\lambda _9\right)v_\eta ^4+\left(2\lambda _2+\lambda _9\right)v_\eta ^2v_\chi ^2+\left(\lambda _4+\lambda _5\right)v_\eta ^2v_\chi ^2\right]fv_\eta v_\rho v_\chi }{v_W\left(v_\eta ^2+v_\chi ^2\right)}},`$ (23)
$`cH20PP`$ $`=`$ $`iv_\eta {\displaystyle \frac{2v_\rho \left[\left(2\lambda _2\lambda _4+\lambda _9\right)v_\chi ^2+\left(\lambda _6\lambda _5+\lambda _9\right)v_\eta ^2\right]fv_\eta v_\chi }{v_W\left(v_\eta ^2+v_\chi ^2\right)}},`$ (24)
where $`\theta `$ is the angle between the Higgs and the incident electron in the CM frame, where the coupling constant of the $`Z`$ boson to $`Z`$ and $`H_1^0`$ are the standard ones and the coupling constant of the $`Z`$ boson to $`Z`$ and $`H_2^0`$ does not exist.
The total width of the Higgs $`H_1^0`$ into quarks, leptons, $`W^+W^{}`$, $`ZZ`$, $`ZZ^{}`$, $`Z^{}Z^{}`$ gauge bosons, $`H_2^0H_2^0`$, $`H_1^{}H_1^+`$, $`H_2^{}H_2^+`$, $`h^0h^0`$, $`H_2^0H_3^0`$ Higgs bosons, $`V^{}V^+`$ charged bosons, $`U^{}U^{++}`$ double charged bosons, $`H_2^0Z`$, $`H_2^0Z^{}`$ bosons and $`H^{}H^{++}`$ double charged Higgs bosons, are, respectively, given by
$`\mathrm{\Gamma }\left(H_1^0\mathrm{all}\right)`$ $`=`$ $`\mathrm{\Gamma }_{H_1^0q\overline{q}}+\mathrm{\Gamma }_{H_1^0\mathrm{}^{}\mathrm{}^+}+\mathrm{\Gamma }_{H_1^0W^+W^{}}+\mathrm{\Gamma }_{H_1^0ZZ}+\mathrm{\Gamma }_{H_1^0Z^{}Z}+\mathrm{\Gamma }_{H_1^0Z^{}Z^{}}+`$ (25)
$`+\mathrm{\Gamma }_{H_1^0H_2^0H_2^0}+\mathrm{\Gamma }_{H_1^0H_1^{}H_1^+}+\mathrm{\Gamma }_{H_1^0H_2^{}H_2^+}+\mathrm{\Gamma }_{H_1^0h^0h^0}+\mathrm{\Gamma }_{H_1^0H_2^0H_3^0}+`$
$`\mathrm{\Gamma }_{H_1^0V^{}V^+}+\mathrm{\Gamma }_{H_1^0U^{}U^+}+\mathrm{\Gamma }_{H_1^0H_2^0Z}+\mathrm{\Gamma }_{H_1^0H_2^0Z^{}}+\mathrm{\Gamma }_{H_1^0H^{}H^{++}},`$
where we have for each the widths given above that
$`\mathrm{\Gamma }_{H_1^0q\overline{q}}`$ $`=`$ $`{\displaystyle \frac{3\sqrt{14M_q^2/M_{H_1^0}^2}}{16\pi M_{H_1^0}}}{\displaystyle \frac{M_q^2}{v_W^2}}(M_{H_1^0}^22M_q^2),`$ (26)
$`\mathrm{\Gamma }_{H_1^0\mathrm{}\mathrm{}+}`$ $`=`$ $`{\displaystyle \frac{\sqrt{14M_{\mathrm{}}^2/M_{H_1^0}^2}}{16\pi M_{H_1^0}}}{\displaystyle \frac{M_{\mathrm{}}^2}{v_W^2}}(M_{H_1^0}^22M_{\mathrm{}}^2),`$ (27)
$`\mathrm{\Gamma }_{H_1^0WW+}`$ $`=`$ $`{\displaystyle \frac{\sqrt{14M_W^2/M_{H_1^0}^2}}{8\pi }}{\displaystyle \frac{g^2M_W^2}{M_{H_1^0}}}\left(3{\displaystyle \frac{M_{H_1^0}^2}{M_W^2}}+{\displaystyle \frac{M_{H_1^0}^4}{4M_W^4}}\right),`$ (28)
$`\mathrm{\Gamma }_{H_1^0ZZ}`$ $`=`$ $`{\displaystyle \frac{\sqrt{14M_Z^2/M_{H_1^0}^2}}{8\pi \mathrm{cos}_{\theta _W}^2}}{\displaystyle \frac{g^2M_Z^2}{M_{H_1^0}}}\left(3{\displaystyle \frac{M_{H_1^0}^2}{M_Z^2}}+{\displaystyle \frac{M_{H_1^0}^4}{4M_Z^4}}\right),`$ (29)
$`\mathrm{\Gamma }_{H_i^0Z^{}Z}`$ $`=`$ $`{\displaystyle \frac{\sqrt{1\left(\frac{M_Z^{}+M_Z}{M_{H_i^0}}\right)^2}\sqrt{1\left(\frac{M_Z^{}M_Z}{M_{H_i^0}}\right)^2}}{8\pi M_{H_i^0}}}(cZZ^{}H_i^0)^2\times `$ (31)
$`\times \left({\displaystyle \frac{5}{2}}+{\displaystyle \frac{1}{4}}{\displaystyle \frac{M_Z^2}{M_Z^{}^2}}+{\displaystyle \frac{1}{4}}{\displaystyle \frac{M_Z^{}^2}{M_Z^2}}+{\displaystyle \frac{1}{4}}{\displaystyle \frac{M_{H_i^0}^4}{M_Z^{}^2M_Z^2}}{\displaystyle \frac{1}{2}}{\displaystyle \frac{M_{H_i^0}^2}{M_Z^2}}{\displaystyle \frac{1}{2}}{\displaystyle \frac{M_{H_i^0}^2}{M_Z^{}^2}}\right),`$
$`\mathrm{\Gamma }_{H_1^0Z^{}Z^{}}`$ $`=`$ $`{\displaystyle \frac{\sqrt{14M_Z^2/M_{H_1^0}^2}g^4(1+3t_W^2)^2(12t_W^2v_\eta ^2(1+3t_W^2)+v_\chi ^2)^2}{576\pi v_\chi ^2M_{H_1^0}}}\times `$ (33)
$`\times \left(3{\displaystyle \frac{M_{H_1^0}^2}{M_Z^{}^2}}+{\displaystyle \frac{1}{4}}{\displaystyle \frac{M_{H_1^0}^4}{M_Z^{}^4}}\right),`$
$`\mathrm{\Gamma }_{H_1^0H_2^0H_2^0}`$ $`=`$ $`{\displaystyle \frac{\sqrt{14M_{H_2^0}^2/M_{H_1^0}^2}}{4\pi M_{H_1^0}}}\left({\displaystyle \frac{\lambda _4(v_\eta ^4+v_\rho ^4)+2v_\eta ^2v_\rho ^2(2\lambda _4+2\lambda _2+3\lambda _1)}{v_\chi ^3}}\right)^2,`$ (34)
$`\mathrm{\Gamma }_{H_1^0H_1^{}H_1^+}`$ $`=`$ $`{\displaystyle \frac{\sqrt{14M_{H_1^\pm }^2/M_{H_1^0}^2}}{4\pi M_{H_1^0}}}\left({\displaystyle \frac{(\lambda _4+\lambda _7)(v_\eta ^4+v_\rho ^4)+2v_\eta ^2v_\rho ^2(\lambda _1+\lambda _2+\lambda _7)}{v_\chi ^3}}\right)^2,`$ (35)
$`\mathrm{\Gamma }_{H_1^0H_2^{}H_2^+}`$ $`=`$ $`{\displaystyle \frac{\sqrt{14M_{H_2^\pm }^2/M_{H_1^0}^2}}{16\pi M_{H_1^0}}}\times `$ (37)
$`\times \left({\displaystyle \frac{(\lambda _4v_\chi ^2+\lambda _6v_\rho ^2)v_\eta ^22(\lambda _5+\lambda _8)v_\rho ^42(2\lambda _1+\lambda _8)v_\rho ^2v_\chi ^2+fv_\eta v_\rho v_\chi }{v_W(v_\rho ^2+v_\chi ^2)}}\right)^2,`$
$`\mathrm{\Gamma }_{H_1^0h_0h_0}`$ $`=`$ $`{\displaystyle \frac{\sqrt{14M_{h_0}^2/M_{H_1^0}^2}}{4\pi M_{H_1^0}}}\left({\displaystyle \frac{\lambda _5v_\rho ^2+\lambda _6v_\eta ^2}{v_\chi }}\right)^2,`$ (38)
$`\mathrm{\Gamma }_{H_i^0H_2^0H_3^0}`$ $`=`$ $`{\displaystyle \frac{\sqrt{1\left(\frac{M_{H_2^0}+M_{H_3^0}}{M_{H_i^0}}\right)^2}\sqrt{1\left(\frac{M_{H_2^0}M_{H_3^0}}{M_{H_i^0}}\right)^2}}{16\pi M_{H_i^0}}}\times `$ (40)
$`\times \left({\displaystyle \frac{4(\lambda _5\lambda _6)v_\eta v_\rho v_\chi +f(v_\eta ^2v_\rho ^2)}{v_\chi ^2}}\right)^2,`$
$`\mathrm{\Gamma }_{H_i^0V^{}V^+}`$ $`=`$ $`{\displaystyle \frac{\sqrt{14M_{V^\pm }^2/M_{H_i^0}^2}}{8\pi M_{H_i^0}}}(cHi0VPM)^2\left(3{\displaystyle \frac{M_{H_i^0}^2}{M_{V^\pm }^2}}+{\displaystyle \frac{M_{H_i^0}^4}{4M_{V^\pm }^4}}\right),`$ (41)
$`\mathrm{\Gamma }_{H_i^0U^{}U^{++}}`$ $`=`$ $`{\displaystyle \frac{\sqrt{14M_{U^{\pm \pm }}^2/M_{H_i^0}^2}}{8\pi M_{H_i^0}}}(cHi0UPP)^2\left(3{\displaystyle \frac{M_{H_i^0}^2}{M_{U^{\pm \pm }}^2}}+{\displaystyle \frac{M_{H_i^0}^4}{4M_{U^{\pm \pm }}^4}}\right),`$ (42)
$`\mathrm{\Gamma }_{H_1^0H_2^0Z}`$ $`=`$ $`{\displaystyle \frac{\sqrt{1\left(\frac{M_{H_2^0}+M_Z}{M_{H_1^0}}\right)^2}\sqrt{1\left(\frac{M_{H_2^0}M_Z}{M_{H_1^0}}\right)^2}}{4\pi M_{H_1^0}}}\left({\displaystyle \frac{gM_Zv_\eta v_\rho }{M_Wv_\chi }}\right)^2\times `$ (44)
$`\times \left({\displaystyle \frac{M_Z^2}{4}}+{\displaystyle \frac{M_{H_2^0}^2}{4M_Z^2}}{\displaystyle \frac{M_{H_2^0}^2M_{H_1^0}^2}{2M_Z^2}}+{\displaystyle \frac{M_{H_1^0}^4}{4M_Z^2}}{\displaystyle \frac{M_{H_2^0}^2}{2}}{\displaystyle \frac{M_{H_1^0}^2}{2}}\right),`$
$`\mathrm{\Gamma }_{H_1^0H_2^0Z^{}}`$ $`=`$ $`{\displaystyle \frac{3\sqrt{1\left(\frac{M_{H_2^0}+M_Z^{}}{M_{H_1^0}}\right)^2}\sqrt{1\left(\frac{M_{H_2^0}M_Z^{}}{M_{H_1^0}}\right)^2}}{4\pi M_{H_1^0}}}\left({\displaystyle \frac{gv_\eta v_\rho t_W^2}{v_\chi ^2\sqrt{1+3t_W^2}}}\right)^2\times `$ (46)
$`\times \left({\displaystyle \frac{M_Z^{}^2}{4}}+{\displaystyle \frac{M_{H_2^0}^2}{4M_Z^{}^2}}{\displaystyle \frac{M_{H_2^0}^2M_{H_1^0}^2}{2M_Z^{}^2}}+{\displaystyle \frac{M_{H_1^0}^4}{4M_Z^{}^2}}{\displaystyle \frac{M_{H_2^0}^2}{2}}{\displaystyle \frac{M_{H_1^0}^2}{2}}\right),`$
$`\mathrm{\Gamma }_{H_i^0H^{}H^{++}}`$ $`=`$ $`{\displaystyle \frac{\sqrt{14M_{H^{\pm \pm }}^2/M_{H_i^0}^2}}{16\pi M_{H_i^0}}}(cHi0HPP)^2,`$ (47)
where using (15) to (22) and putting i=1,2, we will have the total width for $`H_1^0`$ and part of the total width for $`H_2^0`$. The total width of the Higgs $`H_2^0`$ into quarks, leptons, Z Z, ZZ gauge bosons, $`H_1^{}H_1^+`$, $`H_2^{}H_2^+`$, $`h^0h^0`$, $`H_1^0H_3^0`$ higgs bosons, $`V^{}V^+`$ charged bosons, $`U^{}U^{++}`$ double charged bosons, $`H_1^0Z`$, $`H_1^0Z^{}`$ bosons and $`H^{}H^{++}`$ double charged Higgs bosons, is given by
$`\mathrm{\Gamma }\left(H_2^0\mathrm{all}\right)`$ $`=`$ $`\mathrm{\Gamma }_{H_2^0q\overline{q}}+\mathrm{\Gamma }_{H_2^0\mathrm{}^{}\mathrm{}^+}+\mathrm{\Gamma }_{H_2^0Z^{}Z}+\mathrm{\Gamma }_{H_2^0Z^{}Z^{}}+\mathrm{\Gamma }_{H_2^0H_1^{}H_1^+}+`$ (50)
$`+\mathrm{\Gamma }_{H_2^0H_2^{}H_2^+}+\mathrm{\Gamma }_{H_2^0h^0h^0}+\mathrm{\Gamma }_{H_2^0H_1^0H_3^0}+\mathrm{\Gamma }_{H_2^0V^{}V^+}+\mathrm{\Gamma }_{H_2^0U^{}U^+}+`$
$`+\mathrm{\Gamma }_{H_2^0H_1^0Z}+\mathrm{\Gamma }_{H_2^0H_1^0Z^{}}+\mathrm{\Gamma }_{H_2^0H^{}H^{++}},`$
where we have for the remaining part of the $`H_2^0`$
$`\mathrm{\Gamma }_{H_2^0b\overline{b}}`$ $`=`$ $`{\displaystyle \frac{3\sqrt{14M_b^2/M_{H_2^0}^2}}{16\pi M_{H_2^0}}}{\displaystyle \frac{M_b^2v_\rho ^2}{v_W^2v_\eta ^2}}(M_{H_2^0}^22M_b^2),`$ (51)
$`\mathrm{\Gamma }_{H_2^0c\overline{c}(t\overline{t})}`$ $`=`$ $`{\displaystyle \frac{3\sqrt{14M_c^2/M_{H_2^0}^2}}{16\pi M_{H_2^0}}}{\displaystyle \frac{M_c^2v_\eta ^2}{v_W^2v_\rho ^2}}(M_{H_2^0}^22M_c^2),`$ (52)
$`\mathrm{\Gamma }_{H_2^0\tau ^{}\tau ^+}`$ $`=`$ $`{\displaystyle \frac{\sqrt{14M_\tau ^2/M_{H_2^0}^2}}{16\pi M_{H_2^0}}}{\displaystyle \frac{M_\tau ^2v_\rho ^2}{v_W^2v_\eta ^2}}(M_{H_1^0}^22M_\tau ^2),`$ (53)
$`\mathrm{\Gamma }_{H_2^0Z^{}Z^{}}`$ $`=`$ $`{\displaystyle \frac{\sqrt{14M_Z^2/M_{H_2^0}^2}g^4t_W^4}{\pi M_{H_2^0}}}{\displaystyle \frac{v_\eta ^2v_\rho ^2}{v_\chi ^2}}\left(3{\displaystyle \frac{M_{H_2^0}^2}{M_Z^{}^2}}+{\displaystyle \frac{1}{4}}{\displaystyle \frac{M_{H_2^0}^4}{M_Z^{}^4}}\right),`$ (54)
$`\mathrm{\Gamma }_{H_2^0H_1^{}H_1^+}`$ $`=`$ $`{\displaystyle \frac{\sqrt{14M_{H_1^\pm }^2/M_{H_2^0}^2}}{4\pi M_{H_2^0}}}\left(v_\eta v_\rho {\displaystyle \frac{(\lambda _42\lambda _1)v_\eta ^2+v_\rho ^2(2\lambda _2\lambda _4)}{v_\chi ^3}}\right)^2,`$ (55)
$`\mathrm{\Gamma }_{H_2^0H_2^{}H_2^+}`$ $`=`$ $`{\displaystyle \frac{\sqrt{14M_{H_2^\pm }^2/M_{H_2^0}^2}}{16\pi M_{H_2^0}}}\times `$ (57)
$`\times \left(v_\rho {\displaystyle \frac{2(\lambda _5+\lambda _8\lambda _6)v_\eta v_\rho ^2+2(2\lambda _1\lambda _4)v_\eta v_\chi ^2+2\lambda _8v_\rho v_\chi ^2+fv_\rho v_\chi }{v_W(v_\rho ^2+v_\chi ^2)}}\right)^2,`$
$`\mathrm{\Gamma }_{H_2^0h_0h_0}`$ $`=`$ $`{\displaystyle \frac{\sqrt{14M_{h_0}^2/M_{H_2^0}^2}}{4\pi M_{H_2^0}}}\left({\displaystyle \frac{(\lambda _6\lambda _5)v_\eta v_\rho }{v_\chi }}\right)^2,`$ (58)
$`\mathrm{\Gamma }_{H_2^0H_1^0Z}`$ $`=`$ $`{\displaystyle \frac{\sqrt{1\left(\frac{M_{H_1^0}+M_Z}{M_{H_2^0}}\right)^2}\sqrt{1\left(\frac{M_{H_1^0}M_Z}{M_{H_2^0}}\right)^2}}{4\pi M_{H_2^0}}}\left({\displaystyle \frac{gM_Zv_\eta v_\rho }{M_Wv_\chi }}\right)^2\times `$ (60)
$`\times \left({\displaystyle \frac{M_Z^2}{4}}+{\displaystyle \frac{M_{H_1^0}^2}{4M_Z^2}}{\displaystyle \frac{M_{H_1^0}^2M_{H_2^0}^2}{2M_Z^2}}+{\displaystyle \frac{M_{H_2^0}^4}{4M_Z^2}}{\displaystyle \frac{M_{H_1^0}^2}{2}}{\displaystyle \frac{M_{H_2^0}^2}{2}}\right),`$
$`\mathrm{\Gamma }_{H_2^0H_1^0Z^{}}`$ $`=`$ $`{\displaystyle \frac{3\sqrt{1\left(\frac{M_{H_1^0}+M_Z^{}}{M_{H_2^0}}\right)^2}\sqrt{1\left(\frac{M_{H_1^0}M_Z^{}}{M_{H_2^0}}\right)^2}}{4\pi M_{H_2^0}}}\left({\displaystyle \frac{gv_\eta v_\rho t_W^2}{v_\chi ^2\sqrt{1+3t_W^2}}}\right)^2\times `$ (62)
$`\times \left({\displaystyle \frac{M_Z^{}^2}{4}}+{\displaystyle \frac{M_{H_2^0}^2}{4M_Z^{}^2}}{\displaystyle \frac{M_{H_2^0}^2M_{H_1^0}^2}{2M_Z^{}^2}}+{\displaystyle \frac{M_{H_1^0}^4}{4M_Z^{}^2}}{\displaystyle \frac{M_{H_2^0}^2}{2}}{\displaystyle \frac{M_{H_1^0}^2}{2}}\right),`$
The total width of the $`Z^{}`$ boson, whose one part was already calculated in cieto , is
$`\mathrm{\Gamma }\left(Z^{}\mathrm{all}\right)`$ $`=`$ $`\mathrm{\Gamma }_{Z^{}P^{}P^+}+\mathrm{\Gamma }_{Z^{}\mathrm{}_i^{}\mathrm{}_i^+}+\mathrm{\Gamma }_{Z^{}\nu _i\overline{\nu }_i}+\mathrm{\Gamma }_{Z^{}q\overline{q}\left(J\overline{J}\right)}+\mathrm{\Gamma }_{Z^{}X^{}X^+}+\mathrm{\Gamma }_{Z^{}H_1^0H_1^0}+`$
$`+\mathrm{\Gamma }_{Z^{}H_2^0H_2^0}+\mathrm{\Gamma }_{Z^{}H_1^{}H_1^+}+\mathrm{\Gamma }_{Z^{}H_2^{}H_2^+}+\mathrm{\Gamma }_{Z^{}H_1^0H_2^0}+\mathrm{\Gamma }_{Z^{}H_1^0Z}+\mathrm{\Gamma }_{Z^{}H_2^0Z},`$
where $`i=e,\mu `$ and $`\tau `$, $`X^\pm =V^\pm `$ or $`U^{\pm \pm }`$ and we have for the others particles the relations
$`\mathrm{\Gamma }_{Z^{}H_1^0H_1^0}`$ $`=`$ $`{\displaystyle \frac{\sqrt{14M_{H_1^0}^2/M_Z^{}^2}}{2304\pi M_Z^{}}}\left(g{\displaystyle \frac{v_W^2+6v_\eta ^2t_W^2}{v_W^2(1+3t_W^2)}}\right)^2(M_Z^{}^24M_{H_1^0}^2),`$ (63)
$`\mathrm{\Gamma }_{Z^{}H_2^0H_2^0}`$ $`=`$ $`{\displaystyle \frac{\sqrt{14M_{H_2^0}^2/M_Z^{}^2}}{2304\pi M_Z^{}}}\left(g{\displaystyle \frac{v_W^2+6v_\eta ^2t_W^2}{v_W^2(1+3t_W^2)}}\right)^2(M_Z^{}^24M_{H_2^0}^2),`$ (64)
$`\mathrm{\Gamma }_{Z^{}H_1^{}H_1^+}`$ $`=`$ $`{\displaystyle \frac{\sqrt{14M_{H_1^\pm }^2/M_Z^{}^2}}{576\pi M_Z^{}}}\left({\displaystyle \frac{gv_\rho ^2(1+6t_W^2)+v_\eta ^2}{v_\chi ^2(1+3t_W^2)}}\right)^2(M_Z^{}^24M_{H_1^\pm }^2),`$ (65)
$`\mathrm{\Gamma }_{Z^{}H_2^{}H_2^+}`$ $`=`$ $`{\displaystyle \frac{\sqrt{14M_{H_2^\pm }^2/M_Z^{}^2}}{576\pi M_Z^{}}}\left({\displaystyle \frac{gv_\rho ^2(1+6t_W^2)+v_\eta ^2}{v_\chi ^2(1+3t_W^2)}}\right)^2(M_Z^{}^24M_{H_2^\pm }^2).`$ (66)
## IV Results and conclusions
In the following we present the cross section for the process $`e^{}e^+ZH_i^0`$ where $`i=1,2`$, for the NLC (500 GeV) and CLIC (1000 GeV). In all calculations it will be taken the following parameters $`M_{J_1}=250`$ GeV, $`M_{J_2}=350`$ GeV, $`M_{J_3}=500`$ GeV, $`M_V^\pm =200`$ GeV, $`M_{U^{\pm \pm }}=200`$ GeV, $`M_{P_a}=200`$ GeV, $`M_Z^{}=600`$ GeV, $`\lambda _i=1`$ where $`i=1,2,\mathrm{},9`$, $`M_{H_i^0}=200`$ GeV where $`i=1,2,3`$, $`M_{H_i^\pm }=200`$ GeV where $`i=1,2`$, $`M_{H^{++}}=200`$ GeV, $`f=1000`$ GeV and the vacuum expectation value $`w=1000`$ GeV.
The mass of $`M_Z^{}`$ taken above is in accord with the estimates of the CDF and D0 experiments, which probes the $`Z^{}`$ masses in the 500-800 GeV range, tait , while the reach of the LHC is superior for higher masses, that is $`1TeV<M_Z^{}5`$ TeV, freitas . With regards to Higgs the LHC is able to discover the Higgs boson with a mass up to 1 TeV and to check its basic properties. In Fig. 1, we show the cross section $`e^{}e^+ZH_1^0`$, this process will be studied in two cases, the one where we put for the vacuum expectation value $`v_\eta =140`$ GeV and the other $`v_\eta =240`$ GeV, respectively.
Considering that the expected integrated luminosity for both colliders will be of order of $`6\times 10^4`$ pb<sup>-1</sup>/yr and $`2\times 10^5`$ pb<sup>-1</sup>/yr, then the statistics we are expecting are the following.
The first collider gives a total of $`3.4\times 10^4`$ events per year for $`v_\eta =140`$ GeV, if we take the mass of the boson $`M_{H_1^0}=360`$ GeV.
Considering that the signal for $`H_1^0Z`$ production will be $`t\overline{t}`$ and $`q\overline{q}`$ and taking into account that the branching ratios for both particles would be $`B(H_1^0t\overline{t})=3.6\%`$ and $`B(Zq\overline{q})=69.9\%`$, see Figs. 5 and 6, we would have approximately $`855`$ events per year.
Comparing this signal with the standard model background, like $`e^{}e^+W^{}W^+,ZZ`$, we note that this background can be easily distinguished and therefore eliminated by measuring the transverse mass of the two pairs of jets, see cazer , but even so there is another small background, such as $`e^{}e^+WWZ`$, but the cross section for this process is suppressed by at least $`\alpha /\mathrm{sin}^2\theta _W`$ relative to the process involving a double gauge boson, so using the COMPHEP pukhov , the total cross section for this process will be equal to $`4.23\times 10^2`$ pb.
The second collider (CLIC) gives a total of $`2.2\times 10^4`$ events per year if we take the same neutral Higgs mass, that is $`M_{H_1^0}=360`$ GeV and considering the same branching ratios for the $`H_1^0`$ and the Z cited above, we would have nearly 553 events per year, for the signals and backgrounds see also cazer .
In Fig. 2, we show the cross section for the production of the same particles as in Fig. 1, in the colliders NLC and CLIC for $`v_\eta =240`$ GeV and with $`M_{H_1^0}=360`$ GeV. We see from these results that we can expect for the first collider a total of $`1.05\times 10^7`$ events per year.
For the second collider, the CLIC, we expect a total of $`2.9\times 10^5`$ events per year, that would be more than enough to establish the existence of the $`H_1^0`$.
It is interesting to note the difference between the cross section for $`v_\eta =140`$ GeV and for $`v_\eta =240`$ GeV, this difference is due to the coupling constant, see (16).
In Fig. 3, it is shown the cross section for the production of $`e^{}e^+H_2^0Z`$, for $`v_\eta =140`$ GeV with mass of $`M_{H_2^0}=360`$ GeV. We see from these results that we can expect for the first collider a total of $`6.6\times 10^4`$ events per year to produce $`H_2^0Z`$. Taking into account that the $`H_2^0`$ and Z will decay in $`t\overline{t}`$ and $`\mathrm{}^{}\mathrm{}^+`$, see Figs. 7 and 8, and considering that the branching ratios for them are $`B(H_2^0t\overline{t})=21.17\%`$ and $`B(Z\mathrm{}^{}\mathrm{}^+)=3.36\%`$, then we will have a total of $`469`$ events per year, however this events will be affected by backgrounds such as $`e^{}e^+q\overline{q},WW,ZZ`$ production, cazer . For the second collider, the CLIC, we expect a total of $`2.4\times 10^4`$ events per year for the mass of $`H_2^0`$ equal to $`700`$ GeV and $`v_\eta =140`$ GeV, considering now that the channel of decay will be $`Zq\overline{q}`$ and $`H_2ZZ^{}`$ with $`Zb\overline{b}`$ and $`Z^{}e^{}e^+`$, which branching ratios are equal to $`B(Zb\overline{b})=15,45\%`$ and $`B(Z^{}e^{}e^+)=5.9\%`$, see Figs. 9 and 10, we will have a total of $`153`$ events per year, that is if we are looking for the signal $`j\overline{j}b\overline{b}e\overline{e}`$, we could discover the $`H_2^0`$ and $`Z^{}`$, the backgrounds for this signal can be $`WZZ`$, $`HZZ`$, while the cross sections are so small $`10^2`$ a detailed simulation of Monte Carlo must be done in all cases to extract the signal from the background.
Fig. 4 exhibits the total cross secction for the production of the same particles as in Fig. 3, in the colliders NLC and CLIC for $`v_\eta =240`$ GeV. We see from these results that we can expect for the first collider a total of $`1.5\times 10^4`$ events per year for $`M_{H_2^0}=360`$ GeV. This cross section is smaller compared with that of Fig. 3 by a factor of $`0.227`$. This difference between the cross section for $`v_\eta =140`$ and $`v_\eta =240`$ is due to the coupling constant (17). If we want to look for a signal such as $`e^{}e^+H_2^0Zt\overline{t}\mathrm{}^{}\mathrm{}^+`$ we must multiply $`469\times 0.227`$ that gives 106 events. We also have that the CLIC can produce a total of $`5.3\times 10^3`$ for the mass of $`M_{H_2}=700`$ GeV and for $`v_\eta =240`$ GeV, that is, this cross section is smaller by a factor of $`0.22`$ compared with the same process but for $`v_\eta =140`$ GeV, then the signal $`e^{}e^+H_2^0ZZZ^{}Zq\overline{q}(b\overline{b}e^{}e^+)`$ would give a total of $`34`$ events per year. So, we can conclude that the branching fraction measurements could tell us if the Higgs is standard or not.
We can also produce the Higgs bosons via the W-fusion, in which the Higgs bosons are formed in WW collisions and in association with neutrinos, that is
$$e^+e^{}H+\nu \overline{\nu },$$
two mechanisms are responsible for this production, namely, Higgs-strahlung with Z decays to the three types of neutrinos and WW fusion cazer ; petcov ; altare ; kramer ; kniehl ; boss , being this last the dominant one for larger Higgs mass. Detailed analysis of this production will be given elsewhere cieto1 .
In summary, we have shown in this work that in the context of the 3-3-1 model the signatures for neutral Higgs bosons can be significant in both the NLC and in the CLIC colliders, however a detailed simulation of Monte Carlo must be done in all cases.
###### Acknowledgements.
We would like to thank Prof. R. O. Ramos for a careful reading of the manuscript.
|
warning/0506/math0506280.html
|
ar5iv
|
text
|
# Stability of Relative Equilibria with Singular Momentum Values in Simple Mechanical Systems
## 1. Introduction
A relative equilibrium in a dynamical system with symmetry is a point in phase space for which its dynamical evolution is contained in a group orbit. The study of relative equilibria in symmetric Hamiltonian systems has been around for a long time, with its origins in the field of analytical dynamics and more recently using the modern symplectic and Poisson geometric setup. Relative equilibria are important since they are the analogues to equilibrium states in systems with symmetry, formalized with the action of a Lie group on the phase space. In physical applications, the only observable equilibrium states are those which are stable under small perturbations. Similarly, in the symmetric context, the only observable relative equilibria are those which are stable in some adequate sense. Based in Noetherโs Theorem, geometrized in the property of the invariance of the level sets of the momentum map, the notion of stability generally adopted is that of *$`G_\mu `$-stability* introduced in , and that is closely related to the Lyapunov stability of the induced Hamiltonian flow on the reduced phase space.
In the field of analytical dynamics, the classical Routh Theorem gives conditions on the stability of steady motions which keep stationary the value of a first integral of a dynamical system for fixed values of the others (see for instance and the treatments based on the Routh Theorem in ). Relative equilibria are seen in this context as steady motions for systems having cyclic coordinates due to the existence of a symmetry group, for which the components of the momentum map together with the energy provide a set of first integrals. In the last decades, the implementation of these principles within the field of Geometric Mechanics has been studied. This has produced methods (like the Energy-Momentum Method and the Energy-Casimir Method , see also for an overview) to test the stability of relative equilibria in Hamiltonian systems for arbitrary symmetry groups and momentum values. These methods exploit Noetherโs Theorem and the symplectic and Poisson geometry of the phase space. In the case that the relative equilibrium under study lies in a regular value $`\mu `$ of the momentum map, the Energy-Momentum Method of provides a technique to test its stability modulo the action of $`G_\mu `$, the stabilizer of the momentum value $`\mu `$ under the coadjoint representation of $`G`$. This was generalized in and to also cover the case when the momentum value is singular (this happens when the symmetry group does not act freely on phase space), assuming $`G_\mu `$ is compact. Also, in and stability of relative equilibria satisfying several other hypotheses is investigated.
A very important particular kind of Hamiltonian system is the class of *simple mechanical systems*, paradigmatic of Classical Mechanics, since many Hamiltonian system of physical interest lie in this category or can be obtained from a simple mechanical system by some suitable reduction process. These have as phase space the cotangent bundle of a Riemannian manifold (called configuration space) equipped with its canonical symplectic form, and the Hamiltonian function is of the type โkinetic plus potentialโ energy, where the kinetic energy is given by the norm obtained from the Riemannian metric, and the potential energy is the pullback of a function defined on the base. Symmetry in this systems is implemented by the lift of an isometric action on the base that preserves the potential energy. This big amount of extra structure with respect to general Hamiltonian systems on arbitrary symplectic or Poisson phase spaces implies that in simple mechanical systems everything is constructible from the knowledge of the configuration space, its Riemannian structure, the action of the symmetry group on it and the choice of an invariant potential energy. Therefore, it is reasonably to expect that the stability methods referred previously will particularize in a way that the involved computational complexity will simplify considerably, in that one would work at the level of the configuration space, instead of on its twice dimensional cotangent bundle.
In the case of regular relative equilibria, this refinement of the Energy-Momentum Method has been worked out in , and the obtained stability test for relative equilibria in simple mechanical systems is known as reduced Energy-Momentum Method. Its conditions for $`G_\mu `$-stability are reduced from the level of phase space to the level of configuration space. This method has the highest degree of sophistication among the different stability tests available in the literature of symmetric Hamiltonian systems, and as part of it, it provides a block-diagonalization technique that allows to express the linearization of the Hamiltonian vector field at a relative equilibrium in a way adapted both to the symmetry of the system and to the fibered structure of the phase space. This block-diagonalization yields also further simplifications in the stability analysis.
Surprisingly, in the very frequent and important case of singular momentum values such a refinement for simple mechanical systems has not been studied in detail, and thus the application range of the reduced Energy-Momentum Method is severely limited. Indeed, the literature of applications of the theory of relative equilibria is full of examples in which singular relative equilibria of simple mechanical systems are studied with general geometric and Hamiltonian techniques which neglect their extra structure, in particular for the stability analysis. This paper provides a solution to this situation by obtaining a generalization to singular momentum values of the reduced Energy-Momentum Method and its main features.
In Section 2 we quickly review the theory of relative equilibria for general Hamiltonian systems and simple mechanical systems, and we collect some of the standard results on their stability by geometric methods. Section 3 is a necessary technical interlude on the properties and geometry of a distinguished symplectic component of the linear slice for a cotangent-lifted action, and most of our subsequent results will rely on this section. In Section 4 our main result, Theorem 4.1, is stated, providing an extension of the reduced Energy-Momentum Method of applicable to relative equilibria with singular momentum values. Section 5 applies this result to a classical example of a relative equilibrium with a singular momentum value in a well-known simple mechanical system consisting of an axisymmetric rigid body with a fixed point in an homogeneous gravity field. It is shown how the application of our method simplifies the stability analysis with respect to the application of the methods developed for general Hamiltonian systems. In Section 6 we extend the block-diagonalization result of to the singular case, in Corollary 6.2 and Proposition 6.2. Finally, Section 7 puts in context our results with related work in the literature. In particular it is shown how the block-diagonal expression for the symplectic matrix of Proposition 6.2 particularizes in the regular case to the normal form obtained in , and a comparison is also made between our results and the Lagrangian Block-Diagonalization method of Lewis .
## 2. Relative Equilibria and simple mechanical systems
Let $`(๐ซ,\omega )`$ be a smooth finite dimensional symplectic manifold with symplectic form $`\omega `$ and $`G`$ a finite dimensional Lie group acting smoothly, properly and in a Hamiltonian fashion on $`(๐ซ,\omega )`$ with $`\mathrm{Ad}^{}`$-equivariant momentum map $`๐:๐ซ๐ค^{}`$. Given a $`G`$-invariant Hamiltonian function $`hC^G(๐ซ)`$, a point $`z๐ซ`$ is called a *relative equilibrium* (for $`h`$) if its Hamiltonian evolution lies inside a group orbit. Equivariance of $`๐`$ and Noetherโs Theorem imply that the Hamiltonian evolution of $`z`$ is described as the orbit of $`z`$ by a one-parameter subgroup of $`G`$ generated by a Lie algebra element $`\xi `$ which belongs to $`๐ค_\mu ๐ค`$, where $`\mu =๐(z)`$ and $`๐ค_\mu `$ is the Lie algebra of $`G_\mu `$, the stabilizer of $`\mu `$ for the coadjoint representation of $`G`$. The element $`\xi `$ is called a *velocity* of the relative equilibrium. Using the usual notation for the infinitesimal action of $`๐ค`$ on $`๐ซ`$ the condition for $`z`$ to be a relative equilibrium is written as $`X_h(z)=\xi _๐ซ(z)`$, where $`X_h`$ is the Hamiltonian vector field associated to the function $`h`$.
If the stabilizer $`G_z`$ of $`z`$ is not discrete then there is a degeneracy in the choice of a velocity for a given relative equilibrium, since any representative of the class $`[\xi ]๐ค/๐ค_z`$ produces the same orbit of $`z`$. In any case, by equivariance of $`๐`$, the inclusion $`G_zG_\mu `$ holds. The quintuple $`(๐ซ,\omega ,G,๐,h)`$ will be called in short a symmetric Hamiltonian system.
The following definition introduced in is generally adopted as the correct notion of stability of relative equilibria in Hamiltonian systems, generalizing in the Hamiltonian context the concept of Lyapunov stability of fixed equilibria for flows of vector fields.
###### Definition 2.1.
A relative equilibrium $`z`$ with momentum $`\mu =๐(z)`$ is said to be $`G_\mu `$-stable if for every $`G_\mu `$-invariant neighbourhood $`U`$ of $`G_\mu z`$ there exists a neighbourhood $`O`$ of $`z`$ such that the Hamiltonian orbit of $`O`$ lies in $`U`$.
In a method for testing stability of relative equilibria with singular momentum values in Hamiltonian systems is developed, generalizing the Energy-Momentum Method of for relative equilibria with discrete stabilizers. We quote here the main result, due to its importance in the subsequent development of the paper. For that, given an element $`\xi ๐ค`$, define the *augmented Hamiltonian* $`h_\xi C^{\mathrm{}}(๐ซ)`$ as
(2.1)
$$h_\xi (z)=h(z)๐(z),\xi .$$
It well-known that $`z๐ซ`$ is a relative equilibrium for the symmetric Hamiltonian system $`(๐ซ,\omega ,G,๐,h)`$ with velocity $`\xi `$ if and only if $`z`$ is a critical point of $`h_\xi `$. Also recall that since the Hamiltonian $`G`$-action is proper, any stabilizer for this action, in particular $`G_z`$, must be compact (see ). Since $`๐ค_z๐ค_\mu `$ we can choose a $`G_z`$-equivariant splitting of $`๐ค_\mu `$ as $`๐ค_\mu =๐ค_z๐ค_z^{}.`$
###### Theorem 2.1 (,).
Let $`(๐ซ,\omega ,G,๐,h)`$ be a symmetric Hamiltonian system and $`z๐ซ`$ a relative equilibrium with stabilizer $`G_z`$ and velocity $`\xi `$. Assume that $`๐(z)=\mu `$ and that $`G_\mu `$ is compact. Then $`\xi ๐ค_\mu `$. Let $`\xi ^{}`$ be the projection of $`\xi `$ onto some $`G_z`$-invariant complement of $`๐ค_z`$ in $`๐ค_\mu `$ (always available by compactness of $`G_z`$). If $`๐_z^2h_\xi ^{}\text{ }V_s`$ is definite for some (and hence any) complement $`V_s`$ to $`๐ค_\mu z`$ in $`\mathrm{ker}T_z๐`$, then $`z`$ is $`G_\mu `$-stable.
In this theorem the ambiguity in the velocity introduced by the stabilizer of the relative equilibrium appears explicitly. In typical computations, testing this condition over all possible $`G_z`$-invariant complements of $`๐ค_z`$ in $`๐ค_\mu `$ gives the sharpest stability results (see and the example in Section 5). There is an infinite number of choices for the space $`V_s`$ in Theorem 2.1, and any of them is called the (maximal) *symplectic normal space* at $`z`$, since it is a maximal symplectic subspace of the symplectic orthogonal to the group orbit at $`z`$.
In this paper we will study a particular case of Hamiltonian systems of great interest in Classical Mechanics. This is the class of the so-called symmetric *simple mechanical systems*, which are symmetric Hamiltonian systems where $`๐ซ`$ is $`T^{}Q`$, the cotangent bundle of a smooth, finite-dimensional Riemannian manifold $`(Q,,)`$ equipped with its canonical symplectic form $`\omega `$ and $`G`$ is a finite-dimensional Lie group acting properly and isometrically on $`Q`$ and by cotangent lifts on $`T^{}Q`$.
Following , the Hamiltonian function is constructed in the following way: let $`V`$ be a smooth $`G`$-invariant function on $`Q`$ and call $`\overline{V}=\tau ^{}VC^G(T^{}Q)`$, where $`\tau :T^{}QQ`$ is the cotangent bundle projection. We will refer to both $`\overline{V}`$ and $`V`$ as the *potential energy*. Let $`๐ฝ\mathrm{L}:TQT^{}Q`$ be the *Legendre map* associated to $`,`$, defined by the formula
(2.2)
$$๐ฝ\mathrm{L}(v_x),w_x=v_x,w_x,v_x,w_xT_xQ.$$
The Legendre map is a $`G`$-equivariant vector bundle isomorphism covering the identity on $`Q`$. With it, we can define the *kinetic energy* $`KC^G(๐ซ)`$ as
$$K(p_x)=\frac{1}{2}๐ฝ\mathrm{L}^1(p_x),๐ฝ\mathrm{L}^1(p_x),p_xT_x^{}Q.$$
Finally, the Hamiltonian $`h`$ is defined by
(2.3)
$$h=K+\overline{V}.$$
With respect to the canonical symplectic structure of $`T^{}Q`$ the cotangent-lifted action of $`G`$ is Hamiltonian, with equivariant momentum map defined by the expression
(2.4)
$$๐(p_x),\xi =p_x,\xi _Q(x),\xi ๐ค.$$
The symplectic manifold $`T^{}Q`$ is the phase space of the Hamiltonian system, while the base $`Q`$ is usually called *configuration space*. Accordingly, for any point $`p_x`$ in $`T^{}Q`$, the projection $`x=\tau (p_x)`$ is called the configuration point (or base point) of $`p_x`$.
A key feature of simple mechanical systems is that both their geometric and dynamical properties are entirely constructed using the knowledge of the Riemannian manifold $`(Q,,)`$, the isometric $`G`$-action on it and the choice of a potential energy $`V`$. Thus, one could reasonably expect that the implementation on this class of Hamiltonian systems of the stability test given in Theorem 2.1 should simplify accordingly, and yield easier computations at the level of $`Q`$ and $`G`$ instead of the bigger space $`๐ซ=T^{}Q`$. The obtention of such a refinement to simple mechanical systems of this stability test is the main result of this paper, Theorem 4.1.
###### Remark 2.1.
There are several treatments of this problem in the literature. Simo *et al.* develop in a sophisticated test, that particularizes Theorem 2.1, called the reduced Energy-Momentum Method. This test gives sufficient conditions for the stability of relative equilibria in simple mechanical systems provided the configuration point of the cotangent relative equilibrium has a discrete stabilizer (which is the same as to require that its momentum value is regular). The main advantage of this is that it is constructed specifically for this class of systems, and this fact reflects in less computational difficulties than the application of the main method, Theorem 2.1, designed for general Hamiltonian systems. In a Lagrangian analogue of the results of is obtained, being valid also for relative equilibria of a larger class of mechanical systems. Here, based in Theorem 2.1, we produce a method for testing stability of relative equilibria in simple mechanical systems that could be seen as a generalization of the reduced Energy-Momentum Method to the singular case, i.e. without requiring discrete stabilizers of configuration points or regular momentum values.
## 3. A cotangent-bundle adapted splitting of the symplectic normal space
In this section we describe a realization $`V_sT_{p_x}(T^{}Q)`$ of the symplectic normal space at a relative equilibrium $`p_x`$ of a simple mechanical system, as well as a convenient cotangent-bundle adapted splitting of $`V_s`$ which will be extremely useful for the remaining constructions. Most of the results of this section are merely expository, and a complete description including proofs and the obtention of the symplectic normal space at points $`p_x`$ of general form, not only relative equilibria, can be found in .
One of the geometric objects that will be extensively used is the *locked inertia tensor* $`๐`$, a family of bilinear positive semi-definite symmetric forms on $`๐ค`$ defined by
(3.1)
$$๐(x)(\xi ,\eta )=\xi _Q(x),\eta _Q(x),\xi ,\eta ๐ค,xQ.$$
Note that at each point $`x`$, the kernel of $`๐(x)`$ is precisely $`๐ค_x`$, the Lie algebra of the stabilizer of $`x`$. Therefore $`๐(x)`$ is a well defined inner product on $`๐ค`$ only at points of $`Q`$ where the action is locally free. The locked inertia tensor satisfies the following invariance and infinitesimal invariance properties (see ):
(3.2)
$$๐(gx)(\mathrm{Ad}_g\xi ,\mathrm{Ad}_g\eta )=๐(x)(\xi ,\eta )$$
(3.3)
$$(๐๐\lambda _Q(x))(\xi ,\eta )+๐(x)(\mathrm{ad}_\lambda \xi ,\eta )+๐(x)(\xi ,\mathrm{ad}_\lambda \eta )=0,$$
for every $`gG,xQ`$ and $`\xi ,\eta ,\lambda ๐ค`$. Note that $`G_{p_x}G_x`$ by the equivariance of $`\tau `$. Let $`xQ`$ be the base point of an element $`p_xT^{}Q`$ and denote by $`H=G_x`$ its stabilizer. Using the $`G`$-invariant Riemannian metric on $`Q`$ we can form the splitting
(3.4)
$$T_xQ=๐คx๐$$
where $`๐=(๐คx)^{}`$, and it is usually called a linear slice (for the $`G`$-action at $`x`$). Hence, $`๐`$ is the space of directions orthogonally complementary to the group orbit. This is obviously a $`H`$-invariant splitting for the induced linear $`H`$-action on $`T_xQ`$.
Next we choose a $`G_{p_x}`$-invariant splitting of the Lie algebra $`๐ค`$ of the form
(3.5)
$$๐ค=๐ฅ๐ฏ.$$
This is always possible by the properness of the $`G`$-action on $`Q`$, which implies that $`G_{p_x}`$ is compact. A concrete way of choosing (3.5) will be introduced in (3.14). For any element $`\xi ๐ค`$ we write in a unique way $`\xi =\xi ^๐ฅ+\xi ^๐ฏ`$, relative to this splitting. The space $`๐ฏ`$ collects the elements of $`๐ค`$ that generate nontrivial orbits of $`x`$.
Noting that $`๐ฏ๐คx`$ by the isomorphism $`\xi \xi _Q(x)`$, we can compose this identification with (3.4) and dualize, to get
(3.6)
$$T_xQ๐ฏ๐\mathrm{and}T_x^{}Q๐ฏ^{}๐^{}.$$
Associated to the Riemannian structure on $`Q`$, there is an Ehresmann connection on $`T^{}Q`$, for which the connection map at $`p_x`$, $`K:T_{p_x}(T^{}Q)T_x^{}Q`$ is defined as
(3.7)
$$K(X)=\frac{D_c^{}}{Dt}\text{ }t=0\widehat{c}(t)XT_{p_x}(T^{}Q),$$
the covariant differential of $`\widehat{c}(t)`$ along $`c(t)=\tau (c(t))`$ relative to the Levi-Civita connection $``$. Here $`\widehat{c}(t)`$ is any local curve $`\widehat{c}:(ฯต,ฯต)T^{}Q`$ projecting to $`c(t)`$ and satisfying $`i)\widehat{c}(0)=p_x`$, and $`ii)\frac{d}{dt}\text{ }t=0\widehat{c}(t)=X`$. This connection map $`K`$ at $`p_x`$ is a $`G_{p_x}`$-equivariant linear map, which combined with the differential at $`p_x`$ of the cotangent bundle projection $`\tau `$ yields a $`G_{p_x}`$-equivariant linear isomorphism $`\mathrm{\Psi }:T_{p_x}(T^{}Q)T_xQT_x^{}Q=T^{}(T_xQ)`$ defined by
$$\mathrm{\Psi }(X)=(T_{p_x}\tau (X),K(X))XT_{p_x}(T^{}Q)$$
(see ). We call vectors at $`p_x`$ lying in the kernel of $`T_{p_x}\tau `$ *vertical*, since they are tangent to the cotangent fiber through $`p_x`$. Analogously, those elements of $`T_{p_x}(T^{}Q)`$ which are in the kernel of $`K`$ are called *horizontal*, and are identified through $`T_{p_x}\tau `$ with vectors tangent to $`Q`$ at $`x`$. We now compose the above isomorphism $`\mathrm{\Psi }`$ with the two dual isomorphisms (3.6) to get a new one
(3.8)
$$I:T_{p_x}(T^{}Q)(๐ฏ๐)(๐ฏ^{}๐^{})$$
which can be explicitly expressed as $`I(X)=(\eta ,a;\nu ,\alpha ),`$ for the unique $`\eta ,a,\nu ,\alpha `$ satisfying
$$\begin{array}{ccc}T_{p_x}\tau (X)\hfill & =\hfill & a+\eta _Q(x)\hfill \\ K(X)\hfill & =\hfill & \alpha +๐ฝ\mathrm{L}\left(\left(\widehat{๐}_0^1(\nu )\right)_Q(x)\right).\hfill \end{array}$$
Here $`\widehat{๐}_0`$ denotes the restriction of $`๐`$ to $`๐ฏ`$, according to (3.5). Note that now $`\widehat{๐}_0`$ becomes a well-defined inner product in $`๐ฏ`$ and thus also a linear isomorphism $`\widehat{๐}_0:๐ฏ๐ฏ^{}๐ฅ^{}`$. We can therefore work in the image of $`I`$, which we call $`I`$-representation, instead of on $`T_{p_x}(T^{}Q)`$, and that is what we will do in the rest of the paper. Note that in this identification the space of vertical and horizontal vectors is expressed, respectively, as
$$(0,0;\nu ,\alpha )\alpha ๐^{},\nu ๐ฏ^{},\mathrm{and}(\eta ,a;0,0)a๐,\eta ๐ฏ.$$
The isomorphism $`I`$ is $`G_{p_x}`$-equivariant with respect to the linear action on the target space given by
(3.9)
$$g(\eta ,a;\nu ,\alpha )=(\mathrm{Ad}_g\eta ,ga,\mathrm{Ad}_{g^1}^{}\nu ,g\alpha ),$$
where $`ga`$ and $`g\alpha `$ denote, respectively, the restriction to $`G_{p_x}`$ of the linear representation of $`H`$ on $`๐`$ and its contragredient representation on $`๐^{}`$. This action is well defined since $`G_{p_x}H`$ and $`๐`$, $`๐ฏ`$, and their duals are $`H`$-invariant by construction.
In order to obtain a convenient characterization of the symplectic normal space $`V_sT_{p_x}(T^{}Q)`$ at a relative equilibrium in the $`I`$-representation, and also for future reference, we quote some technical results introduced in . There it is proved that it is possible to extend vectors $`v๐`$ to local vector fields $`\overline{v}`$ defined in a neighbourhood of $`x`$, in a way adapted to the $`G`$-action, and such that the family of vector fields $`\lambda _Q,\overline{v}`$, for any $`\lambda ๐ค`$ and $`v๐`$ spans $`T_x^{}Q`$ at every $`x^{}`$ near $`x`$.
We will sketch here the obtention of the local field $`\overline{v}`$. Recall that by Palaisโ Tube Theorem we can construct an invariant tubular neighbourhood of an orbit $`GxQ`$ as follows: Let $`H=G_x`$ act on $`G\times ๐`$ as $`h(g,s)=(gh^1,hs)`$. Let $`G\times _H๐`$ be the quotient space for this action. Then there is a $`H`$-invariant open ball $`U๐`$ centered at the origin such that the map $`\sigma :G\times _HUQ`$ defined by
(3.10)
$$\sigma ([g,s])=g\mathrm{exp}_xs$$
is a diffeomorphism onto a $`G`$-invariant neighbourhood $`O`$ of $`Gx`$. Here $`\mathrm{exp}_x:T_xQQ`$ denotes the exponential map associated to $`,`$. This diffeomorphism is $`G`$-equivariant with respect to the given action on $`Q`$ and the $`G`$-action on $`G\times _H๐`$ defined by $`g^{}[g,s]=[g^{}g,s]`$.
Now choose any inner product on $`๐ค`$ such that the splitting (3.5) is an orthogonal direct sum. Extend this inner product by right translations to a $`H`$-invariant Riemannian metric on $`G`$. Then we can interpret $`๐ฏ`$ as a linear slice at the identity for the free $`H`$-action on $`G`$ given by $`(h,g)gh^1`$. It follows that if we call $`\mathrm{exp}_e`$ the exponential map for this metric on $`G`$, there is a small $`H`$-invariant neighbourhood of $`e`$ in $`G`$ such that every $`g`$ belonging to it can be written as $`g=\mathrm{exp}_e\xi ^๐ฏh^1`$ for unique elements $`\xi ^๐ฏ๐ฏ`$ and $`hH`$. Using the Tube Theorem every element in a small neighbourhood of $`x`$ in $`Q`$ (not $`G`$-invariant in general) can be expressed as $`x^{}=\sigma ([\mathrm{exp}_e\xi ^๐ฏh^1,s])`$ for unique elements $`\xi ^๐ฏ๐ฏ,hH`$ and $`sU`$. It is easy to prove that for any $`x^{}=\sigma ([\mathrm{exp}_e\xi ^๐ฏh^1,s])`$ near $`x`$, the formula
$$F_v^t(x^{})=\sigma ([\mathrm{exp}_e\xi ^๐ฏh^1,s+thv])$$
defines a flow $`F_v^t`$ for any $`vU๐`$. The associated local field $`\overline{v}`$ is then defined as
(3.11)
$$\overline{v}(x^{})=\frac{d}{dt}\text{ }t=0F_v^t(x^{})$$
for every $`x^{}`$ near $`x`$. In the following theorem we collect the most important properties of this family of local vector fields. For the proof, see .
###### Theorem 3.1.
Let $`v,w๐`$ and $`\eta ,\lambda ,\xi ,\xi _i,\xi _j๐ค`$. Then
* $`_{\xi _{i}^{}{}_{Q}{}^{}}\xi _{j}^{}{}_{Q}{}^{}(x),\lambda _Q(x)=\frac{1}{2}\{(๐๐\xi _{i}^{๐ฏ}{}_{Q}{}^{}(x))(\xi _j,\lambda )๐(x)(\xi _i^๐ฏ,[\xi _j,\lambda ])\}`$
* $`_{\xi _{i}^{}{}_{Q}{}^{}}\xi _{j}^{}{}_{Q}{}^{}(x),w=\frac{1}{2}(๐๐w)(\xi _i^๐ฏ,\xi _j)`$
* $`_{\xi _Q}\overline{v}(x),\lambda _Q(x)=\frac{1}{2}(๐๐v)(\xi ^๐ฏ,\lambda )`$
* $`_{\overline{v}}\xi _Q(x),\lambda _Q(x)=\frac{1}{2}(๐๐v)(\xi ,\lambda )`$
* $`_{\overline{v}}\xi _Q(x),w=_{\xi _Q}\overline{v}(x),w+\xi ^๐ฅv,w`$.
Here, $`\xi ^๐ฏ,\xi _i^๐ฏ`$ and $`\xi ^๐ฅ`$ denote the projections of elements of $`๐ค`$ onto $`๐ฏ`$ and $`๐ฅ`$ according to (3.5).
### Notation:
We will introduce for any $`v๐`$ a linear map $`C(v):๐ฏ๐^{}`$ defined as
(3.12)
$$C(v)(\xi ^๐ฏ),w_๐=_{\xi _Q}\overline{v}(x),w,$$
where $`,_๐`$ is the restriction of the metric on $`Q`$ to $`๐T_xQ`$. Note that $`C`$ is not linear in $`v`$ since it depends on the concrete extension $`\overline{v}`$. We will also employ the following notation: if $`W`$ is a linear subspace of the linear space $`V`$ and $`\iota :WV`$ its inclusion, we will write $`_W:V^{}W^{}`$ for its dual projection.
We fix from now on a point of the form $`p_x=๐ฝ\mathrm{L}(\xi _Q(x))`$ with $`G_x=H`$. It follows from (2.4) and the definition of the locked inertia tensor that $`p_x`$ has momentum $`\mu =๐(p_x)=๐(x)(\xi )=\widehat{๐}_0(\xi ^๐ฏ)`$. The reason for this choice of $`p_x`$ will be clear in the following section, where it is explained why every relative equilibrium of a simple mechanical system must be precisely of this form.
### Remark.
It is a well-known property of points of the form $`p_x=๐ฝ\mathrm{L}(\xi _Q(x))`$ that $`G_{p_x}=HG_\mu `$. This follows immediately from the relation $`G_{p_x}=H_{p_x}`$ and identifying $`p_x=(0,\mu )๐^{}๐ฏ^{}`$ using the $`H`$-isomorphism (3.6). For general points of $`T^{}Q`$ one has only an inclusion $`G_{p_x}HG_\mu `$.
We now make a concrete choice for the complement $`๐ฏ`$ in (3.5), as well as for other relevant subspaces of $`๐ค`$. Start by choosing a $`G_{p_x}`$-invariant complement $`๐ญ`$ to $`๐ค_{p_x}`$ in $`๐ค_\mu `$, i.e.
(3.13)
$$๐ค_\mu =๐ค_{p_x}๐ญ.$$
Next, define a $`G_{p_x}`$-invariant complement $`๐ฑ`$ to $`๐ฅ๐ญ`$ in $`๐ค`$ in such a way that defining $`๐ฏ=๐ญ๐ฑ`$ we have that $`๐ญ`$ and $`๐ฑ`$ are orthogonal with respect to the restricted locked inertia tensor $`\widehat{๐}_0`$. We can then write
(3.14)
$$๐ค=๐ฅ๐ฏ=๐ฅ๐ญ๐ฑ,$$
and hence we have constructed the splitting $`(\text{3.5})`$.
Let us define the following subspace of $`๐ค`$
(3.15)
$$๐ฎ^\mu =\{\lambda ๐ฑ:_๐ฅ[\mathrm{ad}_\lambda ^{}\mu ]=0\}.$$
This space will play an important role in our characterization of $`V_s`$, and it can be proved (see ) that it is isomorphic to the symplectic normal space at $`\mu `$ for the restriction to $`H`$ of the coadjoint action of $`G`$ on $`๐ช_\mu `$, the coadjoint orbit containing $`\mu `$. Note also that by using (3.3) we can write $`๐ฎ^\mu =\{\lambda ๐ฑ:_๐ฅ[(๐๐\lambda _Q(x))(\xi ^๐ฏ)=0]\}`$.
As a particular case of Theorem 6.1 in the space $`\mathrm{ker}T_{p_x}๐`$ consists in the $`I`$-representation in the elements $`(\eta ,a;\nu ,\alpha )(๐ฏ๐)(๐ฏ^{}๐^{})`$ satisfying
$$\nu ,\lambda ^๐ฏ\frac{1}{2}\left\{\left(๐๐\xi _Q(x)\right)(\lambda ^๐ฏ,\eta )\left(๐๐a\right)(\lambda ^๐ฏ,\xi )\mathrm{ad}_{\lambda ^๐ฏ}^{}\mu ,\eta \right\}+\mathrm{ad}_{\lambda ^๐ฅ}^{}\mu ,\eta =0$$
for every $`\lambda ^๐ฏ๐ฏ`$ and $`\lambda ^๐ฅ๐ฅ`$. Also, in the $`I`$-representation
$$๐ค_\mu p_x=\{(\lambda ,0;\frac{1}{2}_๐ฏ\left[(๐๐\xi _Q(x))(\lambda )\right],\frac{1}{2}_๐\left[(๐๐())(\lambda ,\xi )\right]):\lambda ๐ญ\}.$$
From the above two expressions it is easy to obtain that the symplectic normal space $`V_s`$, a complement to $`๐ค_\mu p_x`$ in $`\mathrm{ker}T_{p_x}๐`$, can be chosen to be
(3.16)
$$\begin{array}{ccc}V_s\hfill & =\hfill & \{(\lambda ,a;\frac{1}{2}_๐ฏ[(๐๐\xi _Q^๐ฏ(x))(\lambda )+\mathrm{ad}_\lambda ^{}\mu (๐๐a)(\xi ^๐ฏ)],\gamma \hfill \\ & \hfill & \frac{1}{2}_๐\left[(๐๐())(\xi ^๐ฏ,\lambda )\right]+C(a)(\xi ^๐ฏ),_๐):\lambda ๐ฎ^\mu ,a๐,\gamma ๐^{}\},\hfill \end{array}$$
with $`๐ฎ^\mu `$ defined in (3.15) and $`๐`$ in (3.4). The symplectic normal space $`V_s`$ is $`G_{p_x}`$-invariant by construction with respect to the action (3.9) in its ambient space (see ).
## 4. Stability of singular relative equilibria in simple mechanical systems
In the following we will be in the setup of Section 2 and fix a simple mechanical system $`h=K+\overline{V}`$ as in (2.3). In this framework, once an element $`\xi ๐ค`$ is chosen, we can separate the augmented Hamiltonian (2.1) into a kinetic and a potential part as
$$h_\xi =K_\xi +\overline{V_\xi },\text{where}$$
$$K_\xi (p_x)=\frac{1}{2}p_x\chi ^\xi (x)^2\text{and}V_\xi (x)=V(x)\frac{1}{2}๐(x)(\xi ,\xi ).$$
As for the potential energy we have used the notation $`\overline{V_\xi }=\tau ^{}V_\xi `$. The one-form $`\chi ^\xi `$ is defined by
(4.1)
$$\chi ^\xi (x)=๐ฝ\mathrm{L}(\xi _Q(x)).$$
The functions $`K_\xi `$ and $`V_\xi `$ are called the *augmented kinetic energy* and *augmented potential energy* respectively. Recall now (see for instance Theorem 4.1.2 and Proposition 4.2.1 in ) that with the introduction of these two auxiliary functions we have the following characterization of relative equilibria:
###### Proposition 4.1.
Let $`p_xT^{}Q`$. The following are equivalent:
* $`p_x`$ is a relative equilibrium of (2.3) with momentum $`\mu `$ and velocity $`\xi ๐ค_\mu `$
* $`p_x`$ is a critical point of the augmented Hamiltonian $`h_\xi `$
* $`p_x`$ is simultaneously a critical point of $`K_\xi `$ and $`\overline{V_\xi }`$
* $`p_x=๐ฝ\mathrm{L}(\xi _Q(x))`$ and $`x`$ is a critical point of $`V_\xi `$.
Note that (iv) restricts the form of phase space points candidates to be relative equilibria of (2.3) to be of the form $`p_x=๐ฝ\mathrm{L}(\xi _Q(x))`$. Thus is the reason for studying in detail in the previous section the symplectic normal space only at this class of points.
Let $`p_x=๐ฝ\mathrm{L}(\xi _Q(x))T_x^{}Q`$ be a relative equilibrium for the simple mechanical system (2.3) with momentum $`๐(p_x)=\mu `$. We call $`H=G_x`$ and we choose a $`(G_{p_x}=HG_\mu )`$-invariant inner product on $`๐ค`$ relative to which we construct the splittings (3.13) and (3.14). Note that by hypothesis $`x`$ is a critical point of $`V_\xi ^{}`$ for any $`\xi ^{}๐ค_\mu `$ such that $`[\xi ^{}\xi ]=0๐ค_\mu /๐ค_{p_x}`$. In other words, any such $`\xi ^{}`$ is a velocity for the relative equilibrium $`p_x`$. In particular, this happens for $`\xi ^{}๐ญ`$, the projection of $`\xi `$ onto $`๐ญ`$ according to (3.13).
Let $`\delta pT_{p_x}(T^{}Q)`$ be a tangent vector at $`p_x`$. We will write its horizontal and vertical components as $`\delta p^H=T_{p_x}\tau (\delta p)T_xQ`$ and $`\delta p^V=K(\delta p)T_x^{}Q`$ respectively. It is clear that if $`I(\delta p)=(\lambda ,b;\nu ,\gamma )`$ is the $`I`$-representation of $`\delta p`$ one has that
$$\begin{array}{ccc}\delta p^H\hfill & =\hfill & \lambda _Q(x)+b\hfill \\ \delta p^V\hfill & =\hfill & ๐ฝ\mathrm{L}\left((\widehat{๐}_0^1(\nu ))_Q(x)\right)+\gamma .\hfill \end{array}$$
Also, using (3.6) we can express the horizontal and vertical variations $`\delta p^H`$ and $`\delta p^V`$ as elements of $`๐ฏ๐`$ and $`๐ฏ^{}๐^{}`$ respectively like
$$\begin{array}{ccc}\delta p^H\hfill & =\hfill & (\lambda ,b)\hfill \\ \delta p^V\hfill & =\hfill & (\nu ,\gamma ).\hfill \end{array}$$
We will use both notations indistinctly.
Finally, for a curve $`c(t)Q`$ with $`c(0)=x`$ we will write $`\mathrm{Hor}_{p_x}(c(t))`$ for its horizontal lift to $`T^{}Q`$ at the point $`p_x`$ with respect to the Levi-Civita connection. Equivalently, $`\mathrm{Hor}_{p_x}(c(t))`$ is the parallel translation of $`p_x`$ along the curve $`c(t)`$.
###### Lemma 4.1.
Let $`p_x=๐ฝ\mathrm{L}(\xi _Q(x))T_x^{}Q`$ be a relative equilibrium for the simple mechanical system (2.3) with momentum $`\mu `$ and velocity $`\xi ๐ค_\mu `$. Let $`\xi ^{}`$ the orthogonal projection of $`\xi `$ onto $`๐ญ`$ according to (3.13). Then, for any $`\delta p_1,\delta p_2T_{p_x}(T^{}Q)`$
* $`๐_{p_x}^2\overline{V_\xi ^{}}(\delta p_1,\delta p_2)=๐_x^2V_\xi ^{}(\delta p_1^H,\delta p_2^H)`$
* $`๐_{p_x}^2K_\xi ^{}(\delta p_1,\delta p_2)=\delta p_1^V(T_x\chi ^\xi ^{}\delta p_1^H)^V,\delta p_2^V(T_x\chi ^\xi ^{}\delta p_2^H)^V`$.
###### Proof.
(i) follows immediately by the definition of $`\delta p^H`$ and noting that $`\overline{V_\xi ^{}}=\tau ^{}V_\xi ^{}`$. To prove (ii) we will consider horizontal and vertical vectors separately. Let $`I(\delta p_i)=(0;\delta p_i^V)`$ for $`i=1,2`$. Then
$$\begin{array}{ccc}๐_{p_x}^2K_\xi ^{}(\delta p_1,\delta p_2)\hfill & =\hfill & \frac{1}{2}\frac{d}{ds}s=0\frac{d}{dt}t=0p_x+t\delta p_1^V+s\delta p_2^V\chi ^\xi ^{}(x)^2\hfill \\ & =\hfill & \frac{1}{2}\frac{d}{ds}s=0\frac{d}{dt}t=0t\delta p_1^V+s\delta p_2^V(x)^2\hfill \\ & =\hfill & \delta p_1^V,\delta p_2^V.\hfill \end{array}$$
If $`I(\delta p_1)=(0;\delta p_1^V)`$ and $`I(\delta p_2)=(\delta p_2^H;0)`$, let $`c_{\delta p_2^H}^t(x)`$ be any smooth curve satisfying $`c_{\delta p_2^H}^0(x)=x`$ and $`\frac{d}{dt}\text{ }t=0c_{\delta p_2^H}^t(x)=\delta p_2^H`$. Then
$$\begin{array}{ccc}๐_{p_x}^2K_\xi ^{}(\delta p_1,\delta p_2)\hfill & =\hfill & \frac{1}{2}\frac{d}{ds}s=0\frac{d}{dt}t=0\mathrm{Hor}_{p_x+t\delta p_1^V}(c_{\delta p_2^H}^s(x))\chi ^\xi ^{}(c_{\delta p_2^H}^s(x))^2\hfill \\ & =\hfill & \frac{d}{dt}t=0_{\delta p_2^H}\chi ^\xi ^{}(x),p_x+t\delta p_1^V\chi ^\xi ^{}(x).\hfill \end{array}$$
But since $`p_x`$ is a relative equilibrium with velocity $`\xi `$, then $`p_x=\chi ^\xi ^{}(x)`$ and then
$$๐_{p_x}^2K_\xi ^{}(\delta p_1,\delta p_2)=_{\delta p_2^H}\chi ^\xi ^{}(x),\delta p_1^V.$$
Finally, consider two variations of the form $`I(\delta p_1)=(\delta p_1^H;0)`$ and $`I(\delta p_2)=(\delta p_2^H;0)`$. Then
$$\begin{array}{c}๐_{p_x}^2K_\xi ^{}(\delta p_1,\delta p_2)=\hfill \\ =\frac{1}{2}\frac{d}{ds}s=0\frac{d}{dt}t=0\mathrm{Hor}_{\mathrm{Hor}_{p_x}(c_{\delta p_1^H}^t(x))}(F_{\stackrel{~}{\delta p_2^H}}^s(c_{\delta p_1^H}^t(x)))\chi ^\xi ^{}(F_{\stackrel{~}{\delta p_2^H}}^s(c_{\delta p_1^H}^t(x)))^2\hfill \\ =\frac{d}{dt}t=0_{\stackrel{~}{\delta p_2^H}}\chi ^\xi ^{}(c_{\delta p_1^H}^t(x)),\mathrm{Hor}_{p_x}(c_{\delta p_1^H}^t(x))\chi ^\xi ^{}(c_{\delta p_1^H}^t(x))\hfill \\ =_{\delta p_2^H}\chi ^\xi ^{}(x),_{\delta p_1^H}\chi ^\xi ^{}(x).\hfill \end{array}$$
Where if $`\delta p_2=(\lambda ,b)`$, then $`\stackrel{~}{\delta p_2^H}`$ is the local vector field $`\overline{b}+\lambda _Q`$, and $`F_{\stackrel{~}{\delta p_2^H}}^s`$ denotes its flow. Recalling now the definition (3.7) of the operator $`K`$, for any $`vT_xQ`$ and local extension $`\stackrel{~}{v}`$ one has
$$_v\chi ^\xi ^{}(x)=_{\stackrel{~}{v}}\chi ^\xi ^{}(x)=K(T_x\chi ^\xi ^{}v)=(T_x\chi ^\xi ^{}v)^V,$$
and the result is proved. โ
Since every horizontal variation $`\delta p^H`$ can be written as the sum of two contributions one coming from $`๐ฏ`$ and the other from $`๐`$, that is $`\delta p^H=\zeta _Q(x)+b`$, for $`\zeta ๐ฏ,b๐`$, we can consider these two contributions separately and thus obtain concrete expressions for $`(T_x\chi ^\xi ^{}\delta p^H)^V`$.
###### Lemma 4.2.
Let $`\zeta ๐ฏ`$ and $`b๐`$, and identify $`T_x^{}Q`$ with $`๐ฏ^{}๐^{}`$ by the isomorphism (3.6). Then,
* $`\begin{array}{c}(T_x\chi ^\xi ^{}\zeta _Q(x))^V=\hfill \\ (\frac{1}{2}_๐ฏ\left[\left(๐๐\xi _Q^{}(x)\right)(\zeta )+\mathrm{ad}_\zeta ^{}\mu +2\left(๐๐\zeta _Q(x)\right)(\xi ^{})\right],\frac{1}{2}_๐\left[\left(๐๐()\right)(\xi ^{},\zeta )\right]).\hfill \end{array}`$
* $`(T_x\chi ^\xi ^{}b)^V=(\frac{1}{2}_๐ฏ\left[(๐๐b)(\xi ^{})\right],C(b)(\xi ^{}),_๐)`$.
###### Proof.
The proof is just an immediate consequence of Theorem 3.1, in particular of items (i), (ii), (iv), and (v). We will just prove (ii). Recalling that $``$ is a metric connection, then
$$(T_x\chi ^\xi ^{}b)^V=_{\overline{b}}\left(๐ฝ\mathrm{L}(\xi _Q^{})\right)(x)=๐ฝ\mathrm{L}(_{\overline{b}}\xi _Q^{})(x).$$
For $`\lambda ๐ฏ`$ we have, from item (iv) of Theorem 3.1
$$(T_x\chi ^\xi ^{}b)^V,\lambda _Q(x)=_{\overline{b}}\xi _Q^{}(x),\lambda _Q(x)=\frac{1}{2}(๐๐b)(\xi ^{},\lambda ).$$
Similarly, if $`w๐`$, then by item (v) of the above referred theorem we have
$$(T_x\chi ^\xi ^{}b)^V,w=_{\overline{b}}\xi _Q^{}(x),w=C(b)(\xi ^{}),w_๐.$$
This yields result (ii) The proof of (i) is identical, including some manipulations using the infinitesimal equivariance property of the locked inertia tensor given in (3.3). โ
We apply now the results obtained so far in order to produce a singular version of the reduced Energy-Momentum Method of . Consider $`๐ซ=T^{}Q`$ with its canonical symplectic form in the statement of Theorem 2.1, where $`(Q,,)`$ is a Riemannian manifold on which the Lie group $`G`$ acts by isometries. The Hamiltonian action on $`T^{}Q`$ is the cotangent lift of the action on $`Q`$, and the Hamiltonian is given by (2.3) defining a simple mechanical system. Let $`\xi `$ be an element of the Lie algebra $`๐ค`$ of $`G`$. Fix $`p_x=๐ฝ\mathrm{L}(\xi _Q(x))`$ with momentum $`\mu `$. Assume that
* $`x`$ is a critical point of $`V_\xi `$, and
* $`G_\mu `$ is compact.
Then $`p_x`$ is a relative equilibrium for our simple mechanical system with velocity $`\xi `$ satisfying $`\xi ๐ค_\mu `$ and Theorem 2.1 can be applied to study its $`G_\mu `$-stability. That is, we need to determine when $`๐_{p_x}^2h_\xi ^{}\text{ }V_s`$ is definite. For that, we will use the characterization of the symplectic normal space $`V_s`$, given in (3.16) which establishes a linear isomorphism $`\kappa :๐ฎ^\mu ๐๐^{}V_sT_{p_x}T^{}Q(๐ฏ๐)(๐ฏ^{}๐^{})`$ explicitly expressed as
(4.2)
$$\begin{array}{ccc}\kappa (\lambda ,a,\gamma )=(\lambda ,a\hfill & ;\hfill & \frac{1}{2}_๐ฏ\left[(๐๐\xi _Q^{}(x))(\lambda )+\mathrm{ad}_\lambda ^{}\mu (๐๐a)(\xi ^{})\right],\hfill \\ & & \gamma \frac{1}{2}_๐\left[(๐๐())(\xi ^{},\lambda )\right]+C(a)(\xi ^{}),_๐).\hfill \end{array}$$
This map is $`G_{p_x}`$-equivariant with respect to the action on $`๐ฎ^\mu ๐๐^{}`$ given by
$$g(\lambda ,a,\gamma )=(\mathrm{Ad}_g\lambda ,ga,g\gamma ),$$
and the action (3.9) on $`(๐ฏ๐)(๐ฏ^{}๐^{})`$.
###### Definition 4.1.
At a relative equilibrium $`p_x=๐ฝ\mathrm{L}(\xi _Q(x))`$, with momentum $`\mu `$ velocity $`\xi ๐ค_\mu `$, define the linear subspace $`\mathrm{\Sigma }`$ of $`T_xQ`$ isomorphic to $`๐ฎ^\mu ๐`$, as
(4.3)
$$\mathrm{\Sigma }=\{\lambda _Q(x)+aT_xQ:\lambda ๐ฎ^\mu ,a๐\},$$
where $`๐ฎ^\mu `$ is defined in (3.15).
For any two vectors $`v_1,v_2T_xQ`$, define the correction term as the symmetric bilinear form on $`T_xQ`$ defined by
(4.4)
$$\mathrm{corr}_\xi (x)(v_1,v_2)=_๐ฏ\left[(๐๐v_1)(\xi )\right],\widehat{๐}_0^1\left(_๐ฏ\left[(๐๐v_2)(\xi )\right]\right).$$
We can now state the main result of this section.
###### Theorem 4.1 (reduced Energy-Momentum Method).
Suppose that $`p_x=๐ฝ\mathrm{L}(\xi _Q(x))`$ is a relative equilibrium of the simple mechanical system (2.3) with momentum $`\mu `$ and velocity $`\xi ๐ค_\mu `$. Assume that $`G_\mu `$ is compact, $`G_x=H`$ and that a $`(HG_\mu )`$-invariant splitting $`๐ค_\mu =๐ค_{p_x}๐ญ`$. Let $`\xi ^{}`$ be the orthogonal projection of $`\xi `$ onto $`๐ญ`$. Then if
* $`dimQdimG+dimG_x>0`$, and
* $`\left(๐_x^2V_\xi ^{}+\mathrm{corr}_\xi ^{}(x)\right)\text{ }\mathrm{\Sigma }`$ is positive definite
or
* $`dimQdimG+dimG_x=0`$, and
* $`\left(๐_x^2V_\xi ^{}+\mathrm{corr}_\xi ^{}(x)\right)\text{ }\mathrm{\Sigma }`$ is definite (positive or negative)
then the relative equilibrium is $`G_\mu `$-stable.
###### Proof.
According to Theorem 2.1 the relative equilibrium is stable provided $`๐_z^2h_\xi ^{}\text{ }V_s`$ is definite. As both $`K_\xi ^{}`$ and $`\overline{V_\xi ^{}}`$ have a critical point at $`p_x`$ ($`(iii)`$ in Proposition 4.1), and since $`h_\xi ^{}=K_\xi ^{}+\overline{V_\xi ^{}}`$, then
$$๐_{p_x}^2h_\xi ^{}\text{ }V_s=\left(๐_{p_x}^2K_\xi ^{}+๐_{p_x}^2\overline{V_\xi ^{}}\right)\text{ }V_s.$$
With the isomorphism $`\kappa `$ in (4.2) we can compute each of these Hessians in $`V_s`$ parameterized by elements in $`๐ฎ^\mu ๐๐^{}`$.
Let us compute first $`๐_{p_x}K_\xi ^{}\text{ }V_s`$. Let $`\lambda ,\lambda _1,\lambda _2๐ฎ^\mu ,a,a_1,a_2๐`$ and $`\beta ,\beta _1,\beta _2๐^{}`$. Note that using (4.2) we have that the vertical variation $`\delta p^V`$ corresponding to an element $`\lambda ๐ฎ^\mu `$, i.e. $`I(\delta p)=\kappa (\lambda ,0,0)`$, is
$$\delta p^V=(\frac{1}{2}_๐ฏ\left[(๐๐(\xi ^{})_Q(x))(\lambda )+\mathrm{ad}_\lambda ^{}\mu \right],\frac{1}{2}_๐[(๐๐())(\xi ^{},\lambda )]).$$
Using Lemma 4.2 we have
$$\delta p^V(T_x\chi ^\xi ^{}\delta p^H)^V=(_๐ฏ\left[(๐๐\lambda _Q(x))(\xi ^{})\right],0).$$
Similarly, if $`a๐`$ we have for $`I(\delta p)=\kappa (0,a,0)`$
$$\delta p^V=(\frac{1}{2}_๐ฏ\left[(๐๐a)(\xi ^{})\right],C(a)(\xi ^{}),_๐),$$
and then
$$\delta p^V(T_x\chi ^\xi ^{}\delta p^H)^V=(_๐ฏ\left[(๐๐a)(\xi ^{})\right],0).$$
Finally, if $`\beta ๐^{}`$ for $`I(\delta p)=\kappa (0,0,\beta )`$ we have $`\delta p^V=(0,\beta )`$ and $`\delta p^H=(0,0)`$ so
$$\delta p^V(T_x\chi ^\xi ^{}\delta p^H)^V=(0,\beta ).$$
According to (ii) in Lemma 4.1 we can now write, for $`\lambda ,\lambda _1,\lambda _2๐ฎ^\mu ,a,a_1,a_2๐`$ and $`\beta ,\beta _1,\beta _2๐^{}`$
$$\begin{array}{ccc}๐_{p_x}^2K_\xi ^{}(\kappa (\lambda _1,0,0),\kappa (\lambda _2,0,0))\hfill & =\hfill & _๐ฏ[(๐๐\lambda _{1}^{}{}_{Q}{}^{}(x))(\xi ^{})],\widehat{๐}_0^1_๐ฏ[(๐๐\lambda _{2}^{}{}_{Q}{}^{}(x))(\xi ^{})]\hfill \\ ๐_{p_x}^2K_\xi ^{}(\kappa (\lambda ,0,0),\kappa (0,a,0))\hfill & =\hfill & _๐ฏ[(๐๐\lambda _Q(x))(\xi ^{})],\widehat{๐}_0^1_๐ฏ[(๐๐a)(\xi ^{})]\hfill \\ ๐_{p_x}^2K_\xi ^{}(\kappa (0,a_1,0),\kappa (0,a_2,0))\hfill & =\hfill & _๐ฏ[(๐๐a_1)(\xi ^{})],\widehat{๐}_0^1_๐ฏ[(๐๐a_2)(\xi ^{})]\hfill \\ ๐_{p_x}^2K_\xi ^{}(\kappa (0,0,\beta _1),\kappa (0,0,\beta _2))\hfill & =\hfill & \beta _1,\beta _1_๐^{}\hfill \\ ๐_{p_x}^2K_\xi ^{}(\kappa (0,0,\beta ),\kappa (\lambda ,0,0))\hfill & =\hfill & 0\hfill \\ ๐_{p_x}^2K_\xi ^{}(\kappa (0,0,\beta ),\kappa (0,a,0))\hfill & =\hfill & 0.\hfill \end{array}$$
Where $`,_๐^{}`$ is the inner product in $`๐^{}`$ induced from $`,_๐`$ via the Riemannian Legendre map (2.2). We compute now the remaining contribution, the Hessian of the augmented potential energy. Using Lemma 4.1 it is immediate to obtain
$$\begin{array}{ccc}๐_{p_x}^2\overline{V_\xi ^{}}(\kappa (\lambda _1,a_1,0),\kappa (\lambda _2,a_2,0))\hfill & =\hfill & ๐_x^2V_\xi ^{}(\lambda _{1}^{}{}_{Q}{}^{}(x)+a_1,\lambda _{2}^{}{}_{Q}{}^{}(x)+a_2)\hfill \\ ๐_{p_x}^2\overline{V_\xi ^{}}(\kappa (\lambda ,a,0),\kappa (0,0,\beta ))\hfill & =\hfill & 0\hfill \\ ๐_{p_x}^2\overline{V_\xi ^{}}(\kappa (0,0,\beta _1),\kappa (0,0,\beta _2))\hfill & =\hfill & 0\hfill \end{array}$$
for every $`\lambda ,\lambda _1\lambda _2๐ฎ^\mu ,a,a_1,a_2๐,\beta ,\beta _1,\beta _2๐^{}`$.
Therefore, $`๐_{p_x}^2h_\xi ^{}\text{ }V_s`$ block-diagonalizes in the two blocks $`\kappa (๐ฎ^\mu ๐\{0\})`$ and $`\kappa (\{0\}\{0\}๐^{})`$ as
$$\begin{array}{ccc}๐_{p_x}^2h_\xi ^{}V_s(\kappa (\lambda _1,a_1,0),\kappa (\lambda _2,a_2,0))\hfill & =\hfill & \left(๐_x^2V_\xi ^{}+\mathrm{corr}_\xi ^{}(x)\right)(\lambda _{1}^{}{}_{Q}{}^{}(x)+a_1,\lambda _{2}^{}{}_{Q}{}^{}(x)+a_2)\hfill \\ ๐_{p_x}^2h_\xi ^{}V_s(\kappa (\lambda ,a,0),\kappa (0,0,\beta ))\hfill & =\hfill & 0\hfill \\ ๐_{p_x}^2h_\xi ^{}V_s(\kappa (0,0,\beta _1),\kappa (0,0,\beta _2))\hfill & =\hfill & \beta _1,\beta _1_๐^{}\hfill \end{array}$$
for every $`\lambda ,\lambda _1\lambda _2๐ฎ^\mu ,a,a_1,a_2๐,\beta ,\beta _1,\beta _2๐^{}`$. From the above expression and Definition 4.1 it follows that the bilinear form $`๐_{p_x}^2h_\xi ^{}\text{ }V_s`$ is equivalent to the pair of bilinear forms $`\left(๐_x^2V_\xi ^{}+\mathrm{corr}_\xi ^{}(x)\right)\text{ }\mathrm{\Sigma }`$ and $`,_๐^{}`$.
Let us examine now when $`๐_{p_x}^2h_\xi ^{}\text{ }V_s`$ is definite. Since the block $`,_๐^{}`$ is positive-definite or trivial, there are two possible scenarios. (i) $`dim๐=0`$ and (ii) $`dim๐>0`$. Note that $`dim๐=0`$ if and only if the dimension of the orbit $`Gx`$ equals $`dimQ`$. Recall also that $`dimGx=dimGdimG_x`$. In this case $`๐_{p_x}^2h_\xi ^{}\text{ }V_s`$ consists only in the block $`\left(๐_x^2V_\xi ^{}+\mathrm{corr}_\xi ^{}(x)\right)\text{ }\mathrm{\Sigma }`$ and the relative equilibrium is $`G_\mu `$-stable provided this block is definite (positive or negative).
For the other possibility, if $`dim๐0`$ then $`dimQ>dimGx`$ and the block $`,_๐^{}`$ is positive definite since the metric on $`Q`$ is Riemannian, so $`๐_{p_x}^2h_\xi ^{}\text{ }V_s`$ is definite if and only if the block $`\left(๐_x^2V_\xi ^{}+\mathrm{corr}_\xi ^{}(x)\right)\text{ }\mathrm{\Sigma }`$ is positive definite. This completes the proof of the theorem. โ
### Remarks.
* It is easy to see that the above theorem particularizes in the regular case ($`G_{p_x}=G_x=0`$) to the main result of the reduced Energy-Momentum Method, (see page 35 in ). In the regular case, one has to test the stability of the Hessian $`๐_x^2V_\mu `$ restricted to $`(๐ค_\mu x)^{}`$, where $`\mu `$ is the momentum value of the relative equilibrium under study and $`V_\mu `$ is Smaleโs *amended potential energy* (see ), defined as
$$V_\mu (x)=V(x)+\frac{1}{2}\mu ,๐^1(x)(\mu ).$$
Obviously this function is not well defined at those points $`xQ`$ such that $`dim๐ค_x>0`$, since the locked inertia tensor is not invertible. Hence, it is not possible to define the Hessian of the amended potential in the singular setting. However, the following relation is to be noted (see , equation (2.28)), which holds in the regular case at a relative equilibrium
$$๐_x^2V_\mu (v_1,v_2)=๐_x^2V_\xi (v_1,v_2)+(๐๐v_1)(\xi ),๐^1(x)\left[(๐๐v_2)(\xi )\right].$$
This suggests that what we are testing in Theorem 4.1 is exactly the singular analogue of the Hessian of the amended potential, even when we cannot talk about the amended potential itself.
* In the regular case, according to our definition (3.14) of $`๐ฑ`$, $`\mathrm{\Sigma }=(๐ค_\mu x)^{}`$ with respect to $`,`$. In the presence of isotropy for the base point $`x`$ of our relative equilibria, it can be seen that $`\mathrm{\Sigma }`$ is the orthogonal complement to $`๐ค_\mu x`$ within the space of admissible variations (see (7.1) and the proof of Proposition 6.1). Therefore, conditions $`(ii)`$ in Theorem 4.1 are tested in a space $`\mathrm{\Sigma }`$ which is orthogonally complementary (with respect to the kinetic energy metric) to the drift orbit $`G_\mu x`$.
## 5. An example: The sleeping Lagrange top
We will apply our singular version of the reduced Energy-Momentum Method to the study of the relative equilibrium known as the *sleeping Lagrange top*. This problem has been extensively studied in the literature of Classical Mechanics; see in particular for a geometric perspective of the problem using the Hamiltonian and symplectic formalism. Here we show the advantages of the method stated in Theorem 4.1 when studying stability in simple mechanical systems. Indeed, taking into account the extra cotangent bundle structure of the problem actually leads to simpler calculations for obtaining stability results when compared with the non-adapted methods constructed for general symmetric Hamiltonian systems.
The Lagrange top is a symmetric simple mechanical system defined on the configuration space $`Q=\mathrm{SO}(3)`$, in the usual representation by orthogonal $`3\times 3`$ real matrices with determinant 1, and being the symmetry group $`G=๐^2`$. We use the right trivialization for $`T\mathrm{SO}(3)`$ given by the isomorphism $`T\mathrm{SO}(3)๐ฐ๐ฌ(3)\times \mathrm{SO}(3)`$, i.e. $`\delta g=\xi g`$ with $`\xi ๐ค`$, and identifying $`๐ฐ๐ฌ(3)`$ with $`^3`$ under the inverse of the usual isomorphism given by $`^3๐ฎ\widehat{๐ฎ}๐ฐ๐ฌ(3)`$ (as a $`3\times 3`$ matrix algebra), defined by $`\widehat{๐ฎ}๐ฏ=๐ฎ\times ๐ฏ,๐ฏ^3`$. Analogous considerations hold to obtain the trivialization of the phase space for the problem: $`T^{}\mathrm{SO}(3)^3\times \mathrm{SO}(3)`$. Then we can write the Hamiltonian for this system as
(5.1)
$$H(\pi ,\mathrm{\Lambda })=\frac{1}{2}\pi E_\mathrm{\Lambda }^1\pi +mgl\mathrm{\Lambda }๐_\mathrm{๐}๐_\mathrm{๐},$$
where $`\mathrm{\Lambda }Q,\pi ^3`$, and $`(\pi ,\mathrm{\Lambda })T_\mathrm{\Lambda }^{}Q,E=\mathrm{diag}(i,i,i_3)`$ and $`E_\mathrm{\Lambda }=\mathrm{\Lambda }E\mathrm{\Lambda }^t`$. Finally, $`m,g,l`$ are physical constants of the model.
The group $`๐^2`$ is identified with $`S^1\times S^1`$ where, in our matricial representation, both copies of the circle group act by rotations around $`๐_\mathrm{๐}`$. Its action on $`Q`$ is given by $`(L,R)\mathrm{\Lambda }=L\mathrm{\Lambda }R^t`$ and the action on the phase space is by cotangent lifts. We do not actually need the (simple) expression for the cotangent-lifted action since our methods rely finally on the geometry of the action of $`G`$ on $`Q`$ once they were constructed in this spirit. Let $`(l,r)^2=\mathrm{Lie}(๐^2)`$, then the infinitesimal action of $`^2`$ on $`Q`$ is given in the right trivialization of $`T\mathrm{SO}(3)`$ by $`(l,r)_Q(\mathrm{\Lambda })=(l๐_\mathrm{๐}r\mathrm{\Lambda }๐_\mathrm{๐},\mathrm{\Lambda })`$.
To finish with these preliminaries, let us show the Riemannian structure of $`Q`$ associated to the kinetic energy of the problem. If $`V_1=(v_1,\mathrm{\Lambda })`$ and $`V_2=(v_2,\mathrm{\Lambda })`$ are two tangent vectors to $`Q`$ at $`\mathrm{\Lambda }`$, then
(5.2)
$$V_1,V_2(\mathrm{\Lambda })=v_1E_\mathrm{\Lambda }v_2$$
which is easily checked to be a $`๐^2`$-invariant symmetric contravariant bilinear tensor on $`Q`$. Note also from the expression (5.1) of the Hamiltonian, that the potential energy is given by $`V(\mathrm{\Lambda })=mgl\mathrm{\Lambda }๐_\mathrm{๐}๐_\mathrm{๐}`$.
It is well known that the phase space point $`p_I=๐ฝ\mathrm{L}((l,r)_Q(I))`$, where $`I`$ is the identity $`3\times 3`$ matrix, is a relative equilibrium of this system, known as the sleeping Lagrange top. Our aim is to study its stability. First of all, note that since $`\mathrm{\Lambda }=I`$, then $`(l,r)_Q(I)=(\zeta ๐_\mathrm{๐},I)`$, where the number $`\zeta =lr`$ uniquely determines the infinitesimal generator, and it is physically interpreted as the angular velocity of the rigid body modelled by this system.
The first thing we need to compute is the locked inertia tensor. It easily follows from the previous expressions that
$$\begin{array}{c}(l_1,r_1)_Q(\mathrm{\Lambda }),(l_2,r_2)_Q(\mathrm{\Lambda })=(l_1๐_\mathrm{๐}r_1\mathrm{\Lambda }๐_\mathrm{๐})E_\mathrm{\Lambda }(l_2๐_\mathrm{๐}r_2\mathrm{\Lambda }๐_\mathrm{๐})\hfill \\ \text{ }\hfill \\ =(l_1,r_1)\left(\begin{array}{cc}๐_\mathrm{๐}E_\mathrm{\Lambda }๐_\mathrm{๐}& ๐_\mathrm{๐}\mathrm{\Lambda }E๐_\mathrm{๐}\\ ๐_\mathrm{๐}E\mathrm{\Lambda }^t๐_\mathrm{๐}& ๐_\mathrm{๐}E๐_\mathrm{๐}\end{array}\right)\left(\begin{array}{c}l_2\\ r_2\end{array}\right)\hfill \end{array}$$
and then
$$๐(\mathrm{\Lambda })=\left(\begin{array}{cc}๐_\mathrm{๐}E_\mathrm{\Lambda }๐_\mathrm{๐}& ๐_\mathrm{๐}\mathrm{\Lambda }E๐_\mathrm{๐}\\ ๐_\mathrm{๐}E\mathrm{\Lambda }^t๐_\mathrm{๐}& ๐_\mathrm{๐}E๐_\mathrm{๐}\end{array}\right).$$
Thus at the configuration $`\mathrm{\Lambda }=I`$ of our relative equilibrium we have
$$๐(I)=\left(\begin{array}{cc}i_3& i_3\\ i_3& i_3\end{array}\right)$$
and so the momentum value of the relative equilibrium is
$$\mu =๐(I)(l,r)=i_3(\zeta ,\zeta ).$$
As the configuration point of our relative equilibrium is the identity element of $`\mathrm{SO}(3)`$, then its isotropy group is $`H=G_I=S^1`$, regarded as the diagonal embedding of $`S^1`$ in the 2-torus. For the momentum isotropy, just by noting that the group is Abelian and thus the coadjoint representation is trivial, we obtain $`G_\mu =๐^2`$. Finally, using the characterization $`G_{p_x}=G_xG_\mu `$ we also get $`G_{p_x}=G_x=S^1`$.
The next step is to choose a $`G_{p_x}`$-invariant complement to $`๐ค_{p_x}`$ in $`๐ค_\mu =^2`$ in order to implement the splitting (3.13) and obtain the velocity $`(l,r)^{}`$. We will obtain it by orthogonality with respect to a $`G_{p_x}`$-invariant inner product in $`๐ค_\mu `$. Since the adjoint action is also trivial any inner product is $`S^1`$-invariant. We can choose the family of inner products
$$๐=\left(\begin{array}{cc}k& 0\\ 0& 1\end{array}\right).$$
Since $`๐ค_{p_x}`$ is generated by $`(1,1)`$, its orthogonal complement $`๐ญ`$ with respect to $`๐`$ is generated by the normalized vector $`๐ค=\frac{1}{\sqrt{k(1+k)}}(1,k)`$, so for the infinitesimal generator of our relative equilibrium, $`\xi =(l,r)`$, we have that $`\xi ^{}`$ is the orthogonal projection of $`\xi `$ onto $`๐ญ`$, i.e. $`\xi ^{}=๐(\xi ,๐ค)๐ค=\zeta (\frac{1}{1+k},\frac{k}{1+k})`$.
The augmented potential energy $`V_\xi ^{}(\mathrm{\Lambda })=V(\mathrm{\Lambda })\frac{1}{2}๐(\mathrm{\Lambda })(\xi ^{},\xi ^{})`$ corresponding to this relative equilibrium is now written as
$$V_\xi ^{}(\mathrm{\Lambda })=mgl\mathrm{\Lambda }๐_\mathrm{๐}๐_\mathrm{๐}\frac{\zeta ^2}{2(1+k)^2}\left[๐_\mathrm{๐}E_\mathrm{\Lambda }๐_\mathrm{๐}+2ki_3๐_\mathrm{๐}\mathrm{\Lambda }๐_\mathrm{๐}+k^2i_3\right].$$
We now compute the derivative of the augmented potential. According to our right trivialization, a tangent vector $`\delta \mathrm{\Lambda }T_\mathrm{\Lambda }\mathrm{SO}(3)`$ can be written as $`\delta \mathrm{\Lambda }=\widehat{ฯต}\mathrm{\Lambda }`$, where $`ฯต`$ is a vector in $`^3`$. Then, after some vector calculus manipulations we get
(5.3)
$$๐V_\xi ^{}(\mathrm{\Lambda })\delta \mathrm{\Lambda }=mgl๐_\mathrm{๐}(ฯต\times \mathrm{\Lambda }๐_\mathrm{๐})\frac{\zeta ^2}{(1+k)^2}\left[ki_3๐_\mathrm{๐}(ฯต\times \mathrm{\Lambda }๐_\mathrm{๐})๐_\mathrm{๐}E_\mathrm{\Lambda }(ฯต\times ๐_\mathrm{๐})\right]$$
which vanishes at $`\mathrm{\Lambda }=I`$, since $`๐ฝ\mathrm{L}((l,r)_Q(I))`$ is a relative equilibrium. The Hessian of $`V_\xi ^{}`$ at $`\mathrm{\Lambda }=I`$ is
$$\begin{array}{c}๐_I^2V_\xi ^{}(\delta \mathrm{\Lambda }_1,\delta \mathrm{\Lambda }_2)=(ฯต_1\times ๐_\mathrm{๐})(ฯต_2\times ๐_\mathrm{๐})\left(\frac{(ki_3+i_3)\zeta ^2}{(1+k)^2}mgl\right)\hfill \\ \frac{\zeta ^2}{(1+k)^2}(ฯต_1\times ๐_\mathrm{๐})E(ฯต_2\times ๐_\mathrm{๐}).\hfill \end{array}$$
A straightforward computation shows that the correction term (4.4) vanishes.
As the group is Abelian, then $`๐ค_\mu =๐ค`$ and $`๐ฎ^\mu =(0,0)`$, so $`\mathrm{\Sigma }=๐`$. Then by Theorem 4.1 the relative equilibrium is $`G_\mu `$-stable if $`๐_I^2V_\xi ^{}\text{ }๐`$ is positive-definite. It is then necessary to obtain the linear slice for the toral action on $`\mathrm{SO}(3)`$ at $`\mathrm{\Lambda }=I`$ with respect to the Riemannian metric (5.2). A tangent vector $`\delta \mathrm{\Lambda }T_I\mathrm{SO}(3)`$ written as $`\delta \mathrm{\Lambda }=\widehat{ฯต}I`$ is orthogonal to $`๐คI`$ if and only if
$$ฯตE๐_\mathrm{๐}=0$$
and thus we have
$$๐=\{\delta \mathrm{\Lambda }T_I\mathrm{SO}(3):ฯต\mathrm{span}\{(1,0,0),(0,1,0)\}\}.$$
To check the positive definiteness of $`๐_I^2V_\xi ^{}\text{ }๐`$ is then equivalent to showing that the $`2\times 2`$ matrix
(5.4)
$$\left(\frac{(ki_3+i_3i)\zeta ^2}{(1+k)^2}mgl\right)I_2$$
is positive-definite, where $`I_2`$ is the identity matrix in $`^2`$; so the condition for stability is satisfied if
(5.5)
$$\zeta ^2>\frac{(1+k)^2mgl}{ki_3+i_3i}$$
### Remark.
The above expression is exactly the one obtained in , page 718, by using the (singular) Energy-Momentum Method for general Hamiltonian systems in arbitrary symplectic manifolds, i.e. Theorem 2.1 in this paper. In that work, the same condition for the $`๐^2`$-stability of the sleeping Lagrange top is obtained after computing algebraically the eigenvalues of a $`4\times 4`$ matrix ($`๐_{p_x}^2h_\xi \text{ }V_s`$), which was not put in block-diagonal form by applying their general stability criterion. With our method the same expression follows easily from the unique eigenvalue of the scalar matrix (5.4). This shows the potential power of employing methods adapted to the cotangent bundle structure in the study of simple mechanical systems. In particular in practical stability problems with higher dimensional configuration spaces the implementation of Theorem 2.1 could involve checking the definiteness of much more complicated matrices, forcing the use of numerical methods in some situations where the reduced Energy-Momentum Method developed in this paper (Theorem 4.1 and also Corollary 6.2) could offer simpler or even exact results.
For the sake of completeness we will sharpen the stability condition (5.5), following . Since in the stability condition $`k`$ appears, which is related to $`๐`$, the sharpest (or optimal) stability condition will be the lowest value of $`\zeta ^2`$ among all possible $`k`$. This is an straightforward optimization problem in elementary calculus, so by differentiating the expression
$$f(k)=\frac{(1+k)^2mgl}{ki_3+i_3i},$$
we find that the minimum value is reached at $`k=\frac{2ii_3}{i_3}`$. So the sharpest stability condition yields the well-known lower bound for the angular velocity
$$\zeta ^2>\frac{4mgli}{i_3^2}.$$
## 6. The singular Arnold form and Block-Diagonalization
In the previous section we used the realization of the symplectic normal space $`V_s`$ at a relative equilibrium given in (3.16) and which has been shown to be isomorphic to $`๐ฎ^\mu ๐๐^{}`$ by the isomorphism $`(\text{4.2})`$. This was helpful to simplify the study of the definiteness of $`๐_{p_x}^2h_\xi ^{}\text{ }V_s`$, obtaining a block-diagonal structure with one block being positive-definite or trivial, reducing the problem to study the definiteness of $`\left(๐_x^2V_\xi ^{}+\mathrm{corr}_\xi ^{}(x)\right)\text{ }\mathrm{\Sigma }`$. In this section we pursue the study of the symplectic normal space $`V_s`$ by obtaining new block-diagonal forms for $`๐_{p_x}^2h_\xi ^{}\text{ }V_s`$ and also for the symplectic matrix $`\mathrm{\Omega }`$ at a relative equilibrium. Recall that in the $`I`$-representation, the symplectic matrix $`\omega (p_x)`$ of $`T_{p_x}(T^{}Q)`$ has the following form (see and ). If $`(\eta _1,a_1;\nu _1,\alpha _1),(\eta _2,a_2;\nu _2,\alpha _2)I(T_{p_x}(T^{}Q))=(๐ฏ๐)(๐ฏ^{}๐^{})`$ then
(6.1)
$$\mathrm{\Omega }((\eta _1,a_1;\nu _1,\alpha _1),(\eta _2,a_2;\nu _2,\alpha _2))=\nu _2,\eta _1+\alpha _2,a_1\nu _1,\eta _2\alpha _1,a_2.$$
In the following we will use a parametrization of $`V_s`$ different from (4.2). This, when available, will present the extra advantage of putting $`๐_{p_x}^2h_\xi ^{}\text{ }V_s`$ in block-diagonal form consisting of three blocks instead of two, as in Theorem 4.1, hence simplifying the stability analysis further. We will start by defining a singular analogue of the Arnold form, (see and to see how the Arnold form arises in the study of the stability of regular relative equilibria of simple mechanical systems). Hereafter, we assume that we will be working under the same conditions and hypotheses as in the previous section. In particular, we will always have $`p_x=๐ฝ\mathrm{L}(\xi _Q(x))`$ as a relative equilibrium of the simple mechanical system (2.3) with momentum $`\mu `$ and base point isotropy $`G_x=H`$.
###### Definition 6.1.
The singular Arnold form at a relative equilibrium with base point $`x`$ and momentum $`\mu `$ is the bilinear form on $`๐ฎ^\mu `$, $`\mathrm{Ar}:๐ฎ^\mu \times ๐ฎ^\mu `$ defined by
$$\mathrm{Ar}(\lambda _1,\lambda _2)=\mathrm{ad}_{\lambda _1}^{}\mu ,\mathrm{\Lambda }(x,\mu )(\lambda _2),$$
where the map $`\mathrm{\Lambda }(x,\mu ):๐ฎ^\mu ๐ฏ`$ is defined by
$$\mathrm{\Lambda }(x,\mu )(\lambda )=\widehat{๐}_0^1\left(\mathrm{ad}_\lambda ^{}\mu \right)+_๐ฏ^{}\left[\mathrm{ad}_\lambda \left(\widehat{๐}_0^1\mu \right)\right].$$
### Remark.
To the best of our knowledge, the first time a singular analogue of the Arnold form appears in the literature is in , equation (3.56), under the name of โgeneralized Arnold formโ. In that work, this object is defined in the context of general Lagrangian systems and the Lagrangian Block-Diagonalization method. Also see Section 7 for a comparison of other results in .
We will now define another two spaces which will be useful for the obtention of block-diagonal expressions. The motivation for the introduction of these spaces is, following the ideas in , that provided a non-degeneracy condition is satisfied, $`๐_{p_x}^2h_\xi ^{}\text{ }V_s`$ will further block-diagonalized, refining the conditions of Theorem 4.1.
Recalling the identification $`\kappa :๐ฎ^\mu ๐๐^{}V_s`$, let us define the following subspace of $`๐ฎ^\mu ๐`$:
(6.2)
$$w_{\text{int}}=\{(\lambda ^b,b)๐ฎ^\mu ๐:(๐๐(\lambda _Q^b(x)+b))(\xi ^{})๐ญ^{}\}.$$
Also, using the map $`(\lambda ,b)\lambda _Q(x)+b`$ which maps isomorphically $`๐ฎ^\mu ๐`$ onto $`\mathrm{\Sigma }T_xQ`$, we define the following subspaces of $`\mathrm{\Sigma }`$:
(6.3)
$$\mathrm{\Sigma }_{\text{rig}}=\{\lambda _Q(x):\lambda ๐ฎ^\mu \},\mathrm{and}$$
(6.4)
$$\mathrm{\Sigma }_{\text{int}}=\{\lambda _Q^b(x)+b:(\lambda ^b,b)w_{\text{int}}\}.$$
Note that we have the following obvious identifications:
$$\begin{array}{ccc}\mathrm{\Sigma }_{\text{rig}}\hfill & \hfill & ๐ฎ^\mu \hfill \\ \mathrm{\Sigma }_{\text{int}}\hfill & \hfill & w_{\mathrm{int}}.\hfill \end{array}$$
Following we can give the following interpretation for these two spaces: $`\mathrm{\Sigma }`$ is seen as the space of all admissible variations orthogonal to the infinitesimal drift directions $`๐ค_\mu x`$. This space has a contribution $`\mathrm{\Sigma }_{\text{rig}}`$ which corresponds to variations in $`\mathrm{\Sigma }`$ which generate group motions, i.e. regarding our systems as a โrigid bodyโ without internal structure. On the contrary, the subspace $`\mathrm{\Sigma }_{\text{int}}`$ corresponds to all the variations of our system that are purely internal, i.e. variations in โshapeโ, not coming from the symmetry group.
###### Proposition 6.1.
If the Arnold form is non-degenerate then $`\mathrm{\Sigma }=\mathrm{\Sigma }_{\mathrm{rig}}\mathrm{\Sigma }_{\mathrm{int}}.`$
For the proof of this proposition, we will need the following lemma
###### Lemma 6.1.
For every $`\lambda ๐ฎ^\mu `$ and $`v๐`$
* $`_๐ฅ\left[(๐๐\lambda _Q(x))(\xi ^{})\right]=0`$
* $`_๐ฅ\left[(๐๐v)(\xi ^{})\right]=0`$.
###### Proof.
For (i), using (3.3), for any $`\zeta ๐ฅ`$ one has
$$\begin{array}{ccc}(๐๐\lambda _Q(x))(\xi ^{},\zeta )\hfill & =\hfill & ๐(x)(\mathrm{ad}_\lambda \xi ^{},\zeta )๐(\xi ^{},\mathrm{ad}_\lambda \zeta )\hfill \\ & =\hfill & \mathrm{ad}_\lambda ^{}\mu ,\zeta =0,\hfill \end{array}$$
where the second equality follows since $`\mathrm{ker}๐(x)=๐ฅ`$ and $`๐(x)(\xi ^{})=\mu `$.
For (ii), making $`\lambda =\zeta ๐ฅ`$ in item (iii) of Theorem 3.1, we have $`(๐๐v)(\xi ^{},\zeta )=0`$ for every $`\zeta ๐ฅ`$. โ
###### Proof.
(of the proposition) It is clear that $`\mathrm{\Sigma }_{\mathrm{rig}}\mathrm{\Sigma }`$ and $`\mathrm{\Sigma }_{\mathrm{int}}\mathrm{\Sigma }`$ so we have to prove $`\mathrm{\Sigma }_{\mathrm{rig}}\mathrm{\Sigma }_{\mathrm{int}}=0`$ and $`\mathrm{\Sigma }_{\mathrm{rig}}+\mathrm{\Sigma }_{\mathrm{int}}=\mathrm{\Sigma }`$. Let $`0\lambda ๐ฎ^\mu `$. Then $`\lambda _Q(x)\mathrm{\Sigma }_{\mathrm{rig}}\mathrm{\Sigma }_{\mathrm{int}}`$ if and only if $`(๐๐(x)\lambda _Q(x))(\xi ^{},ฯต)=0`$ for every $`ฯต๐ฑ+๐ฅ`$. By $`(i)`$ in Lemma 6.1, this holds if and only if the same condition is satisfied for every $`ฯต๐ฑ`$. Using (3.3), this is equivalent to
$$\begin{array}{ccc}0\hfill & =\hfill & ๐(x)(\mathrm{ad}_\lambda \xi ^{},ฯต)+๐(x)(\xi ^{},\mathrm{ad}_\lambda ฯต)=๐(x)\left(\mathrm{ad}_\lambda \left(\widehat{๐}_0^1\mu \right)\right)+\mathrm{ad}_\lambda ^{}\mu ,ฯต\hfill \\ & =\hfill & \widehat{๐}_0\left(_๐ฏ^{}\left[\mathrm{ad}_\lambda \left(\widehat{๐}_0^1\mu \right)\right]\right)+\mathrm{ad}_\lambda ^{}\mu ,ฯต=\widehat{๐}_0\left(_๐ฏ^{}\left[\left(\mathrm{ad}_\lambda \widehat{๐}_0^1\mu \right)\right]+\widehat{๐}_0^1(\mathrm{ad}_\lambda ^{}\mu )\right),ฯต\hfill \end{array}$$
for every $`ฯต๐ฑ`$, regarding that $`\mathrm{ad}_\lambda ^{}\mu ๐ฏ`$ if $`\lambda ๐ฎ^\mu `$. Since $`\widehat{๐}_0`$ is an isomorphism, this condition is the same as $`\mathrm{\Lambda }(x,\mu )(\lambda )๐ญ๐ค_\mu `$, but then the Arnold form would be degenerate, which is a contradiction.
To prove that $`\mathrm{\Sigma }_{\mathrm{rig}}+\mathrm{\Sigma }_{\mathrm{int}}=\mathrm{\Sigma }`$ let us note the following: if we call
$$๐=\{\delta qT_xQ:_๐ฅ\left[(๐๐\delta q)(\xi ^{})\right]=0\},$$
then by the definitions of $`๐ฑ`$ (3.14), $`๐ฎ^\mu `$ (3.15), and by Lemma 6.1 we have that $`\mathrm{\Sigma }`$ is precisely the orthogonal complement to $`๐ค_\mu x`$ in $`๐`$. The rest of the proof is then a consequence of Proposition 3.7 in . โ
###### Corollary 6.1.
If the Arnold form is non-degenerate then the map $`\stackrel{~}{\kappa }:๐ฎ^\mu \mathrm{\Sigma }_{\mathrm{int}}๐^{}V_s`$ defined as
(6.5)
$$\stackrel{~}{\kappa }(\lambda ,(\lambda _Q^a(x)+a),\gamma )=\kappa (\lambda +\lambda ^a,a,\gamma )$$
for every $`\lambda ๐ฎ^\mu ,(\lambda ^a,a)w_{\mathrm{int}}`$ and $`\gamma ๐^{}`$ is a $`G_{p_x}`$-equivariant isomorphism.
We will assume now that the Arnold form is non-degenerate and then we will study the symplectic matrix $`\mathrm{\Omega }`$ of $`V_s`$ and $`๐_{p_x}^2h_\xi ^{}`$.
###### Proposition 6.2 (Block-Diagonalization forms).
If the Arnold form is non-degenerate, under the isomorphism $`\stackrel{~}{\kappa }`$ of Corollary 6.1 we have the following expressions for the symplectic matrix $`\mathrm{\Omega }`$ and $`๐_{p_x}^2h_\xi ^{}\text{ }V_s`$:
(6.6)
$$\begin{array}{ccccc}& & ๐ฎ^\mu & \mathrm{\Sigma }_{\mathrm{int}}& ๐^{}\\ \mathrm{\Omega }& =& (\begin{array}{c}\mathrm{\Xi }\\ \mathrm{\Psi }^t\\ 0\end{array}& \begin{array}{c}\mathrm{\Psi }\\ S_\mu \\ \mathrm{๐}\end{array}& \begin{array}{c}0\\ \mathrm{๐}\\ 0\end{array})\end{array}$$
and
(6.7)
$$\begin{array}{ccccc}& & ๐ฎ^\mu & \mathrm{\Sigma }_{\mathrm{int}}& ๐^{}\\ ๐_{p_x}^2h_\xi ^{}V_s& =& (\begin{array}{c}\mathrm{Ar}\\ 0\\ 0\end{array}& \begin{array}{c}0\\ (๐_x^2V_\xi ^{}+\mathrm{corr}_\xi ^{}(x))\mathrm{\Sigma }_{\mathrm{int}}\\ 0\end{array}& \begin{array}{c}0\\ 0\\ ,_๐^{}\end{array})\end{array},$$
where the entries of $`\mathrm{\Omega }`$ are
$$\begin{array}{ccc}\hfill \mathrm{\Xi }(\lambda _1,\lambda _2)& =\hfill & \mu ,\mathrm{ad}_{\lambda _1}\lambda _2\hfill \\ \hfill \mathrm{\Psi }(\lambda ,(\lambda _Q^b(x)+b))& =\hfill & \mu ,\mathrm{ad}_\lambda \lambda ^b\hfill \\ \hfill S_\mu ((\lambda _Q^a(x)+a),(\lambda _Q^b(x)+b))& =\hfill & \mu ,\mathrm{ad}_{\lambda ^a}\lambda ^b+C(b)(\xi ^{}),a_๐\hfill \\ & & C(a)(\xi ^{}),b_๐.\hfill \end{array}$$
###### Proof.
The form for $`\mathrm{\Omega }`$ follows trivially from (6.1) and the definition of $`\stackrel{~}{k}`$ from (6.5). For $`๐_{p_x}^2h_\xi ^{}\text{ }V_s`$, and recalling its block-diagonal form showed in the proof of Theorem 4.1, the only two things that we must check are
* $`(๐_x^2V_\xi ^{}+\mathrm{corr}_\xi ^{}(x))(\lambda _{1}^{}{}_{Q}{}^{}(x),\lambda _{2}^{}{}_{Q}{}^{}(x))=\mathrm{Ar}(\lambda _1,\lambda _2)`$, and
* $`(๐_x^2V_\xi ^{}+\mathrm{corr}_\xi ^{}(x))(\lambda _Q(x),\delta q)=0`$,
for $`\lambda ,\lambda _1,\lambda _2๐ฎ^\mu `$ and $`\delta q\mathrm{\Sigma }_{\mathrm{int}}`$.
To prove (i) recall that the potential energy $`V`$ is $`G`$-invariant and then
$$(๐_x^2V_\xi ^{}+\mathrm{corr}_\xi ^{}(x))(\lambda _{1}^{}{}_{Q}{}^{}(x),\lambda _{2}^{}{}_{Q}{}^{}(x))=\left(\frac{1}{2}๐_x^2\left(๐(x)(\xi ^{},\xi ^{})\right)+\mathrm{corr}_\xi ^{}(x)\right)(\lambda _{1}^{}{}_{Q}{}^{}(x),\lambda _{2}^{}{}_{Q}{}^{}(x)).$$
We compute now both terms in the right hand side of the above expression. For the first one,
$$\begin{array}{ccc}\frac{1}{2}๐_x^2\left(๐(x)(\xi ^{},\xi ^{})\right)(\lambda _{1}^{}{}_{Q}{}^{}(x),\lambda _{2}^{}{}_{Q}{}^{}(x))\hfill & =\hfill & \frac{1}{2}\lambda _{2}^{}{}_{Q}{}^{}((๐๐\lambda _{1}^{}{}_{Q}{}^{}(x))(\xi ^{},\xi ^{}))(x)\hfill \\ & =\hfill & \lambda _{2}^{}{}_{Q}{}^{}(๐(\mathrm{ad}_{\lambda _1}\xi ^{},\xi ^{}))(x)\hfill \\ & =\hfill & (๐๐\lambda _{2}^{}{}_{Q}{}^{}(x))(\mathrm{ad}_{\lambda _1}\xi ^{},\xi ^{})\hfill \\ & =\hfill & ๐(x)(\mathrm{ad}_{\lambda _2}(\mathrm{ad}_{\lambda _1}\xi ^{}),\xi ^{})\hfill \\ & & ๐(x)(\mathrm{ad}_{\lambda _1}\xi ^{},\mathrm{ad}_{\lambda _2}\xi ^{}).\hfill \end{array}$$
Now for the second,
$$\begin{array}{ccc}\mathrm{corr}_\xi ^{}(x)(\lambda _{1}^{}{}_{Q}{}^{}(x),\lambda _{2}^{}{}_{Q}{}^{}(x))\hfill & =\hfill & _๐ฏ\left[(๐๐\lambda _{1}^{}{}_{Q}{}^{}(x))(\xi ^{})\right],\widehat{๐}_0^1_๐ฏ\left[(๐๐\lambda _{2}^{}{}_{Q}{}^{}(x))(\xi ^{})\right]\hfill \\ & =\hfill & ๐(x)(\mathrm{ad}_{\lambda _1}\xi ^{})+\mathrm{ad}_{\lambda _1}^{}\mu ,\widehat{๐}_0^1\left[๐(x)(\mathrm{ad}_{\lambda _2}\xi ^{})+\mathrm{ad}_{\lambda _2}^{}\mu \right]\hfill \\ & =\hfill & ๐(x)(\mathrm{ad}_{\lambda _2}\xi ^{},\mathrm{ad}_{\lambda _1}\xi ^{})+\mathrm{ad}_{\lambda _1}^{}\mu ,\mathrm{ad}_{\lambda _2}\xi ^{}\hfill \\ & & +\mathrm{ad}_{\lambda _2}^{}\mu ,\mathrm{ad}_{\lambda _1}\xi ^{}+\mathrm{ad}_{\lambda _1}^{}\mu ,\widehat{๐}_0^1(\mathrm{ad}_{\lambda _2}^{}\mu ),\hfill \end{array}$$
where we have used that $`๐(x)(\eta _1,\eta _2)=๐(x)(\eta _1^๐ฏ,\eta _2^๐ฏ)`$ $`\eta _1,\eta _2๐ค,`$ since $`\mathrm{ker}๐(x)=๐ฅ`$. Also, $`\mathrm{ad}_\lambda ^{}\mu ,\eta =\mathrm{ad}_\lambda ^{}\mu ,\eta ^๐ฏ`$ $`\eta ๐ค,`$ since $`\lambda ๐ฎ^\mu `$, which means that $`\mathrm{ad}_\lambda ^{}\mu ๐ฅ^{}`$. Finally, noticing that
$$\mathrm{ad}_{\lambda _2}^{}\mu ,\mathrm{ad}_{\lambda _1}\xi ^{}=๐(x)(\xi ^{},\mathrm{ad}_{\lambda _2}(\mathrm{ad}_{\lambda _1}\xi ^{})),$$
and putting both contributions together, we obtain
$$\begin{array}{ccc}(๐_x^2V_\xi ^{}+\mathrm{corr}_\xi ^{}(x))(\lambda _{1}^{}{}_{Q}{}^{}(x),\lambda _{2}^{}{}_{Q}{}^{}(x))\hfill & =\hfill & \mathrm{ad}_{\lambda _1}^{}\mu ,\mathrm{ad}_{\lambda _2}\xi ^{}+\widehat{๐}_0^1(\mathrm{ad}_{\lambda _2}^{}\mu )\hfill \\ & =\hfill & \mathrm{ad}_{\lambda _1}^{}\mu ,_๐ฏ\left[\mathrm{ad}_{\lambda _2}\left(\widehat{๐}_0^1\mu \right)\right]+\widehat{๐}_0^1(\mathrm{ad}_{\lambda _2}^{}\mu )\hfill \\ & =\hfill & \mathrm{Ar}(\lambda _1,\lambda _2),\hfill \end{array}$$
since $`\mathrm{ad}_{\lambda _1}^{}\mu ๐ฅ^{}`$ and $`\xi ^{}=\widehat{๐}_0^1\mu `$.
Again for (ii) because the potential energy $`V`$ is $`G`$-invariant we have
$$\begin{array}{c}\left(๐_x^2V_\xi ^{}+\mathrm{corr}_\xi ^{}(x)\right)(\lambda _Q(x),\lambda _Q^a(x)+a)=\hfill \\ =\left(\frac{1}{2}๐_x^2\left(๐(x)(\xi ^{},\xi ^{})\right)+\mathrm{corr}_\xi ^{}(x)\right)(\lambda _Q(x),\lambda _Q^a(x)+a).\hfill \end{array}$$
We will start by computing the contribution of the correction term:
$$\begin{array}{ccc}\mathrm{corr}_\xi ^{}(x)(\lambda ,\lambda _Q^a(x)+a)\hfill & =\hfill & _๐ฏ\left[(๐๐\lambda _Q(x))(\xi ^{})\right],\widehat{๐}_0^1_๐ฏ\left[(๐๐(\lambda _Q^a(x)+a))(\xi ^{})\right]\hfill \\ & =\hfill & _๐ฏ\left[๐(x)(\mathrm{ad}_\lambda \xi ^{})\right]+\mathrm{ad}_\lambda ^{}\mu ,\widehat{๐}_0^1_๐ฏ\left[(๐๐(\lambda _Q^a(x)+a))(\xi ^{})\right]\hfill \\ & =\hfill & ๐(x)(\mathrm{ad}_\lambda \xi ^{})+\mathrm{ad}_\lambda ^{}\mu ,\widehat{๐}_0^1(๐๐(\lambda _Q^a(x)+a)(\xi ^{}),\hfill \end{array}$$
since $`\mathrm{ker}๐(x)=๐ฅ`$ and by Lemma 6.1 $`_๐ฅ[(๐๐(\lambda _Q^a(x)+a)(\xi ^{})]=0.`$ For the first term we have
$$\frac{1}{2}๐_x^2\left(๐(x)(\xi ^{},\xi ^{})\right)(\lambda ,\lambda _Q^a(x)+a)=(๐๐(\lambda _Q^a(x)+a))(\mathrm{ad}_\lambda \xi ^{},\xi ^{}),$$
and so we finally obtain
$$(๐_x^2V_\xi ^{}+\mathrm{corr}_\xi ^{}(x))(\lambda _Q(x),\lambda _Q^a(x)+a)=\mathrm{ad}_\lambda ^{}\mu ,\widehat{๐}_0^1(๐๐(\lambda _Q^a(x)+a)(\xi ^{}).$$
This expression is zero since by construction $`(๐๐(\lambda _Q^a(x)+a))(\xi ^{})`$ annihilates $`๐ฑ`$ if $`(\lambda ^a,a)w_{\mathrm{int}}`$ and $`\widehat{๐}_0^1(\mathrm{ad}_\lambda ^{}\mu )๐ฑ`$ for every $`\lambda ๐ฎ^\mu `$. To see this, note that the image of $`\widehat{๐}_0^1`$ is in $`๐ฏ`$ and hence $`๐(x)(\widehat{๐}_0^1(\mathrm{ad}_\lambda ^{}\mu ),\zeta )=\mathrm{ad}_\lambda ^{}\mu ,\zeta =\mathrm{ad}_\zeta ^{}\mu ,\lambda =0,`$ for every $`\zeta ๐ค_\mu `$, in particular if $`\zeta ๐ญ`$. โ
An inspection of the form of $`๐_{p_x}^2h_\xi ^{}\text{ }V_s`$ in (6.7) together with Theorem 4.1 leads to the following sharper result concerning the $`G_\mu `$-stability of $`p_x`$.
###### Corollary 6.2 (Block-diagonalization and stability).
In the hypothesis of Theorem 4.1, and assuming that the Arnold form in non-degenerate, if:
* $`dimQdimG+dimG_x>0`$
* $`\left(๐_x^2V_\xi ^{}+\mathrm{corr}_\xi ^{}(x)\right)\text{ }\mathrm{\Sigma }_{\mathrm{int}}`$ is positive definite
* the Arnold form is positive definite,
or
* $`dimQdimG+dimG_x=0`$
* the Arnold form is definite (positive or negative),
then the relative equilibrium is $`G_\mu `$-stable.
### Remark.
In , Theorem 2.7, equivalent results to our Proposition 6.2 and Corollary 6.2 are obtained as a consequence of their reduced Energy-Momentum Method, for the particular case $`G_x=\{e\}`$ (regular relative equilibria).
## 7. Some remarks on the stability results
### The residual symmetry sub-blocking.
In this subsection we will use the fact that the symplectic normal space $`V_s`$ supports a linear representation of $`G_{p_x}`$ in order to improve our block-diagonalization results. We start with the following lemma (see for a proof):
###### Lemma 7.1.
Let $`hH=G_x,\xi ๐ฏ`$ and $`v,wT_xQ`$, then
* $`\mathrm{Ad}_{h^1}^{}\left[(๐๐v)(\xi )\right]=(๐๐(hv))(\mathrm{Ad}_h\xi )`$
* $`C(hv)(\mathrm{Ad}_h\xi ),hw_๐=C(v)(\xi ),w_๐`$.
As an immediate consequence of this and just by regarding their definitions (4.3), (3.15), (6.4), the spaces $`\mathrm{\Sigma },๐ฎ^\mu `$ and $`\mathrm{\Sigma }_{\text{int}}`$ are $`G_{p_x}`$-invariant. We will use a tool from representation theory known as the *isotypic decomposition* of a linear space acted linearly upon a compact Lie group to take advantage of the residual symmetry group $`G_{p_x}`$ in order to further block-diagonalize $`๐_{p_x}^2h_\xi ^{}\text{ }V_s`$. For this, we need the following definition, which is taken from .
###### Definition 7.1.
Let $`K`$ be a compact Lie group acting linearly on a (real and finite dimensional) linear space $`N`$. The isotypic decomposition of $`N`$ is the unique decomposition
$$N=N_1\mathrm{}N_r,$$
where each $`N_i`$ is the direct sum of all $`K`$-isomorphic irreducible subspaces of $`N`$, and it is called an *isotypic component* of $`N`$.
The isotypic decomposition of a linear space satisfies the following remarkable property (see ): If $`B`$ is a $`K`$-invariant bilinear form on $`N`$ (that is, $`B(gv_1,gv_2)=B(v_1,v_2)`$ for every $`gK,v_1,v_2N`$) then $`B\text{ }N_i\times N_j=0`$ for every pair of isotypic components of $`N`$ with $`N_iN_j`$. Therefore the expression of $`B`$ block-diagonalizes with respect to the isotypic decomposition of $`N`$. We will apply this property to the bilinear form given by $`๐_x^2V_\xi ^{}+\mathrm{corr}_\xi ^{}(x)`$.
###### Lemma 7.2.
Let $`p_x`$ be a relative equilibrium of the simple mechanical system (2.3) with velocity $`\xi `$. Then the bilinear form $`๐_x^2V_\xi ^{}+\mathrm{corr}_\xi ^{}(x)`$ is $`G_{p_x}`$-invariant.
###### Proof.
We have to prove separately the invariance of each term. For $`๐_x^2V_\xi ^{}`$ the result follows if we prove that $`V_\xi ^{}`$ is $`G_{p_x}`$-invariant. Since $`V`$ is $`G`$-invariant, and so $`G_{p_x}`$-invariant, we only need to prove that the function $`\frac{1}{2}๐()(\xi ^{},\xi ^{})`$ is $`G_{p_x}`$-invariant. Recall that at a relative equilibrium, besides the characterization $`G_{p_x}=G_\mu G_x`$, one also has $`G_{p_x}=\{hG_x:\mathrm{Ad}_h\xi ^{}=\xi ^{}\},`$ which follows trivially from the property
$$gG_{p_x}g๐ฝ\mathrm{L}(\xi _Q^{}(x))=๐ฝ\mathrm{L}(\xi _Q^{}(x)),$$
since $`\xi ^{}๐ญ๐ฏ`$ and by definition our relative equilibrium can be written as $`p_x=๐ฝ\mathrm{L}(\xi _Q(x))=๐ฝ\mathrm{L}(\xi _Q^{}(x))`$. Recall also the invariance property of the locked inertia tensor (3.2). Then, for every $`x^{}Q`$ and $`hG_{p_x}`$, one has
$$\frac{1}{2}๐(hx^{})(\xi ^{},\xi ^{})=\frac{1}{2}๐(x^{})(\mathrm{Ad}_{h^1}\xi ^{},\mathrm{Ad}_{h^1}\xi ^{})=\frac{1}{2}๐(x^{})(\xi ^{},\xi ^{}).$$
For the correction term, recall from Lemma 7.1, that if $`hG_{p_x}`$ and $`\delta qT_xQ`$, then
$$(๐๐(h\delta q))(\xi ^{})=\mathrm{Ad}_{h^1}^{}\left((๐๐\delta q)(\xi ^{})\right),$$
since $`\mathrm{Ad}_h\xi ^{}=\xi ^{}`$. Note also that $`\widehat{๐}_0:๐ฏ๐ฏ^{}`$ is a $`G_x`$-equivariant isomorphism, that is $`\widehat{๐}_0\mathrm{Ad}_h=\mathrm{Ad}_{h^1}^{}\widehat{๐}_0.`$ Then, given $`\delta q_1,\delta q_2T_xQ`$ and $`hG_{p_x}`$, we have
$$\begin{array}{cc}\mathrm{corr}_\xi ^{}(x)(h\delta q_1,h\delta q_2)\hfill & =(๐๐(h\delta q_1))(\xi ^{}),\widehat{๐}_0^1\left((๐๐(h\delta q_2))(\xi ^{})\right)\hfill \\ & =\mathrm{Ad}_{h^1}^{}\left((๐๐\delta q_1)(\xi ^{})\right),\widehat{๐}_0^1\left(\mathrm{Ad}_{h^1}^{}\left((๐๐\delta q_2)(\xi ^{})\right)\right)\hfill \\ & =\mathrm{Ad}_{h^1}^{}\left((๐๐\delta q_1)(\xi ^{})\right),\mathrm{Ad}_h\left(\widehat{๐}_0^1\left((๐๐\delta q_2)(\xi ^{})\right)\right)\hfill \\ & =(๐๐\delta q_1)(\xi ^{}),\widehat{๐}_0^1\left((๐๐\delta q_2)(\xi ^{})\right)\hfill \\ & =\mathrm{corr}_\xi ^{}(x)(\delta q_1,\delta q_2)\hfill \end{array}$$
Thus, in the hypothesis of Theorem 4.1 we can use the fact that $`\mathrm{\Sigma }`$ is a $`G_{p_x}`$-invariant subspace of $`T_xQ`$, and then testing the definiteness of $`(๐_x^2V_\xi ^{}+\mathrm{corr}_\xi ^{}(x))\text{ }\mathrm{\Sigma }`$ is equivalent to testing the definiteness of every restriction $`(๐_x^2V_\xi ^{}+\mathrm{corr}_\xi ^{}(x))\text{ }\mathrm{\Sigma }_i`$ where
$$\mathrm{\Sigma }=\mathrm{\Sigma }_1\mathrm{}\mathrm{\Sigma }_r$$
is the isotypic decomposition of $`\mathrm{\Sigma }`$. Analogously, if the Arnold form is not degenerate, to test the conditions for stability of Corollary 6.2 is equivalent to test definiteness of $`\mathrm{Ar}_{๐ฎ_i^\mu }`$ and $`(๐_x^2V_\xi ^{}+\mathrm{corr}_\xi ^{}(x))\text{ }\mathrm{\Sigma }_{\mathrm{int}}^{}{}_{i}{}^{}`$ for each of the isotypic components of $`๐ฎ^\mu `$ and $`\mathrm{\Sigma }_{\mathrm{int}}`$, respectively.
### Nature of the stability results.
In Theorems 2.1 and 4.1 we have imposed the compactness of the momentum isotropy subgroup $`G_\mu `$. However, this compactness condition can be weakened by assuming that $`\mu `$ is *split*, and this is how the original theorems are stated in and . A momentum value $`\mu `$ is called split if there exists a $`G_\mu `$-invariant complement to $`๐ค_\mu `$ in $`๐ค`$. Obviously this is the case if $`G_\mu `$ is compact, since in this case one can define this complement to be the orthogonal complement to $`๐ค_\mu `$ with respect to some $`G_\mu `$-invariant inner product on $`๐ค`$. Likewise, if $`G`$ itself is compact or Abelian, then every momentum value is automatically split.
In the most general situation, if the relative equilibrium under study has not a split momentum value, Theorem 4.1 and Corollary 6.2 are still applicable, but in that case one does not obtain conditions for $`G_\mu `$-stability, only for the weaker notion of *leafwise stability*. A relative equilibrium is called leafwise stable if it is $`G_\mu `$-stable for the restriction of the Hamiltonian flow to $`๐^1(\mu )`$. The reason for this nomenclature is that in the free case, a relative equilibrium $`z`$ with momentum $`\mu `$ for the symmetric Hamiltonian system $`(๐ซ,\omega ,G,๐,h)`$ is leafwise stable if the point $`[z]๐ซ/G`$ in the orbit space is a Lyapunov stable equilibrium for the reduced Hamiltonian system on the symplectic leave $`๐^1(\mu )/G_\mu ๐ซ/G`$, rather than being stable in the full Poisson quotient $`๐ซ/G`$. The results of guarantee that if $`\mu `$ is split, then leafwise stability of $`z`$ implies Lyapunov stability of $`[z]`$ in $`๐ซ/G`$, and, hence, so do Theorem 4.1 and Corollary 6.2. See for a more detailed explanation of these concepts.
### The reduced Energy-Momentum Method and Lagrangian Block- <br>Diagonalization.
In a method was constructed for testing the stability of relative equilibria of symmetric Lagrangian systems. In that work, the techniques of the reduced Energy-Momentum Method of are translated to systems defined on the tangent bundle $`TQ`$ of the configuration space and developed for general Lagrangian systems invariant under a possibly non free, tangent lifted action. We briefly explain the relationship of the results of applied to simple mechanical systems with our work. See for more details on the Lagrangian Block-Diagonalization method.
Let $`LC^G(TQ)`$ be a function on the tangent bundle of $`Q`$ invariant under the tangent lift of a proper action of the Lie group $`G`$ on $`Q`$. This function is called a Lagrangian. There is a well-known procedure to obtain a bundle map $`๐ฝ\mathrm{L}:TQT^{}Q`$ constructed from $`L`$ (no Riemannian structure is in principle available on $`Q`$). In the case $`๐ฝ\mathrm{L}`$ is a diffeomorphism, the Lagrangian is called hyper-regular, and one can pull-back the canonical symplectic form from $`T^{}Q`$ to $`TQ`$ and define a Hamiltonian system on $`TQ`$ (see for details). Given an element $`\xi ๐ค`$, the locked Lagrangian $`L_\xi C^{\mathrm{}}(Q)`$ is defined as
$$L_\xi (x)=L(\xi _Q(x)).$$
Also, the locked momentum map is defined as the map $`I_\xi :Q๐ค^{}`$ that satisfies
$$I_\xi (x),\eta =\frac{d}{dt}\text{ }t=0L_{\xi +t\eta }(x)$$
for every $`\eta ๐ค`$. We will also need the definition of the space of admissible configuration variations at a point $`xQ`$, which is
(7.1)
$$๐=\{\delta qT_xQ:๐I_\xi \delta q๐ค_x^{}\},$$
for a fixed element $`\xi ๐ค`$. Finally, given $`xQ`$ and $`\xi ๐ค`$ the linearized momentum map $`I_x:๐ค๐ค^{}`$ is defined as
$$I_x(\eta )=\frac{d}{dt}\text{ }t=0I_{\xi +t\eta }(x)$$
for every $`\xi ,\eta ๐ค`$. We will denote its generalized inverse by $`\stackrel{~}{I}_x^1:\mathrm{range}I_x๐ค/\mathrm{ker}I_x`$.
The Lagrangian Block-Diagonalization method gives sufficient conditions for formal stability of relative equilibria in Lagrangian systems. Under some assumptions, formal stability implies $`G_\mu `$-stability. Here we shall not be concerned with those differences, in order to provide a comparison between the Lagrangian Block-Diagonalization and our singular reduced Energy-Momentum Method. Furthermore, one needs a technical condition relating $`I_\xi `$ and $`I_x`$ at a relative equilibrium in order to be able to apply the method (see (3.15) in ). However, for the particular case of Lagrangian systems defining simple mechanical systems this is automatically satisfied and thus it is not necessary for our comparison objective. So we assume that the Lagrangian Block Diagonalization method is applicable in order to simplify the exposition.
The next proposition collects the results of concerning relative equilibria and their $`G_\mu `$-stability. Let $`๐ `$ be the symmetric bilinear form on $`T_xQ`$ induced by the hyper-regular Lagrangian $`L`$ and defined by
$$๐ (v_x,w_x)=\frac{d}{dt}\text{ }t=0\frac{d}{ds}\text{ }s=0L(tv_x+sw_x).$$
By the hyper-regularity hypothesis of $`L`$, $`๐ (,)`$ is non-degenerate.
###### Proposition 7.1 (Lagrangian Block-Diagonalization, ).
Let $`LC^G(TQ)`$ be a hyper-regular Lagrangian invariant under the tangent lifted action of the Lie group $`G`$ on $`Q`$. A point $`v_xTQ`$ is a relative equilibrium for the Lagrangian system defined on $`TQ`$ by $`L`$ if there is an element $`\xi ๐ค`$ such that $`v_x=\xi _Q(x)`$ and $`x`$ is a critical point of $`L_\xi `$. If the bilinear form
$$=\left(๐_x^2L_\xi +๐I_\xi (),\stackrel{~}{I_x}^1\left(๐I_\xi ()\right)\right)\text{ }๐$$
is positive (negative) semi-definite with kernel $`๐ค_\mu x`$ and $`๐ _{|(๐คx)^{}}`$ is positive (negative) definite then the relative equilibrium is $`G_\mu `$-stable.
We now study Lagrangians defining simple mechanical systems. It is a standard fact that if we are given a $`G`$-invariant Riemannian metric $`,`$ on $`Q`$ and a $`G`$-invariant function $`VC^G(Q)`$, then the Lagrangian formulation of the associated simple mechanical system (2.3) is
(7.2)
$$L(v_x)=\frac{1}{2}v_x^2V(x).$$
For $`L`$ of the form (7.2) it is straightforward to compute
$$\begin{array}{cc}L_\xi \hfill & =V_\xi \hfill \\ I_\xi (x)\hfill & =๐(x)(\xi )\hfill \\ I_x\hfill & =๐(x)\hfill \\ ๐ \hfill & =,.\hfill \end{array}$$
Therefore, relative equilibria of the simple mechanical system (7.2) are defined by a velocity $`\xi ๐ค`$ and a critical point $`xQ`$ of $`V_\xi `$, recovering the well-know result for critical points of the augmented Hamiltonian in simple mechanical systems.
To see that the stability conditions of Proposition 7.1 are then equivalent to those given in Theorem 4.1, one only needs to prove that at a relative equilibrium, the space $`\mathrm{\Sigma }T_xQ`$ of Definition 4.1 is indeed a complement to $`๐ค_\mu x`$ in $`๐`$, but this follows from Lemma 6.1 and it has been already used in the proof of Proposition 6.1. This shows that our stability result, Theorem 4.1, is a Hamiltonian version of the Lagrangian Block-Diagonalization method applied to simple mechanical systems. In the same way, one can see that the extra block-diagonalization construction carried out in subsection 3.3 of is a consequence of the splitting of $`V_s`$ in rigid and internal subspaces of Section 6 in this paper.
### The normal form for the symplectic matrix.
Besides the convenient form for $`๐_{p_x}^2h_\xi ^{}\text{ }V_s`$ given in Proposition 6.2 which allowed us to refine Theorem 4.1 and re-express it as Corollary 6.2, it is important to note the particular expression for the symplectic matrix $`\mathrm{\Omega }`$ of the symplectic normal space $`V_s`$ identified with $`๐ฎ^\mu \mathrm{\Sigma }_{\text{int}}๐^{}`$. In the free case (see and ), the explicit form for the symplectic matrix, together with the one for $`๐_{p_x}^2h_\xi ^{}\text{ }V_s`$, allows the authors to obtain the linearization of the Hamiltonian vector field at a relative equilibrium. This is an important observation, since the study of this linearized vector field has applications in the spectral and linear stability (and instability) of the relative equilibrium under study, as well as for the identification of possible bifurcations from it.
It seems that despite the generalization to the Lagrangian side and non-free actions in of the stability results provided by the reduced Energy-Momentum Method of , this feature has not been studied in detail for group actions with singularities and the expression for the symplectic matrix obtained in Proposition 6.2 for relative equilibria with non-discrete configuration isotropies cannot be found in the literature. Here we prove that our expression (6.6) coincides in the regular (free) case with the one obtained in , equations (2.83) and (2.85).
Indeed (and if we assume that the Arnold form is non-degenerate) given two elements $`(\eta _i,(\lambda _Q^{a_i}(x)+a_i),\beta _i)๐ฎ^\mu w_{\mathrm{int}}๐^{}V_s`$, $`i=1,2`$, we will write $`\delta q_i=(\lambda ^{a_i})_Q(x)+a_i`$ and $`\alpha (\delta q_i)=\widehat{๐}_0^1(๐(๐ฝ\mathrm{L}(\delta q_i)))=\lambda ^{a_i}`$. Then we have the following:
###### Proposition 7.2.
The expression for the symplectic matrix $`\mathrm{\Omega }`$ of Proposition 6.2 is equivalent to
$$\begin{array}{ccc}\mathrm{\Omega }((\eta _1,\delta q_1,\beta _1),(\eta _2,\delta q_2,\beta _2))\hfill & =\hfill & \mu ,[\eta _1,\eta _2][\eta _1,\alpha (\delta q_2)]+[\eta _2,\alpha (\delta q_1)]\hfill \\ & & +\beta _2,\delta q_1\beta _1,\delta q_2๐\chi ^\xi ^{}(x)(\delta q_1,\delta q_2),\hfill \end{array}$$
with $`\chi ^\xi ^{}`$ defined in (4.1). This coincides in the regular case with equation (2.83) in .
###### Proof.
Recall from Proposition 6.2 that we have
$$\begin{array}{ccc}\mathrm{\Omega }((\eta _1,\delta q_1,\beta _1),(\eta _2,\delta q_2,\beta _2))\hfill & =\hfill & \mu ,[\eta _1,\eta _2][\eta _1,\lambda ^{a_2}]+[\eta _2,\lambda ^{a_1}]\hfill \\ & & +\beta _2,a_1\beta _1,a_2\mu ,[\lambda ^{a_1},\lambda ^{a_2}]\hfill \\ & & +C(a_2)(\xi ^{}),a_1_๐C(a_1)(\xi ^{}),a_2_๐.\hfill \end{array}$$
Now note from the definition of $`\alpha `$, that $`[\eta _i,\alpha (\delta q_j)]=[\eta _i,\lambda ^{a_j}]`$. Also, since $`\beta _i๐^{}=(๐คx)^{},i=1,2`$, one has that $`\beta _i,\delta q_j=\beta _i,a_j`$. Therefore, the proposition will be proved if we show that
$$๐\chi ^\xi ^{}(x)(\delta q_1,\delta q_2)=\mu ,[\lambda ^{a_1},\lambda ^{a_2}]C(a_2)(\xi ^{}),a_1_๐+C(a_1)(\xi ^{}),a_2_๐.$$
To see this, we choose local extensions $`X=(\lambda ^{a_1})_Q+\overline{a}_1`$ and $`Y=(\lambda ^{a_2})_Q+\overline{a}_2`$ of $`\delta q_1`$ and $`\delta q_2`$ near $`x`$ and then we use the formula for the exterior derivative
$$๐\chi ^\xi ^{}(X,Y)=X\left(\chi ^\xi ^{}(Y)\right)Y\left(\chi ^\xi ^{}(X)\right)\chi ^\xi ^{}([X,Y]).$$
We will need the following lemma, proved in , which shows additional properties of the family of local vector fields introduced in Section 3.
###### Lemma 7.3.
The local vector fields defined by (3.11) satisfy
* $`[\overline{v}_a,\overline{v}_b](x)=0`$ for every $`v_a,v_b๐`$
* $`[\lambda _Q,\overline{v}](x)=0`$ for every $`\lambda ๐ฏ,v๐`$
* There is a small enough open neighbourhood $`Ox`$ such that for every $`x^{}GxO,v๐`$ and $`\xi ๐ค`$, $`\overline{v}(x^{}),\xi _Q(x^{})=0.`$
* Let $`x^{}=\sigma ([g,s])`$ with $`\sigma `$ defined in (3.10) and $`\lambda ๐ฏ`$, then
$$\lambda _Q(x^{})=\frac{d}{dt}\text{ }\mathrm{t}=0\sigma ([(\mathrm{exp}_et\lambda )g,s]).$$
Recall the definition $`\chi ^\xi ^{}=๐ฝ\mathrm{L}(\xi _Q^{})`$ and let us compute separately the group orbit and slice contributions for $`๐\chi ^\xi ^{}(x)(X,Y)`$ using the usual properties of the Levi-Civita connection.
$$\begin{array}{cc}๐\chi ^\xi ^{}(x)(a_1,a_2)\hfill & =\overline{a}_1(\xi _Q^{},\overline{a}_2)(x)\overline{a}_2(\xi _Q^{},\overline{a}_1)(x)\hfill \\ & =_{\overline{a}_1}\xi _Q^{}(x),a_2+\xi _Q^{}(x),_{\overline{a}_1}\overline{a}_2(x)\hfill \\ & _{\overline{a}_2}\xi _Q^{}(x),a_1\xi _Q^{}(x),_{\overline{a}_2}\overline{a}_1(x)\hfill \\ & =C(a_1)(\xi ^{}),a_2_๐C(a_2)(\xi ^{}),a_1_๐,\hfill \end{array}$$
where we have used the definition of $`C`$ (3.12), together with item (v) in Theorem 3.1 and the fact that
$$\begin{array}{c}\xi _Q^{}(x),_{\overline{a}_1}\overline{a}_2(x)\xi _Q^{}(x),_{\overline{a}_2}\overline{a}_1(x)\hfill \\ =\xi _Q^{}(x),T(\overline{a}_1,\overline{a}_2)(x)+\xi _Q^{}(x),[\overline{a}_1,\overline{a}_2](x)]=0\hfill \end{array}$$
by item (i) of Lemma 7.3 as well as noting that the Levi-Civita connection has zero torsion ($`T(X,Y)=_XY_YX[X,Y]=0`$, for every pair of vector fields $`X,Y`$). In the same way we can compute
$$\begin{array}{cc}๐\chi ^\xi ^{}(x)((\lambda ^{a_i})_Q(x),a_j)\hfill & =(\lambda ^{a_i})_Q(\xi _Q^{},\overline{a}_j)(x)\overline{a}_j(\xi _Q^{},(\lambda ^{a_i})_Q)(x)\hfill \\ & \chi ^\xi ^{}\left([(\lambda ^{a_i})_Q,\overline{a}_j]\right)(x)=(๐๐a_j)(\xi ^{},\lambda ^{a_i}),\hfill \end{array}$$
since by (iii) and (ii) in Lemma 7.3 the first and last contributions vanish. Finally
$$\begin{array}{c}๐\chi ^\xi ^{}(x)((\lambda ^{a_1})_Q(x),(\lambda ^{a_2})_Q(x))=\hfill \\ (\lambda ^{a_1})_Q(\xi _Q^{},(\lambda ^{a_2})_Q)(x)(\lambda ^{a_2})_Q(\xi _Q^{},(\lambda ^{a_1})_Q)(x)\hfill \\ +\xi _Q^{}(x),([\lambda ^{a_1},\lambda ^{a_2}])_Q(x)=(๐๐(\lambda ^{a_1})_Q(x))(\xi ^{},\lambda ^{a_2})\hfill \\ (๐๐(\lambda ^{a_2})_Q(x))(\xi ^{},\lambda ^{a_1})+\mu ,[\lambda ^{a_1},\lambda ^{a_2}].\hfill \end{array}$$
Putting together all the contributions we obtain
$$\begin{array}{cc}๐\chi ^\xi ^{}(x)(\delta q_1,\delta q_2)\hfill & =\mu ,[\lambda ^{a_1},\lambda ^{a_2}]+C(a_1)(\xi ^{}),a_2_๐C(a_2)(\xi ^{}),a_1_๐\hfill \\ & +(๐๐\delta q_1)(\xi ^{},\lambda ^{a_2})(๐๐\delta q_2)(\xi ^{},\lambda ^{a_1}).\hfill \end{array}$$
The last two terms of the above expression vanish since for $`i=1,2`$, $`\lambda ^{a_i}๐ฎ^\mu ๐ฑ`$ and $`(\lambda ^{a_i},a_i)w_{\mathrm{int}}`$, and by its definition (6.2) $`_๐ฑ\left[(๐๐\delta q_1)(\xi ^{})\right]=0`$. This finishes the proof. โ
### Acknowledgements.
I would like to thank Esmeralda Sousa-Dias, Matt Perlmutter and Tanya Schmah for many illuminating discussions over the past years which have helped me during the realization of this project. I would also like to thank CDS of the California Institute of Technology for their hospitality and facilities during a research visit in November 2004 when part of this project was carried out and in particular Jerrold Marsden for several useful suggestions and comments on this research during that period. This research was partially supported by the EU through funds for the European Research Training Network โMASIEโ, vide contract HPRN-CT-2000-00113. Finally, I would like to thank the referees for their comments on the manuscript.
|
warning/0506/cond-mat0506101.html
|
ar5iv
|
text
|
# Magnetic Soliton and Soliton Collisions of Spinor Bose-Einstein Condensates in an Optical Lattice
## I Introduction
Recently, spinor Bose-Einstein condensates (BECs) trapped in optical potentials have received much attention in both experimental Stenger ; Anderson ; Liu and theoretical studies Ho . Spinor BECs have internal degrees of freedom due to the hyperfine spin of the atoms which liberate a rich variety of phenomena such as spin domains Miesner and textures Ohmi . When the potential valley is so deep that the individual sites are mutually independent, spinor BECs at each lattice site behave like spin magnets and can interact with each other through both the light-induced and the static, magnetic dipole-dipole interactions. These site-to-site dipolar interactions can cause the ferromagnetic phase transition Pu ; Kevin leading to a โmacroscopicโ magnetization of the condensate array and the spin-wave like excitation Pu ; Zhang analogous to the spin-wave in a ferromagnetic spin chain. For the real spin chain, the site-to-site interaction is caused mainly by the exchange interaction, while the dipole-dipole interaction is negligibly small. For the spinor BECs in the optical lattice, the exchange interaction is absent. The individual spin magnets are coupled by the magnetic and the light-induced dipole-dipole interactions Zhang which are no longer negligible due to the large number of atoms $`N`$ at each lattice site, typically of the order of 1000 or more. Therefore, the spinor BECs in an optical lattice offer a totally new environment to study spin dynamics in periodic structures. The magnetic soliton excited by the interaction between the spin waves Zhang is an important and interesting phenomenon in spinor BECs. In this paper, we demonstrate that the magnetic soliton and elastic soliton collision are admitted for spinor BECs in a one-dimensional optical lattice and are controllable by adjusting of the light-induced and the magnetic dipole-dipole interactions.
The Heisenberg model of spin-spin interactions is considered as the starting point for understanding many complex magnetic structures in solids. In particular, it explains the existence of ferromagnetism and antiferromagnetism at temperatures below the Curie temperature. The magnetic soliton Kosevich , which describes localized magnetization, is an important excitation in the Heisenberg spin chain Tjon ; Li ; Ablowitz ; Huang . The Haldane gap Haldane of antiferromagnets has been reported in integer Heisenberg spin chain. By means of the neutron inelastic scattering Kjems78 and electron spin resonance Asano00 , the magnetic soliton has already been probed experimentally in quasi-one dimensional magnetic systems. Solitons can travel over long distances with neither attenuation nor change of shape, since the dispersion is compensated by nonlinear effects. The study of soliton has been conducted in as diverse fields as particle physics, molecular biology, geology, oceanography, astrophysics, and nonlinear optics. Perhaps the most prominent application of solitons is in high-rate telecommunications with optical fibers. However, the generation of controllable solitons is an extremely difficult task due to the complexity of the conventional magnetic materials. The spinor BECs seems an ideal system to serve as a new test ground for studying the nonlinear excitations of spin waves both theoretically and experimentally.
The outline of this paper is organized as follows: In Sec. II the Landau-Lifshitz equation of spinor BEC in an optical lattice is derived in detail. Next, we obtain the one-soliton solution of spinor BEC in an optical lattice. The result shows that the time-oscillation of the amplitude and the size of soliton can be controlled by adjusting of the light-induced dipole-dipole interaction. We also present that the magnetization varies with time periodically. In Sec. VI, the general two-soliton solution for spinor BEC in an optical lattice is investigated. Analysis reveals that elastic soliton collision occurs and there is a phase exchange during collision. Finally, our concluding remarks are given in Sec. V.
## II Landau-Lifshitz equation of spinor BEC in an optical lattice
The dynamics of spinor BECs trapped in an optical lattice is primarily governed by three types of two-body interactions: spin-dependent collision characterized by the $`s`$-wave scattering length, magnetic dipole-dipole interaction (of the order of Bohr magneton $`\mu _B`$), and light-induced dipole-dipole interaction adjusted by the laser frequency in experiment. Our starting point is the Hamiltonian describing an $`F=1`$ spinor condensate at zero temperature trapped in an optical lattice, which is subject to the magnetic and the light-induced dipole-dipole interactions and is coupled to an external magnetic field via the magnetic dipole Hamiltonian $`H_B`$ Ho ; Miesner ; Ohmi ; Pu ,
$`H`$ $`=`$ $`{\displaystyle \underset{\alpha }{}}{\displaystyle ๐๐ซ\widehat{\psi }_\alpha ^{}(๐ซ)[\frac{\mathrm{}^2^2}{2m}+U_L(๐ซ)]\widehat{\psi }_\alpha (๐ซ)}`$ (1)
$`+{\displaystyle \underset{\alpha ,\beta ,\upsilon ,\tau }{}}{\displaystyle ๐๐ซ๐๐ซ^{}\widehat{\psi }_\alpha ^{}(๐ซ)\widehat{\psi }_\beta ^{}(๐ซ^{})\left[U_{\alpha \upsilon \beta \tau }^{coll}(๐ซ,๐ซ^{})+U_{\alpha \upsilon \beta \tau }^{dd}(๐ซ,๐ซ^{})\right]\widehat{\psi }_\tau (๐ซ^{})\widehat{\psi }_\upsilon (๐ซ)}+H_B,`$
where $`\widehat{\psi }_\alpha \left(r\right)`$ is the field annihilation operator for an atom in the hyperfine state $`|f=1,m_f=\alpha `$, $`U_L(๐ซ)`$ is the lattice potential, the indices $`\alpha ,\beta ,\upsilon ,\tau `$ which run through the values $`1,0,1`$ denote the Zeeman sublevels of the ground state. The parameter $`U_{\alpha \upsilon \beta \tau }^{coll}(๐ซ,๐ซ^{})`$ describes the two-body ground-state collisions and $`U_{\alpha \upsilon \beta \tau }^{dd}(๐ซ,๐ซ^{})`$ includes the magnetic dipole-dipole interaction and the light-induced dipole-dipole interaction.
When the optical lattice potential is deep enough there is no spatial overlap between the condensates at different lattice sites. We can then expand the atomic field operator as $`\widehat{\psi }\left(๐ซ\right)=_n`$ $`_{\alpha =0,\pm 1}\widehat{a}_\alpha \left(n\right)\varphi _n\left(๐ซ\right)`$, where $`n`$ labels the lattice sites, $`\varphi _n(๐ซ)`$ is the condensate wave function for the $`n`$th microtrap and the operators $`\widehat{a}_\alpha (n)`$ satisfy the bosonic commutation relations $`[\widehat{a}_\alpha (n),\widehat{a}_\beta ^{}(l)]=\delta _{\alpha \beta }\delta _{nl}`$. It is assumed that all Zeeman components share the same spatial wave function. If the condensates at each lattice site contain the same number of atoms $`N`$, the ground-state wave functions for different sites have the same form $`\varphi _n\left(๐ซ\right)=\varphi _n\left(๐ซ๐ซ_n\right)`$.
In this paper we consider a one-dimensional optical lattice along the $`z`$-direction, which we also choose as the quantization axis. In the absence of spatial overlap between individual condensates, and neglecting unimportant constants, we can construct the effective spin Hamiltonian Zhang ; Pu as
$$H=\underset{n}{}[\lambda _a^{}\widehat{๐}_n^2\underset{ln}{}\lambda _{nl}๐_n๐_l+2\underset{ln}{}\lambda _{nl}^{ld}\widehat{S}_n^z\widehat{S}_l^z\gamma \widehat{S}_n๐],$$
(2)
where $`\lambda _{nl}=2\lambda _{nl}^{ld}+\lambda _{nl}^{md}`$, $`\lambda _{nl}^{md}`$ and $`\lambda _{nl}^{ld}`$ represent the magnetic and the light-induced dipole-dipole interaction respectively. The direction of the magnetic field $`๐`$ is along the one-dimensional optical lattice and $`\gamma =g_F\mu _B`$ is the gyromagnetic ratio. The spin operators are defined as $`๐_n=\widehat{a}_\alpha ^{}(n)๐
_{\alpha \upsilon }\widehat{a}_\upsilon (n)`$, where $`๐
`$ is the vector operator for the hyperfine spin of an atom, with components represented by $`3\times 3`$ matrices in the $`|f=1,m_f=\alpha `$ subspace. The first term in Eq. (2) is resulted from the spin-dependent interatomic collisions at a given site, with $`\lambda _a^{}=(1/2)\lambda _ad^3r\left|\varphi _n(๐ซ)\right|^4`$, where $`\lambda _a`$ characterizes the spin-dependent $`s`$-wave collisions. The second and the third terms describe the site to site spin coupling induced by the static magnetic field dipolar interaction and the light-induced dipole-dipole interaction. For $`\lambda _{nl}0`$, the transfer of transverse excitation from site to site is allowed, resulting in the distortion of the ground-state spin structure. This distortion can propagate and hence generate spin waves along the atomic spin chain. For an optical lattice created by blue-detuned laser beams, the atoms are trapped in the dark-field nodes of the lattice and the light-induced dipole-dipole interaction is very small Kevin . However, this small light-induced dipole-dipole interaction induces the amplitude and size of the soliton varying with time periodically as we will show in the following section.
From Hamiltonian (2) we can derive the Heisenberg equation of motion at $`k`$th site for the spin excitations. When the optical lattice is infinitely long and the spin excitations are in the long-wavelength limit, i.e., the continuum limit, $`S_kS(z,t)`$, we obtain the Landau-Lifshitz equation of a spinor BECs in an optical lattice as follows:
$`{\displaystyle \frac{S^x}{t}}`$ $`=`$ $`{\displaystyle \frac{2\lambda }{\mathrm{}}}[a^2(S^y{\displaystyle \frac{^2}{z^2}}S^zS^z{\displaystyle \frac{^2}{z^2}}S^y)4{\displaystyle \frac{\lambda ^{ld}}{\lambda }}S^yS^z]+{\displaystyle \frac{\gamma BS^y}{\mathrm{}}},`$
$`{\displaystyle \frac{S^y}{t}}`$ $`=`$ $`{\displaystyle \frac{2\lambda }{\mathrm{}}}[a^2(S^z{\displaystyle \frac{^2}{z^2}}S^xS^x{\displaystyle \frac{^2}{z^2}}S^z)+4{\displaystyle \frac{\lambda ^{ld}}{\lambda }}S^zS^x]{\displaystyle \frac{\gamma BS^x}{\mathrm{}}},`$
$`{\displaystyle \frac{S^z}{t}}`$ $`=`$ $`{\displaystyle \frac{2\lambda }{\mathrm{}}}[a^2(S^x{\displaystyle \frac{^2}{z^2}}S^yS^y{\displaystyle \frac{^2}{z^2}}S^x)],`$ (3)
where we assume that all nearest-neighbor interactions are the same, namely $`\lambda _{<nl>}=\lambda `$, which is a good approximation in the one-dimensional optical lattice for the large lattice constant Konotop . In a rotating frame around $`z`$-axis with angular frequency $`\frac{\gamma B}{\mathrm{}}`$ the spin vector $`S`$ is related with the original one by the transformation
$$S^x=S^x^{}\mathrm{cos}(\frac{\gamma B}{\mathrm{}}t)+S^y^{}\mathrm{sin}(\frac{\gamma B}{\mathrm{}}t),S^y=S^y^{}\mathrm{cos}(\frac{\gamma B}{\mathrm{}}t)S^x^{}\mathrm{sin}(\frac{\gamma B}{\mathrm{}}t).$$
(4)
Thus the Landau-Lifshitz equation (3) in the rotating coordinate system can be written as
$`{\displaystyle \frac{}{t}}S^x`$ $`=`$ $`S^y{\displaystyle \frac{^2}{z^2}}S^zS^z{\displaystyle \frac{^2}{z^2}}S^y16\rho ^2S^yS^z,`$
$`{\displaystyle \frac{}{t}}S^y`$ $`=`$ $`S^z{\displaystyle \frac{^2}{z^2}}S^xS^x{\displaystyle \frac{^2}{z^2}}S^z+16\rho ^2S^zS^x,`$
$`{\displaystyle \frac{}{t}}S^z`$ $`=`$ $`S^x{\displaystyle \frac{^2}{z^2}}S^yS^y{\displaystyle \frac{^2}{z^2}}S^x,`$ (5)
where $`\rho ^2=\lambda ^{ld}/(4\lambda )`$, and the superscript ( ) is omitted for the sake of pithiness. The dimensionless time $`t`$ and coordinate $`z`$ in Eq. (5) are scaled in unit $`2\lambda /\mathrm{}`$ and $`a`$ respectively, where $`a`$ denotes the lattice constant. Also, the terms including the external magnetic field in Eq. (3) have been eliminated with the help of the transformation.
## III One-soliton solution of spinor BEC in an optical lattice
The equation (5) has a form of the Landau-Lifshitz (LL) type which is similar to the LL equation for a spin chain with an easy plane anisotropy Note1 . By introducing a particular parameter Huang Huang showed that the Jost solutions can be generated and the Lax equations are satisfied, and moreover constructed Darboux transformation matrices. An explicit expression of the one-soliton solution in terms of elementary functions of $`z`$ and $`t`$ was reported. Here using the similar method in Ref. Ablowitz ; Huang we obtain both the one-and two-soliton solutions (for detail see the appendix) denoted by $`๐(n)`$ with $`n=1,2`$ of Eq. (5) in the following form:
$`S_n^x`$ $`=`$ $`1{\displaystyle \frac{1}{\mathrm{\Lambda }_n}}\left(\chi _{2,n}+2\chi _{3,n}\mathrm{sin}^2\mathrm{\Phi }_n\right),`$
$`S_n^y`$ $`=`$ $`{\displaystyle \frac{1}{\mathrm{\Lambda }_n}}\left(\chi _{1,n}\eta _n\mathrm{cosh}\mathrm{\Theta }_n\mathrm{sin}\mathrm{\Phi }_n+\chi _{2,n}\mathrm{sinh}\mathrm{\Theta }_n\mathrm{cos}\mathrm{\Phi }_n\right),`$
$`S_n^z`$ $`=`$ $`{\displaystyle \frac{1}{\mathrm{\Lambda }_n}}\left(\chi _{1,n}\mathrm{cosh}\mathrm{\Theta }_n\mathrm{cos}\mathrm{\Phi }_n+\chi _{2,n}\eta _n\mathrm{sinh}\mathrm{\Theta }_n\mathrm{sin}\mathrm{\Phi }_n\right),`$ (6)
where the parameters in the solution are defined by
$`\mathrm{\Lambda }_n`$ $`=`$ $`\mathrm{cosh}^2\mathrm{\Theta }_n+\chi _{3,n}\mathrm{sin}^2\mathrm{\Phi }_n,`$
$`\mathrm{\Theta }_n`$ $`=`$ $`2\kappa _{4,n}\left(zV_ntz_n\right),`$
$`\mathrm{\Phi }_n`$ $`=`$ $`2\kappa _{3,n}z\mathrm{\Omega }_nt+\varphi _n,`$
$`V_n`$ $`=`$ $`2\left(\kappa _{1,n}+{\displaystyle \frac{\kappa _{3,n}}{\kappa _{4,n}}}\kappa _{2,n}\right),`$
$`\mathrm{\Omega }_n`$ $`=`$ $`4\left(\kappa _{1,n}\kappa _{3,n}\kappa _{2,n}\kappa _{4,n}\right),`$ (7)
with $`\kappa _{1,n}=\mu _n(1+\rho ^2/|\zeta _n|^2)`$, $`\kappa _{2,n}=\nu _n(1\rho ^2/|\zeta _n|^2)`$, $`\kappa _{3,n}=\mu _n(1\rho ^2/|\zeta _n|^2)`$, $`\kappa _{4,n}=\nu _n(1+\rho ^2/|\zeta _n|^2)`$, $`\eta _n=(|\zeta _n|^2+\rho ^2)/(|\zeta _n|^2\rho ^2)`$, $`\chi _{1,n}=\left(2\mu _n\nu _n\right)/|\zeta _n|^2`$, $`\chi _{2,n}=\left(2\nu _n^2\right)/|\zeta _n|^2`$, and $`\chi _{3,n}=\left(4\rho ^2\nu _n^2\right)/(|\zeta _n|^2\rho ^2)^2`$. The one-soliton solution, namely $`๐(1)`$ , is simply that
$$S^x(1)=S_1^x;S^y(1)=S_1^y;S^z(1)=S_1^z.$$
(8)
The parameter $`V_1`$ denotes the velocity of envelope motion of the magnetic soliton. The real constants $`z_1`$ and $`\varphi _1`$ represent the center position and the initial phase respectively. The parameter $`\zeta _1=\mu _1+i\nu _1`$ is eigenvalue with $`\mu _1`$, $`\nu _1`$ being the real and imaginary parts. The one-soliton solution (8) describes a spin precession characterized by four real parameters: velocity $`V_1`$, phase $`\mathrm{\Phi }_1`$, the center coordinate of the solitary wave $`z_1`$ and initial phase $`\varphi _1`$. From the one-soliton solution we obtain the properties of the soliton: (i) both the amplitude and the size of the soliton vary with time periodically, as shown in figure 1, in which we demonstrate graphically the dynamics of soliton with the parameters chosen as $`\mu _1=0.45`$ , $`\nu _1=0.7`$, $`\rho ^2=0.375`$, $`z_1=14`$ and $`\varphi _1=0.5`$. This property is resulted from the term $`\rho `$ in Eq. (5) which is determined by the light-induced dipole-dipole interaction. This significant observation from the one-soliton solution shows that the time-oscillation of the amplitude and the size of soliton can be controlled in practical experiment by adjusting of the light-induced dipole-dipole interaction. (ii) the magnetization defined by $`M_3=_{\mathrm{}}^{\mathrm{}}๐z(1S_1^x)`$ varies with time periodically as shown in figure 2. These properties are similar to that of the Heisenberg spin chain with an easy plane anisotropy where the dipolar coupling is typically several orders of magnitude weaker than the exchange coupling and thus would correspond to Curie temperatures much below the observed values. Hence the contribution of the dipolar coupling to the spin wave can be neglected in practice. However, for the spinor BEC in the optical lattice the exchange interaction is absent and the individual spin magnets are coupled by the magnetic and the light-induced dipole-dipole interactions. Due to the large number of atoms $`N`$ at each lattice site, these site to site interactions, despite the large distance between sites, explain the natural existence of magnetic soliton which agrees with the results in Refs. Pu ; Zhang .
To see closely the physical significance of one-soliton solution, it is helpful to show the parameter-dependence of Euler angles of the magnetization vector which in a spherical coordinate is
$$S_1^x(z,t)=\mathrm{cos}\theta ,\text{ }S_1^y+iS_1^z=\mathrm{sin}\theta \mathrm{exp}(i\phi ).$$
(9)
From Eqs. (6) and (8) we find
$`\mathrm{cos}\theta `$ $`=`$ $`1{\displaystyle \frac{\frac{2\nu _1^2}{\left|\zeta _1\right|^2}+2\chi _{3,1}\mathrm{sin}^2\mathrm{\Phi }_1}{\mathrm{cosh}^2\left[ฯ_1^1\left(zV_1tz_1\right)\right]+\chi _{3,1}\mathrm{sin}^2\mathrm{\Phi }_1}},`$
$`\phi `$ $`=`$ $`{\displaystyle \frac{\pi }{2}}+\mathrm{arctan}\left(\eta _1\mathrm{tan}\mathrm{\Phi }_1\right)+\mathrm{arctan}\left(\mathrm{tanh}\mathrm{\Theta }_1\right),`$ (10)
where $`ฯ_1=1/(2\kappa _{4,1})`$ and the maximal amplitude $`A_M=2\left(\nu _1^2/\left|\zeta _1\right|^2+\left|\chi _{3,1}\right|\right)`$. When $`\left|\zeta _1\right|^2>>\rho ^2`$, the phase $`\phi `$ can be rewritten as
$$\phi =\frac{\pi }{2}+\varphi _1+k_1z\mathrm{\Omega }_1t+\mathrm{arctan}\left(\mathrm{tanh}\mathrm{\Theta }_1\right),$$
(11)
where the wave number $`k_1=2\kappa _{3,1}`$ and the frequency of magnetization precession $`\mathrm{\Omega }_1`$ are related by the dispersion law
$$\mathrm{\Omega }_1=k_1\left(k_1+4\rho ^2\mu _1/\left|\zeta _1\right|^2\right)4\kappa _{2,1}\kappa _{4,1}.$$
(12)
We also see that the position of minimum of energy spectrum $`\mathrm{\Omega }_{1,\mathrm{min}}=0`$ is located at $`k_{1\mathrm{min}}=(\sqrt{\left(2\rho ^2\mu _1\right)^2+4\kappa _{2,1}\kappa _{4,1}\left|\zeta _1\right|^4}2\rho ^2\mu _1)/\left|\zeta _1\right|^2`$.
If amplitude $`A_M`$ approaches zero, namely $`\nu _10`$, the parameter $`ฯ_1`$ diverges and Eq. (10) takes the asymptotic form
$$\mathrm{cos}\theta 1,\text{ }\phi \frac{\pi }{2}+\varphi _1+k_1z\mathrm{\Omega }_1t,$$
(13)
indicating a small linear solution of magnon. In this case the dispersion law reduces to $`\mathrm{\Omega }_1=k_1\left(k_1+4\rho ^2\mu _1/\left|\zeta _1\right|^2\right)`$.
## IV Elastic soliton collision for spinor BEC in an optical lattice
The magnetic soliton collision in spinor BECs is an interesting phenomenon in spin dynamics. Here in terms of a Darboux transformation we first of all give the two-soliton solution (for detail see the appendix) of Eq. (5)
$`S^x\left(2\right)`$ $`=`$ $`S_1^xS_2^x+R_3S_1^y+R_5S_1^z,`$
$`S^y\left(2\right)`$ $`=`$ $`S_1^xS_2^y+R_4S_1^y+R_6S_1^z,`$
$`S^z\left(2\right)`$ $`=`$ $`S_1^xS_2^z+R_1S_1^y+R_2S_1^z,`$ (14)
where $`S_n^x`$, $`S_n^y`$ and $`S_n^z`$ $`\left(n=1,2\right)`$ are defined in Eq. (6) and $`R_j`$ $`\left(j=1,2,\mathrm{}6\right)`$ take form as follows:
$`R_1`$ $`=`$ $`{\displaystyle \frac{1}{\mathrm{\Lambda }_2}}\left(\chi _{1,2}\mathrm{cosh}\mathrm{\Theta }_2\mathrm{sinh}\mathrm{\Theta }_2\chi _{2,2}\eta _2\mathrm{cos}\mathrm{\Phi }_2\mathrm{sin}\mathrm{\Phi }_2\right),`$
$`R_2`$ $`=`$ $`1{\displaystyle \frac{\chi _{2,2}}{\mathrm{\Lambda }_2}}\left(\mathrm{cosh}^2\mathrm{\Theta }_2\mathrm{sin}^2\mathrm{\Phi }_2\right),`$
$`R_3`$ $`=`$ $`{\displaystyle \frac{1}{\mathrm{\Lambda }_2}}\left(\chi _{1,2}\eta _2\mathrm{cosh}\mathrm{\Theta }_2\mathrm{sin}\mathrm{\Phi }_2\chi _{2,2}\mathrm{sinh}\mathrm{\Theta }_2\mathrm{cos}\mathrm{\Phi }_2\right),`$
$`R_4`$ $`=`$ $`1{\displaystyle \frac{1}{\mathrm{\Lambda }_2}}\left[\chi _{2,2}\mathrm{sinh}^2\mathrm{\Theta }_2+\left(\chi _{2,2}+2\chi _{3,2}\right)\mathrm{sin}^2\mathrm{\Phi }_2\right],`$
$`R_5`$ $`=`$ $`{\displaystyle \frac{1}{\mathrm{\Lambda }_2}}(\chi _{2,2}\eta _2\mathrm{sin}\mathrm{\Phi }_2\mathrm{sinh}\mathrm{\Theta }_2\chi _{1,2}\mathrm{cosh}\mathrm{\Theta }_2\mathrm{cos}\mathrm{\Phi }_2),`$
$`R_6`$ $`=`$ $`{\displaystyle \frac{1}{\mathrm{\Lambda }_2}}(\chi _{2,2}\eta _2\mathrm{sin}\mathrm{\Phi }_2\mathrm{cos}\mathrm{\Phi }_2+\chi _{1,2}\mathrm{cosh}\mathrm{\Theta }_2\mathrm{sinh}\mathrm{\Theta }_2),`$ (15)
where $`\mathrm{\Theta }_2`$, $`\mathrm{\Phi }_2`$, $`\mathrm{\Omega }_2`$, $`\mathrm{\Lambda }_2`$, $`\eta _2`$ and $`\chi _{m,2}`$ $`\left(m=1,2,3\right)`$ are defined in Eq. (7). The solution (14) describes a general elastic scattering process of two solitary waves with different center velocities $`V_1`$ and $`V_2`$, different phases $`\mathrm{\Phi }_1`$ and $`\mathrm{\Phi }_2`$. Before collision, they move towards each other, one with velocity $`V_1`$ and shape variation frequency $`\mathrm{\Omega }_1`$ and the other with $`V_2`$ and $`\mathrm{\Omega }_2`$. In order to understand the nature of two-soliton interaction, we analyze the asymptotic behavior of two-soliton solution (14). Asymptotically, the two-soliton waves (14) can be written as a combination of two one-soliton waves (8) with different amplitudes and phases. The asymptotic form of two-soliton solution in limits $`t\mathrm{}`$ and $`t\mathrm{}`$ is similar to that of the one-soliton solution (8). In order to analyze the asymptotic behavior of two-soliton solutions (14) we show first of all the asymptotic behavior of $`S_n^x`$, $`S_n^y`$, $`S_n^z\left(n=1,2\right)`$, and $`R_j`$($`j=1,2,\mathrm{}6`$) in the corresponding limits $`t\pm \mathrm{}`$ from Eqs. (6) and (15)
$`R_1`$ $``$ $`\pm \chi _{1,2},\text{ }R_21\chi _{2,2},\text{ }R_30,`$
$`R_4`$ $``$ $`1\chi _{2,2},\text{ }R_50,\text{ }R_6\chi _{1,2},`$
$`S_n^x`$ $``$ $`1,\text{ }S_n^y0,\text{ }S_n^z0,\text{as }t\pm \mathrm{}.`$ (16)
Without loss of generality, we assume that $`\kappa _{4,n}>0`$ $`\left(n=1,2\right)`$ and $`V_1>V_2`$ which corresponds to a head-on collision of the solitons. For the above parametric choice, the variables $`\mathrm{\Theta }_n`$($`n=1,2`$) for the two-soliton behave asymptotically as (i) $`\mathrm{\Theta }_10`$, $`\mathrm{\Theta }_2\pm \mathrm{}`$, as $`t\pm \mathrm{}`$; and (ii) $`\mathrm{\Theta }_20`$, $`\mathrm{\Theta }_1\mathrm{}`$, as $`t\pm \mathrm{}`$. This leads to the following asymptotic forms for the two-soliton solution. (For the other choices of $`\kappa _{4,n}`$ and $`V_n`$, similar analysis can be performed straightforwardly).
(i) Before collision, namely the case of limit $`t\mathrm{}`$.
(a) Soliton 1 ($`\mathrm{\Theta }_10`$, $`\mathrm{\Theta }_2\mathrm{}`$).
$$\left(\begin{array}{c}S^x\left(2\right)\\ S^y\left(2\right)\\ S^z\left(2\right)\end{array}\right)\left(\begin{array}{c}S_1^x\\ \mathrm{sin}\theta \mathrm{cos}\left(\phi \varphi _\mathrm{\Delta }\right)\\ \mathrm{sin}\theta \mathrm{sin}\left(\phi \varphi _\mathrm{\Delta }\right)\end{array}\right),$$
(17)
where $`\varphi _\mathrm{\Delta }=\mathrm{arctan}\left[2\mu _2\nu _2/\left(\mu _2^2\nu _2^2\right)\right]`$ and the parameters $`\theta `$ and $`\phi `$ are defined in Eq. (10).
(b) Soliton 2 ( $`\mathrm{\Theta }_20`$, $`\mathrm{\Theta }_1\mathrm{}`$).
$$\left(\begin{array}{c}S^x\left(2\right)\\ S^y\left(2\right)\\ S^z\left(2\right)\end{array}\right)\left(\begin{array}{c}S_2^x\\ S_2^y\\ S_2^z\end{array}\right),$$
(18)
(ii) After collision, namely the case of limit $`t\mathrm{}`$.
(a) Soliton 1 ($`\mathrm{\Theta }_10`$, $`\mathrm{\Theta }_2\mathrm{}`$).
$$\left(\begin{array}{c}S^x\left(2\right)\\ S^y\left(2\right)\\ S^z\left(2\right)\end{array}\right)\left(\begin{array}{c}S_1^x\\ \mathrm{sin}\theta \mathrm{cos}\left(\phi +\varphi _\mathrm{\Delta }\right)\\ \mathrm{sin}\theta \mathrm{sin}\left(\phi +\varphi _\mathrm{\Delta }\right)\end{array}\right),$$
(19)
(b) Soliton 2 ( $`\mathrm{\Theta }_20`$, $`\mathrm{\Theta }_1\mathrm{}`$).
$$\left(\begin{array}{c}S^x\left(2\right)\\ S^y\left(2\right)\\ S^z\left(2\right)\end{array}\right)\left(\begin{array}{c}S_2^x\\ S_2^y\\ S_2^z\end{array}\right).$$
(20)
here for the expressions of solitons before and after collision, $`S_n^x`$, $`S_n^y`$ and $`S_n^z`$ $`\left(n=1,2\right)`$ are defined in Eq. (6). Analysis reveals that there is no amplitude exchange among three components $`S^x`$, $`S^y`$ and $`S^z`$ for soliton 1 and soliton 2 during collision. However, from Eqs. (17) and (19) one can see that there is a phase exchange $`2\varphi _\mathrm{\Delta }`$ between two components $`S^y`$ and $`S^z`$ for soliton 1 during collision. This elastic collision between two magnetic solitons in the optical lattice is different from that of coupled nonlinear Schrรถdinger equations Kanna . It shows that the information held in each soliton will almost not be disturbed by each other in soliton propagation. These properties may have potential application in future quantum communication. It should be noted that the inelastic collision may appear if the influence of higher-order terms in Eq. (2) is considered.
## V Conclusion
Magnetic soliton dynamics of spinor BECs in an optical lattice is studied in terms of a modified Landau-Lifshitz equation which is derived from the effective Hamiltonian of a pseudospin chain. The soliton solutions are obtained analytically and the elastic collision of two solitons is demonstrated. The significant observation is that time-oscillation of the soliton amplitude and size can be controlled by adjusting of the light-induced dipole-dipole interactions.
It should be interesting to discuss how to create the magnetic soliton and how to detect such magnetic soliton in experiment. In the previous work Miesner using Landau-Zener rf-sweeps at high fields (30 G) a condensate was prepared in the hyperfine state $`|f=1,m_f=0`$, i.e. the the ground state of the spinor BECs. Then the atoms of the ground state can be excited to the hyperfine state $`|f=1,m_f=\pm 1`$ by laser light experimentally. Therefore the excited state of the spinor BECs, i.e. the magnetic soliton can be created. As can be seen from Fig. 1, the spatial-temporal spin variations in the soliton state are significant. This makes it possible to take a direct detection of the magnetic soliton of spinor BECs. By counting the difference numbers of the population between the spin $`+1`$ and $`1`$ Zeeman sublevel, the average of spin component $`<S^z>`$ is measured directly. While transverse components can be measured by use of a short magnetic pulse to rotate the transverse spin component to the longitudinal direction. Any optical or magnetic method which can excite the internal transitions between the atomic Zeeman sublevels can be used for this purpose. In current experiments in optical lattices, the lattice number is in the range of $`10`$-$`100`$, and each lattice site can accommodate a few thousand atoms. This leads to a requirement for the frequency measurement precision of about 10-100 kHz. This is achievable with current techniques. We can also see that the detection of the magnetic soliton of the spinor BECs is different from that of the Heisenberg spin chain.
The magnetic soliton of spinor BECs in an optical lattice is mainly caused by the magnetic and the light-induced dipole-dipole interactions between different lattice sites. Since these long-range interactions are highly controllable the spinor BECs in optical lattice which is an exceedingly clean system can serve as a test ground to study the static and dynamic aspects of soliton excitations.
## VI Acknowledgement
This work was supported by the NSF of China under Grant Nos. 10475053, 60490280, 90406017 and provincial overseas scholar foundation of Shanxi.
## VII Appendix
The corresponding Lax equations for the Eq. (5) are written as
$$_zG(z,t)=LG(z,t),_tG(z,t)=MG(z,t),$$
(21)
where
$`L`$ $`=`$ $`iฯตS^z\sigma _3i\varsigma \left(S^x\sigma _1+S^y\sigma _2\right),`$
$`M`$ $`=`$ $`i2\varsigma ^2S^z\sigma _3+i2\varsigma ฯต\left(S^x\sigma _1+S^y\sigma _2\right)i\varsigma \left(S^y_zS^zS^z_zS^y\right)\sigma _1`$ (22)
$`i\varsigma \left(S^z_zS^xS^x_zS^z\right)\sigma _2iฯต\left(S_1_zS^yS^y_zS^x\right)\sigma _3,`$
here $`\sigma _j`$($`j=1,2,3`$) is Pauli matrix, the parameters $`ฯต`$ and $`\varsigma `$ satisfy the relation $`ฯต^2=\varsigma ^2+4\rho ^2`$. Thus Eq. (5) can be recovered from the compatibility condition $`_tL_xM+[L,M]=0`$. We introduce an auxiliary parameter $`q`$ such that
$$ฯต=2\rho \frac{q+q^1}{qq^1},\varsigma =2\rho \frac{2}{qq^1},$$
(23)
and the complex parameter is defined by $`q=\left(\zeta +\rho \right)/\left(\zeta \rho \right)`$.
It is easily to see that $`S_0=(1,0,0)`$ is a simplest solution of Eq. (5). Under this condition the corresponding Jost solution of Eq. (21) can be written as
$$G_0=U\mathrm{exp}\left\{i\varsigma \left(z2ฯตt\right)\sigma _3\right\},$$
(24)
where $`U=\frac{1}{2}\left\{Ii\left(\sigma _1+\sigma _2+\sigma _3\right)\right\}`$ with $`I`$ denoting unit matrix. In the following we construct the Darboux matrix $`D_n\left(q\right)`$ by using the following recursion relation
$$G_n\left(q\right)=D_n\left(q\right)G_{n1}\left(q\right),\text{ }n=1,2,3,\mathrm{},$$
(25)
where $`D_n\left(q\right)`$ has poles. Since
$$ฯต\left(\overline{q}\right)=\overline{ฯต\left(q\right)},\varsigma \left(\overline{q}\right)=\overline{\varsigma \left(q\right)},L\left(\overline{q}\right)=\sigma _1\overline{L\left(q\right)}\sigma _1,M\left(\overline{q}\right)=\sigma _1\overline{M\left(q\right)}\sigma _1,$$
(26)
we then have
$$G_0\left(\overline{q}\right)=i\sigma _1\overline{G_0\left(q\right)},G_n\left(\overline{q}\right)=i\sigma _1\overline{G_n\left(q\right)},D_n\left(\overline{q}\right)=\sigma _1\overline{D_n\left(q\right)}\sigma _1,$$
(27)
where the overbar denotes complex conjugate. Suppose that $`q_n`$ is a simple pole of $`D_n\left(q\right),`$ then $`\overline{q}_n`$ is also a pole of $`D_n\left(q\right)`$. If $`D_n\left(q\right)`$ has only these two simple poles we have
$`D_n\left(q\right)`$ $`=`$ $`C_nP_n\left(q\right),`$ (28)
$`P_n\left(q\right)`$ $`=`$ $`I+{\displaystyle \frac{q_n\overline{q}_n}{qq_n}}P_n+{\displaystyle \frac{q_n\overline{q}_n}{q+\overline{q}_n}}\stackrel{~}{P}_n,`$ (29)
where $`C_n`$, $`P_n`$, and $`\stackrel{~}{P}_n`$ are $`2\times 2`$ matrix independent of $`q`$, the terms $`\left(q_n\overline{q}_n\right)C_nP_n`$ and $`\left(q_n\overline{q}_n\right)C_n\stackrel{~}{P}_n`$ are residues at $`q_n`$ and $`\overline{q}_n`$, respectively. From Eq (27), we have
$$C_n=\sigma _1\overline{C}_n\sigma _1,\stackrel{~}{P}_n=\sigma _1\overline{P_n}\sigma _1.$$
(30)
From Eqs. (22) and (24) we see that
$$L(q)=L^{}\left(\overline{q}\right),M(q)=M^{}\left(\overline{q}\right),G_0^1(q)=G_0^{}\left(\overline{q}\right),$$
(31)
and hence we have
$$G_n^1(q)=G_n^{}\left(\overline{q}\right),D_n^1(q)=D_n^{}\left(\overline{q}\right)=P_n^{}\left(\overline{q}\right)C_n^{}.$$
(32)
Since $`D_n\left(q\right)D_n^1\left(q\right)=D_n\left(q\right)D_n^1\left(\overline{q}\right)=I`$, it has no poles. Then we obtain
$$P_nP_n^{}\left(\overline{q}_n\right)=0,\text{ }P_n\left(IP_n^{}+\frac{\overline{q}_nq_n}{2q_n}\stackrel{~}{P}_n^{}\right)=0,$$
(33)
which shows the degeneracy of $`P_n`$. One can write $`P_n=(\begin{array}{cc}g_n\hfill & w_n\hfill \end{array})^T(\begin{array}{cc}\mathrm{{\rm Y}}_n\hfill & \xi _n\hfill \end{array})`$ where the superscript $`T`$ means transpose. Substituting this expression into (33) we obtain
$`P_n\left(q\right)`$ $`=`$ $`{\displaystyle \frac{1}{\mathrm{\Delta }_n\left(qq_n\right)\left(q+\overline{q}_n\right)}}\left(\begin{array}{cc}\overline{q}_n\left|\mathrm{{\rm Y}}_n\right|^2+q_n\left|\xi _n\right|^2& 0\\ 0& q_n\left|\mathrm{{\rm Y}}_n\right|^2+\overline{q}_n\left|\xi _n\right|^2\end{array}\right)`$ (44)
$`\times \{q^2\left(\begin{array}{cc}q_n\left|\mathrm{{\rm Y}}_n\right|^2+\overline{q}_n\left|\xi _n\right|^2& 0\\ 0& \overline{q}_n\left|\mathrm{{\rm Y}}_n\right|^2+q_n\left|\xi _n\right|^2\end{array}\right)+q(q_n^2\overline{q}_n^2)\left(\begin{array}{cc}0& \overline{\mathrm{{\rm Y}}}_n\xi _n\\ \overline{\xi }_n\mathrm{{\rm Y}}_n& 0\end{array}\right)`$
$`\left|q_n|^2\left(\begin{array}{cc}q_n\left|\xi _n\right|^2+\overline{q}_n\left|\mathrm{{\rm Y}}_n\right|^2& 0\\ 0& q_n\left|\mathrm{{\rm Y}}_n\right|^2+\overline{q}_n\left|\xi _n\right|^2\end{array}\right)\right\},`$
where
$$\mathrm{\Delta }_n=\left|q_n\right|^2\left(\left|\mathrm{{\rm Y}}_n\right|^2+\left|\xi _n\right|^2\right)^2+\left|\overline{q}_nq_n\right|^2\left|\mathrm{{\rm Y}}_n\right|^2\left|\xi _n\right|^2.$$
(45)
To determine $`\xi _n`$ and $`\mathrm{{\rm Y}}_n,`$ we substitute (25) into (21) and take the limit $`qq_n`$ and then obtain
$`_zD_n\left(q\right)`$ $`=`$ $`L_n\left(q\right)D_n\left(q\right)D_n\left(q\right)L_{n1}\left(q\right),`$
$`_tD_n\left(q\right)`$ $`=`$ $`M_n\left(q\right)D_n\left(q\right)D_n\left(q\right)M_{n1}\left(q\right).`$ (46)
Because of the degeneracy of $`P_n,`$ the second factor of the right-hand sides of Eq. (46), namely, ($`\begin{array}{cc}\mathrm{{\rm Y}}_n\hfill & \xi _n\hfill \end{array})G_{n1}(q_n)`$ must appear in the left-hand side in its original form and, hence, it is independent of $`z`$ and $`t`$. We simply let
$$\left(\begin{array}{cc}\mathrm{{\rm Y}}_n\hfill & \xi _n\hfill \end{array}\right)=\left(\begin{array}{cc}b_n\hfill & 1\hfill \end{array}\right)G_{n1}^1\left(q_n\right),$$
(47)
here $`b_n`$ is a constant. Hence, the Darboux matrices $`D_n\left(q\right)`$ have been determined recursively, except for $`C_n.`$ By a simple algebraic procedure, it is seen that $`\mathrm{\Delta }_n`$ is always non-vanishing regardless of the values $`z`$ and $`t.`$ This shows the regularity of $`P_n`$ and then $`P_n\left(q\right)`$. In the limit as $`q1`$, from Eq. (23) we have
$$ฯต\left(q\right),\varsigma \left(q\right)2\rho \frac{1}{q1}+O\left(1\right),$$
and then from Eq. (46) we obtain
$$S\left(n\right)\sigma =D_n\left(1\right)\left[S\left(n1\right)\sigma \right]D_n^{}\left(1\right),n=1,2,3,\mathrm{}.$$
(48)
Considering the Eq. (31) and Eq. (32) we get
$$C_nC_n^{}=I,$$
(49)
which shows that the matrix $`C_n`$ is diagonal with the help of the Eq. (30) and
$$\left(C_n\right)_{11}=\left(\overline{C_n}\right)_{22},\left|\left(C_n\right)_{11}\right|=1,$$
(50)
then we can write $`C_n=\mathrm{exp}\left(i\omega _n\sigma _3/2\right)`$ which is real and characterizes the rotation-angle of spin in the $`xy`$-plane. It is necessary to mention that $`\omega _n`$ may he dependent on $`z`$ and $`t`$. To determine $`\omega _n`$ one must examine the Lax equations carefully. Since $`\mathrm{exp}\left(i\omega _n\sigma _3/2\right)`$ denotes a rotation around the $`z`$-axis, it does not affect the value of $`S^z`$. Substituting (28) into (46) and taking the limits as $`q\mathrm{}`$ and $`q0`$ respectively, we obtain
$`_z\left\{C_nP_n\left(0\right)\right\}`$ $`=`$ $`L_n\left(q\right)\left\{C_nP_n\left(0\right)\right\}\left\{C_nP_n\left(0\right)\right\}L_{n1}\left(q\right),`$
$`_z\left\{C_n\right\}`$ $`=`$ $`i2\rho S^z\left(n\right)\sigma _3\left\{C_n\right\}+\left\{C_n\right\}i2\rho S^z\left(n1\right)\sigma _3.`$
Comparing these two equations, we derive
$$C_n=\left(\mathrm{\Delta }_n\right)^{\frac{1}{2}}\left(\begin{array}{cc}q_n\left|\mathrm{{\rm Y}}_n\right|^2+\overline{q}_n\left|\xi _n\right|^2& 0\\ 0& \overline{q}_n\left|\mathrm{{\rm Y}}_n\right|^2+q_n\left|\xi _n\right|^2\end{array}\right).$$
(51)
The Eq. (47) gives
$$\mathrm{{\rm Y}}_n=f_n+if_n^1,\xi _n=f_nif_n^1,$$
(52)
where
$$f_n^2=\mathrm{exp}\left(\mathrm{\Theta }_n+i\mathrm{\Phi }_n\right),$$
here the parameters $`\mathrm{\Theta }_n`$ and $`\mathrm{\Phi }_n`$ are defined in Eq. (7). Setting $`n=1`$ and substituting the Eqs. (28), (44), (51), and (52) into (48) we can obtain the one-soliton solution (8). Setting $`n=2`$ and with the similar procedure the expression of the two-soliton solution (14) is derived.
Figure Captions
Figure 1 The amplitude and size of soliton in Eq. (8) vary periodically with time, where $`\mu _1=0.45`$, $`\nu _1=0.7`$, $`z_1=14`$, $`\varphi _1=0.5`$, $`\rho ^2=0.375`$. The unit for time $`t`$ is $`2\lambda /\mathrm{}`$ and $`a`$ for space $`z`$.
Figure 2 The magnetization $`M_3`$ (the integral $`_{\mathrm{}}^{\mathrm{}}๐z(1S_1^x)`$) vary with time periodically, where $`\mu _1=0.45`$, $`\nu _1=0.7`$, $`z_1=14`$, $`\varphi _1=0.5`$, $`\rho ^2=0.375`$. The unit for time $`t`$ is $`2\lambda /\mathrm{}`$ and $`a`$ for space $`z`$.
|
warning/0506/math0506482.html
|
ar5iv
|
text
|
# Subgroups of direct products of elementarily free groups
## 1 Limit Groups
Limit groups arise naturally from several points of view. Most geometrically, they are the finitely generated groups whose Cayley graph can be obtained as the pointed Gromov-Hausdorff limit of a sequence of Cayley graphs of a fixed free group (with a varying choice of generating set of fixed finite cardinality) . Limit groups are precisely those finitely generated groups $`L`$ that are fully residually free: for any finite subset $`TL`$ there exists a homomorphism from $`L`$ to a free group that is injective on $`T`$. It is in this guise that limit groups were studied extensively by Kharlampovich and Myasnikov , .
The name limit group was introduced by Zlil Sela to emphasise the fact that these are precisely the groups that arise when one takes limits of stable sequences of homomorphisms $`\varphi _n:GF`$, where $`G`$ is an arbitrary finitely generated group and $`F`$ is a free group; stable means that for each $`gG`$ either $`I_g=\{n:\varphi _n(g)=1\}`$ or $`J_g=\{n:\varphi _n(g)1\}`$ is finite, and the limit of $`(\varphi _n)`$ is the quotient of $`G`$ by $`\{g|I_g|=\mathrm{}\}`$. A good background reference source for limit groups is .
### 1.1 $`\omega `$-residually free towers
Our results rely heavily on Selaโs version (, 1.12; cf ) of the fundamental theorem that characterizes limit groups and elementarily free groups in terms of $`\omega `$-residually free towers.
###### Definition 1.1
An $`\omega `$-rft space of height $`h`$ is defined by induction on $`h`$. An $`\omega `$-rft group is the fundamental group of an $`\omega `$-rft space.
An $`\omega `$-rft space of height $`0`$ is the wedge (1-point union) of a finite collection of circles, closed hyperbolic surfaces and tori $`๐^n`$ (of arbitrary dimension), except that the closed surface of Euler characteristic $`1`$ is excluded<sup>1</sup><sup>1</sup>1dropping this exclusion, and the corresponding one in quadratic blocks, would not affect the results of our paper.
An $`\omega `$-rft space $`Y_n`$ of height $`h`$ is obtained from an $`\omega `$-rft space $`Y_{h1}`$ of height $`h1`$ by attaching a block of one of the following types:
Abelian: $`Y_h`$ is obtained from $`Y_{h1}๐^m`$ by identifying a coordinate circle in $`๐^m`$ with any loop $`c`$ in $`Y_{h1}`$ such that $`c`$ is a maximal abelian subgroup of $`\pi _1Y_{h1}`$.
Quadratic: One takes a connected, compact surface $`\mathrm{\Sigma }`$ that is either a punctured torus or has Euler characteristic at most $`2`$, and obtains $`Y_h`$ from $`Y_{h1}\mathrm{\Sigma }`$ by identifying each boundary component of $`\mathrm{\Sigma }`$ with a homotopically non-trivial loop in $`Y_{h1}`$; these identifications must be chosen so that there exists a retraction $`r:Y_hY_{h1}`$, and $`r_{}(\mathrm{\Sigma })\pi _1Y_{h1}`$ must be non-abelian.
An $`\omega `$-rft space is called hyperbolic if no tori are used in its construction.
###### Theorem 1.1
(; see also , ) A group is elementarily free if and only if it is the fundamental group of a hyperbolic $`\omega `$-rft space.
###### Theorem 1.2
(\[20, 1.12\]; see also ) Limit groups are precisely the finitely generated subgroups of $`\omega `$-rft groups.
A useful sketch proof of the latter theorem can be found in .
This powerful theorem allows one to prove many interesting facts about limit groups by induction on height.
###### Definition 1.2
The height of a limit group $`S`$ is the minimal height of an $`\omega `$-rft group that has a subgroup isomorphic to $`S`$.
Our approach to understanding an arbitrary limit group $`S`$ will be to embed it in the fundamental group of an $`\omega `$-rft group $`L`$, take the splitting of $`L`$ given by Lemma 1.4 below, and then decompose $`S`$ as a graph of groups corresponding to its action on the Bass-Serre tree of this splitting (subsection 1.3 below). We refer the reader to and for background on the Bass-Serre theory of groups acting on trees, which is used extensively throughout this paper.
### 1.2 Elementarily-free versus hyperbolic limit groups
A straightforward application of the local gluing lemma (, II.11.3) allows one to deduce from the inductive description given above that every hyperbolic $`\omega `$-rft supports a locally CAT$`(1)`$ metric and every $`\omega `$-rft supports a locally CAT$`(0)`$ metric. In particular elementarily free groups are hyperbolic. Moreover, an induction on height, combined with some elementary Bass-Serre theory, shows that any finitely generated subgroup of an elementarily free group $`L`$ is the fundamental group of a locally CAT$`(1)`$ subcomplex of a suitable covering space of the tower space for $`L`$. Thus:
###### Lemma 1.3
All finitely generated subgroups of elementarily free groups are hyperbolic limit groups.
###### Remark 1.1
We emphasize that the embedding in Theorem 1.2 does not preserve hyperbolicity. For this reason, our proof of Theorem 0.1 does not extend to arbitrary hyperbolic limit groups.
### 1.3 Graph of groups decompositions of elementarily free groups
Recall that a graph of groups decomposition is said to be $`n`$-acylindrical if in the action on the associated Bass-Serre tree, the stabilizer of any edge-path of length greater than $`n`$ is trivial.
###### Lemma 1.4
If $`L`$ is the fundamental group of an $`\omega `$-rft space $`Y=Y_h`$ of height $`h1`$, then $`L`$ is the fundamental group of a 2-vertex graph of groups: one of the vertices is $`\pi _1Y_{h1}`$ and the other is a free or free-abelian group of finite rank at least 2; the edge groups are maximal infinite cyclic subgroups of the second vertex group. This decomposition is 2-acylindrical.
Proof. The splitting described is that which the Seifert-van Kampen Theorem associates to the decomposition of $`Y`$ into $`Y_{h1}`$ and the final block in the construction.
If the final block is quadratic, then the edge groups are precisely the peripheral subgroups of the surface $`\mathrm{\Sigma }`$. As such they are maximal and form a malnormal family: if $`C,C^{}\mathrm{\Sigma }`$ are two cyclic subgroups in the conjugacy classes of two (not necessarily distinct) edge groups and $`x\mathrm{\Sigma }C`$, then $`x^1CxC^{}=\{1\}`$.
Any edge-path in the Bass-Serre tree of length greater than 2 must contain a vertex with stabilizer a conjugate of $`\mathrm{\Sigma }`$, and the intersection of the stabilisers of the incident edges will be of the form $`x^1CxC^{}`$. Thus the splitting is 2-acylindrical.
Suppose now that the final block in the construction of $`Y=Y_h`$ is abelian. A straightforward use of the $`\omega `$-residually free condition yields the following facts about limit groups: first, if two non-trivial elements of an $`\omega `$-residually free group commute and are conjugate, then they are equal; second, if $`x,y,z`$ are non-trivial and $`[x,y]=[y,z]=1`$, then $`[x,z]=1`$. It follows from the first of these facts that if $`A`$ is an abelian subgroup and $`yAzAz^1`$, then $`[y,z]=1`$. If $`y1`$, it then follows from the second fact that $`z`$ commutes with $`A`$. Thus maximal abelian subgroups of limit groups are malnormal.
By construction, the edge stabilizer in our splitting is maximal-abelian in $`Y_{h1}`$. Hence it is malnormal in $`Y_{h1}`$ and the splitting is 2-acylindrical. $`\mathrm{}`$
###### Corollary 1.5
If $`\mathrm{\Gamma }`$ is a non-cyclic, finitely generated, freely indecomposable subgroup of an elementarily free group $`G`$, then $`\mathrm{\Gamma }`$ is the fundamental group of a 2-acylindrical graph of groups, in which one of the vertex groups is the fundamental group of a compact surface $`\mathrm{\Sigma }`$ and the incident edge groups are distinct peripheral subgroups of $`\mathrm{\Sigma }`$.
Proof. By Selaโs Theorem 1.1, $`G`$ is the fundamental group of some $`\omega `$-rft space $`Y`$. If $`Y`$ can be chosen of height $`0`$, then $`\mathrm{\Gamma }`$ is the fundamental group of a closed surface $`\mathrm{\Phi }`$ of Euler characteristic at most $`2`$. In this case the splitting of $`\mathrm{\Phi }`$ along any nontrivial $`2`$-sided simple closed curve induces the desired decomposition of $`\mathrm{\Gamma }`$. Otherwise, apply Lemma 1.4 to $`G`$, let $`T`$ be the minimal $`\mathrm{\Gamma }`$-invariant subtree of the resulting Bass-Serre tree for $`G`$, and take the resulting graph-of-groups decomposition for $`\mathrm{\Gamma }`$ with underlying graph $`T/\mathrm{\Gamma }`$. $`\mathrm{}`$
### 1.4 Subgroup Separability
Let $`๐ซ`$ be a class of groups, e.g. free or finite. Let $`\mathrm{\Gamma }`$ be a group. A subgroup $`S\mathrm{\Gamma }`$ is closed in the pro-$`๐ซ`$ topology if for every $`xS`$ there exists a homomorphism $`f:\mathrm{\Gamma }F`$ where $`F๐ซ`$ and $`f(x)f(S)`$.
If $`\{1\}`$ is closed in the pro-$`๐ซ`$ topology then $`\mathrm{\Gamma }`$ is said to be residually $`๐ซ`$.
Notation: If all infinite cyclic subgroups $`S\mathrm{\Gamma }`$ are closed in the profinite topology, then we say that $`\mathrm{\Gamma }`$ is $``$-separable.
###### Proposition 1.6
In a finitely generated free group $`F`$, every finitely generated subgroup is closed in the pro-finite topology.
Proof. This follows easily from Marshall Hallโs theorem , which states that every finitely generated subgroup is a free factor of a subgroup of finite index in $`F`$. $`\mathrm{}`$
###### Corollary 1.7
If $`S\mathrm{\Gamma }`$ is closed in the pro-free topology then it is closed in the pro-finite topology.
###### Lemma 1.8
If $`\mathrm{\Gamma }`$ is a residually free group with no non-cyclic abelian subgroups, then $`\mathrm{\Gamma }`$ is $``$-separable.
Proof. Let $`C=c\mathrm{\Gamma }`$ and suppose that $`\gamma \mathrm{\Gamma }`$ with $`\gamma C`$. There are two cases to consider: either $`[\gamma ,c]1`$, or $`H=c,\gamma `$ is cyclic.
In the first case, since $`\mathrm{\Gamma }`$ is residually finite, there exists a homomorphism $`f:\mathrm{\Gamma }Q`$ where $`Q`$ is finite and $`f([\gamma ,c])1`$; in particular $`\gamma f(C)`$.
In the second case, since $`\mathrm{\Gamma }`$ is residually free, there exists a homomorphism $`\varphi :\mathrm{\Gamma }F`$ where $`F`$ is free such that $`\varphi (\gamma )1`$. Since $`\varphi |_H`$ is injective, $`\varphi (\gamma )\varphi (C)`$, and since $`F`$ is $``$-separable, there exists a finite quotient such that the image of $`\gamma `$ does not lie in the image of $`C`$. $`\mathrm{}`$
###### Corollary 1.9
Hyperbolic limit groups (in particular elementarily free groups) are $``$-separable.
A slight variation on the preceding proof shows that if a group $`\mathrm{\Gamma }`$ is residually free, then each of its maximal abelian subgroups is closed in the pro-free topology. Conversations with Zlil Sela and Henry Wilton convinced us that, in the light of Theorem 1.2, it is not difficult to show that all abelian subgroups of limit groups are closed in the pro-finite (indeed pro-free) topology.
## 2 Some Bass-Serre Theory
As noted in the previous section, an elementarily free group acts $`2`$-acylindrically on the Bass-Serre tree arising from its $`\omega `$-rf tower decomposition. In this section we deduce from this that every nontrivial normal subgroup contains an element that acts hyperbolically on the tree โ a fact that will be important later.
Recall that an automorphism $`\gamma `$ of a tree $`T`$ is hyperbolic if it has no fixed points, and that the unique $`\gamma `$-minimal subtree of $`T`$ is then an isometrically embedded line $`A_\gamma `$ called the axis of $`\gamma `$. In contrast, an automorphism with at least one fixed point is elliptic.
###### Proposition 2.1
Let $`\mathrm{\Gamma }`$ be a group acting on a tree $`T`$.
1. If $`\alpha ,\beta \mathrm{\Gamma }`$ are hyperbolic with disjoint axes $`A_\alpha `$ and $`A_\beta `$, then $`\alpha \beta `$ is hyperbolic and its axis contains the unique shortest arc from $`A_\alpha `$ to $`A_\beta `$.
2. If $`\alpha ,\beta \mathrm{\Gamma }`$ are elliptic with $`\mathrm{Fix}(\alpha )\mathrm{Fix}(\beta )=\mathrm{}`$ then $`\alpha \beta `$ is hyperbolic.
3. If a finite family of convex subsets in $`T`$ intersects pairwise, then the intersection of the entire family is non-empty.
4. If $`\mathrm{\Gamma }`$ is finitely generated then either $`\mathrm{\Gamma }`$ fixes a point of $`T`$, or else $`\mathrm{\Gamma }`$ contains hyperbolic isometries.
5. If the action of $`\mathrm{\Gamma }`$ is $`n`$-acylindrical for some $`n`$, then either $`\mathrm{\Gamma }`$ fixes a point of $`T`$, or else $`\mathrm{\Gamma }`$ contains hyperbolic isometries.
6. If $`\mathrm{\Gamma }`$ contains hyperbolic elements, then the union of the axes of such elements is the unique minimal $`\mathrm{\Gamma }`$-invariant subtree of $`T`$.
Proof.
1: Choose an edge $`e`$ that lies in the arc joining $`A_\alpha `$ to $`A_\beta `$. Let $`X,Y`$ denote the components of $`T\{e\}`$ containing $`A_\alpha ,A_\beta `$ respectively. Note that $`\alpha ^{\pm 1}(Ye)X`$ while $`\beta ^{\pm 1}(Xe)Y`$. Thus $`e`$ is contained in the geodesic path from $`(\alpha \beta )^n(e)`$ to $`(\alpha \beta )^m(e)`$ for all $`n,m>0`$, and the result follows.
2: A similar argument applies, replacing $`A_\alpha ,A_\beta `$ by $`\mathrm{Fix}(\alpha ),\mathrm{Fix}(\beta )`$ respectively.
3: An inductive argument reduces us to the case of three convex sets. Choose a point in each of the three pairwise intersections, and consider the geodesic triangle in $`T`$ with these points as vertices. Each of our three sets contains one of the sides of the triangle. And since we are in a tree, the three sides have a common point.
4: By 2, if $`\mathrm{\Gamma }`$ is generated by a finite set of elliptics then either the product of some pair of these generators is hyperbolic or else the fixed-point sets of each pair intersect non-trivially, in which case it follows from 3 that $`\mathrm{\Gamma }`$ has a fixed point.
5: Without loss of generality, we may assume that $`T`$ is a minimal $`\mathrm{\Gamma }`$-tree. We will reach a contradiction by assuming that $`\mathrm{\Gamma }`$ has no hyperbolic elements and no fixed point.
Consider the union $`U`$ of the sets $`\mathrm{Fix}(\gamma )`$ as $`\gamma `$ ranges over $`\mathrm{\Gamma }\{1\}`$. We claim that $`U`$ is convex, hence a subtree. Indeed, given $`x_1,x_2U`$ there exist $`\gamma _1,\gamma _2N`$ such that $`x_i\mathrm{Fix}(\gamma _i)`$ for $`i=1,2`$, and since $`\gamma _1\gamma _2`$ is not hyperbolic, $`\mathrm{Fix}(\gamma _1)\mathrm{Fix}(\gamma _2)\mathrm{}`$ by 2. Thus $`[x_1,x_2]`$ lies in the convex set $`\mathrm{Fix}(\gamma _1)\mathrm{Fix}(\gamma _2)`$. The tree $`U`$ is nonempty and $`\mathrm{\Gamma }`$-invariant, so $`U=T`$.
Given an edge $`e`$ of $`T`$, we may choose vertices $`x,yT`$ in distinct components of $`T\{e\}`$ a distance at least $`n`$ from $`e`$. Such vertices exist because $`T`$ does not have vertices of valence 1 (since removing the edges incident at such vertices would yield a proper invariant subtree), and therefore each edge in the complement of $`e`$ can be extended to an infinite ray in its component of $`T\{e\}`$.
Let $`\alpha ,\beta \mathrm{\Gamma }`$ be non-trivial elements such that $`x\mathrm{Fix}(\alpha )`$ and $`y\mathrm{Fix}(\beta )`$. Then $`e`$ lies in the convex set $`\mathrm{Fix}(\alpha )\mathrm{Fix}(\beta )`$. If $`e`$ lies in the convex set $`\mathrm{Fix}(\alpha )`$, then so does the segment joining $`x`$ to $`e`$. But then this segment is fixed by $`\alpha `$, contradicting the fact that $`\mathrm{\Gamma }`$ acts $`n`$-acylindrically.
6: This follows from 1. $`\mathrm{}`$
###### Corollary 2.2
If an action of a group $`\mathrm{\Gamma }`$ on a tree $`T`$ is minimal and $`n`$-acylindrical for some $`n>0`$, and $`\mathrm{\Gamma }`$ has no global fixed point in $`T`$, then every non-trivial normal subgroup $`N\mathrm{\Gamma }`$ contains hyperbolic elements.
Proof. Since $`N`$ is normal in $`\mathrm{\Gamma }`$, the subset $`\mathrm{Fix}(N)`$ of $`T`$ is convex and $`\mathrm{\Gamma }`$-invariant, so is either the whole of $`T`$ or is empty. In the first case, $`T`$ must have finite diameter by the $`n`$-acylindrical property. But then $`\mathrm{\Gamma }`$ has a global fixed point in $`T`$, contrary to hypothesis. Hence $`N`$ has no global fixed point in $`T`$, and the result follows from Proposition 2.1(5). $`\mathrm{}`$
###### Remark 2.1
The following example illustrates the difficulties one faces in trying to sharpen the above statement. Consider the action of the HNN-extension $`B=a,ttat^1=a^2`$ of the cyclic group $`a`$ on its Bass-Serre tree. This action is cocompact with cyclic edge stabilisers. The normal closure $`N`$ of $`a`$ does not contain any hyperbolic elements.
## 3 Hyperbolics are almost stable letters
The proof of the following theorem is based on an idea introduced by John Stallings in the context of a free group acting freely on its Cayley tree (see also \[17, Theorem 2.2\] and \[12, Lemma 15.22\]). This result is a restatement of Theorem 0.3.
###### Theorem 3.1
Let $`\mathrm{\Gamma }`$ be a group that acts minimally on a tree $`T`$ and let $`e`$ be an edge whose stabiliser $`A\mathrm{\Gamma }`$ is closed in the pro-finite topology.
If $`t\mathrm{\Gamma }`$ is a hyperbolic isometry whose axis contains $`e`$, then there exists a subgroup $`H\mathrm{\Gamma }`$ of finite index such that $`H`$ is an HNN-extension with stable letter $`t`$ and amalgamated subgroup $`AH`$.
Proof. Consider the segment $`\lambda `$ of the axis of $`t`$ that begins with the edge $`e`$ and ends with the edge $`t(e)`$. Let
$$e=g_0(e),g_1(e),\mathrm{},g_n(e)=t(e)$$
be the finitely many edges in $`\lambda `$ that belong to the $`\mathrm{\Gamma }`$-orbit of $`e`$. The elements $`g_i`$ are not well-defined (unless $`A=1`$). But the left cosets $`g_iA`$ are well-defined and pairwise distinct.
Since $`A\mathrm{\Gamma }`$ is closed in the profinite topology, there exists a finite-index normal subgroup $`K\mathrm{\Gamma }`$ such that the left cosets $`g_0(AK),\mathrm{},g_n(AK)`$ are pairwise distinct.
Bass-Serre theory expresses $`\mathrm{\Gamma }`$ as the fundamental group of a graph of groups, where one of the edge-groups is $`A`$. Using the construction described by Scott and Wall in \[18, Proposition 3.6\], we can construct a classifying space $`X`$ for $`\mathrm{\Gamma }`$ as a graph of aspherical spaces.
Thus $`\mathrm{\Gamma }`$ acts freely on the universal cover $`\stackrel{~}{X}`$ of $`X`$, and there is a $`\mathrm{\Gamma }`$-equivariant map $`f:\stackrel{~}{X}T`$ such that $`E=f^1(e)`$ is a product (โcylinderโ) $`\stackrel{~}{X}_A\times (0,1)`$ for some $`K(A,1)`$-space $`X_A`$. We identify $`\stackrel{~}{X}_A`$ with $`\stackrel{~}{X}_A\times \{\frac{1}{2}\}E`$, and choose a point $`\stackrel{~}{x}_0\stackrel{~}{X}_A`$ as a base-point for $`\stackrel{~}{X}`$. We also choose a path $`\stackrel{~}{\tau }`$ in $`\stackrel{~}{X}`$ from $`\stackrel{~}{x}_0`$ to $`t(\stackrel{~}{x}_0)`$. Since $`\stackrel{~}{X}_A`$, and each of its $`\mathrm{\Gamma }`$-translates, has a bicollared neighbourhood in $`\stackrel{~}{X}`$, we may assume that $`\stackrel{~}{\tau }`$ is transverse to $`g(\stackrel{~}{X}_A)`$ for each $`g\mathrm{\Gamma }`$. Moreover, if we further assume that the total number of points of intersection of $`\stackrel{~}{\tau }`$ with $`_gg(\stackrel{~}{X}_A)`$ is as small as possible, then $`\stackrel{~}{\tau }`$ crosses $`g(\stackrel{~}{X}_A)`$ transversely in a single point for $`g=g_1,\mathrm{},g_{n1}`$, meets $`g(\stackrel{~}{X}_A)`$ in one of its endpoints for $`g=g_0=1`$ and $`g=g_n=t`$, and is disjoint from $`g(\stackrel{~}{X}_A)`$ for all other $`g`$.
Let $`x`$ be the image of $`\stackrel{~}{x}`$ in $`X`$, and $`\tau `$ the image of $`\stackrel{~}{\tau }`$ in $`X`$. We use $`x`$ as the base-point for $`X`$. Then $`\tau `$ is a closed path in $`X`$ representing the element $`t`$ of $`\pi _1(X,x)\mathrm{\Gamma }`$.
Consider the covering space $`Y=\stackrel{~}{X}/K`$ of $`X`$ corresponding to the normal subgroup $`K`$. Let $`E_0`$ denote the image of $`E`$ in $`Y`$, and let $`X_A^{}`$ be the image of $`\stackrel{~}{X}_A`$. The quotient group $`\mathrm{\Gamma }/K`$ acts on $`Y`$, and for each $`g\mathrm{\Gamma }`$, $`E_0`$ and $`(gK)(E_0)`$ either coincide (if $`gAK`$) or are disjoint. In particular, $`E_0=(g_0K)(E_0),E_1=(g_1K)(E_0),\mathrm{},E_n=(g_nK)(E_0)`$ are pairwise disjoint. Let $`y_0`$ be the image in $`Y`$ of $`\stackrel{~}{x}_0`$, and $`\tau ^{}`$ the image in $`Y`$ of $`\stackrel{~}{\tau }`$. Then $`\tau ^{}`$ is a path from $`y_0`$ to $`(tK)(y_0)`$ that intersects each $`(g_iK)(X_A^{})`$ in precisely one point.
We cut $`E_0`$ and $`E_n`$ along the subspace $`X_A^{}`$ and its translate $`(tK)(X_A^{})=(g_nK)(X_A^{})`$, creating four (โboundaryโ) subspaces $`_{}E_0,_{}E_n`$ and $`_+E_0,_+E_n`$, each homeomorphic to $`X_A^{}`$. This cutting transforms $`Y`$ to a space $`Y^{}`$, and $`\tau ^{}`$ becomes a path in $`Y^{}`$ from the copy of $`y_0`$ in $`_+E_0`$ to the copy of $`(tK)(y_0)`$ in $`_{}E_n`$.
We form a new space $`Z`$ from $`Y^{}`$ by identifying $`_{}E_0`$ to $`_+E_n`$, and $`_+E_0`$ to $`_{}E_n`$, using the restriction of the homeomorphism $`(tK):E_0E_n`$. Note that this identification in particular identifies the two endpoints of $`\tau ^{}`$ to a single point $`z`$, which we may regard as the base-point of $`Z`$. The image of $`\tau ^{}`$ in $`Z`$ is thus a closed path $`\overline{\tau }`$ based at $`z`$. The covering map $`YX`$ induces a covering map $`ZX`$. (This $`|G:K|`$-fold cover is not necessarily connected or regular.)
Consider the component $`Z_0`$ of $`Z`$ that contains the copy $`๐`$ of $`X_A^{}`$ that is the image of $`_{}E_0`$ and $`_+E_n`$. Then $`z๐Z_0`$.
The restriction of $`ZX`$ to $`Z_0`$ is a covering of $`X`$ corresponding to a finite index subgroup $`H=\pi _1(Z_0,z)\pi _1(X,x)=\mathrm{\Gamma }`$ that contains $`t`$, since $`\overline{\tau }`$ is a closed path in $`Z_0`$ which projects to $`\tau `$.
Since this loop $`\overline{\tau }`$ representing $`tH=\pi _1(Z_0,z)`$ crosses $`๐`$ transversely precisely once, the Seifert-van Kampen decomposition of $`\pi _1(Z_0,z)`$ expresses $`H`$ as an HNN-extension with stable letter $`t`$ and associated subgroup $`\pi _1(๐)=AH`$. $`\mathrm{}`$
###### Corollary 3.2
Let $`\mathrm{\Gamma }`$ be a group that acts minimally on a tree $`T`$. Let $`e`$ be an edge whose stabiliser $`A\mathrm{\Gamma }`$ is closed in the pro-finite topology.
If $`N\mathrm{\Gamma }`$ is a normal subgroup that contains a hyperbolic isometry, then there exists a subgroup $`H\mathrm{\Gamma }`$ of finite index and an element $`tNH`$ such that $`H`$ is an HNN-extension with stable letter $`t`$ and amalgamated subgroup $`AH`$.
Proof. Since $`N`$ is normal, the union $`UT`$ of the axes of the hyperbolic elements in $`N`$ is $`\mathrm{\Gamma }`$-invariant. But $`U`$ is a subtree (Proposition 2.1(6)) and the action of $`\mathrm{\Gamma }`$ is minimal, so $`U=T`$. Hence there exists a hyperbolic element $`tN`$ whose axis contains $`e`$. $`\mathrm{}`$
### 3.1 The curve-lifting lemma
We mentioned in the introduction that Theorem 3.1 generalizes Scottโs result that every non-trivial element in the fundamental group of a closed surface can be represented by a simple closed curve in some finite-sheeted covering of that surface. In our proof of Theorem 0.1 we shall need the following refinement of this fact.
###### Lemma 3.3
Let $`\mathrm{\Sigma }`$ be a compact surface with non-positive Euler characteristic, $`X`$ a space with $``$-separable fundamental group, and $`f:\mathrm{\Sigma }X`$ a $`\pi _1`$-injective map. If $`w`$ is a non-trivial element of $`\pi _1\mathrm{\Sigma }`$, then there exists a finite-sheeted cover $`\overline{X}`$ of $`X`$, and a simple closed curve $`\alpha `$ in the induced cover $`\overline{\mathrm{\Sigma }}`$ of $`\mathrm{\Sigma }`$, such that the image of $`\alpha `$ in $`\mathrm{\Sigma }`$ represents $`w`$.
Proof. Since $`f`$ is $`\pi _1`$-injective, we can identify $`\pi _1(\mathrm{\Sigma })`$ with the subgroup $`f_{}(\pi _1(\mathrm{\Sigma }))`$ of $`\pi _1(X)`$. We do so implicitly throughout the proof without further comment.
Suppose first that $`w`$ is a proper power of some element $`u\pi _1\mathrm{\Sigma }`$; say $`w=u^n`$. By $``$-separability, there is a subgroup of finite index in $`\pi _1(X)`$ that contains $`w=u^n`$ but contains none of $`u,u^2,\mathrm{},u^{n1}`$. Replacing $`X`$ and $`\mathrm{\Sigma }`$ by the corresponding finite covers, we may assume that $`w`$ is not a proper power in $`\pi _1\mathrm{\Sigma }`$. We fix a constant-curvature Riemannian metric on $`\mathrm{\Sigma }`$ such that the boundary of $`\mathrm{\Sigma }`$ is totally geodesic. With respect to this metric, the conjugacy class of $`w`$ is represented by a closed geodesic $`\beta `$ with transverse self-intersection.
If $`\beta `$ is not an embedding, we can express it as a concatenation of paths $`\beta =\beta _1\beta _2\beta _3`$ where $`\beta _2`$ is an embedded closed path, which must be essential in $`\mathrm{\Sigma }`$ since it is a based geodesic in a non-positively curved metric. Moreover, the conjugacy class represented by $`\beta _2`$ has a trivial intersection with $`w`$, because the elements of this cyclic subgroup are represented by closed geodesics $`\beta ^n`$ which are strictly longer than $`\beta _2`$.
Since $`\pi _1\mathrm{\Sigma }\pi _1X`$ is injective, and $`\pi _1X`$ is $``$-separable, there is a finite sheeted cover of $`X`$ such that $`\beta `$ lifts to a closed geodesic in the induced cover of $`\mathrm{\Sigma }`$ but the corresponding lift of $`\beta _2`$ is not closed. This lift of $`\beta `$ therefore has fewer double points than $`\beta `$.
Repeating this process, we eventually arrive at a finite cover of $`X`$ such that $`\beta `$ lifts to an embedded closed geodesic $`\alpha `$ in the induced cover of $`\mathrm{\Sigma }`$. The proof is complete. $`\mathrm{}`$
## 4 The double-coset lemma
In the introduction, we explained that our proof of Theorem 0.1 has the advantage over that when homological finiteness fails, one can construct explicit topological cycles that demonstrate the lack of finiteness. That construction is contained in the following proof.
###### Theorem 4.1 (Double Coset Lemma)
For $`i=1,\mathrm{},n`$, let $`G_i`$ be an HNN-extension with stable letter $`w_i`$ and associated subgroups $`A_i`$ and $`B_i`$. Let $`G=_{i=1}^nG_i`$, $`A=_{i=1}^nA_i`$, and let $`LG`$ be a subgroup containing $`j_i(w_i)`$ ($`i=1,\mathrm{},n`$), where $`j_i:G_iG`$ is the canonical injection. Suppose also that $`p_i(L)=G_i`$ for all $`i`$, where $`p_i:GG_i`$ is the canonical projection. Then, the $`n`$-th homology group $`H_n(L,)`$ contains a subgroup isomorphic to
$$_LG_A$$
(the $``$ vector space with basis the set of double cosets $`\{LgA,gG\}`$).
Proof. We can construct, for each $`i`$, a $`K(G_i,1)`$-complex $`X_i`$ by starting with a classifying space for the base of the HNN extension and forming a mapping torus corresponding to the given isomorphism $`A_iB_i`$. In this construction, $`w_i`$ corresponds to a $`1`$-cell $`W_i`$ with both endpoints at the base-point of $`X_i`$, and $`W_i`$ appears only in the boundaries of those $`2`$-cells corresponding to defining relations $`w_i^1aw_i=b`$ for a set of generators $`a`$ of $`A_i`$. There is no loss of generality in assuming that each of the generators $`a,b`$ corresponds to a single 1-cell, and hence the 2-cells involving $`w_i`$ have attaching maps of length 4.
Let $`X=X_1\times \mathrm{}\times X_n`$, let $`\stackrel{~}{X}`$ be the universal cover of $`X`$, upon which $`G`$ acts on the left, and let $`\overline{X}=L\backslash \stackrel{~}{X}`$ be the covering complex corresponding to the subgroup $`LG`$.
We work with cellular chains.
Let $`L_i:=\{gG_ij_i(g)L\}`$. Then $`j_i(L_i)`$ is the intersection of the kernels of $`p_k|_L:LG_k`$ for $`ki`$, so is normal in $`L`$. Hence $`L_i=p_ij_i(L_i)`$ is normal in $`p_i(L)=G_i`$. Since also $`j_i(G_i)`$ commutes with $`j_k(G_k)`$ for $`ik`$, it follows that $`j_i(L_i)`$ is normal in $`G`$. But $`w_iL_i`$, so all conjugates of $`j_i(w_i)`$ belong to $`j_i(L_i)L`$. Hence all the lifts of $`W_i`$ in $`\overline{X}`$ are $`1`$-cycles.
Now let $`W`$ denote the cellular $`n`$-cycle $`W_1\times \mathrm{}\times W_n`$ in $`X`$. By the above, all the lifts of $`W`$ to $`\overline{X}`$ are also $`n`$-cycles. We can identify these lifts with right cosets of $`L`$ in $`G`$ as follows. Choose an $`n`$-chain $`\stackrel{~}{W}`$ in $`\stackrel{~}{X}`$ that covers $`W`$. Then the orbit of $`W`$ under the $`G`$-action consists of pairwise distinct $`n`$-chains $`g(\stackrel{~}{W})`$ for $`gG`$, and $`g(\stackrel{~}{W})`$ and $`h(\stackrel{~}{W})`$ cover the same $`n`$-chain in $`\overline{X}`$ if and only if $`Lg=Lh`$.
These chains in $`\overline{X}`$, as has been observed above, are in fact $`n`$-cycles, so generate a subgroup $`M`$ of $`H_n(\overline{X},)=H_n(L,)`$. This subgroup is then a homomorphic image of the free $``$-module on the set of right cosets $`Lg`$ of $`L`$ in $`G`$, in other words, $`(L\backslash G)=_LG`$. The kernel of the corresponding homomorphism is the intersection of this group of $`n`$-cycles with the group of (cellular) $`(n+1)`$-boundaries of $`\overline{X}`$.
Now the boundary of an $`(n+1)`$-cell $`\alpha `$ of $`X`$ involves $`W`$ only if that $`(n+1)`$-cell is a cube formed as the product of a $`2`$-cell $`\beta `$ of some $`X_i`$ whose boundary involves $`W_i`$ and of the $`(n1)`$ $`1`$-cells $`W_k`$ with $`ki`$. (We arranged in the first paragraph that $`\beta `$ have an attaching map of length 4, with two sides corresponding to $`W_i`$.) Therefore a lift $`\stackrel{~}{\alpha }`$ of $`\alpha `$ in $`\stackrel{~}{X}`$ has the combinatorial structure of a cube, and the coefficient of the $`n`$-dimensional face $`\stackrel{~}{W}`$ in the cellular $`n`$-chain $`(\stackrel{~}{\alpha })`$ is $`1a`$ for some $`aA_i`$.
Hence $`M`$ has a homomorphic image isomorphic to the quotient of
$$_LG$$
by the submodule generated by $`\{(_LG)(1a)aA\}`$. But this quotient is just
$$_LG_A.$$
Since this is a free $``$-module, the epimorphism
$$M_LG_A$$
splits, so $`M`$, and hence also $`H_n(L,)`$, has a subgroup isomorphic to
$$_LG_A,$$
as claimed. $`\mathrm{}`$
We will use the double-coset lemma in the proof of our main result, where we shall need the following elementary properties of double cosets in our calculations.
###### Lemma 4.2
Let $`G`$ be a group, $`g`$ an element of $`G`$, and $`A,B,C`$ subgroups of $`G`$ such that $`BC`$. Then the intersection of $`C`$ and the double coset $`AgB`$ is either empty or has the form $`(AC)cB`$ for some $`cC`$.
Proof. Suppose that $`c_1,c_2AgBC`$. Then we can write $`c_i=a_igb_i`$ for $`i=1,2`$. Then $`(a_2a_1^1)=c_2b_2^1b_1c_1^1AC`$, so $`c_2(AC)c_1B`$. Hence, for any $`cAgBC`$, we have $`AgBC(AC)cB`$. The converse inclusion is immediate, using the equation $`AcB=AgB`$. $`\mathrm{}`$
###### Corollary 4.3
If $`A,B,C`$ are subgroups of a group $`G`$ such that $`BC`$ and $`|A\backslash G/B|<\mathrm{}`$, then $`|(AC)\backslash C/B|<\mathrm{}`$.
###### Lemma 4.4
Let $`G,H`$ be groups, $`A,B`$ subgroups of $`G`$ and $`g`$ an element of $`G`$. If $`\varphi :GH`$ is a homomorphism, then $`\varphi (AgB)=\varphi (A)\varphi (g)\varphi (B)`$.
###### Corollary 4.5
If $`A,B`$ are subgroups of a group $`G`$ such that $`|A\backslash G/B|<\mathrm{}`$, and $`\varphi :GH`$ is a homomorphism, then $`|\varphi (A)\backslash \varphi (G)/\varphi (B)|<\mathrm{}`$.
## 5 The Main Theorem
In this section we prove the main theorem, Theorem 0.1, in the following somewhat stronger form.
###### Theorem 5.1
Let $`G_1,\mathrm{},G_n`$ be subgroups of elementarily free groups and let $`\mathrm{\Gamma }G_1\times \mathrm{}\times G_n`$ be a subgroup. Then the following are equivalent:
1. there exist finitely generated subgroups $`\widehat{G}_iG_i`$, for $`i=1,\mathrm{},n`$, such that $`\mathrm{\Gamma }`$ is isomorphic to a finite-index subgroup of $`\widehat{G}_1\times \mathrm{}\times \widehat{G}_n`$;
2. $`\mathrm{\Gamma }`$ is of type $`FP_n()`$;
3. for each $`k=1,\mathrm{},n`$ and each finite-index subgroup $`\mathrm{\Gamma }_0\mathrm{\Gamma }`$, the $`k`$-th homology $`H_k(\mathrm{\Gamma }_0,)`$ is finite dimensional over $``$.
Proof. Finitely generated subgroups of limit groups are limit groups, and hence of type $`FP_{\mathrm{}}`$. Thus any subgroup of finite index in a direct product of finitely many such groups is also of type $`FP_{\mathrm{}}`$, so (i) implies (ii). It is clear that (ii) implies (iii), so it remains only to prove that (iii) implies (i).
Let $`G:=G_1\times \mathrm{}\times G_n`$, and let $`p_i:GG_i`$ denote the canonical projection onto the $`i`$-th factor, for $`i=1,\mathrm{},n`$.
Replacing each $`G_i`$ with $`p_i(\mathrm{\Gamma })`$, we may assume that $`p_i(\mathrm{\Gamma })=G_i`$ for all $`i`$. In particular, each $`H_1(G_i,)`$ is a homomorphic image of $`H_1(\mathrm{\Gamma },)`$, and so finite-dimensinal over $``$. Hence each $`G_i`$ is finitely generated, by \[7, Theorem 2\].
By abuse of notation, we identify $`G_i`$ with the normal subgroup
$$\{1\}\times \mathrm{}\times \{1\}\times G_i\times \{1\}\times \mathrm{}\times \{1\}$$
of $`G`$. Then we define $`L_i=G_i\mathrm{\Gamma }`$, and note that $`L_i`$ is normal in $`\mathrm{\Gamma }`$, and hence that $`L_i=p_i(L_i)`$ is normal in $`G_i=p_i(\mathrm{\Gamma })`$.
If $`L_n=\{1\}`$, then the natural projection $`G_1\times \mathrm{}\times G_nG_1\times \mathrm{}\times G_{n1}`$ is injective on $`\mathrm{\Gamma }`$, so $`\mathrm{\Gamma }`$ is isomorphic to a subgroup of $`G_1\times \mathrm{}\times G_{n1}\times \{1\}`$. An obvious induction reduces us to the case where all of the $`L_i`$ are nontrivial. We will be done if we can show that $`|G_i:L_i|<\mathrm{}`$ for each $`i`$.
If we replace $`G_n`$ by a finite-index subgroup $`\widehat{G}_n`$, and replace each $`G_j`$ by $`\widehat{G}_j=p_jp_n^1(\widehat{G}_n)`$ and $`\mathrm{\Gamma }`$ by $`\mathrm{\Gamma }(\widehat{G}_1\times \mathrm{}\times \widehat{G}_n)`$, then neither our hypotheses nor our desired conclusion is disturbed. (And likewise with $`G_i`$ in place of $`G_n`$.) We shall take advantage of this freedom several times in the sequel without further comment.
We first use this freedom to reduce to the case where none of the $`G_i`$ is cyclic. (This could also be done by appealing to \[7, Theorem 3\].) If $`G_n`$ is cyclic, then $`L_n`$ has finite index in $`G_n`$ since $`L_n\{1\}`$, so we may assume that $`L_n=G_n`$. In this case $`\mathrm{\Gamma }`$ splits as a direct product $`\mathrm{\Gamma }^{}\times `$ for some $`\mathrm{\Gamma }^{}G_1\times \mathrm{}\times G_{n1}`$. The Kรผnneth formula and the homological hypothesis on $`\mathrm{\Gamma }`$ tell us that $`H_k(\mathrm{\Gamma }_0,)`$ is finite dimensional for each $`k=1,\mathrm{},n1`$ and for each finite-index subgroup $`\mathrm{\Gamma }_0\mathrm{\Gamma }^{}`$. By induction on $`n`$, we may assume that the theorem is true for $`\mathrm{\Gamma }_0`$, from which it follows for $`\mathrm{\Gamma }`$.
Henceforth we assume that none of the $`G_i`$ is cyclic. The structure result for subgroups of elementarily free groups, Corollary 1.5, tells us that each $`G_i`$ is either freely decomposable or can be expressed as the fundamental group of a $`2`$-acylindrical graph of groups in which one of the vertex groups is the (nonabelian free) fundamental group of a surface $`\mathrm{\Sigma }_i`$ with boundary, and the incident vertex groups are distinct peripheral subgroups of $`\pi _1\mathrm{\Sigma }_i`$.
If $`G_i`$ is freely decomposable, we regard it as the fundamental group of a nontrivial graph of groups with trivial edge groups (which is in particular $`1`$-acylindrical).
The next step concerns those $`G_i`$ which are freely indecomposable, and for which $`\pi _1\mathrm{\Sigma }_iL_i`$ is nontrivial. For such $`i`$, we may choose a nontrivial element $`a_i`$ in $`\pi _1\mathrm{\Sigma }_iL_i`$. By the curve-lifting lemma, Lemma 3.3, we may assume (after replacing $`G_i`$ by a finite-index subgroup), that $`a_i`$ is represented by a simple closed curve in $`\mathrm{\Sigma }_i`$. Cutting $`\mathrm{\Sigma }_i`$ along this curve produces a refinement of the graph-of-groups structure of $`G_i`$, in which $`a_i`$ is an edge group. Moreover, this refinement remains $`2`$-acylindrical.
For those $`G_i`$ that are freely decomposable, we define $`a_i=1`$, while for the remaining $`G_i`$ we take $`a_i`$ to be a generator of a peripheral subgroup of $`\pi _1\mathrm{\Sigma }_i`$ that is an edge group in the graph-of-groups decomposition of $`G_i`$.
In all cases, $`G_i`$ has a $`2`$-acylindrical graph-of-groups decomposition with an edge group $`a_i`$. Moreover $`a_iL_i`$ whenever either $`G_i`$ is freely decomposable or $`\pi _1\mathrm{\Sigma }_iL_i\{1\}`$.
By Corollary 1.9, $`a_i`$ is closed in the profinite topology on $`G_i`$; by Corollary 2.2, the normal subgroup $`L_i`$ contains an element that acts hyperbolically on the Bass-Serre tree of the decomposition; hence we may apply Corollary 3.2. After replacing the $`G_i`$ by finite-index subgroups, we may assume that each $`G_i`$ has an HNN-decomposition with associated subgroup $`a_i`$ and stable letter $`w_iL_i`$.
By hypothesis $`H_n(\mathrm{\Gamma },)`$ is finite-dimensional over $``$ , so it follows from the double-coset lemma, Theorem 4.1, that $`|\mathrm{\Gamma }\backslash G/A|<\mathrm{}`$, where
$$A=a_1\times \mathrm{}\times a_n.$$
We now split the proof into two cases.
Case 1. Suppose that $`a_iL_i`$ for each $`i`$.
Since $`L_i`$ is normal in $`G_i`$, it follows that $`gAg^1\mathrm{\Gamma }`$ for all $`gG`$, so that $`\mathrm{\Gamma }gA=\mathrm{\Gamma }g`$, so $`|G:\mathrm{\Gamma }|=|\mathrm{\Gamma }\backslash G/A|<\mathrm{}`$, and the result follows.
Case 2. Suppose that (possibly after renumbering) $`a_1L_1`$.
Then, by our choice of $`a_1`$, there is a free surface group $`F=\pi _1\mathrm{\Sigma }_1G_1`$ with $`a_1F`$ and $`FL_1=\{1\}`$. Define $`B=G_1\times a_2\times \mathrm{}\times a_n`$. Then $`AB`$ and
$$\left|\mathrm{\Gamma }\backslash G/A\right|<\mathrm{},$$
so Corollary 4.3 gives
$$\left|(\mathrm{\Gamma }B)\backslash B/A\right|<\mathrm{}.$$
Hence Corollary 4.5 gives
$$\left|p_1(\mathrm{\Gamma }B)\backslash G_1/a_1\right|<\mathrm{},$$
since $`p_1(B)=G_1`$ and $`p_1(A)=a_1`$. But $`a_1F`$, so by Corollary 4.3 again we have
$$\left|[Fp_1(\mathrm{\Gamma }B)]\backslash F/a_1\right|<\mathrm{}.$$
Now $`G_1`$ is normal in $`B`$ with $`B/G_1`$ abelian. Hence $`L_1=\mathrm{\Gamma }G_1`$ is normal in $`\mathrm{\Gamma }B`$ with $`(\mathrm{\Gamma }B)/L_1`$ abelian. Hence $`L_1=p_1(L_1)`$ is normal in $`p_1(\mathrm{\Gamma }B)`$ with $`p_1(\mathrm{\Gamma }B)/L_1`$ abelian. Finally, it follows that $`FL_1`$ is normal in $`Fp_1(\mathrm{\Gamma }B)`$, with $`[Fp_1(\mathrm{\Gamma }B)]/(FL_1)`$ abelian.
But $`FL_1=\{1\}`$. Hence $`Fp_1(\mathrm{\Gamma }B)`$ is an abelian subgroup of the free group $`F`$, and so cyclic: say $`Fp_1(\mathrm{\Gamma }B)=b`$. Then $`F`$ is a non-abelian free group, and $`a_1,bF`$ such that $`|b\backslash F/a_1|<\mathrm{}`$, which is absurd.
This contradiction completes the proof.
$`\mathrm{}`$
Authorsโ addresses
| Martin R. Bridson | | James Howie |
| --- | --- | --- |
| Department of Mathematics | | Department of Mathematics |
| Imperial College London | | Heriot-Watt University |
| London SW7 2AZ | | Edinburgh EH14 4AS |
| m.bridson@imperial.ac.uk | | J.Howie@hw.ac.uk |
|
warning/0506/hep-ph0506227.html
|
ar5iv
|
text
|
# Conditions for CP-Violation in the General Two-Higgs-Doublet Model
## I Introduction
The Standard Model (SM) posits the existence of a single complex hypercharge-one Higgs doublet hhg . Due to the form of the Higgs potential, one component of this Higgs scalar acquires a vacuum expectation value and the SU(2)$`\times `$U(1) electroweak symmetry is spontaneously broken to U(1)<sub>EM</sub>. Hermiticity requires that the parameters of the SM Higgs potential are real. Consequently, the resulting bosonic sector of the electroweak theory is CP-conserving. CP-violation enters through the Yukawa couplings of the Higgs field to fermions. Although there are many potentially complex parameters in the Higgs couplings to three generations of quarks and leptons, one can redefine the fermion fields (to absorb unphysical phases). The end result is one CP-violating parameterโthe Cabibbo-Kobayashi-Maskawa angle ckm .
There are a number of motivations for considering extended Higgs sectors. For example, the minimal supersymmetric extension of the Standard Model requires two complex Higgs doublets susyhiggs . In this paper, we consider the most general two-Higgs-doublet extension of the Standard Model. This model possesses two identical complex, hypercharge-one Higgs doublets. In contrast to the Standard Model, the scalar Higgs potential of the two-Higgs-doublet model (2HDM) contains potentially complex parameters branco . Consequently, the purely bosonic sector can exhibit explicit CP-violation (prior to the introduction of the fermions and the attendant complex Higgs-fermion Yukawa couplings). However as above, not all complex phases are physical. In this paper, we exhibit the necessary and sufficient conditions for an explicitly CP-conserving 2HDM scalar potential.
The procedure for determining whether the Higgs potential explicitly violates CP is in principle straightforward. The Higgs potential parameters are initially defined with respect to two identical Higgs fields $`\mathrm{\Phi }_1`$ and $`\mathrm{\Phi }_2`$. However, one can always choose to change the basis (in the two-dimensional Higgs โflavorโ space) by defining two new (orthonormal) linear combinations of $`\mathrm{\Phi }_1`$ and $`\mathrm{\Phi }_2`$. In this new basis, all the Higgs potential parameters are modified. The Higgs potential is explicitly CP-violating if and only if no choice of basis exists in which all the Higgs potential parameters are simultaneously real.<sup>1</sup><sup>1</sup>1We find it convenient and illuminating to give an explicit proof of this oft-stated result in Appendix A. If (at least) one basis choice exists in which all Higgs potential parameters are real, then the Higgs potential is explicitly CP-conserving. Henceforth, we designate any such basis as a real basis. CP-violation in the scalar sector might still arise if the scalar field vacuum is not time-reversal invariant. In this case, CP is spontaneously broken Lee:1973iz .
Given an arbitrary Higgs potential, it may not be possible to determine by inspection whether a real basis exists. Since there exist four potentially complex parameters in the Higgs potential, one must in general solve a set of four non-linear equations (requiring that these four parameters are real in some specific basis to be determined). Thus, we propose another technique for answering the question of whether a special basis exists in which all Higgs potential parameters are real. Our procedure makes use of the technology introduced in ref. davidson based on invariant combinations of Higgs potential parameters. By definition, these invariants are basis-independent quantities; i.e., they do not depend on the initial basis choice for $`\mathrm{\Phi }_1`$ and $`\mathrm{\Phi }_2`$. We then search for potentially complex invariants.
Four potentially complex (basis-independent) invariants govern the CP-property of the 2HDM scalar potential. If any one of these four invariants possesses a non-zero imaginary part, then the 2HDM scalar potential is explicitly CP-violating. CP is explicitly conserved if and only if all four invariants are real. In the latter case, a real basis must exist (even though an explicit form for the transformation that produces such a basis is not determined). Two of the invariants were found by diagrammatic techniques in ref. davidson . Recently, three of the four invariants were also employed in brs . Other earlier simple (basis-dependent) conditions proposed for the existence of explicit CP-violation in the Higgs potential Grzadkowski:1999ye ; ghdecoupling turn out to be sufficient but not necessary for an explicitly CP-conserving Higgs potential.
Finally, we note that in the discussion above, we have not addressed the question of the minimization of the Higgs potential. This determines the vacuum expectation values (vevs) of the two Higgs fields,<sup>2</sup><sup>2</sup>2We shall always assume that the Higgs potential parameters are chosen such that the scalar minimum of interest preserves U(1)<sub>EM</sub>. which are basis-dependent quantities. The two vevs can in general be complex, although one can absorb these complex phases by phase redefinitions of the individual scalar fields Ginzburg:2004vp . As shown in Appendix F, the Higgs sector is fully CP-conserving if and only if there exists a real basis in which the Higgs vacuum expectation values are simultaneously real. The latter can be established by examining three additional invariants (initially introduced in ref. lavoura ) that depend explicitly on the vevs.
In Section II, the basis-independent formalism for the 2HDM developed in ref. davidson is reviewed. In Section III, we exhibit a set of four independent potentially complex invariants constructed from the Higgs sector parameters. We then prove that the imaginary parts of these four invariants vanish if and only if the 2HDM scalar potential explicitly conserves the CP symmetry. The proof of this theorem relies on a number of important lemmas that are proved in Appendices C and D. The power of this theorem is demonstrated by exhibiting three simple 2HDM models with complex parameters that are CP-conserving. In Section IV we provide some insight into how the set of four complex invariants was discovered by surveying all potentially complex $`n`$th-order invariants for $`n6`$. The manifest reality of all invariants of order three or less is demonstrated explicitly in Appendix E. Thus, one must search for invariants of order $`n4`$ to find candidates that are potentially complex. From the results of our survey, we deduce a number of general features of the potentially complex invariants of arbitrary order. To determine whether an explicitly CP-conserving Higgs potential exhibits spontaneous CP-violation, one must additionally consider basis-independent quantities, initially introduced in ref. lavoura , that depend on the Higgs vevs. Finally, a brief discussion of future directions and concluding remarks are given in Section VI.
## II The Higgs potential of the Two Higgs doublet model
Consider the most general two-Higgs doublet extension of the Standard Model (2HDM) hhg ; Diaz:2002tp . Let $`\mathrm{\Phi }_1`$ and $`\mathrm{\Phi }_2`$ denote two complex $`Y=1`$, SU(2)$`_L`$ doublet scalar fields. The most general SU(2)$`{}_{L}{}^{}\times `$U(1)<sub>Y</sub> invariant scalar potential is given by (see, e.g., ref. Haber:1993an )
$`๐ฑ`$ $`=`$ $`m_{11}^2\mathrm{\Phi }_1^{}\mathrm{\Phi }_1+m_{22}^2\mathrm{\Phi }_2^{}\mathrm{\Phi }_2[m_{12}^2\mathrm{\Phi }_1^{}\mathrm{\Phi }_2+\mathrm{h}.\mathrm{c}.]`$ (1)
$`+\frac{1}{2}\lambda _1(\mathrm{\Phi }_1^{}\mathrm{\Phi }_1)^2+\frac{1}{2}\lambda _2(\mathrm{\Phi }_2^{}\mathrm{\Phi }_2)^2+\lambda _3(\mathrm{\Phi }_1^{}\mathrm{\Phi }_1)(\mathrm{\Phi }_2^{}\mathrm{\Phi }_2)+\lambda _4(\mathrm{\Phi }_1^{}\mathrm{\Phi }_2)(\mathrm{\Phi }_2^{}\mathrm{\Phi }_1)`$
$`+\{\frac{1}{2}\lambda _5(\mathrm{\Phi }_1^{}\mathrm{\Phi }_2)^2+[\lambda _6(\mathrm{\Phi }_1^{}\mathrm{\Phi }_1)+\lambda _7(\mathrm{\Phi }_2^{}\mathrm{\Phi }_2)]\mathrm{\Phi }_1^{}\mathrm{\Phi }_2+\mathrm{h}.\mathrm{c}.\},`$
where $`m_{11}^2`$, $`m_{22}^2`$, and $`\lambda _1,\mathrm{},\lambda _4`$ are real parameters and $`m_{12}^2`$, $`\lambda _5`$, $`\lambda _6`$ and $`\lambda _7`$ are potentially complex parameters. We assume that the parameters of the scalar potential are chosen such that the minimum of the scalar potential respects the U(1)$`_{\mathrm{EM}}`$ gauge symmetry. Then, the scalar field vacuum expectations values are of the form
$$\mathrm{\Phi }_1=\frac{1}{\sqrt{2}}\left(\begin{array}{c}0\\ v_1\end{array}\right),\mathrm{\Phi }_2=\frac{1}{\sqrt{2}}\left(\begin{array}{c}0\\ v_2e^{i\xi }\end{array}\right),$$
(2)
where $`v_1`$ and $`v_2`$ are real and non-negative, $`0|\xi |\pi `$, and
$$v^2v_1^2+v_2^2=\frac{4m_W^2}{g^2}=(246\mathrm{GeV})^2.$$
(3)
In writing eq. (2), we have used a global U(1)<sub>Y</sub> hypercharge rotation to eliminate the phase of $`v_1`$.
Since the scalar doublets $`\mathrm{\Phi }_1`$ and $`\mathrm{\Phi }_2`$ have identical SU(2)$`\times `$U(1) quantum numbers, one is free to define two orthonormal linear combinations of the original scalar fields. The parameters appearing in eq. (1) depend on a particular basis choice of the two scalar fields. Relative to an initial (generic) basis choice, the scalar fields in the new basis are given by $`\mathrm{\Phi }^{}=U\mathrm{\Phi }`$ davidson , where $`U`$ is a U(2) matrix:<sup>3</sup><sup>3</sup>3This U(2) transformation has also been recently exploited in ref. Ginzburg:2004vp .
$$U=e^{i\psi }\left(\begin{array}{cc}\mathrm{cos}\theta & e^{i\xi }\mathrm{sin}\theta \\ e^{i\chi }\mathrm{sin}\theta & e^{i(\chi \xi )}\mathrm{cos}\theta \end{array}\right).$$
(4)
Note that the phase $`\psi `$ has no effect on the scalar potential parameters, since this corresponds to a global hypercharge rotation.
With respect to the new $`\mathrm{\Phi }^{}`$-basis, the scalar potential takes on the same form given in eq. (1) but with new coefficients $`m_{ij}^{\mathrm{\hspace{0.17em}2}}`$ and $`\lambda _j^{}`$. For the general U(2) transformation of eq. (4) with $`\mathrm{\Phi }^{}=U\mathrm{\Phi }`$, the scalar potential parameters ($`m_{ij}^{\mathrm{\hspace{0.17em}2}}`$, $`\lambda _i^{}`$) are related to the original parameters ($`m_{ij}^2`$, $`\lambda _i`$) by:
$`m_{11}^{\mathrm{\hspace{0.17em}2}}`$ $`=`$ $`m_{11}^2c_\theta ^2+m_{22}^2s_\theta ^2\mathrm{Re}(m_{12}^2e^{i\xi })s_{2\theta },`$ (5)
$`m_{22}^{\mathrm{\hspace{0.17em}2}}`$ $`=`$ $`m_{11}^2s_\theta ^2+m_{22}^2c_\theta ^2+\mathrm{Re}(m_{12}^2e^{i\xi })s_{2\theta },`$ (6)
$`m_{12}^{\mathrm{\hspace{0.17em}2}}e^{i\chi }`$ $`=`$ $`\frac{1}{2}(m_{11}^2m_{22}^2)s_{2\theta }+\mathrm{Re}(m_{12}^2e^{i\xi })c_{2\theta }+i\mathrm{Im}(m_{12}^2e^{i\xi }).`$ (7)
and
$`\lambda _1^{}`$ $`=`$ $`\lambda _1c_\theta ^4+\lambda _2s_\theta ^4+\frac{1}{2}\lambda _{345}s_{2\theta }^2+2s_{2\theta }\left[c_\theta ^2\mathrm{Re}(\lambda _6e^{i\xi })+s_\theta ^2\mathrm{Re}(\lambda _7e^{i\xi })\right],`$ (8)
$`\lambda _2^{}`$ $`=`$ $`\lambda _1s_\theta ^4+\lambda _2c_\theta ^4+\frac{1}{2}\lambda _{345}s_{2\theta }^22s_{2\theta }\left[s_\theta ^2\mathrm{Re}(\lambda _6e^{i\xi })+c_\theta ^2\mathrm{Re}(\lambda _7e^{i\xi })\right],`$ (9)
$`\lambda _3^{}`$ $`=`$ $`\frac{1}{4}s_{2\theta }^2\left[\lambda _1+\lambda _22\lambda _{345}\right]+\lambda _3s_{2\theta }c_{2\theta }\mathrm{Re}[(\lambda _6\lambda _7)e^{i\xi }],`$ (10)
$`\lambda _4^{}`$ $`=`$ $`\frac{1}{4}s_{2\theta }^2\left[\lambda _1+\lambda _22\lambda _{345}\right]+\lambda _4s_{2\theta }c_{2\theta }\mathrm{Re}[(\lambda _6\lambda _7)e^{i\xi }],`$ (11)
$`\lambda _5^{}e^{2i\chi }`$ $`=`$ $`\frac{1}{4}s_{2\theta }^2\left[\lambda _1+\lambda _22\lambda _{345}\right]+\mathrm{Re}(\lambda _5e^{2i\xi })+ic_{2\theta }\mathrm{Im}(\lambda _5e^{2i\xi })s_{2\theta }c_{2\theta }\mathrm{Re}[(\lambda _6\lambda _7)e^{i\xi }]`$ (12)
$`is_{2\theta }\mathrm{Im}[(\lambda _6\lambda _7)e^{i\xi })],`$
$`\lambda _6^{}e^{i\chi }`$ $`=`$ $`\frac{1}{2}s_{2\theta }\left[\lambda _1c_\theta ^2\lambda _2s_\theta ^2\lambda _{345}c_{2\theta }i\mathrm{Im}(\lambda _5e^{2i\xi })\right]+c_\theta c_{3\theta }\mathrm{Re}(\lambda _6e^{i\xi })+s_\theta s_{3\theta }\mathrm{Re}(\lambda _7e^{i\xi })`$ (13)
$`+ic_\theta ^2\mathrm{Im}(\lambda _6e^{i\xi })+is_\theta ^2\mathrm{Im}(\lambda _7e^{i\xi }),`$
$`\lambda _7^{}e^{i\chi }`$ $`=`$ $`\frac{1}{2}s_{2\theta }\left[\lambda _1s_\theta ^2\lambda _2c_\theta ^2+\lambda _{345}c_{2\theta }+i\mathrm{Im}(\lambda _5e^{2i\xi })\right]+s_\theta s_{3\theta }\mathrm{Re}(\lambda _6e^{i\xi })+c_\theta c_{3\theta }\mathrm{Re}(\lambda _7e^{i\xi })`$ (14)
$`+is_\theta ^2\mathrm{Im}(\lambda _6e^{i\xi })+ic_\theta ^2\mathrm{Im}(\lambda _7e^{i\xi }),`$
where
$$\lambda _{345}\lambda _3+\lambda _4+\mathrm{Re}(\lambda _5e^{2i\xi }).$$
(15)
These equations exhibit the following features. If $`m_{11}^2=m_{22}^2`$ and $`m_{12}^2=0`$ in some basis then these two conditions are true in all bases. Likewise, if $`\lambda _1=\lambda _2`$ and $`\lambda _7=\lambda _6`$ in some basis then these latter two conditions are true in all bases.
We noted previously that the parameters $`m_{12}^2`$, $`\lambda _5`$, $`\lambda _6`$ and $`\lambda _7`$ are potentially complex. We now pose the following question: does there exist a so-called real basis in which all the scalar potential parameters are real? In general, the existence of a real basis cannot be ascertained by inspection. In particular, starting from an arbitrary basis, it may be quite difficult to determine whether or not there is a choice of $`\theta ,\chi ,\xi `$ above such that all the primed parameters are real. However, in this paper we will show, using the basis-independent techniques described in ref. davidson , that there is a straightforward procedure for determining whether a real basis exists. To accomplish this goal, we write the scalar Higgs potential of the 2HDM following refs. branco and davidson :
$$๐ฑ=Y_{a\overline{b}}\mathrm{\Phi }_{\overline{a}}^{}\mathrm{\Phi }_b+\frac{1}{2}Z_{a\overline{b}c\overline{d}}(\mathrm{\Phi }_{\overline{a}}^{}\mathrm{\Phi }_b)(\mathrm{\Phi }_{\overline{c}}^{}\mathrm{\Phi }_d),$$
(16)
where the indices $`a`$, $`\overline{b}`$, $`c`$ and $`\overline{d}`$ run over the two-dimensional Higgs flavor space and
$$Z_{a\overline{b}c\overline{d}}=Z_{c\overline{d}a\overline{b}}.$$
(17)
Hermiticity of $`๐ฑ`$ implies that
$$Y_{a\overline{b}}=(Y_{b\overline{a}})^{},Z_{a\overline{b}c\overline{d}}=(Z_{b\overline{a}d\overline{c}})^{}.$$
(18)
Under a global U(2) transformation, $`\mathrm{\Phi }_aU_{a\overline{b}}\mathrm{\Phi }_b`$ (and $`\mathrm{\Phi }_{\overline{a}}^{}\mathrm{\Phi }_{\overline{b}}^{}U_{b\overline{a}}^{}`$), where $`U_{b\overline{a}}^{}U_{a\overline{c}}=\delta _{b\overline{c}}`$, and the tensors $`Y`$ and $`Z`$ transform covariantly: $`Y_{a\overline{b}}U_{a\overline{c}}Y_{c\overline{d}}U_{d\overline{b}}^{}`$ and $`Z_{a\overline{b}c\overline{d}}U_{a\overline{e}}U_{f\overline{b}}^{}U_{c\overline{g}}U_{h\overline{d}}^{}Z_{e\overline{f}g\overline{h}}`$. The use of barred indices is convenient for keeping track of which indices transform with $`U`$ and which transform with $`U^{}`$. We also introduce the U(2)-invariant tensor $`\delta _{a\overline{b}}`$, which can be used to contract indices. In this notation, one can only contract an unbarred index against a barred index. For example,
$$Z_{a\overline{d}}^{(1)}\delta _{b\overline{c}}Z_{a\overline{b}c\overline{d}}=Z_{a\overline{b}b\overline{d}},Z_{c\overline{d}}^{(2)}\delta _{b\overline{a}}Z_{a\overline{b}c\overline{d}}=Z_{a\overline{a}c\overline{d}}.$$
(19)
With respect to the $`\mathrm{\Phi }`$-basis of the unprimed scalar fields, we have:
$`Y_{11}=m_{11}^2,Y_{12}=m_{12}^2,`$
$`Y_{21}=(m_{12}^2)^{},Y_{22}=m_{22}^2,`$ (20)
and
$`Z_{1111}=\lambda _1,Z_{2222}=\lambda _2,`$
$`Z_{1122}=Z_{2211}=\lambda _3,Z_{1221}=Z_{2112}=\lambda _4,`$
$`Z_{1212}=\lambda _5,Z_{2121}=\lambda _5^{},`$
$`Z_{1112}=Z_{1211}=\lambda _6,Z_{1121}=Z_{2111}=\lambda _6^{},`$
$`Z_{2212}=Z_{1222}=\lambda _7,Z_{2221}=Z_{2122}=\lambda _7^{}.`$ (21)
For ease of notation, we have omitted the bars from the barred indices in eqs. (II) and (II). Since the tensors $`Y_{a\overline{b}}`$ and $`Z_{a\overline{b}c\overline{d}}`$ exhibit tensorial properties with respect to global U(2) rotations in the Higgs flavor space, one can easily construct invariants with respect to the U(2) by forming U(2)-scalar quantities.
In section III, we shall argue that the scalar potential is CP-conserving if and only if a real basis exists. In this case, all possible U(2)-invariant scalars are manifestly real. Conversely, if the scalar potential explicitly violates CP, then there must exist at least one manifestly complex U(2)-scalar invariant. We shall exhibit the simplest set of independent potentially complex U(2)-scalar invariants that can be employed to test for explicit CP-invariance or non-invariance of the 2HDM scalar potential.
## III Complex invariants and the conditions for a CP-conserving 2HDM scalar potential
Given an arbitrary 2HDM Higgs potential, we have already noted that the scalar potential possesses a number of potentially complex parameters. We would like to determine in general whether this scalar potential is explicitly CP-violating or CP-conserving. The answer to this question is governed by a simple theorem:
Theorem 1: The Higgs potential is explicitly CP-conserving if and only if a basis exists in which all Higgs potential parameters are real. Otherwise, CP is explicitly violated.
Although Theorem 1 is well-known and often stated in the literature, its proof is usually given under the assumption that a convenient basis has been chosen in which the CP transformation laws of the scalar fields assume a particularly simple form branco . In Appendix A, we provide a general proof of Theorem 1 that does not make any assumption about the initial choice of scalar field basis. As already noted, it may be difficult to determine whether a basis exists in which all Higgs potential parameters are real. Thus, we would like to reformulate Theorem 1 in a basis-independent language. That is, we propose to express the conditions for an explicitly CP-violating (or conserving) Higgs potential in terms of basis-independent invariants.
Before presenting the basis-independent version of Theorem 1, let us first enumerate the number of independent CP-violating phases that exist among the scalar potential parameters of the 2HDM. In eq. (1), we have noted four potentially complex parameters: $`Y_{12}m_{12}^2`$, $`\lambda _5`$, $`\lambda _6`$ and $`\lambda _7`$. Naively, it appears that there are three independent CP-violating phases, since one can always perform a phase rotation on one of the Higgs fields to render one of the complex parameters real. However, this conclusion is not correct, since one can utilize a larger SU(2) global symmetry to absorb additional phases.<sup>4</sup><sup>4</sup>4As previously noted, a U(1) hypercharge global rotation leaves all the scalar parameters unchanged; that is, the angle $`\psi `$ in eq. (4) has no effect. If one chooses $`\psi =\frac{1}{2}(\xi \chi )`$, then the matrix $`U`$ given in eq. (4) is an SU(2) matrix. An SU(2) global rotation is parameterized by one angle and two phases. This can be used to remove one real parameter and two phases from the initial ten real parameters and four phases that make up the scalar potential parameters. Thus, ultimately, the number of physical parameters of the scalar potential must be given by nine real parameters and two phases. Equivalently, there can only be two independent complex parameters among the physical parameters that describe the scalar potential.
This result can be derived in a very simple and direct fashion as follows davidson . Consider the explicit forms of $`Z^{(1)}`$ and $`Z^{(2)}`$ defined in eq. (19):
$$Z^{(1)}=\left(\begin{array}{cc}\lambda _1+\lambda _4& \lambda _6+\lambda _7\\ \lambda _6^{}+\lambda _7^{}& \lambda _2+\lambda _4\end{array}\right),Z^{(2)}=\left(\begin{array}{cc}\lambda _1+\lambda _3& \lambda _6+\lambda _7\\ \lambda _6^{}+\lambda _7^{}& \lambda _2+\lambda _3\end{array}\right).$$
(22)
Note that $`Z^{(1)}`$ and $`Z^{(2)}`$ are hermitian matrices that commute so that they can be simultaneously diagonalized by a unitary matrix. It therefore follows that there exists a basis in which $`Z^{(1)}`$ and $`Z^{(2)}`$ are simultaneously diagonal; that is, $`\lambda _7=\lambda _6`$. Once this basis is established, it is clear that the phase of $`\lambda _6`$ and $`\lambda _7`$ can be removed by a U(1) phase rotation of $`\mathrm{\Phi }_2`$. Thus, a basis can always be found in which only two parameters $`Y_{12}`$ and $`\lambda _5`$ are complex. Moreover, the total number of independent real parameters is nine (since in a basis where $`\lambda _7=\lambda _6`$, only one of these two parameters is an independent degree of freedom). This matches the counting of parameters given in the previous paragraph.
Based on this parameter counting, one is tempted to conclude that there should be only two independent potentially complex invariants. Nevertheless, this intuition is misleading. The correct statement is summarized by the following theorem.
Theorem 2: The necessary and sufficient conditions for an explicitly CP-conserving 2HDM scalar potential consist of the (simultaneous) vanishing of the imaginary parts of four potentially complex invariants:
$`I_{Y3Z}`$ $``$ $`\mathrm{Im}(Z_{a\overline{c}}^{(1)}Z_{e\overline{b}}^{(1)}Z_{b\overline{e}c\overline{d}}Y_{d\overline{a}}),`$ (23)
$`I_{2Y2Z}`$ $``$ $`\mathrm{Im}(Y_{a\overline{b}}Y_{c\overline{d}}Z_{b\overline{a}d\overline{f}}Z_{f\overline{c}}^{(1)}),`$ (24)
$`I_{6Z}`$ $``$ $`\mathrm{Im}(Z_{a\overline{b}c\overline{d}}Z_{b\overline{f}}^{(1)}Z_{d\overline{h}}^{(1)}Z_{f\overline{a}j\overline{k}}Z_{k\overline{j}m\overline{n}}Z_{n\overline{m}h\overline{c}}),`$ (25)
$`I_{3Y3Z}`$ $``$ $`\mathrm{Im}(Z_{a\overline{c}b\overline{d}}Z_{c\overline{e}d\overline{g}}Z_{e\overline{h}f\overline{q}}Y_{g\overline{a}}Y_{h\overline{b}}Y_{q\overline{f}}).`$ (26)
Henceforth, the imaginary parts of potentially complex invariants shall be referred to as $`I`$-invariants.
The case of $`\lambda _1=\lambda _2`$ and $`\lambda _7=\lambda _6`$ is a special isolated point in the scalar potential parameter space. In particular, when $`\lambda _1=\lambda _2`$ and $`\lambda _7=\lambda _6`$, the matrices $`Z^{(1)}`$ and $`Z^{(2)}`$ are both proportional to the unit matrix. Thus, if both equalities $`\lambda _1=\lambda _2`$ and $`\lambda _7=\lambda _6`$ are true in one basis, then they must also be true in all bases \[as previously noted below eq. (8)\]. Thus, Theorem 2 breaks up into two distinct cases:
(i) For the isolated point $`\lambda _1=\lambda _2`$ and $`\lambda _7=\lambda _6`$, $`I_{Y3Z}=I_{2Y2Z}=I_{6Z}=0`$ is automatic \[see eqs. (28)โ(30)\]. In this case, the necessary and sufficient condition for an explicitly CP-conserving 2HDM scalar potential is simply given by $`I_{3Y3Z}=0`$.
(ii) Away from the special isolated point of case (i), only three of the $`I`$-invariants need be considered. Specifically, at any other point of the parameter space, the necessary and sufficient conditions for an explicitly CP-conserving 2HDM scalar potential are given by<sup>5</sup><sup>5</sup>5If eq. (27) is satisfied in case (ii), then it follows that $`I_{3Y3Z}=0`$. Thus, the latter is not needed as a separate requirement.
$$I_{Y3Z}=I_{2Y2Z}=I_{6Z}=0.$$
(27)
It is trivial to prove that the above conditions are necessary for explicit CP-conservation. If any of the above $`I`$-invariants \[eqs. (23)โ(26)\] are non-zero, then we can immediately conclude that no basis exists in which all scalar potential parameters are real. Thus, by Theorem 1, the scalar potential would be CP-violating. The proof that the conditions of Theorem 2 are sufficient for explicit CP-conservation will now be given, with further details provided in Appendices C and D.
First, we must prove that all four $`I`$-invariants listed in eqs. (23)โ(26) are required in the formulation of Theorem 2. This may be accomplished by exhibiting four different models in which only one of the four $`I`$-invariants is non-zero. In Sections IV.A and IV.C, we give explicit forms for these four $`I`$-invariants in a generic basis \[see eqs. (39), (41), (47) and (48), respectively\]. However, as already noted below eq. (22), it is always possible to choose a basis in which $`\lambda _7=\lambda _6`$. This basis is not unique, since further basis transformations can be performed while maintaining $`\lambda _7=\lambda _6`$. In any such basis, three of the $`I`$-invariants take particularly simple forms:
$`I_{Y3Z}`$ $`=`$ $`(\lambda _1\lambda _2)^2\mathrm{Im}(Y_{12}\lambda _6^{}),`$ (28)
$`I_{2Y2Z}`$ $`=`$ $`(\lambda _1\lambda _2)\left[\mathrm{Im}(Y_{12}^2\lambda _5^{})+(Y_{11}Y_{22})\mathrm{Im}(Y_{12}\lambda _6^{})\right],`$ (29)
$`I_{6Z}`$ $`=`$ $`(\lambda _1\lambda _2)^3\mathrm{Im}(\lambda _6^2\lambda _5^{}).`$ (30)
The expression for $`I_{3Y3Z}`$ in this basis is more complicated:
$`I_{3Y3Z}=2\mathrm{Im}(Y_{12}^3\lambda _6(\lambda _5^{})^2)4\mathrm{Im}(Y_{12}^3(\lambda _6^{})^3)+[(Y_{11}Y_{22})^26|Y_{12}|^2](Y_{11}Y_{22})\mathrm{Im}(\lambda _6^2\lambda _5^{})`$
$`+\left[(\lambda _1\lambda _3\lambda _4)(\lambda _2\lambda _3\lambda _4)+2|\lambda _6|^2|\lambda _5|^2\right](Y_{11}Y_{22})\mathrm{Im}(Y_{12}^2\lambda _5^{})`$
$`+\left\{(\lambda _1\lambda _2)^2Y_{11}Y_{22}+(4|\lambda _6|^22|\lambda _5|^2)\left[(Y_{11}Y_{22})^2|Y_{12}|^2\right]\right\}\mathrm{Im}(Y_{12}\lambda _6^{})`$
$`(\lambda _1+\lambda _22\lambda _32\lambda _4)\{(Y_{11}Y_{22})\mathrm{Im}(Y_{12}^2(\lambda _6^{})^2)\mathrm{Im}(Y_{12}^3\lambda _5^{}\lambda _6^{})`$
$`+[(Y_{11}Y_{22})^2|Y_{12}|^2]\mathrm{Im}(Y_{12}\lambda _6\lambda _5^{})\}.`$ (31)
Working in the $`\lambda _7=\lambda _6`$ basis, we consider the four models:
1. $`Y_{a\overline{b}}=0`$ and $`\lambda _1\lambda _2`$ ;
2. $`\lambda _6=0`$ , $`\lambda _1\lambda _2`$ and $`Y_{11}=Y_{22}`$ ;
3. $`\lambda _5=0`$ , $`\lambda _1\lambda _2`$ , $`Y_{11}=Y_{22}=0`$ and Re($`Y_{12}\lambda _6^{})=0`$ ;
4. $`\lambda _1=\lambda _2`$ .
Then, in model 1, $`I_{Y3Z}=I_{2Y2Z}=I_{3Y3Z}=0`$ whereas $`I_{6Z}`$ is potentially non-zero. In model 2, $`I_{Y3Z}=I_{6Z}=I_{3Y3Z}=0`$ whereas $`I_{2Y2Z}`$ is potentially non-zero. In model 3, $`I_{2Y2Z}=I_{6Z}=I_{3Y3Z}=0`$ whereas $`I_{Y3Z}`$ is potentially non-zero.<sup>6</sup><sup>6</sup>6Note that if $`\lambda _5=0`$ and $`Y_{11}=Y_{22}`$, then $`I_{3Y3Z}=\left\{(\lambda _1\lambda _2)^2Y_{11}Y_{22}16[\mathrm{Re}(Y_{12}\lambda _6^{})]^2\right\}\mathrm{Im}(Y_{12}\lambda _6^{})`$. Finally, in model 4, $`I_{Y3Z}=I_{6Z}=I_{2Y2Z}=0`$ whereas $`I_{3Y3Z}`$ is potentially non-zero. Thus, we have exhibited four separate models in which CP is violated explicitly, and in each case only one of the four $`I`$-invariants is non-zero. This illustrates that all four $`I`$-invariants are needed to test whether the Higgs potential explicitly conserves or violates CP.
The requirement of four $`I`$-invariants in the formulation of Theorem 2 seems to be in conflict with our previous observation that the number of physical parameters of the 2HDM includes only two phases \[see discussion surrounding eq. (22)\]. However, one can show that for any particular model, at most two $`I`$-invariants need be considered. To verify this assertion, we first transform to a basis in which $`\lambda _7=\lambda _6`$ and where $`\lambda _6`$ (and therefore $`\lambda _7`$) are real.<sup>7</sup><sup>7</sup>7This is always possible as shown below eq. (22). Then there are a number of cases to consider. (i) If $`\lambda _1=\lambda _2`$, then $`I_{3Y3Z}=0`$ implies that the Higgs sector is explicitly CP-conserving. (ii) If $`\lambda _1\lambda _2`$ and $`Y_{12}`$, $`\lambda _5`$ and $`\lambda _6`$ are non-vanishing, then $`I_{Y3Z}=I_{6Z}=0`$ implies that the Higgs sector is explicitly CP-conserving. (iii) If $`\lambda _1\lambda _2`$ and two of the quantities $`Y_{12}`$, $`\lambda _5`$ and $`\lambda _6`$ are non-zero while the third vanishes, then only one $`I`$-invariant need be considered. Specifically, for $`\lambda _5=0`$ \[$`\lambda _6=0`$\], $`I_{Y3Z}=0`$ \[$`I_{2Y2Z}=0`$\] guarantees a CP-conserving Higgs sector, whereas for $`Y_{12}=0`$, $`I_{6Z}=0`$ guarantees a CP-conserving Higgs sector. Thus, we have shown that it is sufficient to examine at most two $`I`$-invariants to determine whether all four $`I`$-invariants \[eqs. (23)โ(26)\] simultaneously vanish.<sup>8</sup><sup>8</sup>8Of course, to take advantage of this observation in practice, one must be able to take the original model and transform to a basis where $`\lambda _7=\lambda _6`$ is real. In general, this may be difficult (and require a numerical computation). Thus, in order to test for explicit CP-violation, it is often simpler to directly evaluate all four $`I`$-invariants in the original basis.
To complete the proof of Theorem 2, we must show that if the four $`I`$-invariants given by eqs. (23)โ(26) vanish, then one can find a basis where all Higgs potential parameters are real. The proof is most easily carried out by first transforming to a basis in which $`\lambda _7=\lambda _6`$ and where $`\lambda _6`$ (and therefore $`\lambda _7`$) are real. In this basis, the cases of $`\lambda _1=\lambda _2`$ and $`\lambda _1\lambda _2`$ must be treated separately. First, we consider the case where $`\lambda _1\lambda _2`$. If $`\lambda _7=\lambda _60`$, then $`I_{6Z}=0`$ \[eq. (30)\] implies that $`\lambda _5`$ is also real in this basis, and $`I_{Y3Z}=0`$ \[eq. (28)\] implies that $`Y_{12}`$ is real. We have therefore achieved a basis in which all scalar potential parameters are real. If $`\lambda _6=\lambda _7=0`$, then one can perform a phase rotation on one of the scalar fields so that $`\lambda _5`$ is real, with $`Y_{12}`$ potentially complex. In this new basis, if $`\lambda _50`$ then $`I_{2Y2Z}=0`$ implies that $`Y_{12}`$ is either real or purely imaginary. In the latter case, eqs. (12)โ(14) show that a U(2) transformation \[see eq. (4)\] with parameters $`\xi =\pi /2`$, $`\mathrm{sin}2\theta =0`$ and $`\chi =0`$ yields a basis in which $`\lambda _6^{}=\lambda _7^{}=0`$, and both $`\lambda _5^{}=\lambda _5`$ and $`Y_{12}^{}`$ are real. Finally, if $`\lambda _5=\lambda _6=\lambda _7=0`$, then one can absorb any phase of $`Y_{12}`$ into a phase redefinition of one of the scalar fields.
Next, we consider the case where $`\lambda _1=\lambda _2`$ in a basis where $`\lambda _7=\lambda _6`$. In this case, it is always possible to make a further change of basis so that $`\lambda _5`$, $`\lambda _6`$ and $`\lambda _7`$ are real (this assertion is Lemma 2, which is proved in Appendix C).<sup>9</sup><sup>9</sup>9In Appendix C, Lemma 3 demonstrates why the condition of $`\lambda _1=\lambda _2`$ is crucial to the proof of Lemma 2. In this latter basis where $`Y_{12}`$ is potentially complex but all other scalar potential parameters are real, eq. (31) yields the following form for the only potentially non-vanishing invariant $`I_{3Y3Z}`$:
$`I_{3Y3Z}=2\mathrm{Im}Y_{12}\left[\lambda _5^2+\lambda _5(\lambda _1\lambda _3\lambda _4)2\lambda _6^2\right]`$
$`\times \left[4\lambda _6(\mathrm{Re}Y_{12})^2(\lambda _3+\lambda _4+\lambda _5\lambda _1)(Y_{11}Y_{22})\mathrm{Re}Y_{12}\lambda _6(Y_{11}Y_{22})^2\right].`$ (32)
Then, $`I_{3Y3Z}=0`$ implies that one of the following three conditions must be true in a basis where all the $`\lambda _i`$ are real: (i) $`Y_{12}`$ is real; (ii) the quantity $`\lambda _5^2+\lambda _5(\lambda _1\lambda _3\lambda _4)2\lambda _6^2=0`$; or (iii) the quantity $`4\lambda _6(\mathrm{Re}Y_{12})^2(\lambda _3+\lambda _4+\lambda _5\lambda _1)(Y_{11}Y_{22})\mathrm{Re}Y_{12}\lambda _6(Y_{11}Y_{22})^2=0`$. In Appendix D, we prove Lemma 4 which demonstrates that if $`Y_{12}`$ is complex and either condition (ii) or condition (iii) holds, then it is possible to find a basis in which $`Y_{12}`$ is real, while maintaining the reality of $`\lambda _5`$, $`\lambda _6`$ and $`\lambda _7`$. Hence it follows that if $`I_{3Y3Z}=0`$, then there exists a basis in which all 2HDM scalar potential parameters are real.<sup>10</sup><sup>10</sup>10The U(2) rotation required to go to this basis is explicitly constructed in Appendix D. The proof of Theorem 2 is now complete.
It is instructive to compare the results of Theorem 2 to one of the basis-dependent conditions that has been proposed in the literature. In a generic basis, a sufficient set of conditions for an explicitly CP-conserving 2HDM scalar potential is:
$$\mathrm{Im}(Y_{12}^2\lambda _5^{})=\mathrm{Im}(Y_{12}\lambda _6^{})=\mathrm{Im}(Y_{12}\lambda _7^{})=\mathrm{Im}(\lambda _5^{}\lambda _6^2)=\mathrm{Im}(\lambda _5^{}\lambda _7^2)=\mathrm{Im}(\lambda _6^{}\lambda _7)=0,$$
(33)
where $`Y_{12}m_{12}^2`$. Clearly, if eq. (33) is satisfied, then a simple phase rotation of one of the scalar fields easily produces a basis in which all the scalar potential parameters are real. However, eq. (33) is not necessary for CP-conservation. In particular, the following statement is generally false: โthe Higgs potential is explicitly CP-violating if one or more of the quantities listed in eq. (33) are non-vanishing.โ This is most easily demonstrated by the following exercise. Start with a model in which all potentially complex scalar potential parameters are real. Then, change the basis with a generic U(2) transformation \[eq. (4)\]. In a typical case, the resulting parameters $`Y_{12}^{}`$, $`\lambda _5^{}`$, $`\lambda _6^{}`$, and $`\lambda _7^{}`$ in the new basis are complex, and one or more of the quantities listed in eq. (33) are non-vanishing. Thus, eq. (33) is not a necessary condition for an explicitly CP-conserving Higgs potential.<sup>11</sup><sup>11</sup>11An example that illustrates the same point is a model in which $`\lambda _1=\lambda _2`$ and $`\lambda _7=\lambda _6`$. Lemma 2 of Appendix C implies that we can transform to a basis in which all the $`\lambda _i`$ are real. Nevertheless, in this basis, eq. (III) implies that it is possible to have an explicitly CP-conserving model with $`I_{3Y3Z}=0`$ and $`\mathrm{Im}Y_{12}0`$.
Despite the relative simplicity of the forms for $`I_{Y3Z}`$, $`I_{2Y2Z}`$, $`I_{6Z}`$ and $`I_{3Y3Z}`$ in the $`\lambda _7=\lambda _6`$ basis, realistic models rarely conform to this particular basis choice. The power of the basis-independent formulation of Theorem 2 thus becomes evident when considering models where the transformation from the generic basis to the $`\lambda _7=\lambda _6`$ basis is not particularly simple. Fortunately, we possess expressions for these $`I`$-invariants in a generic basis \[see eqs. (39), (41), (47) and (48)\], so there is no compelling need to explicitly perform this change of basis. For purposes of illustration, let us consider three special models. In model (i),
$$\lambda _1=\lambda _2,\lambda _6=\lambda _7\mathrm{and}Y_{11}=Y_{22},$$
(34)
where $`Y_{12}`$, $`\lambda _5`$ and $`\lambda _6`$ have arbitrary phases. In model (ii),
$$\lambda _1+\lambda _2=2(\lambda _3+\lambda _4),\lambda _5=0\mathrm{and}\lambda _6=\lambda _7,$$
(35)
where $`Y_{12}`$ and $`\lambda _6`$ have arbitrary phases. In model (iii),
$$\lambda _1=\lambda _2,\lambda _6=\lambda _7^{},Y_{11}=Y_{22}\mathrm{and}Y_{12},\lambda _5\mathrm{real},$$
(36)
where $`\lambda _6`$ has an arbitrary phase. Model (iii) arises by imposing on the Higgs potential a discrete permutation symmetry that interchanges $`\mathrm{\Phi }_1`$ and $`\mathrm{\Phi }_2`$ rebelo .
In the three models above, we have used eqs. (39), (41), (47) and (48) in the generic basis to verify that $`I_{Y3Z}=I_{2Y2Z}=I_{6Z}=I_{3Y3Z}=0`$. Thus models (i), (ii) and (iii) are explicitly CP-conserving. These three models also provide examples of explicitly CP-conserving 2HDM potentials where eq. (33) is not satisfied. Nevertheless, having verified that all the $`I`$-invariants vanish, one is assured of the existence of some basis choice for each model for which all Higgs potential parameters are real.
Here, we provide one explicit example in the case of model (iii). Starting from the generic basis specified in eq. (36), we perform a U(2) transformation \[eq. (4)\] with $`\theta =\pi /4`$ and $`\xi =0`$. Then, eqs. (7), (13) and (14) yield $`m_{12}^{\mathrm{\hspace{0.17em}2}}=\lambda _6^{}=\lambda _7^{}=0`$, while eq. (12) implies that:
$$\lambda _5^{}e^{2i\chi }=\frac{1}{2}(\lambda _1\lambda _3\lambda _4+\lambda _5)2i\mathrm{Im}\lambda _6.$$
(37)
It is now a simple matter to adjust $`\chi `$ so that $`\lambda _5^{}`$ is real. Thus, we have exhibited the U(2) transformation that produces the โreal basisโ of model (iii) in which all scalar potential parameters are real. Applying this U(2) transformation to the fields, it is easy to check that the resulting real basis exhibits a discrete symmetry $`\mathrm{\Phi }_1^{}\mathrm{\Phi }_1^{}`$, $`\mathrm{\Phi }_2^{}\mathrm{\Phi }_2^{}`$. Models that respect the latter discrete symmetry are manifestly CP-invariant since $`\lambda _5^{}`$ is the only potentially complex parameter, whose phase can be rotated away by an appropriate phase rotation of $`\mathrm{\Phi }_2^{}e^{i\chi }\mathrm{\Phi }_2^{}`$.
## IV A survey of complex invariants
In general, it is possible to construct an $`n`$th order invariant quantity for any integer value of $`n`$, where $`n`$ is the total number of $`Y`$โs and $`Z`$โs that appears in the invariant. The vast majority of such invariants are manifestly real. In this section, we focus on those invariants that are potentially complex.
The necessary and sufficient conditions for CP-conservation have been presented in Theorem 2 and depend on only four potentially complex invariants given by eqs. (23)โ(26). However, new potentially complex $`n`$th order invariants arise at every order (for $`n>4`$) that cannot be expressed in terms of lower-order invariants. Nevertheless, Theorem 2 guarantees that if the $`I`$-invariants of eqs. (23)โ(26) vanish, then the imaginary parts of all potentially complex invariants must vanish. In particular, we have explicitly verified the following statements:
1. All invariants (of arbitrary order) that are either independent of $`Z`$ or linear in $`Z`$ are manifestly real.
2. All invariants of cubic order or less are manifestly real.
3. Any quartic (i.e., fourth-order) $`I`$-invariant is a real linear combination of $`I_{Y3Z}`$ and $`I_{2Y2Z}`$.
4. Any fourth or higher-order $`I`$-invariant that is quadratic in $`Z`$ is proportional to $`I_{2Y2Z}`$.
5. Any fifth-order $`I`$-invariant vanishes if $`I_{Y3Z}=I_{2Y2Z}=0`$.
6. Any sixth-order $`I`$-invariant that is independent of $`Y`$ is proportional to $`I_{6Z}`$. Moreover, if $`Y_{a\overline{b}}=0`$ then any $`I`$-invariant of arbitrary order vanishes if $`I_{6Z}=0`$.
7. Any sixth order $`I`$-invariant that is both cubic in $`Y`$ and $`Z`$ respectively is a real linear combination of $`I_{3Y3Z}`$ and lower-order invariants that vanish if $`I_{Y3Z}=I_{2Y2Z}=0`$.
8. Any sixth order $`I`$-invariant that is either linear or quadratic in $`Y`$ vanishes if $`I_{Y3Z}=I_{2Y2Z}=0`$.
Finally, we reiterate that:
1. Any $`I`$-invariant of arbitrary order vanishes if $`I_{Y3Z}=I_{2Y2Z}=I_{6Z}=I_{3Y3Z}=0`$.
This last result is a consequence of Theorem 2. The explicit verification of statements 1โ8 is based on a systematic study of potentially complex U(2)-invariant scalars made up of the tensors $`Y_{a\overline{b}}`$ and $`Z_{a\overline{b}c\overline{d}}`$. This study gives us further confidence that the ultimate conclusion given by statement 9 above is correct.
We begin this study by noting that for $`n=1`$, the only invariants are $`\mathrm{Tr}Y`$, $`\mathrm{Tr}Z^{(1)}`$ and $`\mathrm{Tr}Z^{(2)}`$, all of which are manifestly real. For $`n=2`$, the possible quadratic invariants include the products of the first order invariants and $`\mathrm{Tr}(Y^2)`$, $`\mathrm{Tr}(YZ^{(1)})`$, $`\mathrm{Tr}(YZ^{(2)})`$, $`\mathrm{Tr}(Z^{(i)}Z^{(j)})`$ \[$`(i,j)=(1,1),(1,2),(2,2)`$\], $`\mathrm{Tr}Z^{(31)}Z_{a\overline{b}c\overline{d}}Z_{b\overline{a}d\overline{c}}`$ and $`\mathrm{Tr}Z^{(32)}Z_{a\overline{b}c\overline{d}}Z_{d\overline{a}b\overline{c}}`$, where $`Z^{(31)}`$ and $`Z^{(32)}`$ are introduced in eq. (115).<sup>12</sup><sup>12</sup>12Note that the determinants of $`Y`$, $`Z^{(1)}`$ and $`Z^{(2)}`$ are also quadratic invariants, but these can be expressed in terms of invariants already given above due to the identity $`detM\frac{1}{2}[(\mathrm{Tr}M)^2\mathrm{Tr}(M^2)]`$ which is satisfied by any $`2\times 2`$ matrix. By inspection, all such quadratic invariants are manifestly real. Turning to the cubic invariants, the enumeration of all possible cases becomes significantly more complex. Nevertheless, as shown in Appendix E, it is still possible to verify by hand that all cubic invariants are manifestly real. Thus, in order to find a potentially complex invariant, one must examine invariants of fourth order and higher. At this point, an explicit hand calculation becomes infeasible, and we must employ a computer algebra program such as Mathematica to assist in the analysis. For example, consider all possible invariants that are independent of $`Y_{a\overline{b}}`$ (such invariants will be called $`Z`$-invariants). One can use Mathematica to evaluate the imaginary part of each invariant by explicitly considering invariants which consist of $`n`$-fold products of $`Z`$โs. These invariants are of the form:
$$Z_{a_1\overline{b}_1c_1\overline{d}_1}Z_{a_2\overline{b}_2c_2\overline{d}_2}\mathrm{}Z_{a_n\overline{b}_nc_n\overline{d}_n},$$
(38)
where one chooses the indices $`\{b_1,d_1,b_2,d_2,\mathrm{},b_n,d_n\}`$ to be a particular permutation of $`\{a_1,c_1,a_2,c_2,\mathrm{},a_n,c_n\}`$, and then sums over the repeated indices as usual. By considering all possible permutations, one generates all $`(2n)!`$ possible invariants (many of which are trivially related to others in the complete list of invariants). One can automate the computation with a Mathematica program and compute the imaginary part of all $`(2n)!`$ invariants subject to the constraints of computer time. The procedure can be generalized to include some number of $`Y_{a\overline{b}}`$. In particular, it is easy to show (without computer assistance) that all invariants that are independent of $`Z`$ (such invariants will be called $`Y`$-invariants) are manifestly real, due to the hermiticity property of $`Y_{a\overline{b}}`$.
### IV.1 Fourth-order potentially complex invariants
Among the quartic invariants, we first construct all possible quartic $`Z`$-invariants. By an explicit Mathematica computation, we were able to show that all $`8!=40,320`$ quartic $`Z`$-invariants are manifestly real.
We next search for potentially complex quartic invariants that are linear in $`Y`$. We display one potentially non-zero $`I`$-invariant below:
$`I_{Y3Z}`$ $``$ $`\mathrm{Im}(Z_{a\overline{c}}^{(1)}Z_{e\overline{b}}^{(1)}Z_{b\overline{e}c\overline{d}}Y_{d\overline{a}})`$ (39)
$`=`$ $`2(|\lambda _6|^2|\lambda _7|^2)\mathrm{Im}[Y_{12}(\lambda _6^{}+\lambda _7^{})]+(\lambda _1\lambda _2)\left[\mathrm{Im}(Y_{12}\mathrm{\Lambda }^{})\mathrm{Im}[Y_{12}\lambda _5^{}(\lambda _6+\lambda _7)]\right]`$
$`+(Y_{11}Y_{22})\left[\mathrm{Im}[\lambda _5^{}(\lambda _6+\lambda _7)^2](\lambda _1\lambda _2)\mathrm{Im}(\lambda _7^{}\lambda _6)\right],`$
where
$$\mathrm{\Lambda }(\lambda _2\lambda _3\lambda _4)\lambda _6+(\lambda _1\lambda _3\lambda _4)\lambda _7.$$
(40)
Using Mathematica, we have evaluated the imaginary part of all 7!= 5,040 possible invariants that are linear in $`Y`$ and cubic in $`Z`$. We find that the result either vanishes or is equal to $`\pm I_{Y3Z}`$.
Next, we examine potentially complex quartic invariants that are quadratic in both $`Y`$ and $`Z`$. We display one potentially non-zero $`I`$-invariant below:
$`I_{2Y2Z}`$ $``$ $`\mathrm{Im}(Y_{a\overline{b}}Y_{c\overline{d}}Z_{b\overline{a}d\overline{f}}Z_{f\overline{c}}^{(1)})`$ (41)
$`=`$ $`(\lambda _1\lambda _2)\mathrm{Im}(Y_{12}^2\lambda _5^{})(Y_{11}Y_{22})\left[\mathrm{Im}(Y_{12}\mathrm{\Lambda }^{})+\mathrm{Im}(Y_{12}\lambda _5^{}(\lambda _6+\lambda _7))\right]`$
$`\mathrm{Im}[(Y_{12}\lambda _6^{})^2]+\mathrm{Im}[(Y_{12}\lambda _7^{})^2]+\left[(Y_{11}Y_{22})^22|Y_{12}|^2\right]\mathrm{Im}(\lambda _7^{}\lambda _6).`$
Moreover, we find as before that the imaginary parts of all such invariants (there are 6!=720 invariants that are quadratic in both $`Y`$ and $`Z`$) either vanish or are equal to $`\pm I_{2Y2Z}`$.
It is easy to show that quartic invariants that are cubic in $`Y`$ (and therefore linear in $`Z`$) are manifestly real. In particular, there are only two such invariants that are not a product of lower order invariants: $`\mathrm{Tr}(Y^3Z^{(1)})`$ and $`\mathrm{Tr}(Y^3Z^{(2)})`$. Both these invariants are manifestly real due to the hermiticity properties of $`Y`$, $`Z^{(1)}`$ and $`Z^{(2)}`$.
### IV.2 Fifth order potentially complex invariants
We begin by constructing all possible fifth-order $`Z`$-invariants. Again, with the help of Mathematica, we found that all $`10!=3,628,800`$ $`Z`$-invariants are manifestly real. Next, we considered the $`Y4Z`$-invariants, i.e. the fifth-order invariants that are linear in $`Y`$. After computing the imaginary parts of all $`9!=362,880`$ such invariants, we found that only one genuinely new potentially complex invariant emerged. The corresponding $`I`$-invariant is :
$`I_{Y4Z}`$ $`=`$ $`\mathrm{Im}[Z_{a\overline{b}}^{(2)}Z_{b\overline{a}c\overline{d}}Z_{d\overline{e}}^{(2)}Z_{e\overline{c}f\overline{g}}Y_{g\overline{f}}]`$ (42)
$`=`$ $`\lambda _4I_{Y3Z}+(\lambda _1\lambda _2)\mathrm{Im}[Y_{12}(\lambda _6^{}+\lambda _7^{})^2(\lambda _6^{}\lambda _7^{})]`$
$`+\mathrm{Im}[Y_{12}\lambda _5^{}(\lambda _6^2\lambda _7^{}\lambda _7^2\lambda _6^{})]+\mathrm{Im}[Y_{12}\lambda _5(\lambda _6^{\mathrm{\hspace{0.17em}2}}\lambda _7^{\mathrm{\hspace{0.17em}2}})(\lambda _6^{}+\lambda _7^{})]`$
$`+\frac{1}{2}\left((\lambda _1\lambda _2)(\lambda _1+\lambda _22\lambda _32\lambda _4)+|\lambda _7|^2|\lambda _6|^2\right)\mathrm{Im}[Y_{12}\lambda _5^{}(\lambda _6+\lambda _7)]`$
$`\frac{1}{2}\left((\lambda _1\lambda _2)^2|\lambda _6|^2|\lambda _7|^2\right)\mathrm{Im}[Y_{12}\lambda _5^{}(\lambda _6\lambda _7)]`$
$`+\frac{1}{2}(\lambda _1\lambda _2)(2|\lambda _5|^2|\lambda _6|^2|\lambda _7|^2)\mathrm{Im}[Y_{12}(\lambda _6^{}+\lambda _7^{})]`$
$`+\frac{1}{2}(\lambda _1\lambda _2)(|\lambda _6|^2|\lambda _7|^2)\mathrm{Im}[Y_{12}(\lambda _6^{}\lambda _7^{})]`$
$`+\frac{1}{2}(Y_{11}Y_{22})[4(|\lambda _6|^2|\lambda _7|^2)\mathrm{Im}(\lambda _6\lambda _7^{})+(\lambda _1\lambda _2)\mathrm{Im}[\lambda _5^{}(\lambda _7^2\lambda _6^2)]`$
$`+(\lambda _1+\lambda _22\lambda _32\lambda _4)\mathrm{Im}[\lambda _5^{}(\lambda _6+\lambda _7)^2]],`$
where $`I_{Y3Z}`$ is given by eq. (39). In addition, we have explicitly verified that the imaginary parts of all potentially complex $`Y4Z`$-invariants reduce to a linear combination of $`I_{Y4Z}`$ and the product of $`I_{Y3Z}`$ times a linear combination of $`\mathrm{Tr}[Z^{(1)}]`$ and $`\mathrm{Tr}[Z^{(2)}]`$.
The fact that $`I_{Y4Z}`$ is a โnewโ $`I`$-invariant means that one cannot express $`I_{Y4Z}`$ as a sum of terms, each of which is the imaginary part of a product of lower-order invariants. Nevertheless, one can show that if $`I_{Y3Z}=I_{2Y2Z}=0`$, then it follows that $`I_{Y4Z}=0`$. This is most easily accomplished in the basis where $`\lambda _7=\lambda _6`$. In this basis, eq. (42) simplifies enormously:
$$I_{Y4Z}=(\lambda _1\lambda _2)^2\left[\lambda _4\mathrm{Im}(Y_{12}\lambda _6^{})\mathrm{Im}(Y_{12}\lambda _6\lambda _5^{})\right].$$
(43)
If $`Y_{12}=0`$ in the $`\lambda _7=\lambda _6`$ basis then $`I_{Y4Z}=0`$. Alternatively, if $`Y_{12}0`$, then we make use of:
$$\mathrm{Im}(Y_{12}\lambda _6\lambda _5^{})=\frac{1}{|Y_{12}|^2}\left[\mathrm{Im}(Y_{12}^2\lambda _5^{})\mathrm{Re}(Y_{12}\lambda _6^{})\mathrm{Im}(Y_{12}\lambda _6^{})\mathrm{Re}(Y_{12}^2\lambda _5^{})\right].$$
(44)
Since $`I_{Y3Z}=I_{2Y2Z}=0`$ implies that either $`\lambda _1=\lambda _2`$ or $`\mathrm{Im}(Y_{12}^2\lambda _5^{})=\mathrm{Im}(Y_{12}\lambda _6^{})=0`$ \[see eqs. (28) and (29)\], one can again conclude that $`I_{Y4Z}=0`$. Having proved that the invariant $`I_{Y4Z}`$ vanishes in one basis, it immediately follows that $`I_{Y4Z}=0`$ in all basis choices.
Similarly, we have analyzed the $`2Y3Z`$-invariants, i.e., the fifth-order invariants that are quadratic in $`Y`$ and cubic in $`Z`$. Again, we have computed the imaginary parts of all $`40,320`$ such invariants. We have explicitly verified that any potentially complex fifth-order invariant of this type is a linear combination of $`I_{Y3Z}`$ (with coefficient proportional to $`\mathrm{Tr}Y`$), $`I_{2Y2Z}`$ (with coefficient proportional to a linear combination of $`\mathrm{Tr}[Z^{(1)}]`$ and $`\mathrm{Tr}[Z^{(2)}`$\]) and one new potentially complex invariant form. A particular choice for the new $`I`$-invariant is:
$$I_{2Y3Z}=\mathrm{Im}[Z_{a\overline{c}b\overline{e}}Z_{c\overline{f}d\overline{b}}Z_{e\overline{g}f\overline{h}}Y_{g\overline{a}}Y_{h\overline{d}}].$$
(45)
One could write out the explicit expression for $`I_{2Y3Z}`$ \[as we did in eq. (42) for $`I_{Y4Z}`$\]. However, for our purposes, it is sufficient to give the form of $`I_{2Y3Z}`$ in the $`\lambda _7=\lambda _6`$ basis:
$`I_{2Y3Z}`$ $`=`$ $`(\lambda _1\lambda _2)[4\mathrm{I}\mathrm{m}[Y_{12}^2\lambda _6^{\mathrm{\hspace{0.17em}2}}]+2(Y_{11}Y_{22})\mathrm{Im}[Y_{12}\lambda _5^{}\lambda _6](\lambda _1+\lambda _22\lambda _3)\mathrm{Im}[Y_{12}^2\lambda _5^{}]`$ (46)
$`2\lambda _4(Y_{11}Y_{22})\mathrm{Im}(Y_{12}\lambda _6^{})].`$
Again, it must be emphasized that $`I_{2Y3Z}`$ is a โnewโ $`I`$-invariant in the sense that one cannot express $`I_{2Y3Z}`$ as a sum of terms, each of which is the imaginary part of a product of lower-order invariants. Nevertheless, $`I_{Y3Z}=I_{2Y2Z}=0`$ implies that $`I_{2Y3Z}=0`$.<sup>13</sup><sup>13</sup>13This is easily verified after noting that $`\mathrm{Im}(Y_{12}^2\lambda _6^2)=2\mathrm{I}\mathrm{m}(Y_{12}\lambda _6^{})\mathrm{Re}(Y_{12}\lambda _6^{})`$.
The remaining cases are easily treated. We explicitly verified that any fifth-order invariants that is cubic in $`Y`$ and quadratic in $`Z`$ is proportional to $`(\mathrm{Tr}Y)I_{2Y2Z}`$. It is also simple to show that all fifth-order invariants that are linear in $`Z`$ are manifestly real. In particular, the only two inequivalent invariants of this type that are not products of lower order invariants are $`Z_{a\overline{b}c\overline{d}}Y_{b\overline{a}}^2Y_{d\overline{c}}^2`$ and $`Z_{a\overline{b}c\overline{d}}Y_{b\overline{c}}^2Y_{d\overline{a}}^2`$. By explicit calculation, using the hermiticity properties of $`Y`$ and $`Z`$, it is straightforward to verify that both these invariants are real. We have previously noted that all pure $`Y`$-invariants are manifestly real. This completes the proof that all potentially complex fifth-order invariants are linear combinations of $`I_{Y3Z}`$ and $`I_{2Y2Z}`$ or forms that vanish when $`I_{Y3Z}=I_{2Y2Z}=0`$. That is, the consideration of potentially complex fifth order invariants does not establish any new independent conditions for CP violation.
### IV.3 Sixth-order potentially complex invariants
Two new independent conditions for CP violation arise from the study of sixth-order potentially complex invariants. We begin by constructing all possible sixth-order $`Z`$-invariants. It is here that we encounter the first potentially complex $`Z`$-invariants. One potentially non-zero $`I`$-invariant is:
$`I_{6Z}`$ $``$ $`\mathrm{Im}(Z_{a\overline{b}c\overline{d}}Z_{b\overline{f}}^{(1)}Z_{d\overline{h}}^{(1)}Z_{f\overline{a}j\overline{k}}Z_{k\overline{j}m\overline{n}}Z_{n\overline{m}h\overline{c}})`$ (47)
$`=`$ $`2|\lambda _5|^2\mathrm{Im}[(\lambda _7^{}\lambda _6)^2]\mathrm{Im}[\lambda _{5}^{}{}_{}{}^{\mathrm{\hspace{0.17em}2}}(\lambda _6\lambda _7)(\lambda _6+\lambda _7)^3]+(\lambda _1\lambda _2)|\lambda _5|^2\mathrm{Im}[\lambda _5^{}(\lambda _6+\lambda _7)^2]`$
$`+2\mathrm{I}\mathrm{m}(\lambda _7^{}\lambda _6)\left[|\lambda _5|^2[|\lambda _6|^2+|\lambda _7|^2(\lambda _1\lambda _2)^2]2(|\lambda _6|^2|\lambda _7|^2)^2\right]`$
$`(\lambda _1\lambda _2)\mathrm{Im}(\lambda _5^{}\mathrm{\Lambda }^2)2(|\lambda _6|^2|\lambda _7|^2)\mathrm{Im}[\lambda _5^{}\mathrm{\Lambda }(\lambda _6+\lambda _7)]`$
$`+(\lambda _1\lambda _2)[\mathrm{Im}\left[\mathrm{\Lambda }(\lambda _7\lambda _6^{\mathrm{\hspace{0.17em}2}}+\lambda _6\lambda _7^{\mathrm{\hspace{0.17em}2}}|\lambda _7|^2\lambda _6^{}|\lambda _6|^2\lambda _7^{})\right]`$
$`+2\mathrm{I}\mathrm{m}\left[\lambda _5((|\lambda _6|^2+|\lambda _7|^2)\lambda _6^{}\lambda _7^{}\lambda _7\lambda _6^{\mathrm{\hspace{0.17em}3}}\lambda _6\lambda _7^{\mathrm{\hspace{0.17em}3}})\right]],`$
where $`\mathrm{\Lambda }`$ is defined in eq. (40).
Theorem 2 implies that if $`Y_{a\overline{b}}=0`$ and $`I_{6Z}=0`$, then any $`Z`$-invariant is real. Consequently, the imaginary part of any sixth-order $`Z`$ invariant must be equal to $`cI_{6Z}`$, for some real constant $`c`$. Our proof of Theorem 2 in section III leaves no doubt as to the veracity of this conclusion. Nevertheless, it is instructive to check this assertion explicitly. Unfortunately, a complete survey of all possible $`12!=479,001,600`$ sixth-order complex $`Z`$-invariants is beyond the capability of our desktop computers. However, we were able to examine roughly nine million sixth-order $`Z`$-invariants, and in these cases the imaginary part of each sixth-order $`Z`$-invariant either vanishes or is equal to $`\pm I_{6Z}`$ or $`\pm 2I_{6Z}`$.
If $`\lambda _1\lambda _2`$ in a basis where $`\lambda _7=\lambda _6`$, then Theorem 2 implies that $`I_{Y3Z}=I_{2Y2Z}=I_{6Z}=0`$ is a necessary and sufficient condition for an explicitly CP-conserving 2HDM scalar potential. However, the case of $`\lambda _1=\lambda _2`$ and $`\lambda _7=\lambda _6`$ (where $`I_{Y3Z}=I_{2Y2Z}=I_{6Z}=0`$ is automatic) must be treated separately. In this latter case, the condition for CP violation depends on an independent invariant that first arises at sixth order and is made up of three $`Y`$ and three $`Z`$ factors (henceforth denoted as $`3Y3Z`$-invariants).
Thus, we have constructed all possible $`3Y3Z`$-invariants and examined their imaginary parts. Of course, some of these will simply be linear combinations of lower-order invariants already examined. A complete survey of the imaginary part of all possible 9!=362,880 $`3Y3Z`$-invariants yields one new independent $`I`$-invariant. A representative choice is:
$`I_{3Y3Z}`$ $`=`$ $`\mathrm{Im}(Z_{a\overline{c}b\overline{d}}Z_{c\overline{e}d\overline{g}}Z_{e\overline{h}f\overline{q}}Y_{g\overline{a}}Y_{h\overline{b}}Y_{q\overline{f}})`$ (48)
$`=`$ $`(Y_{11}Y_{22})\left[(\lambda _1\lambda _3\lambda _4)(\lambda _2\lambda _3\lambda _4)|\lambda _5|^2+|\lambda _6|^2+|\lambda _7|^2\right]\mathrm{Im}[Y_{12}^2\lambda _5^{}]`$
$`+[(Y_{11}Y_{22})^2|Y_{12}|^2]|\lambda _5|^2\mathrm{Im}[Y_{12}(\lambda _7^{}\lambda _6^{})]Y_{11}Y_{22}(\lambda _1\lambda _2)\mathrm{Im}[Y_{12}\mathrm{\Lambda }^{}]`$
$`+2[(Y_{11}Y_{22})^2+Y_{11}Y_{22}|Y_{12}|^2]\left[|\lambda _7|^2\mathrm{Im}(Y_{12}\lambda _6^{})|\lambda _6|^2\mathrm{Im}(Y_{12}\lambda _7^{})\right]`$
$`+2Y_{11}Y_{22}\left[|\lambda _7|^2\mathrm{Im}(Y_{12}\lambda _7^{})|\lambda _6|^2\mathrm{Im}(Y_{12}\lambda _6^{})\right]`$
$`+(\lambda _1\lambda _2)Y_{11}Y_{22}\mathrm{Im}[Y_{12}\lambda _5^{}(\lambda _6+\lambda _7)][(Y_{11}Y_{22})^2|Y_{12}|^2]\mathrm{Im}(Y_{12}\lambda _5^{}\stackrel{~}{\mathrm{\Lambda }})`$
$`(Y_{11}Y_{22})\{(Y_{11}Y_{22}+|Y_{12}|^2)[\mathrm{Im}[\lambda _5^{}(\lambda _6^2+\lambda _7^2)](\lambda _1\lambda _2)\mathrm{Im}(\lambda _6\lambda _7^{})]`$
$`+(Y_{11}^2+Y_{22}^24|Y_{12}|^2)\mathrm{Im}[\lambda _5^{}\lambda _6\lambda _7](\lambda _1+\lambda _22\lambda _32\lambda _4)\mathrm{Im}[Y_{12}^2\lambda _6^{}\lambda _7^{}]\}`$
$`+\mathrm{Im}[Y_{12}^3\lambda _5^{}\stackrel{~}{\mathrm{\Lambda }}^{}]+2\mathrm{Im}[Y_{12}^3\lambda _6^{}\lambda _7^{}(\lambda _6^{}\lambda _7^{})]+\mathrm{Im}[Y_{12}^3(\lambda _5^{})^2(\lambda _6\lambda _7)],`$
where $`\mathrm{\Lambda }`$ is defined in eq. (40) and
$$\stackrel{~}{\mathrm{\Lambda }}(\lambda _2\lambda _3\lambda _4)\lambda _6(\lambda _1\lambda _3\lambda _4)\lambda _7.$$
(49)
We have explicitly verified that the imaginary part of any $`3Y3Z`$ invariant is a real linear combination of $`I_{3Y3Z}`$, $`(\mathrm{Tr}Y)I_{2Y3Z}`$, $`[\mathrm{Tr}Y]^2I_{Y3Z}`$, $`[\mathrm{Tr}Y^2]I_{Y3Z}`$, $`\mathrm{Tr}[YZ^{(1)}]I_{2Y2Z}`$, $`\mathrm{Tr}[YZ^{(2)}]I_{2Y2Z}`$ and $`(\mathrm{Tr}Y\mathrm{Tr}Z^{(1)})I_{2Y2Z}`$.<sup>14</sup><sup>14</sup>14Note that $`\mathrm{Tr}Y\mathrm{Tr}Z^{(2)}=\mathrm{Tr}Y\mathrm{Tr}Z^{(1)}\mathrm{Tr}[Y(Z^{(1)}Z^{(2)})]`$. In a basis where $`\lambda _7=\lambda _6`$, $`I_{3Y3Z}`$ reduces to the expression given by eq. (31). Indeed, $`I_{3Y3Z}`$ is non-zero in explicitly CP-violating models with $`\lambda _7=\lambda _6`$ and $`\lambda _1=\lambda _2`$, which confirms that it is a necessary ingredient in the formulation of Theorem 2.
Among other sixth order invariants, all $`6Y`$ and $`Z5Y`$ invariants are manifestly real. A $`2Z4Y`$ invariant is potentially complex, but its imaginary part must be proportional to some linear combination of $`(\mathrm{Tr}Y)^2I_{2Y2Z}`$ and $`(\mathrm{Tr}Y^2)I_{2Y2Z}`$. This leaves two interesting cases: the $`Y5Z`$ and $`2Y4Z`$ invariants, which we now consider in more detail.
A partial scan of the imaginary part of $`10!=3628800`$ $`2Y4Z`$-invariants and $`11!=39916800`$ $`Y5Z`$-invariants has been performed, and our results yield two genuinely new potentially complex invariants, whose imaginary parts we designate by $`I_{2Y4Z}`$ and $`I_{Y5Z}`$, respectively. The resulting expressions in a generic basis are quite complicated and not very illuminating. Hence, here we provide only the explicit forms in a basis where $`\lambda _7=\lambda _6`$:
$`I_{2Y4Z}`$ $`=`$ $`\mathrm{Im}(Z_{b\overline{c}}^{(2)}Z_{c\overline{e}d\overline{f}}Z_{e\overline{q}f\overline{r}}Z_{g\overline{b}h\overline{d}}Y_{q\overline{g}}Y_{r\overline{h}})`$ (50)
$`=`$ $`(\lambda _1\lambda _2)\{(\lambda _1+\lambda _2)\mathrm{Im}(Y_{12}^2\lambda _6^{\mathrm{\hspace{0.17em}2}})(\lambda _1\lambda _2|\lambda _5|^22|\lambda _6|^2)\mathrm{Im}(Y_{12}^2\lambda _5^{})`$
$`+\left[2(\lambda _1Y_{11}\lambda _2Y_{22})(\lambda _3+\lambda _4)(Y_{11}Y_{22})\right]\mathrm{Im}(Y_{12}\lambda _6\lambda _5^{})`$
$`+\left[2|\lambda _5|^2(Y_{11}Y_{22})(\lambda _3+\lambda _4)(\lambda _1Y_{11}\lambda _2Y_{22})\right]\mathrm{Im}(Y_{12}\lambda _6^{})`$
$`[(Y_{11}Y_{22})^22|Y_{12}|^2]\mathrm{Im}(\lambda _6^2\lambda _5^{})\},`$
and
$`I_{Y5Z}`$ $`=`$ $`\mathrm{Im}(Z_{b\overline{c}}^{(1)}Z_{c\overline{b}d\overline{e}}Z_{e\overline{d}f\overline{g}}Z_{g\overline{q}}^{(1)}Z_{q\overline{f}r\overline{s}}Y_{s\overline{r}})`$ (51)
$`=`$ $`(\lambda _1\lambda _2)^2\{(Y_{11}Y_{22})\mathrm{Im}(\lambda _6^2\lambda _5^{})(\lambda _1+\lambda _2)\mathrm{Im}(Y_{12}\lambda _6\lambda _5^{})`$
$`+[\lambda _4(\lambda _1+\lambda _2)+\lambda _4^2|\lambda _5|^2]\mathrm{Im}(Y_{12}\lambda _6^{})\}.`$
If $`Y_{12}0`$ in the $`\lambda _7=\lambda _6`$ basis, then we can use:
$$\mathrm{Im}(\lambda _6^2\lambda _5^{})=\frac{1}{|Y_{12}|^2}\left[\mathrm{Re}(Y_{12}\lambda _6^{})\mathrm{Im}(Y_{12}\lambda _6\lambda _5^{})\mathrm{Re}(Y_{12}\lambda _6\lambda _5^{})\mathrm{Im}(Y_{12}\lambda _6^{})\right]$$
(52)
along with eqs. (28), (29) and (44) to conclude that both $`I_{2Y4Z}`$ and $`I_{Y5Z}`$ vanish if $`I_{Y3Z}=I_{2Y2Z}=0`$.<sup>15</sup><sup>15</sup>15Although this result is demonstrated in the $`\lambda _7=\lambda _6`$ basis, the conclusion must hold for all basis choices. However, if $`Y_{12}=0`$ in the $`\lambda _7=\lambda _6`$ basis, then all invariants of $`n`$th order with $`n5`$ are real. In the latter case, both $`I_{2Y4Z}`$ and $`I_{Y5Z}`$ can still be non-vanishing, which demonstrates that these are new $`I`$-invariants. Nevertheless, by the same argument as before, we may conclude that if $`I_{Y3Z}=I_{2Y2Z}=I_{6Z}=0`$, then both $`I_{2Y4Z}`$ and $`I_{Y5Z}`$ must vanish. For this reason, $`I_{2Y4Z}`$ and $`I_{Y5Z}`$ need not be independently considered in the formulation of Theorem 2.<sup>16</sup><sup>16</sup>16However, it is not possible to express either $`I_{2Y4Z}`$ or $`I_{Y5Z}`$ as a linear combination of $`I_{6Z}`$ $`I_{Y3Z}`$ and $`I_{2Y2Z}`$ with corresponding coefficients that are invariant quantities. In particular, $`I_{6Z}`$ is included in the statement of Theorem 2, since (unlike $`I_{2Y4Z}`$ and $`I_{Y5Z}`$) $`I_{6Z}`$ can be non-zero even when $`Y_{a\overline{b}}=0`$.
We have verified<sup>17</sup><sup>17</sup>17Our conclusion is based on a partial scan of about two million invariants. However, the arguments in the next sub-section strongly suggest that the following results apply to all $`2Y4Z`$ and $`Y5Z`$ invariants. that the imaginary part of any $`2Y4Z`$ invariant can be expressed as a real linear combination of $`I_{2Y4Z}`$, $`I_{2Y3Z}`$, $`I_{2Y2Z}`$ and $`I_{Y3Z}`$. Likewise, the imaginary part of any $`Y5Z`$ invariant can be expressed as a real linear combination of $`I_{Y5Z}`$, $`I_{Y4Z}`$ and $`I_{Y3Z}`$. In both cases, each of the corresponding coefficients of the linear combination of terms are real invariant quantities. As an explicit illustrative example, we have verified:
$`\mathrm{Im}\left[Z_{b\overline{c}}^{(2)}Z_{c\overline{b}d\overline{e}}Z_{e\overline{d}f\overline{g}}Z_{g\overline{q}h\overline{r}}Y_{q\overline{f}}Y_{r\overline{h}}\right]=I_{2Y4Z}\frac{1}{2}\left[\mathrm{Tr}(Z^{(1)}Z^{(2)})\frac{1}{2}[\mathrm{Tr}Z^{(1)}]^2+Z_{a\overline{c}b\overline{d}}Z_{c\overline{a}d\overline{b}}\right]I_{2Y2Z}`$
$`+\frac{1}{4}\mathrm{Tr}Z^{(1)}I_{2Y3Z}+\frac{1}{2}\mathrm{Tr}YI_{Y4Z}\frac{1}{2}\left[\mathrm{Tr}(Z^{(1)}Y)+\frac{1}{2}\mathrm{Tr}Y\mathrm{Tr}Z^{(2)}\right]I_{Y3Z}.`$ (53)
### IV.4 General results for $`๐`$th-order potentially complex invariants
The analyzes of Sections IV.A and IV.B permit us to conjecture a number of results that we expect to hold for complex invariants of arbitrary order. These results provide a method for identifying the number of โnewโ potentially complex invariants at any order. As before, we define a โnewโ $`n`$th order $`I`$-invariant to be one that cannot be written as a sum of terms, each of which is the imaginary part of a product of known invariants of order $`n`$. By this definition, โnewโ $`I`$-invariants arise at each order (for $`n4`$). However, as previously stated, if $`I_{Y3Z}=I_{2Y2Z}=I_{6Z}=I_{3Y3Z}=0`$, then any new $`I`$-invariant that arises must also vanish.
Consider an arbitrary $`n`$th order $`I`$-invariant $`I_{pYqZ}`$ made up of $`p`$ factors of $`Y`$ and $`q=np`$ factors of $`Z`$. In a basis where $`\lambda _7=\lambda _6`$, for $`p3`$
$$I_{pYqZ}=(\lambda _1\lambda _2)^{3p}\mathrm{Im}P(Y_{12},\lambda _5,\lambda _6),$$
(54)
where $`P`$ is a polynomial of its arguments and their complex conjugates constructed such that each term in the sum contains $`p`$ factors of $`Y_{a\overline{b}}`$ and $`q+p3`$ factors of the $`\lambda _i`$, with the constraint that the weight of each term in the sum is zero. Here, we define the weight $`w`$ according to the rules: $`w(Y_{12})=+1`$, $`w(\lambda _5)=+2`$, $`w(\lambda _6)=+1`$, $`w(x^{})=w(x)`$ for any $`x`$ and $`w(xy)=w(x)+w(y)`$ for any $`x`$, $`y`$.<sup>18</sup><sup>18</sup>18Formally, the weight $`w=w(x)`$ for any scalar potential parameter $`x`$ is defined such that $`xe^{iw\theta }x`$ under a redefinition of one of the scalar fields by $`\mathrm{\Phi }_1e^{i\theta }\mathrm{\Phi }_1`$. Of course, $`w=0`$ for any real scalar potential parameter.
The polynomial $`P`$ possesses one additional property of note: it does not vanish in the limit of $`\lambda _1=\lambda _2`$ (assuming that $`P0`$ in general). That is, the behavior of $`I_{pYqZ}`$ in the $`\lambda _1\lambda _2`$ limit is specified explicitly in eq. (54). If $`p>3`$, then $`I_{pYqZ}=0`$. For example at sixth order, $`\mathrm{Tr}(Y^2)I_{2Y2Z}`$ is a potentially non-vanishing $`I`$-invariant with $`p=4`$, but this does not constitute a new $`I`$-invariant by the above definition.
Eq. (54) is consistent with all the results of Sections IV.A and IV.B. It also provides an explanation for the absence of complex invariants of low order. For example, if we apply eq. (54) and attempt to construct $`I_{5Z}`$, we would need to find a polynomial $`P`$ with a non-zero imaginary part that is quadratic in the $`\lambda _i`$. No such polynomial exists, and we conclude that $`I_{5Z}=0`$. We can also use eq. (54) to predict the results of higher order invariants. For example, all seventh and eighth order $`Z`$-invariants must be proportional to $`I_{6Z}`$ (a result that we have confirmed by limited scanning). However, a โnewโ $`Z`$-invariant arises at ninth order, which in the $`\lambda _7=\lambda _6`$ basis must have an imaginary part that is a linear combination of $`I_{6Z}P_3(\lambda _i)`$ and $`(\lambda _1\lambda _2)^3\mathrm{Im}[(\lambda _6^2\lambda _5^{})^2]`$, where $`P_3(\lambda _i)`$ is a real cubic polynomial of the $`\lambda _i`$. Although this is a new $`I`$-invariant, it clearly vanishes when $`I_{6Z}=0`$.
Finally, eq. (54) strongly suggests that there is only one new $`2Y4Z`$ $`I`$-invariant and one new $`Y5Z`$ $`I`$-invariant, since in each case, only one new term, $`\mathrm{Im}(\lambda _6^2\lambda _5^{})`$ arises that did not appear in lower-order invariants (in the $`\lambda _7=\lambda _6`$ basis).<sup>19</sup><sup>19</sup>19Unfortunately, this argument fails to explain the existence of only one $`3Y3Z`$ $`I`$-invariant, a fact that has been confirmed only by a complete scan over all possible invariants of this type.
## V Implications for spontaneous CP-violation
If a Higgs potential is explicitly CP-conserving, then there exists a so-called โreal basisโ in which all the Higgs potential parameters are real. A theory with an explicitly CP-conserving Higgs sector may be CP-violating if the vacuum does not respect the CP-symmetry. In this case, we say that CP is spontaneously broken Lee:1973iz . To determine whether CP is spontaneously broken, one must check whether the vacuum is invariant under time reversal. We assert the following theorem, which is proved in Appendix F:
Theorem 3: Given an explicitly CP-conserving Higgs potential, the vacuum is time-reversal invariant if and only if a real basis exists in which the Higgs vacuum expectation values are real.
Theorem 3 requires one to verify the existence or nonexistence of a basis with certain properties. However, these theorems can be reformulated in a basis-independent language. Here, we follow ref. lavoura , and introduce three U(2)-invariants davidson :
$`\frac{1}{2}v^2J_1`$ $``$ $`\widehat{v}_{\overline{a}}^{}Y_{a\overline{b}}Z_{b\overline{d}}^{(1)}\widehat{v}_d,`$ (55)
$`\frac{1}{4}v^4J_2`$ $``$ $`\widehat{v}_{\overline{b}}^{}\widehat{v}_{\overline{c}}^{}Y_{b\overline{e}}Y_{c\overline{f}}Z_{e\overline{a}f\overline{d}}\widehat{v}_a\widehat{v}_d,`$ (56)
$`J_3`$ $``$ $`\widehat{v}_{\overline{b}}^{}\widehat{v}_{\overline{c}}^{}Z_{b\overline{e}}^{(1)}Z_{c\overline{f}}^{(1)}Z_{e\overline{a}f\overline{d}}\widehat{v}_a\widehat{v}_d,`$ (57)
where $`\mathrm{\Phi }_a^0v\widehat{v}_a/\sqrt{2}`$, with $`v=246`$ GeV and $`\widehat{v}`$ is a unit vector in the complex two-dimensional Higgs flavor space. The scalar potential minimum condition is easily derived from eq. (16):
$$\widehat{v}_{\overline{a}}^{}[Y_{a\overline{b}}+\frac{1}{2}v^2Z_{a\overline{b}c\overline{d}}\widehat{v}_{\overline{c}}^{}\widehat{v}_d]=0.$$
(58)
Thus, we may eliminate $`Y`$ in the expressions for $`J_1`$ and $`J_2`$:
$`J_1`$ $``$ $`\widehat{v}_{\overline{a}}^{}\widehat{v}_{\overline{e}}^{}Z_{a\overline{b}e\overline{f}}Z_{b\overline{d}}^{(1)}\widehat{v}_d\widehat{v}_f,`$ (59)
$`J_2`$ $``$ $`\widehat{v}_{\overline{b}}^{}\widehat{v}_{\overline{c}}^{}\widehat{v}_{\overline{g}}^{}\widehat{v}_{\overline{p}}^{}Z_{b\overline{e}g\overline{h}}Z_{c\overline{f}p\overline{r}}Z_{e\overline{a}f\overline{d}}\widehat{v}_a\widehat{v}_d\widehat{v}_h\widehat{v}_r.`$ (60)
Since $`\mathrm{Im}Y_{12}`$ is determined by the scalar potential minimum conditions in terms of $`\mathrm{Im}\lambda _{5,6,7}`$, one is left with three potentially complex parameters in a basis where $`\widehat{v}`$ is real. These are in one-to-one correspondence with $`J_1`$, $`J_2`$ and $`J_3`$.
Theorem 4: Consider the 2HDM scalar potential in some arbitrary basis. Assume that the minimum of the scalar potential preserves U(1)<sub>EM</sub>. Then, the Higgs sector is CP-conserving (i.e., no explicit nor spontaneous CP-violation is present) if $`J_1`$, $`J_2`$ and $`J_3`$ defined in eqs. (55)โ(57) are real lavoura .
If the Higgs sector is CP-conserving, then according to Theorem 3 some basis must exist in which the Higgs potential parameters and the Higgs field vacuum expectation values are simultaneously real. But in that case, we may immediately conclude that the invariant quantities $`J_1`$, $`J_2`$ and $`J_3`$ must be real. Conversely, the reality of $`J_1`$, $`J_2`$ and $`J_3`$ provide sufficient conditions for a CP-invariant Higgs sector. This result is proven in refs. lavoura and branco ,<sup>20</sup><sup>20</sup>20In fact, there are at most two independent relative phases among $`J_1`$, $`J_2`$ and $`J_3`$. However, as shown in ref. davidson , there are cases where two of the three invariants are real and only one has an non-vanishing imaginary part, which shows that one must check all three invariants in order to determine whether the Higgs sector is CP-invariant. and we do not repeat the proof here.
Note that eqs. (55)โ(57) are considerably simpler than the invariants that govern explicit CP-violation of the Higgs potential \[eqs. (23)โ(26)\]. However, these two sets of invariants serve different purposes. To answer the question of whether the Higgs sector is CP-invariant, one must first choose a basis and minimize the scalar potential. Having found $`\widehat{v}_a`$, one may now compute $`J_1`$, $`J_2`$ and $`J_3`$. If these invariants are all real, then the Higgs potential is explicitly CP-invariant and there is no spontaneous CP-violation. If at least one of the invariants $`J_1`$, $`J_2`$ and $`J_3`$ is complex, then the Higgs sector is CP-violating. However, in this latter case, one must evaluate the four $`I`$-invariants given in eqs. (23)โ(26) to determine whether CP is spontaneously or explicitly broken. If these four $`I`$-invariants all vanish, then CP is spontaneously broken. If at least one of these is non-zero, then CP is explicitly broken. These conclusions are summarized in our final theorem:
Theorem 5: The necessary and sufficient conditions for spontaneous CP-violation in the 2HDM are: (i) $`I_{Y3Z}=I_{2Y2Z}=I_{6Z}=I_{3Y3Z}=0`$, and (ii) at least one of the three invariants $`J_1`$, $`J_2`$, and/or $`J_3`$ possesses a non-vanishing imaginary part. If (i) is not satisfied then (ii) is necessarily true, and the CP-violation is explicit. If (ii) is not satisfied, then (i) is necessarily true, and the Higgs sector is CP-conserving.
We provide two simple examples. First, ref. chinese considers a model in which $`m_{12}^2=\lambda _6=\lambda _7=0`$ and $`\lambda _5`$ is real and positive. Minimizing the scalar potential yields a purely imaginary $`v_2/v_1`$. Nevertheless, a simple relative phase redefinition of the two Higgs fields by $`\pi /2`$ yields a real basis with real vacuum expectation values. (In the new basis, $`\lambda _5^{}<0`$ and all other Higgs potential parameters are unmodified.) Hence, this model is CP-conserving.
Second, consider a Higgs potential that satisfies eq. (36), with $`\lambda _6`$ real, which was proposed in ref. rebelo . That is, all scalar potential parameters of this model are real, and the Higgs potential is explicitly CP-conserving. In this case, a minimum of the scalar potential exists where $`v_1=v_2`$ and the relative phase of the two vevs, $`\xi 0`$. That is, we may write $`\sqrt{2}\widehat{v}=(e^{i\xi /2},e^{i\xi /2})`$. Nevertheless, ref. rebelo proved that this model is CP-conserving. We may explicitly verify this assertion by performing a U(2) transformation given by eq. (4) with $`\psi =\xi /2`$, $`\chi =\pi /2`$ and $`\theta =\pi /4`$. We find that $`\lambda _5^{}=\lambda _5`$, $`m_{12}^{\mathrm{\hspace{0.17em}2}}=m_{12}^2\mathrm{sin}\xi `$, $`\lambda _6^{}=\lambda _7^{}=\lambda _6\mathrm{sin}\xi `$ are all real and $`\widehat{v}^{}=(1,0)`$. Thus, we have established a basis in which all scalar potential parameters and the vacuum expectation values are simultaneously real.
Of course, the absence of spontaneous CP-breaking in both examples can also be confirmed by checking that the invariants $`J_1`$, $`J_2`$ and $`J_3`$ are all real.
## VI Conclusions
The connection between the CP property of a general scalar potential and the parameters of the potential and vacuum expectation values of the Higgs fields is governed by two well-known theorems. The first, proven here as Theorem 1, states that the Higgs sector is explicitly CP-conserving if and only if there exists a real basis, that is choice of basis (in the Higgs โflavorโ space) in which all the scalar potential parameters are real. The second theorem, proven here as Theorem 3, states that the vacuum is CP-invariant, implying the absence of both explicit and spontaneous CP violation, if and only if there exists a real basis in which the Higgs vacuum expectation values are real. In this paper, we have established a simple procedure for determining whether or not a general 2HDM is explicitly CP-conserving by employing a set of four potentially complex basis-independent invariant combinations of the Higgs potential parameters. At least one of these invariants possesses a non-vanishing imaginary part if and only if no real basis exists.
The imaginary parts of the four complex basis-independent invariants that govern the explicit CP-violation properties of the 2HDM scalar potential are $`I_{Y3Z}`$ \[eq. (39)\], $`I_{2Y2Z}`$ \[eq. (41)\], $`I_{6Z}`$ \[eq. (47)\] and $`I_{3Y3Z}`$ \[eq. (48)\]. We have shown that a real basis exists, implying that the 2HDM potential is explicitly CP-conserving, if and only if $`I_{Y3Z}=I_{2Y2Z}=I_{6Z}=I_{3Y3Z}=0`$. We refer to these invariant imaginary parts as $`I`$-invariants.
Note that the above conditions are not sufficient to guarantee that the scalar sector conserves CP, since the minimization of the scalar potential may generate complex vacuum expectation values (vevs). As stated above, if the vevs possess a non-zero relative phase in all real basis choices, then the model spontaneously breaks CP. One can formulate basis independent conditions for spontaneous CP-violation. First, one must prove that the Higgs sector is explicitly CP-conserving (the corresponding invariant conditions have been given above). Spontaneous CP-violation depends on the properties of the Higgs field vevs, $`v_a`$, which can be combined with the Higgs potential parameters to construct additional invariant quantities. Such invariant conditions have been previously obtained in ref. lavoura , and are exhibited in section V. Combining the information from these two classes of invariant conditions, one can distinguish between explicit and spontaneous CP-violation in the 2HDM.
The phenomenological consequences of our invariants will be considered in a forthcoming paper. To apply the basis-independent technology to experimental studies, one would have to examine various CP-violating observables and express them in terms of our invariant quantities. The LHC would provide the first possible arena for such studies. However, the number of Higgs observables that could be extracted from LHC analyzes is limited. We anticipate that Higgs-mediated CP-violating effects are likely to be small, and their extraction will surely require precision measurements. A future high energy $`e^+e^{}`$ linear collider such as the ILC could provide the required luminosity and precision to begin a program of CP-violating Higgs phenomenology. We plan on examining possible CP-violating observables and determining their sensitivity to the $`I`$-invariants. This analysis will require a better understanding of the relation of the $`I`$-invariants to the mixing of CP-even/CP-odd neutral Higgs boson eigenstates.
Perhaps the most attractive 2HDM model is the one associated with the minimal supersymmetric extension of the Standard Model (MSSM) mssm . Indeed, the tree-level Higgs sector of the MSSM is CP-conserving. However, when loop-effects are included, supersymmetry breaking effects, which enter via the loops, can impart non-trivial phases to parameters of the effective 2HDM scalar potential Haber:1993an ; cpsusy .<sup>21</sup><sup>21</sup>21These phases would be directly related to phases of fundamental complex MSSM parameters such as the supersymmetric-conserving $`\mu `$-term, and the supersymmetry-breaking gaugino Majorana mass terms and matrix $`A`$-parameters. One can therefore express the $`I`$-invariants in terms of fundamental MSSM parameters. This may lead to relations among the four $`I`$-invariants introduced above, depending on the model of supersymmetry breaking.
Ultimately, if nature employs a 2HDM as an effective theory of electroweak symmetry breaking, it will be crucial to determine whether Higgs-mediated CP-violation exists and determine its structure. By devising experimental probes of the four $`I`$-invariants, we hope to provide a model-independent technique for elucidating the fundamental theory that is responsible for Higgs sector dynamics.
###### Acknowledgements.
We have greatly benefited from discussions with Sacha Davidson and her participation in the initial stages of this work. We are also grateful for the hospitality of the Aspen Center for Physics, where this project was initiated. This work was supported in part by the U.S. Department of Energy.
## Appendix A Existence of a Real Basis
In this appendix, we prove Theorem 1 that was quoted at the beginning of section III.
Theorem 1: The Higgs potential is explicitly CP-conserving if and only if a basis exists in which all Higgs potential parameters are real. Otherwise, CP is explicitly violated.
A basis in which all Higgs potential parameters are real will be called a real basis. In order to prove Theorem 1, one can either consider the most general CP transformation laws of the scalar fields or invoke the CPT theorem cpt and consider the most general scalar field transformation laws under time-reversal. Here we choose the latter procedure.<sup>22</sup><sup>22</sup>22In ref. branco , the CP transformation of the scalar fields in the real basis are used to prove that the scalar Lagrangian is CP-invariant. Following ref. Branco:1983tn , we note that the form for the action of the anti-unitary time-reversal operator $`๐ฏ`$ on a set of scalar field multiplets is given by
$$๐ฏ\mathrm{\Phi }_a(\stackrel{\mathbf{}}{๐},t)๐ฏ^1=e^{i\psi }(U_T)_{a\overline{b}}\mathrm{\Phi }_b(\stackrel{\mathbf{}}{๐},t),๐ฏ\mathrm{\Phi }_{\overline{a}}^{}(\stackrel{\mathbf{}}{๐},t)๐ฏ^1=\mathrm{\Phi }_{\overline{b}}^{}(\stackrel{\mathbf{}}{๐},t)(U_T^{})_{b\overline{a}}e^{i\psi }.$$
(61)
where $`U_T`$ is a symmetric unitary matrix that depends on the choice of basis. The arbitrary phase factor $`e^{i\psi }`$ corresponds to the freedom to make U(1)<sub>Y</sub> transformations.<sup>23</sup><sup>23</sup>23More generally, the time reversal operator is defined modulo SU(2)$`\times `$U(1)<sub>Y</sub> gauge transformations that leave the Lagrangian invariant (and hence do not modify the scalar potential parameters). To prove that $`U_T`$ is symmetric, we apply the time reversal operator twice and use the well known result that $`๐ฏ^2\mathrm{\Phi }_a(\stackrel{\mathbf{}}{๐},t)๐ฏ^2=\mathrm{\Phi }_a(\stackrel{\mathbf{}}{๐},t`$); that is, $`๐ฏ^2=1`$ when applied to a bosonic field carruthers . Applying this result to eq. (61) yields $`U_T^{}U_T=I`$, due to the anti-unitarity of $`๐ฏ`$. Since $`U_T`$ is unitary, it follows that $`U_T`$ must satisfy $`U_T^T=U_T`$. The (canonical) kinetic energy terms of the scalar field theory are automatically time-reversal invariant. It then follows that the scalar Lagrangian is time-reversal invariant if the scalar potential satisfies:<sup>24</sup><sup>24</sup>24We henceforth omit exhibiting the explicit dependence of the fields on the space-time coordinates.
$$๐ฏ๐ฑ(\mathrm{\Phi },\{p\})๐ฏ^1=๐ฑ(U_T\mathrm{\Phi },\{p^{}\})=๐ฑ(\mathrm{\Phi },\{p\}),$$
(62)
where $`\left\{p\right\}`$ represents the Higgs potential parameters appearing in $`๐ฑ`$, and the complex conjugated parameters $`\left\{p^{}\right\}`$ appear above due to the anti-unitarity of $`๐ฏ`$. If eq. (62) is satisfied, then the action is invariant under time-reversal transformations.
Suppose that a basis exists in which all the Higgs potential parameters are real. In this case, we may choose $`U_T=1`$, in which case eq. (62) is trivially satisfied. To complete the proof of Theorem 1, we must show that a basis exists in which all the Higgs potential parameters are real if eq. (62) is satisfied. First, we examine the quadratic part of the Higgs potential, which we can write in matrix notation as:
$$๐ฑ_2=\mathrm{\Phi }^{}Y\mathrm{\Phi },$$
(63)
where $`Y`$ is a hermitian matrix. Time reversal invariance of $`๐ฑ_2`$ requires
$$๐ฏ\mathrm{\Phi }^{}Y\mathrm{\Phi }๐ฏ^1=\mathrm{\Phi }^{}U_T^{}Y^{}U_T\mathrm{\Phi }=\mathrm{\Phi }^{}Y\mathrm{\Phi },$$
(64)
where we have used $`๐ฏY๐ฏ^1=Y^{}`$. Eq. (64) implies that
$$U_T^{}Y^{}U_T=Y.$$
(65)
As shown in Appendix B, since $`U_T`$ is unitary and symmetric, we can write
$$U_T=V^TV,$$
(66)
where $`V`$ is unitary (but not necessarily symmetric). As a result, eq. (65) will be true if
$$V^{}V^{}Y^{}V^TV=Y,$$
(67)
which can be converted to
$$(VYV^{})^{}=VYV^{}.$$
(68)
That is $`Y^{}VYV^{}`$ is real. But, $`Y^{}`$ is simply $`Y`$ in the new basis $`\mathrm{\Phi }^{}=V\mathrm{\Phi }`$. Thus, there exists a basis in which the parameters of $`๐ฑ_2`$ are real.
A similar computation can be performed for the rest of the terms appearing in the scalar potential. In particular, if we write the quartic part of the Higgs potential as:
$$๐ฑ_4=\frac{1}{2}Z_{a\overline{b}c\overline{d}}(\mathrm{\Phi }_{\overline{a}}^{}\mathrm{\Phi }_b)(\mathrm{\Phi }_{\overline{c}}^{}\mathrm{\Phi }_d),$$
(69)
then the analog of eq. (65) is
$$(U_T^{})_{e\overline{a}}(U_T)_{b\overline{f}}(U_T^{})_{g\overline{c}}(U_T)_{d\overline{h}}Z_{a\overline{b}c\overline{d}}^{}=Z_{e\overline{f}g\overline{h}}.$$
(70)
We again apply eq. (66) and conclude that
$$[V_{p\overline{a}}V_{b\overline{q}}^{}V_{r\overline{c}}V_{d\overline{s}}^{}Z_{a\overline{b}c\overline{d}}]^{}=V_{p\overline{e}}V_{f\overline{q}}^{}V_{r\overline{g}}V_{h\overline{s}}^{}Z_{e\overline{f}g\overline{h}}.$$
(71)
That is, the unitary transformation $`V`$ produces the basis in which all the Higgs potential parameters are real.
Conversely, if no basis exists in which the Higgs potential parameters are real, then no unitary matrix $`V`$ exists such that eqs. (68) and (71) are simultaneously satisfied. Following the above proof in the backward direction, one can conclude that no choice of a unitary symmetric matrix $`U_T`$ exists that satisfies eq. (62).
In some cases (see below), more than one suitable time-reversal operator exists. Any one of these operators can be used to demonstrate that the Higgs potential is explicitly CP-invariant. Nevertheless, in order to ascertain that the Higgs sector is invariant under CP, it is necessary to verify that the vacuum is also CP-invariant (equivalently time-reversal invariant). In particular, the vacuum may select out a unique time-reversal operator, as shown in Appendix F. (If the vacuum is non-invariant with respect to all possible candidate time-reversal operators, then time-reversal invariance is spontaneously broken.) Thus, it is important to consider the possible non-uniqueness in the definition of $`๐ฏ`$ given in eq. (61).
For an explicitly CP-conserving Higgs potential, a real basis must exist. However, the real basis is not unique. In particular, given a real $`\mathrm{\Phi }^{}`$-basis, there exists an O(2)$`\times ๐`$ subgroup of U(2) consisting of $`2\times 2`$ unitary matrices $`W_{a\overline{b}}`$ such that the scalar potential parameters remain real under $`\mathrm{\Phi }_a^{}\mathrm{\Phi }_a^{\prime \prime }=W_{a\overline{b}}\mathrm{\Phi }_b^{}`$. Here, $`๐`$ is the maximal discrete subgroup of U(2) that is a symmetry of the Higgs Lagrangian. In addition, one is free to make U(1)<sub>Y</sub> phase rotations, which simply reflects the fact that $`U_T`$ is only defined up to an overall phase. If $`๐`$ is trivial, then $`W`$ is an orthogonal transformation and $`U_T=I`$ (up to an overall phase) in any real basis. If $`๐`$ is nontrivial, then $`W^TWe^{i\eta }I`$ (for any phase choice $`\eta `$), in which case the choice of $`U_T`$ in the definition of the time reversal operator is not unique (modulo gauge transformations).
To amplify these remarks, we suppose that in the original $`\mathrm{\Phi }`$-basis another anti-unitary operator $`\stackrel{~}{๐ฏ}`$ exists that is a potential candidate for the time-reversal operator. In particular, suppose that there exists a symmetric unitary matrix $`\stackrel{~}{U}_Te^{i\eta }U_T`$ such that<sup>25</sup><sup>25</sup>25That is, if $`\stackrel{~}{U_T}U_T`$ in the $`\mathrm{\Phi }`$-basis, for any choices of the phases $`\stackrel{~}{\psi }`$ and $`\psi `$, then $`\stackrel{~}{๐ฏ}`$ and $`๐ฏ`$ are distinct and equally valid choices for the time reversal operator.
$$\stackrel{~}{๐ฏ}\mathrm{\Phi }_a(\stackrel{\mathbf{}}{๐},t)\stackrel{~}{๐ฏ}^1=e^{i\stackrel{~}{\psi }}(\stackrel{~}{U}_T)_{a\overline{b}}\mathrm{\Phi }_b(\stackrel{\mathbf{}}{๐},t),๐ฑ(\mathrm{\Phi },\{p\})=๐ฑ(\stackrel{~}{U}_T\mathrm{\Phi },\{p^{}\}).$$
(72)
Then, the analysis above implies that there exists a unitary matrix $`\stackrel{~}{V}`$ such that $`\stackrel{~}{U}=\stackrel{~}{V}^T\stackrel{~}{V}`$, and $`\mathrm{\Phi }^{\prime \prime }=\stackrel{~}{V}\mathrm{\Phi }`$ is also a real basis. In this case, the real $`\mathrm{\Phi }^{}`$-basis and the real $`\mathrm{\Phi }^{\prime \prime }`$-basis are related by $`\mathrm{\Phi }^{\prime \prime }=W\mathrm{\Phi }^{}`$ where $`W=\stackrel{~}{V}V^1`$. It follows that $`WW^T=\stackrel{~}{V}U^1\stackrel{~}{V}^Te^{i\eta }I`$ (for any phase choice $`\eta `$). Thus, the existence of $`\stackrel{~}{๐ฏ}๐ฏ`$ implies that the discrete group $`๐`$ is nontrivial. Likewise, one can show that $`W^TW=[V^1]^T\stackrel{~}{U}V^1e^{i\eta }I`$.
Given $`U_T`$ in the $`\mathrm{\Phi }`$-basis, we may determine the form of this matrix in any real basis. For example, inserting $`\mathrm{\Phi }^{}=V\mathrm{\Phi }`$ into eq. (61) and making use of eq. (66), we find
$$๐ฏ\mathrm{\Phi }_a^{}(\stackrel{\mathbf{}}{๐},t)๐ฏ^1=e^{i\psi }\mathrm{\Phi }_a^{}(\stackrel{\mathbf{}}{๐},t).$$
(73)
That is, in the $`\mathrm{\Phi }^{}`$-basis, $`U_T^{}=I`$. Eq. (62) then implies that this is a real basis. Now, let us transform to the real basis $`\mathrm{\Phi }^{\prime \prime }=W\mathrm{\Phi }^{}`$. A similar computation yields
$$๐ฏ\mathrm{\Phi }_a^{\prime \prime }(\stackrel{\mathbf{}}{๐},t)๐ฏ^1=e^{i\psi }(WW^T)_{a\overline{b}}^1\mathrm{\Phi }_b^{\prime \prime }(\stackrel{\mathbf{}}{๐},t).$$
(74)
where $`U_T^{\prime \prime }=(WW^T)^1I`$ in the $`\mathrm{\Phi }^{\prime \prime }`$-basis. Similarly, if we identify $`\stackrel{~}{๐ฏ}`$ as the time-reversal operator, we find that $`\stackrel{~}{U}_T^{}=W^TWI`$ and $`\stackrel{~}{U}_T^{\prime \prime }=I`$. We may assemble all possible real bases into classes. Each class is in one-to-one correspondence with the elements of the discrete group $`๐`$. In the class of real bases associated with the identity element of $`๐`$, the corresponding $`U_T=I`$. In all other classes of real bases, the corresponding $`U_TI`$.
If $`๐`$ is trivial, so that $`W`$ is an orthogonal transformation \[up to an overall phase that can be absorbed, e.g., into the multiplicative phase factor in eq. (74)\], then $`U_T=I`$ in any real basis. In this case, the definition of the time reversal operator $`๐ฏ`$ is unique (modulo gauge transformations).
Finally, we note that the existence of a non-trivial discrete subgroup $`๐`$ imposes strong constraints on the parameters of the Higgs potential. Consider a real $`\mathrm{\Phi }^{}`$-basis and a real $`\mathrm{\Phi }^{\prime \prime }`$-basis related by $`\mathrm{\Phi }^{\prime \prime }=W\mathrm{\Phi }^{}`$. It then follows that $`Y^{\prime \prime }=WY^{}W^{}`$. By assumption, $`Y^{}`$ and $`Y^{\prime \prime }`$ are real. A short computation then yields the vanishing of the following commutators:
$$[Y^{},W^TW]=[Y^{\prime \prime },WW^T]=0.$$
(75)
A similar constraint arises from the requirement that both $`Z^{}`$ and $`Z^{\prime \prime }`$ are real. Using these results, it is straightforward to verify that eq. (72) is satisfied for $`\stackrel{~}{U}_T^{}=W^TW`$ in the $`\mathrm{\Phi }^{}`$-basis and eq. (62) is satisfied for $`U_T^{\prime \prime }=(WW^T)^1`$ in the $`\mathrm{\Phi }^{\prime \prime }`$-basis.
## Appendix B A proof of a result from matrix analysis
In the proof of Theorems 1 and 4, the following lemma is required:
Lemma 1: A complex $`n\times n`$ matrix $`U`$ is unitary and symmetric if and only if there is a complex $`n\times n`$ unitary matrix $`V`$ such that $`U=V^TV`$.
This result is given as problem 17 on p. 215 of ref. horn . Here, we give an explicit proof. Clearly if $`V`$ is unitary it follows that $`U`$ is unitary and symmetric. Thus, we focus on the proof that given $`U`$, the unitary matrix $`V`$ exists. Lemma 1 is a special case of the Takagi factorization of a complex symmetric matrix (see pp. 204โ206 of ref. horn ). Namely, for any complex symmetric matrix $`M`$, there exists a unitary matrix $`V`$ such that $`M=V^TDV`$, where $`D`$ is a real nonnegative diagonal matrix whose elements are given by the nonnegative square roots of the eigenvalues of $`MM^{}`$.<sup>26</sup><sup>26</sup>26The Takagi factorization of a complex symmetric matrix is the basis for the mass diagonalization of a general Majorana fermion mass matrix mohapatra . Applying the Takagi factorization to a unitary matrix $`M=U`$ (i.e., $`UU^{}=I`$), it immediately follows that $`D=I`$. Hence, $`U=V^TV`$ for some unitary matrix $`V`$.
The matrix $`V`$ is not unique. In particular, if $`U=V^TV`$ then $`U=W^TW`$, where the unitary matrix $`W=KV`$ and $`K`$ is an arbitrary orthogonal matrix. However, the proof of Theorem 4 simply requires the existence of $`V`$, which has been proven above.
## Appendix C Does a basis exist in which all the $`\lambda _i`$ are real?
Lemma 2: If the parameters of the 2HDM satisfy the relations: $`\lambda _1=\lambda _2`$ and $`\lambda _7=\lambda _6`$, then one can always transform to a new basis in which $`\lambda _5^{}`$ and $`\lambda _7^{}=\lambda _6^{}`$ are all real.
We begin with eq. (12) and eq. (13) and require that the imaginary parts of $`\lambda _5^{}`$ and $`\lambda _6^{}`$ are zero. We assume that $`\lambda _60`$ (if $`\lambda _7=\lambda _6=0`$, it is trivial to transform to a basis where $`\lambda _5`$ is real by rephasing one of the scalar fields). Moreover, without loss of generality, we may assume that $`\lambda _6`$ is real by rephasing one of the scalar fields appropriately.<sup>27</sup><sup>27</sup>27Since $`\lambda _1=\lambda _2`$ and $`\lambda _7=\lambda _6`$, it follows that $`\lambda _7^{}=\lambda _6^{}`$, and further consideration of $`\lambda _7`$ is unnecessary. If $`\lambda _5`$ is also real after the rephasing, we are done. If not, we write $`\lambda _5|\lambda _5|e^{i\theta _5}`$ and obtain
$`\mathrm{Im}\lambda _5^{}`$ $`=`$ $`\frac{1}{2}f_b\mathrm{sin}2\chi +f_a\mathrm{cos}2\chi ,`$ (76)
$`\mathrm{Im}\lambda _6^{}`$ $`=`$ $`\frac{1}{4}f_d\mathrm{sin}\chi +\frac{1}{2}f_c\mathrm{cos}\chi ,`$ (77)
where
$`f_a`$ $`=`$ $`|\lambda _5|c_{2\theta }\mathrm{sin}(\theta _5+2\xi )2\lambda _6s_{2\theta }\mathrm{sin}\xi ,`$ (78)
$`f_b`$ $`=`$ $`(\lambda _1\lambda _3\lambda _4)s_{2\theta }^2+|\lambda _5|(2s_{2\theta }^2)\mathrm{cos}(\theta _5+2\xi )2\lambda _6s_{4\theta }\mathrm{cos}\xi ,`$ (79)
$`f_c`$ $`=`$ $`|\lambda _5|s_{2\theta }\mathrm{sin}(\theta _5+2\xi )+2\lambda _6c_{2\theta }\mathrm{sin}\xi ,`$ (80)
$`f_d`$ $`=`$ $`[|\lambda _5|\mathrm{cos}(\theta _5+2\xi )\lambda _1+\lambda _3+\lambda _4)]s_{4\theta }+4\lambda _6c_{4\theta }\mathrm{cos}\xi .`$ (81)
As before, we abbreviate $`s_{4\theta }\mathrm{sin}4\theta `$, $`c_{4\theta }\mathrm{cos}4\theta `$, etc. We proceed to solve $`\mathrm{Im}\lambda _5^{}=0`$, which yields an equation for $`\mathrm{cot}2\chi `$, and $`\mathrm{Im}\lambda _6^{}=0`$, which yields an equation for $`\mathrm{cot}\chi `$:
$`\mathrm{cot}2\chi `$ $`=`$ $`{\displaystyle \frac{f_b}{2f_a}}`$ (82)
$`\mathrm{cot}\chi `$ $`=`$ $`{\displaystyle \frac{f_d}{2f_c}},`$ (83)
Under the assumption that $`f_a0`$ and $`f_c0`$, we can eliminate $`\chi `$ by employing the well known identity
$$\mathrm{cot}2\chi =\frac{\mathrm{cot}^2\chi 1}{2\mathrm{cot}\chi },$$
(84)
which leads to the following result:
$$G(\theta ,\xi )f_a(f_d^24f_c^2)2f_bf_cf_d=0.$$
(85)
We wish to prove that there exists at least one $`\theta `$ and $`\xi `$ that solves eq. (85). From any such solution, we may compute $`\chi `$ from eqs. (82) and (83). This would then provide the elements of the U(2) transformation matrix that yields the basis in which all the $`\lambda _i`$ are real.
To prove that a solution to $`G(\theta ,\xi )=0`$ exists, we note that
$`f_a(\theta =0,\xi )`$ $`=`$ $`f_a(\theta =\pi /2,\xi )=|\lambda _5|\mathrm{sin}(\theta _5+2\xi ),`$ (86)
$`f_b(\theta =0,\xi )`$ $`=`$ $`+f_b(\theta =\pi /2,\xi )=2|\lambda _5|\mathrm{cos}(\theta _5+2\xi ),`$ (87)
$`f_c(\theta =0,\xi )`$ $`=`$ $`f_c(\theta =\pi /2,\xi )=2\lambda _6\mathrm{sin}\xi ,`$ (88)
$`f_d(\theta =0,\xi )`$ $`=`$ $`+f_d(\theta =\pi /2,\xi )=4\lambda _6\mathrm{cos}\xi ,`$ (89)
from which it follows that
$$G(0,\xi )=G(\pi /2,\xi )=16\lambda _6^2|\lambda _5|\mathrm{sin}\theta _5.$$
(90)
This means that $`G`$ will have at least one sign change as a function of $`\theta `$. Hence, for any value of $`\xi `$ there exists a value of $`\theta `$ for which $`G(\theta ,\xi )=0`$. Thus, we have proved the existence of a U(2) transformation that results in a basis in which all the $`\lambda _i`$ are real.
The assumption above that $`f_a0`$ and $`f_c0`$ for values of $`\theta `$ and $`\xi `$ at which $`G(\theta ,\xi )=0`$ is not strictly necessary. For example, if $`f_a=0`$ (but $`f_b0`$), then one can rewrite eq. (82) in terms of $`\mathrm{tan}2\chi `$. We then end up again with eq. (85). The only special cases that need be considered are: (i) $`f_a=f_b=0`$ and (ii) $`f_c=f_d=0`$. If (i) and (ii) both hold, then we immediately conclude that $`\lambda _5^{}`$ and $`\lambda _6^{}`$ are real and we are finished. If only (i) \[only (ii)\] holds, then we simply use eq. (83) \[eq. (82)\] to determine $`\chi `$, and we are finished.
We have used Lemma 2 in the proof of Theorem 2 (see Section III). It is instructive to examine the necessity of the condition of $`\lambda _1=\lambda _2`$ in the proof of Lemma 2. For this reason, we prove a second lemma.
Lemma 3: If $`\lambda _1\lambda _2`$ and $`\mathrm{Im}(\lambda _5^{}\lambda _6^2)0`$ in a basis where $`\lambda _7=\lambda _60`$, then it is impossible to transform to a basis in which $`\lambda _5^{}`$, $`\lambda _6^{}`$ and $`\lambda _7^{}`$ are all real.
The proof of Lemma 3 is trivial using invariants. Namely, in a basis where $`\lambda _7=\lambda _60`$, we use eq. (30) to conclude that $`I_{6Z}0`$. Hence in this case, there is no basis in which all the $`\lambda _i`$ are real.
Even without invariants, it is not difficult to show that no basis exists in which all the $`\lambda _i`$ are real. We first rephase one of the scalar fields such that the resulting value of $`\lambda _6`$ is real. In this basis, $`\lambda _5|\lambda _5|e^{i\theta _5}`$, where $`\theta _50(\mathrm{mod}\pi )`$. We then use eqs. (13) and (14) in the case of $`\lambda _7=\lambda _6`$ to obtain
$$\mathrm{Im}(\lambda _6^{}+\lambda _7^{})=\frac{1}{2}\mathrm{sin}\chi s_{2\theta }(\lambda _1\lambda _2).$$
(91)
Since $`\lambda _1\lambda _2`$, it follows that $`\mathrm{Im}\lambda _6^{}=\mathrm{Im}\lambda _7^{}=0`$ implies that either $`\mathrm{sin}2\theta =0`$ or $`\mathrm{sin}\chi =0`$.
If $`\mathrm{sin}2\theta =0`$, then eqs. (12) and (13) yield
$`\lambda _5^{}e^{2i\chi }`$ $`=`$ $`|\lambda _5|e^{\pm i(2\xi +\theta )},`$ (92)
$`\lambda _6^{}e^{i\chi }`$ $`=`$ $`|\lambda _5|e^{\pm i\theta },`$ (93)
where the choice of sign above corresponds to the sign of $`\mathrm{cos}2\theta `$. Thus,
$$\frac{\lambda _5^{}}{\lambda _6^{\mathrm{\hspace{0.17em}2}}}=\frac{|\lambda _5|}{\lambda _6^2}e^{\pm i\theta _5},$$
(94)
and we see that no basis exists in which $`\lambda _5^{}`$ and $`\lambda _6^{}`$ are simultaneously real.
Next, suppose that $`\mathrm{sin}2\theta 0`$ and $`\mathrm{sin}\chi =0`$. Eqs. (12) and (13) then yield
$`\mathrm{Im}(\lambda _5^{})`$ $`=`$ $`|\lambda _5|c_{2\theta }\mathrm{sin}(\theta _5+2\xi )2\lambda _6s_{2\theta }\mathrm{sin}\xi ,`$ (95)
$`\mathrm{Im}(\lambda _6^{})`$ $`=`$ $`\frac{1}{2}|\lambda _5|s_{2\theta }\mathrm{sin}(\theta _5+2\xi )+\lambda _6c_{2\theta }\mathrm{sin}\xi .`$ (96)
If $`\mathrm{sin}\xi =0`$, then eq. (96) reduces to $`\mathrm{Im}(\lambda _6^{})=\frac{1}{2}|\lambda _5|\mathrm{sin}2\theta \mathrm{sin}\theta _50`$. Thus, we must have $`\mathrm{sin}\xi 0`$ if $`\lambda _6^{}`$ is real. Then, using eqs. (95) and (96) to set $`\mathrm{Im}(\lambda _5^{})=\mathrm{Im}(\lambda _6^{})=0`$ yields
$$\mathrm{tan}2\theta =\mathrm{cot}2\theta =\frac{|\lambda _5|\mathrm{sin}(\theta _5+2\xi )}{2\lambda _6\mathrm{sin}\xi }.$$
(97)
However, eq. (97) implies that $`\mathrm{tan}^22\theta =1`$, which is impossible. Once again, we conclude that no basis exists in which $`\lambda _5^{}`$ and $`\lambda _6^{}`$ are simultaneously real. The proof of Lemma 3 is now complete.
## Appendix D Proof of Lemma 4
Consider the special isolated point of the Higgs parameter space in which $`\lambda _1=\lambda _2`$ and $`\lambda _7=\lambda _6`$. By Lemma 2, we may assume without loss of generality that all the $`\lambda _i`$ are real. Thus, $`Y_{12}`$ remains as the only potentially complex parameter. Lemma 4 provides the conditions under which it is possible to find a new basis in which all the Higgs potential parameters are real.
Lemma 4: If the parameters of the 2HDM satisfy the relations: $`\lambda _1=\lambda _2`$ and $`\lambda _7=\lambda _6`$, and the basis is chosen such that all the $`\lambda _i`$ are real and $`Y_{12}`$ is complex, then there exists a new basis in which all the Higgs potential parameters are real if and only if (at least) one of the following two conditions is satisfied:
$$\lambda _5^2+\lambda _5(\lambda _1\lambda _3\lambda _4)2\lambda _6^2=0,$$
(98)
and/or
$$4\lambda _6(\mathrm{Re}Y_{12})^2(\lambda _3+\lambda _4+\lambda _5\lambda _1)(Y_{11}Y_{22})\mathrm{Re}Y_{12}\lambda _6(Y_{11}Y_{22})^2=0.$$
(99)
It is easy to prove that if neither eq. (98) nor eq. (99) is satisfied, then there is no basis in which all Higgs potential parameters are real. The latter conclusion follows directly from $`I_{3Y3Z}0`$, which is a consequence of eq. (III). Thus, we focus on the inverse statement: if either eq. (98) or eq. (99) is satisfied, then there exists a basis in which all the Higgs potential parameters are real.
Suppose that eq. (98) is satisfied, under the assumption that all the $`\lambda _i`$ are real (for $`\lambda _1=\lambda _2`$ and $`\lambda _7=\lambda _6`$) and $`Y_{12}`$ is complex. We search for a U(2) transformation to a new basis in which the $`\lambda _i^{}`$ and $`Y_{12}^{}`$ are real. It will be sufficient to consider solutions with $`\chi =\pi /2`$. At this point, we assume that $`\lambda _60`$ (we shall treat the case of $`\lambda _6=0`$ separately). Then, we demand that $`\theta `$ is the solution (as a function of $`\xi `$) of the following equation:
$$\lambda _6\mathrm{sin}2\theta =\lambda _5\mathrm{cos}2\theta \mathrm{cos}\xi .$$
(100)
Using eq. (76) with $`\chi =\pi /2`$ and real $`\lambda _5`$, it is easy to check that eq. (100) implies that $`\mathrm{Im}\lambda _5^{}=0`$. Next, using eq. (77) with $`\chi =\pi /2`$ and real $`\lambda _5`$ yields:
$$\mathrm{Im}\lambda _6^{}=\frac{1}{4}(\lambda _5\mathrm{cos}2\xi \lambda _1+\lambda _3+\lambda _4)\mathrm{sin}4\theta \lambda _6\mathrm{cos}4\theta \mathrm{cos}\xi .$$
(101)
Using eq. (100), we obtain
$`\mathrm{sin}4\theta `$ $`=`$ $`2\mathrm{sin}2\theta \mathrm{cos}2\theta ={\displaystyle \frac{2\lambda _5\mathrm{cos}^22\theta \mathrm{cos}\xi }{\lambda _6}},`$ (102)
$`\mathrm{cos}4\theta `$ $`=`$ $`\mathrm{cos}^22\theta \mathrm{sin}^22\theta =\mathrm{cos}^22\theta \left(1{\displaystyle \frac{\lambda _5^2\mathrm{cos}^2\xi }{\lambda _6^2}}\right).`$ (103)
Inserting these results into eq. (101) and simplifying the resulting expression yields
$$\mathrm{Im}\lambda _6^{}=\frac{\mathrm{cos}^22\theta \mathrm{cos}\xi }{2\lambda _6}\left[\lambda _5^2+\lambda _5(\lambda _1\lambda _3\lambda _4)2\lambda _6^2\right].$$
(104)
Thus, using eq. (98), we see that $`\mathrm{Im}\lambda _6^{}=0`$ for any value of $`\xi `$. We now choose $`\xi `$ in order that $`\mathrm{Im}Y_{12}^{}=0`$. Using eq. (7) with $`\chi =\pi /2`$ and $`Y_{12}|Y_{12}|e^{i\theta _{12}}`$, we find
$$2|Y_{12}|\mathrm{cos}2\theta \mathrm{cos}(\theta _{12}+\xi )=(Y_{11}Y_{22})\mathrm{sin}2\theta .$$
(105)
Using eq. (100) to eliminate $`\theta `$, we end up with
$$\mathrm{tan}\xi =\mathrm{cot}\theta _{12}\frac{\lambda _5(Y_{11}Y_{22})}{2\lambda _6|Y_{12}|\mathrm{sin}\theta _{12}}.$$
(106)
Finally, we treat the case of $`\lambda _6=0`$. We may assume that $`\lambda _50`$ (otherwise, a simple rephasing of one of the Higgs fields is sufficient to yield a real $`Y_{12}`$). In this case, we choose $`\chi =\xi =\pi /2`$. Then, $`\mathrm{Im}\lambda _5^{}=0`$ is satisfied \[see eq. (100)\] for arbitrary $`\theta `$. Inserting $`\xi =\pi /2`$ into eq. (101) yields
$$\mathrm{Im}\lambda _6^{}=\frac{1}{4}(\lambda _5+\lambda _1\lambda _3\lambda _4)\mathrm{sin}4\theta =0,$$
(107)
after using eq. (98) with $`\lambda _6=0`$ and $`\lambda _50`$. We now choose $`\theta `$ in order that $`\mathrm{Im}Y_{12}^{}=0`$. After putting $`\xi =\pi /2`$ in eq. (105), the end result is
$$\mathrm{cot}2\theta =\frac{Y_{22}Y_{11}}{2|Y_{12}|\mathrm{sin}\theta _{12}}.$$
(108)
To summarize, if eq. (98) is satisfied, we have exhibited a U(2) transformation \[eq. (4) with $`\chi =\pi /2`$, $`\theta `$ given by the solution to eq. (100) and $`\xi `$ given by eq. (106) if $`\lambda _60`$, and $`\chi =\xi =\pi /2`$ and $`\theta `$ given by the solution to eq. (108) if $`\lambda _6=0`$\] such that all Higgs potential parameters are real in the transformed basis.<sup>28</sup><sup>28</sup>28Other U(2) transformations with $`\chi \pi /2`$ can also produce a basis where all Higgs potential parameters are real. For example, a numerical analysis suggests that if $`\chi 0(\mathrm{mod}\pi )`$ and $`\lambda _60`$, then one can choose $`\theta `$ as a function of $`\xi `$ such that $`\mathrm{Im}\lambda _5^{}=0`$. Using this choice for $`\theta `$, one again finds that $`\mathrm{Im}\lambda _6^{}=0`$ as a consequence of eq. (98), independently of the value of $`\xi `$. Finally, $`\xi `$ can be chosen to yield $`\mathrm{Im}Y_{12}^{}=0`$. Of course, only one solution for $`(\chi ,\theta ,\xi )`$ must be exhibited to prove the validity of Lemma 4.
Next, suppose that eq. (99) is satisfied, under the assumption that all the $`\lambda _i`$ are real (for $`\lambda _1=\lambda _2`$ and $`\lambda _7=\lambda _6`$) and $`Y_{12}`$ is complex. We again search for a U(2) transformation to a new basis in which the $`\lambda _i`$ are still real and $`Y_{12}`$ is real. In this case, we choose $`\chi =\pi /2`$ and $`\xi =\pi `$. For this choice, $`\mathrm{Im}\lambda _5^{}=0`$ is automatic (independently of the value of $`\theta `$). Again, we first assume that $`\lambda _60`$ (the case of $`\lambda _6=0`$ is treated separately). Then, the constraints $`\mathrm{Im}Y_{12}^{}=0`$ \[eq. (105)\] and $`\mathrm{Im}\lambda _6^{}=0`$ \[eq. (101)\] reduce to
$`\mathrm{cot}2\theta `$ $`=`$ $`{\displaystyle \frac{Y_{22}Y_{11}}{2|Y_{12}|\mathrm{cos}\theta _{12}}},`$ (109)
$`\mathrm{cot}4\theta `$ $`=`$ $`{\displaystyle \frac{\lambda _5\lambda _1+\lambda _3+\lambda _4}{4\lambda _6}},`$ (110)
respectively. Using the double-angle formula analogous to eq. (84), we may combine eqs. (109) and (110) to yield the following constraint:
$$4\lambda _6|Y_{12}|^2\mathrm{cos}^2\theta _{12}+(\lambda _5\lambda _1+\lambda _3+\lambda _4)(Y_{22}Y_{11})|Y_{12}|\mathrm{cos}\theta _{12}\lambda _6(Y_{22}Y_{11})^2=0,$$
(111)
which is identical to eq. (99), which is assumed to be satisfied. Thus, eqs. (109) and (110) are consistent and provide a solution for $`\theta `$.
Finally, we examine the case of $`\lambda _6=0`$. In this case, eq. (99) reduces to:
$$(\lambda _1\lambda _3\lambda _4\lambda _5)(Y_{11}Y_{22})\mathrm{cos}\theta _{12}=0.$$
(112)
The case of $`\lambda _1\lambda _3\lambda _4\lambda _5=0`$ (with $`\lambda _6=0`$) is equivalent to eq. (98) and has already been treated. Thus, it is sufficient to examine only the cases of $`Y_{11}=Y_{22}`$ and $`\mathrm{cos}\theta _{12}=0`$. In both cases, we may choose $`\chi =\pi /2`$ and $`\xi =\pi `$ as before. Then, it is easy to check that if $`\mathrm{cos}2\theta =0`$ in the case of $`Y_{11}=Y_{22}`$ and $`\mathrm{sin}2\theta =0`$ in the case of $`\mathrm{cos}\theta _{12}=0`$, the U(2) transformation yields $`\mathrm{Im}\lambda _5^{}=\mathrm{Im}Y_{12}^{}=0`$.
To summarize, if eq. (99) is satisfied, we have exhibited a U(2) transformation \[e.g., eq. (4) with $`\chi =\pi /2`$, $`\xi =\pi `$ and $`\theta `$ given by the solution to eq. (109) if $`\lambda _60`$\] such that all Higgs potential parameters are real in the transformed basis.
Thus, we have explicitly constructed a U(2) transformation that renders all Higgs potential parameters real if either eq. (98) or eq. (99) is satisfied. Consequently, $`I_{3Y3Z}=0`$, and it follows that if $`\lambda _1=\lambda _2`$ and $`\lambda _7=\lambda _6`$, then the condition $`I_{3Y3Z}=0`$ is the necessary and sufficient condition for an explicitly CP-conserving Higgs potential. This concludes the proof of Lemma 4.
## Appendix E All Cubic Invariants are Real
In this appendix, we examine invariants constructed from the $`Y_{a\overline{b}}`$ and $`Z_{a\overline{b}c\overline{d}}`$. We show that all invariants that are at most cubic in the $`Z`$โs and independent of $`Y`$ are real. Similarly, we demonstrate that invariants that are linear in $`Y`$ and at most quadratic in the $`Z`$โs are real. Finally, we prove that invariants that are linear in $`Z`$ and quadratic in the $`Y`$โs are real.
First, we introduce some notation. We consider all possible non-trivial<sup>29</sup><sup>29</sup>29That is, we omit tensors that can be expressed as a product of a scalar quantity times $`Z_{a\overline{b}}^{(m)}`$, $`m=1,2`$ (e.g., $`Z_{c\overline{c}d\overline{d}}Z_{e\overline{e}a\overline{b}}[\mathrm{Tr}Z^{(2)}]Z_{a\overline{b}}^{(2)}`$.) second-rank tensors that are quadratic in the $`Z`$โs. Using the symmetry properties of the $`Z`$โs, we find six tensors of this kind:
$`Z_{c\overline{d}}^{(11)}Z_{a\overline{b}}^{(1)}Z_{b\overline{a}c\overline{d}},Z_{c\overline{d}}^{(12)}Z_{a\overline{b}}^{(1)}Z_{b\overline{d}c\overline{a}},`$ (113)
$`Z_{c\overline{d}}^{(21)}Z_{a\overline{b}}^{(2)}Z_{b\overline{a}c\overline{d}},Z_{c\overline{d}}^{(22)}Z_{a\overline{b}}^{(2)}Z_{b\overline{d}c\overline{a}},`$ (114)
$`Z_{c\overline{d}}^{(31)}Z_{a\overline{b}e\overline{d}}Z_{b\overline{a}c\overline{e}},Z_{c\overline{d}}^{(32)}Z_{a\overline{b}e\overline{d}}Z_{c\overline{a}b\overline{e}},`$ (115)
where $`Z^{(1)}`$ and $`Z^{(2)}`$ are defined in eq. (19). A quick computation shows that the $`Z^{(m)}`$ ($`m=1,2`$) and the $`Z^{(pn)}`$ ($`p=1,2,3`$ and $`n=1,2`$) are hermitian; that is,
$$Z_{a\overline{b}}^{(m)}=[Z_{b\overline{a}}^{(m)}]^{},Z_{a\overline{b}}^{(pn)}=[Z_{b\overline{a}}^{(pn)}]^{}.$$
(116)
Next, we consider all possible non-trivial<sup>30</sup><sup>30</sup>30Again, we omit those tensors that are products of simpler tensors (e.g., $`Z_{a\overline{b}b\overline{d}}Z_{c\overline{c}e\overline{f}}Z_{a\overline{d}}^{(1)}Z_{e\overline{f}}^{(2)}`$). fourth-rank tensors that are quadratic in the $`Z`$โs. These fall into a number of different classes. First, we have:
$`Z_{a\overline{b}c\overline{d}}^{(1)}Z_{a\overline{b}e\overline{f}}Z_{c\overline{d}f\overline{e}},Z_{a\overline{b}c\overline{d}}^{(4)}Z_{a\overline{d}f\overline{e}}Z_{c\overline{b}e\overline{f}},`$ (117)
$`Z_{a\overline{b}c\overline{d}}^{(2)}Z_{a\overline{f}e\overline{d}}Z_{f\overline{b}c\overline{e}},Z_{a\overline{b}c\overline{d}}^{(5)}Z_{a\overline{e}f\overline{b}}Z_{c\overline{f}e\overline{d}},`$ (118)
$`Z_{a\overline{b}c\overline{d}}^{(3)}Z_{a\overline{f}c\overline{e}}Z_{f\overline{b}e\overline{d}},Z_{a\overline{b}c\overline{d}}^{(6)}Z_{a\overline{f}c\overline{e}}Z_{e\overline{b}f\overline{d}}.`$ (119)
These fourth rank tensors possess the same symmetry and hermiticity properties as $`Z_{a\overline{b}c\overline{d}}`$, that is:
$$Z_{a\overline{b}c\overline{d}}^{(n)}=Z_{c\overline{d}a\overline{b}}^{(n)},[Z_{a\overline{b}c\overline{d}}^{(n)}]^{}=Z_{b\overline{a}d\overline{c}}^{(n)}.$$
(120)
Note that $`Z_{a\overline{b}c\overline{d}}^{(n+3)}Z_{c\overline{b}a\overline{d}}^{(n)}`$ for $`n=1,2,3`$. The second class of rank-four tensors consists of:
$`\stackrel{~}{Z}_{a\overline{b}c\overline{d}}^{(1)}Z_{a\overline{b}e\overline{f}}Z_{c\overline{e}f\overline{d}},\stackrel{~}{Z}_{a\overline{b}c\overline{d}}^{(3)}Z_{a\overline{e}f\overline{d}}Z_{c\overline{b}e\overline{f}},`$ (122)
$`\stackrel{~}{Z}_{a\overline{b}c\overline{d}}^{(2)}Z_{a\overline{e}f\overline{b}}Z_{c\overline{d}e\overline{f}},\stackrel{~}{Z}_{a\overline{b}c\overline{d}}^{(4)}Z_{a\overline{d}e\overline{f}}Z_{c\overline{e}f\overline{b}}.`$
Note that $`\stackrel{~}{Z}_{a\overline{b}c\overline{d}}^{(n+2)}\stackrel{~}{Z}_{c\overline{b}a\overline{d}}^{(n)}`$ for $`n=1,2`$. Unlike $`Z_{a\overline{b}c\overline{d}}`$ and $`Z_{a\overline{b}c\overline{d}}^{(n)}`$, the tensors $`\stackrel{~}{Z}_{a\overline{b}c\overline{d}}^{(n)}`$ are not symmetric under interchange of the first and second pair of indices. In particular,
$$\stackrel{~}{Z}_{a\overline{b}c\overline{d}}^{(1)}=\stackrel{~}{Z}_{c\overline{d}a\overline{b}}^{(2)},\stackrel{~}{Z}_{a\overline{b}c\overline{d}}^{(3)}=\stackrel{~}{Z}_{c\overline{d}a\overline{b}}^{(4)}.$$
(123)
Consequently, we must distinguish between two types of hermiticity conditions. For $`n=1,2`$, the $`\stackrel{~}{Z}_{a\overline{b}c\overline{d}}^{(n)}`$ satisfy the hermiticity condition of the first kind:
$$[\stackrel{~}{Z}_{a\overline{b}c\overline{d}}^{(n)}]^{}=\stackrel{~}{Z}_{b\overline{a}d\overline{c}}^{(n)},n=1,2,$$
(124)
whereas for $`n=3,4`$, the $`\stackrel{~}{Z}_{a\overline{b}c\overline{d}}^{(n)}`$ satisfy the hermiticity condition of the second kind:
$$[\stackrel{~}{Z}_{a\overline{b}c\overline{d}}^{(n)}]^{}=\stackrel{~}{Z}_{d\overline{c}b\overline{a}}^{(n)},n=3,4.$$
(125)
The final class of rank-four tensors involve $`Z_{a\overline{b}}^{(n)}`$ ($`n=1,2`$). These are:
$`Z_{a\overline{b}c\overline{d}}^{(1n)}Z_{a\overline{b}c\overline{f}}Z_{f\overline{d}}^{(n)},Z_{a\overline{b}c\overline{d}}^{(5n)}Z_{a\overline{b}f\overline{d}}Z_{c\overline{f}}^{(n)},`$ (126)
$`Z_{a\overline{b}c\overline{d}}^{(2n)}Z_{c\overline{b}a\overline{f}}Z_{f\overline{d}}^{(n)},Z_{a\overline{b}c\overline{d}}^{(6n)}Z_{c\overline{b}f\overline{d}}Z_{a\overline{f}}^{(n)},`$ (127)
$`Z_{a\overline{b}c\overline{d}}^{(3n)}Z_{c\overline{d}a\overline{f}}Z_{f\overline{b}}^{(n)},Z_{a\overline{b}c\overline{d}}^{(7n)}Z_{c\overline{d}f\overline{b}}Z_{a\overline{f}}^{(n)},`$ (128)
$`Z_{a\overline{b}c\overline{d}}^{(4n)}Z_{a\overline{d}c\overline{f}}Z_{f\overline{b}}^{(n)},Z_{a\overline{b}c\overline{d}}^{(8n)}Z_{a\overline{d}f\overline{b}}Z_{c\overline{f}}^{(n)}.`$ (129)
These tensors possess neither the hermiticity nor the symmetry properties of $`Z_{a\overline{b}c\overline{d}}`$. Instead, we have (for $`n=1,2`$):
$$[Z_{a\overline{b}c\overline{d}}^{(mn)}]^{}=Z_{b\overline{a}d\overline{c}}^{(m+4,n)},m=1,\mathrm{}4.$$
(130)
Note that $`Z_{a\overline{b}c\overline{d}}^{(2n)}Z_{c\overline{b}a\overline{d}}^{(1n)}`$, $`Z_{a\overline{b}c\overline{d}}^{(3n)}Z_{c\overline{d}a\overline{b}}^{(1n)}`$, and $`Z_{a\overline{b}c\overline{d}}^{(4n)}Z_{a\overline{d}c\overline{b}}^{(1n)}`$ are all distinct due to the lack of symmetry under the interchange of indices.
We proceed to examine all possible quadratic and cubic scalar $`Z`$-invariants. The quadratic scalar $`Z`$-invariants are obtained by summing over the indices of the tensors defined above in all possible allowed ways. However, note that the two-index tensors are hermitian, and any four-index tensor summed over two indices yields a two-index hermitian tensor. Hence any quadratic $`Z`$-invariant is the trace of an hermitian tensor and is hence real. We thus turn to the (non-trivial) cubic $`Z`$-invariants. These must be of the form $`Z_{a\overline{b}}^{(n)}X_{b\overline{a}}`$ ($`n=1`$ or 2) where $`X_{b\overline{a}}`$ is one of the quadratic second-rank tensors defined above, or of the form $`Z_{a\overline{b}c\overline{d}}X_{b\overline{a}d\overline{c}}`$, where $`X_{b\overline{a}d\overline{c}}`$ is one of the quadratic fourth-rank tensors defined above. But, for any hermitian second-rank tensor, $`X_{b\overline{a}}`$, the quantity $`Z_{a\overline{b}}^{(n)}X_{b\overline{a}}`$ is real. Similarly, for any fourth-rank tensor $`X_{b\overline{a}d\overline{c}}`$ that either satisfies the hermiticity conditions of the first or second kind \[see eqs. (124) and (125)\], the quantity $`Z_{a\overline{b}c\overline{d}}X_{b\overline{a}d\overline{c}}`$ is real. All that remains is to check that the scalar quantities of the form $`Z_{a\overline{b}c\overline{d}}Z_{b\overline{a}d\overline{c}}^{(mn)}`$ are real. This is proved by first establishing the following non-trivial result:
$$Z_{a\overline{b}c\overline{d}}Z_{b\overline{a}d\overline{c}}^{(mn)}=Z_{a\overline{b}c\overline{d}}Z_{b\overline{a}d\overline{c}}^{(m+4,n)}.$$
(131)
We have checked this result explicitly with Mathematica (although a simple analytic proof eludes us). Using eq. (130), it immediately follows that all such $`Z`$-invariants are real. This completes the proof that all cubic $`Z`$-invariants are real.
We next turn to the scalar invariants that are linear in $`Y`$. Since for any hermitian two-index tensor $`X_{b\overline{a}}`$, the quantity $`Y_{a\overline{b}}X_{b\overline{a}}`$ is real, it immediately follows that any scalar invariant that is linear in $`Y`$ and at most quadratic in the $`Z`$โs is real. Finally, consider scalar invariants that are quadratic in the $`Y`$โs. Note that $`Y_{a\overline{b}}Y_{c\overline{d}}`$ has the same hermiticity property as $`Z_{a\overline{b}c\overline{d}}`$, and $`Y_{a\overline{c}}Y_{c\overline{b}}`$ is an hermitian two-index tensor. Thus, any scalar invariant quadratic in the $`Y`$โs and linear in $`Z`$ is real. Hence, we have proven that all cubic invariants are real.
It is instructive to see where the above arguments break down when quartic invariants are considered. The simplest complex scalar invariant that is linear in $`Y`$ is at least cubic in $`Z`$. Indeed,
$$I_{Y3Z}=\mathrm{Im}(Z_{a\overline{c}}^{(1)}Z_{c\overline{d}}^{(11)}Y_{d\overline{a}})=\mathrm{Im}[\mathrm{Tr}(Z^{(1)}Z^{(11)}Y)]$$
(132)
is a potentially complex quartic invariant. Note that although $`Y`$, $`Z^{(1)}`$ and $`Z^{(11)}`$ are all hermitian $`2\times 2`$ matrices, $`I_{Y3Z}`$ is not necessarily real because $`Z^{(1)}`$ and $`Z^{(11)}`$ do not commute. More generally, one can check that all manifestly complex scalar invariants that are linear in $`Y`$ and cubic in $`Z`$ can be written in the form $`\mathrm{Tr}(Z^{(n)}Z^{(pq)}Y)`$ or $`\mathrm{Tr}(Z^{(pq)}Z^{(n)}Y)`$. A simple Mathematica computation reveals that
$$I_{Y3Z}=\mathrm{Im}(Z_{a\overline{c}}^{(n)}Z_{c\overline{d}}^{(pq)}Y_{d\overline{a}})=\mathrm{Im}(Z_{c\overline{d}}^{(pq)}Z_{a\overline{c}}^{(n)}Y_{d\overline{a}})$$
(133)
for all possible values of $`n,q=1,2`$ and $`p=1,2,3`$. The last equality in eq. (133) follows from the hermiticity of the $`Z^{(n)}`$, $`Z^{(pq)}`$ and $`Y`$. Hence, we conclude that the imaginary parts of all complex invariants of this type are equal to $`\pm I_{Y3Z}`$.
The simplest complex scalar invariant that is quadratic in $`Y`$ is at least quadratic in $`Z`$. Indeed,
$$I_{2Y2Z}=\mathrm{Im}(Y_{a\overline{b}}Y_{c\overline{d}}Z_{b\overline{a}d\overline{c}}^{(11)})$$
(134)
is a potentially complex quartic invariant. This quantity is not necessarily real since $`Z_{b\overline{a}d\overline{c}}^{(11)}`$ does not satisfy any hermiticity conditions. More generally, one can check that all manifestly complex scalar invariants that are quadratic in both $`Y`$ and $`Z`$ can be written in the form $`Y_{a\overline{b}}Y_{c\overline{d}}Z_{b\overline{a}d\overline{c}}^{(mn)}`$ (for $`m=1,\mathrm{},8`$ and $`n=1,2`$).<sup>31</sup><sup>31</sup>31In particular, it is straightforward to show that $`Y_{a\overline{b}}Y_{c\overline{d}}Z_{b\overline{a}d\overline{c}}^{(n)}`$ and $`Y_{a\overline{b}}Y_{c\overline{d}}\stackrel{~}{Z}_{b\overline{a}d\overline{c}}^{(n)}`$ are real due to the hermiticity properties of $`Y`$, $`Z^{(n)}`$ and $`\stackrel{~}{Z}^{(n)}`$. A simple Mathematica computation reveals that
$$I_{2Y2Z}=\mathrm{Im}(Y_{a\overline{b}}Y_{c\overline{d}}Z_{b\overline{a}d\overline{c}}^{(mn)})=\mathrm{Im}(Y_{a\overline{b}}Y_{c\overline{d}}Z_{b\overline{a}d\overline{c}}^{(m+4,n)})$$
(135)
for all possible values of $`m=1,\mathrm{},4`$ and $`n=1,2`$ \[where the second equality above is a consequence of eq. (130)\]. Hence, we conclude that the imaginary parts of all complex invariants of this type are equal to $`\pm I_{2Y2Z}`$.
Finally, a comprehensive analytic study of $`n`$th-order pure $`Z`$-invariants for $`n4`$ of the type employed above (in the analysis of the cubic invariants) seems prohibitive. Thus, a systematic Mathematica-aided study was carried out to prove that all fourth and fifth-order $`Z`$-invariants are real.
## Appendix F Time-reversal invariance of the Higgs vacuum
In this appendix, we assume that the Higgs scalar action is explicitly CP-conserving (and hence time-reversal invariant by the CPT theorem). That is, there exists a time reversal operator $`๐ฏ`$ that satisfies eq. (62) (for some choice of $`U_T`$). In this context, we ask whether the Higgs vacuum is time-reversal invariant. However, there is an apparent ambiguity, since as shown in Appendix A there may be a number of distinct choices for the time reversal operator (under which the action is invariant). This ambiguity corresponds to a non-trivial discrete group $`๐`$ that is a symmetry of the scalar Lagrangian. In general, the vacuum is not invariant with respect to $`๐`$. In this case, the vacuum may select one distinct choice for the time reversal operator. We shall denote this choice below by $`๐ฏ`$. That is, the theory is time-reversal invariant if the Higgs scalar action is CP-conserving and the vacuum is invariant with respect to (at least) one of the distinct choices for the time reversal operator. If there is no choice for the time reversal operator such that the vacuum is invariant, then time reversal invariance is spontaneously broken.
We denote the vacuum state by $`|0`$ and define $`\mathrm{\Phi }_a|0|\mathrm{\Phi }`$. The action of the time reversal operator is denoted by:
$$๐ฏ|0|0_T,๐ฏ|\mathrm{\Phi }|\mathrm{\Phi }_T.$$
(136)
The anti-unitarity of $`๐ฏ`$ implies that $`0_T|\mathrm{\Phi }_T=0|\mathrm{\Phi }^{}`$. Invariance of the vacuum under time-reversal invariance implies that $`|0=|0_T`$. Hence $`0|\mathrm{\Phi }_T=0|\mathrm{\Phi }^{}`$. It then follows that:
$$0|๐ฏ\mathrm{\Phi }_a๐ฏ^1|0=0|\mathrm{\Phi }_a|0^{},$$
(137)
after inserting $`๐ฏ๐ฏ^1`$ in the appropriate spot and using $`๐ฏ|0=|0`$. Using eq. (61), we end up with Branco:1983tn :
$$(U_T)_{a\overline{b}}\mathrm{\Phi }_b=\mathrm{\Phi }_a^{},$$
(138)
where $`\mathrm{\Phi }_a0|\mathrm{\Phi }_a|0`$. We can use the above results to prove Theorem 3 of section V.
Theorem 3: Given an explicitly CP-conserving Higgs potential, the vacuum is time-reversal invariant if and only if a real basis exists in which the Higgs vacuum expectation values are real.
We prove this theorem by demonstrating that eq. (138) provides the real basis in which the vacuum expectation values are real. By assumption, eq. (138) is satisfied in the $`\mathrm{\Phi }`$-basis (which may or may not be a real basis). As shown in Appendix B, one can always write $`U_T=V^TV`$, where the unitary matrix $`V`$ is unique up to multiplication on the left by an arbitrary orthogonal matrix. Inserting this result into eq. (138) yields
$$V\mathrm{\Phi }=[V\mathrm{\Phi }]^{},$$
(139)
which implies that the vacuum expectation values are real in the $`\mathrm{\Phi }^{}`$-basis, where $`\mathrm{\Phi }^{}V\mathrm{\Phi }`$. However, eqs. (68) and (71) imply that the $`\mathrm{\Phi }^{}`$-basis is a real basis. Of course, if the vacuum expectation values are real in a basis in which all the Higgs potential parameters are real, then the choice $`U_T=I`$ in eq. (61) yields a viable time-reversal operator. Conversely, if the Higgs scalar action is time-reversal invariant but no real basis exists in which the vacuum expectation values are real, then no viable time reversal transformation law exists. In particular, no choice of $`U_T`$ exists that satisfies eq. (138). This can only imply that $`๐ฏ|0|0`$. In this case, the time reversal symmetry is spontaneously broken. Thus, Theorem 3 is proven.
The conditions for a time-reversal invariant theory can therefore be reformulated. The scalar sector of the theory is time-reversal invariant if a $`U_T`$ exists that satisfies eqs. (62) and (138). In practice, the existence or non-existence of such a $`U_T`$ may be difficult to discern, whereas the corresponding basis-independent conditions quoted in section V are straightforward to implement.
Note that the existence of real bases does not necessarily imply that the vacuum expectation values are real in all possible real basis choices. In Appendix A, we demonstrated that if the scalar action is time-reversal invariant then different choices for $`๐ฏ`$ correspond to different real bases in which $`U_T=I`$. If the time reversal operator is defined according to eq. (61) then $`U_T=V^TV`$ yields a real basis $`\mathrm{\Phi }^{}=V\mathrm{\Phi }`$ in which $`U_T^{}=I`$. Alternatively, if the time reversal operator is defined according to eq. (72), then $`\stackrel{~}{U}_T=\stackrel{~}{V}^T\stackrel{~}{V}`$ yields a real basis $`\mathrm{\Phi }^{\prime \prime }=\stackrel{~}{V}\mathrm{\Phi }`$ in which $`\stackrel{~}{U}_T^{\prime \prime }=I`$. The transformation between these two real bases is $`\mathrm{\Phi }^{\prime \prime }=W\mathrm{\Phi }^{}`$, where $`W`$ spans an O(2)$`\times ๐`$ subgroup of U(2).<sup>32</sup><sup>32</sup>32One can also perform a U(1)<sub>Y</sub> transformation, which does not modify the relative phase of the two vacuum expectation values. In Appendix A, we noted that $`U_T^{\prime \prime }=(WW^T)^1`$ and $`\stackrel{~}{U}_T^{}=W^TW`$. If $`๐`$ is trivial, then $`WW^T=W^TW=I`$ and $`U_T=I`$ (up to an overall phase) in any real basis. Eq. (138) then implies that the vacuum expectation values are relatively real in any real basis (and can be chosen real with an appropriate U(1)<sub>Y</sub> phase rotation). If $`๐`$ is nontrivial, then the vacuum expectation values cannot be relatively real in both the $`\mathrm{\Phi }^{}`$-basis and the $`\mathrm{\Phi }^{\prime \prime }`$-basis if $`\mathrm{\Phi }^{\prime \prime }=W\mathrm{\Phi }^{}`$, where $`WW^Te^{i\eta }I`$.
As a simple example, consider again the model specified by eq. (36) with $`\lambda _6`$ real, which was examined at the end of section V. The $`\mathrm{\Phi }`$-basis in this case is a real basis but the vacuum expectation values, $`\sqrt{2}\widehat{v}=(e^{i\xi /2},e^{i\xi /2})`$, exhibit a nontrivial relative phase for $`\xi 0`$ (mod $`\pi `$). Nevertheless, the Higgs vacuum is time-reversal invariant. In this case, we can explicitly exhibit the matrix $`U_T`$ that satisfies eq. (138) and a unitary matrix $`V`$ such that $`U_T=V^TV`$:
$$U_T=\left(\begin{array}{cc}0& 1\\ 1& 0\end{array}\right),V=\frac{1}{\sqrt{2}}\left(\begin{array}{cc}\mathrm{cos}\theta & \mathrm{sin}\theta \\ \mathrm{sin}\theta & \mathrm{cos}\theta \end{array}\right)\left(\begin{array}{cc}1& 1\\ i& i\end{array}\right),$$
(140)
where $`\theta `$ is an arbitrary angle. Indeed, the matrix $`V`$ transforms the (real) $`\mathrm{\Phi }`$-basis to another real basis in which the vacuum expectation values are real. In particular, the choice of $`\theta =\xi /2`$ yields $`\widehat{v}^{}=V\widehat{v}=(1,0)`$ as noted at the end of section V.
|
warning/0506/hep-ph0506122.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
Low-energy neutrino experiments are providing crucial insight into lepton masses and mixing, though this is still limited in its scope. The most economical model for light neutrino masses is the seesaw model , but even the minimal model with three heavy singlet neutrinos contains a total of 18 parameters in the neutrino sector . So far, neutrino experiments provide us with measurements of only four of these : two squared-mass differences $`\mathrm{\Delta }m_{12}^2,`$ $`\mathrm{\Delta }m_{23}^2,`$ and two neutrino mixing angles $`\theta _{23,12}`$. There are prospects for measuring one more mixing angle, $`\theta _{13}`$ and the CP-violating Maki-Nakagawa-Sakata phase $`\delta `$, as well, perhaps, as the overall neutrino mass scale in cosmological data and one combination of Majorana mass parameters in neutrinoless double-$`\beta `$ decay . However, even these measurements would fall short of providing complete information on the full set of nine parameters that are in principle observable in low-energy neutrino experiments , out of the full total of 18.
Nevertheless, the information available from low-energy neutrino experiments is already striking . The atmospheric mixing angle $`\theta _{23}`$ is close to maximal: $`\mathrm{sin}^22\theta _{23}=1.02\pm 0.04`$, and the solar mixing angle $`\theta _{12}`$ is quite large: $`\mathrm{tan}^2\theta _{12}=0.45\pm 0.05`$. It therefore seems that neutrino mixing must be qualitatively different from the smaller mixing visible between the left-handed quarks, where the largest mixing angle is the original Cabibbo angle: $`\mathrm{sin}\theta _C=0.22`$. Such a difference in the quark and neutrino mixing patterns was not widely expected before the experiments, and has given rise to much theoretical discussion and speculation .
One of the problematic issues in the interpretation of the low-energy neutrino data is the running of neutrino masses and mixing parameters below and above the seesaw mass scales. This renormalization inevitably introduces some โfuzzinessโ in the comparison between low-energy measurements and any specific Ansatz for the mass matrix at the seesaw scale, since the renormalization depends on many of the unknown parameters in the seesaw model. The renormalization group equations (RGEs) have been used to study this running extensively, both numerically and analytically <sup>1</sup><sup>1</sup>1An up-to-date list of references on this very extensive subject can be found in .. As a result, the observed low-energy neutrino mixing can be obtained starting from either a bimaximal or from an almost diagonal neutrino mass matrix at the Grand Unification (GUT) scale $`M_{GUT}`$. In such a situation, understanding the systematical features of the running of neutrino parameters becomes crucial for the interpretation of the neutrino data and for the building of flavour models.
The purpose of this paper is to study comprehensively the dependence of neutrino renormalization effects on all the seesaw parameters, paying special attention to obtaining the correct low-energy neutrino measurements. The running of the effective neutrino mass matrix below the lightest singlet neutrino mass is generally well under control and large renormalization effects can be expected only in the case of degenerate light neutrino masses and, in supersymmetric models, for very large values of $`\mathrm{tan}\beta `$ . However, understanding the renormalization effects above and between the heavy neutrino scales is much more complicated, since new non-trivial dependences on the heavy neutrino Yukawa couplings $`Y_\nu ^{ij}`$ are introduced. Because the flavour structure of the new contribution to the RGEs can be very different from that due to the effective neutrino mass matrix, large effects are possible. Since the couplings $`Y_\nu ^{ij}`$ are largely unknown, a typical top-down approach taken in previous studies has been to fix the neutrino parameters at $`M_{GUT}`$ at some chosen values, then to run them down to the electroweak scale and demonstrate that, for this particular choice, the low-energy neutrino mass matrix is compatible with experimental data.
In this paper we take a bottom-up approach in which we first fix the known low-energy neutrino parameters to their measured values, and evaluate renormalization towards higher scales consistently in the framework of the minimal supersymmetric seesaw model. In our approach, every set of studied neutrino parameters is physical by construction. We parameterize the nine high-energy parameters of the seesaw mechanism using the orthogonal complex matrix $`R`$ , and scan over all the 18 seesaw parameters by generating the unknown parameters (including phases) randomly. We run the neutrino parameters up to the GUT scale and study the dependence of the renormalization effects on $`(i)`$ the other observable low-energy and $`(ii)`$ the high-energy parameters.
We find that significant renormalization effects can occur only when some of the light neutrino masses get comparable contributions from two or three heavy neutrinos $`N_j`$ and/or the light neutrino mass scale is at least moderately degenerate. Because the matrix $`R_{ij}`$, known as the dominance matrix , characterizes the contributions of the heavy neutrino $`N_j`$ to the light neutrino masses $`\nu _i`$, this parametrization turns out to be quite useful for the present study. It has been stated in the literature that the solar angle $`\theta _{12}`$ usually runs more than $`\theta _{13,23}.`$ We find that, for light neutrino masses with a strong normal hierarchy, exactly the opposite occurs. The quark-lepton complementarity relation
$`\theta _C+\theta _{12}`$ $`=`$ $`{\displaystyle \frac{\pi }{4}},`$ (1)
turns out to be very stable while, at the same time, the neutrino angles $`\theta _{13}`$ and $`\theta _{23}`$ may receive renormalization effects larger than the accuracy of plausible future experimental tests. The renormalization of the low-energy oscillation phase $`\delta `$ is generally enhanced compared with the running of mixing angles. Nevertheless, a non-diagonal $`R`$-matrix is needed for a large effect also in this case. An interesting feature is the possible crossing of light-neutrino mass eigenstates, which is accompanied by discrete changes in the neutrino mixing pattern, and is correlated with the $`R`$-matrix parameters.
Our paper is organized as follows. In Section 2 we present calculational details of our study. In Section 3 we present and discuss our results. Finally, some conclusions are drawn in Section 4.
## 2 Running of Neutrino Parameters in the MSSM
The superpotential of the minimal supersymmetric standard model (MSSM) with singlet (right-handed) heavy neutrinos is given by
$`W=D^cY_dQH_1+U^cY_uQH_2+E^cY_eLH_1+N^cY_\nu LH_2+{\displaystyle \frac{1}{2}}N^cMN^c,`$ (2)
where the Yukawa matrices $`Y`$ are general complex $`3\times 3`$ matrices and the $`3\times 3`$ heavy neutrino mass matrix $`M`$ is symmetric. The Yukawa matrices can be diagonalized by bi-unitary transformations $`Y^D=U^{}YV`$, where $`V,U`$ refer to the rotation of the left- and right-chiral fields, respectively. In the case of the symmetric matrix $`M`$, $`U=V^{}`$. To explain the neutrino data naturally, the hierarchy in $`M`$ should preferably be of the same order as the square of the hierarchy in $`Y_\nu `$ . We therefore assume hierarchical heavy-neutrino masses: $`M_1M_2M_3.`$
Integrating out all the heavy singlet neutrinos, one gets the usual dimension-5 effective operator
$`={\displaystyle \frac{1}{2}}\kappa LLH_2H_2,`$ (3)
which after electroweak symmetry breaking gives masses to the light neutrinos:
$`m_\nu (\mu )=\kappa (\mu )v^2\mathrm{sin}^2\beta ,`$ (4)
where $`\mu `$ is the renormalization scale, $`v=174`$ GeV and $`\mathrm{tan}\beta =v_2/v_1`$ is the ratio of the v.e.v.โs of the corresponding Higgs doublets. Above the heaviest neutrino mass scale, $`\mu >M_3,`$ the light-neutrino mass matrix reads
$`m_\nu (\mu )=Y_\nu ^T(\mu )M^1(\mu )Y_\nu (\mu )v^2\mathrm{sin}^2\beta .`$ (5)
Between the heavy-neutrino mass scales, $`M_1M_2M_3,`$ there exist a series of effective theories with, in general, $`n`$ active heavy neutrinos. The tree-level matching conditions between these theories at the neutrino thresholds are
$$\stackrel{\left(n\right)}{\kappa }_{gf}|_{M_n}=\stackrel{\left(n+1\right)}{\kappa }_{gf}|_{M_n}+(\stackrel{\left(n+1\right)}{Y_\nu })_{ng}\stackrel{\left(n+1\right)}{M}_n^1(\stackrel{\left(n+1\right)}{Y}_\nu )_{ng}|_{M_n},$$
(6)
where $`(n)`$ is the number of heavy neutrinos not integrated out. In general, the light-neutrino mass matrix can be written as
$`m_\nu =\left(\stackrel{\left(n\right)}{\kappa }+\stackrel{\left(n\right)}{Y_\nu ^T}\stackrel{\left(n\right)}{M}^1\stackrel{\left(n\right)}{Y}_\nu \right)v^2\mathrm{sin}^2\beta .`$ (7)
Since $`m_\nu `$ and $`Y_e^{}Y_e`$ can be diagonalized with the unitary matrices $`V_\nu `$ and $`V_e`$, respectively, the mixing matrix observable in the low-energy experiments is
$`V_{MNS}=V_e^{}V_\nu .`$ (8)
While $`m_\nu `$ contains 9 physical parameters, $`Y_\nu `$ and $`M`$ together contain 18 parameters. The missing 9 parameters crucially affect the physical observables. These include, for example, renormalization-induced lepton-number-violating processes and electric dipole moments in the supersymmetric seesaw model, as well as the renormalization of the light-neutrino parameters . Therefore, to study the dependence of the renormalization of eq. (7) on $`Y_\nu `$, we parametrize $`Y_\nu `$ with the complex orthogonal matrix $`R`$ . In the basis in which $`M`$ is diagonal, we write
$`\stackrel{\left(n\right)}{Y}_\nu =(\stackrel{\left(n\right)}{M}^D)^{\frac{1}{2}}\stackrel{\left(n\right)}{R}(m_\nu ^D)^{\frac{1}{2}}V_\nu ^{}(v\mathrm{sin}\beta )^1,`$ (9)
where the matrix $`R`$ is parametrized in terms of three *complex* angles $`\theta _{12}^R`$, $`\theta _{13}^R`$ and $`\theta _{23}^R`$:
$$R=\left(\begin{array}{ccc}c_{12}^Rc_{13}^R& s_{12}^Rc_{13}^R& s_{13}^R\\ c_{23}^Rs_{12}^Rs_{23}^Rs_{13}^Rc_{12}^R& c_{23}^Rc_{12}^Rs_{23}^Rs_{13}^Rs_{12}^R& s_{23}^Rc_{13}^R\\ s_{23}^Rs_{12}^Rc_{23}^Rs_{13}^Rc_{12}^R& s_{23}^Rc_{12}^Rc_{23}^Rs_{13}^Rs_{12}^R& c_{23}^Rc_{13}^R\end{array}\right),$$
(10)
where $`s_{ij}^R\mathrm{sin}\theta _{ij}^R`$ and $`c_{ij}^R\mathrm{cos}\theta _{ij}^R`$. Since $`Y_\nu `$ and $`M`$ are renormalized, obviously also $`\stackrel{\left(n\right)}{R}`$, which consists of $`n`$ rows, runs with energy. The RGEs for $`Y_\nu `$ and $`M`$ can be found in , and the RGEs for $`R`$ were calculated in . Using these, $`\stackrel{\left(n\right)}{R}`$ has to be evaluated at every heavy neutrino threshold when the matching is performed.
The scale dependence of the effective/combined quantities in (7) is characterized by the differential equation
$`16\pi ^2{\displaystyle \frac{d\stackrel{\left(n\right)}{X}}{dt}}`$ $`=`$ $`(Y_e^{}Y_e+\stackrel{\left(n\right)}{Y}_\nu ^{}\stackrel{\left(n\right)}{Y}_\nu )^T\stackrel{\left(n\right)}{X}+\stackrel{\left(n\right)}{X}(Y_e^{}Y_e+\stackrel{\left(n\right)}{Y}_\nu ^{}\stackrel{\left(n\right)}{Y}_\nu )^T+`$ (11)
$`(2\text{Tr}(\stackrel{\left(n\right)}{Y}_\nu ^{}\stackrel{\left(n\right)}{Y}_\nu +3Y_u^{}Y_u)6/5g_1^26g_2^2)\stackrel{\left(n\right)}{X},`$
where $`X=\kappa ,Y_\nu ^TM^1Y_\nu .`$ Notice that below the $`M_1`$ scale $`\stackrel{\left(n\right)}{Y}_\nu =0.`$ Therefore one expects large renormalization effects to occur above the heavy-neutrino thresholds for two reasons. First, the Yukawa couplings $`Y_\nu `$ can be large. Secondly, the flavour structure of $`Y_\nu ^{}Y_\nu `$ can be very different from the flavour structure of $`Y_e^{}Y_e`$ and $`\kappa .`$ Both those effects can be traced back to the values of $`R`$ via eq. (9). Approximate analytical solutions to eq. (11) have been derived in , which allow one to understand the generic behaviour of the renormalization effects. However, due to enhancement/suppression factors and possible cancellations, the exact numerical solutions may differ considerably from those estimates.
## 3 Results for Normally-Ordered Light Neutrinos
In this Section we present the results of our study for the case of normally-ordered light-neutrino masses, using the following strategy. We start at $`M_Z`$, where we fix the measured light-neutrino parameters as $`\mathrm{\Delta }m_{12}^2=8.\times 10^5\text{eV}^2,`$ $`\mathrm{\Delta }m_{23}^2=2.2\times 10^3\text{eV}^2,`$ $`\mathrm{tan}^2\theta _{12}=0.41,`$ $`\mathrm{sin}\theta _{23}=0.7`$ and $`\mathrm{sin}\theta _{13}=0.05.`$ We then generate randomly the lightest neutrino mass, the heavy neutrino masses, all the low-energy phases and the initial values for the parameters in the $`R`$ matrix. We run all the relevant quantities up to $`M_{GUT}`$ using the 1-loop RGEs for the minimal supersymmetric seesaw model . At every heavy-neutrino threshold we perform the tree-level matching according to eq. (6). To calculate the values of $`\stackrel{\left(n\right)}{Y}_\nu `$ we use the renormalized values of the $`R`$ and $`M`$. At $`M_{GUT}`$ we calculate the renormalized light-neutrino parameters. We always keep the ordering of the light neutrino masses fixed, $`m_1<m_2<m_3`$ for the normal and $`m_3<m_1<m_2`$ for the inverse ordering. Because of that the physical range for $`\theta _{12}`$ extends up to $`\pi /2.`$
In order to accentuate the renormalization effects due to the low-energy neutrino parameters and the parameter matrix $`R,`$ we do not consider degenerate light neutrinos and we assume an upper limit $`m_1<0.1`$ eV on the lightest neutrino mass. Although the present most stringent limit on the overall light-neutrino mass scale scale coming from astrophysics and cosmology is a factor of 2 to 3 weaker, such precision can easily be achieved in the future cosmological experiments . We also minimize the renormalization effects of large $`\mathrm{tan}\beta `$ studied in by working with the relatively small value of $`\mathrm{tan}\beta =5.`$ Instead, we study how the large values of $`Y_\nu `$ affect the renormalization effects.
### 3.1 Renormalization of the Mixing Angles
We start by studying the running of the light-neutrino mixing angles. In Fig. 1 we plot the neutrino mixing angles $`\theta _{ij}`$ at $`M_{GUT}`$ as functions of the lightest neutrino mass $`m_1(M_Z)`$ for $`R(M_Z)=1`$ (left panel) and for randomly generated $`R`$ (right panel). In all the figures the neutrino mass parameters are presented in units of $`eV.`$ The mixing angles $`\theta _{12},`$ $`\theta _{13}`$ and $`\theta _{23}`$ are represented by green (light), blue (dark) and red (medium) dots, respectively. For $`R=1`$ the mixing angles practically do not run: only $`\theta _{12}`$ may change a little for light-neutrino masses close to $`m_1=0.1`$ eV. This is because, for moderate degeneracy, the renormalization of $`\theta _{12}`$ is enhanced by a factor $`m_1^2/\mathrm{\Delta }m_{12}^2.`$ This effect would have been larger for larger values of $`\mathrm{tan}\beta `$ and $`m_1.`$
Turning to the results for the randomly-generated values of $`R`$, a certain pattern of renormalization effects emerges. If $`m_1>\sqrt{\mathrm{\Delta }m_{12}^2},`$ very large changes in the mixing angles may occur. Although $`\theta _{12}`$ tends to change most due to the above-mentioned enhancement factor, also $`\theta _{13}`$ and $`\theta _{23}`$ may gain almost any value. We see that the examples of extreme running considered in the literature are due to having at least a moderately degenerate mass spectrum.
An interesting set of points in Fig. 1 are those gathered around $`\theta _{12}60^{}`$ in the region $`m_1>\sqrt{\mathrm{\Delta }m_{12}^2}.`$ Those correspond to the level crossing of two light-neutrino mass eigenvalues $`m_1`$ and $`m_2`$ due to renormalization. Because by definition $`m_1<m_2,`$ this causes a discrete jump in the value of $`\theta _{12}`$. In the standard parametrization of $`V_{MNS}`$, and with small $`\theta _{13}`$, this implies $`\mathrm{sin}\theta _{12}\mathrm{cos}\theta _{12}^{}`$ and, consequently, $`\theta _{12}^{}=90^{}\theta _{12}.`$ As seen in Fig. 1, this effect is smeared by strong running of $`\theta _{12}`$ and also $`\theta _{13}.`$
In contrast to the previous discussion, if the mass spectrum is strongly hierarchical: $`m_1<\sqrt{\mathrm{\Delta }m_{12}^2},`$ the solar angle is much more stable than the mixing angles $`\theta _{13}`$ and $`\theta _{23}.`$ The latter may vary through a range of almost $`10^{},`$ which is more than the expected precision of future experiments. Moreover, the widths of the $`\theta _{13}`$ and $`\theta _{23}`$ bands in Fig. 1 do not depend on the initial values of the angles $`\theta _{ij}.`$ Thus, discrimination between different flavour models may be possible in principle in the future, if one takes into account renormalization effects. We also note that, for hierarchical light-neutrino masses, the quark-lepton complementarity relation $`\theta _C+\theta _{12}=\pi /4`$ would be maintained with high accuracy at every scale, independently of the unobservable neutrino parameters.
We now study the origins of the effects due to $`R`$. In the left panels of Figs. 2, 3 and 4 we plot the distributions of the neutrino mixing angles $`\theta _{ij}(M_{GUT})`$ as functions of $`m_1(M_Z)`$ for randomly generated complex parameters $`\theta _{12}^R,`$ $`\theta _{13}^R`$ and $`\theta _{23}^R,`$ respectively. In each figure the other parameters in $`R`$ are set to zero. The same neutrino mixing angles are plotted in the right panels of Figs. 234 as functions of the absolute values of the corresponding $`R`$ matrix parameters $`|\theta _{12}^R|,`$ $`|\theta _{13}^R|`$ and $`|\theta _{23}^R|,`$ respectively.
We see in Fig. 2 that non-zero values of $`\theta _{12}^R`$ affect mostly the renormalization of $`\theta _{12}.`$ A large running effect requires also that the overall light neutrino mass scale be high. On the other hand, non-zero values of $`\theta _{13}^R`$ affect mostly the running of $`\theta _{13}`$ and $`\theta _{23},`$ as seen in Fig. 3. Again, a relatively high overall light-neutrino mass scale is required for a significant effect. We also learn from Fig. 3 that the level crossing of light mass eigenvalues is induced by non-zero $`\theta _{13}^R,`$ which strongly affects the running of $`m_1.`$ The parameter $`\theta _{23}^R`$ affects only the running of $`\theta _{13}`$ and $`\theta _{23}.`$ Fig. 4 shows an interesting feature: in this case the running of $`\theta _{13}`$ and $`\theta _{23}`$ does not depend on $`m_1,`$ and significant renormalization effects can be obtained also for very hierarchical light neutrinos.
Comparison of the left and right panels in Figs. 2, 3 and 4 reveals how the renormalization effects depend on the magnitude of the particular parameter $`\theta _{ij}^R.`$ Interestingly, in all the cases the dominant running occurs in the region $`|\theta _{ij}^R|๐ช(1).`$ We recall that $`R`$ can be interpreted to be a dominance matrix , i.e., it shows which heavy neutrino contribution dominates in the mass of a particular light neutrino. Therefore our results imply that, in order to have significant renormalization effects, at least two heavy neutrinos must contribute to one particular light neutrino mass in approximately equal amounts. If the light neutrino masses are dominated by one heavy neutrino each, no large running is possible unless the light neutrinos are degenerate in mass and $`\mathrm{tan}\beta `$ is large.
### 3.2 Renormalization of Masses
The observed hierarchy in the light-neutrino masses, $`\sqrt{\mathrm{\Delta }m_{12}^2/\mathrm{\Delta }m_{23}^2}=0.18,`$ is milder than expected in many flavour models. In GUTs with the simplest scalar sector, the neutrino Yukawa couplings are equal to the up-quark Yukawa couplings at $`M_{GUT}.`$ Contrary to that, phenomenology at the low scale seems to indicate that the neutrino hierarchy is more similar to the less hierarchical down-sector Yukawa couplings, rather than those in the up sector. However, at the moment the lightest neutrino mass and the hierarchy in the heavy-singlet sector are unknown, introducing large uncertainties into such considerations. Therefore it is interesting to study also how the masses of the light neutrinos evolve with energy.
In Fig. 5 we plot the distributions of $`\sqrt{\mathrm{\Delta }m_{12}^2}`$ and $`\sqrt{\mathrm{\Delta }m_{23}^2}`$ at $`M_{GUT}`$ for $`R(M_Z)=1`$ (left panel) and for randomly generated $`R`$ (right panel). The red dot denotes the starting point at $`M_Z`$ from which value all the other points are generated. The hierarchy at the GUT scale tends to be larger than at low energies, although the opposite is also possible for a few points. While for $`R=1`$ both mass differences tend to increase, for random $`R`$ they may also decrease. The abundant points with smaller values of $`\sqrt{\mathrm{\Delta }m_{12}^2}`$ in the right-hand plot correspond to non-zero values of $`\theta _{13}^R.`$ This parameter affects the Yukawa couplings of first generation in such a way that the $`(12)`$ mass difference may run considerably.
### 3.3 Renormalization of CP Observables
Our approach in this study is to fix the known neutrino parameters and to vary the unknown ones randomly. At the moment, the only CP-violating observable in the neutrino sector what we have information about is the baryon asymmetry of the Universe, assuming that the cosmological baryon asymmetry is generated via leptogenesis . In this case, it is possible to constrain one combination of the 6 CP-violating phases in the neutrino sector. However, because one can vary the remaining 5 combinations (and also the unknown CP-conserving neutrino parameters), one cannot make any firm prediction for the neutrino parameters . As the predictions for other renormalization-induced CP-violating observables such as the electric dipole moments of the charged leptons are orders of magnitude smaller than the present experimental bound , no firm constraints come from this sector either. Although Ref. argues that some systematic renormalization effects in leptogenesis are possible due to the running of the effective light neutrino mass matrix, those are already taken into account in the systematic study of .
At the moment, the most realistic possibility seems to be that of measuring the MNS phase $`\delta `$ in future oscillation experiments. The value of this phase, however, is presently unknown. In the left plot of Fig. 6 we present the values of $`\delta (M_{GUT})`$ against the initial values of the phase for $`R(M_Z)=1`$. As seen in the figure there is practically no running of $`\delta `$ in this case. The situation changes considerably for randomly generated $`R`$-matrices, as seen in the right panel of Fig. 6 where we plot the change of the phase, $`\delta (M_{GUT})\delta (M_Z),`$ as a function of the lightest neutrino mass $`m_1.`$ The running of the MNS phase can be numerically significant and, apart from high values of $`m_1,`$ almost independent of the lightest neutrino mass. This behaviour resembles the running of $`\theta _{13,23}`$ in Fig. 1 and can be traced back to the $`1/\theta _{13}`$ enhancement of the running of $`\delta `$ <sup>2</sup><sup>2</sup>2The points around $`\pm 2\pi `$ in Fig. 6 actually correspond to small running modulo $`2\pi `$..
## 4 Renormalization Effects for an Inverted Hierarchy of Light Neutrino Masses
We have repeated the earlier analyses also for an inverted hierarchy of light neutrino masses. Because the solar-neutrino mass scale is now higher than the atmospheric one, the Yukawa couplings of the first generations are generally larger. Therefore, all the effects related to the renormalization of the solar parameters are generally enhanced. This can be seen in Fig. 7, where we plot the neutrino mixing angles at $`M_{GUT}`$ as functions of the lightest neutrino mass $`m_3(M_Z)`$ for $`R(M_Z)=1`$ and for randomly-generated $`R.`$ If $`R(M_Z)=1,`$ the mixing practically does not run even in the inverted-hierarchy case. The exception is in the high-mass region $`m_3(M_Z)0.1`$ eV, when the mass eigenvalues cross and the moderate degeneracy causes discrete jumps in the mixing angles. As expected, for the random choice of $`R`$, the angle $`\theta _{12}`$ runs very strongly. The level-crossing stripe around $`\theta _{12}60^{}`$ exists for all values of $`m_3`$, while the angles $`\theta _{13}`$ and $`\theta _{23}`$ run only for large values of $`m_3`$.
## 5 Discussion and Conclusions
We have studied how the RGE running of neutrino mixing depends on the unknown seesaw parameters. We have taken a bottom-up approach in which we fix the known low-energy neutrino parameters to their measured values. Parametrizing the nine high-energy parameters of the seesaw mechanism via the dominance matrix $`R`$, we have scanned over all the 18 seesaw parameters by generating the unknown parameters randomly. The fact that the matrix $`R_{ij}`$ measures the heavy neutrino $`N_j`$ contribution to the light neutrino $`\nu _i`$ means that this parametrization is particularly valuable for this study. We have compared the results for random $`R`$ elements with the simple case $`R=1`$.
We have found that significant running effects can occur only when some of the light-neutrino masses have comparable contributions from more than one heavy neutrino $`N_i,`$ and the light-neutrino mass scale is at least moderately degenerate. For a normal hierarchy of neutrino masses, the complementarity relation (1) between neutrino and quark mixing angles evolves very little between the GUT scale and the electroweak scale. However, the other oscillation angles $`\theta _{13,23}`$ run rather more than $`\theta _{12}`$ and also beyond the expected measurement errors. In certain cases, we observe level crossing in the light-neutrino mass eigenstates that is accompanied by jumps in the oscillation angles. The running of the CP-violating oscillation phase $`\delta `$ is strong for random $`R`$ but insignificant for $`R=1`$.
The analysis presented here complements the top-down approach often adopted elsewhere. It reveals some of the pitfalls in inferring properties of the neutrino mass matrix generated at the GUT scale from current low-energy measurements alone, in the absence of supplementary theoretical or phenomenological input. We hope that these results may serve as useful aids in the attempt to understand the neutrino mass matrix, which has already revealed several surprises. The results presented here demonstrate that our low-energy measurements are far from telling us the whole story.
Acknowledgment
This work was supported by the ESF Grant 6140 and by the Ministry of Education and Research of the Republic of Estonia. We thank S. Antusch for useful communications.
|
warning/0506/math0506301.html
|
ar5iv
|
text
|
# Introduction
## Introduction
The purpose of this paper is to study, via dimensional reduction, certain holomorphic $`D`$-branes, closely related to torsion-free sheaves, on threefolds $`XC`$ fibered in resolved $`ADE`$ surfaces over a curve. Fibered local CalabiโYau threefolds $`X๐ธ^1`$ of this type, as well as their deformations $`X_s๐ธ^1`$ and extremal transitions, were thoroughly analized in from the point of view of supersymmetric gauge theory. The paper contains an assertion, made explicit in and studied in , that exceptional components of a natural threefold contraction $`X_s\overline{X}_s`$ are classified by irreducible representations of a certain quiver with loop edges, the $`N=1`$ $`ADE`$ quiver (see Figure 3.2 for an example), satisfying a specific set of relations. This statement is in the spirit of Gabrielโs theorem classifying exceptional (not necessarily irreducible) rational curves in resolved $`ADE`$ surfaces in terms of irreducible representations of the corresponding Dynkin quiver.
In this paper we generalize the work of in two directions: we consider holomorphic $`D`$-branes, objects in the derived category of coherent sheaves, instead of exceptional components, and we study the semi-local case: the neighbourhood of a deformed $`ADE`$ fibration $`X_sC`$ over a general curve $`C`$. The main result is Theorem 3.1, which shows that certain holomorphic $`D`$-branes on the fibered threefold $`X_s`$ are classified by representations with relations of a KronheimerโNakajima-type quiver in the category $`Coh(C)`$ of coherent sheaves on the curve $`C`$. In particular, moduli spaces of such holomorphic $`D`$-branes are quiver bundle varieties over $`C`$. If $`C๐ธ^1`$, a further dimensional reduction leads to Theorem 3.4, relating sheaves on the threefold to the zero-dimensional problem of ordinary matrix representations of the $`N=1`$ $`ADE`$ quiver of . The loops in the $`N=1`$ $`ADE`$ quiver arise as the action by multiplication of a parameter $`tH^0(๐ช_{๐ธ^1})`$ on spaces of sections of sheaves on the base $`๐ธ^1`$.
The geometry considered in this paper is non-monodromic, meaning that there is no global nor local monodromy in the fibration of $`ADE`$ surfaces over the curve $`C`$. It appears to be an interesting question to extend the results proved here to these more general cases involving monodromy.
In recent work , the moduli space of certain very special holomorphic $`D`$-branes on resolved $`A_1`$-fibered geometries $`XC`$ has been connected, via imposing a superpotential and going through a large $`N`$ transition, to the Hitchin system on $`C`$. The branes studied in are not of the type classified by our results; they should rather correspond to a complex of quiver representations. Understanding the precise connection between and the present paper is left for future work.
After introducing basic notation in Section 1, Section 2 describes the threefolds we study, and defines some auxiliary sheaves of non-commutative algebras over the curve $`C`$. Section 3 contains our results, in particular the general statement Theorem 3.1 connecting quiver bundles to holomorphic $`D`$-branes on ADE fibrations, as well as the statement for the affine case. Proofs are discussed in Section 4.
## 1. Finite groups of type $`ADE`$ and surfaces
Let $`\mathrm{\Gamma }<\mathrm{SL}(2,)`$ be a finite subgroup of type $`A,D`$ or $`E`$. Let $`๐ฅ_0`$ be the Cartan subalgebra of the finite dimensional Lie algebra of the same type. Fix a set of simple roots $`\{\eta _a:a\mathrm{\Delta }_0\}`$ indexed by nodes of the Dynkin diagram $`\mathrm{\Delta }_0`$, and let $`R_+`$ be the set of positive roots. Let $`๐ฅ`$ be the corresponding affine Cartan with simple roots indexed by nodes of the Dynkin diagram $`\mathrm{\Delta }\mathrm{\Delta }_0`$.
The group ring $`\mathrm{\Gamma }`$ has center $`Z(\mathrm{\Gamma })^\mathrm{\Delta }`$; explicitly, for $`\lambda Z(\mathrm{\Gamma })`$, the isomorphism is obtained by taking the trace of $`\lambda `$ on a set of irreps, indexed by the nodes of $`\mathrm{\Delta }`$ according to the McKay correspondence. There is also a natural identification
$$๐ฅ_0=\{\lambda ^\mathrm{\Delta }|\lambda \delta =0\}๐ฅ^\mathrm{\Delta },$$
where $`\delta =(\delta _a)`$ are the dimensions of the irreps of $`\mathrm{\Gamma }`$.
###### Lemma 1.1.
The centralizer $`C_{\mathrm{GL}(2,)}(\mathrm{\Gamma })`$ of $`\mathrm{\Gamma }`$ in $`\mathrm{GL}(2,)`$ is
1. the full group $`\mathrm{GL}(2,)`$ for type $`A_1`$;
2. a torus $`(^{})^2`$ in $`\mathrm{GL}(2,)`$ for type $`A_n`$ with $`n>1`$;
3. the center $`^{}`$ of $`\mathrm{GL}(2,)`$ for types $`D`$ and $`E`$.
Let $`\overline{Y}=๐ธ^2/\mathrm{\Gamma }`$ be the singular affine quotient, $`Y\overline{Y}`$ its minimal resolution. Exceptional curves in the resolution are in one-to-one correspondence with the nodes of $`\mathrm{\Delta }_0`$, and thus with a set of simple roots of $`๐ฅ_0`$; the positive roots $`\eta R_+`$ correspond to connected, possibly reducible exceptional rational curves. The universal deformations $`๐ด๐ฅ_0`$ and $`\overline{๐ด}๐ฅ_0/W`$ of $`Y`$ and $`\overline{Y}`$, where $`W`$ denotes the Weyl group, are connected by the well known commutative diagram
$$\begin{array}{ccccc}๐ด& & p^{}\overline{๐ด}& & \overline{๐ด}\\ & & & & \\ & & ๐ฅ_0& \stackrel{p}{}& ๐ฅ_0/W.\end{array}$$
## 2. Threefolds: definitions
### 2.1. The geometry
Let $`C`$ be a curve, and let $`๐ฌ`$ be a rank-two vector bundle on $`C`$ whose structure group reduces from $`\mathrm{GL}(2,)`$ to the centralizer $`C_{\mathrm{GL}(2,)}(\mathrm{\Gamma })`$. Thus, by Lemma 1.1,
* for type $`A_1`$, $`๐ฌ`$ is an arbitrary rank-two vector bundle;
* for type $`A_n`$ with $`n>1`$, $`๐ฌ๐ฌ_1๐ฌ_2`$ is the direct sum of two line bundles;
* for types $`D,E`$, $`๐ฌ๐ฌ_0^2`$ for some line bundle $`๐ฌ_0`$.
There is a fiberwise $`\mathrm{\Gamma }`$-action on the total space of the vector bundle $`๐ฌ`$, and the quotient $`\overline{X}=๐ฌ/\mathrm{\Gamma }`$ is a threefold with a curve of compound Du Val singularities along the image of the zero section. Let $`f:X\overline{X}`$ be the crepant resolution, with a map $`\pi :XC`$ whose fibres are minimal resolutions of the corresponding surface singularity, with trivial monodromy in the fibres. The canonical bundle of $`X`$ is
$$\omega _X\pi ^{}(\omega _Cdet๐ฌ^{}).$$
In particular, $`X`$ is CalabiโYau if and only if $`๐ฌ`$ has canonical determinant on $`C`$.
Part of the deformation theory of the threefold $`X`$ was described in . Let $`_0=det๐ฌ๐ฅ_0`$, a vector bundle over $`C`$, and let $`๐ฎ=H^0(C,_0)`$ be its space of sections. Then there is a smooth family of threefolds $`๐ณ๐ฎ`$, with injective KodairaโSpencer map and central fibre $`X_0X`$, together with a fibration $`๐ณC\times S`$ and a contraction $`๐ณ\overline{๐ณ}`$ over $`S`$. Thus, for every $`s๐ฎ`$, the threefold fibre $`X_s`$ possesses a fibration $`\pi _s:X_sC`$ in surfaces and a contraction $`f_s:X_s\overline{X}_s`$ to a singular threefold with compound Du Val singularities. More precisely, for every positive root $`\eta R_+`$ of $`๐ฅ_0`$, there is a map $`p_\eta :_0det๐ฌ`$, whose vanishing locus is a family of root hyperplanes in the $`๐ฅ_0`$ fibers, and we have
###### Lemma 2.1.
Let $`s๐ฎ=H^0(C,_0)`$ be a section of $`_0`$, and let $`\eta R_+`$ be a positive root of $`๐ฅ_0`$. The contraction $`f_s:X_s\overline{X}_s`$ contracts a (possibly reducible) rational curve corresponding to the root $`\eta `$ over a point $`PC`$, if and only if the projected section $`p_\eta (s)H^0(C,det๐ฌ)`$ vanishes at $`PC`$.
Thus if the projected section $`p_\eta (s)`$ is not identically zero for any root $`\eta `$, then $`f_s`$ is a small contraction, contracting rational curves to isolated singularities in certain configurations. If for different roots $`\eta `$, the sections $`p_\eta (s)`$ have different simple zeros, then $`f_s`$ contracts a set of isolated $`(1,1)`$-curves to simple nodes. If the linear system $`det๐ฌ`$ has no base points on $`C`$, then this holds for generic $`s๐ฎ`$.
In the special case $`C๐ธ^1`$, the central fiber $`X_0=๐ธ^1\times Y`$ is CalabiโYau, and its deformations are parameterized by an $`๐ฅ_0`$-valued polinomial $`s๐ฅ_0[t]`$. Under the isomorphism $`๐ฅ_0\{\lambda |s\delta =0\}^\mathrm{\Delta }`$, we can also parameterize deformations by a set of ordinary polinomials $`\mathrm{\Theta }_a[t]`$ indexed by nodes of the affine Dynkin diagram $`\mathrm{\Delta }`$, satisfying $`_a\delta _a\mathrm{\Theta }_a=0`$. The exceptional fibres of $`f_s:X_s\overline{X}_s`$ lie over roots of the various polynomials $`\mathrm{\Theta }_{\eta _a}=\mathrm{\Theta }_a`$, corresponding to simple roots $`\eta _a`$, as well as over roots of their linear combinations $`\mathrm{\Theta }_\eta =_a\mu _a\mathrm{\Theta }_a`$, corresponding to other positive roots $`\eta =_a\mu _a\eta _aR_+`$. For generic choice of parameter $`s๐ฎ`$, equivalently for generic choice of $`\{\mathrm{\Theta }_a\}`$, the polynomials $`\{\mathrm{\Theta }_\eta :\eta R_+\}`$ have distinct simple roots, and the exceptional set of $`f_s:X_s\overline{X}_s`$ consists of isolated $`(1,1)`$-curves.
### 2.2. Sheaves of non-commutative algebras and their sheaves of modules
Given $`(C,๐ฌ)`$, let $`=det๐ฌ๐ฅ`$, a vector bundle on the curve $`C`$ containing $`_0`$ as a subbundle. Given a section $`sH^0(C,)`$, consider the natural composition
$$\sigma _s:๐ฌ^{}๐ฌ^{}\stackrel{^2}{}det๐ฌ^{}\stackrel{s}{}๐ฅ๐ช_C\stackrel{}{}Z(\mathrm{\Gamma })๐ช_C,$$
a family of $`Z(\mathrm{\Gamma })`$-valued symplectic forms in the fibres of the vector bundle $`๐ฌ^{}`$. Also fix, once and for all, a trivializing section $`zH^0(๐ช_C)`$.
###### Definition 2.2.
Let $`๐_s`$ be the sheaf of non-commutative algebras on $`C`$ whose sections on an open set $`UC`$ are
$$๐_s(U)=T๐ฌ^{}(U)\mathrm{\Gamma }/[x_1,x_2]+\sigma _s(x_1,x_2),$$
where $`T๐ฌ^{}(U)`$ is the full tensor algebra of $`๐ฌ^{}(U)`$, $`x_i๐ฌ^{}(U)`$ are local sections, and $`\mathrm{}`$ denotes the two-sided ideal generated by all given expressions. Define also
$$๐ซ_s(U)=T(๐ฌ^{}๐ช_C)(U)\mathrm{\Gamma }/[x_1,x_2]+\sigma _s(x_1,x_2)z^2,[x_i,z],$$
where the fixed section $`zH^0(๐ช_C)`$ commutes with elements of $`\mathrm{\Gamma }`$. The sheaf $`๐ซ_s`$ becomes a sheaf of graded algebras by assigning degree $`1`$ to local sections $`x_i๐ฌ^{}(U)`$ as well as to $`zH^0(๐ช_C)`$; thus its degree-zero piece is
$$๐ซ_{s,0}๐ช_C\mathrm{\Gamma }.$$
###### Remark 2.3.
The sheaf of algebras $`๐_s`$ is a relavitive version of the following non-commutative deformation of the skew group algebra, introduced by CrawleyโBoevey and Holland in , depending on a deformation parameter $`\lambda ๐ฅZ(\mathrm{\Gamma })`$:
$$A_\lambda =x_1,x_2\mathrm{\Gamma }/[x_1,x_2]+\lambda .$$
The graded version is
$$P_\lambda =y_0,y_1,y_2\mathrm{\Gamma }/[y_0,y_i],[y_1,y_2]+\lambda y_0^2.$$
For $`\mathrm{\Gamma }=\{1\}`$, $`\lambda `$ is just a complex number; if $`\lambda 0`$, $`A_\lambda `$ is isomorphic to the first Weyl algebra, whereas $`P_\lambda `$ is a degenerate Sklyanin algebra deforming the algebra of functions on the commutative projective plane $`^2`$. As proved in , for general $`\mathrm{\Gamma }`$ and $`\lambda ๐ฅ_0Z(\mathrm{\Gamma })`$ the algebra $`A_\lambda `$ is finite over its center
$$ZA_\lambda [\overline{Y}_\lambda ].$$
The latter is the coordinate ring of the affine variety $`\overline{Y}_\lambda `$ corresponding to the deformation parameter $`\lambda ๐ฅ_0`$, a deformation of the invariant ring $`[x_1,x_2]^\mathrm{\Gamma }[\overline{Y}]`$. For $`\lambda ๐ฅ๐ฅ_0`$, $`A_\lambda `$ is โgenuinelyโ non-commutative.
By abuse of notation, we will refer to $`๐_s=Proj_C๐ซ_s`$ as the non-commutative projective bundle corresponding to $`s๐ฎ`$, with fibration $`\pi _s:๐_sC`$. Setting $`z=0`$, we have its divisor at infinity
$$i_s:D_s๐_s.$$
The divisor $`D_s`$ has the structure of an ordinary (commutative) $`^1`$-bundle
$$\pi _s|_{D_s}=\tau _s:D_sC$$
equipped with a $`\mathrm{\Gamma }`$-action on the fibres. Its complement $`๐_s=๐_sD_s=Spec_C๐_s`$ is a non-commutative affine bundle.
The sheaf $`๐ซ_s`$ is a sheaf of regular graded algebras in the sense of ; sheaf theory on $`๐_s`$ works in complete analogy with the absolute case discussed in . The category of coherent sheaves $`Coh(๐_s)`$ is by definition the quotient of the category of sheaves of finitely generated graded right $`๐ซ_s`$-modules by the subcategory of sheaves of torsion $`๐ซ_s`$-modules; we will sometimes refer to objects in this category as $`๐ซ_s`$-modules. The trivial module, graded in degree $`n`$, defines the object $`๐ช_{๐_s}(n)Coh(๐_s)`$; given a sheaf $``$, its twists $`(n)`$ are obtained by shifting the grading. We have $`Ext`$ groups as the derived functors of $`Hom`$, and also functors $`xt^i(,๐ช_{๐_s})`$; the latter take values in the category of left $`๐ซ_s`$-modules (compare ).
Pushforward
$$\pi _s:Coh(๐_s)Coh^\mathrm{\Gamma }(C)$$
along the morphism $`\pi _s:๐_sC`$ is defined in the usual way, as the coherent $`\mathrm{\Gamma }`$-sheaf on $`C`$ defined by sections over preimages of open sets of $`C`$, the section spaces being (right) $`\mathrm{\Gamma }`$-modules; the action of $`\mathrm{\Gamma }`$ on $`C`$ is taken to be trivial. The higher pushforwards $`\mathrm{R}^p\pi _s()`$ are the derived functors of $`\pi _s`$. Given a $`๐_s`$-module $``$, we will also use the relative $`Hom`$-functor
$$Hom_C(,):Coh(๐_s)Coh^\mathrm{\Gamma }(C)$$
defined by homomorphisms on preimages of open sets in $`C`$, as well as its derived functors
$$Ext_C^i(,):Coh(๐_s)Coh^\mathrm{\Gamma }(C).$$
We also have a pullback functor
$$\pi _s^{}:Coh^\mathrm{\Gamma }(C)Coh(๐_s)$$
taking a sheaf of (right) $`\mathrm{\Gamma }`$-modules $``$ to the sheaf $`_\mathrm{\Gamma }๐ซ_s`$ of (right) $`๐ซ_s`$-modules. The pair $`(\pi _s^{},\pi _s)`$ forms an adjoint pair as in the commutative case. Similarly, for the inclusion $`i_s:D_s๐_s`$, we have a pullback (restriction) functor
$$i_s^{}:Coh(๐_s)Coh^\mathrm{\Gamma }(D_s),$$
defined by factoring modules of local sections by the ideal $`z`$ (recall that $`z`$ is central), as well as a pushforward
$$i_s:Coh^\mathrm{\Gamma }(D_s)Coh(๐_s),$$
with $`z`$ acting on local sections by zero. There is also a restriction functor to the finite part $`๐_s`$, defined by factoring the ideal $`z1`$.
###### Definition 2.4.
A $`\pi _s`$-free sheaf on $`๐_s`$ is an object $`Coh(๐_s)`$, which admits an embedding
$$\pi _s^{}(๐ฐ)(n)$$
for some $`๐ฐCoh^\mathrm{\Gamma }(C)`$ and $`n`$. A framed $`\pi _s`$-free sheaf $`(,\phi )`$ on $`(๐_s,D_s)`$ is a $`\pi _s`$-free sheaf $``$ on $`๐_s`$, together with a fixed isomorphism
$$\phi :i_s^{}\stackrel{}{}\tau _s^{}๐ฒ,$$
on the divisor $`D_s`$ at infinity, for some $`๐ฒCoh^\mathrm{\Gamma }(C)`$.
###### Remark 2.5.
If $`\pi :๐\{\}`$ is a (non-commutative) projective space over a point, the $`\pi `$-free sheaves are exactly the torsion free ones (compare \[3, Section 2\]). To see this, note that a $`\pi `$-free sheaf is certainly torsion free, since it embeds into a locally free sheaf. Conversely, a torsion free sheaf embeds into some locally free sheaf, which in turn embeds into some $`๐ช_๐^m(n)`$.
###### Lemma 2.6.
If $``$ is $`\pi _s`$-free, then $`L^ji_s^{}=0`$ for $`j>0`$.
#### Proof
As in the commutative case, the structure sheaf $`i_s๐ช_{D_s}`$ has a resolution
$$0๐ช_{๐_s}(1)\stackrel{z}{}๐ช_{๐_s}i_s๐ช_{D_s}0,$$
which implies that $`L^ji_s^{}=0`$ for $`j>1`$ for any $`Coh(๐_s)`$, and also that $`L^1i_s^{}`$ is left exact. If $``$ is $`\pi _s`$-free, applying the latter to an embedding $`\pi _s^{}(๐ฐ)(n)`$ gives the vanishing of $`L^1`$ also. $`\mathrm{}`$
## 3. Threefolds: the results
### 3.1. Twisted quiver representations and quiver sheaves
Recall that, given a quiver with arrows $`ab`$ marked by objects $`O_{ab}๐`$ of an abelian tensor category $`๐`$, a representation of the marked quiver in $`๐`$ consists of a set of objects $`O_a`$ of $`๐`$ associated to nodes, and a set of morphisms $`\phi _{ab}Hom_๐(O_aO_{ab},O_b)`$ associated to the arrows $`ab`$. Representations of a marked quiver in the category $`Coh(X)`$ of an algebraic variety $`X`$ are also called quiver sheaves on $`X`$.
In the specific context of classifying holomorphic $`D`$-branes on the threefold $`X`$ and its deformations, the following quiver marked in $`Coh(C)`$ will arise naturally. The quiver is the standard extended McKay quiver of , obtained from the original one by adding an extra leaf at each node with arrows in both directions. Using the data of the vector bundle $`๐ฌ`$ on $`C`$, we mark this quiver in $`Coh(C)`$ as follows.
* The marked $`A_n`$ quiver for $`n>1`$ is illustrated on Figure 3.1; recall that in this case, there is a decomposition $`๐ฌ=๐ฌ_1๐ฌ_2`$ into a sum of line bundles, since the structure group of $`๐ฌ`$ reduces to the diagonal torus.
* The marked $`A_1`$ quiver consists of only two nodes $`0`$ and $`1`$ and two arrows $`01,10`$ marked by the rank-two bundle $`๐ฌ^{}`$, as well as leaves marked as in the higher $`A_n`$ case.
* For types $`D`$ and $`E`$, arrows between nodes are all marked by the line bundle $`๐ฌ_0^{}`$, where $`๐ฌ=๐ฌ_0^2`$; leaves are marked as before.
### 3.2. The main classification result
###### Theorem 3.1.
Given $`sH^0(C,)`$, there is a 1-to-1 correspondence between the following sets of data.
1. Isomorphism classes of framed $`\pi _s`$-free sheaves $`(,\phi )`$ on $`(๐_s,D_s)`$.
2. Quintuples $`(๐ฑ,๐ฒ,,,๐ฅ)`$, where $`๐ฒ,๐ฑ`$ are coherent $`\mathrm{\Gamma }`$-sheaves on $`C`$, and
$$\begin{array}{ccc}\hfill & & Hom_C^\mathrm{\Gamma }(๐ฑ๐ฌ^{},๐ฑ),\hfill \\ & & \\ \hfill & & Hom_C^\mathrm{\Gamma }(๐ฒ,๐ฑ),\hfill \\ & & \\ \hfill ๐ฅ& & Hom_C^\mathrm{\Gamma }(๐ฑdet๐ฌ^{},๐ฒ),\hfill \end{array}$$
satisfying the following two conditions:
1. the ADHM relation
$$+๐ฅ+s=0Hom_C^\mathrm{\Gamma }(๐ฑdet๐ฌ^{},๐ฑ),$$
where
$$H^0(C,Z(\mathrm{\Gamma })det๐ฌ)Hom_C^\mathrm{\Gamma }(๐ฑdet๐ฌ^{},๐ฑ)$$
is the natural embedding as the central subspace;
2. non-degeneracy: if $`๐ฑ^{}๐ฑ`$ is a $`\mathrm{\Gamma }`$-subsheaf such that $`(๐ฑ^{}๐ฌ^{})๐ฑ^{}`$ and $`๐ฒ๐ฑ^{}`$, then $`๐ฑ^{}=๐ฑ`$.
Sets of quintuples are identified under the action of invertible elements of $`Hom_C^\mathrm{\Gamma }(๐ฑ,๐ฑ)`$.
3. Representations $`(\{๐ฑ_a\},\{๐ฒ_a\},\{_{ab}\},\{_a\},\{๐ฅ_a\})`$ in $`Coh(C)`$ of the marked McKay-type quiver introduced in 3.1, satisfying
1. the ADHM relations
$$\underset{b}{}ฯต_{ab}_{ba}_{ab}+_a๐ฅ_a+s_a=0Hom_C(๐ฑ_adet๐ฌ^{},๐ฑ_a)$$
at each node $`a`$, where $`ฯต_{ab}\{\pm 1\}`$ is a standard assingment of signs to arrows with $`ฯต_{ab}=ฯต_{ba}`$, and $`s_a=P_{\eta _a}(s)`$ is the projected section corresponding to the simple root $`\eta _a`$, and
2. non-degeneracy: if $`\{๐ฑ_a^{}\}`$ is a $``$-invariant set of subsheaves containing the images of $`_a`$โs, then $`๐ฑ_a^{}=๐ฑ_a`$ at all nodes.
Two representations are identified under invertible elements of $`_aHom_C(๐ฑ_a,๐ฑ_a)`$.
If $`s๐ฎ=H^0(C,_0)`$ is a deformation parameter of the threefold $`X=X_0`$, then the same data also parametrizes
1. certain objects in $`๐(CohX_s)`$, the derived category of coherent sheaves on $`X_s`$.
#### Proof
The equivalence $`(1)(2)`$ follows from a version of the relative Beilinson resolution for the non-commutative projective bundle $`๐_sC`$; details are given in Section 4.1. McKayโs definition of the quiver describing the representation theory of $`\mathrm{\Gamma }`$ implies $`(2)(3)`$ in the standard way. Finally the mapping $`(1)(4)`$ in the geometric case $`s๐ฎ=H^0(C,_0)`$ is given by a derived equivalence to be discussed in Section 4.2. $`\mathrm{}`$
###### Remark 3.2.
As $`X=X_0`$ and its deformations $`X_s`$ for $`s๐ฎ`$ are not projective, one needs to rigidify before holomorphic $`D`$-branes, in other words objects in $`๐^b(X_s)`$, have a sensible moduli space. For the central fibre $`X=X_0`$, a crepant resolution of the singular threefold $`๐ฌ/\mathrm{\Gamma }`$, one has a derived equivalence
$$๐(X_0)๐^\mathrm{\Gamma }(๐ฌ)$$
between the derived categories of coherent sheaves on $`X_0`$ and that of $`\mathrm{\Gamma }`$-equivariant sheaves on the total space of the bundle $`๐ฌC`$. One can easily rigidify on the latter by considering $`\mathrm{\Gamma }`$-sheaves on the projective bundle $`๐_0=(๐ฌ๐ช_C)C`$, framed on the divisor at infinity $`D_0=(๐ฌ)๐_0`$. Theorem 3.1 is the appropriate generalization of this approach which also works for deformations: for the analogous problem on $`X_s`$, we consider framed sheaves on the non-commutative projective bundle $`๐_sC`$.
In the surface case, this approach was used earlier in . To quote the result, let $`\lambda Z(\mathrm{\Gamma })`$. Then for $`\mathrm{\Gamma }`$-modules $`V,W`$, Nakajimaโs non-singular quiver variety $`_{V,W,\lambda }`$ parametrizes torsion free sheaves on the non-commutative space $`_\lambda ^2=ProjP_\lambda `$, framed on the commutative $`\mathrm{\Gamma }`$-line at $`\mathrm{}`$. This statement generalizes earlier work of and others. The origin of all such results is of course the ADHM classification of finite-action $`\mathrm{SU}(dim(W))`$-instantons on $`^4`$ of charge $`dim(V)`$.
### 3.3. Some holomorphic $`D`$-branes on $`ADE`$ fibrations over $`๐ธ^1`$
If $`C๐ธ^1`$, Theorem 3.1 can in some cases be re-written in terms of classical quiver representations: representations of a quiver in vector spaces. This will give an interpretation of an assertion of .
Recall that for $`C๐ธ^1`$, a deformation parameter $`s๐ฎ`$ of the central fibre $`X_0=๐ธ^1\times Y`$ can be specified by a set of polynomials $`\{\mathrm{\Theta }_a[t]:a\mathrm{\Delta }\}`$ indexed by the vertices of the affine quiver, subject to $`_a\delta _a\mathrm{\Theta }_a=0`$. The following definition is due to CachazoโKatzโVafa .
###### Definition 3.3.
The affine $`N=1`$ $`ADE`$ quiver is the McKay quiver extended by a loop $`aa`$ at each vertex. For a (finite-dimensional) representation $`(\{V_a\},\{B_{ab}\},\{\mathrm{\Psi }_a\})`$ of this quiver, the ADHM-type relations are
(3.1)
$$\underset{b}{}ฯต_{ab}B_{ba}B_{ab}+\mathrm{\Theta }_a(\mathrm{\Psi }_a)=0Hom(V_a,V_a)$$
at each vertex $`a\mathrm{\Delta }`$ of the quiver, where $`\mathrm{\Theta }_a(\mathrm{\Psi }_a)`$ is to be interpreted as the evaluation of a polynomial on an endomorphism of $`V_a`$, as well as
(3.2)
$$\mathrm{\Psi }_aB_{ba}=B_{ba}\mathrm{\Psi }_bHom(V_a,V_b)$$
along each arrow $`ab`$ of the quiver $`\mathrm{\Delta }`$.
Consider quadruples $`(\{V_a\},\{B_{ab}\},\{\mathrm{\Psi }_a\},๐ฏ_\mathrm{๐})`$, where $`(\{V_a\},\{B_{ab}\},\{\mathrm{\Psi }_a\})`$ is a representation of the affine $`N=1`$ $`ADE`$ quiver satisfying the ADHM-type relations, and $`๐ฏ_\mathrm{๐}V_0`$ is a fixed vector in the vector space attached to the affine node. Call a quadruple non-degenerate if there is no $`(B,\mathrm{\Psi })`$-invariant collection of subspaces $`\{V_a^{}V_a\}`$ with $`๐ฏ_\mathrm{๐}V_0^{}`$.
###### Theorem 3.4.
Equivalence classes of non-degenerate quadruples $`(\{V_a\},\{B_{ab}\},\{\mathrm{\Psi }_a\},๐ฏ_\mathrm{๐})`$ satisfying the ADHM relations, identified under the action of $`_a\mathrm{GL}(V_a)`$, parametrize certain objects in $`๐(CohX_s)`$, holomorphic $`D`$-branes on the threefold $`X_s`$.
#### Proof
Quiver sheaf data on $`C`$ parametrize certain branes on $`X_s`$ by Theorem 3.1. The correspondence between representations of the $`N=1`$ $`ADE`$ quiver and a special class of quiver sheaf data will be discussed in Section 4.3. $`\mathrm{}`$
###### Remark 3.5.
As explained in , the quiver relations (3.1)-(3.2) come from the natural superpotential of the quiver gauge theory on $`\mathrm{\Delta }`$, involving adjoint fields $`\mathrm{\Psi }_a`$ as well as bifundamental fields $`B_{ab}`$.
###### Remark 3.6.
Let the finite $`N=1`$ $`ADE`$ quiver be obtained from the affine one by deleting the affine node. Representations of the finite $`N=1`$ $`ADE`$ quiver, satisfying the ADHM-type relations (3.1)-(3.2), parametrize holomorphic $`D`$-branes supported on exceptional fibres of $`f_s:X_s\overline{X}_s`$. This follows from the statement that the vanishing of the affine component of $`๐ฑ`$ forces all other $`๐ฑ_a`$ be supported on points $`PC`$ at which some projected section $`p_\eta (s)`$ vanishes for some positive root $`\eta R_+`$, in other words on points of the base curve over which the surface fiber $`\pi _s^1(P)`$ contains exceptional curves. Observing that the section $`sH^0(C,(\mathrm{\Gamma })det๐ฌ)`$ is central in $`Hom_C^\mathrm{\Gamma }(๐ฑdet๐ฌ^{},๐ฑ)`$, so commutes with all components of $``$, the latter statement is essentially proved in \[6, 4.1โ4.2\]. This establishes a direct link to , according to which (in the generic case) irreducible representations of the finite $`N=1`$ quiver with the given relations parametrize exceptional components of the contraction $`f_s:X_s\overline{X}_s`$.
## 4. Proofs
### 4.1. The Beilinson argument
The aim of this section is to prove of the equivalence $`(1)(2)`$ of the classification result Theorem 3.1 via an analysis of framed $`\pi _s`$-free sheaves on $`๐_s`$.
Given $`sH^0(C,)`$, recall the sheaf of algebras $`๐ซ_s`$ on the curve $`C`$, and the associated non-commutative bundle $`\pi _s:๐_sC`$. Define $`๐ซ_s`$-modules $`๐ฏ_i`$ by
(4.3)
$$\begin{array}{c}๐ฏ_0=๐ช_{๐_s},\\ \\ 0๐ช_{๐_s}\pi _s^{}(๐ฌ๐ช_C)(1)๐ฏ_10,\\ \\ ๐ฏ_2=\pi _s^{}(det๐ฌ)(3).\end{array}$$
###### Proposition 4.1.
A $`\pi _s`$-free sheaf $``$ on $`๐_s`$, framed on the divisor $`D_s`$, is the cohomology of a monad
$$\pi _s^{}Ext_C^1(๐ฏ_2(1),)(1)\pi _s^{}Ext_C^1(๐ฏ_1,)\pi _s^{}Ext_C^1(๐ฏ_0(1),)(1)$$
of $`๐ซ_s`$-modules.
#### Proof
Given a $`๐ซ_s`$-module $``$, a Koszul duality argument, in an analogous way to the absolute case in \[3, Section 7\] following \[4, Thm 2.6.1\], leads to a Beilinson-type spectral sequence with $`E_1`$ term
$$E_1^{p,q}=\pi _s^{}Ext_C^q(๐ฏ_p(p),)(p),$$
nonzero only for $`2p0,0q2`$, converging to $``$ in the limit. The vanishing results
$$Ext_C^q(๐ฏ_p(p),(1))=0\text{ for }q=0,2,p=1,2$$
which follow from the existence of the framing of $``$ on the divisor $`D_s`$ (compare \[13, Lemma 6.2\]\[3, Lemma 4.2.12\]), reduce the spectral sequence for $`=(1)`$ to the monad given in the statement. Details are left to the reader. $`\mathrm{}`$
We also record an auxiliary result.
###### Lemma 4.2.
There are natural isomorphisms
$$Hom_{๐_s}(\pi _s^{}det๐ฌ^{},\pi _s^{}๐ฌ^{}(1))Hom_{๐_s}(\pi _s^{}๐ฌ^{},๐ช_{๐_s}(1))Hom_C^\mathrm{\Gamma }(๐ฌ^{},๐ฌ^{}๐ช_C).$$
#### Proof
The first isomorphism follows from Lemma 4.3 below. The second one follows from adjunction for the pair $`(\pi _s^{},\pi _s)`$, together with
$$\pi _s๐ช_{๐_s}(1)๐ซ_{s,1}(๐ฌ^{}๐ช_C)\mathrm{\Gamma }Coh^\mathrm{\Gamma }(C),$$
an identity well known from the commutative context. $`\mathrm{}`$
###### Lemma 4.3.
Let $`๐ฌ`$ be a rank-two bundle on a (commutative) space. Then there is a natural isomorphism
$$๐ฌdet๐ฌ^{}๐ฌ^{}.$$
#### Proof
The embedding $`\iota :det๐ฌ^{}๐ฌ^{}๐ฌ^{}`$ induces a natural map
$$Hom(det๐ฌ^{},det๐ฌ^{})Hom(det๐ฌ^{},(๐ฌ^{})^2)Hom(๐ฌdet๐ฌ^{},๐ฌ^{}).$$
The image of the identity of the first $`Hom`$-group gives a natural morphism as in the statement, which can be checked on a local basis to be an isomorphism. $`\mathrm{}`$
Now return to the context of the classification result Theorem 3.1, and consider a quintuple $`(๐ฑ,๐ฒ,,,๐ฅ)`$ as in Theorem 3.1(2); recall that
$$๐ฒ,๐ฑCoh^\mathrm{\Gamma }(C),$$
and
$$\begin{array}{ccc}\hfill & & Hom_C^\mathrm{\Gamma }(๐ฑ๐ฌ^{},๐ฑ),\hfill \\ & & \\ \hfill & & Hom_C^\mathrm{\Gamma }(๐ฒ,๐ฑ),\hfill \\ & & \\ \hfill ๐ฅ& & Hom_C^\mathrm{\Gamma }(๐ฑdet๐ฌ^{},๐ฒ).\hfill \end{array}$$
Let
$$cHom_{๐_s}(\pi _s^{}det๐ฌ^{},\pi _s^{}๐ฌ^{}(1)),dHom_{๐_s}(\pi _s^{}๐ฌ^{},๐ช_{๐_s}(1))$$
denote the images, under the isomorphisms of Lemma 4.2, of the canonical element
$$\mathrm{Id}Hom_C^\mathrm{\Gamma }(๐ฌ^{},๐ฌ^{})Hom_C^\mathrm{\Gamma }(๐ฌ^{},๐ฌ^{}๐ช_C)$$
Note also that we have a fixed section
$$zHom_{๐_s}(๐ช_{๐_s},๐ช_{๐_s}(1)).$$
Define
$$a=\left(\begin{array}{c}\pi _s^{}\left((\mathrm{Id}_๐ฑ\iota )\right)z\pi _s^{}(\mathrm{Id}_๐ฑ)c(1)\\ \pi _s^{}(๐ฅ)z\end{array}\right):\pi _s^{}(๐ฑdet๐ฌ^{})(1)\pi _s^{}(๐ฑ๐ฌ^{}๐ฒ)$$
where $`\iota :det๐ฌ^{}(๐ฌ^{})^2`$ is the natural map. Define similarly
$$b=\left(\begin{array}{cc}\pi _s^{}()z+\pi _s^{}(\mathrm{Id}_๐ฑ)d& \pi _s^{}()z\end{array}\right):\pi _s^{}(๐ฑ๐ฌ^{}๐ฒ)\pi _s^{}(๐ฑ)(1),$$
to obtain the chain of morphisms
(4.4)
$$\pi _s^{}(๐ฑdet๐ฌ^{})(1)\stackrel{a}{}\pi _s^{}(๐ฑ๐ฌ^{}๐ฒ)\stackrel{b}{}\pi _s^{}(๐ฑ)(1).$$
The following result completes the proof of the equivalence $`(1)(2)`$ of the classification result Theorem 3.1.
###### Proposition 4.4.
If the quintuple satisfies the ADHM relation, then (4.4) is a complex of $`๐ซ_s`$-modules. Furthermore, it is a monad defining a framed $`\pi _s`$-free sheaf $``$ if and only if the quintuple $`(๐ฑ,๐ฒ,,,๐ฅ)`$ is non-degenerate. Conversely, every $`\pi _s`$-free $`๐ซ_s`$-module $``$, framed on $`D_s`$, arises from this construction.
#### Proof
The standard direct computation shows that $`ba=0`$ is equivalent to the ADHM relation. The proof of the equivalence of the monad property and non-degeneracy is analogous to the absolute case \[3, Section 4.1\]. For the converse, given a framed sheaf $`(,\phi )`$, let $`๐ฑ=Ext_C^1(๐ช_{๐_s}(1),)`$. Then by Proposition 4.1, $``$ is the middle cohomology of the monad
$$\pi _s^{}(๐ฑdet๐ฌ^{})(1)\pi _s^{}Ext_C^1(๐ฏ_1,)\pi _s^{}๐ฑ(1).$$
The usual arguments \[13, Theorem 6.7\] show that, since $``$ is framed on $`D_s`$, this monad is isomorphic to a monad of the form (4.4) for some quintuple $`(๐ฑ,๐ฒ,,,๐ฅ)`$. $`\mathrm{}`$
### 4.2. A derived equivalence
In this section we complete the proof of Theorem 3.1 by establishing the missing link $`(1)(4)`$.
###### Proposition 4.5.
Let $`s๐ฎ`$ be a deformation parameter of the central fibre $`X=X_0`$. There is a distinguished equivalence of triangulated categories
$$๐(CohX_s)๐(Mod๐_s),$$
where $`Mod๐_s`$ is the category of sheaves of finitely generated right $`๐_s`$-modules, and $`๐()`$ denotes the bounded derived category on both sides.
#### Proof
This assertion is a fibered version of the analogous two-dimensional equivalence proved in , and the proof carries over verbatim. A deformation argument starting from the central fibre $`X=X_0`$ shows that a certain specific component $`M_s`$ of a fine moduli space of torsion sheaves on $`๐_s`$ maps by a semi-small birational map to the singular variety $`\overline{X}_s`$. By , generalizing an argument of , this implies that $`M_s`$ is a crepant resolution of $`\overline{X}_s`$, and one has a derived equivalence
$$๐(CohM_s)๐(Mod๐_s)$$
defined by the universal sheaf. But since $`X_s`$ is the unique crepant resolution of $`\overline{X}_s`$, necessarily $`M_sX_s`$ and the proposition follows. Details are left to the reader. $`\mathrm{}`$
This equivalence gives the mapping $`(1)(4)`$ of Theorem 3.1 from framed $`\pi _s`$-free sheaves on $`๐_s`$ to objects in $`๐(CohX_s)`$. Indeed, a right $`๐ซ_s`$-module can be restricted to the affine part $`๐_s`$ to give a right $`๐_s`$-module, and then mapped using the derived equivalence to an object in $`๐(CohX_s)`$, in other words a holomorphic $`D`$-brane on $`X_s`$.
### 4.3. Fibrations over the affine line
In this section, we take a fibration $`X_sC๐ธ^1`$ and discuss the proof of Theorem 3.4. From Theorem 3.1, we know that certain holomorphic $`D`$-branes on $`X_s`$ are classified by non-degenerate quintuples $`(๐ฑ,๐ฒ,,i,j)`$ satisfying the ADHM equation. Consider the subclass of representations in $`Coh(๐ธ^1)`$ with the simplest possible framing $`๐ฒ๐ช_{๐ธ^1}`$ and $`๐ฑ`$ a torsion $`\mathrm{\Gamma }`$-sheaf on $`๐ธ^1`$. It follows that $`๐ฅ=0`$ and $`H^0(๐ธ^1,๐ฑ^\mathrm{\Gamma })`$. Decompose $`๐ฑ`$ and the map $`B`$ into $`\mathrm{\Gamma }`$-components to obtain torsion sheaves $`๐ฑ_a`$ and sheaf homomorphisms $`_{ab}:๐ฑ_a๐ฑ_b`$ indexed by nodes and edges of the McKay quiver.
Set $`V_a=H^0(๐ธ^1,๐ฑ_a)`$, and let $`B_{ab}=H^0(_{ab}):V_aV_b`$ to be the map on global sections induced by $`_{ab}`$. Let $`๐ฏ_\mathrm{๐}V_0`$ be the section corresponding to $``$. Let also $`\mathrm{\Psi }_a:V_aV_a`$ be the map induced by multiplication by the section $`tH^0(๐ธ^1,๐ช_{๐_1})[t]`$. Theorem 3.4 follows from Theorem 3.1, together with
###### Proposition 4.6.
The map
$$(๐ฑ,๐ช_C,,0,0)(\{V_a\},\{B_{ab}\},\{\mathrm{\Psi }_a\},๐ฏ_\mathrm{๐}V_0)$$
sets up a one-to-one correspondence from this restricted set of quiver ADHM data to representations of the affine $`N=1`$ ADE quiver satisfying the relations (3.1)-(3.2).
#### Proof
Given $`(๐ฑ,)`$, the edge relations (3.2) $`\mathrm{\Psi }_aB_{ba}=B_{ba}\mathrm{\Psi }_b`$ for the data $`(\{V_a\},\{B_{ab}\},\{\mathrm{\Psi }_a\})`$ hold by definition. Further, the ADHM equation for $`(๐ฑ,)`$ is
$$+s=0Hom(๐ฑ,๐ฑdet๐ฌ),$$
which in $`\mathrm{\Gamma }`$-components says that
$$\underset{b}{}ฯต_{ab}_{ba}_{ab}+s_a=0Hom(๐ฑ_a,๐ฑ_a).$$
Replacing $`s_a`$ by the polynomial $`\mathrm{\Theta }_a`$, and remembering that the effect of $`tH^0(๐ช_{๐ธ^1})`$ on $`H^0(๐ฑ)`$ is exactly $`\mathrm{\Psi }_a`$, for global sections we obtain
$$\underset{b}{}ฯต_{ab}B_{ba}B_{ab}+\mathrm{\Theta }_a(\mathrm{\Psi }_a)=0Hom(V_a,V_a)$$
which is exactly relation (3.1) for the node $`a`$.
Conversely, given a representation $`(\{V_a\},\{B_{ab}\},\{\mathrm{\Psi }_a\},๐ฏ_\mathrm{๐}V_0)`$ of the $`N=1`$ ADE quiver, define torsion sheaves attached to the nodes by
$$\begin{array}{ccc}๐ฑ_a=coker(V_a๐ช_{๐ธ^1}& \stackrel{1t\mathrm{\Psi }_a1}{}& V_a๐ช_{๐ธ^1}).\end{array}$$
Using Lemma 4.7 below, for adjacent nodes $`a,b`$ we have a diagram
$$\begin{array}{cccccccccccccc}0& & & & V_a๐ช_{๐ธ^1}& & \stackrel{1t\mathrm{\Psi }_a1}{}& & V_a๐ช_{๐ธ^1}& & & ๐ฑ_a& & 0\\ & & & & B_{ab}1& & & & B_{ab}1& & \\ 0& & & & V_b๐ช_{๐ธ^1}& & \stackrel{1t\mathrm{\Psi }_b1}{}& & V_b๐ช_{๐ธ^1}& & & ๐ฑ_b& & 0\end{array}$$
which, by commutativity $`\mathrm{\Psi }_aB_{ba}=B_{ba}\mathrm{\Psi }_b`$, induces a map $`_{ab}:๐ฑ_a๐ฑ_b`$. The converse of the above argument shows that the ADHM relation follows from the relations (3.1). By Lemma 4.7, the two constructions are inverses to each other. $`\mathrm{}`$
The proof used the elementary
###### Lemma 4.7.
Given a torsion sheaf $`๐ฑ`$ on $`๐ธ^1=Spec[t]`$, let $`V=H^0(๐ธ^1,๐ฑ)`$ and let $`\mathrm{\Psi }:VV`$ be the map given by multiplication by $`tH^0(๐ช_{๐ธ^1})`$. Then the sequence of sheaves
$$\begin{array}{ccccccccc}0& & V๐ช_{๐ธ^1}& \stackrel{1t\mathrm{\Psi }1}{}& V๐ช_{๐ธ^1}& \stackrel{c}{}& ๐ฑ& & 0\end{array}$$
is exact on $`๐ธ^1`$, where $`c:H^0(๐ฑ)๐ช_{๐ธ^1}๐ฑ`$ is the canonical map. Conversely, given a vector space with an endomorphism $`(V,\mathrm{\Psi })`$, the exact sequence defines a torsion sheaf $`๐ฑ`$ on $`๐ธ^1`$, and the two constructions are mutual inverses.
###### Remark 4.8.
In this Lemma, $`๐ฑ๐ช_Z`$ is a structure sheaf of a $`0`$-dimensional subscheme $`Z๐ธ^1`$ if and only if $`\mathrm{\Psi }`$ is a regular endomorphism. Their moduli space is
$$\mathrm{Mat}(n,)//\mathrm{GL}(n,)\{\text{regular endomorphisms}\}/\mathrm{GL}(n,)๐ธ^n(๐ธ^1)^{[n]},$$
where the map is given by taking the coefficients of the characteristic polynomial of $`\mathrm{\Psi }`$, which is also the equation of the corresponding subscheme.
### Acknowledgements
Thanks to Sheldon Katz, Eduard Looijenga, Tom Nevins and Tony Pantev for helpful remarks and correspondence. Special thanks to Ian Grojnowski for many conversations on subjects related to this paper. Support by a European Union Marie Curie Individual Fellowship and by OTKA grant $`\mathrm{\#}046878`$ is also gratefully acknowledged.
Department of Mathematics, Utrecht University
Current address:
Mathematical Institute, University of Oxford
E-mail address: szendroi@maths.ox.ac.uk
|
warning/0506/math0506574.html
|
ar5iv
|
text
|
# Rational, Replacement and Local Invariants of a Group Action
## 1 Introduction
We present algebraic constructions for invariants of a rational group action on an affine space, and relate them to their counterparts in differential geometry. The constructions are algorithmic and can easily be implemented in general purpose computer algebra systems or software specialized in Grรถbner basis computations. This is illustrated by the maple worksheet available at http://www.inria.fr/cafe/Evelyne.Hubert/Publi/rrl\_invariants.html where the examples of the paper are treated.
The first construction is for the computation of a generating set of rational invariants. This generating set is endowed with a simple algorithm to express any rational invariant in terms of them. The construction comes into two variants. In the first one we consider the ideal of the graph of the action as did Rosenlicht , Vinberg & Popov <sup>1</sup><sup>1</sup>1 We are indebted to a referee of the MEGA conference for pointing out this reference that motivated us to push further some of the results. , and Mรผller-Quade & Beth <sup>2</sup><sup>2</sup>2 We would like to thank H. Derksen for suggesting comparison with this reference after we made public a first preprint. . We point out the connections with these previous works in the text. Our proofs are independent and provide an original approach. We show that the coefficients of a reduced Grรถbner basis of the ideal of the graph of the action are invariant. We prove that these coefficients generate the field of rational invariants by exhibiting an algorithm for rewriting any rational invariant in terms of them. The second variant provides a purely algebraic formulation of the geometric construction of a *fundamental set of local invariants* on a smooth manifold proposed by Fels and Olver , as a generalization of Cartanโs moving frame method. It is also computationally more effective as we reduce to zero the dimension of the polynomial ideal for which a reduced Grรถbner basis is computed. This is achieved by adding the ideal of a cross-section to the ideal of the graph.
That latter construction allows to introduce *replacement invariants*, the algebraic counterpart of *normalized invariants* appearing in the geometric construction. A replacement invariant is a tuple of algebraic of functions of rational invariants. Any invariant can be trivially rewritten in their terms by substituting the coordinate functions by the corresponding invariants from this tuple. An *invariantization* map, a computable isomorphism from the set of algebraic functions on the cross-section to the set of algebraic invariants, is defined in terms of replacement invariants.
We use invariantization process to make explicit the connection between the present algebraic construction and the geometric construction of Fels and Olver . We introduce an alternative definition of smooth invariantization which, on one hand, generalizes the one given in and, on the other hand, matches the algebraic construction. We thus provide a bridge between the theory of rational and algebraic invariants and the theory of smooth local invariants in differential geometry.
Diverse fields of application of algebraic invariant theory are presented in \[9, Chapter 5\]. Some of the applications can be addressed with rational invariants. Their present construction together with the simple rewriting algorithm can bring computational benefits. An application of the moving frame method to classical invariant theory was proposed in . In these works, however, the geometric formulation of the method is used without adapting it to the algebraic nature of the problem. A purely algebraic formulation of the moving frame method opens new possibilities of its application in classical invariant theory.
The present algebraic formulation provides a new tool for the investigation of the differential invariants of Lie group actions and their applications to differential systems in the line of . This larger project motivates our choice to consider rational actions. Even if we start with an affine or even linear action on the zeroth order jet space, the prolongation of the action to the higher order jet spaces is usually rational.
The paper is structured as follows. In Section 2 we introduce the action of an algebraic group on the affine space and the graph of the action. This leads to a first construction of a set of generating rational invariants. A second version of the construction is given after the introduction of the cross-section to the orbits in Section 3. This second construction gives rise to the replacement invariants in Section 4, which are used to define a computable invariantization map. In Section 5 we present a geometric construction of local smooth invariants that generalizes the construction of and explicitly relates it to the algebraic construction of the previous sections. Section 6 provides additional examples.
Acknowledgments: We would like to thank Liz Mansfield, Peter Olver and Agnes Szanto for discussing the ideas of the paper during the workshop โDifferential Algebra and Symbolic Computationโ in Raleigh, April 2004, sponsored in part by NSF grants CCR-0306406 and CCF-0347506. We are grateful to Michael Singer for continuing discussion of the project and a number of valuable suggestions.
## 2 Graph of a group action and rational invariants
We give a definition of a rational action of an algebraic group over a field $`๐`$ on an affine space, and formulate two additional hypotheses necessary to our construction. We recall the definition for the graph of the action. It plays a central role in our constructions. The first variant of the algorithm for constructing a generating set of rational invariants, together with an algorithm for expressing any rational invariant in terms of them, is presented in this section.
For exposition convenience we assume that the field $`๐`$ is algebraically closed. The construction proposed in this section relies only on Grรถbner basis computations and thus can be performed in the field of definition of the data (usually $``$ or $`๐ฝ_p`$). Outside of Section 5 the terms *open*, *close* and *closure* refer to the Zariski topology.
### 2.1 Rational action of an algebraic group
We consider an algebraic group that is defined as an algebraic variety $`๐ข`$ in the affine space $`๐^l`$. The group operation and the inverse are given by polynomial maps. The neutral element is denoted by $`e`$. We shall consider an action of $`๐ข`$ on an affine space $`๐ต=๐^n`$.
Throughout the paper $`\lambda =(\lambda _1,\mathrm{},\lambda _l)`$ and $`z=(z_1,\mathrm{},z_n)`$ denote indeterminates while $`\overline{\lambda }=(\overline{\lambda }_1,\mathrm{},\overline{\lambda }_l)`$ and $`\overline{z}=(\overline{z}_1,\mathrm{},\overline{z}_n)`$ denote points in $`๐ข๐^l`$ and $`๐ต=๐^n`$ respectively. The coordinate ring of $`๐ต`$ and $`๐ข`$ are respectively $`๐[z_1,\mathrm{},z_n]`$ and $`๐[\lambda _1,\mathrm{},\lambda _l]/G`$ where $`G`$ is a radical unmixed dimensional ideal. By $`\overline{\lambda }\overline{\mu }`$ we denote the image of $`(\overline{\lambda },\overline{\mu })`$ under the group operation while $`\overline{\lambda }^1`$ denotes the image of $`\overline{\lambda }`$ under the inversion map.
###### Definition 2.1
A rational action of an algebraic group $`๐ข`$ on the affine space $`๐ต`$ is a rational map $`g:๐ข\times ๐ต๐ต`$ that satisfies the following two properties
1. $`g(e,\overline{z})=\overline{z}`$, $`\overline{z}๐ต`$
2. $`g(\overline{\mu },g(\overline{\lambda },z))=g(\overline{\mu }\overline{\lambda },z)`$, whenever both $`(\overline{\lambda },\overline{z})`$ and $`(\overline{\mu }\overline{\lambda },\overline{z})`$ are in the domain of definition of $`g`$.
A rational action is thus uniquely determined by a $`n`$-tuple of rational functions of $`๐(\lambda ,z)`$ whose domain of definition is a dense open set of $`๐ข\times ๐ต`$. We can bring these rational functions to their least common denominator $`h๐[\lambda ,z]`$ without affecting the domain of definition. In the rest of the paper the action is thus given by
$$g(\overline{\lambda },\overline{z})=(g_1(\overline{\lambda },\overline{z}),\mathrm{},g_n(\overline{\lambda },\overline{z}))\text{ for }g_1,\mathrm{},g_nh^1๐[\lambda _1,\mathrm{},\lambda _l,z_1,\mathrm{},z_n]$$
(1)
###### Asumption 2.2
We make the additional assumptions
1. for all $`\overline{z}๐ต`$, $`h(\lambda ,\overline{z})๐[\lambda ]`$ is not a zero-divisor of $`G`$. This says that the domain of definition of $`g_{\overline{z}}:\overline{\lambda }g(\overline{\lambda },\overline{z})`$ contains a non-empty open set of each component of $`๐ข`$.
2. for all $`\overline{\lambda }๐ต`$, $`h(\overline{\lambda },z)๐[z]`$ is different from zero. In other words, for every element $`\overline{\lambda }๐ข`$ there exists $`\overline{z}๐ต`$, such that $`(\overline{\lambda },\overline{z})`$ is in the domain of definition $`g`$.
The following three examples serve as illustration throughout the text.
###### Example 2.3
Scaling. Consider the multiplicative group given by $`G=(1\lambda _1\lambda _2)๐[\lambda _1,\lambda _2]`$. The neutral element is $`(1,1)`$ and $`(\overline{\mu }_1,\overline{\mu }_2)(\overline{\lambda }_1,\overline{\lambda }_2)^1=(\overline{\mu }_1\overline{\lambda }_2,\overline{\mu }_2\overline{\lambda }_1)`$. We consider the scaling action of this group on $`๐^2`$. It is given by the following polynomials of $`๐[\lambda _1,\lambda _2,z_1,z_2]`$: $`g_1=\lambda _1z_1,g_2=\lambda _1z_2.`$
###### Example 2.4
translation+reflection. Consider the group that is the cross product of the additive group and the group of two elements $`\{1,1\}`$, its defining ideal in $`๐[\lambda _1,\lambda _2]`$ being $`G=(\lambda _2^21)`$. The neutral element is $`(0,1)`$ while $`(\overline{\mu }_1,\overline{\mu }_2)(\overline{\lambda }_1,\overline{\lambda }_2)^1=(\overline{\mu }_1\overline{\lambda }_1,\overline{\mu }_2\overline{\lambda }_2).`$ We consider its action on $`๐^2`$ as translation parallel to the first coordinate axis and reflection w.r.t. this axis. It is defined by the following polynomials of $`๐[\lambda _1,\lambda _2,z_1,z_2]`$: $`g_1=z_1+\lambda _1,g_2=\lambda _2z_2.`$
###### Example 2.5
rotation. Consider the special orthogonal group given by $`G=(\lambda _1^2+\lambda _2^21)๐[\lambda _1,\lambda _2]`$ with $`e=(1,0)`$ and $`(\overline{\mu }_1,\overline{\mu }_2)(\overline{\lambda }_1,\overline{\lambda }_2)^1=(\overline{\mu }_1\overline{\lambda }_1+\overline{\mu }_2\overline{\lambda }_2,\overline{\mu }_2\overline{\lambda }_1\overline{\mu }_1\overline{\lambda }_2).`$ Its linear action on $`๐^2`$ is given by the following polynomials of $`๐[\lambda _1,\lambda _2,z_1,z_2]`$:
$$g_1=\lambda _1z_1\lambda _2z_2,g_2=\lambda _2z_1+\lambda _1z_2.$$
An element of the group acts as a rotation around the origin.
### 2.2 Graph of the action and orbits
The *graph of the action* is the image $`๐ช๐ต\times ๐ต`$ of the map $`(\overline{\lambda },\overline{z})(\overline{z},g(\overline{\lambda },\overline{z}))`$ that is defined on a dense open set of $`๐ข\times ๐ต`$. We have $`๐ช=\{(\overline{z},\overline{z}^{})|\overline{\lambda }๐ข\text{s.t.}\overline{z}^{}=g(\overline{\lambda },\overline{z})\}๐ต\times ๐ต`$.
We introduce a new set of variables $`Z=(Z_1,\mathrm{},Z_n)`$ and the ideal $`J=G+(Zg(\lambda ,z))h^1๐[\lambda ,z,Z]`$, where $`(Zg(\lambda ,z))`$ stands for $`(Z_1g_1(\lambda ,z),\mathrm{}`$, $`Z_ng_n(\lambda ,z))`$. The set $`๐ช`$ is dense in its closure $`\overline{๐ช}`$, and $`\overline{๐ช}`$ is the algebraic variety of the ideal:
$$O=J๐[z,Z]=\left(G+(Zg(\lambda ,z))\right)๐[z,Z].$$
Since $`G`$ is radical and unmixed dimensional so is $`J`$ because of the linearity in $`Z`$. If $`G=_{i=0}^\kappa G^{(i)}`$ is the prime decomposition of $`G`$ then we have the following prime decomposition of $`J`$:
$$\left(G+(Zg(\lambda ,z))\right)=\underset{i=0}{\overset{\kappa }{}}\left(G^{(i)}+(Zg(\lambda ,z))\right).$$
The prime ideal $`O^{(i)}=\left(G^{(i)}+(Zg(\lambda ,z))\right)๐[z,Z]`$ is therefore a component of $`O`$. The ideals $`O^{(i)}`$, however, need not be all distinct.
The set $`๐ช`$ is symmetric: if $`(\overline{z},\overline{z}^{})๐ช`$ then $`(\overline{z}^{},\overline{z})๐ช`$. By the NullStellensatz the ideal $`O`$ is also symmetric: $`p(Z,z)O`$ if $`p(z,Z)O`$. Since $`J๐[z]=(0)`$, $`O๐[z]=(0)`$ and therefore $`O๐[Z]=(0)`$ also.
A set of generators, and more precisely a Grรถbner basis , for $`O๐[z,Z]`$ can be computed.
###### Proposition 2.6
Let $`g^{}`$ be the $`n`$-tuple of numerators of $`g`$, that is $`g^{}=hg=(hg_1,\mathrm{},hg_n)\left(๐[\lambda ,z]\right)^n`$. Consider a term order s.t. $`zZ\lambda \{y\}`$ where $`y`$ is a new indeterminate. If $`Q`$ is a Grรถbner basis for $`G+(hZg^{})+(yh1)`$ according to this term order then $`Q๐[z,Z]`$ is a Grรถbner basis of $`O`$ according the induced term order on $`zZ`$.
###### Proof.
Take $`J^{}=(G+(Zg))๐[\lambda ,z,Z]`$ and note that $`J^{}=(G+(hZg^{})):h^{\mathrm{}}`$ where $`g^{}`$ is the numerator of $`g`$. Given a basis $`\mathrm{\Lambda }`$ of $`G`$ and $`g`$ explicitly, a Grรถbner basis of $`J`$ is obtained thanks to \[2, Proposition 6.37, Algorithm 6.6\]. We recognize that $`O`$ is an elimination ideal of $`J^{}`$, namely $`O=J^{}๐[z,Z]`$. A Grรถner basis for $`O`$ is thus obtained by \[2, Proposition 6.15, Algorithm 6.1\]. โ
We mainly use the extension $`O^e`$ of $`O`$ in $`๐(z)[Z]`$. If $`Q`$ is a Grรถbner basis of $`O`$ w.r.t. a term order $`zZ`$ then $`Q`$ is also a Grรถbner basis for $`O^e`$, for the term order induced on $`Z`$ \[2, Lemma 8.93\]. It is nonetheless often preferable to compute a Grรถbner basis of $`O^e`$ over $`๐(z)`$ directly (see Example 6.1).
The *orbit* of $`\overline{z}๐ต`$ is the image $`๐ช_{\overline{z}}`$ of the rational map $`g_{\overline{z}}:๐ข๐ต`$ defined by $`g_{\overline{z}}(\overline{\lambda })=g(\overline{\lambda },\overline{z})`$. We then have the following specialization property (see for instance \[6, Exercise 7\]).
###### Proposition 2.7
Let $`Q`$ be a Grรถbner basis for $`O^e`$ for a given term order on $`Z`$. There is a closed proper subset $`๐ฒ`$ of $`๐ต`$ s.t. for $`\overline{z}๐ต๐ฒ`$ the image of $`Q`$ under the specialization $`z\overline{z}`$ is a Grรถbner basis for the ideal whose variety is the closure of the orbit of $`\overline{z}`$.
Therefore, for $`\overline{z}๐ต๐ฒ`$, the dimension of the orbits of $`\overline{z}`$ is equal to the dimension of $`O^e๐(z)[Z]`$ \[6, Section 9.3, Theorem 8\]. In the rest of the paper this dimension is denoted by $`s`$.
###### Example 2.8
Scaling. Consider the group action of Example 2.3. The set of orbits consists of 1-dimensional punctured straight lines through the origin and a single zero-dimensional orbit, the origin. By elimination on the ideal $`J=(1\lambda _1\lambda _2,Z_1\lambda _1z_1,Z_2\lambda _1z_2)`$ we obtain $`O=(z_1Z_2z_2Z_1)`$. Take $`๐ฒ`$ to consist solely of the origin. For $`\overline{z}๐ต๐ฒ`$ the closure of the orbit of $`\overline{z}`$ is the algebraic variety of $`(\overline{z}_1Z_2\overline{z}_2Z_1)`$
###### Example 2.9
translation+reflection. Consider the group action of Example 2.4. By elimination on the ideal $`J=(\lambda _2^21,Z_1z_1\lambda _1,Z_2\lambda _2z_2)`$ we obtain $`O=(Z_2^2z_2^2)`$. The orbit of a point $`\overline{z}=(\overline{z}_1,\overline{z}_2)`$ with $`\overline{z}_20`$ consists of two lines parallel to the first coordinate axis, while the latter is the orbit of all points with $`\overline{z}_2=0`$
###### Example 2.10
rotation. Consider the group action of Example 2.5. The orbits consist of the origin and the circles with the origin as center. By elimination on the ideal $`J=(\lambda _1^2+\lambda _2^21,Z_1\lambda _1z_1+\lambda _2z_2,Z_2\lambda _2z_1\lambda _1z_2)`$ we obtain $`O=(Z_1^2+Z_2^2z_1^2z_2^2)`$.
### 2.3 Rational invariants
We construct a finite set of generators for the field of rational invariants. Our construction brings out a simple algorithm to rewrite any rational invariant in terms of them. The required operations are restricted to computing a Grรถbner basis and normal forms. Those are implemented in most computer algebra systems. We provide a comparison with related results in .
###### Definition 2.11
A rational function $`r๐(z)`$ is a *rational invariant* if o $`r(g(\lambda ,z))=r(z)modG.`$
The set of rational invariants forms a field<sup>3</sup><sup>3</sup>3Though we do not use this fact but rather retrieve it otherwise, it is worth noting that, as a subfield of $`๐(z)`$, the field of rational invariants is always finitely generated . $`๐(z)^G`$. We show that the coefficients of the Grรถbner basis for $`O^e`$ are invariant and generate $`๐(z)^G`$. The basis is computed using Proposition 2.6.
###### Lemma 2.12
If $`q(z,Z)`$ belongs to $`O`$ then $`q(g(\overline{\lambda },z),Z)`$ belongs to $`O^e`$ for all $`\overline{\lambda }๐ข`$.
###### Proof.
A point $`(\overline{z},\overline{z}^{})๐ช`$ if there exists $`\overline{\mu }๐ข`$ s.t. $`\overline{z}^{}=g(\overline{\mu },\overline{z})`$. Then for a generic $`\overline{\lambda }๐ข`$, $`\overline{z}^{}=g(\overline{\mu }\overline{\lambda }^1,g(\overline{\lambda },\overline{z}))`$. Therefore $`(g(\overline{\lambda },\overline{z}),\overline{z}^{})๐ช`$. Thus if $`q(z,Z)O`$ then $`q(g(\overline{\lambda },\overline{z}),\overline{z}^{})=0`$ for all $`(\overline{z},\overline{z}^{})`$ in $`๐ช`$. By the Hilbert NullStellensatz the numerator of $`q(g(\overline{\lambda },z),Z)`$ belongs to $`O`$ and therefore $`q(g(\overline{\lambda },z),Z)O^e`$. โ
Following \[2, Definition 5.29\] a set of polynomials is reduced, for a given term order, if the leading coefficients of the elements are equal to $`1`$ and each element is in normal form with respect to the others. Given a term order on $`Z`$, a polynomial ideal in $`๐(z)[Z]`$ has a unique reduced Grรถbner basis \[2, Theorem 5.3\].
###### Theorem 2.13
The reduced Grรถbner basis of $`O^e`$ with respect to any term order on $`Z`$ consists of polynomials in $`๐(z)^G[Z]`$.
###### Proof.
Let $`Q=\{q_1,\mathrm{},q_\kappa \}`$ be the reduced Grรถbner basis of $`O^e`$ for a given term order on $`Z`$. By Lemma 2.12 $`q_i(g(\overline{\lambda },z),Z)`$ belongs to $`O^e`$. It has the same support<sup>4</sup><sup>4</sup>4The support here is the set of terms in $`Z`$ with non zero coefficients. as $`q_i`$. As $`q_i(g(\overline{\lambda },z),Z)`$ and $`q_i(z,Z)`$ have the same leading monomial, $`q_i(g(\overline{\lambda },z),Z)q_i(z,Z)`$ is in normal form with respect to $`Q`$. As this difference belongs to $`O^e`$, it must be $`0`$. The coefficients of $`q_i`$ are therefore invariant. โ
Let us note the construction of a generating set of rational invariants proposed by Rosenlicht . In the paragraph before Theorem 2, Rosenlicht points out that the coefficients of the Chow form of $`O^e`$ over $`๐(z)`$ form a set of separating rational invariants. By \[29, Theorem 2\] or \[35, Lemma 2.1\] this set is generating for $`๐(z)^G`$.
Vinberg and Popov showed the existence of a subset of $`๐(z)^G[Z]`$ that generates $`O^e`$ \[35, Lemma 2.4\]. We propose the construction of such a set. They showed furthermore that the set of the coefficients of such a family of generators *separates generic orbits* \[35, Theorem 2.3\] and therefore generates $`๐(z)^G`$ \[29, Theorem 2\],\[35, Lemma 2.1\]. From those results we deduce that the set of coefficients of a reduced Grรถbner basis of $`O^e`$ generates $`๐(z)^G`$. The next theorem provides an alternative proof of this result, providing additionally a rewriting algorithm. To prove generation we indeed exhibit an algorithm that allows to rewrite any rational invariant in terms of the coefficients of a reduced Grรถbner basis.
In the case of linear actions Mรผller-Quade and Beth showed that the coefficient of the Grรถbner basis of $`O^e`$ generate the field of rational invariants. Their proof is based on more general results about the characterization of subfields of $`๐(z)`$ obtained in . Our approach is quite different and more direct. The rewriting algorithm we propose, although it was obtained independently, is nonetheless reminiscent of \[25, Algorithm 1.10\].
###### Lemma 2.14
Let $`\frac{p}{q}`$ be a rational invariant, $`p,q๐[z]`$. Then $`p(Z)q(z)q(Z)p(z)O.`$
###### Proof.
Since $`\frac{p}{q}`$ is an invariant $`\frac{p(\overline{z})}{q(\overline{z})}=\frac{p(g(\overline{\lambda },\overline{z}))}{q(g(\overline{\lambda },\overline{z}))}`$ for all $`(\overline{\lambda },\overline{z})`$ where this expression is defined. Thus $`a(\overline{z}^{},\overline{z})=p(\overline{z}^{})q(\overline{z})q(\overline{z}^{})p(\overline{z})=0`$ for all $`(\overline{z},\overline{z}^{})`$ in $`๐ช=\left\{(\overline{z},\overline{z}^{})\right|\overline{\lambda }๐ข\text{ s. t. }\overline{z}^{}=g(\overline{\lambda },\overline{z})\}๐ต\times ๐ต`$. In other words the polynomial $`a(Z,z)=p(Z)q(z)q(Z)p(z)๐[Z,z]`$ is zero at each point of $`๐ช`$. Since the algebraic variety of $`O`$ is the closure $`\overline{}๐ช`$ of $`๐ช`$ and that $`๐ช`$ is dense in $`\overline{๐ช}`$ we can conclude that $`a(Z,z)O`$ by Hilbert Nullstellensatz. โ
Assume a polynomial ring over a field is endowed with a given term order. A polynomial $`p`$ is in *normal form* w.r.t. a set $`Q`$ of polynomials if $`p`$ involves no term that is a multiple of a leading term of an element in $`Q`$. A *reduction* w.r.t. $`Q`$ is an algorithm that given $`p`$ returns a polynomial $`p^{}`$ in normal form w.r.t. $`Q`$ s.t. $`p=p^{}+_{qQ}a_qq`$ and no leading term of any $`a_qq`$ is larger than the leading term of $`p`$. Such an algorithm is detailed in \[2, Algorithm 5.1\]. It consists in rewriting the terms that are multiple of the leading terms of the elements of $`Q`$ by polynomials involving only terms that are lower. Note that if the leading coefficients of $`Q`$ are $`1`$ then no division occurs. When $`Q`$ is a Grรถbner basis w.r.t. the given term order, the reduction of a polynomial $`p`$ is unique in the sense that $`p^{}`$ is then the only polynomial in normal form w.r.t. $`Q`$ in the equivalence class $`p+(Q)`$.
###### Theorem 2.15
Consider $`\{r_1,\mathrm{},r_\kappa \}๐(z)^G`$ the coefficients of a reduced Grรถbner basis $`Q`$ of $`O^e`$. Then $`๐(z)^G=๐(r_1,\mathrm{},r_\kappa )`$ and we can rewrite any rational invariant $`\frac{p}{q}`$, with $`p,q๐[z]`$, in terms of those as follows.
Take a new set of indeterminates $`y_1,\mathrm{},y_\kappa `$ and consider the set $`Q_y๐[y,Z]`$ obtained from $`Q`$ by substituting $`r_i`$ by $`y_i`$. Let $`a(y,Z)=_{\alpha ^n}a_\alpha (y)Z^\alpha `$ and $`b(y,Z)=_{\alpha ^n}a_\alpha (y)Z^\alpha `$ in $`๐[y,Z]`$ be the reductions<sup>5</sup><sup>5</sup>5For those reductions in $`๐[y,Z]`$ the term order on $`Z`$ is extended to a block order $`yZ`$ so that the set of leading term of $`Q_y`$ is equal to the set of leading terms of $`Q`$. of $`p(Z)`$ and $`q(Z)`$ w.r.t. $`Q_y`$. There exists $`\alpha ^n`$ s.t. $`b_\alpha (r)0`$ and for any such $`\alpha `$ we have $`\frac{p(z)}{q(z)}=\frac{a_\alpha (r)}{b_\alpha (r)}`$.
###### Proof.
It is sufficient to prove the second part of the statement. The Grรถbner basis $`Q`$ is reduced and therefore monic. The sets of leading monomials of $`Q`$ and of $`Q_y`$ are equal. If $`a(y,Z)`$ is the reduction of $`p(Z)`$ w.r.t. $`Q_y`$ then $`a(r,Z)`$, obtained by substituting back $`y_i`$ by $`r_i`$, is the normal form of $`p(Z)`$ w.r.t. $`Q`$. Similarly for $`b(y,Z)`$ and $`q(Z)`$.
As $`O^e๐[Z]=(0)`$, neither $`p(Z)`$ nor $`q(Z)`$ belong to $`O^e`$ and therefore both $`a(r,Z)`$ and $`b(r,Z)`$ are different from $`0`$. By Lemma 2.14 $`q(z)p(Z)p(z)q(Z)modO^e`$ and thus the normal forms of the two polynomials modulo $`O^e`$ are equal: $`q(z)a(r,Z)=p(z)b(r,Z)`$. Thus $`a(r,Z)`$ and $`b(r,Z)`$ have the same support and this latter is non empty since $`a,b0`$. For each $`\alpha `$ in this common support, we have $`q(z)a_\alpha (r)=p(z)b_\alpha (r)`$ and therefore $`\frac{p(z)}{q(z)}=\frac{a_\alpha (r)}{b_\alpha (r)}`$. โ
###### Example 2.16
Scaling. We consider the group action given in Example 2.3. A reduced Grรถbner basis of $`O^e`$ is $`Q=\{Z_2\frac{z_2}{z_1}Z_1\}`$. By Theorem 2.13, $`๐(z_1,z_2)^G=๐(\frac{z_2}{z_1})`$.
Let $`p=z_1^2+4z_1z_2+z_2^2`$ and $`q=z_1^23z_2^2`$. We can check that $`p/q`$ is a rational invariant and we set up to write $`p/q`$ as a rational function of $`r=z_2/z_1`$. To this purpose consider $`P=Z_1^2+4Z_1Z_2+Z_2^2`$ and $`Q=Z_1^23Z_2^2`$ and compute their normal forms $`a`$ and $`b`$ w.r.t. $`\{Z_2yZ_1\}`$ according to a term order where $`Z_1<Z_2`$. We have $`a=(1+4y+y^2)Z_1^2`$ and $`b=(13y^2)Z_1^2`$. Thus
$$\frac{z_1^2+4z_1z_2+z_2^2}{z_1^23z_2^2}=\frac{1+4r+r^2}{13r^2}\text{ where }r=\frac{z_2}{z_1}$$
###### Example 2.17
translation+reflection. We consider the group action given in Example 2.4. A reduced Grรถbner basis of $`O^e`$ is $`Q=\{Z_2^2z_2^2\}`$. By Theorem 2.13, $`๐(z_1,z_2)^G=๐(z_2^2)`$.
###### Example 2.18
Rotation. We consider the group action given in Example 2.5. A reduced Grรถbner basis of $`O^e`$ is $`Q=\{Z_1^2+Z_2^2(z_1^2+z_2^2)\}`$. By Theorem 2.13, $`๐(z_1,z_2)^G=๐(z_1^2+z_2^2)`$.
## 3 Cross-section and rational invariants
Given a cross-section we construct a generating set of rational invariants endowed with a rewriting algorithm. The method is the same as the one presented in previous section but applies to only a section of the graph. In previous section we considered an ideal of the dimension of the generic orbits.Here we consider a zero dimensional ideal. This is computationally advantageous when Grรถbner bases are needed.
We use Noether normalization to prove the existence of a cross-section. The construction thus relies on selecting elements of in an open subset of a certain affine space. Therefore the construction does not entail a deterministic algorithm for the computation of rational invariants. Yet the freedom of choice is extremely fruitful in applicative examples.
Though the presentation is done with an algebraically closed field $`๐`$ that is therefore infinite, the construction is meant to be realized in characteristic zero (i.e. over $``$) or over a sufficiently large field.
### 3.1 Cross-section
Geometrically speaking a *cross-section of degree $`d`$* is a variety that intersects generic orbits in $`d`$ simple points. We give a definition in terms of ideals for it is closer to the actual computations. We give its geometric content in a proposition afterward.
###### Definition 3.1
Let $`P`$ be a prime ideal of $`๐[Z]`$ of complementary dimension to the generic orbits, i.e. if $`O^e`$ is of dimension $`s`$ then $`P`$ is of codimension $`s`$. $`P`$ defines a *cross-section* to the orbits of the rational action $`g:๐ข\times ๐ต๐ต`$ if the ideal $`I^e=O^e+P`$ of $`๐(z)[Z]`$ is radical and zero dimensional. If $`d`$ is the dimension of $`๐(z)[Z]/I^e`$ as a $`๐(z)`$-vector space, we say that $`P`$ defines a *cross-section of degree $`d`$*.
Indeed the algebra $`๐(z)[Z]/I^e`$ is a finite $`๐(z)`$-vector space since $`I^e`$ is zero dimensional \[2, Theorem 6.54\]. A basis for it is provided by the terms in $`Z`$ that are not multiple of the leading terms of a Grรถbner basis of $`I^e`$ \[2, Proposition 6.52\]. Let us note here that an ideal of $`๐(z)[Z]`$ is zero dimensional iff any Grรถbner basis of it has an element whose leading term is $`Z_i^{d_i}`$, for all $`1in`$ \[2, Theorem 6.54\].
The cross-section is thus the variety $`๐ซ`$ of $`P`$. The geometric properties of this variety are explained by the following proposition.
###### Proposition 3.2
Let $`P`$ define a cross-section $`๐ซ`$ of degree $`d`$. There is a closed set $`๐ฎ๐ต`$ s.t. the closure of the orbit of any $`\overline{z}๐ต๐ฎ`$ intersects $`๐ซ`$ in $`d`$ simple points.
###### Proof.
Let $`Q`$ be a reduced Grรถbner basis for $`I^e=O^e+P`$. Similarly to Proposition 2.7, the image $`Q_{\overline{z}}`$ of $`Q`$ under the specialization $`z\overline{z}`$ is a Grรถbner basis for $`O_{\overline{z}}+P`$ in $`๐[Z]`$ for all $`\overline{z}`$ in $`๐ต`$ outside of a closed set $`๐ฒ`$. Thus $`I_{\overline{z}}=O_{\overline{z}}+P`$ is zero dimensional and the dimension of $`๐[Z]/I_{\overline{z}}`$ as a vector space over $`๐`$ is $`d`$.
By the Jacobian criterion for regularity and the prime avoidance theorem \[10, Corollary 16.20 and Lemma3.3\] there is a $`n\times n`$ minor $`f`$ of the Jacobian matrix of $`Q`$ that is not included in any prime divisor of $`I^e`$. Therefore $`f`$ is not a zero divisor in $`๐(z)[Z]/I^e`$ which is a product of fields. There exists thus $`f^{}๐(z)[Z]`$ s.t. $`ff^{}1modI^e`$.
Provided that $`\overline{z}`$ is furthermore chosen so that the denominators of $`f`$ and $`f^{}`$ do not vanish, $`f`$ specializes into a $`n\times n`$ minor $`f_{\overline{z}}`$ of the Jacobian matrix of $`Q_{\overline{z}}`$ and we have $`f_{\overline{z}}f_{\overline{z}}^{}1modI_{\overline{z}}`$ for the specialization $`f_{\overline{z}}^{}`$ of $`f^{}`$. So $`f_{\overline{z}}`$ belongs to no prime divisors of $`I_{\overline{z}}`$ and thus $`I_{\overline{z}}`$ is radical \[10, Corollary 16.20\]. We take $`๐ฎ`$ to be the union of $`๐ฒ`$ with the algebraic set associated to the product of the denominators of $`f`$ and $`f^{}`$. That the number of points of intersection is $`d`$ is shown by \[10, Proposition 2.15\]. โ
That property shows that the cross-sections of degree $`d=1`$ and $`d>1`$ are respectively the sections and the quasi-sections defined in \[35, Paragraph 2.5\]. The existence of quasi-section is insured by \[35, Proposition 2.7\], while a criterion for the existence of a section is described in \[35, Paragraph 2.5 and 2.6\] Our terminology elaborates on the one used in and .
The discussion of \[35, Section 2.5\] shows that $`๐(๐ซ)`$ is isomorphic to $`๐(z)^G`$ when $`๐ซ`$ is a cross-section of degree 1. If $`๐ซ`$ is a cross-section of degree $`d>1`$ then $`๐(๐ซ)`$ is an algebraic extension of $`๐(z)^G`$ of degree $`d`$. In Section 4 we shall come back to those points with a constructive angle that relies on the choice of a cross-section. The viewpoint adopted here is indeed the geometric intuition of the moving frame construction in : almost any algebraic variety of complementary dimension provides a cross-section (of some degree).
The existence of a cross-section is proved by Noether normalization theorem and is linked to an alternative definition of the dimension of an ideal \[30, Section 6.2\].
###### Theorem 3.3
A linear cross-section to the orbit is associated to each point of an open set of $`๐^{s(n+1)}`$, where $`s`$ is the dimension of the generic orbits and $`n`$ the dimension of $`๐ต`$.
###### Proof.
Assume that a Grรถbner basis $`Q`$ of $`O^e`$ w.r.t. a term order $`Z_1,\mathrm{},Z_sZ_{s+1},\mathrm{},Z_n`$ is s.t. an element of $`Q`$ has leading term $`Z_i^{d_i}`$ for some $`d_i\{0\}`$ for all $`s+1in`$ and there is no element of $`Q`$ independent of $`\{Z_{s+1},\mathrm{},Z_n\}`$. Then $`Q`$ is a Grรถbner basis for the extension of $`O^e`$ to $`๐(z)(Z_1,\mathrm{},Z_s)[Z_{s+1},\mathrm{},Z_n]`$ \[2, Lemma 8.93\]. For $`(a_1,\mathrm{},a_s)`$ in an open set of $`๐^s`$ the specialization $`Q_a๐[Z_{s+1},\mathrm{},Z_n]`$ of $`Q`$ under $`Z_ia_i`$ is a Grรถbner basis \[6, Exercise 7\]. Therefore $`Q_a\{Z_1a_1,\mathrm{},Z_sa_s\}`$ is a Grรถbner basis by Buchbergerโs criteria \[2, Theorem 5.48 and 5.66\]. It is a Grรถbner basis of a zero dimensional ideal \[2, Theorem 6.54\]. We can thus take $`P`$ to be generated by $`\{Z_1a_1,\mathrm{},Z_sa_s\}`$.
We can always retrieve the situation assumed above by a change of variables thanks to Noether normalization theorem \[16, Theorem 3.4.1\]. Inspecting the proof we observe that we can choose a change of variables given by a matrix $`(m_{ij})_{1i,jn}`$ with the vector of entries $`m_{ij}`$ in $`๐^{n^2}`$ outside of some algebraically closed set. The set $`\{a_i_{1jn}m_{ij}Z_j|\mathrm{\hspace{0.33em}1}is\}`$ thus defines a cross-section. โ
The choice of a cross section introduces a non deterministic aspect to the algebraic construction proposed in next section. An analysis of the probability of success in characteristic $`0`$ would be based on the measure of a correct test sequence \[13, Theorem 3.5 and 3.7.2\], \[14, Section 3.2\], \[22, Section 4.1\].
We can computationally test if $`๐ซ`$ is a cross-section by checking the properties of $`I^e=\left(G+P+(Zg(\lambda ,z))\right)๐(z)[Z]`$, starting with the computation of its Grรถbner basis. It is nonetheless worth performing the preliminary necessary test of transversality detailed in Section 5.3. It relies on computing the rank of a matrix.
###### Proposition 3.4
Assume that $`P๐[Z]`$ defines a cross-section and that $`O=_{i=0}^\tau O^{(i)}`$ is the prime decomposition of $`O`$. Then
$$O+P=\underset{i=0}{\overset{\tau }{}}(O^{(i)}+P)\text{ and }(O^{(i)}+P)๐[Z]=P.$$
###### Proof.
We can easily check that $`_{i=0}^\tau (O^{(i)}+P)O+P`$ because $`O+P`$ is radical. The converse inclusion is trivial.
For the second equality, note first that $`P(O^{(i)}+P)๐[z,Z]`$. The projection of the variety of $`O^{(i)}๐ต\times ๐ต`$ is thus contained in $`๐ซ`$. We show that the projection is exactly $`๐ซ`$. We can assume that the numbering is such that $`O^{(i)}=((G^{(i)}+(zg(\lambda ,Z)))๐[z,Z]`$ where $`G^{(i)}`$ is a minimal prime of $`G`$ (see Section 2). By Asumption 2.2, for any $`\overline{z}`$ in $`๐ต`$ and therefore in $`๐ซ`$, there exists $`\overline{\lambda }`$ in the variety of $`G^{(i)}`$ s.t. $`g(\overline{\lambda },\overline{z})`$ is defined. Above each point of $`๐ซ`$ there is a point in the variety of $`O^{(i)}`$. โ
### 3.2 Rational invariants revisited
The following theorems provide a construction of a generating set of rational invariants together with an algorithm to rewrite any rational invariant in terms of generators. The method is the same as in Section 2.3 but applied to the ideal $`I^e`$ rather than to $`O^e`$. The computational advantage comes from the fact that $`I^e`$ is zero dimensional.
If $`G`$ is a prime ideal we can actually choose a coordinate cross-section that is $`P`$ can be taken as the ideal generated by a set of the following form: $`\{Z_{j_1}\alpha _1,\mathrm{},Z_{j_s}\alpha _s\}`$ for $`(\alpha _1,\mathrm{},\alpha _s)`$ in $`๐^s`$. In this case we can remove $`r`$ variables for the computation.
###### Theorem 3.5
The reduced Grรถbner basis of $`I^e`$ with respect to any term ordering on $`Z`$ consists of polynomials in $`๐(z)^G[Z]`$.
###### Proof.
The union of a reduced Grรถbner basis of $`O^e`$ and $`P`$ forms a generating set for $`I^e=O^e+P`$. The coefficients of a basis for $`P`$ are in $`๐`$, while the coefficients of a reduced basis for $`O^e`$ belong to $`๐(z)^G`$ due to Theorem 2.13. Since the coefficients of a generating set for $`I^e`$ belong to $`๐(z)^G`$, so do the coefficients of the reduced Grรถbner basis with respect to any term ordering. โ
###### Theorem 3.6
Consider o $`\{r_1,\mathrm{},r_\kappa \}๐(z)^G`$o the coefficients of a reduced Grรถbner basis $`Q`$ of $`I^e`$. Then $`๐(z)^G=๐(r_1,\mathrm{},r_\kappa )`$ and we can rewrite any rational invariant $`\frac{p}{q}`$, with $`p,q๐[z]`$ relatively prime, in terms of those as follows.
Take a new set of indeterminates $`y_1,\mathrm{},y_\kappa `$ and consider the set $`Q_y๐[y,Z]`$ obtained from $`Q`$ by substituting $`r_i`$ by $`y_i`$. Let $`a(y,Z)=_{\alpha ^n}a_\alpha (y)Z^\alpha `$ and $`b(y,Z)=_{\alpha ^n}a_\alpha (y)Z^\alpha `$ in $`๐[y,Z]`$ be the reductions of $`p(Z)`$ and $`q(Z)`$ w.r.t. $`Q_y`$. There exists $`\alpha ^m`$ s.t. $`b_\alpha (r)0`$ and for any such $`\alpha `$ we have $`\frac{p(z)}{q(z)}=\frac{a_\alpha (r)}{b_\alpha (r)}`$.
###### Proof.
We can proceed just as in the proof of Theorem 2.15. We only need to argue additionally that if $`r=\frac{p}{q}๐(z)^G`$, $`p`$ and $`q`$ being relatively prime, then $`p(Z),q(Z)I^e`$. We prove the result for $`p`$, the case of $`q`$ being similar.
By hypothesis $`p(z)q(g(\lambda ,z))q(z)p(g(\lambda ,z))modG`$. Since $`p`$ and $`q`$ are relatively prime, $`p(z)`$ divides $`p(g(\lambda ,z))`$ modulo $`G`$, that is there exists $`\alpha h^1๐[z,\lambda ]`$ s.t. $`p(g(\lambda ,z))\alpha (\lambda ,z)p(z)modG`$. Therefore if $`p`$ vanishes at $`\overline{z}๐ต`$, then it vanishes on $`๐ช_{\overline{z}}`$. Thus if $`pP`$, or equivalently if $`p`$ vanishes on $`๐ซ`$, it vanishes on an open subset of $`๐ต`$ (Proposition 3.2). So $`p`$ must be zero. This is not the case and thus $`pP`$. Since $`I^e๐[Z]=P`$, it is the case that $`p(Z)I^e`$
When $`P`$ defines a cross-section of degree $`1`$, the rewriting trivializes into a *replacement*. Indeed, if the dimension of $`๐(z)[Z]/I^e`$ as a $`๐(z)`$ vector space is $`1`$ then, independently of the chosen term order, the reduced Grรถbner basis $`Q`$ for $`I^e`$ is given by $`\{Z_ir_i(z)|\mathrm{\hspace{0.17em}1}in\}`$ where the $`r_i๐(z)^G`$. In view of Theorem 3.6 $`๐(z)^G=๐(r_1,\mathrm{},r_n)`$ and any rational invariant $`r(z)๐(z)^G`$ can be rewritten in terms of $`r_i`$ by replacing $`z_i`$ by $`r_i`$:
$$r(z_1,\mathrm{},z_n)=r(r_1(z),\mathrm{},r_n(z)),r๐(z)^G.$$
In the next section we generalize this replacement to cross-section of any degree by introducing some special *algebraic invariants*.
###### Example 3.7
scaling. We carry on with the action considered in Example 2.3 and 2.16.
Choose $`P=(Z_11)`$. A reduced Grรถbner basis of $`I^e`$ is given by $`\{Z_11,Z_2\frac{z_2}{z_1}\}`$. We can see that Theorem 3.5 is verified and that $`P`$ defines a cross-section of degree 1. By Theorem 3.6 we know that $`r=z_2/z_1`$ generates the field of rational invariants $`๐(z)^G`$. In this situation, the cross section is of degree $`1`$ and we see that the rewriting algorithm of Theorem 3.6 is a simple replacement. For all $`p๐(z)^G`$ we have $`p(z_1,z_2)=p(1,r)`$.
###### Example 3.8
translation+reflection. We carry on with the action considered in Example 2.4 and 2.17.
Choose $`P=(Z_1Z_2)`$ to define the cross-section. A reduced Grรถbner basis of $`I^e`$ is given by $`\{Z_1Z_2,Z_2^2z_2^2\}`$. The cross-section is thus of degree 2.
###### Example 3.9
rotation. We carry on with the action considered in Example 2.5 and 2.18.
Choose $`P=(Z_2)`$. The reduced Grรถbner basis of $`I^e`$ w.r.t. any term order is $`\{Z_2,Z_1^2(z_1^2+z_2^2)\}`$. We can see that Theorem 2.13 is verified and that $`P`$ defines a cross-section of degree 2. By Theorem 3.6 we know that $`r=z_1^2+z_2^2`$ generates the field of rational invariants $`๐(z)^G`$. In this situation, the rewriting algorithm of Theorem 3.6 consists in substituting $`z_2`$ by $`0`$ and $`z_1^2`$ by $`r`$.
## 4 Replacement invariants and invariantization
Given a cross-section $`๐ซ`$ of degree $`d`$ we introduce $`d`$ distinct $`n`$-tuples of elements that are algebraic over the field of rational invariants. Each $`n`$-tuple has an important *replacement* property: any rational invariant can be rewritten in terms of its components by a simple substitution of the variables by the corresponding elements from the tuple.
The replacement invariants are used to define a process of *invariantization*, that is a projection from the algebraic functions onto the field of algebraic invariants. This projection can be explicitly computed by algebraic elimination. It gives a constructive approach to the isomorphism $`\overline{๐(๐ซ)}\overline{๐(z)}^G`$.
### 4.1 Replacement invariants
Let $`๐ซ`$ be a cross-section of degree $`d`$ defined by a prime ideal $`P`$ of $`๐[Z]`$. The field of rational functions on $`๐ซ`$ is denoted by $`๐(๐ซ)`$. It is the fraction field of the integral domain $`๐[Z]/P=๐[๐ซ]`$. We introduce $`d`$ replacement invariants associated to $`๐ซ`$. We use them to show that $`๐(๐ซ)`$ is an algebraic extension of degree $`d`$ of the field of rational invariants $`๐(z)^G`$.
###### Definition 4.1
An *algebraic invariant* is an element of the algebraic closure $`\overline{๐(z)}^G`$ of $`๐(z)^G`$.
A reduced Grรถbner basis $`Q`$ of $`I^e=O^e+P`$ is contained in $`๐(z)^G[Z]`$ (Theorem 3.5) and therefore is a reduced Grรถbner basis of $`I^G=I^e๐(z)^G[Z]`$. The dimension of $`๐(z)^G[Z]/I^G`$ as a $`๐(z)^G`$-vector space is therefore equal to the dimension $`d`$ of $`๐(z)[Z]/I^e`$ as a $`๐(z)`$-vector space. Consequently the ideal $`I^G`$ has $`d`$ zeros $`\xi =(\xi _1,\mathrm{},\xi _n)`$ with $`\xi _i\overline{๐(z)}^G`$ \[10, Proposition 2.15\]. We call such a tuple $`(\xi _1,\mathrm{},\xi _n)`$ a $`\overline{๐(z)}^G`$-zero of $`I^G`$. A $`\overline{๐(z)}^G`$-zero of $`I^G`$ is a $`\overline{๐(z)}^G`$-zero of $`I^e`$ and conversely.
###### Definition 4.2
A replacement invariant is a $`\overline{๐(z)}^G`$-zero of $`I^G=I^e๐(z)^G[Z]`$, i.e. a $`n`$-tuple $`\xi =(\xi _1,\mathrm{},\xi _n)`$ of algebraic invariants that forms a zero of $`I^e`$.
Thus $`d`$ replacement invariants $`\xi ^{(1)},\mathrm{},\xi ^{(d)}`$ are associated to a cross-section of degree $`d`$. The name owes to next theorem which can be compared with Thomas replacement theorem discussed in \[11, page 38\].
###### Theorem 4.3
Let $`\xi =(\xi _1,\mathrm{},\xi _n)`$ be a replacement invariant. If $`r๐(z)^G`$ then $`r(z_1,\mathrm{},z_n)=r(\xi _1,\mathrm{},\xi _n)`$ in $`\overline{๐(z)}^G`$.
###### Proof.
Write $`r=\frac{p}{q}`$ with $`p,q`$ relatively prime. By Lemma 2.14, $`p(z)q(Z)q(z)p(Z)O^eI^e`$ and therefore $`p(Z)\frac{p(z)}{q(z)}q(Z)=p(Z)r(z)q(Z)I^e`$. Since $`\xi `$ is a zero of $`I^e`$, we have $`p(\xi )r(z)q(\xi )=0`$. In the proof of Theorem 3.5 we saw that $`p(Z),q(Z)`$ can not belong to $`P`$ and therefore cannot be zero divisors modulo $`I^e`$. Thus $`q(\xi )0`$ and the conclusion follows. โ
The field $`๐(\xi )`$, for any replacement invariant $`\xi `$, is an algebraic extension of $`๐(z)^G`$. Indeed $`๐(z)^G๐(\xi )`$ and $`\xi `$ is algebraic over $`๐(z)^G`$. This leads to the following results.
###### Lemma 4.4
$`I^G=I^e๐(z)^G[Z]`$ is a prime ideal of $`๐(z)^G[Z]`$.
###### Proof.
Let $`I^{(1)}`$ and $`I^{(2)}`$ be prime divisors of $`I^G`$ in $`๐(z)^G[Z]`$ and consider replacement invariants $`\xi ^{(1)}`$ and $`\xi ^{(2)}`$ that are $`\overline{๐(z)}^G`$-zeros of $`I^{(1)}`$ and $`I^{(2)}`$ respectively. Due to Theorem 4.3 $`๐(\xi ^{(i)})=๐(z)^G(\xi ^{(i)})`$. There is therefore a $`๐(z)^G`$-isomorphism $`๐(z)^G[Z]/I^{(i)}๐(\xi ^{(i)})`$ for $`i=1`$ or $`2`$. On the other hand we have $`๐(\xi ^{(i)})๐(๐ซ)`$ since $`P`$ is the ideal of all relationships on the components of $`\xi ^{(i)}`$ over $`๐`$ (Proposition 3.4). Thus
$$๐(z)^G[Z]/I^{(1)}๐(\xi ^{(1)})๐(๐ซ)๐(\xi ^{(2)})๐(z)^G[Z]/I^{(2)}.$$
We have an isomorphism between $`๐(z)^G[Z]/I^{(1)}`$ and $`๐(z)^G[Z]/I^{(2)}`$ that leaves $`๐(z)^G`$ fixed and maps the class of $`Z`$ modulo $`I^{(1)}`$ to the class of $`Z`$ modulo $`I^{(2)}`$. Therefore $`I^{(1)}=I^{(2)}`$ so that $`I^G`$ is prime. โ
###### Theorem 4.5
The field $`๐(๐ซ)`$ is an algebraic extension of $`๐(z)^G`$ of degree $`d`$, the degree of the cross-section $`๐ซ`$.
###### Proof.
For any replacement invariant $`\xi `$ we have $`๐(z)^G[Z]/I^G๐(\xi )๐(๐ซ)`$. Since the dimension of $`๐(z)^G[Z]/I^G`$ as $`๐(z)^G`$-vector space is $`d`$, the field $`๐(๐ซ)`$ is an algebraic extension of $`๐(z)^G`$ of degree $`d`$. โ
In particular if $`๐ซ`$ is a cross-section of degree one we have $`๐(๐ซ)๐(z)^G`$. In all cases we have the isomorphism $`\overline{๐(๐ซ)}\overline{๐(z)}^G`$ obtained in \[35, Section 2.5\] by different means.
###### Example 4.6
scaling. Consider the multiplicative group from Example 2.3, 2.8, 2.16. We considered the cross-section of degree 1 defined by $`P=(Z_11)`$. There is single replacement invariant $`\xi =(1,\frac{z_2}{z_1})`$ with rational components, which can be read off the reduced Grรถbner basis of $`I^e=(Z_11,Z_2\frac{z_2}{z_1})`$. One can check that $`r(z_1,z_2)=r(1,\frac{z_2}{z_1})`$ for any $`r๐(z)^G=๐\left(\frac{z_2}{z_1}\right)`$.
###### Example 4.7
translation+reflection. Consider the group action from Example 2.4, 2.9, 2.17, 3.8. We chose the cross-section defined by $`P=(Z_1Z_2)`$ and found that $`๐(z_2^2)`$ was the field of rational invariants. Generic orbits have two components and the cross-section is of degree 2. Since $`I^e=(Z_1Z_2,Z_2^2z_2^2)`$, the two replacement invariants are $`\xi ^{(1)}=(z_2,z_2)`$ and $`\xi ^{(2)}=(z_2,z_2)`$. Though rational functions, their components are not rational invariants but only algebraic invariants. Also $`I^e=(Z_1z_2,Z_2z_2)(Z_1+z_2,Z_2+z_2)`$ is a reducible ideal of $`๐(z)[Z]`$, while $`I^G`$ is an irreducible ideal of $`๐(z)^G[Z]`$.
###### Example 4.8
rotation. Consider the group action from Example 2.5, 2.10, 2.18, 3.9. We chose the cross-section defined by $`P=(Z_2)`$. Here the cross-section is again of degree 2 but the generic orbits have a single component. Since $`I^e=(Z_2,Z_1^2z_1^2z_2^2)`$ the two replacement invariants associated to $`๐ซ`$ are $`\xi ^{(\pm )}=(0,\pm \sqrt{z_1^2+z_2^2})`$.
### 4.2 Invariantization
In this section we introduce invariantization as a projection from the ring of univariate polynomials over $`๐[z]`$ to the ring of univariate polynomials over $`๐(z)^G`$. It depends on the choice of a cross-section and is computable by algebraic elimination. As this projection extends to univariate polynomials over $`๐(๐ซ)`$ it can be understood as the computable counterpart to the isomorphism $`\overline{๐(๐ซ)}\overline{๐(z)}^G`$ that follows from Proposition 4.5.
We assume throughout this section that the field $`๐`$ is of characteristic zero. The ideal of the cross-section $`๐ซ`$ is taken alternatively in $`๐[z]`$ and in $`๐[Z]`$. To avoid confusion we shall use in this section $`P_z`$ and $`P_Z`$ to distinguish the two cases. The localization of $`๐[z]`$ at $`P_z`$ is denoted by $`๐[z]_๐ซ`$. In the proof of Theorem 3.6 we have shown that $`๐(z)^G๐[z]_๐ซ`$.
The first approach for invariantization that draws directly on is to consider a replacement invariant $`\xi `$ associated to $`๐ซ`$ and the following chain of homomorphisms:
$$\begin{array}{ccccc}๐[z]_๐ซ& \stackrel{\pi }{}& ๐(๐ซ)& \stackrel{\varphi _\xi }{}& \overline{๐(z)}^G\\ r(z)& & r(z)+P_z& & r(\xi )\end{array}$$
(2)
The restriction of $`\iota _\xi =\varphi _\xi \pi :๐[z]_๐ซ\overline{๐(z)}^G`$ to $`๐(z)^G`$ is the identity map by Theorem 4.3. We call the image of a rational function $`r(z)๐[z]_๐ซ`$ under $`\iota _\xi `$ its *$`\xi `$ -invariantization*.
If $`๐ซ`$ is a cross-section of degree $`d`$ there are $`d`$ distinct associated replacement invariants $`\xi ^{(1)},\mathrm{},\xi ^{(d)}`$. The image $`\iota _\xi (r(z))=r(\xi )`$ depends on the chosen replacement invariant $`\xi `$. Such is not the case of the minimal polynomial of $`r(\xi )`$ over $`๐(z)^G`$ which depends only on $`๐ซ`$ as we shall see. We therefore define the $`๐ซ`$-invariantization as a map taking a univariate polynomial over $`๐[z]_๐ซ`$ to a univariate polynomial over $`๐(z)^G`$. The connection to the smooth invariantization of is developed in Section 5.
###### Definition 4.9
The $`๐ซ`$-invariantization $`\iota \alpha `$ of a monic univariate polynomial $`\alpha ๐[z]_๐ซ[\zeta ]`$ is the squarefree part of $`_{i=1}^d\alpha (\xi ^{(i)},\zeta )`$, where $`\xi ^{(1)},\mathrm{},\xi ^{(d)}`$ are the $`d`$ replacement invariants associated to the cross-section $`๐ซ`$.
Readers familiar with computer algebra techniques can see that $`\iota \alpha `$ belongs to $`๐(z)^G[\zeta ]`$ with the following line of arguments. The replacement invariants $`\xi ^{(1)},\mathrm{},\xi ^{(d)}`$ are the $`d`$ distinct zeros of the zero dimensional prime ideal $`I^G`$ of $`๐(z)^G[Z]`$. By a transcription of the primitive element theorem, see for instance \[16, Proposition 4.2.2-3\], they are thus the images by a polynomial map $`\psi :\theta (\psi _1(\theta ),\mathrm{},\psi _n(\theta ))`$ over $`๐(z)^G`$ of the roots $`\theta ^{(1)},\mathrm{},\theta ^{(d)}\overline{๐(z)}^G`$ of an irreducible univariate polynomial of degree $`d`$ with coefficients in $`๐(z)^G`$. The coefficients of the polynomial
$$\underset{i=1}{\overset{d}{}}\alpha (\xi ^{(i)},\zeta )=\underset{i=1}{\overset{d}{}}\alpha (\psi (\theta ^{(i)}),\zeta )$$
are elements of the field extension $`๐(z)^G(\theta ^{(1)},\mathrm{},\theta ^{(d)})`$ of $`๐(z)^G`$ that are invariant under all permutations of the $`\theta ^{(i)}`$. By \[34, Section 8.1\] or \[12, Theorem 8.15\], that polynomial belongs to $`๐(z)^G[\zeta ]`$ and thus so does its squarefree part $`\iota \alpha `$ \[34, Section 8.1\].
For a Galois theory oriented reader the details are given below. By definition $`\iota \alpha `$ belongs to the extension $`๐(\xi ^{(1)},\mathrm{},\xi ^{(d)})`$, which we denote by $`๐_\xi `$. Due to Theorem 4.3 $`๐_\xi =๐(z)^G(\xi ^{(1)},\mathrm{},\xi ^{(d)})`$. In order to prove that $`\iota \alpha ๐(z)^G[\zeta ]`$ we will show that this polynomial is preserved by the Galois group of the extension $`๐_\xi ๐(z)^G`$. We need the following proposition.
###### Proposition 4.10
Let $`\{\xi ^{(1)},\mathrm{},\xi ^{(d)}\}`$ be the set of replacement invariants corresponding to a cross-section $`๐ซ`$ of degree $`d`$. Then the field $`๐_\xi =๐(\xi ^{(1)},\mathrm{},\xi ^{(d)})`$ is a splitting field of a univariate polynomial $`\beta (z,\zeta )๐(z)^G[\zeta ]`$ of degree $`d`$. The Galois group of the extension $`๐_\xi ๐(z)^G`$ permutes the $`n`$-tuples $`\xi ^{(1)},\mathrm{},\xi ^{(d)}`$.
###### Proof.
Due to the replacement Theorem 4.3 one has the equality $`๐(\xi ^{(1)})=๐(z)^G(\xi ^{(1)})`$. From Corollary 4.5 it follows that $`๐(z)^G(\xi ^{(1)})`$ is an extension of degree $`d`$ of $`๐(z)^G`$ for $`i=1..d`$. Since $`๐`$ assumed to be of characteristic zero, the components $`\xi _1^{(1)},\mathrm{},\xi _n^{(1)}`$ of n-tuple $`\xi ^{(1)}`$ are separable over $`๐(z)^G`$. Hence there exists a primitive element $`\theta _1๐(\xi ^{(1)})`$, such that $`๐(\xi ^{(1)})=๐(z)^G(\xi ^{(1)})=๐(z)^G(\theta _1)`$, where $`\theta _1`$ is a root of an irreducible univariate polynomial $`\beta (z,\zeta )๐(z)^G[\zeta ]`$ of degree $`d`$ \[5, Theorem 5.4.1\].
Let $`\sigma _{ji}:๐(\xi ^{(i)})๐(\xi ^{(j)})`$ be the $`๐(z)^G`$-isomorphism induced by exchanging $`\xi ^{(i)}`$ and $`\xi ^{(j)}`$. Then $`\theta _j=\sigma _{j1}(\theta _1)`$ is a primitive element of the extension $`๐(\xi ^{(j)})๐(z)^G`$. Indeed, since $`\theta _1`$ is the primitive element of $`๐(z)^G(\xi ^{(1)})`$, for each $`i=1..n`$, there exists polynomial $`\psi _i`$ over $`๐(z)^G`$ such that $`\xi _i^{(1)}=\psi _i(\theta _1)`$. Since $`\sigma _{j1}`$ is a $`๐(z)^G`$-isomorphism, it follows that $`\xi _i^{(j)}=\sigma _{j1}(\xi _i^{(1)})=\sigma _{j1}(\psi _i(\theta _1))=\psi _i(\sigma _{j1}(\theta _1))=\psi _i(\theta _j)`$ for $`i=1..n`$. Thus $`\theta _j`$ is a primitive element of $`๐(\xi ^{(j)})๐(z)^G`$, and so $`๐_\xi =๐(z)^G(\theta _1,\mathrm{},\theta _d)`$
In addition, we proved that $`n`$-tuples $`\xi ^{(1)},\mathrm{},\xi ^{(d)}`$ are images of $`\theta _1,\mathrm{},\theta _d`$ under the polynomial map $`\psi =(\psi _1,\mathrm{}\psi _n):\overline{๐(z)}^G\left[\overline{๐(z)}^G\right]^n`$, where the coefficients of the univariate polynomials $`\psi _1,\mathrm{}\psi _n`$ are in $`๐(z)^G`$. Since $`\xi ^{(1)},\mathrm{},\xi ^{(d)}`$ are distinct tuples, then $`\theta _1,\mathrm{},\theta _d`$ are distinct elements of $`\overline{๐(z)}^G`$. We will now show that $`\theta _1,\mathrm{},\theta _d`$ are roots of the minimal polynomial $`\beta ๐(z)^G[\zeta ]`$ that defines $`\theta _1`$.
Indeed, since the field $`๐(z)^G`$ is fixed under $`\sigma _{j1}`$, for $`j=1..d`$, then so is the polynomial $`\beta `$. Thus $`\theta _j=\sigma _{j1}(\theta _1)`$ are roots of the polynomial $`\beta `$. It follows that $`๐_\xi =๐(z)^G(\theta _1,\mathrm{},\theta _d)`$ is the splitting field of an irreducible univariate polynomial $`\beta ๐(z)^G[\zeta ]`$ of degree $`d`$.
The elements of the $`Gal(๐_\xi /๐(z)^G)`$ permute the roots $`\theta _1,\mathrm{},\theta _d`$ of the polynomial $`\beta `$, and therefore it permutes the tuples $`\xi ^{(j)}=\psi (\theta _j)`$ for all $`j=1..d`$. โ
###### Corollary 4.11
Let $`\alpha (z,\zeta )๐[z]_๐ซ`$ be a univariate polynomial over $`๐[z]_๐ซ`$. Then its $`๐ซ`$-invariantization $`\iota \alpha `$ is a polynomial over $`๐(z)^G`$.
###### Proof.
The Galois group of the extension $`๐_\xi ๐(z)^G`$ induces permutations of the n-tuples $`\xi ^{(1)},\mathrm{},\xi ^{(d)}`$. Thus the polynomial $`p(\zeta )=_{i=1}^d\alpha (\xi ^{(i)},\zeta )๐_\xi [\zeta ]`$ is fixed under $`Gal(๐_\xi /๐(z)^G)`$. Hence its coefficients belong to $`๐(z)^G`$. By definition $`\iota \alpha `$ is the square-free part of $`p(\zeta )`$, and hence it is also fixed under the Galois group, since it has the same roots in $`๐_\xi `$ as $`p(\zeta )`$ itself \[5, Proposition 5.3.8\], and the Galois group permutes these roots. Thus its coefficients of $`\iota \alpha `$ are in $`๐(z)^G`$. โ
The following properties follow directly from the definition of the map $`\iota `$:
1. A $`\overline{๐(z)}^G`$-zero of $`\iota \beta `$ is a $`\overline{๐(z)}^G`$-zero of a $`\beta (\xi ^{(i)},\zeta )`$ and conversely.
2. If $`\beta ๐(z)^G[\zeta ]`$ then $`\iota \beta =\beta `$ since $`\beta (\xi ^{(i)},\zeta )=\beta (z,\zeta )`$ by Theorem 4.3.
3. If $`\alpha \beta modP_z`$ then $`\iota \alpha =\iota \beta `$ since the elements of $`P_z`$ vanish on all $`\xi ^{(i)}`$.
The last property shows that $`\iota `$ induces a map $`\varphi `$ from the set of monic polynomials of $`๐(๐ซ)[\zeta ]`$ to the set monic polynomials of $`๐(z)^G[\zeta ]`$ s.t. $`\iota =\varphi \pi `$.
From the first property it follows that $`\beta (\xi ^{(i)},\zeta )`$ divides $`\iota \beta (z,\zeta )`$ in $`๐(\xi ^{(i)})[\zeta ]๐(z)^G[\zeta ]`$ when $`\beta (\xi ^{(i)},\zeta )`$ is squarefree. Since $`๐(๐ซ)๐(\xi ^{(i)})`$ this amounts to the following proposition that is used in Section 5.
###### Proposition 4.12
Let $`\beta `$ be a monic polynomial of $`๐[z]_๐ซ[\zeta ]`$. If $`\beta `$ is squarefree when considered in $`๐(๐ซ)[\zeta ]`$ then it divides $`\iota \beta (z,\zeta )`$ in $`๐(๐ซ)[\zeta ]`$, that is there exists $`q(z,\zeta )๐[z]_๐ซ[\zeta ]`$ s.t. $`\iota \beta (z,\zeta )q(z,\zeta )\beta (z,\zeta )modP_z`$.
Also we recognize in the definition of the invariantization map the norm of a polynomial in a algebraic extension \[12, Section 8.8\]. We reformulate the results extending those of that text namely:
* $`\iota \beta `$ can be computed by algebraic elimination.
* if $`\beta (\xi ^{(i)},\zeta )`$ is the minimal polynomial over $`๐(\xi ^{(i)})\overline{๐(z)}^G`$ of an element in $`\overline{๐(z)}^G`$, then $`\iota \beta `$ is the minimal polynomial of this element over $`๐(z)^G`$
The algebraic elimination to compute $`\iota \beta `$ can be performed by several techniques. For a strict generalization of \[12, Section 8.8\] one could introduce a resultant formula, as developed in . We propose here a formulation in terms of elimination ideals.
###### Proposition 4.13
Let $`\beta ๐[z]_๐ซ[\zeta ]`$ be a monic polynomial. Then its $`๐ซ`$-invariantization $`\iota \beta `$ is the squarefree part of the monic generator of $`(I^G+\alpha (Z,\zeta ))๐(z)^G[\zeta ]`$ where $`\alpha (z,\zeta )๐[z][\zeta ]`$ is the numerator of $`\beta `$.
###### Proof.
The leading coefficient of $`\alpha (Z,\zeta )๐[Z][\zeta ]`$ does not belong to $`P_Z`$, and therefore it does not belong to $`I^G`$. It follows that $`(I^G+\alpha (Z,\zeta ))๐(z)^G[\zeta ](0)`$ since $`I^G`$ is zero-dimensional.
Let $`\gamma (z,\zeta )`$ be the monic generator of $`(I^G+\alpha (Z,\zeta ))๐(z)^G[\zeta ]`$. We first prove that $`\iota \beta `$ divides the squarefree part of $`\gamma (z,\zeta )`$. The fact that $`\gamma (z,\zeta )`$ belongs to $`I^G+\alpha (Z,\zeta )`$ can be written as $`\gamma (z,\zeta )q(z,Z,\zeta )\alpha (Z,\zeta )modI^G`$ where $`q(z,Z,\zeta )๐(z)^G[Z,\zeta ]`$. Substituting $`\xi ^{(i)}`$ for $`Z`$ we have $`\gamma (z,\zeta )=q^{}(z,\xi ^{(i)},\zeta )\beta (\xi ^{(i)},\zeta )`$ where $`q(z,\xi ^{(i)},\zeta )`$ and $`q^{}(z,\xi ^{(i)},\zeta )`$ differ by the factor in $`๐[\xi ^{(i)}]`$ that distinguishes $`\alpha (\xi ^{(i)},\zeta )`$ from $`\beta (\xi ^{(i)},\zeta )`$. Therefore all the factors $`\beta (\xi ^{(i)},\zeta )`$ of $`\iota \beta `$ divide $`\gamma (z,\zeta )`$. Since $`\iota \beta `$ is the squarefree product of $`\beta (\xi ^{(i)},\zeta )`$ it divides the squarefree part of $`\gamma (z,\zeta )`$.
Conversely, we prove that the squarefree part of $`\gamma (z,\zeta )`$ divides $`\iota \beta `$. The $`\overline{๐(z)}^G`$-zeros of $`\alpha (Z,\zeta )+I^G`$ are the $`(n+1)`$-tuples $`(\xi ^{(i)},f_{i,j})`$, where $`f_{i,j}`$, $`1j\mathrm{deg}\beta `$, are the roots of $`\beta (\xi ^{(i)},\zeta )`$. Since $`\gamma (z,\zeta )`$ belongs to $`\alpha (Z,\zeta )+I^G`$ its set of $`\overline{๐(z)}^G`$-roots includes all the $`f_{i,j}`$. Thus $`\gamma `$ and $`\iota \beta `$ have the same set of roots. Therefore the squarefree part of $`\gamma `$ divides $`\iota \beta `$
Note that the monic generator of $`(I^G+\alpha (Z,\zeta ))๐(z)^G[\zeta ]`$ is the monic generator of $`(I^e+\alpha (Z,\zeta ))๐(z)[\zeta ]`$. This latter is an element of the reduced Grรถbner basis of $`\left(\alpha (Z,\zeta )+I^e\right)`$ w.r.t a term order that eliminates $`Z`$. It follows from Proposition 3.5 that it belongs to $`๐(z)^G[\zeta ]`$. Therefore computations over $`๐(z)`$ lead to the correct reasult over $`๐(z)^G`$.
The last proposition provides the computable counterpart of the isomorphism $`\overline{๐(๐ซ)}\overline{๐(z)}^G`$, elements of $`\overline{๐(๐ซ)}`$ or $`\overline{๐(z)}^G`$ being represented by irreducible monic polynomials over $`๐(๐ซ)`$ or $`๐(z)^G`$ respectively.
###### Proposition 4.14
Let $`\alpha `$ be a monic polynomial of $`๐[z]_๐ซ[\zeta ]`$. The polynomial $`\iota \alpha ๐(z)^G[\zeta ]`$ is irreducible if and only if $`\alpha `$ is a power of an irreducible polynomial when considered in $`๐(๐ซ)[\zeta ]`$.
###### Proof.
Note that $`\iota (\beta \gamma )`$, for $`\beta ,\gamma ๐[z]_๐ซ[\zeta ]`$, is the squarefree part of the product $`\iota \beta \iota \gamma `$. So if $`\alpha `$ considered in $`๐(๐ซ)[\zeta ]`$ is the product of two relatively prime factors then $`\iota \alpha `$ cannot be irreducible.
We can replace $`\alpha `$ by its squarefree part when considered in $`๐(๐ซ)[\zeta ]`$ without loss of generality and thus assume for the converse implication that $`\alpha (z,\zeta )`$ is irreducible there. Let $`\overline{\alpha }๐[z][\zeta ]`$ be obtained from $`\alpha `$ by cleaning up the denominators. Then $`\overline{\alpha }(Z,\zeta )`$ is irreducible modulo $`I^G`$ so that $`\left(\overline{\alpha }(Z,\zeta )+I^G\right)`$ is prime. The monic generator $`\iota \alpha `$ of $`\left(\alpha (Z,\zeta )+I^G\right)๐(z)[\zeta ]`$ is thus irreducible. โ
The following example illustrates various properties of the $`๐ซ`$-invariantization map $`\iota `$.
###### Example 4.15
scaling. We consider the scaling action defined in Example 2.3 and the cross-section defined by the ideal $`P_Z=(Z_1^2+Z_2^21)`$. It is a cross-section of degree $`2`$. We have $`I^e=(Z_1^2\frac{z_1^2}{z_1^2+z_2^2},Z_2\frac{z_2}{z_1}Z_1)`$ and therefore the two replacement invariants are
$$\xi ^{(\pm )}=(\frac{\pm z_1}{\sqrt{z_1^2+z_2^2}},\frac{\pm z_2}{\sqrt{z_1^2+z_2^2}}).$$
The invariantization of $`\alpha =\zeta z_1`$ is $`\iota \alpha =\zeta ^2\frac{z_1^2}{z_1^2+z_2^2}`$. We have $`\iota \alpha =(\zeta +z_1)\alpha +\frac{z_1^2}{z_1^2+z_2^2}(z_1^2+z_2^21)(\zeta +z_1)\alpha modP_z`$. We obtained $`\iota \alpha `$ by computing the reduced Grรถbner basis of the ideal $`(\zeta Z_1,Z_1^2\frac{z_1^2}{z_1^2+z_2^2},Z_2\frac{z_2}{z_1}Z_1)`$ with a term order that eliminates $`Z_1`$ and $`Z_2`$. Note that, although $`\alpha `$ defines a polynomial function, its invariantization defines two algebraic invariants $`\pm \frac{z_1}{\sqrt{z_1^2+z_2^2}}`$.
The invariantization of $`\beta =\zeta ^3+\zeta ^2+z_2\zeta +1`$ is $`\iota \beta =\zeta ^6+2\zeta ^5+\zeta ^4+2\zeta ^3+\frac{z_2^2+2z_1^2}{z_1^2+z_2^2}\zeta ^2+1`$. We have $`\iota \beta (\zeta ^3+\zeta ^2z_2\zeta +1)\beta modP_z`$.
In the next two instances the monic polynomial is equal modulo $`P_z`$ to a polynomial in $`\overline{๐(z)}^G[\zeta ]`$. As a consequence, the invariantization equals to the original polynomial modulo $`P_z`$
The polynomial $`\gamma =\zeta z_1^2`$ is equal to its $`๐ซ`$-invariantization $`\iota \gamma =\zeta \frac{z_1^2}{z_1^2+z_2^2}\gamma modP_z`$.
The irreducible polynomial $`\delta =\zeta ^2\frac{z_1^2+z_2^21}{z_2^2}\zeta \frac{z_1^2}{z_2^2}`$ becomes a reducible modulo $`P_z`$: $`\delta \zeta ^2\frac{z_1^2}{z_2^2}modP_z`$. Its invariantization is thus reducible: $`\iota \delta =(\zeta \frac{z_1}{z_2})(\zeta +\frac{z_1}{z_2})\delta modP_z`$.
## 5 Local invariants and the moving frame construction
In this section we connect the algebraic algorithms presented in this paper with their original source of inspiration, the Fels-Olver moving frame construction . It is shown in that in the case of a *locally free smooth action* of a Lie group $`๐ข`$ on a manifold $`๐ต`$, a choice of local cross-section corresponds to a local $`๐ข`$-equivariant map $`\rho :๐ต๐ข`$. This map provides a generalization of the classical geometrical moving frames<sup>6</sup><sup>6</sup>6For this reason the map $`\rho `$ is called *moving frame* in . We adopt the term *a moving frame map.* . A moving frame map gives rise to an *invariantization* process, a projection from the set of smooth functions to the set of local invariants.
We introduce an alternative definition of the smooth invariantization process which, on one hand, generalizes the definition given in to non-free, semi-regular actions and, on the other hand, can be effectively reformulated in the algebraic context. We make explicit comparisons with both the moving frame and the algebraic constructions in Section 5.6 and Section 5.5 respectively.
In this section we consider real smooth manifolds. All statements and constructions from this section are applicable to complex manifolds. In the latter case all maps and functions are assumed to be meromorphic.
### 5.1 Local action of a Lie group on a smooth manifold
We consider a Lie group $`๐ข`$, with identity denoted $`e`$, and a smooth manifold $`๐ต`$ of dimension $`n`$. We first review the necessary facts and terminology from the theory of Lie group actions on smooth manifolds. Our presentations is based on .
###### Definition 5.1
A local action of a Lie group $`๐ข`$ on a smooth manifold $`๐ต`$ is a smooth map $`g:\mathrm{\Omega }๐ต`$, where $`\mathrm{\Omega }\{e\}\times ๐ต`$ is an open subset of $`๐ข\times ๐ต`$, and the map $`g`$ satisfies the following two properties:
1. $`g(e,\overline{z})=\overline{z}`$, $`\overline{z}๐ต`$.
2. $`g(\overline{\mu },g(\overline{\lambda },z))=g(\overline{\mu }\overline{\lambda },z)`$, for all $`\overline{z}๐ต`$ and $`\overline{\lambda },\overline{\mu }๐ข`$ s. t. $`(\overline{\lambda },\overline{z})`$ and $`(\overline{\mu }\overline{\lambda },\overline{z})`$ are in $`\mathrm{\Omega }`$.
The *orbit* of $`\overline{z}๐ต`$ is the image $`๐ช_{\overline{z}}`$ of the smooth map $`g_{\overline{z}}:๐ข๐ต`$ defined by $`g_{\overline{z}}(\overline{\lambda })=g(\overline{\lambda },\overline{z})`$. The domain of $`g_{\overline{z}}`$ is an open subset of $`๐ข`$ containing $`e`$.
For every point $`\overline{z}๐ต`$ the differential $`dg_{\overline{z}}:T๐ข|_eT๐ต|_{\overline{z}}`$ maps the tangent space of $`๐ข`$ at $`e`$ to the tangent space of $`๐ต`$ at the point $`\overline{z}`$. The tangent space $`T๐ข|_e`$ can be identified with the Lie algebra $`๐ค`$ of $`๐ข`$. Let $`\widehat{v}๐ค`$ then $`v(\overline{z})=dg_{\overline{z}}(\widehat{v})`$ is a smooth vector field on $`๐ต`$, called the infinitesimal generator of the $`๐ข`$-action corresponding to $`\widehat{v}`$. The set of all infinitesimal generators for a $`๐ข`$-action form a Lie algebra, such that the map $`\widehat{v}v`$ is a Lie algebra homomorphism.
By $`\mathrm{exp}(ฯตv,\overline{z}):\times ๐ต๐ต`$ we denote the flow of $`v`$. The flow is defined as an integral curve of the vector field $`v`$ with the initial condition $`\overline{z}`$. One can prove that every point of the connected component of the orbit $`๐ช_{\overline{z}}^0\overline{z}`$ can be reached from $`\overline{z}`$ by a composition of flows of a finite number of infinitesimal generators.
Let $`\widehat{v}_1,\mathrm{},\widehat{v}_\kappa `$, where $`\kappa s`$ is the dimension of the group, be a basis of the Lie algebra of $`๐ข`$. Then the infinitesimal generators $`v_1,\mathrm{},v_\kappa `$ span the tangent space to the orbits at each point of $`๐ต`$.
###### Definition 5.2
An action of a Lie group $`๐ข`$ on a smooth manifold $`๐ต`$ is semi-regular if all orbits have the same dimension.
Throughout this section the action is assumed to be semi-regular. The dimension of the orbits is denoted by $`s`$.
### 5.2 Local invariants
We give definitions of local invariants and fundamental sets of those. We prove that the existence of a fundamental set of local invariants follows from the existence of a flat coordinate system. The proof is based on standard arguments from differential geometry.
###### Definition 5.3
A smooth function $`f`$, defined on an open subset $`๐ฐ๐ต`$, is a local invariant if $`v(f)=0`$ for any infinitesimal generator $`v`$ of the $`๐ข`$-action on $`๐ฐ`$.
Equivalently $`f(\mathrm{exp}(\epsilon v,\overline{z}))=f(\overline{z})`$ for all $`\overline{z}๐ฐ`$, all infinitesimal generator $`v`$, and all real $`\epsilon `$ sufficiently close to zero. If the group $`๐ข`$ is connected, the function $`f`$ is continuous on $`๐ต`$, and the condition of Definition 5.3 is satisfied at every point of $`๐ต`$ then $`f`$ is a global invariant on $`๐ต`$ due to \[26, Proposition 2.6\]. In what follows we neither assume $`f`$ to be continuous outside of $`๐ฐ`$, nor $`๐ข`$ to be connected.
A collection of smooth functions $`f_1,\mathrm{},f_l`$ are functionally *dependent* on a manifold $`๐ต`$ if for each point $`\overline{z}๐ฐ`$ there exists on open neighborhood an $`๐ฐ`$ and a non-zero differentiable function $`F`$ in $`l`$ variables such that $`F(f_1,\mathrm{},f_l)=0`$ on $`๐ฐ`$. From the implicit function theorem it follows that $`f_1,\mathrm{},f_l`$ are functionally dependent on $`๐ฐ`$ if and only if the rank of the corresponding Jacobian matrix is less than $`l`$ at each point of $`๐ต`$. We say that functions $`f_1,\mathrm{},f_l`$ are *independent* on $`๐ต`$ if they are not dependent when restricted to any open subset of $`๐ต`$. As it is commented in \[26, p85\] functional dependence and functional independence on $`๐ต`$ do not exhaust the range of possibilities, except for analytic functions. Throughout the section the term independent functions means functionally independent functions. Finally we say that $`f_1,\mathrm{},f_l`$ are independent at a point $`\overline{z}๐ต`$ if the rank of the corresponding Jacobian matrix is maximal at $`\overline{z}`$. Independence at $`\overline{z}`$ implies independence on some open neighborhood of this point. If $`๐ฐ`$ is an open subset of $`๐ต`$ and $`f_1,\mathrm{},f_n`$ are independent at each point of $`๐ต`$, then these functions provide a coordinate system on $`๐ฐ`$.
###### Definition 5.4
A collection of local invariants on $`๐ฐ`$ forms a *fundamental set* if they are functionally independent, and any local invariant on $`๐ฐ`$ can be expressed as a smooth function of the invariants from this set.
The Lie algebra of infinitesimal generators provides an integrable distribution<sup>7</sup><sup>7</sup>7An integrable distribution is a collection of smooth vector fields, whose span over the ring of smooth functions is closed with respect to Lie bracket. of smooth vector-fields on $`๐ต`$, whose integral manifolds are orbits. For a semi-regular action this distribution is of constant rank $`s`$, the dimension of the orbits. It follows from Frobenius theorem that in an open neighborhood $`๐ฐ`$ of each point there exists a coordinate system $`x_1,\mathrm{},x_s,y_1,\mathrm{},y_{ns}`$ such that the connected components of the orbits on $`๐ฐ`$ are level sets of the last $`ns`$ coordinates \[31, p. 262\] and \[26, Theorem 1.43\]. Such coordinate system is called flat, or straightening. The proof of the following theorem establishes that $`y_1,\mathrm{},y_{ns}`$ form a fundamental set of local invariants.
###### Theorem 5.5
Let $`๐ข`$ be a Lie group acting semi-regularly on an $`n`$-dimensional manifold $`๐ต`$. Let $`s`$ be the dimension of the orbits. In the neighborhood of each point $`\overline{z}๐ต`$ there exists a fundamental set of $`ns`$ local invariants.
###### Proof.
By Frobenius theorem there exists a flat coordinate system $`x_1,\mathrm{},x_s`$, $`y_1,\mathrm{},y_{ns}`$ in a neighborhood $`๐ฐ\overline{z}`$. The connected components of the orbits on $`๐ฐ`$ coincide with the level sets of the last $`ns`$ coordinate functions. Thus $`y_1,\mathrm{},y_{ns}`$ are constant on the connected components of the orbits, and therefore they are local invariants, being smooth and functionally independent by definition of a coordinate system. It remains to show that any other invariant is locally expressible in terms of them. Let $`v`$ be an infinitesimal generator of the group action. Since $`v(y_i)=0`$ for $`i=1..(ns)`$ then $`v=_{i=1}^sv(x_i)\frac{}{x_i}`$ is a linear combination of the first $`s`$ basis vector fields. Let $`v_1=_{i=1}^sa_{1i}\frac{}{x_i},\mathrm{},v_\kappa =_{i=1}^sa_{\kappa i}\frac{}{x_i}`$ be a basis of infinitesimal generators of the group action. Without loss of generality we may assume that the first $`s`$ generators $`v_1,\mathrm{},v_s`$ are linearly independent at each point of $`๐ฐ`$. Let $`f(x_1\mathrm{},x_s,y_1,\mathrm{}y_{ns})`$ be a local invariant, then $`v_j(f)=_{i=1}^ra_{ji}\frac{f}{x_i}=0`$ for $`j=1..s`$. This is a homogeneous system of $`s`$ linear equation with $`s`$ unknowns $`\frac{f}{x_1},\mathrm{},\frac{f}{x_s}`$. Since $`v_1,\mathrm{},v_s`$ are linearly independent at each point, the rank of the system is maximal. Thus $`\left(\frac{f}{x_1}=0,\mathrm{},\frac{f}{x_n}=0\right)`$ is the only solution. Hence $`f`$ is a function of invariants $`y_1,\mathrm{},y_{ns}`$. โ
The existence of a fundamental set of local invariants, therefore, follows from the existence of a flat coordinate system. The proof is not constructive however. The invariantization process, introduced in next section, leads to a different characterization of a fundamental set of invariants. Invariantization, and therefore fundamental invariants, can be effectively computed either by the algorithms of Section 4, in the case of a rational action of an algebraic group (see Section 5.5), or by the moving frame method of , in the case of a locally free action of a Lie group (see Section 5.6).
### 5.3 Local cross-section and smooth invariantization
We define local cross-sections to the orbits and show that a local cross-section passing through any given point can easily be constructed. A local cross-section gives rise to an equivalence relationship on the ring of smooth functions such that any class has a single representative that is a local invariant. This leads to an *invariantization map*, a projection from the ring of smooth functions to the ring of local invariants. It generalizes the invariantization process defined in to semi-regular actions. Although a possibility of such generalization is indicated in the remarks of \[11, Section 4\], the precise definitions and theorems, appearing in this section, are new.
###### Definition 5.6
An embedded submanifold $`๐ซ`$ of $`๐ต`$ is a *local cross-section* to the orbits if there is an open set $`๐ฐ`$ of $`๐ต`$ such that
* $`๐ซ`$ intersects $`๐ช_{\overline{z}}^0๐ฐ`$ at a unique point $`\overline{z}๐ฐ`$, where $`๐ช_{\overline{z}}^0`$ is the connected component of $`๐ช_{\overline{z}}๐ฐ`$, containing $`\overline{z}`$.
* for all $`\overline{z}๐ซ๐ฐ`$, $`๐ช_{\overline{z}}^0`$ and $`๐ซ`$ are transversal and of complementary dimensions.
The second condition in the above definition is equivalent to the following condition on tangent spaces: $`T_{\overline{z}}๐ต=T_{\overline{z}}๐ซT_{\overline{z}}๐ช_{\overline{z}}`$, $`\overline{z}๐ซ๐ฐ`$.
An embedded submanifold of codimension $`s`$ is locally given as the zero set of $`s`$ independent functions. Assume that $`h_1(z),\mathrm{},h_s(z)`$ define $`๐ซ`$ on $`๐ฐ`$. The tangent space at a point of $`๐ซ`$ is the kernel of the Jacobian matrix $`J_h`$ at this point. A basis of infinitesimal generators $`v_1,\mathrm{},v_\kappa `$, where $`\kappa s`$ is the dimension of the group, span the tangent space to the orbits at each point of $`๐ซ`$. Therefore the submanifold $`๐ซ`$ is a local cross-section if and only if the span of the infinitesimal generators $`v_1,\mathrm{},v_\kappa `$ has a trivial intersection with the kernel of $`J_h`$ on $`๐ซ`$. Equivalently:
$$\text{ the rank of the }s\times \kappa \text{ matrix }\left(v_j(h_i)\right)_{i=1..s}^{j=1..\kappa }=J_hV\text{ equals to }s\text{ on }๐ซ,$$
(3)
where $`V`$ is the $`n\times \kappa `$ matrix, whose $`i`$-th column consists of the coefficients of the infinitesimal generator $`v_i`$ in a local coordinate system. In the next theorem we prove the existence of a local cross-section through every point. The first paragraph of the proof provides a simple practical algorithm to construct a coordinate local cross-section through a point. An algebraic counterpart of this statements is given by Theorem 3.3.
###### Theorem 5.7
Let $`๐ข`$ act semi-regularly on $`๐ต`$. Through every point $`\overline{z}๐ต`$ there is a local cross-section that is defined as the level set of $`s`$ coordinate functions.
###### Proof.
Let $`V`$ be the $`n\times \kappa `$ matrix of the coefficients of the infinitesimal generators $`v_1,\mathrm{},v_\kappa `$ relative to a coordinate system $`z_1,\mathrm{},z_k`$. The rank of $`V`$ equals to the dimension of the orbits $`s`$. Thus there exist $`s`$ rows of $`V`$ that form an $`s\times \kappa `$ submatrix $`\widehat{V}`$ of rank $`s`$ at the point $`\overline{z}`$, and therefore it has rank $`s`$ on an open neighborhood $`๐ฐ_1\overline{z}`$. Assume that these rows correspond to coordinate $`z_{i_1},\mathrm{},z_{i_s}`$. Let $`(c_1,\mathrm{},c_n)`$ be coordinates of the point $`\overline{z}`$, then functions $`h_1=z_{i_1}c_{i_1},\mathrm{},h_s=z_{i_s}c_{i_s}`$ satisfy condition (3). The common zero set $`๐ซ`$ of these functions contains $`\overline{z}`$.
It remains to prove that there exists a neighborhood $`๐ฐ\overline{z}`$ such that $`๐ซ`$ intersects each connected component of the orbits on $`๐ฐ`$ at a unique point. Let $`x_1,\mathrm{},x_s,y_1,\mathrm{},y_{ns}`$ be a flat coordinate system in an open neighborhood $`๐ฐ_2\overline{z}`$. Due to Theorem 5.5 $`y_1,\mathrm{},y_{ns}`$ are independent local invariants. We will show that functions $`z_{i_1},\mathrm{},z_{i_s},y_1,\mathrm{},y_{ns}`$ provide a coordinate system an open set $`๐ฐ=๐ฐ_1๐ฐ_2`$ containing $`\overline{z}`$. Without loss of generality we may assume that $`\{z_{i_1},\mathrm{},z_{i_s}\}=\{z_1,\mathrm{},z_s\}`$ are the first $`s`$ coordinates. In terms of flat coordinates $`z_i=F_i(x,y),i=1..s`$, where $`F_i`$ are smooth functions on $`๐ฐ_2`$. Since $`v_i(y_j)=0`$ for $`i=1..\kappa ,j=1..ns`$, then
$$\left(v_j(z_i)\right)_{i=1..s}^{j=1..\kappa }=\left(\frac{F_i}{x_r}\right)_{i=1..s}^{r=1..s}\left(v_j(x_r)\right)_{r=1..s}^{j=1..\kappa }.$$
(4)
We note that $`\left(v_i(z_j)\right)_{j=1..s}^{i=1..\kappa }=\widehat{V}`$ is $`s\times \kappa `$ matrix of rank $`s`$ at each point of $`๐ฐ`$. Matrix $`\left(v_j(x_r)\right)_{r=1..s}^{j=1..\kappa }`$ also has maximal rank $`s`$ on $`๐ฐ`$. Therefore the matrix $`\left(\frac{F_i}{x_r}\right)_{i=1..s}^{r=1..s}`$ is invertible on $`๐ฐ`$. By looking at the rank of the corresponding Jacobian matrix in flat coordinates, we conclude that functions $`z_1,\mathrm{},z_s,y_1,\mathrm{},y_{ns}`$ are independent at each point of $`๐ฐ`$, and therefore define a coordinate system on $`๐ฐ`$.
By construction all points on $`๐ซ`$ have the same $`z`$-coordinates. Thus two distinct points of $`๐ซ`$ must differ by at least one of the $`y`$-coordinates. Since $`y`$ coordinates are constant on the connected components of the orbits on $`๐ฐ`$, distinct points of $`๐ซ`$ belong to distinct connected components of the orbits. โ
Given a cross-section on $`๐ฐ`$ one can define a projection from the set of smooth functions on $`๐ฐ`$ to the set of local invariants.
###### Definition 5.8
Let $`๐ซ`$ be a local cross-section to the orbits on an open set $`๐ฐ`$. Let $`f`$ be a smooth function on $`๐ฐ`$. The *invariantization* $`\overline{\iota }f`$ of $`f`$ is the function on $`๐ฐ`$ that is defined, for $`\overline{z}๐ฐ`$, by $`\overline{\iota }f(\overline{z})=f(\overline{z}_0),`$ where $`\overline{z}_0=๐ช_{\overline{z}}^0๐ซ`$.
In other words, the invariantization of a function $`f`$ is obtained by spreading the values of $`f`$ on $`๐ซ`$ along the orbits. The next theorem shows that $`\overline{\iota }f`$ is the unique local invariant with the same values on $`๐ซ`$ as $`f`$.
###### Theorem 5.9
Let a Lie group $`๐ข`$ act semi-regularly on a manifold $`๐ต`$, and let $`๐ซ`$ be a local cross-section. Then $`\overline{\iota }f`$ is the unique local invariant defined on $`๐ฐ`$ whose restriction to $`๐ซ`$ is equal to the restriction of $`f`$ to $`๐ซ`$. In other words $`\overline{\iota }f|_๐ซ=f|_๐ซ`$.
###### Proof.
For any $`\overline{z}๐ฐ`$ and small enough $`\epsilon `$ the point $`\mathrm{exp}(\epsilon v,\overline{z})`$ belongs to the same connected component $`๐ช_{\overline{z}}^0`$. Let $`\overline{z}_0=๐ช_{\overline{z}}^0๐ซ`$. Then $`\overline{\iota }f\left(\mathrm{exp}(\epsilon v,\overline{z})\right)=f(\overline{z}_0)=\overline{\iota }f(\overline{z})`$, and thus $`\overline{\iota }f`$ is a local invariant. By definition $`\overline{\iota }f(\overline{z}_0)=f(\overline{z}_0)`$ for all $`\overline{z}_0๐ซ`$.
In order to show its smoothness we write $`\overline{\iota }f`$ in terms of flat coordinates $`x_1,\mathrm{},x_s`$, $`y_1,\mathrm{},y_{ns}`$. By probably shrinking $`๐ฐ`$, we may assume that $`๐ซ`$ is given by the zero-set of smooth independent functions $`h_1(x_1,\mathrm{},x_s,y_1,\mathrm{},y_{ns}),\mathrm{},`$ $`h_s(x_1,\mathrm{},x_s,y_1,\mathrm{},y_{ns})`$. From the transversality condition (3) and local invariance of $`y`$โs, it follows that the first $`s`$ columns of the Jacobian matrix $`J_h`$ form a submatrix of rank $`s`$. Thus the cross-section $`๐ซ`$ can be described as a graph
$`x_1=p_1(y_1,\mathrm{},y_{ns}),\mathrm{},x_s=p_s(y_1,\mathrm{},y_{ns})`$, where $`p_1,\mathrm{},p_s`$ are smooth functions. Then the function
$$\overline{\iota }f(x_1,\mathrm{},x_s,y_1,\mathrm{},y_{ns})=f(p_1(y_1,\mathrm{},y_{ns}),\mathrm{},p_s(y_1,\mathrm{},y_{ns}),y_1,\mathrm{},y_{ns})$$
is smooth, as a composition of smooth functions.
To prove the uniqueness, assume that an invariant function $`q`$ has the same values on $`๐ซ`$ as $`f`$, then the invariant function $`h=\overline{\iota }fq`$ has zero value on $`๐ซ`$. A point $`\overline{z}๐ฐ`$ can be reached from $`\overline{z}_0=๐ซ๐ช_{\overline{z}}^0`$ by a composition of flows defined by infinitesimal generators. Without loss of generality, we may assume that it can be reached by a single flow $`\overline{z}=\mathrm{exp}(ฯตv,\overline{z}_0)`$, where $`\mathrm{exp}(\epsilon v,\overline{z}_0)๐ช_{\overline{z}}^0`$ for all $`0\epsilon ฯต`$. From the invariance of $`h`$ it follows that $`h\left(\mathrm{exp}(ฯตv,\overline{z}_0)\right)=h(\overline{z}_0)=0`$. Thus $`q(z)=\overline{\iota }f(z)`$ on $`๐ฐ`$. โ
Theorem 5.9 allows us to view the invariantization process as a projection from the set of smooth functions on $`๐ฐ`$ to the equivalence classes of functions with the same value on $`๐ซ`$. Each equivalence class contains a unique local invariant. The algebraic counterpart of this point of view is described in Section 4.2.
The invariantization of differential forms can be defined in a similar implicit manner. It has been shown in that the essential information about the differential ring of invariants and the structure of differential forms can be computed from the infinitesimal generators of the action and the equations that define the cross-section, without explicit formulas for invariants.
### 5.4 Normalized and fundamental invariants
The *normalized invariants* introduced in are the invariantizations of the coordinate functions. They have the replacement property. In the algebraic context they correspond to *replacement* invariants defined in Section 4. This correspondence is made precise by Proposition 5.15. We show that a set of normalized invariants contains a fundamental set of local invariants.
All results of this subsection are stated under the following assumptions. A manifold $`๐ซ`$ is a local cross-section to the $`s`$-dimensional orbits of a semi-regular $`๐ข`$-action on an open $`๐ฐ๐ต`$, and $`\overline{\iota }`$ is the corresponding invariantization map. The set $`๐ฐ`$ is a single coordinate chart on $`๐ต`$ with coordinate functions $`z_1,\mathrm{},z_n`$. By possibly shrinking $`๐ฐ`$ we may assume that $`๐ซ`$ is the zero set of $`s`$ independent smooth functions.
Since our definition of invariantization differs from we restate and prove the replacement theorem.
###### Theorem 5.10
If $`f(z_1,\mathrm{},z_n)`$ is a local invariant on $`๐ฐ`$ then $`f(\overline{\iota }z_1,\mathrm{},\overline{\iota }z_n)=f(z_1,\mathrm{},z_n)`$.
###### Proof.
Since o $`\overline{\iota }z_1|_๐ซ=z_1|_๐ซ,\mathrm{},\overline{\iota }z_n|_๐ซ=z_n|_๐ซ`$, then $`f(\overline{\iota }z_1,\mathrm{},\overline{\iota }z_n)|_๐ซ=f(z_1,\mathrm{},z_n)|_๐ซ`$. Thus functions $`f(\overline{\iota }z_1,\mathrm{},\overline{\iota }z_n)`$ and $`f(z_1,\mathrm{},z_n)`$ are both local invariants and have the same value on $`๐ซ`$. By Theorem 5.9 they coincide. โ
###### Lemma 5.11
Let $`๐ซ`$ be a local cross-section on $`๐ฐ`$, given as the zero set of $`s`$ independent functions $`h_1,\mathrm{},h_s`$. Then $`h_1(\overline{\iota }z_1,\mathrm{},\overline{\iota }z_n)=0,\mathrm{},h_s(\overline{\iota }z_1,\mathrm{},\overline{\iota }z_n)=0`$ on $`๐ฐ`$. If for a differentiable $`n`$-variable function $`f`$ we have $`f(\overline{\iota }z_1,\mathrm{},\overline{\iota }z_n)0`$ on an open subset of $`๐ฐ`$, then there exits open $`๐ฒ๐ฐ`$ such that $`๐ฒ๐ซ\mathrm{}`$ and at each point of $`๐ฒ๐ซ`$ functions $`f`$, $`h_1,\mathrm{},h_s`$ are not independent.
###### Proof.
Since $`h(\overline{\iota }z)|_๐ซ=\overline{\iota }h(z)|_๐ซ`$ and both functions are invariants, one has $`h(\overline{\iota }z)=\overline{\iota }h(z)`$ by Theorem 5.9. The latter is zero since $`h|_๐ซ=0`$. Assume now that there exits a differentiable function $`f`$ and an open subset of $`๐ฑ๐ฐ`$ such that $`f(\overline{\iota }z_1,\mathrm{},\overline{\iota }z_n)0`$ on $`๐ฑ`$. Since $`f(\overline{\iota }z)=\overline{\iota }f(z)`$ is invariant, there exists an open $`๐ฒ๐ฑ`$ such that $`f(\overline{\iota }z_1,\mathrm{},\overline{\iota }z_n)0`$ on $`๐ฒ`$ and $`๐ฒ๐ซ\mathrm{}`$. We conclude that $`f(z_1,\mathrm{},z_n)0`$ on $`๐ซ๐ฒ`$. In this case $`f`$ cannot be independent of $`h_1,\mathrm{},h_s`$ at any point of $`๐ซW`$ since otherwise this would imply that $`๐ซ`$ is of dimension less then $`ns`$. โ
###### Theorem 5.12
Let $`๐ซ`$ be a local cross-section on $`๐ฐ`$, given as the zero set of $`s`$ independent functions. The set $`\{\overline{\iota }z_1,\mathrm{},\overline{\iota }z_n\}`$ of the invariantizations of the coordinate functions $`z_1,\mathrm{},z_n`$ contains a fundamental set of $`ns`$ local invariants on $`๐ฐ`$.
###### Proof.
Due to the implicit function theorem, after a possible shrinking $`๐ฐ`$ and renumbering of the coordinate functions, we may assume that $`๐ซ`$ is the zero set of the functions $`h_1(z)=z_1p_1(z_{s+1},\mathrm{},z_n),\mathrm{},h_s(z)=z_sp_s(z_{s+1},\mathrm{},z_n)`$. Therefore $`\overline{\iota }z_1=p_1(\overline{\iota }z_{s+1},\mathrm{},\overline{\iota }z_n),\mathrm{},\overline{\iota }z_s=p_k(\overline{\iota }z_{s+1},\mathrm{},\overline{\iota }z_n)`$ by Theorem 5.9. From Theorem 5.10 we can conclude that any local invariant can be written in terms of $`\overline{\iota }z_{s+1},\mathrm{},\overline{\iota }z_n`$. Since for every differentiable non-zero $`ns`$-variable function $`f`$, functions $`f(z_{s+1},\mathrm{},z_n),h_1(z),\mathrm{},h_s(z)`$ are independent at every point of $`๐ฐ`$, then by Lemma 5.11, $`\overline{\iota }z_{s+1},\mathrm{},\overline{\iota }z_n`$ are functionally independent on $`๐ฐ`$. โ
### 5.5 Relation between the algebraic and the smooth constructions
We establish a connection between the smooth and the algebraic constructions. We show that the normalized invariants (Section 5.4) can be viewed as smooth representatives of the replacement invariants (Section 4.1), and that algebraic invariantization (Section 4.2) provides a constructive approach to smooth invariantization (Section 5.3).
To be at the intersection of the hypotheses of the smooth and the algebraic settings we consider a real algebraic group, that is the set of real points of an algebraic group defined<sup>8</sup><sup>8</sup>8This implicitly means that we know the ideal $`G`$ (Section 2.1) from a set of generators with coefficients in $``$. over $``$. It is a real Lie group \[32, the Proposition in Chapter 3, Section 2.1.2\]. Lie groups appearing in applications often satisfy this property. We also assume that the local action is given by a rational map (1), in Section 2.1, that satisfies Asumption 2.2. This guarantees semi-regularity of the action on an open set $`๐ต`$ of $`^n`$ as the orbits of non-maximal dimension are contained in an algebraic set defined by minors of the matrix $`V`$ of (3), in Section 5.3.
In Section 2 to 4 we assumed for convenience of writing that the field of coefficients $`๐`$ was algebraically closed. Yet the algebraic constructions of those sections require no extension of the field of definition of the group or the action. With the initial data described above, Theorem 2.13 produces a set of rational invariants in $`(z)^G`$ that generate $`(z)^G`$ by Theorem 2.15.
Rational invariants are obviously local invariants. We show that so are smooth representatives of algebraic invariants. The following definition formalizes the notion of a smooth representative of an algebraic function.
###### Definition 5.13
A smooth map $`F:๐ฐ^k`$ is a smooth zero of $`\{p_1,\mathrm{},p_\kappa \}(z)[\zeta _1,\mathrm{},\zeta _k]`$ if the coefficients of the $`p_i`$ are well defined on $`๐ฐ`$ and $`p_i(\overline{z},F(\overline{z}))=0`$ for all $`\overline{z}๐ฐ`$. In this case we also say that $`F`$ is a smooth zero of the ideal $`(p_1,\mathrm{},p_\kappa )`$.
###### Proposition 5.14
Assume $`F:๐ฐ^k`$ is a smooth zero of $`\{p_1,\mathrm{},p_\kappa \}(z)^G[\zeta _1,\mathrm{},\zeta _k]`$. If $`(p_1,\mathrm{},p_\kappa )`$ is a zero dimensional ideal then the components of $`F`$ are local invariants.
###### Proof.
Let $`p(z)^G[\zeta ]`$, that is $`p(z,\zeta )=_{\alpha ^n}a_\alpha (z)\zeta ^\alpha `$, where $`a_\alpha (z)(z)^G`$. Assume that $`p(\overline{z},F(\overline{z}))=0`$ for all $`\overline{z}๐ฐ`$. For any $`\overline{z}๐ฐ`$ and an infinitesimal generator $`v`$ there exits $`ฯต>0`$, such that $`\mathrm{exp}(\epsilon v,\overline{z})๐ฐ`$ whenever $`|\epsilon |<ฯต`$. Then $`p(\mathrm{exp}(\epsilon v,\overline{z}),F(\mathrm{exp}(\epsilon v,\overline{z})))=_{\alpha ^n}a_\alpha (\mathrm{exp}(\epsilon v,\overline{z}))F(\mathrm{exp}(\epsilon v,\overline{z}))^\alpha =0`$. Since the coefficients $`a_\alpha `$ are invariant $`_{\alpha ^n}a_\alpha (\overline{z})F(\mathrm{exp}(\epsilon v,\overline{z}))^\alpha =0`$ for all $`\overline{z}๐ฐ`$ and small enough $`\epsilon `$. Thus for a fixed point $`\overline{z}`$ all the values $`F(\mathrm{exp}(\epsilon v,\overline{z}))`$ for all sufficiently small $`\epsilon `$ are the common roots of the set of polynomials $`\{p_1,\mathrm{},p_\kappa \}`$. Since by the assumption the number of roots is finite, we conclude that $`F(\mathrm{exp}(\epsilon v,\overline{z}))=F\left(\mathrm{exp}(0v,\overline{z})\right)=F(\overline{z})`$ and thus the components of $`F(z)`$ are local invariants. โ
It follows from Theorem 5.7 that, through every point of $`๐ต`$, there exists a local cross-sections defined by linear equations over $``$. Conversely, we can consider a cross-section $`๐ซ`$, defined over $``$, that has non singular real points, meaning that the real part has the same dimension as the complex part. For any point $`\overline{z}๐ต๐ซ`$ where the rank of the matrix (3) does not drop, there is a neighborhood $`๐ฐ`$ on which $`๐ซ`$ defines a local cross-section, and such points are dense in $`๐ซ`$.
The $`(z)^G`$-zero of the zero dimensional ideal $`I^G=(G+P+(zg(\lambda ,z)))(z)^G[Z]`$ are precisely the replacement invariants. According to the previous proposition the smooth zeros of this ideal are local invariants. We characterize the tuple of normalized invariants as one of them.
###### Theorem 5.15
Let $`๐ซ`$ be an algebraic cross-section which, when restricted to an open set $`๐ฐ`$, defines a smooth cross-section. The tuple of normalized invariants $`\overline{\iota }z=(\overline{\iota }z_1,\mathrm{},\overline{\iota }z_n)`$ is the smooth zero of the ideal $`I^G`$ whose components agree with the coordinate functions on $`๐ซ๐ฐ`$.
###### Proof.
Let $`\overline{z}๐ฐ`$ be an arbitrary point, and let $`\overline{z}_0`$ be the point of intersection of $`๐ซ`$ with the connected component of $`๐ช_{\overline{z}}๐ฐ`$, containing $`\overline{z}`$. Then there exists $`\overline{\lambda }`$ in the connected component of the identity of $`๐ข`$, such that $`\overline{z}_0=\overline{\lambda }\overline{z}`$ so that $`(\overline{z},\overline{z}_0)`$ is a zero of the ideal $`I=O+P`$. By definition $`\overline{\iota }z(\overline{z})=\overline{z}_0`$ and therefore $`(\overline{z},\overline{\iota }z(\overline{z}))`$ is a zero of the ideal $`I`$ for all $`\overline{z}๐ฐ`$. Equivalently $`\overline{\iota }z`$ is a smooth zero of $`I^G`$. By Theorem 5.9 it is the unique tuple of local invariants that agree with the coordinate functions on $`๐ซ๐ฐ`$. โ
Therefore a replacement invariant not only generates algebraic invariants but their smooth representatives also generate local invariants.
###### Example 5.16
scaling. The action defined in Example 2.3 corresponds to the following action of the multiplicative group $`^{}`$:
$$g:\begin{array}{ccc}^{}\times ^2& & ^2\hfill \\ (\lambda ,z_1,z_2)& & (\lambda z_1,\lambda z_2).\hfill \end{array}$$
The action is semi-regular on $`^2\{(0,0)\}`$.
In Example 3.7 we chose the cross-section $`๐ซ`$ defined by $`z_1=1`$. The cross-section being of degree 1 there is a single associated replacement invariant that corresponds to the tuple $`(1,\frac{z_2}{z_1})`$ of rational invariants.
Let $`๐ฐ=\{(z_1,z_2)^2|z_10\}`$. The components of the smooth map $`F:๐ฐ^2`$ s.t. $`F(z_1,z_2)=(1,\frac{z_2}{z_1})`$ are the normalized invariants for the local cross-section $`๐ซ๐ฐ`$.
###### Example 5.17
translation+reflection. The action defined in Example 2.4 corresponds to the following action of the Lie group $`\times \{1,1\}`$ given by
$$g:\begin{array}{ccc}\times \{1,1\}\times ^2& & ^2\hfill \\ (\lambda _1,\lambda _2,z_1,z_2)& & (z_1+\lambda _1,\lambda _2z_2).\hfill \end{array}$$
The action is semi-regular on $`^2`$.
In Example 3.8 we chose the cross-section $`๐ซ`$ defined by $`z_2=z_1`$. There are two replacement invariants associated to $`๐ซ`$: $`\xi ^{(\pm )}=(\pm z_2,\pm z_2)`$. They both correspond to smooth maps $`F^{(\pm )}:^2^2`$ the components of which are local invariants.
Only $`(z_2,z_2)`$ coincides with the coordinate functions on $`๐ซ`$, that defines a local cross-section on $`๐ฐ=^2`$. The normalized invariants are thus $`(z_2,z_2)`$.
###### Example 5.18
rotation The action defined in Example 2.5 corresponds to the following action of the additive group $``$ given by
$$g:\begin{array}{ccc}\times ^2& & ^2\hfill \\ (t,z_1,z_2)& & (\frac{1t^2}{1+t^2}z_1\frac{2t}{1+t^2}z_2,\frac{2t}{1+t^2}z_1+\frac{1t^2}{1+t^2}z_2).\hfill \end{array}$$
The action is semi-regular on $`^2\{(0,0)\}`$.
In Example 3.9 we chose the cross-section $`๐ซ`$ defined by $`z_2=0`$. The replacement invariants associated to the cross-section $`๐ซ`$ are the $`\overline{(z)}^G`$-zeros of the ideal $`I^G=(Z_2,Z_1^2(z_1^2+z_2^2))`$.
The smooth maps $`F^{(\pm )}:^2\{(0,0)\}^2`$ s.t. $`F^{(\pm )}(z_1,z_2)=(0,\pm \sqrt{z_1^2+z_2^2})`$ are smooth zeros of $`I^G`$. Their components are thus local invariants.
The cross-section $`๐ซ`$ defines a local cross-section for instance on $`๐ฐ=^2\{(z_1,z_2)|z_1=0,z_20\}`$. As $`F^{(+)}|_{๐ซ๐ฐ}=z_1`$, the tuple of normalized invariants are $`(0,\sqrt{z_1^2+z_2^2})`$ on $`๐ฐ`$.
We conclude this section by linking the smooth invariantization and the algebraic invariantization introduced in Section 4.2. Recall that the algebraic invariantization was a map that associated a univariate polynomial over $`(z)^G`$ to univariate polynomials over $`๐[z]_๐ซ`$ (Definition 4.9).
###### Theorem 5.19
Let $`๐ซ`$ be an algebraic cross-section which, when restricted to an open set $`๐ฐ`$, defines a local cross-section. Let $`f:๐ฐ`$ be a smooth zero of a univariate polynomial $`\beta ๐(z)[\zeta ]`$. The smooth invariantization $`\overline{\iota }f`$ of $`f`$ is a smooth zero of the algebraic $`๐ซ`$-invariantization $`\iota \beta (z)^G[\zeta ]`$ of $`\beta `$.
###### Proof.
The polynomial $`\iota \beta (z,\zeta )=_{i=1}^kb_i(z)\zeta ^i`$, where $`b_i๐(z)^G`$. Any point $`\overline{z}๐ฐ`$ can obtained from the point $`\overline{z}_0๐ซ`$ by a composition of flows along infinitesimal generators of the group action. The argument will not change if we assume that $`\overline{z}=\mathrm{exp}(\epsilon v,\overline{z}_0)`$ is obtained by the flow along a single vector field. Then from the invariance of $`b_i(z)`$ and local invariance of $`\overline{\iota }f(z)`$ it follows that $`\overline{z}๐ฐ`$:
$`\iota \beta (\overline{z},\overline{\iota }f(\overline{z}))`$ $`=`$ $`{\displaystyle \underset{i=1}{\overset{k}{}}}b_i\left(\mathrm{exp}(\epsilon v,\overline{z}_0)\right)f\left(\mathrm{exp}(\epsilon v,\overline{z}_0)\right)^i`$
$`=`$ $`{\displaystyle \underset{i=1}{\overset{k}{}}}b_i(\overline{z}_0)\overline{\iota }f(\overline{z}_0)^i=\iota \beta (\overline{z}_0,\overline{\iota }f(z_0)),\text{ where }\overline{z}_0๐ซ๐ฐ.`$
From Proposition 4.12 it follows that $`\iota \beta `$ is divisible by $`\beta `$ when restricted to $`๐ซ`$. Thus $`\iota \beta (\overline{z}_0,f(\overline{z}_0))=0,\overline{z}_0๐ซ๐ฐ`$, since $`\beta (\overline{z},f(\overline{z}))0`$ on $`๐ฐ`$. It follows that $`\overline{\iota }f(z)`$ is a smooth zero of a polynomial $`\overline{\iota }\beta (z,\zeta )๐(z)^G[\zeta ]`$. โ
In particular if $`r(z)`$ is a rational function that is well defined on $`๐ฐ`$, then its smooth invariantization $`\overline{\iota }r(z)`$ is a smooth zero of the $`๐ซ`$-invariantization $`\iota (\zeta r(z))`$ of the polynomial $`\zeta r(z)`$. To discriminate the right one we only need to check that its value coincide with the one of $`r(z)`$ on $`๐ซ๐ฐ`$.
### 5.6 Moving frame map
We show that the invariantization map described in Section 4.2 generalizes the invariantization process described in . The latter is restricted to locally-free actions, and is based on the existence of a local $`๐ข`$-equivariant map $`\rho :๐ฐ๐ข`$. Although local freeness of the action guarantees the existence of $`\rho `$, due to the implicit function theorem, it might not be explicitly computable. We review the Fels-Olver construction, and prove that in the case of locally free actions it is equivalent to the one presented in Section 5.3.
###### Definition 5.20
An action of a Lie group $`๐ข`$ on a manifold $`๐ต`$ is *locally free* if for every point $`\overline{z}๐ต`$ its isotropy group $`๐ข_{\overline{z}}=\{\overline{\lambda }๐ข|\overline{\lambda }\overline{z}=\overline{z}\}`$ is discrete.
Local freeness implies semi-regularity of the action, the dimension of each orbit being equal to the dimension of the group. Theorem 4.4 from , can be restated as follows in the case of locally free actions.
###### Theorem 5.21
A Lie group $`๐ข`$ acts locally freely on $`๐ต`$ if and only if every point of $`๐ต`$ has an open neighborhood $`๐ฐ`$ such that there exists a map $`\rho :๐ฐ๐ข`$ that makes the following diagram commute. Here the map $`\overline{\mu }\overline{\mu }\overline{\lambda }^1`$ is chosen for the action of $`๐ข`$ on itself, and $`\overline{\lambda }`$ is taken in a suitable neighborhood (depending on the point of $`๐ฐ`$) of the identity in $`๐ข`$.
The map $`\rho `$ is locally $`๐ข`$-equivariant, $`\rho (\overline{\lambda }\overline{z})=\rho \overline{\lambda }^1`$ for $`\overline{\lambda }`$ sufficiently close to the identity, and is called *a moving frame map*. If $`๐ซ`$ is a cross-section, then the equation
$$\rho (\overline{z})\overline{z}๐ซ,$$
(5)
uniquely defines $`\rho (\overline{z})`$ in a sufficiently small neighborhood of the identity. In particular, $`\rho (\overline{z}_0)=e`$ for all $`\overline{z}_0๐ซ`$. Reciprocally, a moving frame map defines a local cross-section to the orbits: $`๐ซ=\{\rho (\overline{z})\overline{z}|\overline{z}๐ฐ\}๐ฐ`$.
In local coordinates, Condition (5) gives rise to implicit equations for expressing the group parameters in terms of the coordinate functions on the manifold. When the group acts locally freely, the local existence of smooth solutions is guaranteed by the transversality condition and the implicit function theorem. Since the implicit function theorem is not constructive, we might nonetheless not be able to obtain explicit formulas for the solution.
In \[11, Definition 4.6\] the invariantization of a function $`f`$ on $`๐ฐ`$ is defined as the function whose value at a point $`\overline{z}๐ฐ`$ is equal to $`f(\rho (\overline{z})\overline{z})`$. Next proposition shows that this moving frame based definition of invariantization is equivalent to Definition 5.8 given in terms of cross-section. The advantage of the latter definition is that it is not restricted to locally free actions.
###### Proposition 5.22
Let $`\rho `$ be a moving frame map on $`๐ฐ`$. Then $`\overline{\iota }f(\overline{z})=f(\rho (\overline{z})\overline{z}).`$
###### Proof.
Local invariance of $`f(\rho (z)z)`$ follows from the local equivariance of $`\rho `$, i. e. for $`\overline{\lambda }`$ sufficiently close to the identity:
$$f(\rho (\overline{\lambda }\overline{z})(\overline{\lambda }\overline{z}))=f(\rho (\overline{z})\overline{\lambda }^1(\overline{\lambda }\overline{z}))=f(\rho (\overline{z})\overline{z}.$$
Since $`\rho (z_0)=e`$ then $`f(\rho (\overline{z}_0)\overline{z}_0)=f(\overline{z}_0)`$ for all $`\overline{z}_0๐ซ`$. Thus $`f(\rho (z)z)`$ is locally invariant and equals to $`f`$, when restricted to $`๐ซ`$. The conclusion follows from Theorem 5.9. โ
Thus the moving frame map offers an approach to invariantization that is constructive up to the resolution of the implicit equations given by (5). In the algebraic case the moving frame map is defined by the ideal
$$M^e=\left(G+P+(Zg(\lambda ,z))\right)(z)[\lambda ].$$
Indeed, if $`(\overline{z},\overline{\lambda })`$ is a zero of $`M=M^e[z,\lambda ]`$, in an appropriate open set of $`๐ต\times ๐ข`$, then $`\overline{\lambda }\overline{z}๐ซ`$. The action is locally free if and only if $`M^e`$ is zero dimensional.
In this case, the smooth zero $`F:๐ฐ๐ข`$ of $`M^e`$, that is the identity of the group when restricted to $`๐ซ`$, provides a moving frame map $`\rho `$ on $`๐ฐ`$.
If one can obtain the map $`\rho `$ explicitly, the invariantization map can be computed using Proposition 5.22. Even in this favorable case, the expression for $`\rho `$ often involves algebraic functions which can prove difficult to manipulate symbolically. The purely algebraic approach proposed in Section 4 is more suitable for symbolic computation.
## 6 Additional examples
We first consider a linear action of $`SL_2`$ on $`๐^7`$ taken from . That latter paper presents an algorithm to compute a set of generators of the algebra of polynomial invariants for the linear action of a reductive group. The ideal $`O=(G+(Zg(\lambda ,z)))๐[z,Z]`$, where now $`g`$ is a polynomial map that is linear in $`z`$, is also central in the construction as a set of generators of $`๐[z]^G`$ is obtained by applying the Reynolds operator, which is a projection from $`๐[z]`$ to $`๐[z]^G`$, to generators of $`O+(Z_1,\mathrm{},Z_n)`$, the ideal of the null cone.
The fraction field of $`๐[z]^G`$ is included in $`๐(z)^G`$ but does not need to be equal. Conversely there is no known algorithm to compute $`๐[z]^G=๐(z)^G๐[z]`$ from the knowledge of a set of generators of $`๐(z)^G`$.
###### Example 6.1
We consider the linear action of $`SL_2`$ on $`๐^7`$ given by the following polynomials of $`๐[\lambda _1,\mathrm{},\lambda _4,z_1,\mathrm{},z_7]`$:
$$\begin{array}{c}g_1=\lambda _1z_1+\lambda _2z_2,g_2=\lambda _3z_1+\lambda _4z_2\\ g_3=\lambda _1z_3+\lambda _2z_4,g_4=\lambda _3z_3+\lambda _4z_4\\ g_5=\lambda _1^2z_5+2\lambda _1\lambda _2z_6+\lambda _2^2z_7,\\ g_6=\lambda _3\lambda _1z_5+\lambda _1\lambda _4+\lambda _2\lambda _3z_6+\lambda _2\lambda _4z_7,\\ g_7=\lambda _3^2z_5+2\lambda _3\lambda _4z_6+\lambda _4^2\end{array}$$
the group being defined by $`G=(\lambda _1\lambda _4\lambda _2\lambda _31)๐[\lambda _1,\lambda _2,\lambda _3,\lambda _4]`$.
The cross-section defined by $`P=(Z_1+1,Z_2,Z_3)`$ is of degree one. The reduced Grรถbner basis (for any term order) of the ideal $`I^e๐(z)[Z]`$ is indeed given by $`\{Z_1+1,Z_2,Z_3,Z_4r_2,Z_5r_3,Z_6r_4,Z_7r_1\}`$ where
$$\begin{array}{c}r_1=z_7z_{1}^{}{}_{}{}^{2}2z_2z_6z_1+z_{2}^{}{}_{}{}^{2}z_5,r_2=z_3z_2z_1z_4,\\ r_3=\frac{z_{3}^{}{}_{}{}^{2}z_72z_6z_4z_3+z_5z_{4}^{}{}_{}{}^{2}}{\left(z_1z_4z_3z_2\right)^2},r_4=\frac{z_1z_6z_4z_1z_3z_7+z_3z_2z_6z_2z_5z_4}{z_1z_4z_3z_2}\end{array}$$
By Theorem 3.6, $`๐(z)^G=๐(r_1,r_2,r_3,r_4)`$. In this case the rewriting of any rational invariant in terms of $`r_1,r_2,r_3,r_4`$ consists simply of the substitution of $`(z_1,z_2,z_3,z_4,z_5,z_6,z_7)`$ by $`(1,0,0,r_2,r_3,r_4,r_1)`$. We illustrate this by rewriting the five generating polynomial invariants computed in in terms of $`r_1,r_2,r_3,r_4`$:
$$\begin{array}{c}z_{2}^{}{}_{}{}^{2}z_52z_2z_6z_1+z_7z_{1}^{}{}_{}{}^{2}=r_1,z_3z_2z_1z_4=r_2,\\ z_{3}^{}{}_{}{}^{2}z_72z_6z_4z_3+z_5z_{4}^{}{}_{}{}^{2}=r_3r_{2}^{}{}_{}{}^{2},z_1z_3z_7z_3z_2z_6+z_2z_5z_4z_1z_6z_4=r_4r_2,\\ z_{6}^{}{}_{}{}^{2}z_7z_5=r_{4}^{}{}_{}{}^{2}r_1r_3,\end{array}$$
The reduced Grรถbner basis of $`O^e`$, relative to the total degree order with ties broken by reverse lexicographical order, has 9 elements:
$$\begin{array}{c}Z_{6}^{}{}_{}{}^{2}Z_7Z_5+r_1r_3r_{4}^{}{}_{}{}^{2},Z_6Z_4+r_3r_2Z_2r_4Z_4Z_3Z_7,\\ Z_5Z_4Z_3Z_6+r_3r_2Z_1r_4Z_3,Z_3Z_2Z_1Z_4r_2,\\ Z_2Z_6Z_1Z_7+r_4Z_2\frac{r_1}{r_2}Z_4,Z_2Z_5+Z_1r_4Z_6Z_1\frac{r_1}{r_2}Z_3,\\ Z_{2}^{}{}_{}{}^{2}+\frac{r_1}{r_3r_{2}^{}{}_{}{}^{2}}Z_{4}^{}{}_{}{}^{2}\frac{Z_7}{r_3}2\frac{r_4}{r_3r_2}Z_4Z_2,Z_{1}^{}{}_{}{}^{2}\frac{Z_5}{r_3}2\frac{r_4}{r_3r_2}Z_3Z_1+\frac{r_1}{r_3r_{2}^{}{}_{}{}^{2}}Z_{3}^{}{}_{}{}^{2}\\ Z_2Z_1\frac{r_4}{r_3}\frac{Z_6}{r_3}+\frac{r_1}{r_3r_{2}^{}{}_{}{}^{2}}Z_4Z_32\frac{r_4}{r_3r_2}Z_4Z_1,\end{array}$$
Though this Grรถbner basis is obtained without much difficulty, the example illustrates the advantage obtained by considering the construction with a cross-section: $`I^e`$ has a much simpler reduced Grรถbner basis than $`O^e`$.
We finally take a classical example in differential geometry: the Euclidean action on the second order jets of curves. The variables $`x,y_0,y_1,y_2`$ stand for the independent variable, the dependent variable, the first and the second derivatives respectively. We shall recognize the curvature as the non constant component of a replacement invariant.
###### Example 6.2
We consider the group defined by $`G=(\alpha ^2+\beta ^21,ฯต^21)๐[\alpha ,\beta ,a,b,ฯต]`$. The neutral element is $`(1,0,0,0,1)`$, the group operation is
$`(\alpha ^{},\beta ^{},a^{},b^{},ฯต^{})(\alpha ,\beta ,a,b,ฯต)=(\alpha \alpha ^{}\beta \beta ^{},\beta \alpha ^{}+\alpha \beta ^{},a+\alpha a^{}\beta b^{},b+\alpha a^{}+\alpha b^{},ฯตฯต^{})`$ and the inverse map $`(\alpha ,\beta ,a,b)^1=(\alpha ,\beta ,\alpha ab\beta ,\beta a\alpha b,ฯต)`$. The rational action on $`๐^4`$ we consider is given by the rational functions:
$$\begin{array}{c}g_1=\alpha x\beta y_0+a,g_2=ฯต\beta x+ฯต\alpha y_0+b,\\ g_3=\frac{\beta +\alpha y_1}{\alpha \beta y_0},g_4=\frac{y_2}{(\alpha \beta y_0)^3}.\end{array}$$
We have
$$O=\left(\left(1+y_1^2\right)^3Y_2^2\left(1+Y_1^2\right)^3y_2^2\right)$$
and if we consider the the cross section defined by $`P=(X,Y_0,Y_1)`$ the reduced Grรถbner basis of $`I^e=O^e+P`$ is
$$\{X,Y_0,Y_1,Y_2^2\frac{y_2^2}{(1+y_1^2)^3}\}.$$
According to Theorem 2.15 or Theorem 3.6
$$๐(z)^G=๐\left(\frac{y_2^2}{(1+y_1^2)^3}\right).$$
The two replacement invariants $`\xi =(\xi _1,\xi _2,\xi _3,\xi _4)`$ associated to the cross-sections are given by
$$\xi _1=0,\xi _2=0,\xi _3=0,\xi _4=\pm \sqrt{\frac{y_2^2}{(1+y_1^2)^3}}.$$
|
warning/0506/hep-th0506093.html
|
ar5iv
|
text
|
# Lagrange structure and quantization
## 1. Introduction
We suggest a method for path integral quantization of classical theories whose equations of motion are not necessarily variational. The key idea behind the method is that any classical dynamics can be uniformly converted into an equivalent topological field theory based on the action principle. Below in the introduction we informally comment on the main ingredients of the construction to give a preliminary impression of the quantization method we propose.
In the first place, we introduce a notion of *Lagrange structure* that generalizes the standard Lagrangian formalism more or less in the same sense as the Poisson geometry generalizes the symplectic one. The Lagrange structure does not require the equations of motion to be Lagrangian in usual sense, i.e. no integrating multiplier is assumed to exist bringing the equations to the variational form. The main ingredient of the Lagrange structure is the *Lagrange anchor* (denoted by $`V`$) that would be the inverse to an integrating multiplier $`\mathrm{\Lambda }`$ if the latter existed. The Lagrange anchor is required to satisfy a chain of compatibility conditions involving the equations of motion and the gauge generators for the gauge systems. If the anchor is invertible, these conditions will be equivalent to that providing the inverse matrix $`\mathrm{\Lambda }=V^1`$ to be an integrating multiplier for the equations of motion. Given the Lagrange structure, one can perform the path integral quantization of the classical theory even when the anchor is not invertible, and therefore, the dynamics do not admit any action functional.
Another explanation for the origin of the Lagrange structure is provided by the BV formalism , . The standard BV method describes the (gauge) system in terms of field-antifield supermanifold endowed with a master action and the canonical antibracket. Due to the Jacobi identity, the corresponding antibracket is automatically compatible with the BRST differential (associated with the master action) in the sense of the Leibnitz rule. For this standard case, the Lagrange anchor appears in the theory as an odd bivector defining canonical antibracket between fields and anti-fields. Similarly, as we show, the generic Lagrange structure equips an appropriately superextended original space with a BRST differential and a (non-canonical, weak) antibracket. The BRST differential carries all the information about the original classical theory, whereas the antibracket contains the ingredients needed for quantization. To make the formalism working (i.e. sufficient for the path integral quantization), the antibracket is required to satisfy the graded Jacobi identity in a โweakโ sense, i.e. up to homotopy w.r.t. the BRST differential. In this algebraic setting, all the compatibility conditions between classical equations of motion and Lagrange anchor are encoded in the graded Leibnitz rule for the BRST differential and the weak antibracket. As the antibracket can be degenerate, the BRST differential is not necessarily anti-hamiltonian vector field, in contrast to the standard Lagrangian theory.
Regarding the relaxed Jacobi identity for the weak antibracket, the Lagrange structure can be considered as Lagrangian counterpart of the weak Poisson structure studied in . Various particular types of the weak Poisson brackets were studied earlier in , , and recently in . The deformation quantization of weak Poisson manifolds was given in Ref. making use of superextension to the Kontsevich formality theorem , which had not been proved at that moment. An exhaustive proof has been given to the superformality theorem in the recent paper in connection with the deformation quantization of $`P_{\mathrm{}}`$-structure (=weak Poisson structure).
From the pure algebraic viewpoint, the defining relations for the Lagrange structure can be regarded as structure relations for $`L_{\mathrm{}}`$-algebra , . This $`L_{\mathrm{}}`$-algebra is a particular example of the homotopy analog of Schouten (= odd Poisson, Gerstenhaber) algebras studied in Ref. in the context of higher derived brackets. The higher (derived) brackets were also studied in various related contexts in Refs. .
As a next step towards quantization, we work out a BRST description for the Lagrange structure of a generic (i.e. not necessarily Lagrangian) classical theory. Making use of this BRST formalism, we construct a topological sigma-model along the lines of the AKSZ approach and its further development<sup>1</sup><sup>1</sup>1There is a great number of works devoted to the topological sigma-models. We mention only those references which are the most relevant for the construction we use. , , , . The master equation for the sigma-model action reproduces the defining relations for the Lagrange structure of the original theory much similar as the master equation for the Poisson sigma-model reproduces the Jacobi identity for the Poisson bivector , . By construction, the dynamics of the topological sigma-model is variational and we prove it is equivalent to the dynamics of the original classical theory. Quantizing this topological field theory one gets an arbitrary (not necessarily Lagrangian) original theory quantized. If the original theory had an action, the transition amplitude for the sigma-model can be explicitly integrated out in the bulk, resulting in the standard BV answer for the path integral of the original theory, with the BV master action built in a usual way from the original action.
Notice that the trivial Lagrange anchor $`V=0`$ is always admissible for any equations of motion. The trivial anchor, however, results in trivial quantization, with no quantum fluctuations appeared: The path integral for the topological sigma-model is reduced to integration only over the classical trajectories of the original theory. In this sense, the transition amplitudes in (non-Lagrangian) theories with trivial Lagrange anchor are similar to the classical transition amplitudes studied by Gozzi et al. for theories based on the action principle. In the general case, the quantum fluctuations will be trivial only for those degrees of freedom which belong to the kernel of the Lagrange anchor. The quantization technique we develop is explicitly covariant and does not require any special coordinate system adjusted for separating the anchor kernel from the other degrees of freedom<sup>2</sup><sup>2</sup>2Moreover, such an adjusted coordinate system does not exist if the rank of the anchor varies over the configuration space. However, this separation, whenever it is possible, allows for a simple interpretation. For example, if the degrees of freedom belonging to the kernel of the Lagrange anchor can be explicitly excluded from the classical equations of motion it is natural to consider them as auxiliary variables. The equations will become variational for the remaining degrees of freedom. The auxiliary degrees of freedom do not fluctuate in this case, while the other ones are quantized in the usual way with an action defined on the reduced space.. In this respect, the proposed path integral quantization is analogous to the deformation quantization of Poisson manifolds: the degrees of freedom from the kernel of the (regular) Poisson bivector correspond to the center of $``$-algebra upon quantization, remaining โnonquantizedโ in this sense.
Let us comment on the paper composition. In the next Section we elaborate on the origin of the Lagrange anchor and the related structures from the viewpoint of classical dynamics. In Sect. 3 we introduce the notion of the regular Lagrange structure and discuss its connection with strongly homotopy Schouten algebras ($`S_{\mathrm{}}`$-algebras for short). In Sect. 4, the BRST embedding is worked out for the Lagrange structure and the existence theorem is proved for the corresponding master equation. As a byproduct, we obtain one more geometric interpretation of the Lagrange structure as an infinitesimal deformation of certain Lagrangian submanifold in the cotangent bundle over the space of trajectories. In Sect. 5, proceeding along the lines of the AKSZ procedure, we construct a topological sigma-model related to the BRST complex which have been built for the Lagrange structure in previous section. Being effectively localized on the boundary, the dynamics of this topological sigma-model in $`d+1`$ dimensions are proved to be equivalent to the original $`d`$-dimensional theory. Quantizing the topological sigma model we get the path integral quantization of the original (non-Lagrangian) dynamics. Finally, in Sect. 6, we consider several examples illustrating the quantization formalism proposed. As the first example, we show that the quantization of the usual Lagrangian gauge theory, being performed by our method, is equivalent to the standard BV-quantization. Next, as the second example, we detail our quantization method for the general classical theory given by a set of independent (not necessarily Lagrangian) equations of motion without gauge symmetry. As the third example we consider the systems with the equations of motion being the first-order ODEs $`\dot{x}{}_{}{}^{i}=h^i(x)`$. For these systems, we identify purely algebraic (i.e. containing no time derivatives) Lagrange anchors with the Poisson bivectors compatible to the vector $`h`$. When the Poisson bivector is nondegenerate, the equations of motion are Hamiltonian and can be derived from the first-order action functional, while in the degenerate case, these equations are not necessarily either Hamiltonian, or variational. However, our method provides a natural embedding of this theory into the equivalent topological sigma-model, so the theory can be quantized through this embedding. The fourth example deals with the Maxwell electrodynamics formulated in terms of the strength tensor of electromagnetic field. The Maxwell equations for the strength tensor are known to be non-Lagrangian and linearly dependent. We find an explicitly covariant Lagrange anchor for these equations and quantize them following our general prescription. As a result, we obtain a path integral giving precisely the same transition amplitude as usual Faddeev-Popov path integral defined in terms of the electromagnetic potentials. This exemplifies the way in which our quantization method can be made applicable to non-Lagrangian field theories formulated from outset in terms of strength tensors, like the high-spin gauge fields .
## 2. Lagrange structure: a preliminary exposition
In this section, we give a down-to-earth explanation to the origin of the Lagrange anchor and related structures. It is the structures which are behind the path integral quantization of dynamical systems whose equations of motion do not necessarily follow from the action principle. More rigorous consideration of the subject is given in Sect 3.
In the standard Lagrangian formalism, the mechanical system is specified by an action functional $`S:M`$ defined on the space of all trajectories (histories) $`M`$ over the configuration space of the system. The true physical trajectories are postulated to deliver local minima to $`S`$ that leads to the equations of motion of the form
(1)
$$T_i(x)_iS(x)=0,$$
where $`x^i`$ are local coordinates on $`M`$. When all the critical points of $`S(x)`$ are non-degenerate (and hence isolated) one said about a non-degenerate Lagrangian theory; otherwise there can exist continuous families of trajectories satisfying the same boundary conditions, in which case one says about a gauge invariant (or degenerate) Lagrangian theory.
The equations of motion (1) can be understood as defined by an exact 1-form $`T=dS`$ on an infinite-dimensional manifold $`M`$. An immediate consequence of this interpretation is the Helmholtz criterion
(2)
$$dT=0,$$
that is a necessary condition for the equations $`T=0`$ to come from the action principle. Due to the Poincarรฉ lemma, the closedness of $`T`$ ensures the existence of a local action functional $`S_U`$ defined on any contractible open set $`UM`$ such that $`T|_U=dS_U`$. If $`x_0U`$ is a solution to $`T=0`$, one can use $`S_U`$ to perform the quasi-classical (= perturbative in $`\mathrm{}`$) quantization of the system โnear the classical trajectory $`x_0`$โ.
So, the standard Lagrangian formalism requires the equations of motion to satisfy two conditions: (i) they must be components of a one-form on the cotangent bundle to the space of trajectories, and (ii) this one-form has to be closed, i.e satisfying the Helmholtz condition (2). Of course, there are many physically interesting systems whose equations of motion do not satisfy even the first condition, not to mention the second one. The first natural step towards generalizing the Lagrangian formalism is to replace the cotangent bundle $`T^{}M`$, where the equations (1) take the values, by an arbitrary vector bundle $`M`$, that we term a *dynamics bundle*. The space of true physical trajectories is then identified with zero locus of some section in the dynamics bundle: $`T\mathrm{\Gamma }()`$. (Relaxing the Helmholtz condition will be the next step, addressed after relation (7)). If $`e^a`$ is a local frame of sections of $``$ over a trivializing coordinate chart $`UM`$, so that $`T=T_a(x)e^a`$, then instead of (1) we get the equations of motion in the form:
(3)
$$T_a(x)=0.$$
Notice that we do not assume that $`dimT^{}M=dim`$, so the โnumberโ of equations is allowed to be less or greater than $`dimM`$. (Of course, in the infinite dimensional context, the notion of dimension needs clarification. An appropriate definition can be done, for example, in the case of local theories, i.e. when (3) is a system of PDEโs.)
The question arises what might be an analogue for the integrability condition (2) that can ensure the existence of a local action for the equations (3). To answer this question, let us first consider the case when $`dim=dimT^{}M`$. In this case, the answer is given by existence of a vector bundle isomorphism $`\mathrm{\Lambda }:T^{}M`$ such that $`T^{}=\mathrm{\Lambda }(T)`$ is a closed 1-form. In terms of local coordinates this reads
(4)
$$T_i^{}(x)=\mathrm{\Lambda }_i^a(x)T_a(x),$$
where $`\mathrm{\Lambda }_i^a(x)`$ is a non-degenerate matrix. The equations $`T^{}=0`$ are obviously equivalent to Eqs.(3) in the sense that both have the same solutions. Checking the closedness condition for the 1-form $`T^{}`$ leads to the following relations:
(5)
$$dT^{}=0dT_a=\mathrm{\Lambda }^bC_{ab}^dT_d+G_{ab}\mathrm{\Lambda }^b.$$
Here we consider $`T_a`$ and $`\mathrm{\Lambda }^a=\mathrm{\Lambda }_i^adx^i`$ as a collection of $`0`$\- and $`1`$-forms defined on a coordinate chart $`U`$ and labelled by index $`a`$. The structure functions $`G_{ab}`$ and $`C_{ab}^c`$ are, respectively, symmetric and antisymmetric in indices $`a,b`$. In particular, the functions $`C_{ab}^c`$ enter to the Maurer-Cartan equation for the basis $`1`$-forms $`\mathrm{\Lambda }^a`$:
(6)
$$d\mathrm{\Lambda }^a=C_{bc}^a\mathrm{\Lambda }^b\mathrm{\Lambda }^c.$$
It is the question of finding the โintegrating multiplierโ $`\mathrm{\Lambda }`$ or investigating obstructions to its existence that the inverse problem of variational calculus deals with. As soon as $`\mathrm{\Lambda }`$ is known, one can define a local action $`S(x)`$ such that $`dS=T_a\mathrm{\Lambda }^aT^{}`$. Having the local action at hands, one can perform a quasi-classical path integral quantization in a vicinity of any classical solution.
Since $`\mathrm{\Lambda }:T^{}M`$ is an isomorphism of vector bundles, there is the inverse map $`V=\mathrm{\Lambda }^1:^{}TM`$ defining (and defined by) a section $`V=V_a^i(x)e^a_i\mathrm{\Gamma }(TM)`$. The integrability condition (5) is then equivalent to the following relation in terms of $`V_a`$ and $`T_a`$:
(7)
$$V_a^i_iT_bV_b^i_iT_a=C_{ab}^dT_d.$$
Now one may forget about $`\mathrm{\Lambda }`$, taking the last relation as a definition of the integrability condition, valid for an arbitrary vector bundle $`M`$. In doing so, one has no need to require the homomorphism $`V:TM`$ to be of constant rank over $`M`$. It is the map $`V`$ subject to relations (7) which we shall call the *Lagrange anchor*. The general and precise definition of the Lagrange anchor is given in the next section. When the equations of motion are defined by a 1-form on $`M`$, i.e $`=TM`$, the relation (7) still remains much less restrictive for the dynamics than the Helmholtz condition (5), as the anchor $`V_j^i(x)`$ satisfying (7), is not required to be invertible. The usual Lagrangian dynamics (1) corresponds to the special case where $`=TM`$ and $`V_i^j=\delta _i^j=\mathrm{\Lambda }_i^j`$.
For the general Lagrange anchor (7), no integrating multiplier $`\mathrm{\Lambda }=V^1`$ is required to exist. So, in general, existence of the Lagrange anchor (7) does not mean existence of a local action $`S_U`$ in the vicinity $`U`$ of given solution $`x_0`$. Nonetheless, if $`x_0`$ is a regular point, in the sense that the ranks of matrices $`(V_a^i)`$ and $`(_iT_a)`$ are constant over $`U`$, the following statement holds true:
###### Proposition 2.1.
Given a pair of sections $`(T,V)`$ satisfying (7), then for any regular solution $`x_0M`$ of (3) one can find a coordinate system $`(y^1,\mathrm{},y^m,z^1,\mathrm{},z^k)`$ centered at $`x_0`$ together with a set of local functions $`S(y)`$, $`E^1(y),\mathrm{},E^k(y)`$ such that equations $`T_a(y,z)=0`$ are equivalent to
$$\frac{S(y)}{y^I}=0,z^J=E^J(y),$$
where $`k=\mathrm{rank}(_iT_a(x_0))\mathrm{rank}(G_{ab}(x_0))`$ and $`G_{ab}=V_a^i_iT_b`$.
We prove this proposition in Sect. 4.6. It is natural to call the function $`S(y)`$, depending on a part of degrees of freedom, a partial action. On the surface defined by the equations for $`z`$โs<sup>3</sup><sup>3</sup>3In fact, these equations are not required to be explicitly solved w.r.t. $`z`$โs, just a unique existence is required for the solution with appropriate initial data., the dynamics become Lagrangian for the other degrees of freedom, more or less in the same sense as the Hamiltonian dynamics become symplectic upon reduction to a symplectic leaf of a regular Poisson structure. It should be noted that the path integral quantization method developed in this paper does not require any special coordinate system. Moreover, the method is insensitive to the rank of the matrix $`V_a^i(x)`$ and remains applicable to the cases where no (partial) action can exist.
Let $`\mathrm{\Sigma }`$ denote the set of all solutions, i.e. $`\mathrm{\Sigma }=\{pM|T_a(p)=0\}`$. Equations of motion (3) are called dependent if there is a vector bundle $`_1M`$ and a bundle homomorphism $`Z:_1`$ such that $`T\mathrm{ker}Z`$ and $`\mathrm{rank}(Z|_\mathrm{\Sigma })0`$. In terms of local coordinates this means
(8)
$$Z_A^aT_a0,$$
where the section $`Z=Z_A^ae_ae^A\mathrm{\Gamma }(^{}_1)`$ does not vanish on $`\mathrm{\Sigma }`$ identically.
Equations of motion (3) are said to be gauge invariant if there exists a vector bundle $`_1M`$ together with a bundle homomorphism $`R:_1TM`$ such that the corresponding section $`R=R_\alpha ^ie^\alpha _i\mathrm{\Gamma }(_1^{}TM)`$ does not vanish on $`\mathrm{\Sigma }`$ identically and
(9)
$$R_\alpha ^i_iT_a|_\mathrm{\Sigma }=0.$$
For an ordinary gauge theory with action $`S`$ and gauge symmetry generators $`R_\alpha ^i_i`$, Rels. (8), (9) take the form
(10)
$$R_\alpha ^i_iS=0,R_\alpha ^i_i_jS|_{dS=0}=0.$$
As is seen, in the ordinary Lagrangian gauge theory, the role of $`R`$โs is two-fold: the same $`R`$โs generate the gauge symmetries of the equations of motion (9) and describe the functional dependence (the Noether identities) between them (8). If the equations follow from the action principle, the gauge symmetry means dependence of equations of motion, and vice versa. In the general (non-Lagrangian) case, the generators $`R`$โs and $`Z`$โs may be completely independent from each other. In particular, it is possible to have dependent but not gauge invariant equations of motion and vice versa.
Rels. (8), (9) considered independently from (7) define a gauge algebra structure irrespectively to existence of the Lagrangian or Hamiltonian formalism. The BRST imbedding for such a generic gauge algebra was systematically described in Ref. along the usual lines of the BRST theory<sup>4</sup><sup>4</sup>4 Also in Ref., the stress was made on consistent combining of the gauge algebra relations (8, 9) and the (weak) Poisson structure. That was aimed at deformation quantization of the generic gauge systems (not necessarily having Poisson structure on $`M`$) through constructing a star-product which is associative only for the on-shell gauge invariants, not for all functions on $`M`$. Examples of the non-Lagrangian and/or non-Hamiltonian models where the BRST description is an efficient tool for studying the classical dynamics, can be found, e.g., in Refs. , . . Examining the compatibility conditions between the defining relations for the gauge algebra (8, 9) and the Lagrange structure (7), one can find rich algebraic and geometric structures that are systematically studied in the following sections. These are the structures which provide the possibility to (path-integral) quantize the system even if its classical dynamics does not admit action principle.
## 3. Regular Lagrange structure and $`S_{\mathrm{}}`$-algebra
To summarize the previous discussion in a more formal way, a classical system is specified by a vector bundle $`M`$ over the space of trajectories $`M`$, and a section $`T\mathrm{\Gamma }()`$ playing the role of equations of motion. The space of true trajectories of the system is then identified with zero locus $`\mathrm{\Sigma }`$ of $`T`$: $`\mathrm{\Sigma }=\{xM|T(x)=0\}`$. In what follows we refer to $`\mathrm{\Sigma }`$ as a shell.
### 3.1. Lagrange structure
By a *Lagrange structure* for a classical system $`(,T)`$ we understand $``$-linear map $`d_{}:\mathrm{\Gamma }(^n)\mathrm{\Gamma }(^{n+1})`$ obeying conditions:
1. $`d_{}T=0`$ ,
2. $`d_{}`$ is a derivation of degree 1, i.e.
$$d_{}(AB)=d_{}AB+(1)^nAd_{}B,A\mathrm{\Gamma }(^n),B\mathrm{\Gamma }(^{}).$$
Here we identify $`\mathrm{\Gamma }(^0)`$ with $`C^{\mathrm{}}(M)`$.
Due to the Leibnitz identity (ii), in each trivializing chart $`UM`$ the operator $`d_{}`$ is completely specified by its action on coordinate functions $`x^i`$ and basis sections $`e^a`$ of $`|_U`$:
(11)
$$d_{}x^i=V_a^i(x)e^a,d_{}e^a=\frac{1}{2}C_{bc}^a(x)e^be^c.$$
Applying $`d_{}`$ to the section $`T=T_ae^a`$, one can see that the property (i) reproduces the integrability condition (7):
$$0=d_{}T=\frac{1}{2}(V_a^i_iT_bV_b^i_iT_aC_{ab}^cT_c)e^ae^b.$$
The first relation from (11) means also that $`d_{}`$ defines a bundle homomorphism $`V:^{}TM`$.
In the particular case where $`d_{}^{}{}_{}{}^{2}=0`$, $`T`$ is nothing but a closed $`1`$-$``$-form associated to the Lie algebroid with the anchor $`V`$. Although for the general Lagrange structure (i)-(ii), the differential $`d_{}`$ is not required to be nilpotent, we call $`V`$ the *Lagrange anchor*. For the Lagrange anchor, the requirement of identical nilpotency is replaced by a relaxed condition $`d_{}^{}{}_{}{}^{2}T=0`$ following from (i). This weaker requirement can have further off-shell consequences that are derived in the next section under certain regularity conditions on $`T`$.
### 3.2. Regularity conditions
Let $`(,T,d_{})`$ be a Lagrange structure with shell $`\mathrm{\Sigma }`$. The Lagrange structure is said to be regular of type $`(m,n)`$ if $`\mathrm{\Sigma }\mathrm{}`$ and there exists a finite chain of vector bundles $`_kM`$ together with $`M`$-bundle homomorphisms
(12)
$$0_m\mathrm{}_1\stackrel{R}{}TM\stackrel{J}{}\stackrel{Z}{}_1\mathrm{}_n0,$$
such that
1. the map $`J`$ is defined by section $`T\mathrm{\Gamma }(T^{}M)`$, where $``$ is any connection on $``$;
2. there is a neighbourhood $`UM`$ of $`\mathrm{\Sigma }`$ such that all the homomorphisms (12) have constant ranks over $`U`$;
3. upon restriction to $`\mathrm{\Sigma }`$, the chain (12) makes an exact sequence.
Several remarks are in order concerning this definition.
Remark 1. The regularity condition ensures that $`\mathrm{\Sigma }M`$ is a smooth submanifold.
Remark 2. When exist, the homomorphisms (12) are not unique off shell. Thinking of these homomorphisms as sections of the corresponding vector bundles,
(13)
$$R=R_\alpha ^ie^\alpha _i,J=_iT_adx^ie^a,Z=Z_A^ae_ae^A,\mathrm{},$$
one can add to them any sections vanishing on $`\mathrm{\Sigma }`$, leaving the properties (a)-(c) unaffected. In particular, making a shift
(14)
$$ZZ+T_aW_A^{ab}e^Ae_b,$$
if necessary, we can always choose $`Z`$ in such a way that $`T\mathrm{ker}Z`$, cf. (8).
Remark 3. In the definition above one can pass from the chain (12) to the transpose one by replacing each vector bundle with its dual and inverting all the arrows. The transpose chain meets the same conditions (a)-(c) as the original one.
Remark 4. The condition (c) means that $`R`$โs and $`Z`$โs, defining the chain links in (12), are to be understood as the generators of gauge symmetry (9) and the generators of Noether identities (8). Having in mind this interpretation, we term $`_1`$ and $`_1`$ the gauge algebra bundle and the Noether identity bundle, respectively. Accordingly, $``$ is referred to as the dynamics bundle .
In this paper we deal mostly with quantization of regular $`(1,1)`$-type Lagrange structures associated to the four-term sequences
(15)
$$0\stackrel{R}{}TM\stackrel{J}{}\stackrel{Z}{}๐ข0.$$
In other words, we consider a set of gauge invariant and linearly dependent equations of motion (3), (8), (9) with the generators of gauge symmetry $`R_\alpha `$ and Noether identities $`Z_A`$ chosen in a linearly independent way. In the ordinary Lagrangian gauge theory the dynamics bundle coincides with the cotangent bundle ($`=T^{}M`$), the Noether identity bundle coincides with the gauge algebra bundle ($`_1=_1`$) and the generators of gauge symmetry coincide with the generators of Noether identities ($`R=Z`$). For the general system of type $`(1,1)`$, any of these coincidences should not necessarily occur, e.g.: the gauge algebra bundle $`_1`$ and the bundle of Noether identities $`_1`$ can be different even by dimension. In Sect. 6.4 we give an example of quantizing the dynamical system of type (0,1) which is not Lagrangian, although it has a nontrivial Lagrange anchor. This means the theory has dependent equations of motion having no gauge symmetry.
### 3.3. Completeness
A regular Lagrange structure $`(,T,d_{})`$ is called complete if
(16)
$$TM=\mathrm{Im}V\mathrm{Im}R,$$
where $`V:^{}TM`$ is the Lagrange anchor corresponding to $`d_{}`$, and $`R`$ is determined by (12). In other words, the completeness means that the tangent bundle is spanned by the Lagrange anchor and the gauge symmetry generators. It is easy to find that the number $`m`$ of Lagrangian equations in Proposition 2.1 is equal to the rank of the matrix $`(R_\alpha ^i,V_a^i)`$. Hence, for a complete Lagrange structure, all the equations of motion turn out to be (locally) Lagrangian. In this paper we consider regular Lagrange structures that are not necessarily complete.
### 3.4. Physical observables
Given a regular Lagrange structure $`(,T,d_{})`$ with the shell $`\mathrm{\Sigma }`$, we say that $`fC^{\mathrm{}}(M)`$ is a trivial function if it vanishes on shell, i.e. $`f|_\mathrm{\Sigma }=0`$. The subspace of trivial functions is denoted by $`C^{\mathrm{}}(M)^{\mathrm{triv}}`$. It follows from the regularity conditions that
$$fC^{\mathrm{}}(M)^{\mathrm{triv}}f=K^aT_a,$$
for some $`K\mathrm{\Gamma }(^{})`$. Function $`fC^{\mathrm{}}(M)`$ is said to be invariant if for any section $`\epsilon =\epsilon ^\alpha e_\alpha \mathrm{\Gamma }(_1)`$ there exist a section $`F\mathrm{\Gamma }(^{}_1^{})`$ such that
(17)
$$\epsilon ^\alpha R_\alpha ^i_if=\epsilon ^\alpha F_\alpha ^aT_aC^{\mathrm{}}(M)^{\mathrm{triv}},$$
Here $`\{R_\alpha \}`$ is an (over)complete basis of gauge generators (9). Again, in view of the regularity conditions one can rewrite (17) in a more compact way: $`R_\alpha ^i_if|_\mathrm{\Sigma }=0`$. The subspace of invariant functions is denoted by $`C^{\mathrm{}}(M)^{\mathrm{inv}}`$. In view of condition (9) the trivial functions are automatically invariant, so we can write
$$C^{\mathrm{}}(M)^{\mathrm{triv}}C^{\mathrm{}}(M)^{\mathrm{inv}}C^{\mathrm{}}(M).$$
Two invariant functions $`f_1,f_2C^{\mathrm{}}(M)^{\mathrm{inv}}`$ are considered as equivalent if they coincide on shell,
(18)
$$f_1f_2f_1f_2=K^aT_aC^{\mathrm{}}(M)^{\mathrm{triv}}.$$
The space of physical observables $`๐ซ`$ is now defined as the quotient of the space of invariant functions by the space of trivial ones,
(19)
$$๐ซ=C^{\mathrm{}}(M)^{\mathrm{inv}}/C^{\mathrm{}}(M)^{\mathrm{triv}}$$
Let us recall that $`M`$ is an infinite dimensional space of trajectories, and $`T_a=0`$ are the differential equations whose solutions are parametrized by initial data. The initial data transformed into each other by the gauge symmetry transformations are considered as equivalent. The space of inequivalent initial data can then be understood as a physical phase space $`M_{phys}`$. In the case where $`M_{phys}`$ happen to be a smooth Hausdorf manifold we have an equivalent definition of $`๐ซ`$ as the space of smooth functions on $`M_{phys}`$, i.e. $`๐ซ=C^{\mathrm{}}(M_{\mathrm{phys}})`$.
### 3.5. $`S_{\mathrm{}}`$ \- algebras
An $`S_{\mathrm{}}`$-algebra ($`S`$ for Schouten) is a $`_2`$-graded, supercommutative and associative algebra $`A`$ endowed with a sequence of odd linear maps $`S_n:A^nA`$ such that
1. $`S_n(\mathrm{},a_k,a_{k+1},\mathrm{})=(1)^{ฯต(a_k)ฯต(a_{k+1})}S_n(\mathrm{},a_{k+1},a_k,\mathrm{})`$,
$`ฯต(a)`$ being the parity of a homogeneous element $`aA`$.
2. $`aS_n(a_1,\mathrm{},a_{n1},a)`$ is a derivation of $`A`$ of the parity $`1+_{k=1}^{n1}ฯต(a_k)(\mathrm{mod}\mathrm{\hspace{0.33em}2})`$.
3. For all $`n0`$,
$$\underset{k+l=n}{}\underset{(k,l)\mathrm{shufle}}{}(1)^ฯตS_{l+1}(S_k(a_{\sigma (1)},\mathrm{},a_{\sigma (k)}),a_{\sigma (k+1)},\mathrm{},a_{\sigma (k+l)})=0,$$
where $`(1)^ฯต`$ is the natural sign prescribed by the sign rule for a permutation of homogeneous elements $`a_1,\mathrm{},a_nA`$.
Recall that a $`(k,l)`$-shuffle is a permutation of indices $`1,2,\mathrm{},k+l`$ satisfying $`\sigma (1)<\mathrm{}<\sigma (k)`$ and $`\sigma (k+1)<\mathrm{}<\sigma (k+l)`$.
When $`S_0=0`$ we say about a flat $`S_{\mathrm{}}`$-algebra. In this case $`S_1:AA`$ is a nilpotent differential, and $`S_2`$ induces an odd Poisson structure on corresponding cohomology. An odd Poisson algebra can thus be regarded as $`S_{\mathrm{}}`$-algebra with bracket $`S_2:AAA`$ and all other $`S_k=0`$. In fact, properties (i) and (iii) characterize $`L_{\mathrm{}}`$-algebras. See for recent discussion of $`S_{\mathrm{}}`$-algebras.
It turns out that any regular Lagrange structure of type $`(m,n)`$ gives rise to an $`S_{\mathrm{}}`$-algebra structure on the supercommutative algebra of sections
(20)
$$A=\mathrm{\Gamma }\left(^{}\underset{k=1}{\overset{m}{}}S^{}(\mathrm{\Pi }^k_k)\underset{l=1}{\overset{n}{}}S^{}(\mathrm{\Pi }^{l+1}_l)\right).$$
Here $`S^{}`$ stands for symmetric tensor powers (in the $`_2`$-graded sense) and $`\mathrm{\Pi }`$ denotes the parity reversion operation, i.e. $`\mathrm{\Pi }`$ is a vector bundle over $`M`$ whose fibers are odd linear spaces. By definition, $`\mathrm{\Pi }^2=\mathrm{id}`$ and $`S^{}(\mathrm{\Pi })=^{}`$.
In the next section, applying the machinery of the BRST theory, we give an explicit description for $`S_{\mathrm{}}`$-algebras associated with $`(1,1)`$-type Lagrange structures. Extension to the $`(m,n)`$-type Lagrange structures is straightforward.
## 4. BRST imbedding
### 4.1. Ambient Poisson supermanifold
Let $`(,T,d_{})`$ be a regular Lagrange structure of type $`(1,1)`$ corresponding to the four-term sequence (15). Following the general line of ideas of BRST theory, we have to realize the original space of trajectories $`M`$ as a body of an appropriate $``$-graded supermanifold $`๐ฉ`$. Let us choose $`๐ฉ`$ to be the total space of the vector bundle
(21)
$$๐ฉ=\mathrm{\Pi }(^{})T^{}M\mathrm{\Pi }(^{})(๐ข๐ข^{}),$$
where $`,`$ and $`๐ข`$ are the bundles of gauge algebra, dynamical equations and the Noether identities respectively (15), see Remark 4 of Sect. 3.2. The base $`M`$ is canonically imbedded into $`๐ฉ`$ as zero section. Besides the Grassman parity, the fibers of each direct summand in (21) are endowed with an additional $``$-grading, called the ghost number. For simplicity, to avoid cumbersome sign factors, we assume the base $`M`$ to be an ordinary (even) manifold, that corresponds to the case of mechanical systems without fermionic degrees of freedom. Then the Grassman parity of the fibers is correlated to $``$-grading in a simple way: the even coordinates have even ghost numbers, while the odd coordinates have odd ghost numbers. We also equip $`๐ฉ`$ with a pair of auxiliary $``$-gradings called the momentum- and resolution degrees ($`m`$\- and $`r`$-degrees, for short) that will be used later for proving the existence theorem for the BRST charge. The information about the gradings of local coordinates is arranged in the table:
| base and fibers | $`M`$ | $`T^{}M`$ | $``$ | $`^{}`$ | $``$ | $`^{}`$ | $`๐ข`$ | $`๐ข^{}`$ |
| --- | --- | --- | --- | --- | --- | --- | --- | --- |
| local coordinates | $`x^i`$ | $`\overline{x}_j`$ | $`c^\alpha `$ | $`\overline{c}_\beta `$ | $`\eta _a`$ | $`\overline{\eta }^b`$ | $`\xi _A`$ | $`\stackrel{}{\overline{\xi }}^B`$ |
| $`ฯต`$=Grassmanโs parity | 0 | 0 | 1 | 1 | 1 | 1 | 0 | 0 |
| $`\mathrm{gh}`$ = ghost number | 0 | 0 | 1 | -1 | -1 | 1 | -2 | 2 |
| $`\mathrm{Deg}`$ = momentum degree | 0 | 1 | 0 | 1 | 0 | 1 | 0 | 1 |
| $`\mathrm{deg}`$ = resolution degree | 0 | 1 | 0 | 2 | 1 | 0 | 2 | 0 |
Table 1
Splitting all the local coordinates into the โposition coordinatesโ $`\phi ^I=(x^i,c^\alpha ,\eta _a,\xi _A)`$ and โmomentaโ $`\overline{\phi }_J=(\overline{x}_i,\overline{c}_\alpha ,\overline{\eta }^a,\overline{\xi }^A)`$, we can write
(22)
$$\begin{array}{cc}\mathrm{gh}(\overline{\phi }_I)=\mathrm{gh}(\phi ^I),\hfill & ฯต(\overline{\phi }_I)=ฯต(\phi ^I),\hfill \\ \mathrm{Deg}(\overline{\phi }_I)=1,\hfill & \mathrm{Deg}(\phi ^I)=0.\hfill \end{array}$$
Though the submanifold $`T^{}M๐ฉ`$ is an ordinary manifold, we do not include the fibers of $`T^{}M`$ into the body of $`\times _2`$-graded manifold $`N`$ and treat $`\overline{x}`$โs as formal variables.
Fixing a linear connection $`=_{}_{}_๐ข`$ on $`๐ข`$, we endow $`๐ฉ`$ with the exact symplectic structure $`\omega =d\mathrm{\Lambda }`$, where
(23)
$$\begin{array}{c}\mathrm{\Lambda }=\overline{x}_idx^i+\overline{c}_\alpha c^\alpha +\overline{\eta }^a\eta _a+\overline{\xi }^A\xi _A,\\ c^\alpha =dc^\alpha +dx^i\mathrm{\Gamma }_{i\beta }^\alpha c^\beta ,\end{array}$$
and similar expressions are assumed for covariant differentials of $`\eta `$โs and $`\xi `$โs. The corresponding Poisson brackets of local coordinates read
(24)
$$\begin{array}{ccc}\{\overline{\eta }^b,\eta _a\}=\delta _a^b,\hfill & \{\overline{x}_i,\eta _a\}=\mathrm{\Gamma }_{ia}^b\eta _b,\hfill & \{\overline{x}_i,\overline{\eta }^b\}=\mathrm{\Gamma }_{ia}^b\overline{\eta }^a,\hfill \\ \{\overline{c}_\alpha ,c^\beta \}=\delta _\alpha ^\beta ,\hfill & \{\overline{x}_i,c^\alpha \}=\mathrm{\Gamma }_{i\beta }^\alpha c^\beta ,\hfill & \{\overline{x}_i,\overline{c}_\beta \}=\mathrm{\Gamma }_{i\beta }^\alpha \overline{c}_\alpha ,\hfill \\ \{\overline{\xi }^A,\xi _B\}=\delta _B^A,\hfill & \{\overline{x}_i,\xi _A\}=\mathrm{\Gamma }_{iA}^B\xi _B,\hfill & \{\overline{x}_i,\overline{\xi }^A\}=\mathrm{\Gamma }_{iB}^A\overline{\xi }^B,\hfill \end{array}$$
$$\begin{array}{cc}\{\overline{x}_i,x^j\}=\delta _i^j,\hfill & \{\overline{x}_i,\overline{x}_j\}=R_{ija}^b\overline{\eta }^a\eta _b+R_{ij\alpha }^\beta c^\alpha \overline{c}_\beta +R_{ijA}^B\overline{\xi }^A\xi _B,\hfill \end{array}$$
and the other brackets vanish. The structure functions determining the Poisson brackets of $`\overline{x}_i`$ and $`\overline{x}_j`$ are just components of the curvature tensor of $``$.
Clearly, the equations $`\overline{\phi }_I=0`$ define the Lagrangian submanifold
(25)
$$=\mathrm{\Pi }()๐ข๐ฉ,$$
and the supercommutative algebra of functions $`C^{\mathrm{}}()`$ is naturally isomorphic to the algebra (20) with $`m=n=1`$.
### 4.2. BRST charge
It turns out that all the ingredients of the Lagrange structure can be naturally interpreted as coefficients of expansion in the fiber coordinates of a single function $`\mathrm{\Omega }C^{\mathrm{}}(๐ฉ)`$, called the BRST charge, such that
(26)
$$\mathrm{gh}(\mathrm{\Omega })=1,ฯต(\mathrm{\Omega })=1,\mathrm{Deg}(\mathrm{\Omega })1.$$
The relations defining the Lagrange structure are generated by and are equivalent to the master equation
(27)
$$\{\mathrm{\Omega },\mathrm{\Omega }\}=0.$$
To get the desired interpretation let us first expand $`\mathrm{\Omega }`$ in the powers of momenta $`\overline{\phi }_I`$:
(28)
$$\mathrm{\Omega }=\underset{n=1}{\overset{\mathrm{}}{}}\mathrm{\Omega }_n,\mathrm{Deg}(\mathrm{\Omega }_n)=n.$$
Substituting this expansion into the master equation (27) and considering it in the first three orders in the momenta, we get the following relations for $`\mathrm{\Omega }_1,\mathrm{\Omega }_2`$:
(29)
$$\{\mathrm{\Omega }_1,\mathrm{\Omega }_1\}=0,\{\mathrm{\Omega }_1,\mathrm{\Omega }_2\}=0,\{\mathrm{\Omega }_2,\mathrm{\Omega }_2\}=2\{\mathrm{\Omega }_1,\mathrm{\Omega }_3\}.$$
The first term $`\mathrm{\Omega }_1=\mathrm{\Omega }^I(\phi )\overline{\phi }_I`$ gives rise to the odd, nilpotent vector field on $``$,
(30)
$$Q\mathrm{\Omega }^I(\phi )\frac{}{\phi ^I}=T_a\frac{}{\eta _a}+c^\alpha R_\alpha ^i\frac{}{x^i}+\eta _aZ_A^a\frac{}{\xi _A}+\mathrm{},$$
carrying all the information about the classical system $`(,T)`$ itself. Evaluating the nilpotency condition $`Q^2=0`$ at the lowest order in $`r`$-degree (see Table 1), one immediately recovers Rels.(8, 9) characterizing (3) as a set of gauge invariant and linearly dependent equations of motion, with $`R`$ and $`Z`$ being the generators of gauge transformations and Noether identities, respectively<sup>5</sup><sup>5</sup>5As it has been already mentioned, these generators, coinciding in the standard Lagrangian case, can be different in general, even by number..
Notice that the odd vector fields with zero square are known as homological. These are essentially equivalent to the notion of $`L_{\mathrm{}}`$-algebras or strong(ly) homotopy Lie algebras , . In physical literature, the homological vector fields usually appear as BRST-differentials associated either with the BV master action , or Hamiltonian BFV-BRST charge , .
The Poisson action of $`\mathrm{\Omega }_1`$ makes $`C^{\mathrm{}}(๐ฉ)`$ a cochain complex with the nilpotent differential
(31)
$$๐ปA=\{\mathrm{\Omega }_1,A\},AC^{\mathrm{}}(๐ฉ).$$
Let us denote by $`(๐ป)=_{n=0}^{\mathrm{}}^n(๐ป)`$ the corresponding cohomology group graded by $`m`$-degree.
The Lagrange anchor $`V:^{}TM`$, associated to the Lagrange structure for the classical system (30) is contained in the next term
(32)
$$\mathrm{\Omega }_2=\mathrm{\Omega }^{IJ}(\phi )\overline{\phi }_I\overline{\phi }_J=\overline{\eta }^aV_a^i\overline{x}_i+\mathrm{}.$$
Rels.(29) characterize $`\mathrm{\Omega }_2`$ as a weak anti-Poisson structure on $``$ (25), i.e. $`\mathrm{\Omega }_1`$-invariant (= $`๐ป`$-closed) odd bivector satisfying the Jacobi identity up to homotopy. The corresponding weak antibracket reads
(33)
$$(a,b)\frac{1}{2}\{\{\mathrm{\Omega }_2,a\},b\},a,bC^{\mathrm{}}().$$
Examining the Jacobi identity for these brackets one finds
(34)
$$\begin{array}{c}(a,(b,c))+(1)^{ฯต(b)ฯต(c)}((a,c),b)+(1)^{ฯต(a)(ฯต(b)+ฯต(c))}((b,c),a)=\\ S_3(๐ปa,b,c)(1)^{ฯต(a)ฯต(b)}S_3(a,๐ปb,c)(1)^{(ฯต(a)+ฯต(b))ฯต(c)}S_3(a,b,๐ปc)\\ ๐ปS_3(a,b,c).\end{array}$$
where we have introduced the following notation:
(35)
$$S_n(a_1,a_2,\mathrm{},a_n)\frac{1}{n!}\{\mathrm{}\{\mathrm{\Omega }_n,a_1\}a_2\},\mathrm{},a_n\},a_kC^{\mathrm{}}().$$
In particular,
(36)
$$\begin{array}{c}S_00,S_1(a)=๐ปa=\mathrm{\Omega }^I_Ia,S_2(a,b)=(a,b)=(1)^{ฯต(a)ฯต(I)}\mathrm{\Omega }^{IJ}_Ja_Ib,\\ S_3(a,b,c)=(1)^{ฯต(a)(ฯต(I)+ฯต(J))+ฯต(b)ฯต(I)}\mathrm{\Omega }_3^{IJK}_Ka_Jb_Ic.\end{array}$$
As is seen, the weak antibracket (33) induces the genuine antibracket on the cohomology group $`^0(๐ป)`$.
As was noticed in , Rel. (35) defines a flat $`S_{\mathrm{}}`$-structure on the supercommutative algebra $`C^{\mathrm{}}()`$: By definition, each $`S_n`$ is a graded-symmetric multi-differentiation of $`C^{\mathrm{}}()`$ and the generalized Jacobi identities for the collection of maps $`\{S_n\}`$ follow from the master equation (27) for the BRST charge $`\mathrm{\Omega }`$.
### 4.3. The unique existence of the BRST charge
In our treatment of the Lagrange structures the BRST charge $`\mathrm{\Omega }`$ is not given a priory \- it arises as a solution to the master equation (27) with prescribed โboundary conditionsโ. By the boundary conditions we mean the starting row of expansion of the BRST charge according to $`r`$-degree:
(37)
$$\mathrm{\Omega }=T_a\overline{\eta }^a+c^\alpha R_\alpha ^i\overline{x}_i+\eta _aZ_A^a\overline{\xi }^A+\overline{\eta }^aV_a^i\overline{x}_i+\mathrm{},$$
where the dots stand for the terms more than quadratic in the fiber coordinates and/or $`r`$-degree $`>1`$. The structure functions $`T`$โs, $`R`$โs, $`Z`$โs and $`V`$โs are naturally identified with the equations of motions, gauge symmetry generators, generators of Noether identities and the Lagrange anchor, respectively.
Given the boundary conditions (37), the existence of a solution to the master equations (27) can be proved by the standard tools of homological perturbation theory , . Let us sketch this proof. Consider the following expansion of the BRST charge according to the $`r`$-degree:
(38)
$$\mathrm{\Omega }=\underset{n=0}{\overset{\mathrm{}}{}}\mathrm{\Omega }^{(n)},\mathrm{deg}\left(\mathrm{\Omega }^{(n)}\right)=n.$$
In particular, the general expressions for the first two terms read
(39)
$$\mathrm{\Omega }^{(0)}=T_a\overline{\eta }^a,\mathrm{\Omega }^{(1)}=c^\alpha R_\alpha ^i\overline{x}_i+\eta _aZ_A^a\overline{\xi }^A+\overline{\eta }^aV_a^i\overline{x}_i+\overline{\eta }^a\eta _bU_{\alpha a}^bc^\alpha +\overline{\eta }^a\overline{\eta }^bW_{ab}^d\eta _d.$$
Substituting (38) into the master equation (27) gives a chain of equations of the form:
(40)
$$\delta \mathrm{\Omega }^{(n+1)}=K_n(\mathrm{\Omega }^{(0)},\mathrm{},\mathrm{\Omega }^{(n)}),n=0,1,2,..,$$
where
(41)
$$\delta =T_a\frac{}{\eta _a}+\eta _aZ_A^a\frac{}{\xi _A}+\overline{x}_iR_\alpha ^i\frac{}{\overline{c}_\alpha }+\overline{\eta }^a_iT_a\frac{}{\overline{x}_i}+\overline{\eta }^a\eta _bU_{\alpha a}^b\frac{}{\overline{c}_\alpha }$$
is a nilpotent operator decreasing the $`r`$-degree by one,
(42)
$$\delta ^2=0,\mathrm{deg}(\delta )=1,$$
and $`K_n`$ involves the brackets of the $`\mathrm{\Omega }`$โs of lower order. The nilpotency condition (42) is due to the following relations (cf. Eqs.(8,9)):
(43)
$$Z_A^aT_a=0,R_\alpha ^i_iT_a=U_{\alpha a}^bT_b.$$
Let $`(\delta )=_{n=0}^{\mathrm{}}_n(\delta )`$ denote the corresponding cohomology group graded by $`r`$-degree. It is not hard to see that the regularity conditions of Sect.3.2, which we assume satisfied for the classical system under consideration, provide acyclicity of $`\delta `$ in strictly positive $`r`$-degrees:
(44)
$$_n(\delta )=0\mathrm{for}n>0.$$
The proof of the last fact is quite standard (see e.g. ) and we leave it to the reader.
On the other hand, expanding the Jacobi identity $`\{\{\mathrm{\Omega },\mathrm{\Omega }\},\mathrm{\Omega }\}=0`$ in terms of $`r`$-degree, one can deduce that the r.h.s. of the $`(n+1)`$-th equation (40) is $`\delta `$-closed provided all the previous equations are satisfied. Therefore, the only equation to check to ensure solvability of (40) is
(45)
$$\delta \mathrm{\Omega }^{(1)}=K_0(\mathrm{\Omega }^{(0)}),$$
where $`K_0(\mathrm{\Omega }^{(0)})\{\mathrm{\Omega }^{(0)},\mathrm{\Omega }^{(0)}\}=0`$. Substituting the explicit expressions (39), (41) into (45), we reproduce the basic relations (43) as well as the integrability condition (7) with $`C_{ab}^d=W_{ab}^d+V_a^i\mathrm{\Gamma }_{ib}^dV_b^i\mathrm{\Gamma }_{ia}^d`$, where $`W_{ab}^d`$ is defined by (39). Thus, Eq.(45) generates all the defining relations for the regular Lagrange structure $`(,T,d_{})`$ of type $`(1,1)`$.
Resolving Eqs.(40) step-by-step, we are obviously free to add to $`n`$-th order solution $`\mathrm{\Omega }^{(n)}`$ any $`\delta `$-closed (and hence exact) term. This ambiguity is not essential, however, as it can always be absorbed by a canonical transformation of the Poisson supermanifold $`N`$. The proof of the last fact is quite standard (see e.g. ), and we omit it here.
### 4.4. Exactness of Lagrange structure
An important observation about the weak antibracket (33), encoding the Lagrange structure, is that $`\mathrm{\Omega }_2`$ determines a trivial $`๐ป`$-cocycle. In other words, for any $`\mathrm{\Omega }_2`$ there is a function
(46)
$$\begin{array}{c}G=G^{ij}(x)\overline{x}_i\overline{x}_j+\mathrm{},\\ \mathrm{Deg}(G)=2,ฯต(G)=0,\mathrm{gh}(G)=0,\end{array}$$
such that $`\mathrm{\Omega }_2=๐ปG`$. The last fact is a straightforward consequence of a more general statement about $`๐ป`$-cohomology.
###### Proposition 4.1.
Let $`(๐ป)=_n^m(๐ป)`$ be the group of $`๐ป`$-cohomology (31), where the numbers $`m`$ and $`n`$ refer to the $`m`$-degree and the ghost number respectively, then $`_n^m(๐ป)=0`$ for all $`m>n`$.
Proof: We start with the following decomposition:
(47)
$$๐ป=\delta +\mathrm{\Delta },$$
where $`\delta `$ is given by (41) and
$$\mathrm{deg}(\delta )=1,\mathrm{deg}(\mathrm{\Delta })0.$$
Thus, $`๐ป`$ is a deformation of $`\delta `$ by terms of higher $`r`$-degree and the statement follows immediately from the acyclicity of $`\delta `$ in positive $`r`$-degree. To make these arguments more explicit let us introduce the contracting homotopy for $`\delta `$, i.e. an operator $`\delta ^{}`$ obeying the property
(48)
$$(\delta \delta ^{}+\delta ^{}\delta )A=A,$$
for all $`A`$ with $`\mathrm{deg}(A)>0`$. Using this operator, one can show that $`๐ป`$-cohomology is localized at zero $`r`$-degree. Indeed, applying $`\delta ^{}`$ to both sides of $`๐ปA=0`$ yields
(49)
$$NA=๐ป\delta ^{}A,N1+(\delta ^{}\mathrm{\Delta }+\mathrm{\Delta }\delta ^{}).$$
Since $`\mathrm{deg}(\delta ^{}\mathrm{\Delta }+\mathrm{\Delta }\delta ^{})>0`$, the operator $`N`$ is invertible and commutes with $`๐ป`$. Thus $`A=๐ป(N^1\delta ^{}A)`$. To complete the proof it remains to note that
(50)
$$\mathrm{deg}(A)\mathrm{Deg}(A)\mathrm{gh}(A).$$
The last inequality can be verified just by comparing the lines of Table 1. $`\mathrm{}`$
Applying now the inequality (50) to $`\mathrm{\Omega }_2`$, we get $`\mathrm{deg}(\mathrm{\Omega }_2)1`$, and hence $`\mathrm{\Omega }_2=๐ปG`$. For a reason that will become clear later on, we call the function $`G`$, determined up to a $`๐ป`$-cocycle, a propagator associated to the weak anti-Poison structure $`\mathrm{\Omega }_2`$.
The proposition above allows us to give another prove of the existence theorem for the BRST charge $`\mathrm{\Omega }`$. Namely, consider the Hamiltonian flow $`\varphi _t`$ generated by the propagator $`G`$. Applying this flow to the symbol of homological vector field $`\mathrm{\Omega }_1`$, determined by Rels. (28, 30), we get a one-parameter family of formal functions $`\mathrm{\Omega }(t)C^{\mathrm{}}(๐ฉ)[[t]]`$ related to each other by a formal canonical transform. By definition,
(51)
$$\mathrm{\Omega }(t)=\varphi _t^{}\mathrm{\Omega }_1=\underset{n=0}{\overset{\mathrm{}}{}}\frac{t^n}{n!}\{G,\{G,\mathrm{},\{G,\mathrm{\Omega }_1\}\mathrm{}\}.$$
Clearly, the function $`\mathrm{\Omega }(1)`$ obeys the boundary condition (37) and solves the master equation (27) whenever $`\mathrm{\Omega }_1`$ does. The problem thus reduces to solving the nilpotency condition for the homological vector field (30). Expanding $`\mathrm{\Omega }_1=\mathrm{\Omega }_1^{(n)}`$ according to the resolution degree, we get a chain of equations
(52)
$$\delta \mathrm{\Omega }_1^{(n+1)}=K_n(\mathrm{\Omega }_1^{(0)},\mathrm{},\mathrm{\Omega }_1^{(n)}),$$
which are closely analogous to Eqs. (40), though with $`\delta `$ which is different from (41). Here
(53)
$$\delta =T_a\frac{}{\eta _a}+\eta _aZ_A^a\frac{}{\xi _A}$$
is the usual Koszul-Tate differential associated with the shell $`\mathrm{\Sigma }`$. The solvability of the system (52) can be easily seen from the acyclicity of $`\delta `$ in strictly positive $`r`$-degree and the invariance of $`\mathrm{\Sigma }`$ under the gauge transformations (9).
### 4.5. Physical observables
Upon the BRST imbedding, the physical observables of original theory are usually identified with certain BRST cohomology in ghost number zero. Let us show that the space of physical observables $`๐ซ`$, defined by (19), is naturally isomorphic to the subgroup $`_0^0(๐ป)(๐ป)`$ generated by the $`๐ป`$-cocycles of the ghost and momentum degree zero. Substituting the general expansion
(54)
$$F=\underset{n=0}{\overset{\mathrm{}}{}}F^{(n)}(\phi )=f(x)+\eta _aF_\alpha ^a(x)c^\alpha +\xi _AF_{\alpha \beta }^A(x)c^\alpha c^\beta +\mathrm{},\mathrm{deg}(F^{(n)})=n,$$
to the $`๐ป`$-closedness condition $`๐ปF=0`$ leads to the sequence of equations
(55)
$$\delta F^{(n+1)}=B_n(F^{(0)},\mathrm{},F^{(n)}),\mathrm{deg}(B_n)=n,$$
where the $`\delta `$ is given by (53). The first equation of this sequence reproduces the condition of on-shell invariance (17). Proceeding by induction in $`n`$ and using the identity $`๐ป^2F=0`$, one can see that the r.h.s. of the $`n`$-th equation (55) is $`\delta `$-closed, provided that Eq. (17) is satisfied. Since the differential $`\delta `$ is acyclic in positive $`r`$-degree, we conclude that (i) any invariant function $`fC(M)^{\mathrm{inv}}`$ is lifted to a $`๐ป`$-cocycle $`FC^{\mathrm{}}(๐ฉ)`$ with $`\mathrm{gh}(F)=\mathrm{Deg}(F)=0`$, and (ii) any two equivalent (in the sense of (18)) functions $`f_1,f_2C^{\mathrm{}}(M)^{\mathrm{inv}}`$ determine the same class of $`๐ป`$-cohomology upon the lift:
(56)
$$F_1F_2=๐ปK,K=K^a\eta _a+\mathrm{}.$$
This establishes an isomorphism between the space of physical observables $`๐ซ`$ of the original theory (19) and the BRST cohomology group $`_0^0(๐ป)`$.
### 4.6. Lagrange structure from the Hamiltonian viewpoint.
In the conventional BFV approach the BRST charge arises as a tool for quantizing first-class constrained Hamiltonian systems. Given a Poisson manifold $`P`$, the first class constraints $`\mathrm{\Theta }_IC^{\mathrm{}}(P)`$ are defined as an (overcomplete) basis in the regular, Poisson-closed ideal of functions vanishing on a coisotropic submanifold $`CP`$. In a more general setting , one can think of $`\{\mathrm{\Theta }_I\}`$ as a section of a (nontrivial) vector bundle $`EP`$, which intersects the base $`P`$ at points of $`C`$. The standard BFV-BRST theory , corresponds to the case of a trivial vector bundle $`E`$. According to the general prescriptions of BFV-BRST theory, to each first class constraint, one has to associate a pair of canonically conjugated ghost variables $`(๐,๐ซ)`$, extending thus the original Poisson manifold $`P`$ to the supermanifold $`\mathrm{\Pi }(EE^{})`$. In the reducible case, i.e when the constraints $`\mathrm{\Theta }_I`$ are functionally dependent, the additional pairs of canonically conjugated variables (ghosts-for-ghosts) must be introduced into the scheme , .
A glance at Table 1 is enough to see that the spectrum of ghost numbers corresponds to Hamiltonian first-class constrained system of a first-stage of reducibility. In order to make this interpretation explicit let us combine the local coordinates with ghost numbers $`1`$ and $`1`$ into the ghost coordinates $`๐^I=(\overline{\eta }^a,c^\alpha )`$ and ghost momenta $`๐ซ_I=(\eta _a,\overline{c}_\alpha )`$, respectively. In this notation the BRST charge (37) can be rewritten as
(57)
$$\mathrm{\Omega }=๐^I\mathrm{\Theta }_I(x,\overline{x})+๐ซ_I\mathrm{\Xi }_A^I(x,\overline{x})\xi ^A+\frac{1}{2}๐ซ_KU_{IJ}^K(x,\overline{x})๐^J๐^I+O(๐ซ^2,\xi ^2),$$
where the expansion coefficients $`\mathrm{\Theta }_I=(\stackrel{~}{T}_a,\stackrel{~}{R}_\alpha )`$ and $`\mathrm{\Xi }_A^I=(\stackrel{~}{Z}_A^a,0)`$, playing the role of first class constraints and their null-vectors, are given by the formal power series in $`\overline{x}`$โs of the form
(58)
$$\begin{array}{c}\stackrel{~}{T}_a(x,\overline{x})=T_a(x)+V_a^i(x)\overline{x}_i+O(\overline{x}^2),\hfill \\ \stackrel{~}{R}_\alpha (x,\overline{x})=R_\alpha ^i(x)\overline{x}_i+O(\overline{x}^2),\hfill \\ \stackrel{~}{Z}_A^a(x,\overline{x})=Z_A^a(x)+O(\overline{x}).\hfill \end{array}$$
At lowest orders in $`๐`$โs the master equation (27) gives the standard involution relations for a set of reducible first-class constraints
(59)
$$\{\mathrm{\Theta }_I,\mathrm{\Theta }_J\}=U_{IJ}^K\mathrm{\Theta }_K,\mathrm{\Xi }_A^I\mathrm{\Theta }_I=0,$$
w.r.t. the canonical Poisson bracket on $`T^{}M`$. From the regularity condition it readily follows that the number of the independent first class constraints $`\mathrm{\Theta }_I`$ is equal to $`dimM`$. In physical terms, one can interpret this fact concluding that the considered Hamiltonian system has no (local) physical degrees of freedom. From the geometrical viewpoint this implies that equations $`\mathrm{\Theta }_I=0`$ define a Lagrangian submanifold $`LT^{}M`$; more accurately, $`L`$ is a formal Lagrangian submanifold as we are not concerned with convergence of the formal series (58).
The first class constraints $`\mathrm{\Theta }_I`$ can also be regarded as a formal deformation of those given by the leading terms of expansions (58) in the โdirectionโ of the Lagrange anchor $`V`$. The involution relations for the โinitialโ constraints $`T_a(x)`$ and $`R_\alpha ^i(x)\overline{x}_i`$ readily follow from nilpotency of the homological vector field (30). The integrability condition (7) results then from the requirement that the deformed constraints (58) have to be the first class as well. From this standpoint, the Lagrange structure can be understood as an infinitesimal of deformation of the Lagrangian submanifold $`L_0T^{}M`$ defined by the equations $`T_a(x)=0`$ and $`R_\alpha ^i(x)\overline{x}_i=0`$. As is shown in the next section, any nonzero Lagrange anchor gives rise to quantum fluctuations of physical observables. In other words, any โclassicalโ deformation of $`L_0T^{}M`$ (in the category of Lagrangian submanifolds in $`T^{}M`$) results in a quantum deformation upon path-integral quantization.
Actually, the results of Sect. 4.4 allows one to interpret the series (58) as Tailorโs expansion in $`\overline{x}`$โs of some smooth functions $`\mathrm{\Theta }_I(x,\overline{x})`$, $`\mathrm{\Xi }_A^I(x,\overline{x})`$ defined in a sufficiently small vicinity $`U`$ of any given point $`pT^{}M`$. This can be proved as follows. Since both the symbol $`\mathrm{\Omega }_1`$ of homological vector field (30) and the generator $`G`$ of canonical transform (51) are smooth function on $`๐ฉ`$ (with polynomial dependence of fiber variables), one can assert that for any $`p๐ฉ`$ there exists a neighborhood $`W`$ together with $`\epsilon >0`$ such that $`\mathrm{\Omega }(t)=\varphi _t\mathrm{\Omega }_1`$ is a smooth function on $`W`$ for all $`t[0,\epsilon )`$. Moreover, shrinking the vicinity $`W๐ฉ`$ along the fibers, one can always choose $`\epsilon >1`$, so that
$$\mathrm{\Theta }_I(x,\overline{x})\frac{\mathrm{\Omega }(1)}{๐^I}|_{๐=\xi =0},\mathrm{\Xi }_A^I(x,\overline{x})\frac{^2\mathrm{\Omega }(1)}{\xi ^A๐ซ_I}|_{๐=\xi =0}$$
are smooth functions on an open subset $`U=WT^{}M`$.
Let $`x_0\mathrm{\Sigma }M`$ be a classical solution and let $`U`$ be a sufficiently small neighbourhood of $`x_0`$ in $`T^{}M`$ for which the equations $`\mathrm{\Theta }_I(x,\overline{x})=0`$ make sense, i.e. determine a Lagrangian submanifold $`LU`$. If $`\mathrm{rank}(V_a^i(x),R_\alpha ^i(x))=m`$ for all $`xUM`$, then one can split the position coordinates and momenta onto two groups, $`x^i=(y^I,z^J)`$ and $`\overline{x}_i=(\overline{y}_I,\overline{z}_J)`$, such that the index $`I`$ runs $`m`$ values, and the Lagrangian submanifold $`LU`$ is determined by the equations
(60)
$$\overline{y}_I=\frac{\mathrm{\Psi }(y,\overline{z})}{y^I},z^J=\frac{\mathrm{\Psi }(y,\overline{z})}{\overline{z}_J},$$
$`\mathrm{\Psi }`$ being the generating function for $`L`$ of the first kind . Indeed, due to the rank condition the equations $`\mathrm{\Theta }_I(x,\overline{x})=0`$ can be explicitly resolved w.r.t. $`m`$ of $`dimM`$ momenta $`\overline{x}_i`$. Then using the rest of the equations one can express $`dimMm`$ position coordinates $`x`$โs in terms of the other variables (as the total number of independent equations equals $`dimM`$.) Since the resolved constraints are always in the abelian involution, one finds immediately that the r.h.s. of these constraints are to be given by the gradient of some function $`\mathrm{\Psi }(y,\overline{z})`$.
Now we can use the local representation (60) for $`L`$ to prove the announced Proposition 2.1. Clearly, the classical equations of motion $`T_a(x)=0`$ are equivalent to $`\mathrm{\Theta }_I(x,\overline{x})=0`$ and $`\overline{x}_i=0`$. (Geometrically speaking, the shell $`\mathrm{\Sigma }`$ is given by the intersection $`LM`$.) Setting in Eqs. (60), $`\overline{x}_i=0`$ yields the following local representation for the shell:
(61)
$$\frac{S(y)}{y^I}=0,z^J=E^J(y),$$
where
(62)
$$S(y)\mathrm{\Psi }(y,0),E^J(y)\frac{\mathrm{\Psi }}{\overline{z}_J}(y,0).$$
A simple linear algebra shows that the number $`k=dimMm`$ of non-Lagrangian equations of motion is defined by the formula of Proposition 2.1.
## 5. Quantization
The usual path-integral quantization deals with computing of quantum averages for the physical observables (= function(al)s on the space of trajectories $`M`$). The quantum average of an observable $`F`$ is given by its integral over $`M`$ with the uniform weight $`e^{\frac{i}{\mathrm{}}S}`$, $`S`$ being the action functional of the system. In the gauge invariant Lagrangian theory, this simple rule is replaced by a more sophisticated BV scheme , realizing the same idea in the presence of gauge invariance. If the original classical system admitted consistent operator BFV-BRST quantization, the BV method can be deduced from the Hamiltonian BFV-BRST scheme. In general, the relationship remains obscure between the BV quantization and the deformation quantization of the (weak) Poisson manifolds, so the BV method is presently viewed as an ad hoc postulate for the path-integral quantization. Obviously, the BV scheme can not be directly applied to quantize the systems having no action functional. In this section, we extend the path-integral quantization method to include not necessarily Lagrangian (gauge) systems.
Let us briefly outline the quantization algorithm we suggest. The starting point is the BRST embedding for the Lagrange structure presented in the previous section. Upon this embedding, the physical observables of the original theory are identified with the BRST cohomology group $`_0^0(๐ป)`$. As the next step, making use of the AKSZ method in the form of Ref. , we construct a topological sigma-model related to this BRST complex. Then we prove that the dynamics of this topological sigma-model are equivalent to the original classical theory; in so doing, the space of physical observables $`_0^0(๐ป)`$ is naturally identified with the boundary observables of the topological sigma-model. The topological sigma-model, being a Lagrangian theory in usual sense, can be quantized by the standard BV prescription, that results in quantizing the original theory which is not necessarily Lagrangian. In a particular case of Lagrangian systems, the sigma-model path-integral can be explicitly localized at the boundary, where it precisely reproduces the BV answer for the original Lagrangian theory.
### 5.1. Topological sigma-model
Consider the $`(1,1)`$-dimensional supermanifold $`=\mathrm{\Pi }TI`$ with boundary associated to the odd tangent bundle of the closed interval $`I=[0,1]`$. The โpointsโ of $``$ are parameterized by one even coordinate $`t[0,1]`$ of ghost number $`0`$ and one odd coordinate $`\theta \mathrm{\Pi }T_tI`$ of ghost number $`1`$. We shall also use the collective notation $`z=(t,\theta )`$. By the boundary of $``$ we mean the two-point set $`=\{z_0,z_1\}`$ constituted by the โend pointsโ $`z_0=(0,0)`$ and $`z_1=(1,0)`$ of the superinterval $``$. The canonical volume element on $``$ is given by $`d^2z=dtd\theta `$.
Consider now the superspace $`๐ฉ^{}`$ of all smooth maps from the source supermanifold $``$ to the target supermanifold $`๐ฉ`$. (The latter is defined by (21).) In terms of local coordinates $`\varphi ^k=(\phi ^I,\overline{\phi }_J)`$ on $`๐ฉ`$ each element $`(\varphi :๐ฉ)๐ฉ^{}`$ defines (and is defined by) a field configuration
(63)
$$\varphi ^k(z)=\varphi ^k(t)+\theta \stackrel{}{\varphi }{}_{}{}^{k}(t).$$
According to the definitions above
(64)
$$ฯต(\varphi ^k)=ฯต(\stackrel{}{\varphi }{}_{}{}^{k})+1,\mathrm{gh}(\varphi ^k(t))=\mathrm{gh}(\stackrel{}{\varphi }{}_{}{}^{k}(t))+1=\mathrm{gh}(\varphi ^k).$$
The action of the topological sigma-model reads
(65)
$$๐ฎ[\varphi ]=_{}d^2z(\mathrm{\Lambda }_k(\varphi )\mathrm{D}\varphi ^k\mathrm{\Omega }(\varphi )).$$
The first and second terms are given here by the pull-backs of the symplectic potential (23) and the BRST charge (37) respectively, and
(66)
$$\mathrm{D}=\theta \frac{}{t}$$
is an odd, nilpotent vector field on $``$ of ghost number 1. Taking into account the Grassman parity of all the factors entering the integrand (65) one can see that $`\mathrm{gh}(๐ฎ)=0`$. The action (65) admits a straightforward BV interpretation that will be given in the next Sect. 5.2.
The Poisson bracket (24) on $`๐ฉ`$ induces the antibracket (i.e. the odd Poisson bracket) on the superspace $`๐ฉ^{}`$. By definition,
(67)
$$(F,G)=_{}d^2z\left(\frac{\delta _rF}{\delta \varphi ^k(z)}\omega ^{km}(\varphi (z))\frac{\delta _lG}{\delta \varphi ^m(z)}\right),$$
for any functionals of fields $`F[\varphi ]`$ and $`G[\varphi ]`$.
The model (65) is called topological since the action $`๐ฎ`$ is required to satisfy the classical master equation
(68)
$$(๐ฎ,๐ฎ)=0.$$
An explicit calculation yields
(69)
$$\frac{1}{2}(๐ฎ,๐ฎ)=_{}d^2z\left(\mathrm{D}\mathrm{\Omega }+\frac{1}{2}\{\mathrm{\Omega },\mathrm{\Omega }\}(\varphi (z))\right)=\mathrm{\Omega }(z_1)\mathrm{\Omega }(z_0).$$
To meet the master equation, we impose the following boundary conditions on the momenta:
(70)
$$\overline{\phi }^J|_{}=0.$$
Then $`\mathrm{\Omega }(z_0)=\mathrm{\Omega }(z_1)=0`$, as $`\mathrm{Deg}(\mathrm{\Omega })1`$ due to the definition (26).
As usual, the classical observables of the topological sigma-model (65) are identified with zero-ghost-number cohomology of the BRST differential $`(๐ฎ,)`$: The functional $`F`$ defines a classical observable iff
(71)
$$\mathrm{gh}(F)=0\mathrm{and}(๐ฎ,F)=0,$$
and two BRST closed functionals $`F`$ and $`G`$ are considered to define the same classical observables ($`FG`$) if they belong to the same class of BRST cohomology, i.e. $`FG=(๐ฎ,H)`$ for some $`H`$.
Of particular interest are the boundary observables. These are constructed from the physical observables $`[F]_0^0(๐ป)=๐ซ`$ of the original gauge theory (see Sect. 4.5) by the rule $`\widehat{F}[\varphi ]=F(\phi (z_1))`$. A simple computation shows that
(72)
$$(๐ฎ,\widehat{F})=\{\mathrm{\Omega },F\}(\phi (z_1))=(๐ปF)(\phi (z_1))=0.$$
If $`F=๐ปG`$, then $`\widehat{F}=(๐ฎ,\widehat{G})`$, and hence $`\widehat{F}0`$. Clearly, replacing the point $`z_1`$ with $`z_0`$ one gets another set of classical observables supported on the other end of the superinterval $``$.
The quantum average of a classical observable $`[F]_0^0(๐ป)`$ corresponding to the original (not necessarily Lagrangian) gauge theory $`(,T,d_{})`$ is defined by the path integral
(73)
$$F=_๐ฉ^{}๐\varphi F(\phi (z_1))e^{\frac{i}{\mathrm{}}๐ฎ[\varphi ]},$$
where the integration measure is normalized in such a way that $`1=1`$. To evaluate this path integral one has to impose a gauge fixing condition and choose an appropriate integration measure. As the sigma-model action entering (73) is a proper solution to the master equation, the gauge fixing procedure is standard for the BV method , i.e. this means to fix any Lagrange surface in the anti-Poisson manifold. The regularization of possible divergencies in (73) does not have any specificity compared to any other sigma-model path integral.
The main result of the section is Relation (73) defining quantum average for a physical observable of any (i.e. not necessarily Lagrangian) dynamical system in terms of the usual path integral for the topological sigma-model. Below we elaborate on the equivalence between the original system and the topological sigma-model at the level of classical dynamics.
### 5.2. Classical equivalence
It was shown in Refs. , that the action of any topological sigma-model on $`(1,1)`$-dimensional supermanifold can be interpreted as the BV master action of a constrained Hamiltonian system. In the case under consideration such a Hamiltonian system has been already constructed, in fact, in Sect. 4.6. To recover this effective Hamiltonian constrained system from the action (65) one should integrate out of $`\theta `$ in (65) and set to zero all the fields with nonzero ghost number. The result will have the form
(74)
$$S_0[x,\overline{x},\lambda ]=_I(\overline{x}_idx^i\lambda ^I\mathrm{\Theta }_I(x,\overline{x})).$$
where we introduced the notation $`\lambda ^I(\stackrel{}{\overline{\eta }}{}_{}{}^{a},\stackrel{}{c}{}_{}{}^{\alpha })`$. Clearly, $`S_0`$ is nothing but the Hamiltonian action functional on the cotangent bundle $`T^{}M`$ with the total Hamiltonian given by the linear combination of the (reducible) first class constraints (58). Upon this identification the 1-forms $`\lambda ^I`$ on $`I`$ play the role of Lagrange multipliers to $`\mathrm{\Theta }`$โs. The action $`S_0`$ is invariant under the standard gauge transformation
(75)
$$\begin{array}{c}\delta _\epsilon x^i=\{x^i,\mathrm{\Theta }_I\}\epsilon ^I,\delta _\epsilon \overline{x}_i=\{\overline{x}_i,\mathrm{\Theta }_I\}\epsilon ^I,\\ \delta _\epsilon \lambda ^I=d\epsilon ^I\lambda ^KU_{KJ}^I\epsilon ^J+\mathrm{\Xi }_A^I\epsilon ^A,\end{array}$$
where $`\epsilon ^I=(\epsilon ^a,\epsilon ^\alpha )`$ and $`\epsilon ^A`$ are the infinitesimal gauge parameters, and the structure functions $`U_{KJ}^I`$, $`\mathrm{\Xi }_A^I`$ are given by (59). The compatibility between the gauge transformations (89) and the boundary conditions $`\overline{x}_i(0)=\overline{x}_i(1)=0`$ implies that $`\epsilon ^I(0)=\epsilon ^I(1)=0`$. The linear dependence of constraints $`\mathrm{\Theta }_I`$ leads to the linear dependence of the gauge algebra generators (75). Indeed, substituting
(76)
$$\epsilon ^I=\mathrm{\Xi }_A^I\varrho ^A,\epsilon ^A=d\varrho ^A\lambda ^IW_{IB}^A\varrho ^B$$
turns (75) to the trivial (on-shell vanishing) gauge transformation:
(77)
$$\begin{array}{c}\delta _\varrho x^i=\frac{\delta S_0}{\delta \lambda ^I}\{x^i,\mathrm{\Xi }_A^I\}\varrho ^A,\delta _\varrho \overline{x}_i=\frac{\delta S_0}{\delta \lambda ^I}\{\overline{x}_i,\mathrm{\Xi }_A^I\}\varrho ^A,\\ \delta _\varrho \lambda ^I=\frac{\delta S_0}{\delta x^i}\{x^i,\mathrm{\Xi }_A^I\}\varrho ^A\frac{\delta S_0}{\delta \overline{x}_i}\{\overline{x}_i,\mathrm{\Xi }_A^I\}\varrho ^A.\end{array}$$
Here we used the definition of the structure function $`W`$
(78)
$$\{\mathrm{\Xi }_B^I,\mathrm{\Theta }_J\}\mathrm{\Theta }_B^KU_{KJ}^I=W_{JB}^A\mathrm{\Xi }_A^I$$
following from the identity $`\{\mathrm{\Xi }_A^I\mathrm{\Theta }_I,\mathrm{\Theta }_J\}=0`$.
Now the BV interpretation of the sigma-model action functional $`๐ฎ=S_0+\mathrm{}`$ becomes obvious: it is just the BV master action corresponding to the theory with reducible first class constraints and the action (74). Upon this interpretation, the fields $`(\overline{\eta }^a,\stackrel{}{\overline{\xi }}{}_{}{}^{A},c^\alpha )`$ are identified with the ghosts corresponding to the infinitesimal gauge parameters $`(\epsilon ^a,\epsilon ^A,\epsilon ^\alpha )`$, $`\overline{\xi }^A`$ are ghosts for ghosts, and the other component fields are antifields to the aforementioned ones including the original gauge fields $`(\stackrel{}{\overline{\eta }}{}_{}{}^{a},\stackrel{}{c}{}_{}{}^{\alpha },x^i,\overline{x}_i)`$.
From the Hamiltonian viewpoint, the model (74) has no (local) degrees of freedom as the first class constraints $`\mathrm{\Theta }_I=0`$ define a Lagrangian submanifold $`LT^{}M`$. This amounts to saying that, given a โtimeโ<sup>6</sup><sup>6</sup>6To avoid confusion, let us recall that it is the โtimeโ $`t`$ which is an auxiliary dimension introduced when the original dynamics (governed by the equations of motion (3)) is embedded into the topological sigma-model dynamics. The original equations of motion (3) (that are defined on the boundary, from the viewpoint of this sigma-model) contain their own evolution parameter. moment $`t_0(0,1)`$, by an appropriate gauge transformation (75) one can always move any point $`(x^i(t_0),\overline{x}_j(t_0))L`$ to any other point of $`L`$ assigning, simultaneously, any given value to $`\lambda (t_0)`$, no matter what were the boundary/initial values of these variables at $`t=0`$ or $`t=1`$. Whereas at the end points of the โtimeโ interval we have the boundary conditions $`\overline{x}_i(0)=\overline{x}_i(1)=0`$ reducing the constraints
$$0=\frac{\delta S_0}{\delta \lambda ^I}=\mathrm{\Theta }_I(x,\overline{x})$$
(the only dynamical equations one has actually to solve) to the original equations of motion $`T_a(x(0))=T_a(x(1))=0`$. So, we can conclude that the dynamical content of the topological sigma-model (65) is equivalent to that of the original (non-Lagrangian) gauge theory.
## 6. Examples
### 6.1. BV field-antifield formalism
In this section we quantize the standard Lagrangian gauge system by the proposed general method that works for not necessarily Lagrangian theories. In this case, the standard BV quantization will be shown to follow from the proposed quantization scheme.
In the BV formalism the classical gauge theory is completely specified by master action $`S(\varphi )`$ defined on an antisymplectic manifold $``$ of fields and antifields $`\varphi ^i`$ and subject to the classical master equation
(79)
$$(S,S)_{}\frac{_rS}{\varphi ^I}\sigma ^{IJ}\frac{_lS}{\varphi ^J}=0,$$
$`\sigma ^{IJ}`$ being the odd bivector dual to the antisymplectic structure $`\sigma =d\varphi ^I\sigma _{IJ}d\varphi ^J`$. Besides the Grassman parity, the supermanifold $``$ is graded by the ghost number and it is additionally required that
(80)
$$\mathrm{gh}(S)=1,\mathrm{gh}((F,G))=\mathrm{gh}(F)+\mathrm{gh}(G)1,$$
for any homogeneous $`F,GC^{\mathrm{}}()`$.
We start with the following simple observation which might be of some interest in its own right: Any Lagrangian gauge theory on $``$ with the master action $`S(\varphi )`$ admits an equivalent reformulation as the topological sigma-model having $``$ as target manifold. The construction is as follows. Let $`๐ฅ=\times \mathrm{\Pi }`$ be the $`(1,2)`$-dimensional supermanifold with boundary, given by the direct product of the superinterval $`=\mathrm{\Pi }TI`$ with coordinates $`z=(t,\theta )`$ and the odd linear space $`\mathrm{\Pi }`$ โparameterizedโ by $`\overline{\theta }`$. The ghost number assignments are given by
(81)
$$\mathrm{gh}(t)=0,\mathrm{gh}(\theta )=1,\mathrm{gh}(\overline{\theta })=1.$$
The โpointsโ of $`๐ฅ`$ are thus the triples $`u=(t,\theta ,\overline{\theta })`$, where $`t[0,1]`$. The boundary of $`๐ฅ`$ is, by definition, a two point set $`๐ฅ=\{u_0,u_1\}`$ constituted by $`u_0=(0,0,0)`$ and $`u_1=(1,0,0)`$.
The field content of the topological sigma-model in question is identified with the superspace $`^๐ฅ`$ of maps from $`๐ฅ`$ to $``$. In terms of local coordinates $`\varphi ^I`$ on $``$, each $`\varphi ^๐ฅ`$ is described by a field configuration
(82)
$$\varphi ^I(u)=\phi ^I(z)+\overline{\theta }\overline{\phi }^I(z)=\phi ^I(t)+\theta \stackrel{}{\phi }{}_{}{}^{I}(t)+\overline{\theta }\overline{\phi }^I(t)+\overline{\theta }\theta \stackrel{}{\overline{\phi }}{}_{}{}^{I}(t),$$
Observe that under the general coordinate transformations on $``$, $`\varphi ^I\stackrel{~}{\varphi }^I(\varphi )`$, the component superfields $`\{\overline{\phi }^I(z)\}`$ behave like the coordinates of a tangent vector to $``$. So, one may think of the superfields $`\varphi ^I(u)=(\phi ^I(z),\overline{\phi }^I(z))`$ as smooth maps from the total space of the odd tangent bundle $`\mathrm{\Pi }TI`$ to that of $`\mathrm{\Pi }T`$.
Let us suppose for a moment that the antisymplectic 2-form is exact i.e. $`\sigma =d\rho `$ as it happens in the conventional BV theory. Then we can define the following action of the topological sigma-model
(83)
$$๐ฎ=_๐ฅd^3u(\rho _I(\varphi )\overline{\mathrm{D}}\varphi ^IS(\varphi )).$$
Here $`d^3u=dtd\theta d\overline{\theta }`$ is a natural integration measure on $`๐ฅ`$, and
(84)
$$\overline{\mathrm{D}}=\frac{}{\overline{\theta }}+\mathrm{D},\mathrm{D}=\theta \frac{}{t}$$
are odd, self-commuting vector fields on $`๐ฅ`$ of ghost number 1. Notice that $`\mathrm{gh}(๐ฎ)=0`$. As will be shown below, action (83) is equivalent, on the one hand, to the general topological sigma-model action (65) constructed for the original Lagrangian equations of motion with the anchor being the unit matrix, and on the other hand to the BV master action for the original Lagrangian theory.
The antibracket on $``$ gives rise to that on $`^๐ฅ`$:
(85)
$$(\varphi ^I(u),\varphi ^J(u^{}))_^๐ฅ=\sigma ^{IJ}(\varphi (u))\delta ^3(uu^{}).$$
Taking the antibracket of the action (83) with itself, we find
(86)
$$(๐ฎ,๐ฎ)_^๐ฅ=d^3u\left(\frac{\delta _r๐ฎ}{\delta \varphi ^I(u)}\sigma ^{IJ}(\varphi (u))\frac{\delta _l๐ฎ}{\delta \varphi ^J(u)}\right)=\frac{\varphi ^I}{\overline{\theta }}\left(\sigma _{IJ}\frac{\varphi ^J}{\overline{\theta }}_IS\right)|_{u_0}^{u_1}.$$
To cancel the boundary terms in the r.h.s. of the last expression we can set
(87)
$$\frac{\varphi ^I}{\overline{\theta }}|_๐ฅ=0\overline{\phi }^I(0)=\overline{\phi }^I(1)=0.$$
With these boundary conditions the functional $`๐ฎ`$ becomes the classical master action on the antisymplectic supermanifold $`^๐ฅ`$. Integrating out of $`\overline{\theta }`$ we get
(88)
$$๐ฎ=_{}d^2z\left(\overline{\phi }^I\sigma _{IJ}\mathrm{D}\phi ^I+\frac{1}{2}\overline{\phi }^I\sigma _{IJ}\overline{\phi }^J+\overline{\phi }^I\frac{S}{\phi ^I}\right).$$
As is seen, the functional $`๐ฎ`$ depends actually on the antisymplectic structure $`\sigma `$, but not on the choice of antisymplectic potential $`\mathrm{\Lambda }`$.
In consequence of the master equation the action (88) is invariant under the abelian gauge transformations
(89)
$$\begin{array}{c}\delta _\epsilon \phi ^I=\epsilon ^I,\hfill \\ \delta _\epsilon \overline{\phi }^J=\mathrm{D}\epsilon ^J+\epsilon ^L\frac{^2S}{\phi ^L\phi ^I}\sigma ^{IJ}+\epsilon ^L\frac{\sigma _{IK}}{\phi ^L}(\overline{\phi }^K+\mathrm{D}\phi ^K)\sigma ^{IJ},\hfill \end{array}$$
with the infinitesimal gauge parameter $`\epsilon ^I(z)`$ subject to the boundary conditions
(90)
$$\epsilon ^I(z)_IS(\phi (z))|_{}=0.$$
Since $`S`$ is a proper solution to the classical master equation (79), only โhalfโ of the gauge parameters $`\epsilon ^i(z)`$ need vanish at the boundary points $`z_0`$ and $`z_1`$.
Notice that the action (88) is identical in form to the action (65). Making identifications
(91)
$$\mathrm{\Lambda }=\overline{\phi }^I\sigma _{IJ}d\phi ^J,\mathrm{\Omega }=\mathrm{\Omega }_1+\mathrm{\Omega }_2=\overline{\phi }^I\frac{S}{\phi ^I}+\frac{1}{2}\overline{\phi }^I\sigma _{IJ}\overline{\phi }^J,$$
one can readily check that the BRST charge $`\mathrm{\Omega }`$ obeys the master equation $`\{\mathrm{\Omega },\mathrm{\Omega }\}=0`$ with the Poisson bracket corresponding to the exact symplectic structure $`\omega =d\mathrm{\Lambda }`$. Moreover, when $`S`$ is the minimal sector master action of an irreducible gauge theory, the spectrum of component fields $`(\phi ^I,\overline{\phi }^J)`$ coincides with that presented in Table 1. We leave the check of details to the reader.
Notice that the โtruncatedโ BRST charge $`\mathrm{\Omega }_1=\overline{\phi }^IS/\phi ^I`$ was first considered in Ref.. In that paper, a very convenient superfield technique was developed to illuminate the relationship between the BFV-BRST charge and the BV master action. In this Section we use a similar technique, although the same results can be derived without superfields.
According to the results of Sect. 4.2, the term $`\mathrm{\Omega }_1`$ defines a homological vector field on $``$:
(92)
$$\begin{array}{c}QF=\{\mathrm{\Omega }_1,F\}=\frac{_rS}{\phi ^I}\sigma ^{IJ}\frac{_lF}{\phi ^J},FC^{\mathrm{}}(),\\ Q^2=0\{\mathrm{\Omega }_1,\mathrm{\Omega }_1\}=0.\end{array}$$
In the case at hand the operator $`Q`$ is nothing but the usual BV-differential associated with the classical master action $`S`$. The second term $`\mathrm{\Omega }_2`$ defines (and is defined by) the original anti-Poisson structure on $``$:
(93)
$$\begin{array}{c}(F,G)=\{\{\mathrm{\Omega }_2,F\},G\},F,G,HC^{\mathrm{}}(),\\ (1)^{ฯต(F)ฯต(H)}(F,(G,H))+cycle(F,G,H)=0\{\mathrm{\Omega }_2,\mathrm{\Omega }_2\}=0.\end{array}$$
Both structures are compatible in the sense of the graded Liebnitz rule
(94)
$$Q(F,G)=(QF,G)+(1)^{ฯต(F)+1}(F,QG)\{\mathrm{\Omega }_1,\mathrm{\Omega }_2\}=0.$$
As we have shown in Sect.4.4, the last relation implies the existence of a function $`G=G^{IJ}\overline{\phi }_I\overline{\phi }_J`$, called the propagator, such that
(95)
$$\mathrm{\Omega }_2=\{\mathrm{\Omega }_1,G\},ฯต(G)=0,\mathrm{gh}(G)=0.$$
Here we set $`\overline{\phi }_I\sigma _{IJ}\overline{\phi }^J`$, so that $`ฯต(\overline{\phi }_I)=ฯต(\phi ^I)`$ and $`\mathrm{gh}(\overline{\phi }_I)=\mathrm{gh}(\phi ^I)`$. When $`S(x)`$ is the action of a system without gauge symmetry, the only nonzero block of the matrix $`G^{IJ}`$ is given by the inverse to the van Vleck matrix $`_i_jS(x)`$ that justifies the term โpropagatorโ.
To establish a classical correspondence between the topological sigma-model (83) and the original gauge theory we first observe that according to (89) the fields $`\phi ^I`$ are purely gauge ones in the interior of $``$, while the fields $`\overline{\phi }^I`$ enter to the action functional $`๐ฎ`$ only in an algebraic way (i.e. without derivatives). This suggests that all the dynamical degrees of freedom are supported at the boundary of the superinterval $``$. Notice also that the form of the action (88) is quite similar to that of Hamiltonian action functional with $`\overline{\phi }^I`$ playing the role of momenta conjugated to the coordinates $`\phi ^I`$. So, to obtain the action functional governing the dynamics of the boundary degrees of freedom we can just eliminate the auxiliary fields $`\overline{\phi }^I`$ from $`๐ฎ`$ by means of their own equations of motion:
(96)
$$\frac{\delta ๐ฎ}{\delta \overline{\phi }^I}=0\overline{\phi }^I=\mathrm{D}\phi ^I+\sigma ^{IJ}\frac{S}{\phi ^J}.$$
As would be expected, the boundary field configurations $`\phi ^I(z_{0,1})`$ define the stationary points of $`S`$, since $`\overline{\phi }|_{}=\mathrm{D}\phi |_{}=0`$. Substituting (96) to (88), we finally get
(97)
$$๐ฎ|_{\delta ๐ฎ/\delta \overline{\phi }=0}=_{}d^2z\mathrm{D}S=S(\phi (z_1))S(\phi (z_0)).$$
The action describes two uncoupled copies of the original gauge theory (one for each end of the superinterval $``$) with $`\times `$ being the total configuration space.
Let us now comment on quantum equivalence. Proceeding to quantization, one assigns the BV configuration space $``$ with a nondegenerate density $`\rho `$ and replaces the classical master equation (79) with the quantum one:
(98)
$$(S,S)_{}=2i\mathrm{}\mathrm{\Delta }_{}S.$$
Here $`\mathrm{\Delta }_{}:C^{\mathrm{}}()C^{\mathrm{}}()`$ is the odd Laplace operator defined by the rule
(99)
$$\mathrm{\Delta }_{}F=\mathrm{div}_\rho X_F,$$
$`X_F=(F,)_{}`$ being the Hamiltonian vector field corresponding to $`FC^{\mathrm{}}()`$. The density $`\rho `$ is chosen in such a way that $`\mathrm{\Delta }_{}^2=0`$. By definition, a quantum observable is a function of fields $`F(\varphi )`$ annihilated by the quantum BRST operator $`\widehat{S}_{\mathrm{}}`$:
(100)
$$\widehat{S}_{\mathrm{}}F=(S,F)_{}i\mathrm{}\mathrm{\Delta }_{}F=0.$$
The quantum average of $`F`$ is defined by the path integral
(101)
$$F_S=_{}๐\varphi \delta (\gamma _a(\varphi ))F(\varphi )e^{\frac{i}{\mathrm{}}S(\varphi )}.$$
where $`๐\varphi `$ is the integration measure associated to $`\rho `$ and equations
(102)
$$\gamma _a(\varphi )=0$$
define an appropriate Lagrange surface in $``$. (In the conventional BV scheme the constraints (102) are required to be in abelian involution, i.e. $`(\gamma _a,\gamma _b)_{}=0`$, though, upon some modifications , a more general involution is also allowed.)
As with the antibracket, the measure density $`\rho `$ on $``$ induces that on the space of fields $`^I`$:
(103)
$$\stackrel{~}{\rho }=\underset{u๐ฅ}{}\rho (\varphi (u)).$$
Then the functional counterpart of the odd Laplacian (99) reads
(104)
$$\mathrm{\Delta }_{^{}}=_๐ฅd^3u\stackrel{~}{\rho }^1\frac{\delta }{\delta \varphi ^I(u)}\stackrel{~}{\rho }\sigma ^{IJ}(\varphi (u))\frac{\delta }{\delta \varphi ^J(u)}.$$
Given the quantum master action $`S`$, we define the action of the topological sigma-model by the same formula (83). The latter is proved to be a solution to the quantum master equation on $`^{}`$ with renormalized Plank constant $`\mathrm{}^{}`$. Indeed, a straightforward computation yields
(105)
$$(๐ฎ,๐ฎ)_{^{}}2i\mathrm{}^{}\mathrm{\Delta }_{^{}}๐ฎ=_{}d^3u\left((S,S)_{}2iC\mathrm{}^{}\mathrm{\Delta }_{}S\right)(\varphi (u)),$$
where $`C=\delta ^3(0)`$ is indefinite โconstantโ. The functional $`๐ฎ`$ will satisfy the quantum master equation if we set $`\mathrm{}^{}=\mathrm{}C^1`$. Notice that formally $`ฯต(C)=0`$ and $`\mathrm{gh}(C)=0`$. To assign a precise meaning for the value $`\delta ^3(0)`$ one has to apply a suitable regularization to the ill-defined Laplace operator (104). Similarly, after the renormalization above any quantum observable $`F(\varphi )`$ of the original gauge theory gives rise to the boundary quantum observable $`\widehat{F}[\varphi ]=F(\varphi (u_1))`$ of the topological sigma-model, i.e. $`\widehat{๐ฎ}_{\mathrm{}^{}}F(\varphi (u_1))=0`$.
To calculate the quantum average of a boundary observable $`\widehat{F}`$ we can apply the Faddeev-Popov recipe to the naive path integral (113). This includes several steps. First one promotes the infinitesimal gauge parameter $`\epsilon ^I(z)`$ to a ghost field $`๐^I(z)`$ with opposite Grassman parity. Then one looks for an appropriate gauge fixing conditions. The explicit structure of the gauge transformations (89) suggests to impose conditions only on the fields $`\phi ^I(z)`$. This can be done in many (equivalent) ways. For example, choosing a symmetric connection $``$ on $``$ we can set
(106)
$$\chi ^I(z)\frac{\phi ^I}{\theta }+\theta \left(\ddot{\phi }^I\mathrm{\Gamma }_{JK}^I(\phi )\dot{\phi }^J\dot{\phi }^K\right)=0.$$
Here $`\mathrm{\Gamma }_{JK}^I`$ are the Cristoffel symbols of $``$ and the overdot stands for the derivative in $`t`$. Let us assume that any two points of affine manifold $`(,)`$ are connected by a unique geodesics. Then equations $`\chi {}_{}{}^{I}(z)=0`$ fix the superfield $`\phi ^I(z)`$ up to arbitrary boundary values $`\phi ^I(z_0),\phi ^I(z_1)`$. To fix the residual gauge symmetry at the boundary we may use the conditions (102):
(107)
$$\gamma _a(\phi (z_0))=\gamma _a(\phi (z_1))=0.$$
Finally, to provide the correct integration measure in the path integral (73) one introduce the antighost fields $`\overline{๐_J}`$. By definition, $`ฯต(\overline{๐}_I)=ฯต(๐^I)`$. Geometrically, the fields $`๐^I`$ are $`\overline{๐}_J`$ can be viewed as taking values in tangent and cotangent spaces (with reverse parity) of the target manifold of fields $`\phi `$โs.
Since we regard the ghost-antighosts fields $`๐`$โs and $`\overline{๐}`$โs to be related with gauge symmetry in the interior of the superinterval $``$, the appropriate boundary conditions for them are
(108)
$$๐^I|_{}=0,\overline{๐}_J|_{}=0.$$
Notice that we do not introduce ghost and antighost fields associated to the residual gauge symmetry at the boundary as such fields are assumed to be already included into the action $`S`$ and the gauges $`\gamma `$โs.
After all these preparations we can write
(109)
$$F_๐ฎ=_{^{}}๐\phi ๐\overline{\phi }๐๐๐\overline{๐}\varrho [\phi ]\delta [\chi (\phi )]\delta [\gamma (\phi (z_0))]\delta [\gamma (\phi (z_1))]F(\phi (z_1))e^{\frac{i}{\mathrm{}}๐ฎ_{FP}}.$$
Here
(110)
$$๐ฎ_{FP}[\phi ,\overline{\phi },๐,\overline{๐}]=๐ฎ[\phi ,\overline{\phi }]+๐ฎ_{\mathrm{gh}}[\phi ,๐,\overline{๐}]$$
is the usual Faddeev-Popov action given by the sum of the initial gauge invariant action and the ghost action
(111)
$$\begin{array}{c}๐ฎ_{\mathrm{gh}}_{}d^2z\overline{๐_I}\left(\delta _๐\chi ^I\right)|_{\chi =0}\\ =_{}d^2z(\overline{๐}^I\frac{๐_I}{\theta }+\theta (_t๐^I_t\overline{๐}{}_{I}{}^{}+R_{IJK}^L(\phi )๐^I\dot{\phi }^J\dot{\phi }^K\overline{๐}_L)),\\ _t๐^I=\dot{๐}{}_{}{}^{I}\dot{\phi }^J\mathrm{\Gamma }_{JK}^I(\phi )๐^K,_t\overline{๐}{}_{I}{}^{}=\dot{\overline{๐}}{}_{I}{}^{}\dot{\phi }^J\mathrm{\Gamma }_{JI}^K(\phi )\overline{๐}{}_{K}{}^{}.\end{array}$$
If one disregards the Grassman nature of the Faddeev-Popov ghosts $`๐`$โs and $`\overline{๐}`$โs, then the second term in (111) is nothing but the Jacobi action for the deviation of geodesics. As to the factor
(112)
$$\varrho [\phi ]\mathrm{sdet}\left(\sigma ^{IJ}(\phi (z))\delta ^2(zw)\right),$$
it is introduced to provide the invariance of the integration measure under the gauge transformations (89). The appearance of this factor can be rigourously justified within BV scheme, but we shall not dwell on this here.
Notice that the path integral is Gaussian in the variables $`\overline{\phi }`$, $`๐`$ and $`\overline{๐}`$, and assumes no actual integration over the interior values of $`\phi `$โs due to the gauge fixing conditions. Integrating successively over all these variables we get
(113)
$$\begin{array}{c}F_๐ฎ=_\times ๐\phi (z_0)๐\phi (z_1)\delta [\gamma (\phi (z_0))]\delta [\gamma (\phi (z_1))]F(\phi (1))e^{\frac{i}{\mathrm{}}(S(\phi (z_1))S(\phi (z_0)))}\\ =(\mathrm{const})_{}๐\phi (z_1)\delta [\gamma (\phi (z_1))]F(\phi (z_1))e^{\frac{i}{\mathrm{}}S(\phi (z_1))}.\end{array}$$
(The role of the Faddeev-Popov ghosts was to compensate the Berezinian resulting from integration of the delta-functional $`\delta [\chi (\phi )]`$ and the factor (112) was exactly compensated by the integration of $`\overline{\phi }`$โs.) Including the inessential overall constant in (113) to the integration measure, we arrive at the desired equality
(114)
$$F_S=\widehat{F}_๐ฎ,$$
where the $`\widehat{F}=F(\varphi (u_1))`$ is the boundary observable of the topological sigma-model (83) constructed from the quantum observable $`F(\varphi )`$ of the original gauge theory (79). The net result, seen from (114), is that the path integral quantization based on the embedding into the topological sigma-model (65) (that does not require the original equations of motion to be Lagrangian) in the Lagrangian case brings precisely the same average values for the observables as in the standard BV-quantization.
### 6.2. General Lagrange structure of type (0,0)
In the previous section we have exemplified the quantization method applying it to a gauge system whose equations of motion are Lagrangian. In this section, we apply the method to a complementary, in a sense, particular case: the system without gauge symmetry and with independent but general (i.e. not necessarily Lagrangian) equations of motion. As will be seen, the quantization method works well in this case too, bringing the reasonable results admitting clear physical interpretation.
Let $`(,T,d_{})`$ be a Lagrange structure associated to a set of independent equations of motion $`T_a(x)=0`$, so that the matrix $`_iT_a`$ is on-shell nondegenerate. According to the general prescription of Sect.4, this classical theory is BRST embedded to the Poisson supermanifold. For the sake of simplicity we assume here that the dynamics bundle $``$ admits a flat connection $`=`$. The action of the topological sigma-model has the following structure:
(115)
$$๐ฎ=d^2z\left(\overline{x}_iDx^i+\overline{\eta }^aD\eta _a\mathrm{\Omega }\right)=S_0+(\mathrm{ghost}\mathrm{terms}),$$
where
(116)
$$S_0=_0^1(\overline{x}^idx^i\lambda ^a\stackrel{~}{T}_a(x,\overline{x})),\lambda ^a\stackrel{}{\overline{\eta }}{}_{}{}^{a},$$
is a Hamiltonian action associated to the first class constraints (cf. (58))
(117)
$$\stackrel{~}{T}(x,\overline{x})=T_a(x)+V_a^i(x)\overline{x}_i+O(\overline{x}^2),\{\stackrel{~}{T}_a,\stackrel{~}{T}_b\}=U_{ab}^c\stackrel{~}{T}_c.$$
Let us further assume that the classical master action (115) meets also the quantum master equation (98) or, what is the same, that $`๐ฎ`$ is annihilated by the (suitably regularized) odd Laplace operator.
In order to write the gauge fixed action we then introduce the non-minimal sector of BV fields: the antighosts $`\zeta _a`$ and the Lagrange multipliers $`\pi _a`$ as well as the corresponding antifields $`\stackrel{}{\zeta }^a`$ and $`\stackrel{}{\pi }^a`$. The Grassman parity and the ghost number assignments of the introduced variables are
(118)
$$\begin{array}{cccc}ฯต(\zeta _a)=1,\hfill & ฯต(\stackrel{}{\zeta }{}_{}{}^{a})=0,\hfill & ฯต(\pi _a)=0,\hfill & ฯต(\stackrel{}{\pi }{}_{}{}^{a})=1,\hfill \\ \mathrm{gh}(\zeta _a)=1,\hfill & \mathrm{gh}(\stackrel{}{\zeta }{}_{}{}^{a})=0,\hfill & \mathrm{gh}(\pi _a)=0,\hfill & \mathrm{gh}(\stackrel{}{\pi }{}_{}{}^{a})=1.\hfill \end{array}$$
The explicit form of the gauge transformation (75) suggests to impose the following gauge-fixing condition on $`\lambda `$โs:
(119)
$$\frac{d}{dt}(\lambda ^a(e))=0,$$
with $`e`$ being a nowhere vanishing vector field on $`[0,1]`$. The gauge fixing fermion associated to (119) is given by
(120)
$$\mathrm{\Psi }=\underset{0}{\overset{1}{}}๐t\zeta _a\dot{\lambda }^a,\mathrm{gh}(\mathrm{\Psi })=1,$$
where we set for simplicity $`e=_t`$. The standard non-minimal BV action $`๐ฎ+_0^1dt\pi _a\stackrel{}{\zeta }^a`$ depends thus on the fields
$$\varphi ^A=(x^i,\overline{x}_i,\lambda ^a,\pi _a,\eta _a,\overline{\eta }^a,\zeta _a)$$
and antifields
$$\varphi _A^{}=(\stackrel{}{x}{}_{}{}^{i},\stackrel{}{\overline{x}}{}_{i}{}^{},\stackrel{}{\lambda }{}_{}{}^{a},\stackrel{}{\pi }{}_{a}{}^{},\stackrel{}{\eta }_a\stackrel{}{\overline{\eta }}{}_{}{}^{a},\stackrel{}{\zeta _a}).$$
Now the gauge fixed action is obtained by restricting the non-minimal BV action to the Lagrangian submanifold $``$:
(121)
$$\varphi _A^{}=\frac{\mathrm{\Psi }}{\varphi ^A}.$$
The last equations allow one to express all the antifields via the fields. The result is
(122)
$$๐ฎ_{\mathrm{gf}}=\underset{0}{\overset{1}{}}dt(\overline{x}_i\dot{x}^i+\pi _a\dot{\lambda }^a+\dot{\overline{\eta }}{}_{}{}^{a}\dot{\zeta }_{a}^{}+\lambda ^a\stackrel{~}{T}_a(x,\overline{x})\lambda ^a\overline{\eta }^bU{}_{ab}{}^{c}(x,\overline{x})\dot{\zeta }_c).$$
The quantum average of the boundary observable $`F(x(1))`$ is defined now by the regularized version of the naive path integral (73)
(123)
$$F=_{}Fe^{\frac{i}{\mathrm{}}๐ฎ_{\mathrm{gh}}}.$$
To elucidate the meaning of the last formula it is instructive to consider the case of trivial Lagrange structure:
(124)
$$V_a^i=0,\stackrel{~}{T}_a(x,\overline{x})=T_a(x),U_{ab}^c=0.$$
Integrating in (123) of $`\overline{x}^i`$, $`\pi _a`$, $`\overline{\eta }^a`$ and $`\zeta _a`$ one finds immediately
(125)
$$\begin{array}{c}F๐x๐\lambda \delta [\dot{x}^i]\delta [\dot{\lambda }^a]F(x(1))e^{\frac{i}{\mathrm{}}{\scriptscriptstyle \lambda ^aT_a(x)}}\\ _Md^nx\delta (T_a(x))F(x)F(x_0)\end{array}$$
where $`x_0`$ is a unique solution to the classical equations of motion $`T_a(x)=0`$. Normalizing the integration measure in such a way that $`1=1`$, we can finally write $`F=F(x_0)`$. We see that the quantum average of $`F`$ involves no quantum corrections in $`\mathrm{}`$ and coincides with value of functional $`F(x)`$ on a given classical trajectory $`x_0M`$. Thus one can regards Rel.(125) as a classical vacuum-vacuum amplitude in the presence of observable. In the context of Hamiltonian mechanics such amplitudes were introduced and studied in Ref. .
In order to relate the above result with conventional formulas of quantum mechanics let us consider the intermediate possibility: the anchor is degenerate but regular at the vicinity of a classical solution $`x_0`$. In this case, due to Proposition 2.1 we can assume the equations of motion to have the form
(126)
$$_IS(y)=0,z^J=E^J(y).$$
For these equations we have the canonical Lagrange anchor $`V=(V^J,V_I)`$, where the vector fields
(127)
$$V^J=0,V_I=\frac{}{y^I}+\frac{E^J}{y^I}\frac{}{z^J}$$
form an abelian distribution. The integrability conditions (7) are obviously satisfied with $`C`$โs equals zero. Substituting these data to the gauge fixed action (122), we arrive at Gaussian path integral for the quantum average (123). As in the previous case the path integral is localized at the boundary. The calculation is rather simple, so we just write the final result
(128)
$$F๐yF(y,E(y))e^{\frac{i}{\mathrm{}}S(y)}=๐y๐zF(y,z)\delta (z^JE^J(y))e^{\frac{i}{\mathrm{}}S(y)}.$$
Here $`y^I=y^I(1)`$, $`z^J=z^J(1)`$. The quantum average is given thus by a superposition of the classical amplitude (125) for the โnon-Lagrangianโ degrees of freedom $`z`$โs and the usual Feynmanโs amplitude associated with the partial action $`S(y)`$.
In the most general case of irregular Lagrange anchor we can use the Feynman perturbation technique to obtain the quasi-classical expansion for the quantum average around the classical solution
$$x_0\mathrm{\Sigma },\overline{x}=0,\lambda =0,\pi =0,\overline{\eta }=0,\zeta =0.$$
Thus we write $`x(t)=x_0+y(t)`$, with a fluctuation field $`y(t)`$, and decompose the gauge fixed action on a free part and interaction, $`๐ฎ_{\mathrm{gf}}=๐ฎ_0+๐ฎ_{\mathrm{int}}`$, with
(129)
$$\begin{array}{c}๐ฎ_0=\underset{0}{\overset{1}{}}๐t\left(\overline{x}_i\dot{y}^i+\pi _a\dot{\lambda }^a+\dot{\overline{\eta }}{}_{}{}^{a}\dot{\zeta }_{a}^{}+\lambda ^a_iT_a(x_0)y^i\right),\hfill \\ ๐ฎ_{\mathrm{int}}=\underset{0}{\overset{1}{}}dt(\lambda ^aV_a^i(x_0)\overline{x}_i\hfill \\ +\lambda ^a\underset{k+l>1}{}\frac{1}{k!l!}_{i_1}\mathrm{}_{i_k}^{j_1}\mathrm{}^{j_l}\stackrel{~}{T}_a(x_0,0)y^{i_1}\mathrm{}y^{i_k}\overline{x}_{j_1}\mathrm{}\overline{x}_{j_l}\hfill \\ \lambda ^a\overline{\eta }^b\dot{\zeta }_c\underset{k,l=0}{\overset{\mathrm{}}{}}\frac{1}{k!l!}_{i_1}\mathrm{}_{i_k}^{j_1}\mathrm{}^{j_l}U{}_{ab}{}^{c}(x_0,0)y^{i_1}\mathrm{}y^{i_k}\overline{x}_{j_1}\mathrm{}\overline{x}_{j_l}).\hfill \end{array}$$
The Feynman propagator is then deduced from the $`๐ฎ_0`$. With account of the boundary conditions
(130)
$$\begin{array}{cc}\overline{x}^i(0)=\overline{x}^i(1)=0,& \overline{\eta }^a(0)=\overline{\eta }^a(1)=0,\\ \pi _a(0)=\pi _a(1)=0,& \zeta _a(0)=\zeta _a(1)=0,\end{array}$$
we find
(131)
$$\begin{array}{c}\lambda ^a(t)y^i(s)_0=i\mathrm{}T^{ai},\hfill \\ \overline{x}_j(t)y^i(s)_0=i\mathrm{}\delta _j^i[t\vartheta (ts)],\hfill \\ \overline{\eta }^b(t)\dot{\zeta }_a(s)_0=i\mathrm{}\delta _a^b[t\vartheta (ts)].\hfill \end{array}$$
Here we use the following definition of $`\vartheta `$-function <sup>7</sup><sup>7</sup>7There is an unavoidable ambiguity in the definition of $`\vartheta (0)`$. The value $`\vartheta (0)`$ contributes to the path integral trough the tadpole diagrams involving $`\overline{x}_j(t)y^i(t)_0`$ and $`\overline{\eta }^b(t)\dot{\zeta }_a(t)_0`$. The advantage of our choice $`\vartheta (0)=0`$ is that it leads to a covariant expression for the first quantum correction (134). A similar problem for the path-integral quantization of the Poisson sigma-model is discussed in Ref. . :
(132)
$$\vartheta (t)=\{\begin{array}{cc}1,\hfill & t>0\text{;}\hfill \\ 0,\hfill & t0\text{.}\hfill \end{array}$$
The quasi-classical expansion for the quantum average of a boundary observable $`F(x(1))`$ is given by
(133)
$$F=Fe^{\frac{i}{\mathrm{}}๐ฎ_{\mathrm{gf}}}=\underset{n=0}{\overset{\mathrm{}}{}}\frac{i^n}{\mathrm{}^nn!}F(๐ฎ_{\mathrm{int}})^ne^{\frac{i}{\mathrm{}}๐ฎ_0}$$
Using the Wick theorem for Gaussian integral we find the following expression for the first quantum correction to the classical average:
(134)
$$F=F(x_0)+\frac{i\mathrm{}}{2}\left[_i(G^{ij}_jF)_i(G^{ij}_jT_a)T^{ak}_kFU_{ab}^bT^{ai}_iF\right](x_0)+O(\mathrm{}^2).$$
Here $``$ is some connection on $`M`$. The symmetric matrix $`G^{ij}V_a^iT^{aj}`$ can be thought of as the Feynman propagator of boundary fields (cf. (46))
(135)
$$y^i(t)y^j(s)=i\mathrm{}G^{ij}(x_0)+O(\mathrm{}^2)$$
The expression for the first quantum correction is explicitly invariant under the general coordinate transformations on $`M`$ and, as one can easily check, does not depend on the choice of connection $``$.
### 6.3. First-order theories
Let $`N`$ be a smooth manifold equipped with a vector field $`h=h^i(x)_i`$. The integral trajectories of $`h`$ are defined by the system of first-order ODEs
(136)
$$T^i(x(t))\dot{x}^i(t)h^i(x(t))=0,$$
where the overdot stands for the derivative in time $`t[t_1,t_2]`$. To identify these equations with the general Eqs. (3) of Sect.2 one should combine the discrete index $`i`$ and the continuous evolution parameter $`t`$ into the one superindex $`a=(i,t)`$. Then the space of histories $`M`$ is the space of all smooth trajectories on $`N`$. Given Eqs. (136), we look for a Lagrange anchor of the form
(137)
$$V^{ij}(t,s)=\alpha ^{ij}(x(t))\delta (ts),$$
where $`\alpha =\alpha ^{ij}_i_j`$ is a contravariant tensor on $`N`$. Substituting the ansatz (137) into the integrability condition (7) we get a set of necessary and sufficient conditions for the anchor $`V`$ to be compatible with equations of motion. These conditions read
(138)
$$\alpha ^{ij}=\alpha ^{ji},[\alpha ,\alpha ]=0,[h,\alpha ]=0.$$
Here the square brackets denote the Schouten commutator of multivector fields. Rels. (138) just say that $`\alpha `$ is a Poisson bivector on $`N`$, and the vector field $`h`$ is a differentiation of the corresponding Poisson algebra.
We thus see that the Poisson structure is a particular example of the Lagrange one. One recovers it by looking for a local, purely algebraic anchor (137) for the first-order ODEs (136). When the Poisson bivector $`\alpha `$ is nondegenerate so is the anchor $`V`$. In that case the equations (136) appear to be Hamiltonian and can be derived from a (local) action functional. Meanwhile, for a degenerate anchor $`V`$ no such action can exist even if the equations (136) are Hamiltonian.<sup>8</sup><sup>8</sup>8 The last relation in (138) is automatically satisfied if $`h`$ is a (locally) Hamiltonian vector field, i.e. $`h=\rho _i\alpha ^{ij}_j`$ with $`\rho =\rho _idx^i`$ being a closed 1-form on $`N`$. For degenerate Poisson bivector the differentiation $`h`$, can be not necessarily Hamiltonian even locally, so the equations (136) are more general than the Hamilton ones.
In accordance with the definitions of Sect. 3, Rels. (136), (137), (138) define a regular Lagrange structure of type $`(0,0)`$. The corresponding BRST charge, being constructed by the general method of Sect. 4, reads
(139)
$$\mathrm{\Omega }=\underset{t_1}{\overset{t_2}{}}๐t\left(\overline{\eta }_i(\dot{x}^ih^i+\alpha ^{ij}\overline{x}_j)+\frac{1}{2}\overline{\eta }_i\overline{\eta }_j_k\alpha ^{ij}\eta ^k\right),$$
Here $`=+\mathrm{\Gamma }`$ is an arbitrary symmetric connection $`N`$. If we set $`\overline{\eta }_i(t_1)=\overline{\eta }_i(t_2)=0`$, then the BRST charge meets the master equation $`\{\mathrm{\Omega },\mathrm{\Omega }\}=0`$ with respect to the following Poisson bracket:
(140)
$$\begin{array}{cc}\{\overline{\eta }^i(t),\eta _j(s)\}=\delta _j^i\delta (ts),\hfill & \{\overline{x}_i(t),\eta _j(s)\}=\mathrm{\Gamma }_{ij}^k\eta _k\delta (ts),\hfill \\ \{\overline{x}_i(t),x^j(s)\}=\delta _i^j\delta (ts),\hfill & \{\overline{x}_i(t),\overline{\eta }^j(s)\}=\mathrm{\Gamma }_{ik}^j\overline{\eta }^k\delta (ts),\hfill \\ \{\overline{x}_i(t),\overline{x}_j(s)\}=R_{ijk}^n\overline{\eta }_n\eta ^k\delta (ts),\hfill & \end{array}$$
$`R_{ijk}^n(x)`$ being the curvature tensor of $``$. Substituting the BRST charge (139) and the symplectic potential for bracket (140) into the general formula (65), one can get the topological sigma model whose dynamics is equivalent to the original first order dynamics (136). Having this topological sigma-model, one can compute the average values (transition amplitudes) for physical observables by the formula (73), even though the original equations (136) are not Hamiltonian.
Notice that the expression (139) would reproduce the BRST charge of the Poisson sigma-model if $`h`$ was set to zero and $`=`$. This might be viewed as a possible answer to the question about the way of incorporating the Hamiltonian into the Poisson sigma-model. The BRST charge (139), containing the covariant derivative of the Poisson bivector and nilpotent with respect to the non-canonical Poisson bracket (140) probably answers to one more question , about the way of incorporating connection in the BFV-BRST quantization of Poisson sigma-model to make it explicitly covariant in the target space. Notice that the โcovariantizationโ of BFV scheme was given in for general first-class constraint systems. The above BRST-BFV formulation for the topological sigma-model can be viewed as a particular case of the covariant formalism of the paper .
This example allows further extension: the first-order equations (136) can be complemented by the constraints reducing the dynamics to a submanifold in $`N`$. Also the (sub)manifold can be factorized by a gauge symmetry. In this case, the ansatz (137) for the Lagrange anchor would result in the requirement for $`\alpha `$ to satisfy the Jacobi identity modulo constraints and gauge generators. Dynamics of this type have been recently studied in , where the deformation quantization method was extended to such systems. Similar manifolds were also studied in . The method of the present paper allows us to define the transition amplitudes for such systems using the gauged version of the sigma-model defined by the BRST charge (139).
### 6.4. Maxwell electrodynamics in the first-order formalism
In this section, we exemplify the general quantization method by the model of Maxwell electrodynamics in first-order formalism. This is a simple example, which demonstrates many characteristic features of more complicated non-Lagangian field-theoretical models.
In the first-order formalism, the electromagnetic field can be described by the strength tensor considered as an independent field, i.e. without use of the electromagnetic potential. The equations of motion for the strength tensor are not Lagrangian. These equations are dependent, i.e. there are Noether identities (8) among them, but there are no gauge transformations for the fields. The dynamics bundle (see Remark 4 of Sect. 3.2) is different from the cotangent bundle, even by dimension. So this model allows us to exemplify how the quantization method can handle with all these features that are impossible in the Lagrangian dynamics. On the other hand, as the Maxwell electrodynamics admits alternative Lagrangian formulation involving the electromagnetic potential, one can check that the quantization performed by our method gives the same results as in the standard Lagrangian formalism.
The Maxwell equations for strength tensor $`F_{\mu \nu }(x)`$ read
(141)
$$T_{1\mu }(x)^\nu F_{\mu \nu }(x)J_\mu (x)=0,T_{2\mu }(x)^\nu \stackrel{~}{F}_{\mu \nu }(x)=0,$$
where $`J_\mu (x)`$ is a conserved electric current (considered as an external source), and
(142)
$$\stackrel{~}{F}_{\mu \nu }=\frac{1}{2}ฯต_{\mu \nu \alpha \beta }F^{\alpha \beta }$$
denotes the dual strength tensor. All indices are risen and lowered by Minkowski metric in $`^{1,3}`$. Due to the antisymmetry of the strength tensor, Eqs. (141) are linearly dependent,
(143)
$$^\mu T_{z\mu }0,z=1,2.$$
To make contact with general notation of the paper introduced in Section 2, and to identify general relations (3), (8) with (141,) (143) in the Maxwell theory, one should collect the discrete and continuous indices into the following superindices: $`a=(z,\mu ,x)`$, $`A=(z,x)`$ and $`i=([\mu \nu ],x)`$.
Consider the Lagrange anchor $`V_a=(V_{1\mu },V_{2\mu })`$, where the Poincarรฉ covariant vector fields
(144)
$$V_{1\mu }=0,V_{2\mu }=d^4y^\nu \delta ^4(xy)\frac{\delta }{\delta \stackrel{~}{F}^{\mu \nu }(y)}.$$
form an abelian distribution. Since
(145)
$$V_{2\mu }T_{2\nu }=\frac{1}{2}(\eta _{\mu \nu }\mathrm{}_\mu _\nu )\delta ^4(xy),V_{2\mu }T_{1\nu }=0,$$
the integrability condition (7) is obviously satisfied. The anchor is regular but not complete (see Sect. 3.3). The physical explanation for this incompleteness is that the second equation in (141), expressing the fact of closedness of 2-form $`F_{\mu \nu }(x)dx^\mu dx^\nu `$, is considered as non-Lagrangian in the sense of Proposition 2.1. Hence, no quantum fluctuations violate the condition $`dF=0`$, and it could be (locally) resolved in terms of electromagnetic potentials $`A=A_\mu (x)dx^\mu `$ both at classical and quantum levels. (Recall that according to Rel. (128) the non-Lagrangian equations of motion enter the path integral as arguments of $`\delta `$-function thereof suppressing any quantum fluctuations of โnon-Lagrangianโ degrees of freedom).
According to the classification of Sect. 3.2, Rels. (141), (143) and (144) describe a regular Lagrange structure of type (0,1). The corresponding BRST charge is given by
(146)
$$\mathrm{\Omega }=d^4x\left(\overline{\eta }^{2\mu }[^\nu \stackrel{~}{F}_{\mu \nu }+^\nu \stackrel{~}{\overline{F}}_{\mu \nu }]+\overline{\eta }^{1\mu }[^\nu F_{\mu \nu }J_\mu ]+\overline{\xi }^z^\mu \eta _{z\mu }\right),$$
$`\overline{F}_{\mu \nu }`$ being the momenta canonically conjugated to the fields $`F^{\mu \nu }`$. Substituting the BRST charge to the general expression (65) for the sigma-model action yields <sup>9</sup><sup>9</sup>9Both the dynamics bundle and the Noether identity bundle are assigned with flat connection.
(147)
$$๐ฎ=\underset{0}{\overset{1}{}}๐t๐\theta \left[\mathrm{\Omega }+d^4x(\overline{F}_{\mu \nu }DF^{\mu \nu }+\overline{\eta }^{z\mu }D\eta _{z\mu }+\overline{\xi }^zD\xi _z)\right].$$
Integrating of $`\theta `$, we get
(148)
$$\begin{array}{c}๐ฎ=\underset{0}{\overset{1}{}}dtd^4x(\overline{F}_{\mu \nu }\dot{F}^{\mu \nu }\overline{\eta }^{z\mu }\dot{\eta }_{z\mu }+\overline{\xi }^z\dot{\xi }_z+\stackrel{}{\overline{\eta }}{}_{}{}^{2\mu }[^\nu \stackrel{~}{F}_{\mu \nu }+^\nu \stackrel{~}{\overline{F}}_{\mu \nu }]+\stackrel{}{\overline{\eta }}{}_{}{}^{1\mu }[^\nu F_{\mu \nu }J_\mu ]\\ \overline{\eta }^{2\mu }^\nu \stackrel{}{\stackrel{~}{\overline{F}}}{}_{\mu \nu }{}^{}\overline{\eta }^{2\mu }^\nu \stackrel{}{\stackrel{~}{F}}{}_{\mu \nu }{}^{}\overline{\eta }^{1\mu }^\nu \stackrel{}{F}{}_{\mu \nu }{}^{}+\overline{\xi }^z^\mu \stackrel{}{\eta }_{z\mu }+\stackrel{}{\overline{\xi }}{}_{}{}^{z}_{}^{\mu }\eta _{z\mu }).\end{array}$$
The action is invariant under the following gauge transformation:
(149)
$$\begin{array}{ccc}\delta F^{\mu \nu }=\frac{1}{2}\epsilon ^{\mu \nu \rho \sigma }_\rho \overline{\epsilon }_\sigma ^2,& \delta \overline{F}_{\mu \nu }=\frac{1}{2}_{[\mu }\overline{\epsilon }_{\nu ]}^1+\frac{1}{2}\epsilon _{\mu \nu \rho \sigma }^\rho \overline{\epsilon }^{2\sigma },& \delta \stackrel{}{\overline{\eta }}{}_{}{}^{z\mu }=\dot{\overline{\epsilon }}^{z\mu }+^\mu \stackrel{}{\overline{\epsilon }}{}_{}{}^{z},\\ \delta \overline{\eta }^{z\mu }=^\mu \overline{\epsilon }^z,& \delta \stackrel{}{\overline{\xi }}{}_{}{}^{z}=\dot{\overline{\epsilon }}^z,& \delta \overline{\xi }^z=0.\end{array}$$
(Here we write out only the transformation formulas for fields.)
From the general viewpoint, (148) is a minimal master action of a gauge theory with linearly dependent gauge-algebra generators. The fields $`\overline{\eta }`$โs and $`\stackrel{}{\overline{\xi }}`$โs are ghosts associated to the reducible gauge symmetry, and the fields $`\overline{\eta }`$โs play the role of ghosts-for-ghosts.
The gauge fixing procedure for the theory at hands is standard . We introduce the โnon-minimal sector of variablesโ constituted by the trivial field-antifield pairs ($`b`$โs, $`\stackrel{}{b}`$โs) and ($`\pi `$โs, $`\stackrel{}{\pi }`$ โs). The spectrum, parity and ghost number of these fields are completely determined by the form of gauge transformations (149) (see for details). An appropriate gauge-fixing fermion is chosen as
(150)
$$\mathrm{\Psi }=\underset{0}{\overset{1}{}}๐td^4x\left(b_{z\mu }\dot{\lambda }^{z\mu }+\dot{b}_z_\mu \lambda ^{z\mu }+b_{1z}_\mu \overline{\eta }^{z\mu }+b_1^{1z}^\mu \dot{b}_{z\mu }\right),\mathrm{gh}(\mathrm{\Psi })=1,$$
where we set $`\lambda ^{z\mu }\stackrel{}{\overline{\eta }}^{z\mu }`$. The gauge-fixed action is obtained by excluding the antifields from the non-minimal master action
(151)
$$S+\underset{0}{\overset{1}{}}dtd^4x(\pi _{z\mu }\stackrel{}{b}{}_{}{}^{z\mu }+\pi _z\stackrel{}{b}{}_{}{}^{z}+\pi _{1z}\stackrel{}{b}{}_{1}{}^{z}+\pi _1^{1z}\stackrel{}{b}{}_{1z}{}^{1}).$$
using the equation $`\varphi _i^{}=\delta \mathrm{\Psi }/\delta \varphi ^i`$, where $`\varphi ^i`$ and $`\varphi _i^{}`$ collectively denote all the fields and antifields respectively. The result is
(152)
$$\begin{array}{c}๐ฎ_{\mathrm{gf}}=d^4x\{\underset{0}{\overset{1}{}}dt[\overline{F}_{\mu \nu }\dot{F}^{\mu \nu }+\overline{\eta }^{z\mu }(\ddot{b}_{z\mu }+_\mu \ddot{b}_z)+\lambda ^{2\mu }(^\nu \stackrel{~}{F}_{\mu \nu }+^\nu \stackrel{~}{\overline{F}}_{\mu \nu })+\lambda ^{1\mu }(^\nu F_{\mu \nu }J_\mu )\\ +b_{1z}\mathrm{}\overline{\xi }^z(^\mu \dot{b}_{z\mu }+\mathrm{}\dot{b}_z)\stackrel{}{\overline{\xi }}{}_{}{}^{z}+\pi _{z\mu }(\dot{\lambda }^{z\mu }+^\mu \dot{b}_1^{1z})\pi _z_\mu \dot{\lambda }^{z\mu }+\pi _{1z}_\mu \overline{\eta }^{z\mu }+\pi _1^{1z}^\mu \dot{b}_{z\mu }]\\ +(\dot{\overline{\eta }}^{z\mu }b_{z\mu }+b_{z\mu }^\mu \stackrel{}{\overline{\xi }}{}_{}{}^{z}+\pi _{z\mu }^\mu b_1^{1z}+\pi _z_\mu \lambda ^{z\mu })|_0^1\}.\end{array}$$
We impose the following boundary conditions at $`t=0,1`$:
(153)
$$\overline{\eta }^{z\mu }=0,\overline{\xi }^z=0,\pi _1^{1z}=0,b_{z\mu }=_\mu b_z,b_1^{1z}=0.$$
The physical observables are just arbitrary functionals of the strength tensor $`F_{\mu \nu }(x)`$ and the quantum average $`๐ช`$ of an observable $`๐ช(F_{\mu \nu })`$ is given by the general expression (123). As with an arbitrary free theory, the ghost dynamics is completely decoupled from the dynamics of physical fields. Integrating over the interior values of ghost fields we arrive at the path integral with action
$$๐ฎ_{\mathrm{gf}}=d^4x\{\underset{0}{\overset{1}{}}dt[\overline{F}_{\mu \nu }\dot{F}^{\mu \nu }+A^{2\mu }(^\nu \stackrel{~}{F}_{\mu \nu }+^\nu \stackrel{~}{\overline{F}}_{\mu \nu })+A^{1\mu }(^\nu F_{\mu \nu }J_\mu )]+\left(b_z\mathrm{}\stackrel{}{\overline{\xi }}{}_{}{}^{z}+\pi _z_\mu A^{z\mu }\right)|_0^1\},$$
where $`A^{z\mu }(x)\lambda ^{z\mu }(x,0)`$. (One can check that all the determinants resulting from the integration are cancelled out.) Then, integrating of the remaining variables $`F_{\mu \nu }`$, $`\overline{F}_{\mu \nu }`$, $`A^{z\mu }`$, $`b_z`$, $`\stackrel{}{\overline{\xi }}^1`$ and $`\pi _z`$, we arrive at the final result
(154)
$$๐ช=(\mathrm{const})๐A๐F๐\pi ๐c๐\overline{c}๐ช(F_{\mu \nu })e^{\frac{i}{\mathrm{}}\stackrel{~}{S}_{\mathrm{gf}}}.$$
Here we introduced the following notation:
(155)
$$\stackrel{~}{S}_{\mathrm{gf}}=d^4x\left(F^{\mu \nu }[_\mu A_\nu +F_{\mu \nu }]+\pi ^\mu A_\mu +_\mu c^\mu \overline{c}\right),$$
(156)
$$A_\mu =A_{1\mu }(x),F_{\mu \nu }=F_{\mu \nu }(x,1),\pi =\pi _1(x,1),c=\stackrel{}{\overline{\xi }}{}_{}{}^{1}(x,1),\overline{c}=b_1(x,1).$$
As is seen the functional (155) is nothing but the gauge-fixed action of Maxwellโs electrodynamics in the first-order formalism: The boundary values of the Lagrange multipliers $`\lambda _{1\mu }`$ are identified with electromagnetic potentials $`A_\mu `$, whereas the boundary values of ghost-for-ghost $`\stackrel{}{\overline{\xi }}^1`$ and $`b_1`$ play the role of the Faddeev-Popov ghosts associated with the gauge invariance $`A_\mu A_\mu +_\mu \phi `$ and the Lorentz gauge $`^\mu A_\mu `$.
Above, one can see how the non-Lagrangian Maxwellโs equations for the strength tensor (141) are quantized strictly following the general prescription of the paper, through quantizing equivalent topological field theory in five dimensions (148). After integrating of the fields in the bulk (that can be done explicitly in this case) the result reduces to the standard expression (154) for the quantum average given by the Faddeev-Popov quantization of electrodynamics in terms of electromagnetic potential.
## 7. Concluding remarks
In this paper we have proposed a method of converting any not necessarily Lagrangian dynamics into equivalent Lagrangian topological field theory, that allows us to path-integral quantize the dynamical system. The Lagrange structure described in Sections 2 and 3 is a key pre-requisite of this quantization method. Given the equations of motion (3), they completely define their gauge symmetries (9) and the Noether identities (8), that are the basic ingredients for constructing the Lagrange structure. The Lagrange anchor (7) is the last input the Lagrange structure needs to be defined. This ingredient can not be uniformly found from the equations of motion, as the compatibility conditions (7) are not so restrictive to determine the anchor in a unique way. This is much similar to the Poisson structure which is not uniquely defined by given first order equations of motion (As is seen from the example of Sect. 6.3, this is more than a general analogy). Different Lagrange anchors can result in different quantization for the same classical theory. The trivial anchor $`V=0`$, which is compatible with any equations of motion, results in a trivial quantization in the sense that the quantum average of any physical observable would coincide with its classical value. Choosing a degenerate Lagrange anchor of constant rank, one implicitly separates the degrees of freedom which are quantized from those which are not. This is a choice which can have physical and/or geometrical motivation, but it can not be justified just by the form of the equations of motion. Even in the simplest case with equations of motion following from the action with no gauge symmetry, one can choose, in principle, not a unit anchor (that would result in the standard Feynman path integral) but a degenerate one that would assign no quantum corrections to some degrees of freedom. And this latter choice might have reasonable physical motivation in some cases, e.g. when some โpartโ of dynamics is to be considered as an effective theory emulating classical background for the other part that demonstrates a quantum behavior.
So, the general conclusion is that the Lagrange structure which is behind the path integral quantization, requires more physical/geometrical data about the system than it is contained in the classical equations of motion. This is not surprising: as quantum theory gives a more detailed description of the model, it has to require more inputs.
There are many physically interesting models, like self-dual Yang-Mills theory, Vasilievโs higher-spin fields , etc., having no Lagrangians. These models can have, however, nontrivial Lagrange structures that could allow to quantize these non-Lagrangian theories. As the equations of motion are not sufficient, in general, to uniquely define the Lagrange anchor, one has to identify in these models some other appropriate structures that could serve as building blocks for constructing the anchor. There are various ideas that can be helpful for making such an identification in local field theories. For example, not so many explicitly relativistic covariant finite order differential operators can be found for a given field, so even the most general ansatz for a local explicitly covariant Lagrange anchor can be amenable to study in many cases. As is seen from the example of section 6.4, there are quite few first order operators for the abelian vector field that can be tested for this role. If the space of trajectories $`M`$ admits any symmetric contravariant second rank tensor $`G^{ij}`$ (degenerate or not, no matter) even not related anyhow with the equations of motion, it can be taken as a propagator of Sect. 4.4, that would define the Lagrange anchor in the form $`V_a^i=_jT_aG^{ji}`$. Even the requirement of symmetry can be relaxed for $`G^{ij}`$ to hold on shell only. If other geometric data are naturally assigned to the system, other schemes can be invented, perhaps, to convert these data into the Lagrange anchor. In this paper we just propose the method to path integral quantize dynamics, given equations of motion and Lagrange anchor.
|
warning/0506/hep-th0506211.html
|
ar5iv
|
text
|
# The Topological ๐_๐ String
## 1 Introduction
Topological strings on Calabi-Yau manifolds describe certain solvable sectors of superstrings. In particular, various BPS quantities in string theory can be exactly computed using their topological twisted version. Also, topological strings provide simplified toy examples of string theories which are still rich enough to exhibit interesting stringy phenomena in a more controlled setting. There are two inequivalent ways to twist the Calabi-Yau $`\sigma `$ model which leads to the celebrated A and B model . The metric is not a fundamental degree of freedom in these models. Instead, the A-model apparently only involves the Kรคhler moduli and the B-model only the complex structure moduli. However, the roles interchange once branes are included, and it has even been conjectured that there is a version of S-duality which maps the A-model to the B-model on the same Calabi-Yau manifold . This is quite distinct from mirror symmetry which relates the A-model on $`X`$ to the B-model on the mirror of $`X`$. Subsequently, several authors found evidence for the existence of seven and/or eight dimensional theories that unify and extend the A and B-models . This was one of our motivations to take a closer look at string theory on seven-dimensional manifolds of $`G_2`$ holonomy and see whether it allows for a topological twist. We were also motivated by other issues, such as applications to M-theory compactifications on $`G_2`$-manifolds, and the possibility of improving our understanding of the relation between supersymmetric gauge theories in three and four dimensions.
In this paper, we study the construction of a topological string theory on a seven dimensional manifold of $`G_2`$ holonomy. Our approach is to define a topological twist of the $`\sigma `$ model on $`G_2`$ manifolds. On such manifolds, the $`(1,1)`$ world-sheet supersymmetry algebra gets extended to a non-linear algebra, which has a $`c=\frac{7}{10}`$ minimal model subalgebra . We use this fact to define the topological twist of the $`\sigma `$ model. This is a particular realization of a more generic result: On an orientable $`d`$ dimensional manifold which has holonomy group $`H`$ which is a subgroup of $`SO(d)`$, the coset CFT $`SO(d)_1/H_1`$ with its chiral algebra appears as a building block of the corresponding sigma model, at least at large volume. It is natural to conjecture that this building block persists at finite volume (i.e. to all orders in $`\alpha ^{}`$). It therefore gives rise to extra structure in the world sheet theory which corresponds to geometrical constructions in the target space. For example, for Calabi-Yau three folds, this extra structure is given by the $`U(1)`$ R-symmetry current, which can be used to Hodge decompose forms of total degree $`p+q`$ into $`(p,q)`$ forms. The exterior derivative has a corresponding decomposition as $`d=+\overline{}`$, and physical states in the world sheet theory correspond to suitable Dolbeault cohomology groups $`H_\overline{}^{}(X,V)`$. A $`G_2`$ manifold has an analogous refinement of the de Rham cohomology . Differential forms can be decomposed into irreducible representations of $`G_2`$. The exterior derivative can be written as the sum of two nilpotent operators $`d=\stackrel{ห}{d}+\widehat{d}`$, where $`\stackrel{ห}{d}`$ and $`\widehat{d}`$ are obtained from $`d`$ by restricting its action on differential forms to two disjoint subsets of $`G_2`$ representations. This leads to a natural question: is there a topologically twisted theory such that the BRST operator in the left (or right) sector maps to $`\stackrel{ห}{d}`$. We will see that the answer to this question is yes, and in this paper, we give the explicit construction of such a theory.
The outline of the paper is as follows. In section 2, we start by reviewing $`\sigma `$ models on target spaces of $`G_2`$ holonomy. We discuss the relation between covariantly constant p-forms on target spaces and holomorphic currents in the world sheet theory: every covariantly constant p-form leads to the existence of a chiral current supermultiplet (at least classically). A $`G_2`$ manifold has a covariantly constant 3 and 4 form leading to extra currents in the chiral algebra extending it from a $`(1,1)`$ super-conformal algebra to a non-linear algebra generated by 6 currents. As expected, this algebra contains the chiral algebra of the coset $`SO(7)_1/(G_2)_1`$, which by itself is another $`๐ฉ=1`$ superconformal algebra with central charge $`c=\frac{7}{10}`$. This is a minimal model, called the tri-critical Ising model, which plays a crucial role in defining the twisted theory. In fact, the tri-critical Ising model is what replaces the $`U(1)`$ R-symmetry of the $`๐ฉ=2`$ superconformal algebra. The full $`c=\frac{21}{2}`$ Virasoro algebra with generators $`L_n`$ splits into two commuting Virasoro algebras, $`L_n=L_n^I+L_n^r`$, with $`L_n^I`$ the generators of the $`c=\frac{7}{10}`$ tri-critical Ising model. This means that we can label highest weight states by their $`L_0^I`$ and $`L_0^r=L_0L_0^I`$ eigenvalues. We also review some facts about the tri-critical Ising model. In the NS sector, there are primary fields of weights $`0,\frac{1}{10}`$, $`\frac{6}{10}`$ and $`\frac{3}{2}`$ and in the Ramond sector, there are two primary fields of weights $`\frac{7}{16}`$ and $`\frac{3}{80}`$. We discuss the fusion rules in this model, which helps us identify the conformal block structure of various fields. This structure plays an important role in definition of the twisted theory.
In section 3, we derive a unitarity bound for the algebra which provides a non-linear inequality (a BPS bound) between the total weight of the state and its tri-critical Ising model weight. We define a notion of chiral primary states for $`G_2`$ sigma model by requiring that they saturate this bound. We also discuss the special chiral primary states in the CFT which correspond to the metric moduli that preserve the $`G_2`$ holonomy.
In section 4, we define the topological twisting of the $`G_2`$ $`\sigma `$-model. We define correlation functions in the twisted theory by relating them to certain correlation functions in the untwisted theory with extra insertion of a certain Ramond sector spin field. The twisting acts on different conformal blocks of the same local operators in a different way. We also define the BRST operator $`Q`$ as a particular conformal block of the original $`๐ฉ=1`$ supercharge. The BRST cohomology consists precisely of the chiral primary states. We discuss the chiral ring, descent relations and a suggestive localization argument which shows that the path integral localizes on constant maps. Finally, we analyze some of the putative properties of the twisted stress tensor of the theory.
In section 5, we go on to discuss the geometric interpretation of the BRST cohomology. To make this connection, we use the fact that p-forms on the $`G_2`$ manifold transforming in different $`G_2`$ representations correspond to operators in the CFT which carry different tri-critical Ising model weight ($`L_0^I`$ eigenvalue). Using this we can identify how the BRST operator acts on p-forms. We find that the BRST cohomology in the left or the right moving sector is a Dolbeault type cohomology of the differential complex $`0\mathrm{\Lambda }_1^0\mathrm{\Lambda }_7^1\mathrm{\Lambda }_7^2\mathrm{\Lambda }_1^30`$ where the differential operator is the usual exterior derivative composed with various projection operators to particular representations of $`G_2`$ as indicated by the subscript. When we combine the left and the right movers, the BRST cohomology is just as a vector space equal to the total de Rham cohomology $`H^{}(M)`$. The BRST cohomology includes the metric moduli that preserve the $`G_2`$ holonomy. These are in one-to-one correspondence with elements of $`H^3(M)`$. We also compute three point functions at genus 0 and show that these can be written as appropriate triple derivatives of a suitable generalization of Hitchinโs functional. To show this, we develop an analogue of special geometry for $`G_2`$ manifolds by defining coordinates on the moduli space of $`G_2`$ metrics as periods of the $`G_2`$ invariant three form and the dual four form. As in the case of Calabi-Yau manifolds, the dual periods are derivatives of a certain pre-potential, which is proportional to the Hitchinโs functional. We also argue that the partition function should be viewed as a wave function in a quantum mechanics corresponding to the phase space $`H^2H^3H^4H^5`$, where the symplectic form is given by integrating the wedge product of two forms over the seven manifold. We also consider the special case of the $`G_2`$ manifold being a product of Calabi-Yau and a circle and show that the twisted $`G_2`$ theory is an interesting and non-trivial combination of the A and the B model.
There is extensive literature about string theory and M-theory compactified on $`G_2`$ manifolds. The first detailed study of the world-sheet formulation of strings on $`G_2`$ manifolds appeared in . The world-sheet chiral algebra was studied in some detail in . For more about type II strings on $`G_2`$ manifolds and their mirror symmetry, see e.g. . A review of M-theory on $`G_2`$ manifolds with many references can be found in .
## 2 $`๐_\mathrm{๐}`$ sigma models
A supersymmetric $`\sigma `$ model on a generic Riemannian manifold has $`(1,1)`$ world sheet supersymmetry. However, existence of covariantly constant p-forms implies the existence of an extended symmetry algebra . This symmetry algebra is a priori only present in the classical theory. Upon quantization, it could either be lost or it could be preserved up to quantum modifications. However, since the extended symmetry is typically crucial for many properties of the theory such as spacetime supersymmetry, it is natural to postulate the extended symmetry survives quantization. To determine the quantum version of the algebra, one can for example study the most general quantum algebra with the right set of generators. For the generators expected in the $`G_2`$ case this was done in (though not with this motivation). It turns out that there is a two-parameter family of algebras with the right generators. By requiring the right value of the total central charge, and by requiring that it contains the tri-critical Ising model (which is crucial for space-time supersymmetry), both parameters are fixed uniquely leading to what we call the $`G_2`$ algebra.
Alternatively, one could have started with the special case of $`^7`$ as a model of a $`G_2`$ manifold in the infinite volume limit. This is simply a theory of free fermions and bosons, and one can easily find a quantum algebra with the right number of generators using the explicit form of the covariantly closed three and four form for $`G_2`$ manifolds written in terms of a local orthonormal frame. From this large volume point of view it is natural to expect the coset $`SO(7)_1/(G_2)_1`$ to appear, since $`SO(7)_1`$ is just a theory of free fermions and bosons. In any case, this leads to the same result for the $`G_2`$ algebra as the approach described in the previous paragraph. In the remainder of this section we will briefly describe the large volume approach.
### 2.1 Covariantly Constant p-forms and Extended Chiral Algebras
We start from a sigma model with $`(1,1)`$ supersymmetry, writing its action in superspace:
$$S=d^2zd^2\theta (G_{\mu \nu }+B_{\mu \nu })D_\theta ๐^\mu D_{\overline{\theta }}๐^\nu $$
(2.1)
where
$$D_\theta =\frac{}{\theta }+\theta \frac{}{z},D_{\overline{\theta }}=\frac{}{\overline{\theta }}+\overline{\theta }\frac{}{\overline{z}}$$
and $`๐`$ is a superfield, which, on shell can be taken to be chiral:
$$๐^\mu =\varphi ^\mu (z)+\theta \psi ^\mu (z)$$
For now, we set $`B_{\mu \nu }=0`$. This model generically has $`(1,1)`$ superconformal symmetry classically. The super stress-energy tensor is given by
$$๐(z,\theta )=G(z)+\theta T(z)=\frac{1}{2}G_{\mu \nu }D_\theta ๐^\mu _z๐^\nu $$
This $`๐ฉ=(1,1)`$ sigma model can be formulated on an arbitrary target space. However, generically the target space theory will not be supersymmetric. For the target space theory to be supersymmetric the target space manifold must be of special holonomy. This ensures that covariantly constant spinors, used to construct supercharges, can be defined. The existence of covariantly constant spinors on the manifold also implies the existence of covariantly constant *p-forms* given by
$$\varphi _{(p)}=ฯต^T\mathrm{\Gamma }_{i_1\mathrm{}i_p}ฯตdx^{i_1}\mathrm{}dx^{i_p}.$$
(2.2)
This expression may be identically zero. The details of the holonomy group of the target space manifold dictate which p-forms are actually present.
The existence of such covariantly constant p-forms on the target space manifold implies the existence of extra elements in the chiral algebra . For example, given a covariantly constant $`p`$ form, $`\varphi _{(p)}=\varphi _{i_1\mathrm{}i_p}dx^{i_p}\mathrm{}dx^{i_p}`$ satisfying $`\varphi _{i_1\mathrm{}i_p}=0`$, we can construct a holomorphic superfield current given by
$$๐_{(p)}(z,\theta )=\varphi _{i_1\mathrm{}i_p}D_\theta ๐^{i_1}\mathrm{}D_\theta ๐^{i_p}$$
which satisfies $`D_{\overline{\theta }}๐_{(p)}=0`$ on shell. In components, this implies the existence of a dimension $`\frac{p}{2}`$ and a dimension $`\frac{p+1}{2}`$ current. For example, on a Kรคhler manifold, the existence of a covariantly constant Kรคhler two form $`\omega =g_{i\overline{j}}(d\varphi ^id\varphi ^{\overline{j}}d\varphi ^{\overline{j}}d\varphi ^i)`$ implies the existence of a dimension 1 current $`J=g_{i\overline{j}}\psi ^i\psi ^{\overline{j}}`$ and a dimension $`\frac{3}{2}`$ current $`G^{}(z)=g_{i\overline{j}}(\psi ^i_z\varphi ^{\overline{j}}\psi ^{\overline{j}}_z\varphi ^i)`$, which add to the $`(1,1)`$ superconformal currents $`G(z)`$ and $`T(z)`$ to give a $`(2,2)`$ superconformal algebra. In fact, there is a non-linear extension of the $`(2,2)`$ algebra even in the case of Calabi-Yau by including generators corresponding to the (anti)holomorphic three-form. This algebra was studied in .
### 2.2 Extended algebra for $`๐_\mathrm{๐}`$ sigma models
A generic seven dimensional Riemannian manifold has $`SO(7)`$ holonomy. A $`G_2`$ manifold has holonomy which sits in a $`G_2`$ subgroup of $`SO(7)`$. Under this embedding, the eight dimensional spinor representation $`\mathrm{๐}`$ of $`SO(7)`$ decomposes into a $`\mathrm{๐}`$ and a singlet of $`G_2`$:
$$\mathrm{๐}\mathrm{๐}\mathrm{๐}$$
The singlet corresponds to a covariantly constant spinor $`ฯต`$ on the manifold satisfying
$$ฯต=0.$$
For $`G_2`$ manifolds (2.2) is non-zero only when $`p=0,3,4`$ and $`7`$ since an anti-symmetrized product of $`p`$ fundamentals ($`\mathrm{๐}`$) of $`SO(7)`$ has a $`G_2`$ singlet for these $`p`$. The zero and the seven forms just correspond to constant functions and the volume form. In addition to these, there is a covariantly constant 3-form $`\varphi ^{(3)}=\varphi _{ijk}^{(3)}dx^idx^jdx^k`$ and its Hodge dual 4-form, $`\varphi ^{(4)}=\varphi _{(3)}=\varphi _{ijkl}^{(4)}dx^idx^jdx^kdx^l`$. By the above discussion, the 3-form implies the existence of a superfield current $`๐_{(3)}(z,\theta )=\varphi _{ijk}^{(3)}D_\theta ๐^iD_\theta ๐^jD_\theta ๐^k\mathrm{\Phi }+\theta K`$. Explicitly, $`\mathrm{\Phi }`$ is a dimension $`\frac{3}{2}`$ current
$$\mathrm{\Phi }=\varphi _{ijk}^{(3)}\psi ^i\psi ^j\psi ^k$$
(2.3)
and $`K`$ is its dimension 2 superpartner
$$K=\varphi _{ijk}^{(3)}\psi ^i\psi ^j\varphi ^k.$$
(2.4)
Similarly, the 4-from implies the existence of a dimension 2 current
$$Y=\varphi _{ijkl}^{(4)}\psi ^i\psi ^j\psi ^k\psi ^l$$
(2.5)
and its dimension $`\frac{5}{2}`$ superpartner
$$N=\varphi _{ijkl}^{(4)}\psi ^i\psi ^j\psi ^k\varphi ^l.$$
(2.6)
However, as it will become clear later, instead of $`Y`$ and $`N`$, it is more useful to use the following basis of chiral currents
$$X=Y\frac{1}{2}G_{ij}\psi ^i\psi ^j$$
(2.7)
and its superpartner
$$M=N\frac{1}{2}G_{ij}\varphi ^i\psi ^j+\frac{1}{2}G_{ij}\psi ^i^2\varphi ^j.$$
(2.8)
So in summary, the $`G_2`$ sigma model has a chiral algebra generated by the following six currents
| | | | |
| --- | --- | --- | --- |
| $`h=\frac{3}{2}`$ | $`G(z)`$ | $`\mathrm{\Phi }(z)`$ | |
| | | | |
| $`h=2`$ | $`T(z)`$ | $`K(z)`$ | $`X(z)`$ |
| | | | |
| $`h=\frac{5}{2}`$ | | | $`M(z)`$ |
| | | | |
These six generators form a closed algebra which appears explicitly e.g. in (see also ). We have reproduced the algebra in appendix B. As explained in the beginning of section 2, the existence of this algebra can be taken as the definition of string theory on $`G_2`$ manifolds.
### 2.3 The Tri-critical Ising Model
An important fact, which will be crucial in almost all the remaining analysis, is that the generators $`\mathrm{\Phi }`$ and $`X`$ form a closed sub-algebra:
$`\mathrm{\Phi }(z)\mathrm{\Phi }(0)`$ $`=`$ $`{\displaystyle \frac{7}{z^3}}+{\displaystyle \frac{6}{z}}X(0)`$
$`\mathrm{\Phi }(z)X(0)`$ $`=`$ $`{\displaystyle \frac{15}{2z^2}}\mathrm{\Phi }(0){\displaystyle \frac{5}{2z}}\mathrm{\Phi }(0)`$
$`X(z)X(0)`$ $`=`$ $`{\displaystyle \frac{35}{4z^4}}{\displaystyle \frac{10}{z^2}}X(0){\displaystyle \frac{5}{z}}X(0).`$
Defining the supercurrent $`G_I=\frac{i}{\sqrt{15}}\mathrm{\Phi }`$ and stress-energy tensor $`T_I=\frac{1}{5}X`$ this is recognized to be the unique $`๐ฉ=1`$ super-conformal algebra of the minimal model with central charge $`c=\frac{7}{10}`$ known as the tri-critical Ising Model. This sub-algebra plays a similar role to the one played by the $`U(1)`$ R-symmetry in the case of Calabi-Yau target spaces. The extended chiral algebra contains two $`๐ฉ=1`$ superconformal sub-algebras: the original one generated by $`(G,T)`$ and the $`๐ฉ=1`$ superconformal sub-algebra generated by $`(\mathrm{\Phi },X)`$.
In fact, with respect to the conformal symmetry, the full Virasoro algebra decomposes in two commuting Virasoro algebras: $`T=T_I+T_r`$ with
$$T_I(z)T_r(w)=\mathrm{regular}.$$
(2.9)
This means we can classify conformal primaries by two quantum numbers, namely its tri-critical Ising model highest weight and its highest weight with respect to $`T_r`$: $`|\mathrm{primary}=|h_I,h_r`$. The Virasoro modules decompose accordingly as
$$_{c=\frac{21}{2}}=_{c=\frac{7}{10}}^I_{c=\frac{98}{10}}^{rest}.$$
(2.10)
Notice that this decomposition is with respect to the Virasoro algebras and not with respect to the $`๐ฉ=1`$ structures, which in fact do not commute. For e.g., the superpartner of $`\mathrm{\Phi }`$ with respect to the full $`๐ฉ=1`$ algebra is $`K`$ whereas its superpartner with respect to the $`๐ฉ=1`$ of the tri-critical Ising model is $`X`$.
### 2.4 Tri-critical Ising and Unitary Minimal Models
We now review a few facts about the tri-critical Ising that we will use later in the paper.
Unitary minimal models are labelled by a positive integer $`p=2,3,\mathrm{}`$ and occur only on the โdiscrete seriesโ at central charges $`c=1\frac{6}{p(p+1)}`$. The tri-critical Ising model is the second member ($`p=4`$) which has central charge $`c=\frac{7}{10}`$. In fact, it is also a minimal model for the $`๐ฉ=1`$ superconformal algebra.
The conformal primaries of unitary minimal models are labelled by two integers $`1n^{}p`$ and $`1n<p`$. Primaries with label $`(n^{},n)`$ and $`(p+1n^{},pn)`$ are identical and should be identified with each other. Therefore, there are in total $`p(p1)/2`$ primaries in the theory. The weights of the primaries are conveniently arranged into a Kac table. The conformal weight of the primary $`\mathrm{\Phi }_{n^{}n}`$ is $`h_{n^{}n}=\frac{[pn^{}(p+1)n]^21}{4p(p+1)}.`$ In the tri-critical Ising model $`(p=4)`$ there are 6 primaries of weights $`0,\frac{1}{10},\frac{6}{10},\frac{3}{2},\frac{7}{16},\frac{3}{80}`$. Below we write the Kac table for the tri-critical Ising model. Beside the Identity operator $`(h=0)`$ and the $`๐ฉ=1`$ supercurrent $`(h=\frac{3}{2})`$ the NS sector (first and third columns) contains a primary of weight $`h=\frac{1}{10}`$ and its $`๐ฉ=1`$ superpartner $`(h=\frac{6}{10})`$. The primaries of weight $`\frac{7}{16},\frac{3}{80}`$ are in the Ramond sector (middle column).
The Hilbert space of the theory decomposes in a similar way, $`=_{n,n^{}}_{n^{},n}\times \stackrel{~}{}_{n^{}n}`$. A central theme in this work is that since the primaries $`\mathrm{\Phi }_{n^{}n}`$ form a closed algebra under the OPE they can be decomposed into conformal blocks which connect two Hilbert spaces. Conformal blocks are denoted by $`\mathrm{\Phi }_{n^{},n,m^{}m}^{l^{},l}`$ which describes the restriction of $`\mathrm{\Phi }_{n^{},n}`$ to a map that only acts from $`_{m^{},m}`$ to $`_{l^{},l}`$. More details can be found in .
An illustrative example, which will prove crucial in what follows, is the conformal block structure of the primary $`\mathrm{\Phi }_{2,1}`$ of weight $`1/10`$. General arguments show that the fusion rule of this field with any other primary $`\mathrm{\Phi }_{n^{}n}`$ is $`\varphi _{(2,1)}\times \varphi _{(n^{},n)}=\varphi _{(n^{}1,n)}+\varphi _{(n^{}+1,n)}.`$ The only non-vanishing conformal blocks in the decomposition of $`\mathrm{\Phi }_{2,1}`$ are those that connect a primary with the primary right above it and the primary right below in the Kac table, namely, $`\varphi _{2,1,n^{},n}^{n^{}1,n}`$ and $`\varphi _{2,1,n^{},n}^{n^{}+1,n}`$. This can be summarized formally by defining the following decomposition<sup>1</sup><sup>1</sup>1Perhaps the notation with $``$ and $``$ is a bit misleading. By $`\mathrm{\Phi }_{2,1}^{}`$, we mean that conformal block of $`\mathrm{\Phi }_{2,1}`$ which maps
$$_0\stackrel{\mathrm{\Phi }_{2,1}^{}}{}_{\frac{1}{10}}\stackrel{\mathrm{\Phi }_{2,1}^{}}{}_{\frac{6}{10}}\stackrel{\mathrm{\Phi }_{2,1}^{}}{}_{\frac{3}{2}}$$
(2.11) This is going down only in the first column of the Kac table, but is actually going up in the third column.
$$\mathrm{\Phi }_{2,1}=\mathrm{\Phi }_{2,1}^{}\mathrm{\Phi }_{2,1}^{}.$$
(2.12)
Similarly, the fusion rule of the Ramond field $`\mathrm{\Phi }_{1,2}`$ with any primary is $`\varphi _{(1,2)}\times \varphi _{(n^{},n)}=\varphi _{(n^{},n1)}+\varphi _{(n^{},n+1)}`$ showing that it is composed of two blocks, which we denote as follows
$$\mathrm{\Phi }_{1,2}=\mathrm{\Phi }_{1,2}^{}\mathrm{\Phi }_{1,2}^+.$$
(2.13)
It is important here to specify on which half of the Kac table we are acting. We take $`\varphi _{(n^{},n)}`$ to be either in the first column or in the top half of the second column, i.e. in the boldface region of table 1. With this restriction we denote by $`\mathrm{\Phi }_{1,2}^{}`$ the conformal block that takes us to the left in the Kac table and $`\mathrm{\Phi }_{1,2}^+`$ the one that takes us to the right. Conformal blocks transform under conformal transformations exactly like the primary field they reside in but are usually not single-valued functions of $`z(\overline{z})`$. This splitting into conformal blocks plays a crucial role in the twisting procedure. The $`+`$ and $``$ labels will be clarified further when we consider the Ramond sector of the full $`G_2`$ algebra in section 7.1 where we see that these labels correspond to Ramond sector ground states with different fermion numbers.
## 3 Chiral Primaries, Moduli and a Unitarity Bound
Having discussed this $`c=\frac{7}{10}`$ subalgebra we now turn to the full $`G_2`$ chiral algebra. We first identify a set of special states which will turn out to saturate a unitarity bound for the full $`G_2`$ algebra. We call these the chiral primary states. This name seems appropriate since the representations built on chiral primary states are โshortโ whereas the generic representation is โlong.โ The chiral primary states include the moduli of the compactification, i.e. the metric and $`B`$-field moduli that preserve the $`G_2`$ holonomy.
### 3.1 Chiral Primary States
The chiral-algebra associated with manifolds of $`G_2`$ holonomy<sup>2</sup><sup>2</sup>2We loosely refer to it as โthe $`G_2`$ algebraโ but it should not be confused with the Lie algebra of the group $`G_2`$. allows us to draw several conclusions about the possible spectrum of such theories. It is useful to decompose the generators of the chiral algebra in terms of primaries of the tri-critical Ising model and primaries of the remainder (2.10). The commutation relations of the $`G_2`$ algebra imply that some of the generators of the chiral algebra decompose as : $`G(z)=\mathrm{\Phi }_{2,1}\psi _{\frac{14}{10}},K(z)=\mathrm{\Phi }_{3,1}\psi _{\frac{14}{10}}`$ and $`M(z)=a\mathrm{\Phi }_{2,1}\chi _{\frac{24}{10}}+b[X_1,\mathrm{\Phi }_{2,1}]\psi _{\frac{14}{10}},`$ with $`\psi ,\chi `$ primaries of the indicated weights in the $`T_r`$ CFT and $`a,b`$ constants.
The Ramond sector ground states on a seven dimensional manifold (so that the corresponding CFT has $`c=21/2`$) have weight $`\frac{7}{16}`$. This implies that these states, which are labelled by two quantum numbers (the weights under the tri-critical part and the remaining CFT), are $`|\frac{7}{16},0`$ and $`|\frac{3}{80},\frac{2}{5}`$. The existence of the $`|\frac{7}{16},0`$ state living just inside the tri-critical Ising model is crucial for defining the topological theory. Coupling left and right movers, the only possible RR ground states compatible with the $`G_2`$ chiral algebra<sup>3</sup><sup>3</sup>3Otherwise the spectrum will contain a 1-form which will enhance the chiral algebra . Geometrically this is equivalent to demanding that $`b_1=0`$. are a single $`|\frac{7}{16},0_L|\frac{7}{16},0_R`$ ground state and a certain number of states of the form $`|\frac{3}{80},\frac{2}{5}_L|\frac{3}{80},\frac{2}{5}_R`$. For a further discussion of the RR ground states see also section 7.1 and appendix C.
By studying operator product expansions of the RR ground states using the fusion rules
$`{\displaystyle \frac{7}{16}}\times {\displaystyle \frac{7}{16}}`$ $`=`$ $`0+{\displaystyle \frac{3}{2}}`$
$`{\displaystyle \frac{7}{16}}\times {\displaystyle \frac{3}{80}}`$ $`=`$ $`{\displaystyle \frac{1}{10}}+{\displaystyle \frac{6}{10}}`$
we get the following โspecialโ NSNS states
$$|0,0_L|0,0_R,|\frac{1}{10},\frac{2}{5}_L|\frac{1}{10},\frac{2}{5}_R,|\frac{6}{10},\frac{2}{5}_L|\frac{6}{10},\frac{2}{5}_R\mathrm{and}|\frac{3}{2},0_L|\frac{3}{2},0_R$$
(3.1)
corresponding to the 4 NS primaries $`\mathrm{\Phi }_{n^{},1}`$ with $`n^{}=1,2,3,4`$ in the tri-critical Ising model. Note that for these four states there is a linear relation between the Kac label $`n^{}`$ of the tri-critical Ising model part and the total conformal weight $`h_{total}=\frac{n^{}1}{2}`$. In fact, in section 3.3, we show that similar to the BPS bound in the $`๐ฉ=2`$ case, primaries of the $`G_2`$ chiral algebra satisfy a (non-linear) bound of the form
$$h_I+h_r\frac{1+\sqrt{1+80h_I}}{8}.$$
(3.2)
which is precisely saturated for the four NS states listed above. We will therefore refer to those states as โchiral primaryโ states. Just like in the case of Calabi-Yau, the $`\frac{7}{16}`$ field maps Ramond ground states to NS chiral primaries and is thus an analogue of the โspectral flowโ operators in Calabi-Yau.
### 3.2 Moduli
It was shown in that the upper components
$$\stackrel{~}{G}_{\frac{1}{2}}|\frac{1}{10},\frac{2}{5}_LG_{\frac{1}{2}}|\frac{1}{10},\frac{2}{5}_R$$
correspond to exactly marginal deformations of the CFT preserving the $`G_2`$ chiral algebra
$$\{G_{\frac{1}{2}},๐ช_{\frac{1}{10},\frac{2}{5}}\}=๐ช_{0,1}.$$
(3.3)
and as such, correspond to the moduli of the $`G_2`$ compactification. As we will see in more detail later, there are $`b_2+b_3`$ such moduli.
Geometrically, the metric moduli are deformations of the metric ($`\delta g_{ij}`$) that preserve Ricci flatness (these deformations also preserve the $`G_2`$ structure). Such deformations satisfy the Lichnerowicz equation:
$$\mathrm{\Delta }_L\delta g_{ij}^2\delta g_{ij}+2R_{mijn}\delta g^{mn}+2R_{(i}^k\delta g_{j)k}=0.$$
(3.4)
That there are $`b_3`$ solutions to this equation (up to diffeomorphisms) can be seen by relating (3.4) to an equation for a three-form $`\omega `$ which is constructed out of $`\delta g`$ via $`\delta g_{ij}`$: $`\omega _{ijk}=\varphi _{l[ij}\delta g_{k]}^l.`$ Indeed, it can be shown that for every solution of (3.4) modulo diffeomorphisms there is a corresponding harmonic three-form:
$$\mathrm{\Delta }_L\delta g=0\mathrm{\Delta }\omega =0.$$
(3.5)
A natural question is if $`\mathrm{\Delta }_L`$ can be written as the square of some first order operator. Such a construction exists if the manifold supports a covariantly constant spinor $`ฯต_0`$. We can construct a spinor valued one-form out of $`\delta g_{ij}`$ as $`\delta g_{ij}(\mathrm{\Gamma }^iฯต_0)dx^j`$. This is a section of S(M) $`T^{}M`$ where $`S(M)`$ is the spin bundle. There is a natural $`D/`$ operator acting on this vector bundle. It can be shown that $`D/^{}D/=\mathrm{\Delta }_L`$, which then reduces (3.4) to
$$D/\left(\delta g_{ij}\mathrm{\Gamma }^iฯต_0dx^j\right)=0$$
(3.6)
which was shown to imply
$$_i\delta g_{jk}\varphi ^{ij}{}_{l}{}^{}=0$$
(3.7)
in . This first order condition for the metric moduli will be beautifully reproduced from our analysis later of the BRST cohomology of our topologically twisted sigma model.
There is another quick way to see how the condition of being chiral primary implies the first order condition (3.7). This is done using the zero mode of the generator $`K(z)`$ of the $`G_2`$ algebra. In the next section we will find that $`K_0=0`$ for chiral primaries using some explicit calculations. One can also show this more generally, since the $`K_0`$ eigenvalue of highest weight states of the $`G_2`$ algebra can be determined in terms of their $`L_0`$ and $`X_0`$ eigenvalues by using the fact that the null ideal in (B.19) has to vanish when acting on such states (see appendix B). Again this leads to the conclusion that $`K_0=0`$ for chiral primaries. Now in the large volume limit the operator $`๐ช_{\frac{1}{10},\frac{2}{5}L}\times ๐ช_{\frac{1}{10},\frac{2}{5}R}`$, correspond to the operator $`\delta g_{ij}\psi _L^i\psi _R^j`$. <sup>4</sup><sup>4</sup>4The tri-critical Ising model weight of this operator can be computed to be $`\frac{1}{10}`$ by taking the OPE of it with $`X`$ and then extracting the second order pole. The $`K_0`$ eigenvalue is then easily extracted from the double pole in the OPE
$$K(z)๐ช_{\frac{1}{10},\frac{2}{5}L}(0)\mathrm{}+\frac{_i\delta g_{jk}\varphi ^{ij}{}_{l}{}^{}๐ช_{\frac{1}{10},\frac{2}{5}L}^{}(0)}{z^2}+\mathrm{}.$$
(3.8)
We see that $`K_0=0`$ implies precisely the first order condition (3.7) which is a nice consistency check of the framework.
### 3.3 A Unitarity Bound
The $`G_2`$ algebra has highest weight representations, made from a highest weight vector that is annihilated by all positive modes of all the generators. First, notice that when acting on highest weight states, the generators $`L_0,X_0`$ and $`K_0`$ commute<sup>5</sup><sup>5</sup>5 The only subtlety is the $`[X_0,K_0]`$ commutator. It does not vanish in general, but it does vanish when acting on highest weight states. so a highest weight state can be labelled by the three eigenvalues $`l_0,x_0,k_0`$ <sup>6</sup><sup>6</sup>6As we mentioned in the previous subsection, $`k_0`$ is determined in terms of $`l_0`$ and $`x_0`$ by requiring the vanishing of the null ideal (B.19) when acting on these states. We ignore this in this subsection, though it does alter the analysis.. In addition, $`l_00`$, $`x_00`$, and $`k_0`$ is purely imaginary. The first two conditions follow from unitarity (recall that $`5X`$ is the stress tensor of the tri-critical Ising model), the last condition follows from the hermiticity conditions on $`K_0`$: $`K_m^{}=K_m`$.
Next, we want to derive some bounds on $`l_0,x_0,k_0`$ that come from unitarity. In particular, we consider the three states $`\{G_{1/2}|l_0,x_0,k_0`$, $`\mathrm{\Phi }_{1/2}|l_0,x_0,k_0`$, $`M_{1/2}|l_0,x_0,k_0\}`$ and we consider the matrix $``$ of inner products of these states with their hermitian conjugates<sup>7</sup><sup>7</sup>7This analysis assumes that $`x_0`$ is strictly negative otherwise $`\mathrm{\Phi }_{\frac{1}{2}}|l_0,0,k_0`$ vanishes. For $`x_0`$ we remove this state and consider the matrix of inner products of the remaining two states, which leads to exactly the same conclusion.. This matrix can be worked out using the commutation relations and we find
$$=\left(\begin{array}{ccc}2l_0& k_0& l_0+2x_0\\ k_0& 6x_0& 5k_0/2\\ 2x_0+l_0& 5k_0/2& l_0/2+4x_08x_0l_0\end{array}\right)$$
(3.9)
This matrix is indeed hermitian, and unitarity implies that the eigenvalues of this matrix should be nonnegative. In particular, the determinant should be nonnegative
$$det=(8l_06x_08l_0x_0)k_0^2+24x_0^2(4l_0^2l_0+x_0).$$
(3.10)
The piece between parentheses before $`k_0^2`$ is always positive, and $`k_0^2`$ is always negative. Therefore we should in particular require that (for $`x_00`$)
$$4l_0^2l_0+x_00$$
(3.11)
which implies
$$l_0\frac{1+\sqrt{116x_0}}{8}.$$
(3.12)
Changing basis to eigenvalues of $`T_r,T_I`$ (see 2.9) the bound (3.12) becomes
$$h_I+h_r\frac{1+\sqrt{1+80h_I}}{8}.$$
(3.13)
This bound will turn out to play an important role. When the bound is saturated, we will call the corresponding state โchiral primaryโ in analogy to states saturating the BPS bound in $`๐ฉ=2`$. Since in the NS sector of the tri-critical Ising model, $`h_I=0,\frac{1}{10},\frac{6}{10},\frac{3}{2}`$ chiral states have total $`h_I+h_r`$ scaling dimension $`0,\frac{1}{2},1,\frac{3}{2}`$ which exactly match the special NSNS states 3.1. We will see that just like for $`๐ฉ=2`$ theories it is exactly those chiral states that survive the topological twist. Indeed, in the Coulomb gas approach they became weight zero after the twist. It is interesting to see that the definition of chiral primaries involves a nonlinear identity. This reflects the fact that the $`G_2`$ chiral algebra is non-linear. Since $`det=0`$ for chiral primaries, a suitable linear combination of the three states used in building $`det`$ vanishes. In other words, chiral primaries are annihilated by a combination of fermionic generators and the representations built from chiral primaries will be smaller than the general representation, as expected for BPS states.
When the bound (3.13) is saturated, $`det`$ can only be nonnegative as long as $`k_0=0`$. Thus, chiral primaries necessarily have $`k_0=0`$, and we will mostly suppress the quantum number $`k_0`$ in the remainder.
## 4 Topological Twist
To construct a topologically twisted CFT, we usually proceed in two steps. First we define a new stress-energy tensor, which changes the quantum numbers of the fields and operators of the theory under Lorentz transformations. Secondly, we identify a nilpotent scalar operator, usually constructed out of the supersymmetry generators of the original theory, which we declare to be the BRST operator. Often this BRST operator can be obtained in the usual way by gauge fixing a suitable symmetry. If the new stress tensor is exact with respect to the BRST operator, observables (which are elements of the BRST cohomology) are metric independent and the theory is called topological. In particular, the twisted stress tensor should have a vanishing central charge.
### 4.1 Review of twisting the Calabi-Yau $`\sigma `$-model
In practice , for the $`๐ฉ=2`$ theories, an n-point correlator on the sphere in the twisted theory can conveniently be defined<sup>8</sup><sup>8</sup>8Up to proper normalization. as a correlator in the *untwisted* theory of the same n operators plus two insertions of a spin-field, related to the space-time supersymmetry charge, that serves to trivialize the spin bundle. For a Calabi-Yau 3-fold target space there are two $`SU(3)`$ invariant spin-fields which are the two spectral flow operators $`๐ฐ_{\pm \frac{1}{2}}`$. This discrete choice in the left and the right moving sectors is the choice between the $`+()`$ twists which results in the difference between the topological $`A/B`$ models.
The action for the $`\sigma `$-model on a Calabi-Yau is given by
$$S=d^2z\frac{1}{2}g_{i\overline{j}}\left(x^i\overline{}x^{\overline{i}}+x^{\overline{i}}\overline{}x^i\right)+g_{i\overline{j}}\left(i\psi _{}^{\overline{j}}D\psi _{}^i+i\psi _+^{\overline{j}}\overline{D}\psi _+^i\right)+R_{i\overline{j}k\overline{l}}\psi _+^i\psi _+^{\overline{j}}\psi _{}^k\psi _{}^{\overline{l}}$$
(4.1)
Twisting this $`\sigma `$-model corresponds to adding a background gauge field for the $`U(1)`$ which acts on the complex fermions. Effectively, we change the covariant derivative from $`D=+\frac{\omega }{2}`$ to $`D^{}=+\frac{\omega }{2}+A`$, where we set the background value of $`A=\frac{\omega }{2}`$. Similarly, $`\overline{D}`$ changes to $`\overline{D}^{}=\overline{}+\frac{\overline{\omega }}{2}\pm \overline{A}`$, where the $`+`$ sign refers to the B twist and the $``$ sign refers to the A twist. This has the effect of changing the action in the following way:
$$\delta S=g_{i\overline{j}}\psi _+^i\psi _+^{\overline{j}}\frac{\overline{\omega }}{2}\pm g_{i\overline{j}}\psi _{}^i\psi _{}^{\overline{j}}\frac{\omega }{2}$$
(4.2)
Just considering the left moving sector, and bosonizing the $`\psi _+`$โs by defining $`g_{i\overline{j}}\psi _+^i\psi _+^{\overline{j}}=i\sqrt{d}\varphi `$, where $`d`$ is the complex dimension of the Calabi-Yau, we find
$$\delta S=g_{i\overline{j}}\psi _+^i\psi _+^{\overline{j}}\frac{\omega }{2}=i\frac{\sqrt{d}}{2}\varphi \omega =+i\frac{\sqrt{d}}{2}\varphi R.$$
On a genus $`g`$ Riemann surface, we can choose $`R`$ such that it has $`\delta `$-function support at $`22g`$ points. So for example, on a sphere, we get
$$e^{\delta S}=e^{i\frac{\sqrt{d}}{2}\varphi (0)}e^{i\frac{\sqrt{d}}{2}\varphi (\mathrm{})}$$
which implies that correlation functions in the twisted theory are related to those in the untwisted theory by $`22g`$ insertions of the operator (also known as the spectral flow operator) $`e^{i\frac{\sqrt{d}}{2}\varphi }`$:
$$\mathrm{}_{\mathrm{twisted}}=e^{i\frac{\sqrt{d}}{2}\varphi (\mathrm{})}\mathrm{}e^{i\frac{\sqrt{d}}{2}\varphi (0)}_{\mathrm{untwisted}}$$
This effectively adds a background charge for the field $`\varphi `$ of magnitude $`Q=\sqrt{d}`$, changing the central charge of the CFT
$$c=\frac{3}{2}\times 2d13Q^2+3d1=0$$
which is what we expect in a topological theory.
### 4.2 The $`๐_\mathrm{๐}`$ Twist On The Sphere
We can apply a similar procedure to the $`G_2`$ $`\sigma `$-model. The role of the operator $`e^{i\frac{\sqrt{d}}{2}\varphi }`$ will be played by the conformal block $`\mathrm{\Phi }_{1,2}^+`$ of the primary with conformal weight $`\frac{7}{16}`$ which creates the state $`|\frac{7}{16},0`$. Notice that this state sits entirely inside the tri-critical Ising model. Indeed, also in the case of Calabi-Yau manifolds, the spectral flow operator $`e^{i\frac{\sqrt{d}}{2}\varphi }`$, sits purely within the $`U(1)=\frac{U(d)}{SU(d)}`$ part. In $`G_2`$ manifolds, the coset $`\frac{SO(7)_1}{(G_2)_1}`$ (with central charge $`\frac{7}{10}`$) plays the same role as the $`U(1)`$ subalgebra in $`๐ฉ=2.`$ We therefore suggest (refining a similar suggestion of ) that correlation functions of the twisted theory are defined in terms of the untwisted theory as
$$\begin{array}{cc}\hfill & V_1(z_1)\mathrm{}V_n(z_n)_{\mathrm{๐๐ ๐๐๐๐๐}}^{\mathrm{๐๐๐๐๐}}\hfill \\ \hfill \underset{i=1}{\overset{n}{}}z_i^{(h_i\stackrel{~}{h_i})}& \mathrm{\Sigma }(\mathrm{})V_1(z_1)\mathrm{}V_n(z_n)\mathrm{\Sigma }(0)_{\mathrm{๐๐๐๐ ๐๐๐๐๐}}^{\mathrm{๐๐๐๐๐}}\hfill \end{array}$$
(4.3)
where, $`(h)\stackrel{~}{h}`$ are the weights with respect to the (un)twisted stress tensor respectively<sup>9</sup><sup>9</sup>9The product $`_{i=1}^nz_i^{(h_i\stackrel{~}{h_i})}`$ comes about from the mapping between the flat cylinder and the sphere. Note that this is not the same as computing the expectation value of $`V_1(z_1)\mathrm{}V_n(z_n)`$ in the Ramond ground state $`\mathrm{\Sigma }(0)|0`$ because we insert the same operator at $`0,\mathrm{}`$ and not an operator and its BPZ conjugate. and $`\mathrm{\Sigma }`$ is the conformal block
$$\mathrm{\Sigma }=\mathrm{\Phi }_{1,2}^+$$
(4.4)
defined in (2.13).
In further arguments were given, using the Coulomb gas representation of the minimal model, that there exists a twisted stress tensor with vanishing central charge. Those arguments, which are briefly reviewed in appendix A, are problematic because the Coulomb gas representation really adds additional degrees of freedom to the minimal model. To properly restrict to the minimal model, one needs to consider cohomologies of BRST operators defined by Felder . The proposed twisted stress tensor of does not commute with Felderโs BRST operators and therefore it does not define a bona fide operator in the minimal model. In addition, a precise definition of a BRST operator for the topological theory was lacking in .
We will proceed differently. We formulate our discussion purely in terms of the tri-critical Ising model itself without ever referring to the Coulomb gas representation, except by way of motivation and intuition. We will propose a BRST operator, study its cohomology, and then use 4.3 to compute correlation functions of BRST invariant observables. The connection to target space geometry will be made. We will then comment on the extension to higher genus and on the existence of a topologically twisted $`G_2`$ string.
### 4.3 The BRST operator
The basic idea is that the topological theory for $`G_2`$ sigma models should be formulated in terms of its (non-local)<sup>10</sup><sup>10</sup>10It should be stressed that this splitting into conformal blocks is non-local in the sense that conformal blocks may be multi-valued functions of z ($`\overline{z}`$). conformal blocks and not in terms of local operators. By using the split (2.12) into conformal blocks, we can split any field whose tri-critical Ising model part contains just the conformal family $`\mathrm{\Phi }_{2,1}`$ into its up and down parts. For example, the $`๐ฉ=1`$ supercurrent $`G(z)`$ can be split as
$$G(z)=G^{}(z)+G^{}(z).$$
(4.5)
We claim that $`G^{}`$ is the BRST current and $`G^{}`$ is a candidate for the for the anti-ghost<sup>11</sup><sup>11</sup>11Incidently, the Coulomb gas representation indeed assigns the expected conformal weights after the twist (see appendix A).. The basic $`๐ฉ=1`$ relation
$$G(z)G(0)=\left(G^{}(z)+G^{}(z)\right)\left(G^{}(0)+G^{}(0)\right)\frac{2c/3}{z^3}+\frac{2T(0)}{z}$$
(4.6)
proves the nilpotency of this BRST current (and of the candidate anti-ghost) because the RHS contains descendants of the identity operator only and has trivial fusion rules with the primary fields of the tri-critical Ising model and so $`(G^{})^2=(G^{})^2=0`$.
An algebraic formulation of the decomposition 4.5 starts from defining projection operators. Any state in the theory can be labelled by its eigenvalues under the two commuting (2.9) Virasoro modes of $`T_I,T_r`$ and perhaps some additional quantum numbers needed to completely specify the state. We denote by $`P_n^{}`$ the projection operator on the sub-space of states whose tri-critical Ising model part lies within the conformal family of one of the four NS primaries $`\mathrm{\Phi }_{n^{},1}`$. The image of $`P_n^{}`$ is $`_{n^{},1}`$ which we abbreviate here to $`_n^{}`$. The corresponding weights of the primary fields in the tri-critical Ising model by $`\mathrm{\Delta }(n^{})`$. Thus, $`\mathrm{\Delta }(1)=0`$, $`\mathrm{\Delta }(2)=\frac{1}{10}`$, $`\mathrm{\Delta }(3)=\frac{6}{10}`$ and $`\mathrm{\Delta }(4)=\frac{3}{2}`$. This is summarized by the equation
$$\mathrm{\Delta }(n^{})=\frac{(2n^{}3)(n^{}1)}{10}.$$
(4.7)
The 4 projectors add to the identity
$$P_1+P_2+P_3+P_4=1$$
(4.8)
because this exhaust the list of possible highest weights in the NS sector of the tri-critical Ising model<sup>12</sup><sup>12</sup>12For simplicity, we will set $`P_n^{}=0`$ for $`n^{}0`$ and $`n^{}5`$, so that we can simply write $`_n^{}P_n^{}=1`$ instead of (4.8)..
We can now define our candidate BRST operator in the NS sector more rigorously
$$Q=G_{\frac{1}{2}}^{}\underset{n^{}}{}P_{n^{}+1}G_{\frac{1}{2}}P_n^{}.$$
(4.9)
The nilpotency $`Q^2=0`$ is easily proved:
$$Q^2=\underset{n^{}}{}P_{n^{}+2}G_{\frac{1}{2}}^2P_n^{}=\underset{n^{}}{}P_{n^{}+2}L_1P_n^{}=0$$
(4.10)
where we could replace the intermediate $`P_{n^{}+1}`$ by the identity because of the property 4.5 and the last equality follows since $`L_1`$ maps each $`_n^{}`$ to itself.
### 4.4 BRST Cohomology and Chiral Operators
Having defined the BRST operator, we can now compute its cohomology. We first derive the condition on the tri-critical Ising model weight $`h_I`$ and its total weight for it to be annihilated by $`Q`$. Then we go on to defining the operator cohomology, which correspond to operators (or conformal blocks of operators) $`๐ช`$ satisfying $`\{Q,๐ช\}=0`$. We mostly work in the NS sector. Perhaps it is more appropriate to work in the Ramond sector since the topological theory computations are done in the Ramond sector of the untwisted theory (see also section 4.9). We assume here that a version of spectral flow exists which will map the NS sector to the Ramond sector. We discuss such a spectral flow in appendix F.
#### 4.4.1 State Cohomology
As a first step in the analysis of the BRST cohomology, we consider the action of $`Q`$ on highest weight states $`|h_I,h_r=|\mathrm{\Delta }(k),h_r`$ of the full algebra. Because $`Q`$ is a particular conformal block of the supercharge $`G_{\frac{1}{2}}`$, to extract the action of $`Q`$ on a state, we first act with $`G_{\frac{1}{2}}`$ on the state and then project on to the term. As discussed previously, the $`๐ฉ=1`$ supercurrent $`G`$ can be decomposed as $`\mathrm{\Phi }_{2,1}\psi _{\frac{14}{10}}`$. The fusion rules of the tri-critical Ising model then imply that
$`G_{1/2}|\mathrm{\Delta }(k),h_r`$ $`=`$ $`c_1|\mathrm{\Delta }(k1),h_r\mathrm{\Delta }(k1)+\mathrm{\Delta }(k){\displaystyle \frac{1}{2}}`$ (4.11)
$`+c_2|\mathrm{\Delta }(k+1),h_r\mathrm{\Delta }(k+1)+\mathrm{\Delta }(k){\displaystyle \frac{1}{2}}`$
where the two states on the right are highest weight states of the $`L_m,X_m`$ subalgebra (but not necessarily of the full $`G_2`$ algebra) and which are normalized to have unit norm. Then by definition
$$Q|\mathrm{\Delta }(k),h_r=c_2|\mathrm{\Delta }(k+1),h_r\mathrm{\Delta }(k+1)+\mathrm{\Delta }(k)\frac{1}{2}.$$
(4.12)
Using the $`G_2`$ algebra (appendix B), we find that
$$\mathrm{\Delta }(k),h_r|G_{1/2}G_{1/2}|\mathrm{\Delta }(k),h_r=2(\mathrm{\Delta }(k)+h_r)=|c_1|^2+|c_2|^2.$$
(4.13)
The first answer is obtained using $`\{G_{1/2},G_{1/2}\}=2L_0`$, the second follows from (4.11). In a similar way we compute
$`\mathrm{\Delta }(k),h_r|G_{1/2}X_0G_{1/2}|\mathrm{\Delta }(k),h_r`$ $`=`$ $`9\mathrm{\Delta }(k)h_r10\mathrm{\Delta }(k)(\mathrm{\Delta }(k)+h_r)`$ (4.14)
$`=`$ $`5\mathrm{\Delta }(k1)|c_1|^25\mathrm{\Delta }(k+1)|c_2|^2.`$
We can use (4.13) and (4.14) to solve for $`c_1`$ and $`c_2`$ up to an irrelevant phase. In particular, we find that the highest weight state is annihilated by $`Q`$, which is equivalent to $`c_2=0`$, if
$$9\mathrm{\Delta }(k)h_r10\mathrm{\Delta }(k)(\mathrm{\Delta }(k)+h_r)=10\mathrm{\Delta }(k1)(\mathrm{\Delta }(k)+h_r).$$
(4.15)
We can rewrite this as
$$\mathrm{\Delta }(k)+h_r=\frac{10\mathrm{\Delta }(k)}{10\mathrm{\Delta }(k)+110\mathrm{\Delta }(k1)}=\frac{k1}{2}=\frac{1+\sqrt{1+80\mathrm{\Delta }(k)}}{8}$$
(4.16)
where we used (4.7). This is precisely the unitarity bound (3.13). Therefore, the only highest weight states that are annihilated by $`Q`$ are the chiral primaries that saturate the unitarity bound. It is gratifying to see a close parallel with the other examples of topological strings in four and six dimensions<sup>13</sup><sup>13</sup>13 Strictly speaking the above derivation is not quite correct for $`k=1,4`$, since $`\mathrm{\Delta }(0)`$ and $`\mathrm{\Delta }(5)`$ do not exist. If they would appear, then the corresponding representations would not be unitary, since they lie outside the Kac table. This implies that the only representations with either $`k=0`$ or $`k=3`$ that can appear in the theory necessarily have $`h_r=0`$, and these are indeed annihilated by the BRST operator.. We have shown so far that all states that are primary under the $`L_m,X_m`$ subalgebra and are annihilated by $`G_{1/2}`$ are annihilated by $`Q`$ if they saturate the unitarity bound. These states, need not be primary with respect to the full $`G_2`$ algebra. This is implied by the condition $`|c_1|^20`$ in (4.13) and (4.14).
Of course, to study the full BRST cohomology, much more work is required, and in particular we would want to prove that BRST closed descendants are always BRST exact. We donโt have such a proof, but some partial evidence is given in section 4.6. In the RR sector it is much easier to analyze the BRST cohomology and there one immediately sees that the cohomology consists of just the RR ground states (see section 4.9).
The geometric meaning of the BRST cohomology will become clear in the next section. In the remainder of this section, we collect various other technical aspects of the twisted CFT. Readers more interested in the more geometrical aspects can jump to section 5.
#### 4.4.2 Operator Cohomology
Let $`๐ช_{n^{},h,\alpha }`$ be the local operator corresponding to the state $`|\mathrm{\Delta }(n^{}),h,\alpha `$.<sup>14</sup><sup>14</sup>14Here $`\alpha `$ is a formal label that might be needed to completely specify a state. Generically, $`Q`$ does not commute with the local operators $`๐ช_{\mathrm{\Delta }(1),0},๐ช_{\mathrm{\Delta }(2),\frac{2}{5}},๐ช_{\mathrm{\Delta }(3),\frac{2}{5}}`$ and $`๐ช_{\mathrm{\Delta }(4),0}`$ corresponding to the chiral states $`|0,0,|\frac{1}{10},\frac{2}{5},|\frac{6}{10},\frac{2}{5},|\frac{3}{2},0`$ (for brevity we will denote those 4 local operators just by their tri-critical Ising model Kac index $`๐ช_i,i=1,2,3,4).`$ This is because the topological $`G_2`$ CFT is formulated not in terms of local operators of the untwisted theory but in terms of non-local conformal blocks. It is straightforward to check that the following blocks,
$$๐_n^{}=\underset{m}{}P_{n^{}+m1}๐ช_n^{}P_m$$
(4.17)
which pick out the maximal โdown componentโ of the corresponding local operator, do commute with $`Q`$ and are thus in its operator cohomology. For example writing explicitly $`Q=P_4G_{\frac{1}{2}}P_3+P_3G_{\frac{1}{2}}P_2+P_2G_{\frac{1}{2}}P_1`$ it follows trivially from the definition of the projectors $`P_IP_J=P_I\delta _{I,J}`$ that $`Q`$ commutes with $`๐_4=P_4๐ช_4P_1.`$ To get some familiarity with the notation we work out another example,
$$\begin{array}{cc}\hfill \{Q,๐_2\}& =\underset{n^{}}{}P_{n^{}+1}\left(G_{\frac{1}{2}}P_n^{}๐ช_2+๐ช_2P_n^{}G_{\frac{1}{2}}\right)P_{n^{}1}\hfill \\ & =\underset{n^{}}{}P_{n^{}+1}\left(\{G_{\frac{1}{2}},๐ช_2\}\right)P_{n^{}1}=\underset{n^{}}{}P_{n^{}+1}๐ช_{\mathrm{\Delta }(1),1}P_{n^{}1}=0\hfill \end{array}$$
(4.18)
where we repeatedly use the property 4.5 and the existence of the marginal operators 3.3. Note that we have not shown that the blocks 4.17 exhaust the $`Q`$ cohomology but presumably this is indeed the case.
This algebraic characterization of the conformal blocks corresponding to chiral primaries fits nicely with the Coulomb gas approach where the tri-critical Ising model vertex operator (i.e. block) of the chiral primaries was identified in A.12 to be exactly the unscreened vertex that created the maximal โdownโ shift in the Kac table.
### 4.5 The Chiral Ring
In a close parallel to what happens in theories with $`๐ฉ=2`$ SUSY, the conformal blocks which commute with $`Q`$ form a ring under the OPE. Due to the simplicity of the tri-critical Ising model there are in fact just two non trivial checks which are $`๐_2(z)๐_2(0)`$ and $`๐_2(z)๐_3(0)`$. For example
$$\begin{array}{cc}\hfill ๐_2(z)๐_3(0)& =P_4๐ช_2(z)P_3๐ช_3(0)P_1=P_4๐ช_2(z)๐ช_3(0)P_1=\hfill \\ & =P_4๐ช_4(0)P_1=๐_4(0).\hfill \end{array}$$
(4.19)
The second equality follows because $`P_1`$ projects on the identity and the third due to the unitarity bound 3.13 (which for chiral primaries is just the linear relation 4.16) implying that in the OPE of two chiral primaries there can be no poles and the leading regular term is automatically also a chiral primary.
### 4.6 An $`\mathrm{๐ฌ๐ฅ}(\mathrm{๐}|\mathrm{๐})`$ Subalgebra
We can construct an interesting $`sl(2|1)`$ subalgebra of the full algebra, whose commutation relations are identical to the lowest modes of the $`N=2`$ algebra. To construct this subalgebra, we define
$$G_r^{}=\underset{k}{}P_{k1}G_rP_k,G_r^{}=\underset{k}{}P_{k+1}G_rP_k,J_0=L_0\{G_{1/2}^{},G_{1/2}^{}\}.$$
(4.20)
Using properties of the $`G_2`$ algebra, and Jacobi identities, we can show that the algebra generated by $`G_{\pm 1/2}^{}`$, $`G_{\pm 1/2}^{}`$, $`L_0`$, $`L_{\pm 1}`$ and $`J_0`$ closes and forms the algebra $`sl(2|1)`$. Notice that $`QG_{1/2}^{}`$ is one of the generators of this algebra. We know that $`sl(2|1)`$ has short and long representations, and any state in the BRST cohomology must necessarily be a highest weight state of a short representation. This shows that $`sl(2|1)`$ descendants are never part of the BRST cohomology. This is a hint that the only elements of the BRST cohomology are the chiral primaries, but to prove this we would need to extend the above reasoning to include also elements which are descendants with respect to the other generators of the $`G_2`$ algebra, or require us to determine the precise form of the antighost and twisted stress tensor.
#### Position Independence of Correlators
Notice that the generators of translations on the plane, namely, $`L_1`$ and $`\stackrel{~}{L}_1`$ are BRST exact:
$$L_1=\{Q,G_{\frac{1}{2}}^{}\}$$
(4.21)
It follows that, in the topological $`G_2`$ theory, genus zero correlation functions of chiral primaries between BRST closed states are position independent. This is a crucial ingredient of topological theories.
### 4.7 A Twisted Virasoro Algebra?
Above, we constructed an $`sl(2|1)`$ algebra, and it is natural to ask if it can be extended to a full $`N=2`$ algebra. This seems unlikely, but one definitely expects to find at least all the modes of a twisted stress-tensor, which is essential for the construction of a topological string theory on higher genus Riemann surfaces. Since genus zero amplitudes are independent of the locations of the operators, this suggests that such a twisted stress tensor should indeed exist.
The construction of the $`sl(2|1)`$ algebra immediately yields a candidate for the twisted stress tensor, namely
$$\stackrel{~}{L}_m\{Q,G_{m+1/2}^{}\}\{Q,G_{m+1/2}\}.$$
(4.22)
This definition seems to work at first sight. For example,
$$\stackrel{~}{L}_1=L_1$$
(4.23)
as expected for a twisted energy-momentum tensor. In addition,
$$[\stackrel{~}{L}_1,\stackrel{~}{L}_m]=(1m)\stackrel{~}{L}_{m1},$$
(4.24)
which is the correct commutation relation for a Virasoro algebra. In addition, $`[\stackrel{~}{L}_m,\stackrel{~}{L}_m]`$ annihilates chiral primaries, as expected for a twisted energy-momentum tensor with zero central charge. However, there is no obvious reason why the other commutation relations should be valid. Some extremely tedious calculations reveal that (assuming that we did not make any mistakes in the lengthy algebra) when acting on primaries of the full $`G_2`$ algebra
$$\stackrel{~}{L}_0|\mathrm{\Delta }(k+1),h_r=\frac{4k2}{4k1}((\mathrm{\Delta }(k+1)+h_r)\frac{k}{2})|\mathrm{\Delta }(k+1),h_r$$
(4.25)
and
$`[\stackrel{~}{L}_2,\stackrel{~}{L}_2]|\mathrm{\Delta }(k+1),h_r=c_k((\mathrm{\Delta }(k+1)+h_r){\displaystyle \frac{k}{2}})\times `$ (4.26)
$`(1485+2868k+2644k^23392k^3640k^4+512k^572k(\mathrm{\Delta }(k+1)+h_r))|\mathrm{\Delta }(k+1),h_r`$
with
$$c_k=\frac{4k2}{(k+1)(2k+3)(4k11)(4k1)^2(4k+9)}.$$
(4.27)
This clearly shows that $`[\stackrel{~}{L}_2,\stackrel{~}{L}_2]4\stackrel{~}{L}_0`$. In addition, we see the shift in $`\stackrel{~}{L}_0`$ would live entirely in the tri-critical part were it not for the prefactor $`(4k2)/(4k1)`$ that appears. Having the twist purely in the tri-critical piece is appealing, as this can easily be implemented in the Coulomb gas formulation, but further work is required to prove that such a twisted energy-momentum tensor indeed exists and is BRST exact. The above proposal is apparently not quite the correct one.
### 4.8 Moduli and Descent Relations
As mentioned in section 3.1 the upper components $`\stackrel{~}{G}_{\frac{1}{2}}|\frac{1}{10},\frac{2}{5}_LG_{\frac{1}{2}}|\frac{1}{10},\frac{2}{5}_R`$ where shown in to be exactly marginal deformations of the CFT preserving the $`G_2`$ chiral algebra. We also saw that they are in one-to-one correspondence with the $`b_3`$ metric moduli of the $`G_2`$ manifold. Once we include the $`B`$-field the number of such moduli will turn out to be $`b_2+b_3`$ as we will see in section 5.2. Since both the ordinary and the topologically-twisted theories should exist on an arbitrary manifold of $`G_2`$ holonomy it is important to check that the moduli space of deformations of the two theories agrees. So far we have seen that the interesting objects in the twisted theory are given in terms of non local objects of the original one. We will now demonstrate that nevertheless the two theories have the same moduli space of deformations. In a fashion identical to 2.12 we can split the local field $`๐ช_2`$ that creates the chiral primary state $`|\frac{1}{10},\frac{2}{5}`$ as
$$๐ช_2=๐ช_2^{}+๐ช_2^{}=\underset{m}{}P_{m+1}๐ช_2P_m+\underset{m}{}P_{m1}๐ช_2P_m.$$
(4.28)
The first term coincides with $`๐_2`$ which corresponds to a chiral operator in the twisted theory so in particular $`\{Q,๐_2\}=0.`$ Also, a computation similar to 4.18 shows that $`\{G_{\frac{1}{2}}^{},๐ช_2^{}\}=0.`$ Using this we compute
$$\begin{array}{cc}\hfill [Q,\{G_{\frac{1}{2}},๐ช_2\}]& =[Q,\{G_{\frac{1}{2}}^{}+G_{\frac{1}{2}}^{},๐ช_2^{}+๐ช_2^{}\}]\hfill \\ & =[Q,\{Q,๐ช_2^{}\}]+[Q,\{G_{\frac{1}{2}}^{},๐_2\}]\hfill \\ & =[Q,\{G_{\frac{1}{2}}^{},๐_2\}]\hfill \\ & =[\{Q,G_{\frac{1}{2}}^{}\},๐_2]\hfill \\ & =[L_1,๐_2]=๐_2.\hfill \end{array}$$
(4.29)
In other words, we showed that $`๐_2=\{Q,\mathrm{something}\}`$, and the something is the $`(1,0)`$-form $`\{G_{1/2},๐ช_2\}`$. This is a conventional operator that does not involve any projectors. If we combine this also with the right-movers, we find that the deformations in the action of the topological string are exactly the same as the deformations of the non-topological string.
### 4.9 The Ramond Sector
We have previously given evidence, though no rigorous proof, that the cohomology in the NS sector of $`G_{1/2}^{}`$ is given by the chiral primaries. In the R sector the situation is somewhat different. There is an obvious candidate for a BRST operator in the R sector, namely $`Q=G_0^{}`$. Perhaps this is an even better candidate, as it is the zero-mode of a field (as it should be in a twisted theory), and because our twisting essentially boils down to doing computations in the R sector. It is not immediately clear that there is an easy map between the action of $`G_0^{}`$ in the R sector and the action of $`G_{1/2}^{}`$ in the NS sector. This would require us to have a suitable isomorphism between the NS and R sector. Such an isomorphism does exist and is sometimes referred to as spectral flow (discussed more in appendix F), however it is not at all clear that this maps $`G_{1/2}^{}`$ to $`G_0^{}`$. It does however map R ground states to chiral primaries, so this is further evidence that the BRST cohomology in the NS sector consists of chiral primaries and nothing else.
As an aside, notice that in the NS sector we found an $`sl(2|1)`$ subalgebra using some of the modes of $`G^{}`$ and $`G^{}`$. In the R sector this is no longer the case. In the R sector the only easy calculation we can readily do is that
$$\{G_0^{},G_0^{}\}=L_0\frac{7}{16}.$$
(4.30)
This in particular implies that the $`G_0^{}`$ cohomology is given by the R ground states. This is an exact statement. Therefore, $`G_0^{}`$ looks like an excellent candidate BRST operator. It also has the nice property that the right hand side of (4.30) is the most natural definition of $`L_0^{\mathrm{twisted}}`$ in the R sector in contrast to the situation in the NS sector.
### 4.10 Localization
It can be shown quite generally that the path integral localizes to fixed points of the BRST symmetry. For the usual case of the A and B model, this implies that only holomorphic and constant maps contribute, respectively. To derive a similar statement for the topological $`G_2`$ sigma model, we start by writing the action as
$`S`$ $`=`$ $`{\displaystyle d^2z\frac{1}{2}g_{IJ}x^I\overline{}x^J}+g_{IJ}\left(i\psi _L^JD\psi _L^I+i\psi _L^JD\psi _L^I+i\psi _R^J\overline{D}\psi _R^I+i\psi _R^J\overline{D}\psi _R^I\right)`$
$`+R_{IJKL}\psi _R^I\psi _R^J\psi _L^K\psi _L^L`$
This action has the fermionic symmetry
$`\delta x^I`$ $`=`$ $`iฯต_L\psi _L^I+iฯต_R\psi _R^I`$
$`\delta \psi _L^I`$ $`=`$ $`ฯต_Lx^Jฯต_R\psi _R^K\mathrm{\Gamma }_{KM}^I\psi _L^M`$
$`\delta \psi _L^I`$ $`=`$ $`ฯต_R\psi _R^K\mathrm{\Gamma }_{KM}^I\psi _L^M`$
$`\delta \psi _R^I`$ $`=`$ $`ฯต_L\psi _L^K\mathrm{\Gamma }_{KM}^I\psi _R^M`$
$`\delta \psi _R^I`$ $`=`$ $`ฯต_R\overline{}x^Jฯต_L\psi _L^K\mathrm{\Gamma }_{KM}^I\psi _R^M`$
The fixed points of this symmetry satisfy $`x^I=\overline{}x^I=0`$, which implies that the path integral localizes on constant maps. Of course, we should take this analysis with a grain of salt: the decomposition of the world sheet fermions $`\psi ^I`$ into conformal blocks $`\psi ^I+\psi ^I`$ is inherently quantum mechanical and hence it is problematic to use this decomposition in path integral arguments. Nevertheless, we take this argument as at least suggestive that we are localizing on constant maps.
## 5 Relation to Geometry
For a $`G_2`$ manifold, differential forms of any degree can be decomposed into irreducible representations of $`G_2`$
$`\mathrm{\Lambda }^0=\mathrm{\Lambda }_1^0`$ $`\mathrm{\Lambda }^1=\mathrm{\Lambda }_7^1`$
$`\mathrm{\Lambda }^2=\mathrm{\Lambda }_7^2\mathrm{\Lambda }_{14}^2`$ $`\mathrm{\Lambda }^3=\mathrm{\Lambda }_1^3\mathrm{\Lambda }_7^3\mathrm{\Lambda }_{27}^3`$
This is described in more detail in Appendix D. In a similar spirit as Hodge theory, this decomposes the cohomology groups as $`H^p=_{}H_{}^p(M)`$ where the sum is over $`G_2`$ representations $``$. The cohomology turns out to depend solely on the representation $``$ and not on the degree $`p`$ . For a proper compact $`G_2`$ manifold, $`H^1(M)=0`$ and so there is no cohomology in the seven dimensional representation of $`G_2`$. Also, $`b_1^3=1`$, corresponding to a unique closed three form $`\varphi `$ which defines the $`G_2`$ structure. There are only two independent Betti numbers left unknown, namely $`b_{14}^2`$ which is equal to the usual second Betti number $`b_2`$ and $`b_{27}^3=b_31`$ with no known restrictions on these numbers.
### 5.1 Dolbeault Complex for $`๐_\mathrm{๐}`$-Manifolds
It is possible to define a refinement of the de Rham complex, in a spirit somewhat similar to Dolbeault cohomology, as follows:
$$0\mathrm{\Lambda }_1^0\stackrel{\stackrel{ห}{D}}{}\mathrm{\Lambda }_7^1\stackrel{\stackrel{ห}{D}}{}\mathrm{\Lambda }_7^2\stackrel{\stackrel{ห}{D}}{}\mathrm{\Lambda }_1^30$$
(5.1)
where $`\stackrel{ห}{D}`$ is the usual exterior derivative when acting on 0-forms, but is the composition of the exterior derivative and projection to the 7 and 1 representations of $`G_2`$ when acting on 1 and 2 forms respectively:
$`\stackrel{ห}{D}(\alpha )`$ $`=`$ $`\pi _7^2(d\alpha )\mathrm{for}\alpha \mathrm{\Lambda }^1`$
$`\stackrel{ห}{D}(\beta )`$ $`=`$ $`\pi _1^3(d\beta )\mathrm{for}\beta \mathrm{\Lambda }^2`$
where the projection operators $`\pi _r^p`$ are defined in appendix D. In local coordinates, these expressions become
$`\left(\stackrel{ห}{D}(\alpha )\right)_{\mu \nu }dx^\mu dx^\nu `$ $`=`$ $`3_{[\mu }A_{\nu ]}\varphi _\rho ^{\mu \nu }\varphi _{\eta \chi }^\rho dx^\eta dx^\chi \alpha =A_\mu dx^\mu `$
$`\left(\stackrel{ห}{D}(\beta )\right)_{\mu \nu \rho }dx^\mu dx^\nu dx^\rho `$ $`=`$ $`_{[\xi }B_{\eta \chi ]}\varphi ^{\xi \eta \chi }\varphi _{\mu \nu \rho }dx^\mu dx^\nu dx^\rho \beta =B_{\mu \nu }dx^\mu dx^\nu `$
We will next see that the cohomology of this differential complex maps to the BRST cohomology in the left (or right) moving sector. The differential operator $`\stackrel{ห}{D}`$ maps to the BRST operator $`G_{\frac{1}{2}}^{}`$. This gives a nice and natural geometric meaning to the BRST operator, and clearly shows we are on the right track.
### 5.2 The BRST Cohomology Geometrically
In the previous section, we argued that the BRST cohomology consists of the chiral primary operators of our conformal field theory. We now proceed to study the sigma model description of these operators and the geometric meaning of the chiral ring.
To determine whether an operator corresponds to a chiral primary, we need to find its $`L_0`$ and $`X_0`$ quantum numbers. Often in topological theories, this calculation can be reduced to operators built out of non-derivative fields only. In our case we also expect this to be the case, since all elements in the cohomology are in one-to-one correspondence to R ground states. Also, the argument that the path integral localizes on constant maps indicates that only zero modes appear.
So we proceed by analyzing the action of the BRST operator at the level of operators that do not contain any derivatives of fields. In the left-moving sector, such operators are in one-to-one correspondence with $`p`$-forms on the target space:
$$\omega _{i_1,\mathrm{},i_p}dx^{i_1}\mathrm{}dx^{i_p}\omega (x^\mu )_{i_1,\mathrm{},i_p}\psi ^{i_1}\mathrm{}\psi ^{i_p}.$$
(5.2)
The same is obviously also true in the right-moving sector, but for simplicity we analyze the left-moving sector first.
The group $`G_2`$ acts on the tangent space of the manifold, and the space of $`p`$-forms at a point can be decomposed in $`G_2`$ representations as explained above. Since $`X_0`$ and $`L_0`$ are $`G_2`$ singlets, they take the same value in each of these representations. Some further explicit calculations <sup>15</sup><sup>15</sup>15As an example, we determine the $`X_0`$ eigenvalue of the operator $`A(X)_\mu \psi ^\mu `$ which corresponds to the one form $`A(X)_\mu dx^\mu `$. Using the expression for $`X(z)`$ in (2.7), the $`X_0`$ eigenvalue is given by the coefficient of the second order pole in the OPE
$$X(z).\left(A(X)_\mu \psi ^\mu (0)\right)\mathrm{}\frac{1}{2}\frac{A(X)_\mu \psi ^\mu }{z^2}+\mathrm{}$$
(5.3) which gives the $`X_0`$ eigen-value of this operator to be $`\frac{1}{2}`$ and the tri-critical Ising model weight $`\frac{1}{10}`$. involving the precise form of $`X_0`$ then reveal that the quantum numbers associated to each representation are
$$\begin{array}{ccccc}& & & & \\ & \mathrm{๐}& \mathrm{๐}& \mathrm{๐๐}& \mathrm{๐๐}\\ & & & & \\ p=0& |0,0& & & \\ & & & & \\ p=1& & |\frac{1}{10},\frac{2}{5}& & \\ & & & & \\ p=2& & |\frac{6}{10},\frac{2}{5}& |0,1& \\ & & & & \\ p=3& |\frac{3}{2},0& |\frac{11}{10},\frac{2}{5}& & |\frac{1}{10},\frac{7}{5}\\ & & & & \\ p=4& |2,0& |\frac{16}{10},\frac{2}{5}& & |\frac{6}{10},\frac{7}{5}\\ & & & & \\ p=5& & |\frac{21}{10},\frac{2}{5}& |\frac{3}{2},1& \\ & & & & \\ p=6& & |\frac{26}{10},\frac{2}{5}& & \\ & & & & \\ p=7& |\frac{7}{2},0& & & \end{array}$$
(5.4)
This table also nicely reflects the two maps which take a $`p`$-form $`\omega `$ into a $`p+3`$ form given by $`\omega \varphi `$ and into a $`p+4`$ form $`\omega \varphi `$ (see appendix D). When restricted to $`G_2`$ representations, these operators are either identically zero or act as isomorphisms. They translate to the action of $`\mathrm{\Phi }_{\frac{3}{2}}`$ and $`X_2`$ at the level of states. Notice that chiral primaries appear only in four places in (5.4), and precisely those differential forms enter into (5.1). Of course, this is not a coincidence, as we will see below.
In order to construct the precise form of these states, we need to project the relevant forms on to appropriate $`G_2`$ representation. All such projectors can be constructed in terms of the three-form $`\varphi `$ and its Hodge dual, as explained in Appendix D. To find their precise form, various identities satisfied by $`\varphi `$ are useful, such as
$`\varphi _c{}_{}{}^{de}\varphi _{de}^{}^f`$ $`=`$ $`{\displaystyle \frac{1}{6}}\delta _c^f`$
$`\varphi _{ab}{}_{}{}^{cd}\varphi _{cd}^{}^e`$ $`=`$ $`{\displaystyle \frac{1}{6}}\varphi _{ab}^e`$
$`\varphi _{ab}{}_{}{}^{c}\varphi _{c}^{}^{de}`$ $`=`$ $`{\displaystyle \frac{2}{3}}\varphi _{ab}{}_{}{}^{de}+{\displaystyle \frac{1}{36}}(\delta _a^d\delta _b^e\delta _a^e\delta _b^d)`$
$`\varphi _{ab}{}_{}{}^{cd}\varphi _{cd}^{}^{ef}`$ $`=`$ $`{\displaystyle \frac{1}{12}}\varphi _{ab}{}_{}{}^{ef}+{\displaystyle \frac{1}{144}}(\delta _a^e\delta _b^f\delta _a^f\delta _b^e)`$
$`\varphi _{abc}\varphi ^{abc}`$ $`=`$ $`{\displaystyle \frac{7}{6}}`$
$`\varphi _{[ab}{}_{}{}^{cd}\varphi _{e]cd}^{}`$ $`=`$ $`\varphi _{abe}`$
$`{\displaystyle \frac{1}{2}}\varphi _{[ab}{}_{}{}^{cd}\varphi _{e]cd}^{}^f`$ $`=`$ $`{\displaystyle \frac{1}{4}}\varphi _{abe}{}_{}{}^{f}.`$ (5.5)
In these equations, antisymmetrization over $`n`$ indices does not include a factor of $`1/n!`$. They are also useful in order to compute the $`X_0`$ eigenvalue in each representation. Notice, however, that the exact quantum $`X_0`$ eigenstates can not in general be written in terms of fields without derivatives, typically one needs to add some quantum corrections involving fewer fermions and a few derivatives as well.
This table allows us to extract the precise action of the BRST operator on the operators that do not involve derivatives. For example,
$$G_{1/2}A_\mu (X)\psi ^\mu =\frac{1}{2}_{[\nu }A_{\mu ]}\psi ^\nu \psi ^\mu +A_\mu (X)X^\mu .$$
(5.6)
In the calculation we get a covariant derivative, however this is equal to the ordinary derivative when acting on forms as an exterior derivative. To extract the action of $`G_{1/2}^{}`$, we first observe the second term has $`X_0=0`$ and therefore only contributes to $`G_{1/2}^{}`$. The first term has a part transforming in the $`\mathrm{๐}`$ of $`G_2`$ and a part transforming in the $`\mathrm{๐๐}`$ of $`G_2`$, and according to (5.4) we need to project on the $`\mathrm{๐}`$ to obtain the action of $`G_{1/2}^{}`$. The relevant projection operator is $`P_{ab}{}_{}{}^{de}=6\varphi _{ab}{}_{}{}^{c}\varphi _{c}^{}^{de}`$, and we finally get
$$G_{1/2}^{}A_\mu (X)\psi ^\mu =3_{[\nu }A_{\mu ]}\varphi ^{\nu \mu }{}_{\rho }{}^{}\varphi _{}^{\rho }{}_{\alpha \beta }{}^{}\psi _{}^{\alpha }\psi ^\beta .$$
(5.7)
It is clear by inspection of table (5.4) that chiral primaries, i.e. non-trivial elements of the BRST cohomology, can either be singlet 0- or 3-forms, or 1- or 2-forms transforming in the $`\mathrm{๐}`$ of $`G_2`$.
By repeating (5.7) for the two form $`B_{\mu \nu }\psi ^\mu \psi ^\nu `$ and the three form $`\varphi _{\mu \nu \alpha }\psi ^\mu \psi ^\nu \psi ^\alpha `$, the kernel of $`Q_{\mathrm{BRST}}`$ in the left-moving sector is then seen to consist of
$`1`$
$`A_\mu \psi ^\mu `$ $`\mathrm{with}`$ $`\varphi _\rho {}_{}{}^{\mu \nu }_{[\mu }^{}A_{\nu ]}=0`$
$`B_{\mu \nu }\psi ^\mu \psi ^\nu `$ $`\mathrm{with}`$ $`\varphi ^{\rho \mu \nu }_{[\rho }B_{\mu \nu ]}=0`$
$`\varphi _{\mu \nu \rho }\psi ^\mu \psi ^\nu \psi ^\rho .`$ (5.8)
We should still remove the image of $`G_{1/2}^{}`$, which means identifying for example
$$A_\mu A_\mu +_\mu C$$
(5.9)
and
$$B_{\alpha \beta }B_{\alpha \beta }+3_{[\nu }D_{\mu ]}\varphi ^{\nu \mu }{}_{\rho }{}^{}\varphi _{}^{\rho }_{\alpha \beta }$$
(5.10)
for arbitrary $`C`$, $`D_\mu `$.
It is interesting to note that the BRST cohomology in the left moving sector is just the Dolbeault type cohomology of the $`\stackrel{ห}{D}`$ operator that we defined in the previous subsection. The BRST operator $`G_{1/2}^{}`$ naturally maps to the operator $`\stackrel{ห}{D}`$. In fact, the table 5.4 reveals the existence of two other differential complexes. One of these is related to the complex in (5.1) by the Hodge duality. The other one is a new complex
$$0\mathrm{\Lambda }_{14}^2\stackrel{\stackrel{~}{D}}{}\mathrm{\Lambda }_7^3\mathrm{\Lambda }_{27}^3\stackrel{\stackrel{~}{D}}{}\mathrm{\Lambda }_7^4\mathrm{\Lambda }_{27}^4\stackrel{\stackrel{~}{D}}{}\mathrm{\Lambda }_{14}^50$$
(5.11)
where the differential operator $`\stackrel{~}{D}`$ is the composition of the ordinary exterior derivative with appropriate projection operators (defined in appendix D). This new complex does not consist of chiral primaries and does not seem to play any role in the twisted theory we are considering, but it would still be interesting to know whether it has a distinguished geometric interpretation.
If we do not combine left and right movers, the cohomology is almost trivial. As we noted earlier, compact $`G_2`$ manifolds have $`b_1=0`$ and therefore there is no cohomology in the seven-dimensional representation of $`G_2`$. As a consequence, only the identity and the three-form survive if we do not include right-movers.
However, once we combine left- and right-movers, we obtain a more interesting cohomology. The two-form $`B`$ and one-form $`A`$ are in one-to-one correspondence via $`B_{\mu \nu }=\varphi _{\mu \nu }{}_{}{}^{\alpha }A_{\alpha }^{}`$ so it is sufficient to consider only the combination of the left- and right moving one-forms. Each of them transforms in the $`\mathrm{๐}`$ of $`G_2`$, and $`\mathrm{๐}\mathrm{๐}=\mathrm{๐}+\mathrm{๐}+\mathrm{๐๐}+\mathrm{๐๐}`$. We get one non-trivial class from $`\mathrm{๐}`$, none from $`\mathrm{๐}`$, $`b_2`$ from $`\mathrm{๐๐}`$ and $`b_31`$ from $`\mathrm{๐๐}`$. In total, we get $`b_2+b_3`$, corresponding to the non-trivial $`B`$-field and metric deformations of the $`G_2`$ manifold. This is indeed the set of moduli that we expect to find in a topological theory. If we replace the left or right movers by a two-form, these results do not change. We also get a contribution to the cohomology from the left-moving zero/three form times the right-moving zero/three form. The total cohomology is
$`0\mathrm{form}\times \mathrm{\hspace{0.17em}\hspace{0.17em}0}\mathrm{form}`$ $``$ $`b_0`$
$`1\mathrm{form}\times \mathrm{\hspace{0.17em}\hspace{0.17em}1}\mathrm{form}`$ $``$ $`b_2+b_3`$
$`2\mathrm{form}\times \mathrm{\hspace{0.17em}\hspace{0.17em}2}\mathrm{form}`$ $``$ $`b_4+b_5`$
$`3\mathrm{form}\times \mathrm{\hspace{0.17em}\hspace{0.17em}3}\mathrm{form}`$ $``$ $`b_7`$ (5.12)
plus another copy of this if we allow the left and right levels not to match each other. Either way, we get one or two copies of the full cohomology $`H^{}(M)`$ of $`M`$.
We can verify whether we recover known results about the metric moduli of $`G_2`$ manifolds. According to the above, metric and $`B`$-field moduli should be given by operators of the form
$$(\delta g_{\mu \nu }+\delta B_{\mu \nu })\psi _R^\mu \psi _L^\nu $$
(5.13)
with
$$\varphi _\alpha {}_{}{}^{\lambda \mu }(_{[\lambda }\delta g_{\mu ]\nu }+_{[\lambda }\delta B_{\mu ]\nu })=0.$$
(5.14)
Metric moduli are indeed known to satisfy this equation (eq. 3.7) as pointed out in . To verify that $`B`$-moduli also satisfy (5.14), we first use the fact that $`\varphi `$ is covariantly constant to rewrite
$$\varphi _\alpha {}_{}{}^{\lambda \mu }(_{[\lambda }\delta B_{\mu ]\nu })=_{[\lambda }(\delta B_{\mu ]\nu }\varphi _\alpha {}_{}{}^{\lambda \mu }).$$
(5.15)
Since $`B`$-moduli transform in the $`\mathrm{๐๐}`$ of $`G_2`$, they also obey (see appendix D)
$$\delta B_{\lambda \mu }\varphi _\alpha {}_{}{}^{\lambda \mu }=0.$$
(5.16)
We can therefore replace the rhs of (5.15) by
$$_{[\lambda }(\delta B_{\mu \nu ]}\varphi _\alpha {}_{}{}^{\lambda \mu })=_{[\lambda }\delta B_{\mu \nu ]}\varphi _\alpha {}_{}{}^{\lambda \mu }=0$$
(5.17)
since $`B`$-moduli are closed two-forms. This shows that the $`B`$-moduli also satisfy (5.14) and the BRST cohomology consists exactly of the metric and the $`B`$-field moduli.
### 5.3 Correlation Functions
In this section we explicitly compute some simple correlation functions in the $`G_2`$ sigma model by working in the classical, large volume approximation.
As we discussed already, the operator cohomology contains only operators that map $`_i`$ to $`_j`$ with $`ij`$. Therefore only a finite set of correlation functions will be nonzero. Letโs first consider the left-movers only, and consider a three-point function of three operators $`๐ช_k=A_\mu ^k\psi ^\mu `$, with $`k=1,2,3`$, and we assume each to be in the BRST cohomology. This boils down to the calculation of
$$V_{\frac{7}{16},+}๐ช_1๐ช_2๐ช_3V_{\frac{7}{16},+}$$
(5.18)
in the untwisted theory. This object turns out to be a 4-point function in the R-sector
$$\mathrm{\Phi }_0๐ช_1๐ช_2๐ช_3_R$$
(5.19)
because $`V_{\frac{7}{16},+}^{}=V_{\frac{7}{16},}=\mathrm{\Phi }_0V_{\frac{7}{16},+}`$. The operator $`\mathrm{\Phi }`$ is $`\varphi _{\alpha \beta \gamma }\psi ^\alpha \psi ^\beta \psi ^\gamma `$, and from the contractions we obtain for the correlator something proportional to
$$\varphi _{\alpha \beta \gamma }g^{\alpha \mu }g^{\beta \nu }g^{\gamma \rho }A_\mu ^1A_\nu ^2A_\rho ^3.$$
(5.20)
The inverse metrics arise due to the fact that in this approximation the fermion two-point function is proportional to the inverse metric.
Combining left and right movers, relabelling everything in terms of metric and $`B`$-field moduli, and including an integral over the seven manifold from the zero mode of $`X^\mu `$, we finally obtain for the three-point function for metric and $`B`$-field moduli
$$_{3\mathrm{point}}=_Md^7x\sqrt{g}\varphi _{\alpha \beta \gamma }(\delta _1g^{\alpha \alpha ^{}}+\delta _1b^{\alpha \alpha \mathrm{`}})(\delta _2g^{\beta \beta ^{}}+\delta _2b^{\beta \beta \mathrm{`}})(\delta _3g^{\gamma \gamma ^{}}+\delta _3b^{\gamma \gamma \mathrm{`}})\varphi _{\alpha ^{}\beta ^{}\gamma ^{}}.$$
(5.21)
To analyze this expression a bit further, we drop the $`B`$-field moduli. In addition, we will take a suitable set of coordinates $`t_i`$ on the moduli space of $`G_2`$ metrics, and denote by $`Y_i`$ the operator corresponding to sending $`t_it_i+\delta t_i`$. In other words, the three-point function reads
$$Y_iY_jY_k=_Md^7x\sqrt{g}\varphi _{\alpha \beta \gamma }\frac{g^{\alpha \alpha ^{}}}{t_i}\frac{g^{\beta \beta ^{}}}{t_j}\frac{g^{\gamma \gamma ^{}}}{t_k}\varphi _{\alpha ^{}\beta ^{}\gamma ^{}}.$$
(5.22)
One might expect, based on general arguments (see e.g. ), that this is the third derivative of some prepotential if suitable โflatโ coordinates are used. For example, consider the manifold $`M=T^7`$ and choose coordinates such that $`\varphi `$ is linear in them. We find that
$$Y_iY_jY_k=\frac{1}{21}\frac{^3}{t_it_jt_k}\varphi \varphi .$$
(5.23)
This strongly suggests that the same results should also be valid on general $`G_2`$ manifolds. In fact, in the next subsection, we will develop a version of โspecial geometryโ for $`G_2`$ manifolds and show that with an appropriate definition of flat coordinates for the moduli space of $`G_2`$ metrics, the three point function can be written as in (5.23)
The action
$$S=\varphi \varphi $$
(5.24)
also appears in , where it was shown that the critical points of this functional, viewed as a functional on the space of three-forms in a given cohomology class, are precisely the three-forms of $`G_2`$ manifolds. It was also the starting point of topological M-theory in , see also . It is tempting to speculate that our topological $`G_2`$ string provides the framework to quantize topological M-theory, which by itself is not yet a well-defined quantum theory.
### 5.4 $`๐_\mathrm{๐}`$ Special Geometry
To prove in full generality a relation between our topological three point function and the Hitchin functional we need to develop a version of โspecial geometryโ for $`G_2`$ manifolds.
First of all we define
$$=\varphi \varphi ,$$
(5.25)
which will be a functional on the space of $`G_2`$ metrics (or on the space of the corresponding three-forms).
The most natural choice for flat coordinates, as our torus example also suggests, is to choose periods, as we do in the case of the six-dimensional topological string. We thus pick a symplectic basis of homology three-cycles $`C_A`$ and dual four cycles $`D^A`$, and define coordinates on the moduli space of $`G_2`$ metrics as
$$t^A=_{C_A}\varphi .$$
(5.26)
For the dual periods we introduce the notation
$$F_A=_{D^A}\varphi .$$
(5.27)
It is perhaps tempting to write
$$\varphi =t^A\chi _A$$
(5.28)
with $`\chi _A`$ a basis of three forms Poincare dual to the four-cycles $`D_A`$. This is not quite correct as the detailed form of $`\varphi `$ will in general differ from (5.28) by an exact three-form. In most calculations, this exact three-form drops out, but it is important to keep in mind that $`\varphi `$ cannot simply be expanded linearly in a given basis of cohomology.
Continuing, we can also write $`F_A`$ as
$$F_A=\varphi _A\varphi $$
(5.29)
Furthermore, by a generalization of the Riemann bilinear identities we find that
$$=t^AF_A.$$
(5.30)
Letโs now take one derivative of $``$. We readily obtain
$$_B=F_B+t^A_BF_A.$$
(5.31)
We can also perform straightforward explicit computations by using the canonical expressions for $`\varphi `$ and $`\varphi `$ in local coordinates:
$`\varphi `$ $`=`$ $`dx^{123}+dx^{145}+dx^{167}+dx^{246}dx^{257}dx^{347}dx^{356}`$ (5.32)
$`\varphi `$ $`=`$ $`dx^{4567}+dx^{2367}+dx^{2345}+dx^{1357}dx^{1346}dx^{1256}dx^{1247}`$ (5.33)
Here, $`dx^{ijk}=e^ie^je^k`$, with $`e^i=e_\mu ^idx^\mu `$ a local orthonormal frame, i.e. a set of vielbeins in which the metric becomes $`g_{\mu \nu }=e_\mu ^ie_\nu ^i`$. To find the variation of various quantities with respect to $`t^A`$, we will need to vary the vielbeins. We notice that up to $`SO(7)`$ rotations rotating the $`e^i`$ into each other
$$_Ae_\mu ^a=\frac{1}{2}_Ah_{\mu \nu }h^{\nu \lambda }e_\lambda ^a.$$
(5.34)
Then, using the explicit expressions for $`\varphi `$ and $`\varphi `$ in terms of the vielbeins, we find
$$F_A=\varphi _A\varphi =\frac{3}{2}_Ah_{\mu \nu }h^{\mu \nu }$$
(5.35)
and
$$t^B_AF_B=_A\varphi \varphi =2_Ah_{\mu \nu }h^{\mu \nu }$$
(5.36)
which implies that
$$t^A_BF_A=\frac{4}{3}F_B$$
(5.37)
We conclude from (5.31) and (5.37) that
$$F_B=\frac{3}{7}_B.$$
(5.38)
Thus we see that $``$ is homogeneous of degree $`7/3`$ in the coordinates $`t^A`$, which can also easily be verified explicitly, but more importantly we have found that the dual periods are the derivatives of a single function, the prepotential $`F`$, which is given by
$$F=\frac{3}{7}.$$
(5.39)
Next we turn to the second derivative of $``$. From the above we readily obtain
$$_A\varphi _B\varphi =_B\varphi _A\varphi =3\varphi _A_B\varphi =\frac{3}{7}_A_B.$$
(5.40)
We can evaluate the first expression most easily, by varying the vielbeins that appear in the standard expression for $`\varphi `$ and $`\varphi `$, and by counting the resulting terms. We find
$$_A\varphi _B\varphi =\frac{1}{2}\sqrt{g}(_Ah_{\mu \nu }h^{\mu \nu }_Bh_{\rho \sigma }h^{\rho \sigma }_Ah_{\mu \nu }h^{\nu \rho }_Bh_{\rho \sigma }h^{\sigma \mu }).$$
(5.41)
On the other hand, by using the third identity in (5.5) we deduce
$$\sqrt{g}\varphi _{abc}_Ah^{aa^{}}_Bh^{bb^{}}h^{cc^{}}\varphi _{a^{}b^{}c^{}}=\frac{1}{36}\sqrt{g}(_Ah_{\mu \nu }h^{\mu \nu }_Bh_{\rho \sigma }h^{\rho \sigma }_Ah_{\mu \nu }h^{\nu \rho }_Bh_{\rho \sigma }h^{\sigma \mu }).$$
(5.42)
Combining (5.40), (5.41) and (5.42) we finally obtain
$$\sqrt{g}\varphi _{abc}_Ah^{aa^{}}_Bh^{bb^{}}h^{cc^{}}\varphi _{a^{}b^{}c^{}}=\frac{1}{42}_A_B.$$
(5.43)
Therefore, the second derivatives of $``$ closely resemble the expression for the three-point function we obtained from the topological string.
Turning finally to the third derivative, this analysis is a bit more tedious. In analogy with (5.40) we have
$$_A\varphi _B_C\varphi =\frac{3}{2}\varphi _A_B_C\varphi =\frac{3}{7}_A_B_C.$$
(5.44)
The first expression is again the most useful one to manipulate, and we do this as before in terms of a representation in a local orthonormal flat frame (i.e. vielbeins). We again use the variation of the vielbein as given in (5.34). We find a new feature, namely we now also will run into double derivatives of the metric, due to the double derivative acting on $`\varphi `$ in the first expression in (5.44). We can get rid of this double derivative as follows. We write $`_{BC}`$ for the double derivative acting on a single vielbeins only. Then it is easy to see that
$$_A\varphi _{BC}\varphi =_{BC}\varphi _A\varphi .$$
(5.45)
Now notice that $`_B_C=_{BC}+_{BC}^{}`$, where $`_{BC}^{}`$ is defined such that the two derivatives never act on the same vielbein. Thus, for example,
$`_{BC}e^1e^2`$ $``$ $`_B_Ce^1e^2+e^1_B_Ce^2`$
$`_{BC}^{}e^1e^2`$ $``$ $`_Be^1_Ce^2+_Ce^1_Be^2`$ (5.46)
and clearly these two add up to $`_B_C`$. Because $`\varphi `$ is linear in $`t^A`$ (5.28), we can replace in (5.45) $`_{BC}\varphi =_B_C\varphi _{BC}^{}\varphi =_{BC}^{}\varphi `$. So we obtain,
$$_A_B_C=\frac{7}{3}\left(_A\varphi _{BC}^{}\varphi _{BC}^{}\varphi _A\varphi \right).$$
(5.47)
In this expression no double derivatives of the metric appear anymore. However it contains a priori all kinds of contractions of the three single derivatives of the metric. To determine the detailed form of the result, we took (5.47), wrote $`\varphi `$ in terms of $`\varphi `$ using the seven-dimensional completely antisymmetric $`ฯต`$ tensor, and expanded (5.47) in terms of all possible contractions that can appear. After a significant amount of tedious algebra we found, quite surprisingly, that almost all terms cancel, and that we are left with the simple final result
$$_A_B_C=21\sqrt{g}\varphi _{abc}_Ah^{aa^{}}_Bh^{bb^{}}_Ch^{cc^{}}\varphi _{a^{}b^{}c^{}}.$$
(5.48)
This proves that our topological three-point function is indeed the third derivative of a single function, which is precisely the Hitchin functional, viewed as a function on the space of $`G_2`$ metrics! Notice that (5.48) is valid both for the rather trivial modulus which corresponds to rescaling $`\varphi `$, as well as for the $`b_31`$ moduli which live in the $`\mathrm{๐๐}`$ of $`G_2`$. For the latter moduli an expression similar to (5.48) was written down in , where it was used to describe fibrations of $`G_2`$ manifolds by coassociative submanifolds. These three-point functions were called Yukawa couplings in that paper, though the relation with the physical Yukawa couplings in M-theory was not given. Our results shows that the cubic coupling (5.48), which is the topological three-point function, is indeed closely related to the physical Yukawa couplings that one obtains in compactifying M-theory on $`G_2`$ manifolds. This is because the Kรคhler potential of the resulting four-dimensional theory is essentially the logarithm of $``$, and Yukawa couplings are given by the third derivative of the Kรคhler potential. A more detailed discussion can be found in section 7.3.
### 5.5 Inclusion of the $`๐`$-field
We next want to see what happens when we include the $`B`$-field. There is only one relevant correlator
$$\sqrt{g}\varphi _{abc}_pB^{aa^{}}_qB^{bb^{}}_Ch^{cc^{}}\varphi _{a^{}b^{}c^{}},$$
(5.49)
since the correlators involving one or three $`B`$-field insertions vanish identically due to symmetry/anti-symmetry properties of the index contractions. We introduced coordinates $`s^p`$ on the space $`H^2(M)`$ of $`B`$-fields, but still need to specify how they are defined. To simplify the above expression, we first observe that since $`B^{bb^{}}`$ lives in the $`\mathrm{๐๐}`$ of $`G_2`$ (the $`B`$-field is a closed two form and the only non-trivial second cohomology transforms as in the $`\mathrm{๐๐}`$ dimensional representation of $`G_2`$), which means $`\varphi _{abb^{}}B^{bb^{}}=0`$. Therefore, we can antisymmetrize over $`a,a^{},b,b^{}`$ in the above expression so that it becomes
$$\frac{1}{24}\sqrt{g}\varphi _{c[ab}\varphi _{a^{}b^{}]c^{}}_pB^{aa^{}}_qB^{bb^{}}_Ch^{cc^{}}.$$
(5.50)
Next, we can use the following identity
$$\varphi _{a[bc}\varphi _{b^{}c^{}]a^{}}=\frac{4}{9}g_{a[b}\varphi _{cb^{}c^{}]a^{}}\frac{4}{9}g_{a^{}[b}\varphi _{cb^{}c^{}]a}\frac{2}{9}\delta _{aa^{}}\varphi _{[bcb^{}c^{}]}$$
(5.51)
which we can prove in a local orthonormal frame. Inserting (5.51) into (5.50) leads to
$$\sqrt{g}\varphi _{abc}_pB^{aa^{}}_qB^{bb^{}}_Ch^{cc^{}}\varphi _{a^{}b^{}c^{}}=\frac{1}{9}\frac{^3}{t^Cs^ps^q}\sqrt{g}\varphi ^{abcd}B_{ab}B_{cd}$$
(5.52)
where it is crucial that we choose our coordinates $`s^a`$ such that the periods of $`BB`$ along all four-cycles are purely quadratic expressions in terms of the $`s^p`$ that do not depend on the $`t^A`$. We can rewrite (5.52) more compactly as
$$\sqrt{g}\varphi _{abc}_pB^{aa^{}}_qB^{bb^{}}_Ch^{cc^{}}\varphi _{a^{}b^{}c^{}}=\frac{1}{216}\frac{^3}{t^Cs^ps^q}BB\varphi $$
(5.53)
which is manifestly invariant under $`BB+dV`$. The expression on the right hand side of (5.53) also appeared in as defining a nice quadratic form on the space of $`B`$-fields, here we see that it arises naturally from the topological $`G_2`$ string. Also notice that this term is purely cubic in the coordinates, so fourth and higher derivatives of this terms will vanish identically.
The final generating functional of all correlation functions is an extension of the Hitchinโs functional to include the B-fields:
$$_{\mathrm{tot}}=\varphi \varphi +\frac{7}{72}BB\varphi .$$
(5.54)
### 5.6 What are we quantizing?
From the above discussion it seems clear that the prepotential $``$ of the topological string theory that we are studying can be viewed as a wave function in the Hilbert space that one obtains by quantization of the symplectic space $`H^2(M,)H^3(M,)H^4(M,)H^5(M,)`$, with symplectic structure $`\omega (\delta \alpha ,\delta \beta )=\delta \alpha \delta \beta `$. For the six-dimensional topological string, this point of view was taken in , see also , and it was shown that this is the natural way to understand the holomorphic anomaly. In our case we do not have a holomorphic anomaly, so it is not clear how compelling the interpretation of $``$ as a wave function is, see also section 8.1. Still, it is interesting to pursue this idea a little bit and therefore we will now briefly study the wave function interpretation restricting to the metric degrees of freedom only, i.e. we restrict ourselves to $`H^3H^4`$.
In order to be able to define suitable covariant derivatives we first define a Kรคhler potential
$$K=\frac{3}{7}\mathrm{log}.$$
(5.55)
This is, up to a numerical factor, precisely the Kรคhler potential of the 4d theory obtained by compactifying M-theory on a $`G_2`$ manifold (see section 7.3). In fact, the expression in (5.53) corresponds to the gauge couplings of the 4d theory<sup>16</sup><sup>16</sup>16More precisely , the gauge couplings are proportional to $`\left(t^A\frac{^3}{t^As^ps^q}BB\varphi \right)`$ and the $`\theta `$ terms are given by $`\left(p^A\frac{^3}{t^As^ps^q}BB\varphi \right)`$ where $`p^A`$ are moduli coming from the $`C`$ field in M-theory:
$$C=\underset{a=1}{\overset{h_2}{}}A^a_aB+p^A_A\varphi $$
Here $`A^a`$ are the $`h_2`$ gauge fields in the four-dimensional theory. We will come back to this in section 7.3. so that at tree level our topological string computes both the Kรคhler potential and the gauge couplings of the low energy effective field theory.
We can use the Kรคhler potential to define a covariant derivative
$$_A\varphi =_A\varphi +_AK\varphi $$
(5.56)
which has the property that $`_A\varphi `$ lives purely in the $`\mathrm{๐๐}`$ of $`G_2`$. In other words, the covariant derivative projects out the $`G_2`$ singlet contribution. Similarly, we can define a covariant derivative of $`\varphi `$ via
$$_A\varphi =_A\varphi +\frac{4}{3}_AK\varphi .$$
(5.57)
A useful observation is that
$$_A\varphi =_A\varphi $$
(5.58)
which can be derived using the calculations done in the preceding sections, but which also follows from the identity
$$\delta \varphi =(\frac{4}{3}\pi _1(\delta \varphi )+\pi _7(\delta \varphi )\pi _{27}(\delta \varphi ))$$
(5.59)
where $`\pi _1,\pi _7`$ and $`\pi _{27}`$ are the appropriate projections on the corresponding $`G_2`$ representations, and $`\delta \varphi `$ is an arbitrary variation.
Turning back to $`H^3H^4`$, we wish to consider the quantization of this space with respect to the symplectic form
$$\omega =_M\delta \alpha _3\delta \alpha _4$$
(5.60)
for $`(\alpha _3,\alpha _4)H^3H^4`$.
The simplest quantization, the analogue of the real polarization in the case of the B-model, is to define
$$p^A=_{C_A}\alpha _3,q_A=_{D^A}\alpha _4$$
(5.61)
for which the symplectic form becomes simply
$$\omega =\underset{A}{}dp^Adq_A.$$
(5.62)
This structure is manifestly independent of the $`G_2`$ structure of the manifold, i.e. it is background independent.
Next, we introduce a different set of coordinates. We pick a fixed reference $`G_2`$ structure $`\varphi `$ and choose
$$(\alpha _3,\alpha _4)=(x^A_A\varphi ,y^A_\varphi _A\varphi ).$$
(5.63)
We put the subscript $`\varphi `$ on $``$ to indicate that this is defined wrt to the reference $`G_2`$ structure. Notice that $`_\varphi _A\varphi `$ is closed, this follows from the identity
$$_A\varphi =_A\varphi \frac{7}{3}_AK\varphi $$
(5.64)
and since $`d\varphi =0`$ it is clear that $`d_A\varphi =0`$ as well, so that the right hand side of (5.64) is indeed closed.
Combining (5.40) and (5.64) we find that the symplectic form becomes
$$\omega =e^{7K/3}_A_BKdx^Ady^B$$
(5.65)
so that after quantization
$$[x^A,y^B]=ie^{7K/3}K^{AB}$$
(5.66)
with $`K^{AB}`$ the inverse of $`K_{AB}_A_BK`$.
As we vary the background the quantization changes. The coordinate $`x^A`$ is independent of the background (in fact, $`x^A=p^A`$ defined in (5.61)), since $`\varphi `$ is linear in the background coordinates $`t^A`$ (up to possible an exact form). However, $`y^A`$ changes. Its variation follows by imposing
$$\frac{\alpha _4}{t^B}=0.$$
(5.67)
After some straightforward algebra we obtain
$$_Ay^D\frac{7}{3}_AKy^D=K_{ABC}K^{CD}y^B,$$
(5.68)
where $`K_{ABC}_A_B_CK`$. It is interesting to observe that the answers are naturally expressed in terms of the Kรคhler potential $`K`$.
Equation (5.68) implies that $`y`$ eigenstates satisfy
$$_A|y=\left(K_{ABC}K^{BD}\frac{}{y^D}Y^C+\frac{7}{3}K_A\frac{}{Y^B}Y^B\right)|y.$$
(5.69)
The topological string wave function $`\psi (y)=\psi _{\mathrm{top}}|y`$ will then satisfy a similar differential equation, given that $`|\psi _{\mathrm{top}}`$ does not depend on the choice of background $`G_2`$ structure. This is the analogue of the holomorphic anomaly for the $`G_2`$ string.
From here on there are many different polarizations one can study. We can combine $`x^A`$ and $`y^A`$ in complex coordinates and work with the corresponding coherent states, to be closer to what we do in the case of a Calabi-Yau manifold. We can also separate out the overall rescalings of the metric and parametrize
$$(\alpha _3,\alpha _4)=(\xi \varphi +x^i_i\varphi ,\zeta \varphi +y^j_\varphi _j\varphi )$$
(5.70)
The symplectic form, in these coordinates, becomes
$$\omega =e^{7K/3}\left(d\xi d\zeta +(_i_jK_iK_jK)dx^idx^j\right)$$
(5.71)
The rest of the analysis will be similar to what we did above and we will not work out the details here. It will be an interesting question to see whether we can use these differential equations to make an educated guess about the higher genus contributions to the wave function.
To summarize, the topological $`G_2`$ string can be viewed as a wave function associated to a certain Lagrangian submanifold of the symplectic space $`H^2H^3H^4H^5`$. The Lagrangian submanifold consists of the points
$$(B,\varphi ,\frac{7}{3}_\varphi \varphi +\frac{7}{72}BB,\frac{7}{36}B\varphi )$$
(5.72)
where $`\varphi `$ runs over the space of $`G_2`$ metrics and $`B`$ over $`H^2(M)`$.
### 5.7 Topological $`๐_\mathrm{๐}`$ strings on $`\mathrm{๐๐}\times ๐^\mathrm{๐}`$.
An interesting example to study is the topological $`G_2`$ string on $`CY\times S^1`$. Because of the $`S^1`$, this seven-manifold is not a generic $`G_2`$ manifold. Whereas generic $`G_2`$ manifolds have no supersymmetric two-cycles, $`CY\times S^1`$ does have such two-cycles and therefore world-sheet instantons will contribute to the theory. In addition, the analysis of the BRST cohomology will be modified since $`H^1(CY\times S^1,)=`$. We will postpone a detailed discussion of these issues to another occasion, and here mainly focus on the metric and $`B`$-field moduli of $`CY\times S^1`$.
Any manifold of the form $`CY\times S^1`$ has a natural $`G_2`$ structure of the form
$`\varphi `$ $`=`$ $`\mathrm{Re}(e^{i\alpha }\mathrm{\Omega })+R\omega d\theta `$
$`\varphi `$ $`=`$ $`R\mathrm{Im}(e^{i\alpha }\mathrm{\Omega })d\theta +{\displaystyle \frac{1}{2}}\omega \omega `$ (5.73)
where $`\theta `$ is a periodic variable with period $`2\pi `$, $`e^{i\alpha }`$ is an arbitrary phase, $`R`$ is the radius of the $`S^1`$, and $`\mathrm{\Omega }`$ and $`\omega `$ are the holomorphic three-form and Kรคhler form on the Calabi-Yau manifold. These are not completely independent, but have to obey
$$i\mathrm{\Omega }\overline{\mathrm{\Omega }}=\frac{4}{3}\omega \omega \omega .$$
(5.74)
The $`G_2`$ BRST complex in say the left-moving sector, acting at the level of zero modes, involves among other the following differentials:
$$\mathrm{\Omega }^0(M,)\stackrel{d}{}\mathrm{\Omega }^1(M,)\stackrel{\varphi d}{}\mathrm{\Omega }^6(M,)\stackrel{d}{}\mathrm{\Omega }^7(M,).$$
(5.75)
where we used the identification of the $`\mathrm{๐}`$ in $`\mathrm{\Omega }^2(M,)`$ with $`\mathrm{\Omega }^6(M,)`$ and of the $`\mathrm{๐}`$ in $`\mathrm{\Omega }^3(M,)`$ with $`\mathrm{\Omega }^7(M,)`$ (see table (5.4)). The complex (5.75) is equivalent to (5.1) for any $`G_2`$ manifold. Thus, the full BRST cohomology is obtained by combining two complexes of the form (5.75), one for the left-movers and one for the right-movers. If we specialize to the case of a Calabi-Yau manifold times a circle using (5.7), (5.75) reduces to a certain complex involving the differential forms on the Calabi-Yau manifold. We are not aware of any literature on Calabi-Yau manifolds where such a complex appears, and this shows that the topological $`G_2`$ twist is not in a straightforward way related to the usual topological twist for Calabi-Yau manifolds.
More generally, complexes of the form (5.75) can be constructed for any special holonomy manifold by simply replacing $`\varphi `$ by a suitable covariantly closed differential form. It is an interesting question whether such complexes give in general rise to a new geometric understanding of special holonomy manifold.
Turning back to the $`CY\times S^1`$ case, the metric moduli of $`CY\times S^1`$ include the $`2h^{1,2}`$ complex structure moduli and $`h^{1,1}`$ Kรคhler moduli of the Calabi-Yau, but also the radius of the circle $`R`$. The total number of metric moduli is therefore $`dimH^3(CY\times S^1,)1`$. The number of three-form moduli is, however, equal to $`dimH^3(CY\times S^1,)`$. The difference is the parameter $`\alpha `$ in (5.7). Strictly speaking $`\alpha `$ does not correspond to an element of the BRST cohomology, and we should therefore remove the period of $`\varphi `$ corresponding to $`\alpha `$ from our consideration, but since nothing turns out to depend on $`\alpha `$ we may as well work with the full set of $`dimH^3(CY\times S^1)`$ periods. The modulus $`R`$ on the other hand is physical, and this has some interesting consequences for the relation between the topological $`G_2`$ string and the A- and B-model topological string on the Calabi-Yau manifold.
To study the topological $`G_2`$ string and its relation to the A- and B-model, we choose a basis of three-cycles $`A^I,B_I`$ with intersection number $`(A^I,B_J)=\delta _J^I`$ on the Calabi-Yau manifold. Similarly, we choose a basis of two-cycles $`C^a`$ and dual four-cycles $`D_a`$. The cycles on $`CY\times S^1`$ are then given by
$`\mathrm{two}\mathrm{cycles}`$ $`:`$ $`C^a`$
$`\mathrm{three}\mathrm{cycles}`$ $`:`$ $`C^a\times S^1,A^I,B_I`$
$`\mathrm{four}\mathrm{cycles}`$ $`:`$ $`D_a,B_I\times S^1,A^I\times S^1`$
$`\mathrm{five}\mathrm{cycles}`$ $`:`$ $`D_a\times S^1.`$ (5.76)
The prepotential of the topological $`G_2`$ string also depends on the $`B`$-field. To take this account we need to improve the four form to
$$\varphi \varphi ^{(4)}R\mathrm{Im}(e^{i\alpha }\mathrm{\Omega })d\theta \frac{1}{2}\mathrm{Re}(\omega +\frac{i}{2}B)(\omega +\frac{i}{2}B).$$
(5.77)
The various periods, which define coordinates on the moduli space of $`G_2`$ metrics, are given by
$`b^a`$ $`=`$ $`{\displaystyle _{C^a}}B`$
$`k^a`$ $`=`$ $`{\displaystyle _{C^a\times S^1}}\varphi `$
$`q^I`$ $`=`$ $`{\displaystyle _{A^I}}\varphi `$
$`p_I`$ $`=`$ $`{\displaystyle _{B_I}}\varphi `$
$`{\displaystyle \frac{3}{7}}{\displaystyle \frac{}{k^a}}`$ $`=`$ $`{\displaystyle _{D_a}}\varphi ^{(4)}`$
$`{\displaystyle \frac{3}{7}}{\displaystyle \frac{}{q^I}}`$ $`=`$ $`{\displaystyle _{B_I\times S^1}}\varphi ^{(4)}`$
$`{\displaystyle \frac{3}{7}}{\displaystyle \frac{}{p_I}}`$ $`=`$ $`{\displaystyle _{A^I\times S^1}}\varphi ^{(4)}`$
$`{\displaystyle \frac{1}{2}}{\displaystyle \frac{}{b^a}}`$ $`=`$ $`{\displaystyle _{D_a\times S^1}}B\varphi .`$ (5.78)
Now, we want to relate these variable to the quantities that appear naturally in the A and the B models on the Calabi-Yau manifold. If we denote by $`^A`$ and $`^B`$ the suitably normalized prepotentials of the A- and B-model, then these obey
$`X^I`$ $`=`$ $`{\displaystyle _{A^I}}\mathrm{\Omega }`$
$`{\displaystyle \frac{^B}{X^I}}`$ $`=`$ $`{\displaystyle _{B_I}}\mathrm{\Omega }`$
$`t^a`$ $`=`$ $`{\displaystyle _{C^a}}\omega +{\displaystyle \frac{i}{2}}B`$
$`{\displaystyle \frac{^A}{t^a}}`$ $`=`$ $`{\displaystyle _{D_a}}(\omega +{\displaystyle \frac{i}{2}}B)^2`$ (5.79)
with $`X^I`$ and $`t^a`$ the complex structure and complexified Kรคhler moduli. By comparing (5.7) and (5.7) we can now determine the relation between $``$ and $`^A`$ and $`^B`$. This is somewhat subtle due to the appearance of the parameter $`R`$ in $`\varphi `$ and $`\varphi ^{(4)}`$. $`R`$ itself is not an independent period but it appears in (5.7) in a non-trivial way. We should also keep in mind that in (5.7) $`\mathrm{\Omega }`$ and $`\omega `$ are constrained by (5.74), so that the variables $`X^I`$ and $`t^a`$ obey a nontrivial constraint. To reformulate this constraint we denote
$$P(X^I,\overline{X}^I)=3i\mathrm{\Omega }\overline{\mathrm{\Omega }},Q(t^a,\overline{t}^a)=4\omega ^3$$
(5.80)
so that the constraint is that $`P(X^I,\overline{X}^I)=Q(t^a,\overline{t}^a)`$. A comparison of the periods yields the following set of equations (we put $`\alpha =0`$ here, but it can be trivially put back into the equations by replacing $`\mathrm{\Omega }e^{i\alpha }\mathrm{\Omega }`$)
$`b^a`$ $`=`$ $`2\mathrm{I}\mathrm{m}(t^a)`$
$`k^a`$ $`=`$ $`2\pi R\mathrm{Re}(t^a)`$
$`q^I`$ $`=`$ $`\mathrm{Re}(X^I)`$
$`p_I`$ $`=`$ $`\mathrm{Re}(_I^B)`$
$`{\displaystyle \frac{3}{7}}{\displaystyle \frac{}{k^a}}`$ $`=`$ $`{\displaystyle \frac{1}{2}}\mathrm{Re}(_a^A)`$
$`{\displaystyle \frac{3}{7}}{\displaystyle \frac{}{q^I}}`$ $`=`$ $`2\pi R\mathrm{Im}(_I^B)`$
$`{\displaystyle \frac{3}{7}}{\displaystyle \frac{}{p_I}}`$ $`=`$ $`2\pi R\mathrm{Im}(X^I)`$
$`{\displaystyle \frac{1}{2}}{\displaystyle \frac{}{b^a}}`$ $`=`$ $`2\pi R\mathrm{Im}(_a^A).`$ (5.81)
To solve this system of equations, we first express $`P(X^I,\overline{X}^I)`$ in terms of $`q^I,p_I`$. As is well-known, in terms of $`q^I,p_I`$ $`P`$ is equal to the Legendre transform of the imaginary part of $`^B`$,
$$P(p_I,q^I)=3i\mathrm{\Omega }\overline{\mathrm{\Omega }}=12(\mathrm{Im}(^B)p_I\mathrm{Im}(X^I))_{q^I=\mathrm{Re}(X^I),p_I=\mathrm{Re}(_I^B)}.$$
(5.82)
We cannot express $`Q(t^a,\overline{t}^a)`$ in terms of $`k^a`$ directly, due to the factor of $`R`$ that appears in the relation between $`k^a`$ and $`t^a`$. However, the following is a function of just the $`k^a`$:
$$S(k^a)=4(2\pi R\omega )^3.$$
(5.83)
The constraint $`P=Q`$ now implies that $`R`$ is a nontrivial function of $`q^I,p_I,k^a`$, given by
$$2\pi R(p_I,q^I,k^a)=\left(\frac{S(k^a)}{P(p_I,q^I)}\right)^{1/3}.$$
(5.84)
We also define
$$T(p_I,q^I,k^a,b^a)=12\mathrm{R}\mathrm{e}(^A)_{t^a=\frac{k^a}{2\pi R(p_I,q^I,k^a)}+\frac{ib^a}{2}}$$
(5.85)
so that
$$S(k^a)=(2\pi R(p_I,q^I,k^a))^3T(p_I,q^I,k^a,b^a)_{b^a=0}.$$
(5.86)
We now claim that
$``$ $`=`$ $`2\pi R(p_I,q^I,k^a)\left({\displaystyle \frac{7}{36}}P(p_I,q^I){\displaystyle \frac{7}{72}}T(p_I,q^I,k^a,b^a)\right)`$ (5.87)
$`=`$ $`{\displaystyle \frac{7}{3}}(2\pi R)\left((\mathrm{Im}(^B)p_I\mathrm{Im}(X^I))+{\displaystyle \frac{1}{2}}\mathrm{Re}(^A)\right).`$
This shows that the prepotential of the topological $`G_2`$ string is indeed a combination of the $`A`$\- and $`B`$-model topological string, but the complex and Kรคhler moduli of the Calabi-Yau manifold get mixed in a rather intricate way due to the presence of the radius $`R`$. $`R`$ is closely related to the volume of the Calabi-Yau manifold, and it would be interesting to see if this is related to and/or can resolve the gravitational anomaly found in the one-loop calculation in the six-dimensional Hitchin system in . The non-trivial role that $`R`$ plays in the above also manifests itself in the analysis of four-dimensional supergravity, see e.g. .
To show that (5.87) solves (5.7) is somewhat complicated due to the dependence of $`R`$ on $`p_I,q^I,k^a`$. However, one may check that
$$\frac{}{(2\pi R)}=\frac{7}{36}(P(p_I,q^I)T(p_I,q^I,k^a,b^a)_{b^a=0})$$
(5.88)
where it is important to differentiate not just the explicit $`R`$ that appears in (5.87), but also the $`R`$ that appears in the definition of $`T`$ in (5.85). The right hand side of (5.88) is precisely the original constraint (5.74) and therefore vanishes identically. In other words, the radius seems to play the role of a Lagrange multiplier that imposes the volume constraint (5.74). Because of this, we can treat $`R`$ as a constant when verifying (5.7), and with this simplification it is straightforward to verify that (5.87) solves (5.7).
From (5.87) we also find, using (5.84) and (5.86), that
$$_{b^a=0}=\frac{7}{12}S(k^a)^{1/3}P(p_I,q^I)^{2/3}.$$
(5.89)
Thus, the topological $`G_2`$ string is not just the sum of $`A`$\- and $`B`$-model, but it can also be written as the product of fractional powers of the $`A`$\- and $`B`$-model. It would be interesting to know whether either the combinations (5.87) and (5.89) have any distinguished meaning for six-dimensional topological strings.
## 6 The Topological $`๐_\mathrm{๐}`$ String
We have so far been considering a topologically twisted $`\sigma `$ model of maps from a sphere into a $`G_2`$ manifold. However, on higher genus Riemann surfaces, there is nothing interesting to compute in the $`\sigma `$-model. To get interesting amplitudes, we need to couple the $`\sigma `$ model to two dimensional gravity, and integrate over the moduli space of Riemann surfaces. This will define the topological $`G_2`$ string. In the following, we first give a preliminary discussion the topological $`\sigma `$-model at higher genus and then construct a measure on the moduli space of Riemann surfaces to define the topological string amplitudes.
### 6.1 Twisting the $`\sigma `$ Model At Higher Genus.
Generalizing the sphere computation to higher genera , n-point correlators on a genus-g Riemann surface in the twisted theory are defined as a correlator in the untwisted theory of the same n operators plus $`(22g)`$ insertions of the spin-field that is related to the space-time supersymmetry charge. For a Calabi-Yau 3-fold target space on a Riemann surface with $`g>1`$ the meaning of the above prescription is to insert $`2g2`$ of the conjugate spectral flow operator ($`e^{i\frac{\sqrt{d}}{2}\varphi }`$ in the notation of section 4.1). To generalize this to the $`G_2`$ situation, we will do something similar. However, there is only a single $`G_2`$ invariant spinfield. This is where the decomposition in conformal blocks in section 2.3 is useful: the spin-field $`\mathrm{\Phi }_{1,2}`$ (which corresponds to the particular Ramond sector ground state $`|\frac{7}{16},0`$) could be decomposed in a block $`\mathrm{\Phi }_{1,2}^+`$ and in a block $`\mathrm{\Phi }_{1,2}^{}`$ (see eq 2.13 and A.8, and also section 7.1). At genus zero we needed two insertions of $`\mathrm{\Phi }_{1,2}^+`$, so the natural guess is that at genus $`g`$ we need $`2g2`$ insertions of $`\mathrm{\Phi }_{1,2}^{}`$. We will demonstrate shortly that with this guess the topological $`G_2`$ strings are indeed โcriticalโ in 7 dimensions.
### 6.2 Topological Strings
To go from a topological $`\sigma `$ model to topological strings, we need to integrate over the moduli space of Riemann surfaces, $`_g`$. To construct a measure on the moduli space of Riemann surfaces, we need an anti-ghost $`G_{}^{}`$, such that $`\{Q,G_{}^{}\}=T`$ where $`T`$ is the twisted stress tensor and $`Q`$ is the BRST operator. We use the notation $`G_{}^{}`$ for the anti-ghost because the conformal block $`G^{}`$ defined previously almost does the job, as discussed in section 4.7. In the following, we assume that a suitable modification $`G_{}^{}`$ of $`G^{}`$ exists which we can use to define the topological string amplitudes. With this important assumption we can define the genus-g free energy $`F_g`$ of the $`G_2`$ topological string by integrating over the $`3g3`$ dimensional moduli space of genus-g Riemann surfaces $`_g`$ along with $`3g3`$ insertions of the anti-ghost folded against Beltrami differentials giving the appropriate measure of integration
$$F_g=__g\underset{i=1}{\overset{3g3}{}}|(\mu _i,G_{}^{})|^2_g$$
(6.1)
where the folded anti-ghosts are defined by integrating them over the genus-g worldsheet against the Beltrami differentials $`(\mu _i,G_{}^{})=d^2z\mu _i(z)G_{}^{}(z)`$.
##### Critical Dimension
The usual topological strings on Calabi-Yau manifolds have a โcritical dimensionโ $`d=6`$ (complex dimension 3). This is because essentially all the higher genus free energies $`F_g`$ vanish when the target space is a complex manifold of (complex) dimension other than 3. The $`G_2`$ string is critical in 7 dimensions. Indeed, we can use the fusion rules of the tri-critical Ising model to show that there is a non-vanishing contribution to correlation functions of $`2g2`$ $`\mathrm{\Phi }_{1,2}`$โs and $`3g3`$ $`G^{}`$. We can also show that their correlation functions are non-zero by considering the Coulomb gas representation of the tri-critical Ising model (which is useful to compute correlation functions). From that perspective the $`2g2`$ insertions of $`\mathrm{\Phi }_{1,2}^{}`$ and $`3g3`$ insertions of $`G^{}`$ yield a total $`\varphi `$ charge of
$$(22g)\frac{5}{2\sqrt{10}}+(3g3)\frac{2}{\sqrt{10}}=(g1)\frac{1}{\sqrt{10}}$$
(6.2)
which is exactly the correct amount needed to cancel the existing background charge ($`\frac{1}{\sqrt{10}}`$) of the tri-critical Ising model on a genus-g Riemann surface. Here we used that the anti-ghost $`G^{}`$ has weight two in the Coulomb gas representation (see appendix A).
The $`G_2`$ topological string partition function is defined as an asymptotic series in a coupling constant $`\lambda `$
$$๐ต=e^{},\mathrm{๐ ๐๐๐๐}=\underset{g=0}{\overset{\mathrm{}}{}}\lambda ^{22g}F_g.$$
(6.3)
The descent relations introduced in section 4.8 enable us to now define correlation functions of chiral primaries just like in the $`๐ฉ=2`$ topological string.
## 7 Physics in Three Dimensions
Since we are discussing type II string theory compactified on a manifold of $`G_2`$ holonomy, we expect the topological $`G_2`$ string to be of relevance for the resulting three-dimensional effective field theory. In this section we will explore some properties of this effective field theory and how they are related to topological $`G_2`$ strings. Since $`G_2`$ compactifications preserve four supercharges, the resulting three-dimensional theory will have $`๐ฉ=2`$ supersymmetry.
### 7.1 Massless fields and the GSO projection
We are dealing with an odd dimensional compactification of string theory. Therefore, the GSO projection is particularly subtle. In order to define it, we need a notion of fermion number. We will first define this in the NS sector of the internal CFT corresponding to the sigma model on the $`G_2`$ manifold. As discussed in some detail in , we can assign a fermion number to a state by assigning a fermion number to the tri-critical Ising part of the state. In the NS sector, there is a tri-critical Ising model notion of fermion number in which we associate fermion number $`(1)^{n+1}`$ for states in Hilbert space $`_n`$. with $`n=1,\mathrm{},4`$ ($`n=1`$ corresponding to the identity, $`n=2`$ to the primary $`\frac{1}{10}`$ etc) . The fermion number in the 3d spacetime part of the compactification in the NS sector is the usual one.
In the R sector, things are less straightforward. In three dimensions, the representations of the Clifford algebra are two-dimensional, and there are no chiral spinors. The same holds true in seven dimensions. Therefore, in order to have a well-defined fermion number, we need to take a reducible representation of the Clifford algebra in three dimensions which consists of two spinors which we will call $`|3,+`$ and $`|3,`$ where the sign indicates fermion number. Similarly, we need two spinors coming from the seven-dimensional part, which we will call $`|7,+`$ and $`|7,`$. The zero modes of the three-dimensional fermions map $`|3,+`$ to $`|3,`$ and vice versa. With this doubling we have a well defined action of $`(1)^F`$ given by $`(1)^F|3,\pm =\pm |3,\pm `$. A similar remark applies to the seven-dimensional part. When we combine the three and seven-dimensional part, we find that if we take all possible combinations, we obtain a reducible representation. The smallest irreducible representation, which still allows for a proper action of $`(1)^F`$, is obtained by taking e.g. the combinations
$`|\chi ,+`$ $`=`$ $`|3,+|7,++|3,|7,`$
$`|\chi ,`$ $`=`$ $`|3,+|7,+|3,|7,+`$ (7.1)
where fermion number acts as $`(1)^F|\chi ,\pm =\pm |\chi ,\pm `$. The GSO projection projects on one of the two chiralities and results in a single two component spinor in three dimensions. From the right movers we get another two-component spinor and this is how we arise at $`N=2`$ supersymmetry in three dimensions. <sup>17</sup><sup>17</sup>17 Notice that this also resolves the peculiar feature that representations in the R sector (discussed in appendix C) of the $`G_2`$ algebra can be one-dimensional, but once we combine left and right movers they should be two-dimensional. As the above shows, the R sector really involves two-dimensional representations, and the left-right sector four-dimensional ones. No strange enhancement is necessary once we combine left and right movers.
If we just quantize the seven-dimensional sigma model, the above suggests that we get two copies of each R representation, together with a label $`\pm `$. The natural interpretation from the point of view of the tri-critical Ising model, is that $`\pm `$ corresponds to the decomposition of $`R`$ ground states in two conformal blocks. In this way, the fusion rules of the tri-critical Ising model can be made to agree with the fermion number assignment, up to an extra minus sign for the product of two fields in the RR sector. For example,
$`[{\displaystyle \frac{7}{16}},\pm ][{\displaystyle \frac{7}{16}},]`$ $`=`$ $`[0,+]`$
$`[{\displaystyle \frac{7}{16}},\pm ][{\displaystyle \frac{7}{16}},\pm ]`$ $`=`$ $`[{\displaystyle \frac{3}{2}},]`$
$`[{\displaystyle \frac{7}{16}},\pm ][{\displaystyle \frac{3}{80}},]`$ $`=`$ $`[{\displaystyle \frac{6}{10}},+]`$
$`[{\displaystyle \frac{7}{16}},\pm ][{\displaystyle \frac{3}{80}},\pm ]`$ $`=`$ $`[{\displaystyle \frac{1}{10}},],`$ (7.2)
etcetera.
Using these fusion rules, it is easy to see that tree level correlation functions only vanish if the total $`(1)^F`$ of the operators in the correlation function is equal to $`(1)^p`$, where $`p=n_R/2`$ is half the number $`n_R`$ of R fields. This applies to both the left and right movers separately. At higher genus correlation functions also involve a choice of spin structure.
We can now also properly define operators like $`G^{}`$ and $`G^{}`$ in the R sector. We decompose the R Hilbert space as
$$_R_{R,1}_{R,2}_{R,3}_{R,4}=_{\frac{7}{16},+}_{\frac{3}{80},}_{\frac{3}{80},+}_{\frac{7}{16},}$$
(7.3)
and define the up and down projections exactly as in the case of the NS sector in terms of the action on $`_i`$. For example, $`G^{}`$ will only map $`_i_{i+1}`$.
### 7.2 Relation of the Topological $`๐_\mathrm{๐}`$ String to Physical Amplitudes
An important application of topological strings stems from the realization that its amplitudes agree with certain amplitudes of the physical superstring. The usual topological strings on Calabi-Yau manifolds compute F-terms in four dimensional compactification of the physical superstrings. A natural question is: What physical amplitudes does the topological $`G_2`$ string compute in three dimensional $`๐ฉ=2`$ compactifications of superstring theories. As we will see, at genus zero, the topological string indeed computes certain Yukawa couplings. However, at higher genus, unlike the usual topological string theories, the topological $`G_2`$ string does not compute F-terms in three dimensions. As we will see, this failure to compute such terms can be traced to the absence of chiral spinors in three dimensions.
Comactification of type II superstrings on $`G_2`$ holonomy manifolds leads to $`๐ฉ=2`$ supergravity in three dimensions, where a single supercharge arises from each world sheet chirality. The (e.g. left moving) supersymmetry generator is constructed according to the standard FMS ansatz
$$Q^\alpha =e^{\frac{\phi }{2}}\left(S_{3+}^\alpha \mathrm{\Sigma }_++S_3^\alpha \mathrm{\Sigma }_{}\right)$$
(7.4)
where $`S_{3\pm }`$ is a spin-field in $`R^{1,2}`$ (corresponding to the states $`|3,\pm `$ in section 7.1) and $`\mathrm{\Sigma }_\pm `$ are operators corresponding to the states $`|7,\pm `$ in section 7.1. Also, $`\phi `$ is the bosonized super-ghost arising in the standard BRST quantization of type II superstrings.
Which physical amplitudes can we possibly relate to the topological string? These should be amplitudes involving Ramond sector vertex operators which, in their $`G_2`$ factor have the field $`\mathrm{\Sigma }_{}`$ inserted an appropriate number of times to give a topological amplitude.<sup>18</sup><sup>18</sup>18In the case of Calabi-Yau 3-folds, analogous amplitudes which are related to the topological string consist of $`2g2`$ gravi-photons, which suggests a F-term in the four dimensional effective action of the form $`W^{2g}`$, where $`W`$ is the Weyl super-multiplet of $`๐ฉ=2`$ supergravity. Here, $`W`$ is the chiral superfield of $`๐ฉ=2`$ supergravity multiplet whose first component is the graviphoton field strength $`T_{\mu \nu }`$. In components, the $`W^{2g}`$ term gives a coupling between two gravitons and $`2g2`$ graviphotons: $`R^2T^{2g2}`$, and it can be shown that the coefficient of this term is the topological string partition function $`F_g(t,\overline{t})`$. In addition, in order to have some non-trivial dynamics in three dimensions, we need a field which sits in $`(\mathrm{๐},!\mathrm{๐})`$ of $`SO(3)\times G_2SO(10)`$. A singlet under the $`SO(3)`$ factor would imply a non-dynamical degree of freedom in three dimensions.
##### The RR sector
The RR vertex operators have spinor bilinears. We are looking for singlets under $`G_2`$. These will come from the spinor bilinears made out of the covariantly constant spinor on the $`G_2`$ manifold. As discussed before, this can only generate a three form or a four form. All other combinations vanish. Then, there remains a unique field which sits in the $`(\mathrm{๐},\mathrm{๐})`$ of $`SO(3)\times G_2`$. For type IIA and type IIB, this corresponds to a scalar field $`\rho `$ such that
$$\begin{array}{cc}\hfill \mathrm{type}\mathrm{IIA}_\mu \rho & =_{M_7}F_{RR}^{(4)}\varphi ,\hfill \\ \hfill \mathrm{type}\mathrm{IIB}_\mu \rho & =_{M_7}F_{RR}^{(5)}\varphi .\hfill \end{array}$$
(7.5)
where $`\varphi `$ is the 3-form that defines the $`G_2`$ structure. The vertex operator (in type IIB) corresponding to these spacetime fields in the $`1/2`$ picture is
$$V^i=e^{\frac{\phi +\stackrel{~}{\phi }}{2}}\left(S_{3+}^\alpha (\tau _{\alpha \beta }^i)\stackrel{~}{S}_{3+}^\beta \mathrm{\Sigma }_+\stackrel{~}{\mathrm{\Sigma }}_++S_3^\alpha (\tau _{\alpha \beta }^i)\stackrel{~}{S}_3^\beta \mathrm{\Sigma }_{}\stackrel{~}{\mathrm{\Sigma }}_{}\right)$$
(7.6)
where (non)tilde denotes (left) right-movers and $`\tau ^i`$ are the Pauli matrices.<sup>19</sup><sup>19</sup>19For type IIA, we need to change $`\stackrel{~}{\mathrm{\Sigma }}_\pm `$ to $`\stackrel{~}{\mathrm{\Sigma }}_{}`$.
At first sight, it might seem that $`2g2`$ insertions of this operator would twist the $`G_2`$ part of the CFT by appropriate insertions of the spin field $`\mathrm{\Sigma }_{}`$. However, this is of course incorrect, because the vertex operator in (7.6) is a sum of two terms. Therefore, in addition to getting terms with $`\mathrm{\Sigma }_{}^{2g2}\stackrel{~}{\mathrm{\Sigma }}_{}^{2g2}`$ which can be simply related to the topological amplitudes, we get terms with $`\mathrm{\Sigma }_+^{2g2}\stackrel{~}{\mathrm{\Sigma }}_+^{2g2}`$ insertions and also all possible cross terms which are non-topological in nature. At a generic genus, generally these non-topological terms are non-vanishing, with the result that the total amplitude is non-topological in nature. For type II strings on Calabi-Yau manifolds, there is a natural way to restrict to one of the two terms in such a vertex operator 7.6, and that is by looking at self-dual (or anti self-dual) graviphoton field strengths. In three dimensions, there is no natural way to restrict to one of the two terms in the vertex operator. Therefore, we conclude that generically, the topological string does not seem to compute F-terms in the three dimensional effective action. There is an exception, though, at genus 0.
### 7.3 Tree level effective action and the topological $`๐_\mathrm{๐}`$ string
In order to describe the three-dimensional effective action it is convenient to first work with 11d supergravity compactification on $`G_2`$ manifolds down to four dimensions. The three dimensional action can then be obtained by a dimensional reduction. The four dimensional theory has $`b_3`$ chiral multiplets and $`b_2`$ vector multiplet. The scalars in the chiral multiplets are complex combinations of the metric moduli and the three form 11 dimensional $`C`$-field moduli: $`S^A=t^A+ip^A`$, where $`p^A`$ is defined in footnote 16. The Kรคhler potential for the scalars is a function of the real part of $`S^A`$ and is given by
$$K(S+\overline{S})=3\mathrm{log}(\frac{1}{7}\varphi \varphi )$$
(7.7)
The kinetic terms for the $`b_2`$ gauge fields are given by
$$\mathrm{Im}d^4xd^2\theta \tau _{ab}W_\alpha ^aW^{\alpha b}$$
(7.8)
which can be dimensionally reduced to three dimensions
$$S=\mathrm{Im}d^3xd^2\theta \tau _{ab}W_\alpha ^aW^{b\alpha }$$
(7.9)
where $`W_\alpha ^a`$ is the field strength superfield, the gauge coupling is $`\tau _{ab}=S^A_A_a_b\left(\frac{36}{7}_{\mathrm{tot}}\right)`$, where $`_{\mathrm{tot}}`$ is defined in eq (5.54) and $`_a=\frac{}{s^a}`$.
This action is written in terms of dimensionally reduced 4d vector multiplet as an integral over a chiral half of superspace. In 3 dimensions, vectors multiplets are dual to the chiral multiplet and it is interesting to determine the Kรคhler potential for these chiral multiplets. To this end, we need to perform the duality transformation and it is convenient to do this directly in superspace. Four-dimensional vector multiplets are not the most convenient way to define gauge theories in three dimensions. Gauge theories in three dimensions are usually formulated in terms of linear multiplets. We therefore first rewrite (7.9) in terms of linear multiplets $`G^a`$ in terms of which the action becomes
$$S=d^3xd^4\theta (\tau _{ab}(S)+\overline{\tau }_{ab}(\overline{S}))G^aG^b.$$
(7.10)
We can write the B-field as $`G^a\omega _a`$ and $`\varphi =(S^A+\overline{S}^A)\chi _A`$, where $`\omega _a`$ and $`\chi _A`$ are bases of $`H^2`$ and $`H^3`$ respectively, of the $`G_2`$ manifold. Then, the superspace action can be formally written as
$$S=d^3xd^4\theta BB\varphi $$
(7.11)
which is exactly the second term which appears in $`_{\mathrm{total}}`$.
To perform the duality transformation explicitly between the linear and the chiral multiplets (see e.g. ), we can even start from a more general action
$$S=d^3xd^4\theta f(G^a,S,\overline{S})$$
(7.12)
This action can be rewritten as
$$S=d^3xd^4\theta f(\stackrel{~}{G}^a,S,\overline{S})\stackrel{~}{G}^a(Y_a+\overline{Y}_a)$$
(7.13)
where the superfields $`\stackrel{~}{G}^a`$ are unconstrained real superfields, and the $`Y_a`$ are chiral superfields. Extremizing the action with respect to $`Y_a`$ constrains $`\stackrel{~}{G}^a`$ to be linear superfields from which we obtain (7.12) back. We can also vary this action with respect to $`\stackrel{~}{G}^a`$ which yields the equation
$$Y_a+\overline{Y}_a=\frac{f(\stackrel{~}{G}^a,S,\overline{S})}{\stackrel{~}{G}^a}$$
(7.14)
By solving for $`\stackrel{~}{G}^a`$ in terms of $`S`$ and $`\overline{S}`$ and substituting in (7.13) gives the dual description in terms of a Kรคhler potential $`K(Y_a+\overline{Y}_a,S,\overline{S})`$ for the chiral multiplets $`Y_a`$:
$$S=d^3xd^4\theta K(Y_a+\overline{Y}_a,S,\overline{S}).$$
(7.15)
Here, $`K`$ is the Legendre transform of $`f`$. For our case (7.10), $`f=\left(\tau _{ab}(S)+\overline{\tau }_{ab}(\overline{S})\right)\stackrel{~}{G}^a\stackrel{~}{G}^b`$, so
$$K(Y_a+\overline{Y}_a,S,\overline{S})=(Y_a+\overline{Y}_a)\left(\mathrm{}\tau (S)^1\right)^{ab}(Y_b+\overline{Y}_b)$$
(7.16)
This is simply the Legendre transform of (7.11) with respect to the $`B`$ field moduli.
## 8 Discussion, open questions and future directions
In this concluding section, we list and discuss several interesting issues and future directions.
### 8.1 The coupling constant
The partition function for the ordinary topological string on Calabi-Yau manifolds is better thought of as a wave function. This picture emerges from the holomorphic anomaly, where the holomorphic anomaly equation is interpreted as describing the change in basis (an infinitesimal fourier transform) in the quantum mechanics whose phase space is given by $`H^3(M)`$ . It remains an interesting question whether the partition function of our topological string should naturally have a wave function interpretation. In our case, there is no corresponding holomorphic anomaly equation. Also, when we consider our topological string on CY $`\times S^1`$, it naturally contains both the holomorphic and anti-holomorphic A and B models. These facts suggests an interpretation as a partition function as opposed to a wave function.
However, we also argued in section 5.6 that we could view the topological $`G_2`$ string as a wavefunction corresponding to a lagrangian submanifold of $`H^2+H^3+H^4+H^5`$. From this perspective, it is interesting to note that we can naturally incorporate the string coupling in the framework. Consider again our function
$$=\frac{1}{g_s^2}\varphi \varphi +\frac{7}{24g_s^2}BB\varphi $$
(8.1)
where we have now included the string coupling constant. We can associate to it a Lagrangian submanifold of $`H^{}(M)`$ which now also includes $`H^0`$ and $`H^7`$, namely
$$(\frac{1}{g_s},B,\varphi ,\frac{}{\varphi },\frac{}{B},\frac{}{\frac{1}{g_s}})$$
(8.2)
In this way the string coupling gets naturally associated to $`H^0(M)`$. This is similar to what is done in the A model. In the B model, the string coupling is related to one particular component of $`H^3`$, namely the one proportional to the holomorphic three form. At first sight, it does not seem to be the case here. However, as discussed in Appendix D, there is an isomorphism between $`H^0`$ and $`H_1^3`$, i.e. those elements of the third cohomology which transform as the singlet under the group $`G_2`$. The moduli space has a projective structure. We can view the $`t^A`$ defined in (5.26) as providing real projective coordinates on the $`b_{27}^3=b_31`$ dimensional moduli space of $`G_2`$ metrics which correspond to deformations of the $`G_2`$ structure which are not rescalings of the metric. The partition function of the topological $`G_2`$ string is then a section of a real line bundle of degree $`\frac{7}{3}`$. Though this is not the structure that we find in the topological string, it may naturally emerge when we try to lift it to M-theory.
### 8.2 Strong coupling limit
The construction of the topological string theory that we have given is a perturbative one. The strong coupling limit and a non-perturbative completion remains an interesting question. A strong coupling limit, if well defined, could naturally be topological M-theory . An obvious strong coupling limit is one where we scale $`\varphi `$ with $`\lambda ^{3/7}`$ and $`g_s`$ with $`\lambda `$, after which we send $`\lambda \mathrm{}`$. This does not change the form of $``$. It is not clear whether the result should be viewed as a string theory. In fact, it is perhaps more appropriate to think of this topological theory as describing certain sector of M-theory compactification on $`G_2`$ manifolds down to 4 dimensions. The number of variables that remain will be one-less compared to the number of variables in three dimensions โ we lose the degree of freedom corresponding to the rescaling of $`\varphi `$, the three-form which defines the $`G_2`$ structure; or equivalently, the string coupling.
Another limit we can study is the theory on $`CY\times S^1`$. In this case we can try to decompactify the $`S^1`$, which is related via a 9-11 flip to the strong coupling limit above. Since $`R`$ depends non-trivially on all moduli, it is not immediately clear what is a natural set of variables that survives. Perhaps we should keep all $`H^3`$ except the class proportional to $`\varphi `$, as we do for the complex structure in the B-model?
### 8.3 Relation to black holes and Hitchin flows
Notice that our function $`P(q^I,p_I)`$ (eq. 5.82) is the Legendre transform of the free energy of the B-model, which is exactly the expression that appears in the recent discussions of the relation between topological strings and black hole entropy . This is perhaps not that surprising given that $`P(q^I,p_I)`$ is the volume of the CY at the horizon of the black hole through the attractor mechanism. Yet, one may wonder whether the circle in the 7d theory on $`CY\times S^1`$ can be interpreted as a Euclidean time direction so that the theory can be directly viewed as a thermal system with nonzero entropy, giving a microscopic description of the black hole entropy. Perhaps our topological twist can be interpreted as counting BPS states in a black hole background.
In , domain wall solutions of $`๐ฉ=2`$ gauged four-dimensional supergravity were constructed, where the supergravity theory was obtained by the dimensional reduction of type IIA on โhalf-flatโ six manifolds. These are manifolds which have a particular type of $`SU(3)`$ structure. The domain walls are determined by flow equations which govern the dependence of scalars (corresponding to the moduli of the internal manifold) in the direction transverse to the domain wall. These flow equations were shown to be equivalent to Hitchinโs flow equations, which implies that the transverse direction to the domain wall combines with the internal manifold to give a $`G_2`$ manifold. A natural question is whether the black hole attractor flows have a similar interpretation in terms of Hitchin flows which may then admit a re-interpretation of these in terms of a manifold with $`G_2`$ structure. We leave this interesting point for a future investigation.
Notice that in M-theory on $`G_2`$ manifolds there are no supersymmetric black holes, so we do not expect the existing relation between topological strings and BPS black holes to generalize to this setup.
### 8.4 An analogue of KS theory?
The topological A and B model are defined perturbatively in an on shell formalism which studies maps from the world sheet to a target space. Perturbative computations can be done using world-sheet methods. However, for the B-model, there is a target space โstring field theoryโ (though for the B-model, this reduces to a field theory), namely the Kodaira Spencer theory which presumably yields exactly the same results as the world-sheet calculations. This is a theory of complex structure deformations of the Calabi-Yau manifold. The fundamental variable of Kodaira Spencer theory corresponds to an infinitesimal change of the complex structure of the Calabi-Yau manifold. The equation of motion of this theory is equivalent to the complex structure being integrable. The action, which can be written down by following the standard rules of string field theory , consists of a quadratic kinetic term and a cubic interaction term. There are no higher point interaction terms since four and higher point correlation functions in the world sheet theory vanish.
One may hope that the target space theory of the topological $`G_2`$ string is a seven dimensional theory of deformations of $`G_2`$ structures, a version of the Kodaira Spencer theory that lives in seven dimensions. The fundamental variable should be an infinitesimal metric deformation, i.e. a symmetric two-tensor $`A_{\mu \nu }`$. If we again follow the standard string field theory logic, the action would take the form
$$S=S_2(A)+S_3(A)$$
(8.3)
with $`S_2(A)A\frac{G_0^{}}{b_0^{}}A=A\frac{G_0^{}}{G_0^{}}A`$ and with
$$S_3(A)=d^7x\sqrt{g}\varphi ^{\alpha \beta \gamma }A_{\alpha \alpha ^{}}A_{\beta \beta ^{}}A_{\gamma \gamma ^{}}\varphi ^{\alpha ^{}\beta ^{}\gamma ^{}}.$$
(8.4)
The equation of motion of this theory, if correct, should correspond to the equation for integrability of $`A`$ to a $`G_2`$ metric. Such a quadratic equation is unknown to us so it would be interesting to study further. Notice that for the A-model such a simple cubic theory does not exist.
There is yet another theory in the case of the B-model which has been proposed as a possible equivalent space-time theory, which is a six-dimensional Hitchin functional. This is proposed in and studied and refined in . In the latter paper it is also pointed out that the six-dimensional Hitchin theory has a one-loop gravitational anomaly which again suggests that complex and Kรคhler moduli cannot be treated independently. This agrees nicely with the analysis of our model on $`CY\times S^1`$ and clearly it is worth trying to understand whether our theory on $`CY\times S^1`$ is free of any such one-loop anomalies. What is confusing and begs for clarification is the fact that the six-dimensional theory has a Kodaira Spencer formulation and a Hitchin formulation and both are supposed to reproduce the prepotential (see also ), whereas in seven dimensions, we only have the prepotential itself and that is the Hitchin functional. It would be quite interesting if the 7d Hitchin functional would also be the effective spacetime theory, since that would mean that prepotential obtained from Hitchinโs functional would again be Hitchinโs functional. We clearly need to sort all this out if we want to make progress in โtopological M theoryโ (see also ).
### 8.5 Branes
Though our theory does not have world-sheet instantons (since there are no supersymmetric 2-cycles), it does have supersymmetric branes, namely $`0,3,4`$ and $`7`$-branes, that will give rise to non-perturbative corrections. Presumably, the formulation of topological M-theory is in terms of topological membranes. However, strings and membranes are dual in seven dimensions. It is for these reasons that the 3 brane is specially interesting. Its world-volume theory is a candidate topological membrane theory that might give rise to an alternative definition of a 7d theory ( see also for further discussions of membranes in $`G_2`$ manifolds). In some examples one can see that membranes should play an important role. For example, if one considers topological strings on orientifolds of CY compactifications, one finds a version of Gromow-Witten invariants coming from oriented and unoriented string world-sheets. As the theory is equivalent to M-theory on $`(CY\times S^1)/_2`$, from the M-theory point of view we are counting membranes wrapping the $`S^1`$ . We leave a detailed discussion of the branes in the theory to a future publication.
### 8.6 Open problems and future directions
There are several further open problems. Perhaps the most important one is to find a twisted stress tensor which is crucial for the definition of the topological string beyond genus zero. It is also interesting to understand the geometric meaning of the higher genus amplitudes. In the case of the A-model, the higher genus amplitudes roughly compute the number of holomorphic maps from a genus $`g`$ Riemann surface into the Calabi-Yau. Such an interpretation is less clear for the B-model for $`g>1`$ (the genus $`0`$ result reproduce the special geometry relations and the genus 1 result is related to the holomorphic Ray-Singer torsion). For example, are there interesting indices (like the elliptic genus) that we can define and study in this context? Perhaps related to this, we would like to understand better the localization arguments.
Mirror symmetry for $`G_2`$ manifolds will be interesting to investigate in the context of our topological twist. A version of mirror symmetry for $`G_2`$ manifolds was studied in . In , an analogue of Witten index was introduced that counts the total number of ground states and not just ground states weighted with $`(1)^F`$, where $`F`$ counts the fermion number. This was defined by using a $`Z_2`$ automorphism $`L`$ of the $`G_2`$ algebra under which the currents $`K`$ and $`\mathrm{\Phi }`$ change signs, and the index was defined as $`\mathrm{Tr}(L(1)^F)`$. This index will count the total number of chiral primary states in our topological theory. In fact, in , it was argued that acting with $`L`$ in the left sector and the identity in the right sector corresponds to the mirror automorphism of the $`G_2`$ algebra, which can then be geometrically interpreted as mirror symmetry for $`G_2`$ manifolds.
We list several other related questions that still remain open. For example, are there other relations to the low energy effective action? Is there a Berkovits formulation in three dimensions? Is the Dolbeault-like complex for $`G_2`$ manifolds that corresponds to the BRST cohomology in the left or the right sector useful in other contexts? It is also perhaps worthwhile to investigate more concrete world-sheet models of theories based on the $`G_2`$ algebra, for example using minimal models and discrete torsion, see e.g. . It is also interesting to extend this construction to more general setting which involve turning on the NS-NS background fields. As discussed in , this setup involves a study of $`G_2\times G_2`$ structures, and it would be interesting to understand how our topological twist is modified in this context.
A natural extension of this work is to study topological strings on spin(7) manifolds. This may reveal interesting extensions of Hitchinโs functionals to such manifolds. We will report these results elsewhere .
### Acknowledgments
It is a pleasure to thank Nathan Berkovits, David Berman, Volker Braun, Robbert Dijkgraaf, Anton Gerasimov, Thomas Grimm, Sergei Gukov, Hirosi Ooguri, Samson Shatashvili, Annamaria Sinkovics and Erik Verlinde for useful discussions. We also thank Sheer El-Showk for finding typographical errors in the earlier version of this paper. This research is partially supported by the stiching FOM.
## Appendix A The Coulomb Gas Representation
A useful (though subtle) representation of minimal models is the โCoulomb gasโ representation. Much of the evidence pointing at a possible topological twisting for $`G_2`$ manifolds was constructed in using this approach. For reasons that will become apparent defining the topological theory in this representation is very difficult. Although we proceeded in the main text to define the topological construction in an independent way which avoids many of the complications of the Coulomb Gas Representation, we summarize it here for completeness as well as for a useful source of intuition for the results we obtained in the main text.
In the Coulomb gas representation minimal model primaries are represented as vertex operators in a theory of a scalar coupled to a background charge. The holomorphic energy momentum tensor in such theories is given by
$$T(z)=\frac{1}{2}\left(\varphi (z)\varphi (z)+iQ^2\varphi (z)\right)$$
(A.1)
with central charge
$$c=13Q^2.$$
(A.2)
Primaries are the โvertex operatorsโ
$$V_{n^{}n}(z)e^{i\alpha _{n^{}n}\varphi (z)}$$
(A.3)
where
$$\alpha _{n^{}n}=\frac{1}{\sqrt{2}}[(n^{}1)\alpha _{}+(n1)\alpha _+].$$
(A.4)
The conformal dimension of these operators
$$h(V_{n^{}n})=\frac{1}{2}\alpha _{n^{}n}(\alpha _{n^{}n}+Q).$$
(A.5)
In the Tri-critical Ising model we choose $`Q=\frac{1}{\sqrt{10}}`$ which sets $`\alpha _+=\frac{4}{\sqrt{10}}`$ and $`\alpha _{}=\frac{5}{\sqrt{10}}`$ and one can easily verify that A.5 correctly reproduce the conformal weights inside the tri-critical Ising model.
An important subtlety arises because one can construct two weight $`1`$ vertex operators $`V_\pm V_{\pm 1,1}=e^{i\sqrt{2}\alpha _\pm }`$ called screening operators. Integrating $`V_\pm `$ against the vertex operators A.3 gives screened vertex operators which have the same conformal weight as A.3 but a different โchargeโ under $`\varphi \varphi +const`$. More precisely, these operators are defined as
$$V_{n^{}n}^{r^{}r}(z)=\underset{i=1}{\overset{r^{}}{}}du_i\underset{j=1}{\overset{r}{}}dv_jV_{n^{}n}(z)V_+(u_1)\mathrm{}V_+(u_r^{})V_{}(v_1)\mathrm{}V_{}(v_r)$$
(A.6)
where the contours of the $`u`$ and $`v`$ integrations have been defined carefully in . Each screened vertex operator $`V_{n^{}n}^{r^{}r}`$ correspond to a different conformal block of the operator $`V_{n^{}n}`$. So, for example, in (2.12), the two conformal blocks, in the Coulomb gas picture are given by
$$\mathrm{\Phi }_{2,1}^{}=P_{}V_{21}^{10}P_{}+P_{}V_{21}^{00}P_{},\mathrm{\Phi }_{2,1}^{}=P_{}V_{21}^{00}P_{}+P_{}V_{21}^{10}P_{}$$
(A.7)
where we have been careful to put in projectors $`P_{}`$ and $`P_{}`$. $`P_{}`$ projects to the states corresponding to the first column of the Kac table and the first two entries of the second column, whereas $`P_{}`$ projects to the last two entries of the middle column and the third column of the Kac table. In this way we unambiguously embed the minimal model Hilbert space in the Hilbert space of the scalar field. Similarly, for the conformal blocks of $`\mathrm{\Phi }_{1,2}`$ we have the following Coulomb gas representations:
$$\mathrm{\Phi }_{1,2}^+=P_{}V_{12}^{00}P_{}+P_{}V_{12}^{01}P_{},\mathrm{\Phi }_{1,2}^{}=P_{}V_{12}^{01}P_{}+P_{}V_{12}^{00}P_{}$$
(A.8)
In the Coulomb gas representation of the Tri-critical Ising model, the field $`\varphi `$ has a background charge $`Q=\frac{1}{\sqrt{10}}`$. If we just consider the subspace of the Hilbert space corresponding to the projection $`P_{}`$, we can write $`P_{}V_{12}^{00}P_{}=e^{i\frac{5}{2\sqrt{1}0}\varphi }`$ and then in this sector, insertions of two $`\mathrm{\Sigma }`$ fields on a sphere effectively changes the background charge from
$$Q=\frac{1}{\sqrt{10}}\frac{6}{\sqrt{10}}$$
(A.9)
The central charge of the total CFT changes from $`c=\frac{21}{2}`$ to zero:
$$c=\frac{3}{2}\times 7=\frac{7}{10}+\frac{98}{10}13(\frac{6}{\sqrt{10}})^2+\frac{98}{10}=0$$
(A.10)
which hints strongly at the existence of a topological theory.
Changing the background charge changes the weights of various fields. The change in weight depends on the charge of the field. In fact, since different conformal blocks of the same field carry different charges, their weights shift by different amounts after the twist. The twisting acts differently on the conformal blocks of the same operator. For example, the new weights of some of the blocks after the twist are
$$\begin{array}{cc}\hfill G^{}& 1,G^{}2\hfill \\ \hfill M^{}& 2,M^{}3\hfill \end{array}$$
(A.11)
Using A.5 one finds the conformal weights of Coulomb gas vertex operators in the twisted theory shifted
$$\begin{array}{cc}\hfill V_{21}^{00}=e^{\frac{2i}{\sqrt{10}}},V_{31}^{00}=e^{\frac{4i}{\sqrt{10}}}& \frac{2}{5}\hfill \\ \hfill V_{31}^{00}=e^{\frac{6i}{\sqrt{10}}},\mathrm{๐}& 0\hfill \\ \hfill V_{21}^{10}e^{\frac{2i}{\sqrt{10}}}& \frac{3}{5}\hfill \end{array}$$
(A.12)
Notice that the blocks corresponding to the unscreened vertex operators in the Coulomb gas representation, dressed with the appropriate weight in the remainder CFT of the โchiralโ states 3.1 become weight $`0`$ after the twist. Similar arguments were used in .
A few words about the Coulomb gas approach are however in order. The Hilbert space of the free theory with a background charge is larger than that of the minimal model. To go from the free theory to the minimal model, we need to consider cohomologies of approach BRST operators defined by Felder . So while the Coulomb gas representation is useful in doing computations, it cannot be used to construct new operators unless they commute with Felderโs BRST operators. We thus emphasize that these arguments should be taken as inspirational rather than rigorous.
## Appendix B The $`G_2`$ Algebra
The $`G_2`$ algebra is given by
$$\{G_n,G_m\}=\frac{7}{2}(n^2\frac{1}{4})\delta _{n+m,0}+2L_{n+m}$$
(B.1)
$$[L_n,L_m]=\frac{21}{24}(n^3n)\delta _{n+m,0}+(nm)L_{n+m}$$
(B.2)
$$[L_n,G_m]=(\frac{1}{2}nm)G_{n+m}$$
(B.3)
$$\{\mathrm{\Phi }_n,\mathrm{\Phi }_m\}=\frac{7}{2}(n^2\frac{1}{4})\delta _{n+m,0}+6X_{n+m}$$
(B.4)
$$[X_n,\mathrm{\Phi }_m]=5(\frac{1}{2}nm)\mathrm{\Phi }_{n+m}$$
(B.5)
$$[X_n,X_m]=\frac{35}{24}(n^3n)\delta _{n+m,0}5(nm)X_{n+m}$$
(B.6)
$$[L_n,X_m]=\frac{7}{24}(n^3n)\delta _{n+m,0}+(nm)X_{n+m}$$
(B.7)
$$\{G_n,\mathrm{\Phi }_m\}=K_{n+m}$$
(B.8)
$$[G_n,K_m]=(2nm)\mathrm{\Phi }_{n+m}$$
(B.9)
$$[G_n,X_m]=\frac{1}{2}(n+\frac{1}{2})G_{n+m}+M_{n+m}$$
(B.10)
$$\{G_n,M_m\}=\frac{7}{12}(n^2\frac{1}{4})(n\frac{3}{2})\delta _{n+m,0}+(n+\frac{1}{2})L_{n+m}+(3nm)X_{n+m}$$
(B.11)
$$[\mathrm{\Phi }_n,K_m]=\frac{3}{2}(mn+\frac{1}{2})G_{n+m}3M_{n+m}$$
(B.12)
$$\{\mathrm{\Phi }_n,M_m\}=(2n\frac{5}{2}m\frac{11}{4})K_{n+m}3:G\mathrm{\Phi }:_{n+m}$$
(B.13)
$$[X_n,K_m]=3(m+1)K_{n+m}+3:G\mathrm{\Phi }:_{n+m}$$
(B.14)
$`[X_n,M_m]`$ $`=`$ $`[{\displaystyle \frac{9}{4}}(n+1)(m+{\displaystyle \frac{3}{2}}){\displaystyle \frac{3}{4}}(n+m+{\displaystyle \frac{3}{2}})(n+m+{\displaystyle \frac{5}{2}})]G_{n+m}`$
$`[5(n+1){\displaystyle \frac{7}{2}}(n+m+{\displaystyle \frac{5}{2}})]M_{n+m}+4:GX:_{n+m}`$
$$[K_n,K_m]=\frac{21}{6}(n^3n)\delta _{n+m,0}+3(nm)(X_{n+m}L_{n+m})$$
(B.16)
$$[K_n,M_m]=[\frac{11}{2}(n+1)(n+m+\frac{3}{2})\frac{15}{2}(n+1)n]\mathrm{\Phi }_{n+m}+3:GK:_{n+m}6:L\mathrm{\Phi }:_{n+m}$$
(B.17)
$`\{M_n,M_m\}`$ $`=`$ $`{\displaystyle \frac{35}{24}}(n^2{\displaystyle \frac{1}{4}})(n^2{\displaystyle \frac{9}{4}})\delta _{n+m,0}+[{\displaystyle \frac{3}{2}}(n+m+2)(n+m+3)`$
$`10(n+{\displaystyle \frac{3}{2}})(m+{\displaystyle \frac{3}{2}})]X_{n+m}+[{\displaystyle \frac{9}{2}}(n+{\displaystyle \frac{3}{2}})(m+{\displaystyle \frac{3}{2}})`$
$`{\displaystyle \frac{3}{2}}(n+m+2)(n+m+3)]L_{n+m}4:GM:_{n+m}+8:LX:_{n+m}`$
An important property of the algebra is the fact that it contains a null ideal, generated by
$$๐ฉ=4(GX)2(\mathrm{\Phi }K)4M^2G.$$
(B.19)
This null ideal has various consequences. For example, it allows us to determine the eigenvalue of $`K_0`$ on highest weight states in terms of their $`L_0`$ and $`X_0`$ eigenvalues. Thus, $`K_0`$ is not an independent quantum number in the theory.
In a two-parameter family of chiral algebras was found, with the same generators as the $`G_2`$ algebra. However, the $`G_2`$ algebra is the only one among this family which has the right central charge $`c=21/2`$ and contains the tri-critical Ising model as a subalgebra. The latter is needed for space-time supersymmetry, and therefore the $`G_2`$ algebra appears to be uniquely fixed by these physical requirements.
The representation theory of the $`G_2`$ algebra was studied in some detail in . Both in the NS and R sector there are short and long representations. We will discuss the representations of the latter in the next section C. In the NS sector the short representations correspond to what we called chiral primaries, whereas in the R sector the short representations correspond to R ground states.
Character formulae for the $`G_2`$ algebra are unknown. In the partition functions for string theory on particular non-compact $`G_2`$ manifolds were found, and from these one can extract candidate character formulas for some of the representations of the $`G_2`$ algebra. It would be nice to have general explicit expressions for the characters. One may try to obtain these by using the fact that the $`G_2`$ algebra can be obtained by quantum Hamiltonian reduction (see e.g. ) from the affine super Lie algebra based on $`D(2,1,\alpha )`$, as suggested in . Following the strategy in one expects that the characters can be expressed in terms of highest weight characters of the $`D(2,1,\alpha )`$ affine super Lie algebra, but we have not explored this in this paper.
## Appendix C R sector
In this section we will be completely pedantic. In the R sector we have the following commutation relations of the zero modes ($`L_0`$ commutes with everything)
$`\{G_0,G_0\}`$ $`=`$ $`2(L_0{\displaystyle \frac{7}{16}})`$
$`\{G_0,\varphi _0\}`$ $`=`$ $`K_0`$
$`[G_0,X_0]`$ $`=`$ $`{\displaystyle \frac{1}{4}}G_0+M_0`$
$`\{G_0,M_0\}`$ $`=`$ $`{\displaystyle \frac{1}{2}}(L_0{\displaystyle \frac{7}{16}})`$ (C.1)
$`[G_0,K_0]`$ $`=`$ $`K_0`$
$`[X_0,K_0]`$ $`=`$ $`{\displaystyle \frac{3}{2}}K_03\varphi _0G_0`$
$`[X_0,\varphi _0]`$ $`=`$ $`0`$
$`[X_0,M_0]`$ $`=`$ $`{\displaystyle \frac{21}{16}}G_0{\displaystyle \frac{9}{4}}M_0+4G_0X_0`$ (C.2)
$`[K_0,\varphi _0]`$ $`=`$ $`{\displaystyle \frac{3}{4}}G_0+3M_0`$
$`[K_0,M_0]`$ $`=`$ $`3G_0K_06\varphi _0(L_0{\displaystyle \frac{7}{16}})`$
$`\{\varphi _0,\varphi _0\}`$ $`=`$ $`{\displaystyle \frac{7}{8}}+6X_0`$
$`\{\varphi _0,M_0\}`$ $`=`$ $`{\displaystyle \frac{7}{4}}K_03G_0\varphi _0`$
$`\{M_0,M_0\}`$ $`=`$ $`{\displaystyle \frac{21}{8}}(L_0{\displaystyle \frac{7}{16}})+8(L_0{\displaystyle \frac{7}{16}})X_04G_0M_0.`$ (C.3)
In addition, there is the operator
$$๐ฉ=\frac{3}{2}M_03K_0\varphi _0+6G_0X_0$$
(C.4)
which should be null when acting on highest weight states. To extract this algebra from the operator product expansion one needs to use a suitable normal ordering prescription. One may check that this algebra is consistent with hermiticity, associativity, and yields the right spectrum for $`X_0`$.
To build representations, we first consider a highest weight vector of the form $`|7/16,h_r`$. One may check that $`(\frac{7}{4}G_0+M_0)|7/16,h_r`$ has $`X_0`$ eigenvalue equal to $`99/16`$. This is outside the Kac table for the tri-critical Ising model. Therefore, this vector has to be null. Given this null vector, we find that the representation a priori has four states remaining. Notice that, as we will discuss momentarily, these representations may still be reducible.
We introduce the basis
$$\left(\begin{array}{c}|7/16,h_r\\ (\frac{17}{4}G_0+M_0)|7/16,h_r\\ \varphi _0|7/16,h_r\\ (\frac{17}{4}G_0+M_0)\varphi _0|7/16,h_r\end{array}\right)$$
(C.5)
In this basis the various generators look like (with $`\widehat{l}=L_0\frac{7}{16}`$)
$`G_0`$ $`=`$ $`\left(\begin{array}{cccc}0& 6\widehat{l}& 0& 0\\ \frac{1}{6}& 0& 0& 0\\ 0& 0& 0& 6\widehat{l}\\ 0& 0& \frac{1}{6}& 0\end{array}\right)`$ (C.10)
$`M_0`$ $`=`$ $`\left(\begin{array}{cccc}0& \frac{27}{2}\widehat{l}& 0& 0\\ \frac{7}{24}& 0& 0& 0\\ 0& 0& 0& \frac{27}{2}\widehat{l}\\ 0& 0& \frac{7}{24}& 0\end{array}\right)`$ (C.15)
$`\varphi _0`$ $`=`$ $`\left(\begin{array}{cccc}0& 0& \frac{49}{8}& 0\\ 0& 0& 0& \frac{7}{8}\\ 1& 0& 0& 0\\ 0& \frac{1}{7}& 0& 0\end{array}\right)`$ (C.20)
$`X_0`$ $`=`$ $`\left(\begin{array}{cccc}\frac{35}{16}& 0& 0& 0\\ 0& \frac{3}{16}& 0& 0\\ 0& 0& \frac{35}{16}& 0\\ 0& 0& 0& \frac{3}{16}\end{array}\right)`$ (C.25)
$`K_0`$ $`=`$ $`\left(\begin{array}{cccc}0& 0& 0& \frac{63}{2}\widehat{l}\\ 0& 0& \frac{7}{8}& 0\\ 0& \frac{36}{7}\widehat{l}& 0& 0\\ \frac{1}{7}& 0& 0& 0\end{array}\right).`$ (C.30)
There is a two-parameter family of possible metrics compatible with unitarity, namely
$$g=\left(\begin{array}{cccc}\frac{8a}{49}& 0& ib& 0\\ 0& \frac{288a\widehat{l}}{49}& 0& 36i\widehat{l}b\\ ib& 0& a& 0\\ 0& 36i\widehat{l}b& 0& 36\widehat{l}a\end{array}\right).$$
(C.31)
These representations are not irreducible. Indeed, we can go to an eigenbasis of $`\varphi _0`$. To do this we define a new basis as
$$\left(\begin{array}{cccc}\frac{7i}{\sqrt{8}}& 0& 1& 0\\ \frac{7i}{\sqrt{8}}& 0& 1& 0\\ 0& \frac{7i}{\sqrt{8}}& 0& 1\\ 0& \frac{7i}{\sqrt{8}}& 0& 1\end{array}\right)\left(\begin{array}{c}|7/16,h_r\\ (\frac{17}{4}G_0+M_0)|7/16,h_r\\ \varphi _0|7/16,h_r\\ (\frac{17}{4}G_0+M_0)\varphi _0|7/16,h_r\end{array}\right).$$
(C.32)
Then the generators become
$`G_0`$ $`=`$ $`\left(\begin{array}{cccc}0& 0& 0& 6\widehat{l}\\ 0& 0& \widehat{l}& 0\\ 0& \frac{1}{6}& 0& 0\\ \frac{1}{6}& 0& 0& 0\end{array}\right)`$ (C.37)
$`M_0`$ $`=`$ $`\left(\begin{array}{cccc}0& 0& 0& \frac{27}{2}\widehat{l}\\ 0& 0& \frac{27}{2}\widehat{l}& 0\\ 0& \frac{7}{24}& 0& 0\\ \frac{7}{24}& 0& 0& 0\end{array}\right)`$ (C.42)
$`\varphi _0`$ $`=`$ $`\left(\begin{array}{cccc}\frac{7i}{\sqrt{8}}& 0& 0& 0\\ 0& \frac{7i}{\sqrt{8}}& 0& 0\\ 0& 0& \frac{i}{\sqrt{8}}& 0\\ 0& 0& 0& \frac{i}{\sqrt{8}}\end{array}\right)`$ (C.47)
$`X_0`$ $`=`$ $`\left(\begin{array}{cccc}\frac{35}{16}& 0& 0& 0\\ 0& \frac{35}{16}& 0& 0\\ 0& 0& \frac{3}{16}& 0\\ 0& 0& 0& \frac{3}{16}\end{array}\right)`$ (C.52)
$`K_0`$ $`=`$ $`\left(\begin{array}{cccc}0& 0& 0& \frac{18i}{\sqrt{2}}\widehat{l}\\ 0& 0& \frac{18i}{\sqrt{2}}\widehat{l}& 0\\ 0& \frac{i}{\sqrt{8}}& 0& 0\\ \frac{i}{\sqrt{8}}& 0& 0& 0\end{array}\right).`$ (C.57)
The metric becomes
$$g=\left(\begin{array}{cccc}c_1& 0& 0& 0\\ 0& c2& 0& 0\\ 0& 0& 36c_2\widehat{l}& 0\\ 0& 0& 0& 36c_1\widehat{l}\end{array}\right)$$
(C.58)
where $`c_1,c_2`$ are arbitrary constants related to $`a,b`$ in some way which is not terribly important. We therefore see that the representation splits into two complex conjugate ones which are each two dimensional. For $`\widehat{l}0`$ this is the complete story, i.e. the zero modes are represented as two complex conjugate two-dimensional representations. One is spanned by the first and fourth vector, the other one by the second and the third.
In the case we have R ground states, i.e. $`\widehat{l}=0`$, we see that the system degenerates further. We can consistently decouple the third and fourth vector and find two complex conjugate one-dimensional representations of the algebra. These correspond to the $`h_I=\frac{7}{16}`$ R ground state that is purely internal. In this representation, $`G_0=M_0=K_0=0`$.
The null module generated by the third and fourth vector also provides two one-dimensional complex conjugate representations. Taking $`c_1`$ and $`c_2`$ to scale as $`1/\widehat{l}`$, we see that this gives rise to one-dimensional representations of the form $`|\frac{3}{80},\frac{2}{5}`$. In these representation also $`G_0=M_0=K_0=0`$.
In short, in the R sector we have massless and massive representations. If we combine the left and right movers, things change a little bit. We cannot use eigenvectors of $`\varphi _0`$ and $`\overline{\varphi }_0`$ with nonzero eigenvalue simultaneously, since that is inconsistent with $`\{\varphi _0,\overline{\varphi }_0\}=0`$. The smallest unitary representation of this algebra is two-dimensional. Therefore, combining left and right massless representations leads to a two-dimensional representation. Combining massless and massive to a four-dimensional representation, and combining two massive representations to a eight-dimensional representation.
## Appendix D Decomposition of differential forms into irreps of $`G_2`$
In this appendix, we review the decomposition of differential forms into irreducible representations of the group $`G_2`$. Our discussion follows the one in
For a $`G_2`$ manifold, differential forms of any degree can be decomposed into irreducible representations of $`G_2`$
$`\mathrm{\Lambda }^0=\mathrm{\Lambda }_1^0`$ $`\mathrm{\Lambda }^1=\mathrm{\Lambda }_7^1`$
$`\mathrm{\Lambda }^2=\mathrm{\Lambda }_7^2\mathrm{\Lambda }_{14}^2`$ $`\mathrm{\Lambda }^3=\mathrm{\Lambda }_1^3\mathrm{\Lambda }_7^3\mathrm{\Lambda }_{27}^3`$
This decomposition is compatible with the Hodge star operation, so $`\mathrm{\Lambda }_m^n=\mathrm{\Lambda }_m^{7n}`$. It is useful to define this decomposition into irreducible representations explicitly.
##### 2-forms and 5-forms
The 2-forms decompose into a 7 and 14 of $`G_2`$. These spaces can be characterized as follows:
$`\mathrm{\Lambda }_7^2`$ $`=`$ $`\{\omega \mathrm{\Lambda }^2;(\varphi \omega )=2\omega \}`$
$`\mathrm{\Lambda }_{14}^2`$ $`=`$ $`\{\omega \mathrm{\Lambda }^2;(\varphi \omega )=\omega \}`$
It is useful to write expressions for projector operators $`\pi _7`$ and $`\pi _{14}`$. These project onto the appropriate subspaces:
$`\pi _7^2(\omega )`$ $`=`$ $`{\displaystyle \frac{\omega +(\varphi \omega )}{3}}`$
$`\pi _{14}^2(\omega )`$ $`=`$ $`{\displaystyle \frac{2\omega (\varphi \omega )}{3}}`$
where the superscript $`2`$ on $`\pi _k^2`$ indicates that this is the projector when acting on 2-forms. In local coordinates, these can be written as
$`(\pi _7^2)_{ab}^{de}`$ $`=`$ $`6\varphi _{ab}^c\varphi _c^{de}=4\varphi _{ab}^{de}+{\displaystyle \frac{1}{6}}(\delta _a^d\delta _b^e\delta _a^e\delta _b^d)`$
$`(\pi _{14}^2)_{ab}^{ef}`$ $`=`$ $`4\varphi _{ab}^{ef}+{\displaystyle \frac{1}{3}}(\delta _a^e\delta _b^f\delta _a^d\delta _b^e)`$
Similarly, for 5 forms, we have the decomposition:
$`\mathrm{\Lambda }_7^5`$ $`=`$ $`\{\omega \mathrm{\Lambda }^5;\varphi \omega =2\omega \}`$
$`\mathrm{\Lambda }_{14}^5`$ $`=`$ $`\{\omega \mathrm{\Lambda }^5,\varphi \omega =\omega \}`$
which implies the projectors
$`\pi _7^5(\omega )`$ $`=`$ $`{\displaystyle \frac{\omega +\varphi \omega }{3}}`$
$`\pi _{14}^5(\omega )`$ $`=`$ $`{\displaystyle \frac{2\omega \varphi \omega }{3}}`$
##### 3-forms and 4-forms
The three forms decompose into 1, 7 and 27 dimensional representations of $`G_2`$. Explicitly, these spaces are given by
$`\mathrm{\Lambda }_1^3`$ $`=`$ $`\{\omega \mathrm{\Lambda }^3:\varphi ((\varphi \omega ))=7\omega \}`$
$`\mathrm{\Lambda }_7^3`$ $`=`$ $`\{\omega \mathrm{\Lambda }^3;(\varphi (\varphi \omega ))=4\omega \}`$
$`\mathrm{\Lambda }_{27}^3`$ $`=`$ $`\{\omega \mathrm{\Lambda }^3;\varphi \omega =\varphi \omega =0\}`$
We also define projection operators:
$`\pi _1^3(\omega )`$ $`=`$ $`{\displaystyle \frac{1}{7}}\varphi ((\varphi \omega ))`$
$`\pi _7^3(\omega )`$ $`=`$ $`{\displaystyle \frac{1}{4}}((\varphi (\varphi \omega )))`$
$`\pi _{27}^3(\omega )`$ $`=`$ $`\omega \pi _1^3(\omega )\pi _7^3(\omega )`$
For four forms, we have the decomposition
$`\mathrm{\Lambda }_1^4`$ $`=`$ $`\{\omega \mathrm{\Lambda }^4:\varphi ((\varphi \omega ))=7\omega \}`$
$`\mathrm{\Lambda }_7^4`$ $`=`$ $`\{\omega \mathrm{\Lambda }^4;\varphi (\varphi \omega )=4\omega \}`$
$`\mathrm{\Lambda }_{27}^4`$ $`=`$ $`\{\omega \mathrm{\Lambda }^4;\varphi \omega =\varphi \omega =0\}`$
and the projectors
$`\pi _1^4(\omega )`$ $`=`$ $`{\displaystyle \frac{1}{7}}\varphi ((\varphi \omega ))`$
$`\pi _7^4(\omega )`$ $`=`$ $`{\displaystyle \frac{1}{4}}(\varphi (\varphi \omega ))`$
$`\pi _{27}^4(\omega )`$ $`=`$ $`\omega \pi _1^4(\omega )\pi _7^4(\omega )`$
There are natural $`G_2`$-equivariant isomorphisms between these spaces. For example, the map $`\omega \varphi \omega `$ is an isomorphism between $`\mathrm{\Lambda }_r^p\mathrm{\Lambda }_r^{p+3}`$ if $`\varphi \omega _p`$ is non-zero when $`\omega \mathrm{\Lambda }_r^p`$:
$`\mathrm{\Lambda }_1^0`$ $``$ $`\mathrm{\Lambda }_1^3\mathrm{\Lambda }_7^1\mathrm{\Lambda }_7^4`$
$`\mathrm{\Lambda }_7^2`$ $``$ $`\mathrm{\Lambda }_7^5\mathrm{\Lambda }_{14}^2\mathrm{\Lambda }_{14}^5`$
$`\mathrm{\Lambda }_7^3`$ $``$ $`\mathrm{\Lambda }_7^6\mathrm{\Lambda }_1^4\mathrm{\Lambda }_1^7`$
Also, the map $`\omega \varphi \omega `$ is an isomorphism between $`\mathrm{\Lambda }_r^p\mathrm{\Lambda }_r^{p+4}`$ when $`\varphi \omega _p`$ is non-zero when $`\omega \mathrm{\Lambda }_r^p`$:
$`\mathrm{\Lambda }_1^0`$ $``$ $`\mathrm{\Lambda }_1^4\mathrm{\Lambda }_7^1\mathrm{\Lambda }_7^5`$
$`\mathrm{\Lambda }_7^2`$ $``$ $`\mathrm{\Lambda }_7^6\mathrm{\Lambda }_1^3\mathrm{\Lambda }_1^7`$
## Appendix E Some correlation functions
We can use the expression (4.3) to compute some correlation functions in the twisted theory in terms of correlation functions of the untwisted theory. For example, the two point function of operators
$$๐ช_2=\mathrm{\Phi }_{2,1}\psi _h,๐ช_3=\mathrm{\Phi }_{3,1}\psi _h$$
can be written in terms of a four-point function of the tri-critical Ising model
$`๐ช_2(z_1)๐ช_3(z_2)`$ $`=`$ $`z_1^{\frac{1}{2}}z_2^1(z_1z_2)^{2h}\times \mathrm{\Phi }_{1,2}(\mathrm{})\mathrm{\Phi }_{2,1}(z_1)\mathrm{\Phi }_{3,1}(z_2)\mathrm{\Phi }_{1,2}(0)_{\mathrm{tri}\mathrm{critical}}`$
$`=`$ $`{\displaystyle \frac{c}{(z_1z_2)^{2h\frac{4}{5}}}}`$
where $`c`$ is a constant. This is independent of position if $`h=\frac{2}{5}`$, which is what we need for the operators $`๐ช_2`$ and $`๐ช_3`$ to be chiral in the topological theory. This correlation functions gets contributions from only one conformal block, precisely the one that is kept in the topological theory. On the other hand, consider the two point function of operators whose tri-critical Ising model weight is $`\frac{1}{10}`$:
$$๐ช=\mathrm{\Phi }_{2,1}\psi _h$$
The two point function of this operator with itself can be written in terms of a four-point function of the tri-critical Ising model:
$`๐ช(z_1)๐ช(z_2)`$ $`=`$ $`z_1^{\frac{1}{2}}z_2^{\frac{1}{2}}(z_1z_2)^{2h}\times \mathrm{\Phi }_{1,2}(\mathrm{})\mathrm{\Phi }_{2,1}(z_1)\mathrm{\Phi }_{2,1}(z_2)\mathrm{\Phi }_{1,2}(0)_{\mathrm{tri}\mathrm{critical}}`$
$`=`$ $`{\displaystyle \frac{c}{(z_1z_2)^{2h+\frac{1}{5}}}}\times {\displaystyle \frac{z_1+z_2}{z_1z_2}}`$
This is not even translationally invariant! However, it is easy to see that the conformal block that contributes to this correlation function is
$$\mathrm{\Phi }_{1,2}๐ช^{}๐ช^{}\mathrm{\Phi }_{1,2}$$
but $`๐ช^{}`$ is not a chiral operator. Correlation functions of chiral operators obey all the properties of a usual CFT. However, correlation functions of non-chiral operators in the twisted theory are not that of a CFT. This is qualitatively different from what happens in the usual $`๐ฉ=2`$ twisting. In that case, the twisted theory makes sense as a CFT, even before we restrict ourselves to chiral operators. This intermediate CFT does not seem to exist for us.
## Appendix F Spectral flow and the twist
Whether or not the twisted stress tensor exists, and if so what its precise form is remains for now an open problem. In the case of Calabi-Yau manifolds, the existence of spectral flow was useful in order to construct the twisted stress tensor, so it is worth considering what precisely the analogue of spectral flow is in our case.
Spectral flow, a word used rather loosely, refers to a particular isomorphism between the R and NS sector of an $`N=2`$ conformal field theory. What it does is easily illustrated in case of a free scalar field $`\phi `$. Denote by $`\widehat{p}=i\phi `$ the zero mode of the momentum operator, and by $`\widehat{x}`$ the conjugate coordinate. Then spectral flow by the amount $`\eta `$ is simply implemented by the operator
$$S_1=e^{i\eta \widehat{x}}.$$
(F.1)
Spectral flow maps representations with momentum eigenvalue $`p`$ to representations with momentum eigenvalue $`p+\eta `$. If we bosonize the $`U(1)`$ current in $`N=2`$ theories then this $`S_1`$ precisely implements what is usually referred to as spectral flow.
This is not quite the same as the statement that some particular R operator generates spectral flow. In that case, we are talking about an operator in the theory, and not a simple object constructed out of zero modes only such as $`S_1`$. It is this full operator, and not $`S_1`$, that appears in the generator of space-time supersymmetry. It is again easy to illustrate this in the case of a free scalar field. Instead of $`S_1`$ we consider the operator
$$S_2=\frac{dz}{z^{\eta q+1}}e^{i\eta \varphi }:_p_{\eta +p}$$
(F.2)
acting on representations with momentum eigenvalue $`p`$ and mapping them to representations of eigenvalue $`p+\eta `$. On highest weight states, $`S_1`$ and $`S_2`$ are identical, but on descendants they are not. The new stress tensors obtained by spectral flow are obtained using $`S_1`$. One can also define new stress tensors using the action of $`S_2`$, simply as $`L_n^{}=S_2^1L_nS_2`$, but this is not usually done. One can explicitly work out the difference between the two prescriptions, but that is not very insightful. The modes of the twisted stress tensors of the A and B-model are linear combinations of the modes of the initial stress tensor and its spectrally flown version. This is spectral flow with respect to $`S_1`$. Whether the twisted stress tensor have any relation to the new stress tensor obtained through $`S_2`$ is not known.
In the case of $`G_2`$ manifolds, the situation is different. We no longer have a version of $`S_1`$, but we do have a version of $`S_2`$, where the exponential of the field is now replaced by the R vertex operator $`V_{7/16,+}`$. It maps chiral primaries to R ground states and vice versa. It should induce an isomorphism between the NS and R sector of the theory, otherwise the theory would not be space-time supersymmetric. In particular, this implies that we can define a new stress tensor in say the NS sector via $`L_n^{}=S_2^1L_nS_2`$. Clearly, $`L_0^{}`$ annihilates all chiral primaries and is a good candidate for a the zero mode of a twisted stress tensor. Whether the highest modes of $`L_n^{}`$ can also be used to construct the modes of a twisted stress tensor still remain to be worked out, even in the case of Calabi-Yau manifolds. We leave this as an interesting direction to explore.
|
warning/0506/hep-th0506215.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
The D1-D5-P system, made of $`N_1`$ D1-branes, $`N_5`$ D5-branes and $`N_P`$ units of momentum, has been an ideal arena for studying microscopic physics of black holes in string theory . The $`N_P=0`$ case (the D1-D5 system) is not classically a black hole because its horizon vanishes at the supergravity level, whereas its microscopic entropy computed from the dual CFT is finite: $`S=2\sqrt{2}\pi \sqrt{N}`$ ($`T^4`$ compactification) and $`S=4\pi \sqrt{N}`$ (K3 compactification), where $`NN_1N_5`$. In a recent beautiful paper by Dabholkar , it was shown that heterotic 4D black hole with classically vanishing horizon, which is dual to the above D1-D5 system compactified on $`T^2\times \mathrm{K3}`$, becomes a black hole with string-size event horizon once stringy $`R^2`$ corrections to the supergravity action are taken into account, and that the $`R^2`$-corrected macroscopic entropy<sup>1</sup><sup>1</sup>1The $`R^2`$-corrected macroscopic entropy was derived in a sequence of papers . of such โsmallโ black hole agrees with the microscopic entropy, as predicted by Sen . Further references for these developments include .
If we consider the D1-D5 system compactified instead on $`S^1\times \mathrm{K3}`$ and add angular momentum $`J=๐ช(N)`$ to it, the brane worldvolume starts to look like a ring from the 5D viewpoint, rather than a pointlike object. Therefore, this D1-D5-J system is expected to be described by a โsmallโ version of the black ring in 5D supergravity , once we consider stringy corrections to the supergravity action. However, a systematic framework for studying stringy corrections to the 5D supergravity, like the one for 4D , has not been available so far, so we cannot directly study the stringy corrections to this system.
Recently, a new connection between 4D and 5D black objects was proposed , which relates the partition function of a 5D black hole/ring with that of a 4D black hole. Although this โ4D-5D connectionโ is based on supergravity analysis, it is expected to hold even if we take into account stringy corrections, since it is natural to expect that a BPS solution interpolating 4D and 5D objects exists even if stringy corrections are included, and the continuous-moduli-independence of entropy is the property of the microscopic theory, not of supergravity. Indeed, the microscopic entropy of 5D black ring agrees with that of 4D black hole including quantum corrections . This suggests that 5D black holes/rings which classically have no horizon will generally develop finite small horizon.
Therefore, one naturally wants to apply the 4D-5D connection to the D1-D5-J ring system on $`S^1\times \mathrm{K3}`$ to study its properties using 4D techniques. However, the 4D-5D connection is not directly applicable, since one cannot get to the duality frame in which one can make use of the 4D-5D connection via simple $`S`$-duality and $`T`$-duality along $`S^1`$ directions in type II. In this note, we map by $`U`$-duality the D1-D5-J system into a duality frame in which the 4D-5D connection is applicable, and relate it with a small non-rotating black hole in 4D. The microscopic entropy counting is well-understood for both these 4D and 5D configurations. On the macroscopic side, the horizon geometry of the 4D small non-rotating black hole is well-understood whereas the geometry of the original D1-D5-J system, which is expected to be a 5D small black ring, is not well-understood. By connecting this D1-D5-J system with the geometry of the 4D small black hole that has a small horizon, we give an indirect evidence of event horizon showing up in the 5D small black ring by stringy corrections to the supergravity action.
The organization of this paper is as follows. In section 2, we study the microscopic entropy of the D1-D5-J system, and clarify the limit in which it can be regarded as a ring with a well-defined profile. In section 3, we present a duality chain which relates the D1-D5-J system and a 4D non-rotating small black hole via the 4D-5D connection. The entropy of the 4D small black hole agrees with the microscopic entropy of the D1-D5-J system, which justifies applying the 4D-5D connection for the small black objects. Also, this 4D small black hole has non-vanishing horizon because of stringy corrections to the supergravity action. These results suggest that event horizon appears for the 5D small black ring by stringy corrections to the supergravity action. In section 4, We conclude with some comments on future directions.
The D1-D5-J small black ring system and its entropy was also studied recently by Kraus and Larsen from a different point of view.
## 2 D1-D5-J system
IN order to understand the charges and dipole charges that appear in the D1-D5-J system, it is easiest to start from heterotic string compactified in $`^{1,4}\times S^1\times T^4`$. Let us take $`^{1,4}`$, $`S^1`$, and $`T^4`$ directions to be $`01234`$, $`5`$, and $`6789`$ directions, respectively. We will write the $`S^1`$ as $`S_5^1`$ henceforth. We wrap $`N_F`$ F1โs along $`S_5^1`$ and put on it $`N_P`$ units of linear momentum along $`S_5^1`$. From Virasoro constraint and BPS condition, we should impose $`N_L=1N_FN_P`$, where $`N_L`$ is the left-moving oscillation number. Since $`N_L>0`$, we should choose $`N_FN_P<0`$. Furthermore, we give $`J=n_p`$ units of angular momentum to the system in the $`\psi `$ direction, where $`\psi `$ is the angular direction on the 1-2 plane. Now the F1 worldvolume is a helix or a coil wound on a cylinder spanned by $`\psi `$ and $`x^5`$, moving upwards along the $`x^5`$ axis. From the 5D (012345) point of view, the system looks like a ring in the $`\psi `$ direction. The configuration is as follows:
$`\begin{array}{cccccc}& & \psi & S_5^1& T^4& \\ & & & & & \\ N_P& P& & & & \\ N_F& F1& & & & \\ n_p& p& & & & \\ n_f& f1& & & & \end{array}`$ (2.6)
Here, โ$``$โ denotes wrapped directions, while โ$``$โ denotes smeared directions. $`P,p`$ denote momenta and $`N_P,n_p`$ are the corresponding momentum numbers. $`F1,f1`$ denote fundamental string and $`N_F,n_f`$ are the corresponding winding numbers. The lowercase letters mean that they are along the contractible direction $`\psi `$. The momentum number $`n_p`$ along $`\psi `$ is nothing but angular momentum number $`J`$. We will refer to this system (2.6) as the FP system.
The maximum angular momentum we can have is $`J_{\mathrm{max}}=N=|N_FN_P|=N_FN_P`$, which is the case when the F1 worldvolume is a perfect helix of radius $`\sqrt{J_{\mathrm{max}}}`$. In the corresponding classical solution , the F1 winds once around the $`\psi `$ circle while it winds $`N_F`$ times the $`x^5`$ direction $`S_5^1`$. Therefore, $`n_f=1`$ if $`J=J_{\mathrm{max}}`$. For $`J<J_{\mathrm{max}}`$, the F1 worldvolume starts to fluctuate around the perfect helix, and it is not obvious that the winding number $`n_f`$ is well-defined. We will discuss later in what limit $`n_f`$ is well-defined.
The system (2.6) is dual to the well-studied D1-D5 system in type IIB compactified on $`S^1\times \mathrm{K3}`$, by the following chain of dualities: heterotic/IIA duality, $`T(5)`$<sup>2</sup><sup>2</sup>2$`T`$-duality along the axis of such helical objects was discussed in ., and then $`S`$. The resulting configuration is:
$`\begin{array}{cccccccc}& & & & \psi & S_5^1& \mathrm{K3}& \\ & & & & & & & \\ N_1& =& N_P& D1& & & & \\ N_5& =& N_F& D5& & & & \\ n_p& =& n_p& p& & & & \\ n_{kk}& =& n_f& kk& & S^1& & \end{array}`$ (2.12)
Now the 6789 direction is K3, and โ$`S^1`$โ means the special circle of the KK monopole. The numbers of D1- and D5-branes are $`N_1=N_P`$ and $`N_5=N_F`$. Note that the relation between the F1 number in heterotic string and the NS5 number in type IIA involves a minus sign . It is consistent with the fact that we need $`N_1N_5>0`$ for susy on the type II side. This system is dual to a 1+1 dimensional $`๐ฉ=(4,4)`$ CFT, which is sigma model with target space $`\mathrm{K3}^N/S_N`$, $`N=N_1N_5`$ .
Now, let us count the entropy of the D1-D5-J system (2.12), or equivalently, the FP system (2.6). Let us here take the FP description, which is a special case of the computation done in . We need to count the number of states generated by 24 left-moving transverse bosons $`\alpha _n^i`$, $`i=1,\mathrm{},24`$, at level $`N=|N_FN_P|=N_FN_P`$ that have angular momentum $`J=n_p`$ in the 1-2 plane.
If we write
$`\alpha _n^{i=1}`$ $`=\sqrt{n/2}(a_n^++a_n^{}),\alpha _n^{i=2}=i\sqrt{n/2}(a_n^+a_n^{}),n=1,2,\mathrm{},`$ (2.13)
then the level $`N`$ and angular momentum $`J`$ are
$`N`$ $`={\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}{\displaystyle \underset{i=1}{\overset{24}{}}}\alpha _n^i\alpha _n^i={\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}\left[n(a_n^+{}_{}{}^{}a_{n}^{+}+a_n^{}{}_{}{}^{}a_{n}^{})+{\displaystyle \underset{i=3}{\overset{24}{}}}\alpha _n^i\alpha _n^i\right],`$ (2.14)
$`J`$ $`=i{\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}{\displaystyle \frac{1}{n}}(\alpha _n^1\alpha _n^2\alpha _n^2\alpha _n^1)={\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}\left[a_n^+{}_{}{}^{}a_{n}^{+}a_n^{}{}_{}{}^{}a_{n}^{}\right].`$ (2.15)
We can compute the entropy $`S(N,J)`$ by studying the partition function
$`Z=\mathrm{Tr}[e^{\beta (N+\lambda J)}]={\displaystyle \underset{N,J}{}}d_{N,J}q^Nz^J.`$ (2.16)
Here $`\lambda `$ is the chemical potential conjugate to $`J`$, and $`q=e^\beta ,z=e^{\beta \lambda }`$. The evaluation of $`d_{N,J}`$ was done in , and the leading term of the entropy for $`N1`$, $`J=๐ช(N)`$ is
$`S`$ $`=\mathrm{log}d_{N,J}=4\pi \sqrt{N|J|}.`$ (2.17)
One sees from (2.17) that the only effect of $`J0`$ is to replace $`N`$ in the entropy formula with $`NJ`$. Here we assumed $`J>0`$, $`J=๐ช(N)`$. This means that, BoseโEinstein condensation of $`J`$ $`a_{n=1}^+`$ particles occurs and the whole angular momentum $`J`$ is carried by the condensate. The remaining particles have level $`NJ`$ but no net angular momentum. In other words, in the ensemble with level $`N`$ and angular momentum $`J>0`$, $`J=๐ช(N)`$, the states that contribute to entropy are of the form
$`\underset{\text{BoseโEinstein condensate}}{\underset{}{(a_{n=1}^+{}_{}{}^{})^J\text{}}}\times \underset{\text{states that are responsible for entropy of the ensemble with level }NJ\text{ and no angular momentum}}{\underset{}{{\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}\left[{\displaystyle \underset{i=\pm ,3\mathrm{}24}{}}(\alpha _n^i)^{N_{ni}}\right]|0}}.`$ (2.18)
The state $`(a_{n=1}^+{}_{}{}^{})^J|0`$, $`J>0`$ represents (classically) a fundamental string that goes once around the circle in the 1-2 plane of radius $`\sqrt{J}`$, which is large if $`J1`$ . Clearly, this state has $`n_f=1`$. Acting on this state by the operator $`_{n=1}^{\mathrm{}}\left[_{i=\pm ,3\mathrm{}24}(\alpha _n^i)^{N_{ni}}\right]`$ in (2.18) makes the fundamental string fluctuate around this circle. If this fluctuation is much smaller than the radius of the circle $`\sqrt{J}`$, the winding number $`n_f`$ is well-defined and $`n_f=1`$. Because $`_{ni}N_{ni}=NJ`$, statistical mechanics tells us that $`N_{ni}\sqrt{NJ}`$ for $`n=๐ช(1)`$. This means that the size of the fluctuation is of order $`(NJ)^{1/4}`$, which is much smaller than the radius of the circle $`\sqrt{J}`$ if
$`N,J1`$ $`,J=๐ช(N).`$ (2.19)
In this limit, the system is expected to become a small black ring,<sup>3</sup><sup>3</sup>3More detailed analysis shows that, if $`J=๐ช(N^\gamma ),`$ $`1/2<\gamma 1`$, BoseโEinstein condensation occurs and the system is expected to become a small black ring in 5D. On the other hand, if $`\gamma 1/2`$, BoseโEinstein does not occur and the system is expected to become a small rotating black hole in 5D . and talking about winding number along $`\psi `$ makes sense, so we can say $`n_f=1`$. This relation between the black hole/ring transition and BoseโEinstein condensation is very reminiscent of the microscopic description of the large black ring in .
If $`J<0`$, then the above argument all goes through if we replace $`(a_{n=1}^+{}_{}{}^{})^J`$ with $`(a_{n=1}^{}{}_{}{}^{})^{|J|}`$. In this case, $`n_f=1`$. So, the $`|J|`$ in (2.17) can be replaced with $`|n_p|=n_fn_p`$.
Therefore, the entropy of the system (2.6) or (2.12) can be written as
$`S_{\mathrm{micro}}`$ $`=4\pi \sqrt{N_FN_Pn_fn_p}=4\pi \sqrt{N_1N_5+n_pn_{kk}},`$ (2.20)
where in the last equality we used the relation between charges listed in (2.12).
The entropy counting can be done also in the D1-D5 frame (2.12) using the dual CFT. If we replace the oscillators $`\alpha _n^i`$ with the chiral primaries of the $`๐ฉ=(4,4)`$ CFT , the counting can be done in a completely identical way.
## 3 4D-5D connection
Now, we would like use the 4D-5D connection in order to deform the 5D small black ring (2.6) (or equivalently (2.12)) into a 4D small black hole.
The obstacle to doing that in a straightforward manner is that $`T`$-duality along the $`S_5^1`$ direction and $`S`$-duality in type II string will not take the system (2.12) to a duality frame in which the 4D-5D connection as derived in is applicable; one cannot avoid having unwanted nonzero charges. Fortunately, type IIA string on $`S^1\times \mathrm{K3}`$ has $`O(5,21;)`$ symmetry as a part of the $`U`$-duality group <sup>4</sup><sup>4</sup>4The full U-duality group is $`O(5,21;)\times _2`$ ., which interchanges charges so that we can use the 4D-5D connection. Or equivalently, we start from heterotic configuration (2.6), where we have $`O(5,21;)`$ symmetry as $`T`$-duality. After $`T`$-dualizing the charges appropriately in heterotic string, we can then go to type II by heterotic/IIA duality.
Either way, after such a chain of dualities, we end up with the following configuration in type IIA:
$`\begin{array}{ccccc}& & \psi & S_5^1& \mathrm{K3}\\ & & & & \\ \hfill N_P& D2& & & \alpha _2\\ \hfill N_F& D2& & & \alpha _3\\ \hfill n_p& p& & & \\ \hfill n_f& ns5& & & \end{array}`$ (3.6)
Here $`\alpha _a`$ are 2-cycles in K3, i.e., $`\alpha _aH_2(\mathrm{K3})`$, $`a=2,3,\mathrm{},23`$ (the reason for reserving the index 1 will become clear shortly). The $`F,P`$ in (2.6) have been mapped into D2-branes wrapping particular two 2-cycles, $`\alpha _2`$ and $`\alpha _3`$.
Which particular two 2-cycles $`\alpha _2`$, $`\alpha _3`$ should the D2-branes wrap? This question can be answered as follows. First, let us recall how the $`O(5,21;)`$ $`T`$-duality group arises in heterotic string. In heterotic string theory on $`S^1\times T^4`$, there are 21 left- and 5 right-moving momenta in the internal directions. These momenta are quantized electric charges from the 5D point of view. They form the 26-dimensional Narain lattice with signature (5,21), and the $`O(5,21;)`$ $`T`$-duality rotates the charges in this lattice. $`N_P`$ and $`N_F`$ in (2.6) are two such electric charges of heterotic string, and in the 2-dimensional sublattice of the Narain lattice in which these charges live, the metric is proportional to $`(\genfrac{}{}{0pt}{}{01}{10})`$ (recall that the left- and right-moving momenta in the $`x^5`$ direction are $`p_{L,R}^5=N_P/R\pm N_FR/\alpha ^{}`$, and the invariant form contains $`(p_L^5)^2(p_R^5)^2=4N_PN_F/\alpha ^{}`$ ). Correspondingly, there is $`O(5,21;)`$ $`U`$-duality group on the IIA side. This $`O(5,21;)`$ group contains $`O(3,19;)`$ subgroup that interchanges the D2-branes wrapping 2-cycles in K3. Because there are 22 2-cycles $`\alpha _a,a=2,\mathrm{},23`$ in K3, the D2-brane charges live in 22-dimensional lattice $`H_2(\mathrm{K3},)`$. The metric for this charge lattice is the intersection number $`C_{ab}`$ of 2-cycles in K3, which is known to have signature $`(3,19)`$ . So, the answer to the question above is: we should choose the 2-cycles $`\alpha _2,\alpha _3`$ so that, in the 2-dimensional sublattice of the lattice $`H_2(\mathrm{K3},)`$ in which they live, the metric should be $`(\genfrac{}{}{0pt}{}{01}{10})`$. In other words, the intersection numbers $`C_{ab}`$ should satisfy
$`C_{22}=C_{33}=0,C_{23}=1.`$ (3.7)
Now we are ready to use the 4D-5D connection. Uplifting (3.6) to M-theory,
$`\begin{array}{cccccccc}& & & & \psi & S_5^1& \mathrm{K3}& S_{10}^1\\ & & & & & & & \\ q_2& =& N_P& M2& & & \alpha _2& \\ q_3& =& N_F& M2& & & \alpha _3& \\ 2J_L^3& =& n_p& p& & & & \\ p^1& =& n_f& m5& & & & \end{array}`$ (3.13)
where $`J_L^3`$ is the $`SU(2)_LSO(4)`$ charge. This can be thought of as a 5D small black ring (internal directions are $`S_5^1,\mathrm{K3},S_{10}^1`$), and is simply a special case of the configurations considered in . So, by the (string corrected) 4D-5D connection, there is a way to continuously deform this 5D ring, by way of a Taub-NUT geometry, into a 5D black string which from the 4D point of view is a black hole. The entropy of the 5D ring and that of the 4D hole are identical since entropy cannot change in such an adiabatic process. In this process, the role of the Taub-NUT is just a โcatalystโ to wind the ring around the M-circle and it can be removed when we have reached the 4D configuration, since the entropy is independent of the D6 charge. Practically, one can say that we can reinterpret the contractible circle $`\psi `$ as a non-contractible M-circle, if the relation between 4d and 5d charges (Eq. (3.20) below) is correctly taken into account. Compactifying now on this non-contractible M-circle, we obtain a 4D small non-rotating black hole:
$`\begin{array}{cccccccc}& & & & S_5^1& \mathrm{K3}& S_{10}^1& \\ & & & & & & & \\ q_2& =& N_P& D2& & \alpha _2& & \\ q_3& =& N_F& D2& & \alpha _3& & \\ q_0& =& n_p& d0& & & & \\ p^1& =& n_f& d4& & & & \end{array}`$ (3.19)
where we used the relation between 5D and 4D charges ,
$`p_{4D}^A=p_{5D}^A,q_A^{4D}=q_A^{5D}3D_{ABC}p^Bp^C=q_A^{5D},q_0^{4D}=2J_L^3.`$ (3.20)
The reserved index 1 is now for the 2-cycle $`T^2=S_5^1\times S_{10}^1`$.
The object with charges (3.19) does not have a classical horizon, but develops a finite horizon when $`R^2`$ corrections are taken into account . The macroscopic (Bekenstein-Hawking-Wald) entropy is
$`S_{\mathrm{macro}}=2\pi \sqrt{{\displaystyle \frac{\widehat{q}_0c_{2A}p^A}{6}}},\widehat{q}_0=q_0+{\displaystyle \frac{1}{12}}\widehat{D}^{AB}q_Aq_B,D_{AB}=D_{ABC}p^C.`$ (3.21)
Here $`c_{2A}`$ are the coefficients of the second Chern class, and $`D_{ABC}=\frac{1}{6}C_{ABC}`$, where $`C_{ABC}`$ is the triple intersection number. The matrix $`\widehat{D}^{AB}`$ is the inverse of $`D_{AB}`$ in the subspace orthogonal to its kernel.
In the present case, $`c_{2,A=1}=24`$ and $`C_{1ab}=C_{ab}`$, $`a,b=1,\mathrm{},23`$ with the only relevant values given in (3.7). Therefore, the macroscopic entropy (3.21) is computed as
$`S=4\pi \sqrt{q_0p^1+q_2q_3}=4\pi \sqrt{N_PN_Fn_pn_f}`$ (3.22)
Thus we reproduced the microscopic entropy (Eq. (2.20)) of the D1-D5-J system from the macroscopic entropy of the 4D small black hole. This supports the validity of the 4D-5D connection that we used to arrive at the 4D configuration (3.19), even when stringy corrections are crucial.
## 4 Conclusion and Discussion
The D1-D5-J system is expected to develop a finite horizon once stringy $`R^2`$ corrections to the supergravity action are taken into account, and become a 5D small black ring. We clarified the limit, $`|J|\sqrt{N}`$, in which the D1-D5-J system can be regarded as a ring with a well-defined profile, and showed that we can relate this system by duality transformations and the 4D-5D connection to a 4D small non-rotating black hole, whose horizon structure is well-understood. This gives an indirect evidence that event horizon appears in the 5D small black ring by stringy corrections. We also checked that the entropy of the 4D small black hole agrees with the microscopic entropy of the D1-D5-J system, which can be regarded as a consistency check.
A novel idea for understanding black hole physics was put forward by Mathur and collaborators (for a review see ), who conjectured that a black hole is not a singularity surrounded by empty space and horizon, but an ensemble of smooth but complicated (possibly quantum) geometries inside the stretched horizon. This bold conjecture has been established for the BPS D1-D5 system, for which microstate geometries have been explicitly written down at the supergravity level .<sup>5</sup><sup>5</sup>5The supergravity microstate solutions obtained in include fluctuations in the noncompact $`^4`$ directions, but not in the internal $`T^4`$ or K3 directions. In this Mathur picture, the entropy of a black hole can be estimated from the area of the โstretched horizon,โ defined as the surface on which the microstate geometries start to differ from each other. In , an ensemble of D1-D5 microstate geometries with angular momentum $`J`$ were studied, and the entropy was correctly estimated to be $`\sqrt{N_1N_5|J|}`$ from this โstretched horizonโ area. It is interesting to study the relation between their โstretched horizonโ and the non-vanishing horizon that we argued to appear in the 5D small black ring by stringy corrections.
The above definition of the stretched horizon by Mathur is a qualitative one, and in order to determine the entropy of a black hole from an area law including the numerical factor, it is desirable to define the stretched horizon in a more precise way. It is also interesting to study the relation between this definition of the stretched horizon and other definitions, for example, as the distance at which a probe D-brane starts to become indistinguishable from the D-branes that compose the background black hole, due to thermal effect .
Another aspect of the D1-D5-J system is that, being dual to a configuration of fundamental heterotic string, it is an example of systems for which the microscopic entropy can be computed very precisely . It is interesting to analyze this system further to study the recently proposed relation between the black hole partition function and the topological string partition function .
## Acknowledgments
It is pleasure to thank Atish Dabholkar and Ashoke Sen for helpful conversations. We also would like to thank Per Kraus for useful comments on the early draft of this note. N.I. would like to thank friends in Harvard and Caltech string theory group for their very nice hospitality where this work was done. The work of M.S. was supported in part by Department of Energy grant DE-FG03-92ER40701 and a Sherman Fairchild Foundation postdoctoral fellowship. Finally we are happy to thank the people of India for their generosity.
|
warning/0506/quant-ph0506259.html
|
ar5iv
|
text
|
# Relaxation of Pseudo pure states : The Role of Cross-Correlations
## I I. Introduction
Quantum information processing (QIP) often requires pure state as the initial state preskill ; chuangbook . Shorโs prime factorizing algorithm shor , Grover search algorithm grover are few examples. Creation of pure state in NMR is not easy due to small gaps between nuclear magnetic energy levels and demands unrealistic experimental conditions like near absolute zero temperature or extremely high magnetic field. This problem has been circumvented by creating a Pseudo Pure State (PPS). While in a pure state all energy levels except one have zero populations, in a PPS all levels except one have equal populations. Since the uniform background populations do not contribute to the NMR signal, such a state then mimics a pure state. Several methods of creating PPS have been developed like spatial averaging cory97 ; cory98 , logical labeling chuang97 ; chuang98 ; kavita00 ; kavita1 , temporal averaging chuangtemp , spatially averaged logical labeling technique (SALLT) maheshpra01 . However pseudo pure state, as well as pure states are not stationary and are destroyed with time as the spin system relaxes toward equilibrium. In QIP there are also cases where one or more qubits are initialized to a suitable state at the beginning of the computation and are used as storage or memory qubits at the end of the computation performed on some other qubitschuangdelay . In these cases it is important for memory qubits to be in the initialized state till the time they are in use since deviation from the initial state adds error to the output result. Since it is not possible to stop decay of a state which is away from equilibrium, alternate strategies like Quantum Error Correction qec , Noiseless subspace ns1 ; ns2 are being tried. Recently Sarthour et al.quadrel has reported a detailed study of relaxation of pseudo pure states and few other states in a quadrupolar system. Here we experimentally examine the lifetime of various pseudo pure states in a weakly J-coupled two qubit system. We find that cross terms (known as cross-correlation) between different pathways of relaxation of a spin can retard the relaxation of certain PPS and accelerate that of others.
In 1946 Bloch formulated the behavior of populations or longitudinal magnetizations when they are perturbed from the equilibrium bloch . The recovery toward equilibrium is exponential for a two level system and for a complex system the recovery involves several time constants redfield . For complex systems the von Neumann-Liouville equation von ; abragam describes mathematically the time evolution of the density matrix in the magnetic resonance phenomena. For system having more than one spin the relaxation is described by a matrix called the relaxation matrix whose elements are linear combinations of spectral densities, which in turn are Fourier transforms of time correlation function anilreview of the fluctuations of the various interactions responsible for relaxation. There exist several different mechanisms for relaxation, such as, time dependent dipole-dipole(DD) interaction, chemical shift anisotropy(CSA), quadrupolar interaction and spin rotation interaction anilreview . The correlation function gives the time correlations between different values of the interactions. The final correlation function has two major parts, namely the โAuto-correlationโ part which gives at two different times the correlation between the same relaxation interaction and the โCross-correlationโ part which gives the time correlation between two different relaxation interactions. The mathematics of cross correlation can be found in detail, in works of Schneider sch1 ; sch2 , Blicharski blicharski and Hubbard hubbard . Recently a few models have been suggested to study the decoherence of the quantum coherence, the off-diagonal elements in density matrix zurek ; cory03 . It can be shown that in absence of r.f. pulses and under secular approximation the relaxation of the diagonal and the off-diagonal elements of the density matrix are independent redfield . Here we study the longitudinal relaxation that is the relaxation of the diagonal elements of the density matrix and the role of cross-correlations in it.
## II II. Theory
### II.1 A. The Pseudo Pure State (PPS) : In terms of magnetization modes
In terms of magnetization modes the equilibrium density matrix of a two spin system is given by cory98 ; abragam ; ernstbook ; jonesgate \[Fig.1\],
$`\chi _{eq}=\gamma _1I_{1Z}+\gamma _2I_{2Z}`$ (1)
where $`\gamma _1`$ and $`\gamma _2`$ are gyro-magnetic ratios of the two spins $`I_1`$ and $`I_2`$ respectively. The density matrix of a general state can be written as,
$`\chi =\pm \alpha \gamma _1I_{1Z}\pm \beta \gamma _2I_{2Z}\pm \nu \gamma _12I_{1Z}I_{2Z}]`$ (2)
which for the condition $`\alpha \gamma _1`$=$`\beta \gamma _2`$=$`\nu \gamma _1`$=K, corresponds to the density matrix of a PPS given by cory98 ,
$`\chi _{pps}=K\left[\pm I_{1Z}\pm I_{2Z}\pm 2I_{1Z}I_{2Z}\right]`$ (3)
where, K is a constant, the value of which depends on the method of creation of PPS.
The first two terms in the right hand side in Eq.2 and Eq.3 are the single spin order modes for the first and second spin respectively while the last term is the two spin order mode of the two spins cory98 . Choosing properly the signs of the modes, the various PPS of a two-qubit system are,
$`\chi _{pps}^{00}=K\left[+I_{1Z}+I_{2Z}+2I_{1Z}I_{2Z}\right]`$
$`\chi _{pps}^{01}=K\left[I_{1Z}+I_{2Z}+2I_{1Z}I_{2Z}\right]`$
$`\chi _{pps}^{10}=K\left[+I_{1Z}I_{2Z}+2I_{1Z}I_{2Z}\right]`$
$`\chi _{pps}^{11}=K\left[+I_{1Z}+I_{2Z}2I_{1Z}I_{2Z}\right]`$ (4)
The relative populations of the states for different PPS are shown in Fig. 2. As seen in Eq.2, in PPS the coefficients of the all three modes are equal. On the other hand equilibrium density matrix does not contain any two spin order mode. To reach Eq.3 starting from Eq.1, the two spin order mode has to be created and at the same time the coefficients of all the modes have to be made equal.
### II.2 B. Relaxation of Magnetization Modes
The equation of motion of modes M is given by anilreview ,
$`{\displaystyle \frac{d}{dt}}\stackrel{}{M}\left(t\right)=\widehat{\mathrm{\Gamma }}\left[\stackrel{}{M}\left(t\right)\stackrel{}{M}\left(\mathrm{}\right)\right]`$ (5)
where $`\widehat{\mathrm{\Gamma }}`$ is the relaxation matrix and $`\stackrel{}{M}\left(\mathrm{}\right)`$ is the equilibrium values of a mode. For a weakly coupled two-spin system relaxing via mutual dipolar interaction and the CSA relaxation, the two dominant mechanism of relaxation of spin half nuclei in liquid state, the above equation takes the form,
$`{\displaystyle \frac{d}{dt}}\left[\begin{array}{c}I_{1Z}\left(t\right)\\ I_{2Z}\left(t\right)\\ 2I_{1Z}I_{2Z}\left(t\right)\end{array}\right]=\left[\begin{array}{ccc}\rho _1& \sigma _{12}& \delta _{1,12}\\ \sigma _{12}& \rho _2& \delta _{2,12}\\ \delta _{1,12}& \delta _{2,12}& \rho _{12}\end{array}\right]\left[\begin{array}{c}I_{1Z}\left(0\right)I_{1Z}\left(\mathrm{}\right)\\ I_{2Z}\left(0\right)I_{2Z}\left(\mathrm{}\right)\\ 2I_{1Z}I_{2Z}\end{array}\right]`$ (15)
where $`\rho _i`$ is the self relaxation rate of the single spin order mode of spin $`๐ข`$, $`\rho _{ij}`$ is the self relaxation rate of the two spin order mode of spin $`๐ข`$ and $`๐ฃ`$, $`\sigma _{ij}`$ is the cross-relaxation (Nuclear Overhouser Effect, NOE) rate between spins $`๐ข`$ and $`๐ฃ`$ and $`\delta _{i,ij}`$ is the cross-correlation term between CSA relaxation of spin $`๐ข`$ and the dipolar relaxation between the spins $`๐ข`$ and $`๐ฃ`$. $`\rho `$ and $`\sigma `$ involve only the auto-correlation terms and $`\delta `$ involves only the cross-correlation termsanilreview . Magnetization modes of one order relaxes to other orders through cross-correlation and in absence of it the relaxation matrix becomes block diagonal within each order. The relaxation of modes are in general dominated by their self relaxation $`\rho `$, but in case of samples having long $`T_1`$, the cross-correlation terms become comparable with self-relaxation and play an important role in relaxation of the spins.
The formal solution of Eq. 5 is given by,
$`\stackrel{}{M}\left(t\right)=\stackrel{}{M}\left(\mathrm{}\right)+\left[\stackrel{}{M}\left(0\right)\stackrel{}{M}\left(\mathrm{}\right)\right]exp\left(\widehat{\mathrm{\Gamma }}t\right)`$ (16)
As time evolution of various modes are coupled, a general solution of the above equation requires diagonalization of the relaxation matrix. However, in the initial rate approximation Eq.7 can be written (for small values of t=$`\tau `$) as,
$`\stackrel{}{M}\left(\tau \right)`$ $`=`$ $`\stackrel{}{M}\left(\mathrm{}\right)+\left[\stackrel{}{M}\left(0\right)\stackrel{}{M}\left(\mathrm{}\right)\right]\left[1\widehat{\mathrm{\Gamma }}\tau \right]`$ (17)
$`=`$ $`\stackrel{}{M}\left(0\right)\widehat{\mathrm{\Gamma }}\tau \left[\stackrel{}{M}\left(0\right)\stackrel{}{M}\left(\mathrm{}\right)\right]`$ (18)
This equation asserts that in the initial rate approximation (for low $`\tau `$), the decay or growth of a mode is linear with time and the initial slope is proportional to the corresponding relaxation matrix element. If the modes are allowed to relax for a longer time, their decay or growth deviates from the linear nature and adopts a multi-exponential behavior to finally reach the equilibriumanilreview .
### II.3 C. Relaxation of Pseudo pure state
Let a two qubit system be in $`|00`$ PPS at t=0.
$`\chi ^{00}\left(0\right)=K\left[I_{1Z}+I_{2Z}+2I_{1Z}I_{2Z}\right]`$ (19)
After time t it will relax to,
$`\chi ^{00}\left(t\right)`$ $`=`$ $`\left(K+\mathrm{\Delta }_1\left(t\right)\right)I_{1Z}+\left(K+\mathrm{\Delta }_2\left(t\right)\right)I_{2Z}+\left(K+\mathrm{\Delta }_{12}\left(t\right)\right)2I_{1Z}I_{2Z}`$ (20)
where $`\mathrm{\Delta }_1\left(t\right)`$,$`\mathrm{\Delta }_2\left(t\right)`$ and $`\mathrm{\Delta }_{12}\left(t\right)`$ are the time dependent deviations of respective modes from their initial values. The deviation of the two spin order can be measured from spectrum of either spin. Eq.20 can also be written as,
$`\chi ^{00}\left(t\right)=\left(K+\mathrm{\Delta }_{12}\left(t\right)\right)\left[I_{1Z}+I_{2Z}+2I_{1Z}I_{2Z}\right]+\left(\mathrm{\Delta }_1\left(t\right)\mathrm{\Delta }_{12}\left(t\right)\right)I_{1Z}+\left(\mathrm{\Delta }_2\left(t\right)\mathrm{\Delta }_{12}\left(t\right)\right)I_{2Z}`$ (21)
The first term is the pseudo pure state with the coefficient decreasing in time while the other two terms are the excesses of the single spin order modes with coefficients increasing in time. For other pseudo pure states Eq.21 becomes,
$`\chi ^{01}\left(t\right)`$ $`=`$ $`\left(K+\mathrm{\Delta }_{12}\left(t\right)\right)\left[I_{1Z}+I_{2Z}+2I_{1Z}I_{2Z}\right]+\left(\mathrm{\Delta }_1\left(t\right)+\mathrm{\Delta }_{12}\left(t\right)\right)I_{1Z}+\left(\mathrm{\Delta }_2\left(t\right)\mathrm{\Delta }_{12}\left(t\right)\right)I_{2Z}`$ (22)
$`\chi ^{10}\left(t\right)`$ $`=`$ $`\left(K+\mathrm{\Delta }_{12}\left(t\right)\right)\left[I_{1Z}I_{2Z}+2I_{1Z}I_{2Z}\right]+\left(\mathrm{\Delta }_1\left(t\right)\mathrm{\Delta }_{12}\left(t\right)\right)I_{1Z}+\left(\mathrm{\Delta }_2\left(t\right)+\mathrm{\Delta }_{12}\left(t\right)\right)I_{2Z}`$ (23)
$`\chi ^{11}\left(t\right)`$ $`=`$ $`\left(K\mathrm{\Delta }_{12}\left(t\right)\right)\left[I_{1Z}+I_{2Z}2I_{1Z}I_{2Z}\right]+\left(\mathrm{\Delta }_1\left(t\right)+\mathrm{\Delta }_{12}\left(t\right)\right)I_{1Z}+\left(\mathrm{\Delta }_2\left(t\right)+\mathrm{\Delta }_{12}\left(t\right)\right)I_{2Z}`$ (24)
In the initial rate approximation (using Eq.18) we obtain for the $`|00`$ pps,
$`\mathrm{\Delta }_1\left(\tau \right)`$ $`=`$ $`\tau \left[\rho _1\left(\gamma _1K\right)+\sigma _{12}\left(\gamma _2K\right)K\delta _{1,12}\right]`$ (25)
$`\mathrm{\Delta }_2\left(\tau \right)`$ $`=`$ $`\tau \left[\sigma _{12}\left(\gamma _1K\right)+\rho _1\left(\gamma _2K\right)K\delta _{2,12}\right]`$ (26)
$`\mathrm{\Delta }_{12}\left(\tau \right)`$ $`=`$ $`\tau \left[\delta _{1,12}\left(\gamma _1K\right)+\delta _{2,12}\left(\gamma _2K\right)K\rho _{12}\right]`$ (27)
Let the coefficients of the PPS term and the two single spin order modes $`I_{1Z}`$ and $`I_{2Z}`$ in Eq.21 be called as $`๐`$,$``$ and $`๐`$ respectively. Fig.3 schematically shows the time evolution of the coefficients $`๐`$,$``$ and $`๐`$ for $`|00`$ PPS. Any coefficient for any PPS at any instant, is simply the initial value plus the total deviation due to the auto and the cross-correlations. For example, $`๐`$ for $`|00`$ PPS at time $`\tau `$, is $`๐^{00}\left(\tau \right)=K+๐_{auto}^{00}\left(\tau \right)+๐_{cc}^{00}(\tau `$), where K is the initial value, and $`๐_{auto}^{00}\left(\tau \right)`$ and $`๐_{cc}^{00}\left(\tau \right)`$ are the deviations at $`\tau `$ due to auto-correlation and cross-correlation parts respectively.
#### II.3.1 (a) Contribution of auto-correlation terms to the deviation
Putting the values of the deviations of different modes obtained from Eq.(25-27) in Eq.(21-24), we obtain the contribution only of auto-correlation terms to the deviation from initial value of the coefficients $`๐`$,$``$ and $`๐`$ under initial rate approximation (at t=$`\tau `$) as,
$`๐_{auto}^{00}\left(\tau \right)=๐_{auto}^{01}\left(\tau \right)=๐_{auto}^{10}\left(\tau \right)=๐_{auto}^{11}\left(\tau \right)=K\rho _{12}\tau `$
$`_{auto}^{00}(\tau )=[\rho _1(\gamma _1K)+\sigma _{12}(\gamma _2K)+K\rho _{12}]\tau `$ ; $`_{auto}^{01}(\tau )=[\rho _1(\gamma _1+K)+\sigma _{12}(\gamma _2K)K\rho _{12}]\tau `$
$`_{auto}^{10}(\tau )=[\rho _1(\gamma _1K)+\sigma _{12}(\gamma _2+K)+K\rho _{12}]\tau `$ ; $`_{auto}^{11}(\tau )=[\rho _1(\gamma _1K)+\sigma _{12}(\gamma _2K)+K\rho _{12}]\tau `$
$`๐_{auto}^{00}(\tau )=[\sigma _{12}(\gamma _1K)+\rho _2(\gamma _2K)+K\rho _{12}]\tau `$ ; $`๐_{auto}^{01}(\tau )=[\sigma _{12}(\gamma _1+K)+\rho _2(\gamma _2K)+K\rho _{12}]\tau `$
$`๐_{auto}^{10}(\tau )=[\sigma _{12}(\gamma _1K)+\rho _2(\gamma _2+K)K\rho _{12}]\tau `$ ; $`๐_{auto}^{11}(\tau )=[\sigma _{12}(\gamma _1K)+\rho _2(\gamma _2K)+K\rho _{12}]\tau `$ (28)
It is evident that in absence of cross-correlations the $`|00`$ and $`|11`$ PPS relax at the same initial rate since $`๐_{auto}^{00}=๐_{auto}^{11}`$, $`_{auto}^{00}=_{auto}^{11}`$ and $`๐_{auto}^{00}=๐_{auto}^{11}`$. However the same is not true for $`|01`$ and $`|10`$ PPS.
#### II.3.2 (b) Contribution of cross-correlation terms to the deviation
The contribution by the cross-correlation terms is given by,
$`๐_{cc}^{00}\left(\tau \right)=+\left[\delta _{1,12}\left(\gamma _1K\right)+\delta _{2,12}\left(\gamma _2K\right)\right]\tau `$
$`๐_{cc}^{01}\left(\tau \right)=+\left[\delta _{1,12}\left(\gamma _1+K\right)+\delta _{2,12}\left(\gamma _2K\right)\right]\tau `$
$`๐_{cc}^{10}\left(\tau \right)=+\left[\delta _{1,12}\left(\gamma _1K\right)+\delta _{2,12}\left(\gamma _2+K\right)\right]\tau `$
$`๐_{cc}^{11}\left(\tau \right)=\left[\delta _{1,12}\left(\gamma _1K\right)+\delta _{2,12}\left(\gamma _2K\right)\right]\tau `$
$`_{cc}^{00}\left(\tau \right)=\left[\delta _{1,12}\gamma _1+\delta _{2,12}\left(\gamma _2K\right)\right]\tau `$ ; $`_{cc}^{01}\left(\tau \right)=+\left[\delta _{1,12}\gamma _1+\delta _{2,12}\left(\gamma _2K\right)\right]\tau `$
$`_{cc}^{10}\left(\tau \right)=\left[\delta _{1,12}\gamma _1+\delta _{2,12}\left(\gamma _2+K\right)\right]\tau `$ ; $`_{cc}^{11}\left(\tau \right)=+\left[\delta _{1,12}\gamma _1+\delta _{2,12}\left(\gamma _2K\right)\right]\tau `$
$`๐_{cc}^{00}\left(\tau \right)=\left[\delta _{1,12}\left(\gamma _1K\right)+\delta _{2,12}\gamma _2\right]\tau `$ ; $`๐_{cc}^{01}\left(\tau \right)=\left[\delta _{1,12}\left(\gamma _1+K\right)+\delta _{2,12}\gamma _2\right]\tau `$
$`๐_{cc}^{10}\left(\tau \right)=+\left[\delta _{1,12}\left(\gamma _1K\right)+\delta _{2,12}\gamma _2\right]\tau `$ ; $`๐_{cc}^{11}\left(\tau \right)=+\left[\delta _{1,12}\left(\gamma _1K\right)+\delta _{2,12}\gamma _2\right]\tau `$ (29)
The important thing is that the presence of cross-correlation can lead to differential relaxation of all PPS. Positive cross-correlation rates $`\delta _{1,12}`$ and $`\delta _{2,12}`$, slow down the relaxation of all the three coefficients for $`|00`$ PPS since ($`๐^{00}\left(\tau \right)>๐_{auto}^{00}\left(\tau \right),^{00}\left(\tau \right)<_{auto}^{00}\left(\tau \right),๐^{00}\left(\tau \right)<๐_{auto}^{00}\left(\tau \right)`$), while make the relaxation of all three coefficients faster for $`|11`$ PPS since ($`๐^{11}\left(\tau \right)<๐_{auto}^{11}\left(\tau \right),^{11}\left(\tau \right)>_{auto}^{11}\left(\tau \right),๐^{11}\left(\tau \right)>๐_{auto}^{11}\left(\tau \right)`$). For $`|01`$ and $`|10`$ PPS cross-correlations give a mixed effect since ($`๐^{01}\left(\tau \right)>๐_{auto}^{01}\left(\tau \right),^{01}\left(\tau \right)>_{auto}^{01}\left(\tau \right),๐^{01}\left(\tau \right)<๐_{auto}^{01}\left(\tau \right)`$) and ($`๐^{10}\left(\tau \right)>๐_{auto}^{00}\left(\tau \right),^{10}\left(\tau \right)<_{auto}^{10}\left(\tau \right),๐^{10}\left(\tau \right)>๐_{auto}^{10}\left(\tau \right)`$). As the contributions of the auto-correlation part for $`|00`$ and $`|11`$ PPS are equal, we have monitored the relaxation behavior only of $`|00`$ and $`|11`$ PPS to study the effect of cross-correlations.
For samples having long $`T_1`$, where the cross-correlations becomes comparable with auto-correlation rates, the four PPS relax with four different rates and the difference increases with the increased value of the cross-correlation terms. The three coefficients $`๐`$,$``$ and $`๐`$ (normalized to the equilibrium line intensities) in terms of proton and fluorine line intensities for $`|00`$ PPS are,
$`๐\left(t\right)={\displaystyle \frac{H^0\left(t\right)H^1\left(t\right)}{H^1\left(\mathrm{}\right)+H^0\left(\mathrm{}\right)}}={\displaystyle \frac{F^0\left(t\right)F^1\left(t\right)}{F^1\left(\mathrm{}\right)+F^0\left(\mathrm{}\right)}}`$
$`\left(t\right)`$ $`=`$ $`{\displaystyle \frac{2H^1\left(t\right)}{H^0\left(\mathrm{}\right)+H^1\left(\mathrm{}\right)}}`$
$`๐\left(t\right)`$ $`=`$ $`{\displaystyle \frac{2F^1\left(t\right)}{F^0\left(\mathrm{}\right)+F^1\left(\mathrm{}\right)}}`$ (30)
and for $`|11`$ PPS are,
$`๐\left(t\right)={\displaystyle \frac{H^1\left(t\right)H^0\left(t\right)}{H^1\left(\mathrm{}\right)+H^0\left(\mathrm{}\right)}}={\displaystyle \frac{F^1\left(t\right)F^0\left(t\right)}{F^1\left(\mathrm{}\right)+F^0\left(\mathrm{}\right)}}`$
$`\left(t\right)`$ $`=`$ $`{\displaystyle \frac{2H^0\left(t\right)}{H^0\left(\mathrm{}\right)+H^1\left(\mathrm{}\right)}}`$
$`๐\left(t\right)`$ $`=`$ $`{\displaystyle \frac{2F^0\left(t\right)}{F^0\left(\mathrm{}\right)+F^1\left(\mathrm{}\right)}}`$ (31)
where, $`H^0`$ and $`H^1`$ are intensities of the two proton transitions, when the fluorine spin is respectively in state $`|0`$ and $`|1`$. Similarly $`F^0`$ and $`F^1`$ are intensities of two fluorine transitions corresponding to the proton spin being respectively in the state $`|0`$ and $`|1`$, as shown in Fig.1 and Fig.6. $`H^0\left(t\right)`$ and $`H^0\left(\mathrm{}\right)`$ give the $`H^0`$ line intensity respectively at time t and at equilibrium. Thus by monitoring the intensities of the two proton and two fluorine transitions as a function of time, one can calculate the coefficient $`A\left(t\right)`$ which is a measure of decay of PPS.
## III III. Simulation
Relaxation of the coefficients $`๐`$,$``$ and $`๐`$ have been simulated using MATLAB, for a weakly coupled $`{}_{}{}^{19}F`$-$`{}_{}{}^{1}H`$ system. The relaxation matrix used for the simulation is,
$`\widehat{\mathrm{\Gamma }}=\left[\begin{array}{ccc}0.3125& 0.02& \delta _{1,12}\\ 0.02& 0.33& \delta _{2,12}\\ \delta _{1,12}& \delta _{2,12}& 0.33\end{array}\right]`$
Fig.4 shows the decay of coefficient $`๐`$ with time. $`๐^{00}`$ and $`๐^{11}`$ show no difference in decay rate in absence of cross-correlation rates. As $`\delta _{1,12}`$ and $`\delta _{2,12}`$ are increased more and more difference in decay rate is observed. Fig.5 shows growth of coefficients $``$ and $`๐`$. As $`\delta _{2,12}`$ is taken smaller than $`\delta _{1,12}`$, difference in decay rate between $`๐^{00}`$ and $`๐^{11}`$ is found to be less than between $`^{00}`$ and $`^{11}`$.
## IV IV. Experimental
All the relaxation measurement were performed on a two qubit sample formed by one fluorine and one proton of 5-fluro 1,3-dimethyl uracil yielding an AX spin system with a J-coupling of 5.8 Hz. Longitudinal relaxation time constants for $`{}_{}{}^{19}F`$ and $`{}_{}{}^{1}H`$ are 6 and 7.2 Sec respectively at room temperature (300K). All the experiments were performed in a Bruker DRX 500 MHz spectrometer where the resonance frequencies for $`{}_{}{}^{19}F`$ and $`{}_{}{}^{1}H`$ are 470.59 MHz and 500.13 MHz respectively. The Pseudo-pure state was prepared by spatial averaging method using J-evolution cory98 . Relaxation of all the three coefficients for $`|00`$ and $`|11`$ PPS has been calculated. Since auto-correlations contribute equally to the relaxation of these two PPS, any difference in relaxation rate can be attributed to cross-correlation rates.
Sample temperature was varied to change the correlation time and hence the cross-correlation rate $`\delta `$. Four different sample temperatures, 300K, 283K, 263K and 253K were used. Fig.6 shows the proton and fluorine spectra obtained using recovery measurement at four different temperatures. The spectra correspond to the initial PPS state and that after an interval of 2.5 sec. Fig.7 shows the longitudinal relaxation times ($`T_1`$) of fluorine and proton as function of temperature obtained from initial part of inversion-recovery experiment. A steady decrease in $`T_1`$ with decreasing temperature indicates that the dynamics of the sample molecule is in the short correlation time limit abragam . In this limit auto as well as cross-correlations increase linearly with decreasing temperature.
All the spectra were fitted to bi-Lorentzian lines in MATLAB and various parameters were extracted using the Origin software. Fig.8 shows the decay of the coefficient $`๐`$ calculated independently from proton and fluorine spectra. At 300K, $`๐^{00}`$ and $`๐^{11}`$ showed almost same rate of decay. As the temperature was gradually lowered, a steady increase in difference in decay rate was observed. This is due to the steady increase in cross-correlation rates with decreasing temperature, which is expected in the short correlation time limit. In Fig.9 the growths of the coefficients $``$ and $`๐`$ are shown. Similar to the coefficient $`๐`$, coefficients $``$ and $`๐`$ also show differences in decay rate between $`|00`$ and $`|11`$ PPS at lower temperatures. The difference between $`^{00}`$ and $`^{11}`$ at any temperature was found to be larger compared to between $`๐^{00}`$ and $`๐^{11}`$. This is expected since, according to Eq.29 the dominant cross-correlation factor in $`^{00}`$ and $`^{11}`$ is $`\delta _{1,12}`$ which is the cross-correlation between CSA of fluorine with fluorine-proton dipolar interaction whereas in $`๐^{00}`$ and $`๐^{11}`$ the dominant factor is $`\delta _{2,12}`$ which is cross-correlation between CSA of proton, which is much less than fluorine, with fluorine-proton dipolar interaction. Thus it is found that at lower temperatures the $`|00`$ PPS decays slower than the $`|11`$ PPS. The dominant difference in the decay rates arises from the cross-correlations between the CSA of the fluorine and the dipolar interaction between the fluorine and the proton spin. To the best of our knowledge this is the first study of its kind where the differential decay of the PPS has been attributed to cross-correlations.
## V V. Conclusion
We have demonstrated here that in samples having long $`T_1`$ cross-correlations plays an important role in determining the rate of relaxation of pseudo pure state. In QIP sometimes one or more qubits having comparatively longer longitudinal relaxation are used as storage or memory qubits. Recently Levitt et al. have demonstrated a long living antisymmetric state arrived by shifting the sample from high to very low magnetic field, suggesting that this long living state could be used as memory qubit levitthigh ; levittlow . In such cases fidelity of computation depends on how much the memory qubits have been deviated from the initialized state at the beginning of the computation till the time they are actually used. Theoretically it is shown here that in presence of cross-correlations, all the four PPS relax with different initial rates. For positive cross-correlations the $`|00`$ PPS relaxes significantly slower than $`|11`$ PPS. It is therefore important to choose a proper initial pseudo pure state according to the sample.
## Acknowledgments
We gratefully acknowledge Prof. K. V. Ramanathan for discussions and Mr. Rangeet Bhattacharyya for his help in data processing. The use of DRX-500 high resolution liquid state spectrometer of the Sophisticated Instrument Facility, Indian Institute of Science, Bangalore, funded by Department of Science and Technology (DST), New Delhi, is gratefully acknowledged. AK acknowledges โDAE-BRNSโ for โSenior Scientist schemeโ, and DST for a research grant.
FIGURE CAPTIONS
Figure 1. (a) Chemical structure of 5-fluro 1,3-dimethyl uracil. The fluorine and the proton spins (shown by circles) are used as the two qubits $`I_1`$ and $`I_2`$ respectively. (b) The energy level diagram of a two qubit system identifying the four states 00,01,10 and 11. Under high temperature and high field approximation abragam the relative equilibrium deviation populations are indicated in the bracket for each level. Assuming this to be a weakly coupled two spin system the deviation populations become proportional to the gyromagnetic ratios $`\gamma _1`$ and $`\gamma _2`$. $`I_j^k`$ refers to the transition of the $`j^{th}`$ spin when the other spin is in state $`|k`$. Thus $`H^0`$ means the proton transition when the fluorine is in state $`|0`$.
Figure 2. Population distribution of different energy levels of a two spin system in different pseudo-pure states. K is a constant whose value depends on the protocol used for the preparation of PPS. (a),(b),(c) and (d) show respectively the $`|00`$,$`|01`$,$`|10`$ and $`|11`$ PPS.
Figure 3. Schematic representation of decay of the coefficient $`๐`$ and growth of the coefficients $``$ and $`๐`$. The magnetization modes are normalized to their respective equilibrium values. In each sub-figure the three bars correspond to the modes $`I_{1Z}`$,$`I_{2Z}`$ and 2$`I_{1Z}I_{2Z}`$ from left to right. The amount of any mode present at any time is directly proportional to the height of the corresponding bar. The numbers provided in the rightmost column represent typical values of the modes. (a) Thermal equilibrium. At thermal equilibrium only $`I_{1Z}`$ and $`I_{2Z}`$ exist. (b) $`|00`$ pseudo-pure state just after creation, where all the three modes are equal in magnitude. For $`|00`$ PPS all modes are of same sign but this is not the case for other PPS \[Eq.3\]. Coefficient $`๐`$ is the common equal amount of all the modes and it is maximum at t=0. (c) The amount of magnetization modes (schematic) at time $`\tau `$, after preparation of the PPS at t=0. The two single spin order modes increase and the two spin order mode decreases from their initial values. (d) The state of various modes at time $`\tau `$, (same as fig.c) redrawn with filled bar to indicate the residual value of $`๐`$. All the three coefficients $`๐`$,$``$ and $`๐`$ are shown. $`๐`$ (shown by the filled bar), which is the measure of the PPS, has come down by the same amount as the two spin order. $``$ (shown by the empty bar) and $`๐`$ (shown by the striped bar) are the residual part of the single spin order modes $`I_{1Z}`$ and $`I_{2Z}`$ respectively.(e) The values of various modes and coefficients after a delay $`\tau ^{}>\tau `$.
Figure 4. Simulation of decay of coefficient $`๐`$. The boxes ($``$) and circles ($``$) correspond to the $`|11`$ and $`|00`$ PPS respectively. In each plot deviation from initial value ($`๐\left(t\right)๐\left(0\right)`$)has been plotted.
Figure 5. Simulation of growth of coefficient $``$ and $`๐`$. The boxes ($``$) and circles ($``$) correspond to the $`|11`$ and $`|00`$ PPS respectively.
Figure 6. Relaxation of Pseudo pure state as monitored on (a) fluorine spin and (b) proton spin of the 5-fluro 1,3-dimethyl uracil at four different temperatures. The top row in (a) and (b) show the equilibrium spectrum at each temperature. With decrease in temperatures the lines broaden due to decreased $`T_2`$. The second row in (a) and (b) show the spectra corresponding to the $`|00`$ PPS, prepared by spatial averaging method using J-evolution. The state of PPS was measured by $`90^o`$ pulse at each spin. The third row in (a) and (b) show the spectra after an interval of 2.5 seconds after creation of the $`|00`$ PPS. The fourth row shows the spectra immediately after creation of $`|11`$ PPS and the fifth row, the spectra after 2.5 seconds.
Figure 7. Longitudinal relaxation time $`T_1`$ of fluorine (a) and proton (b) as function of temperature, measured from the initial part of inversion recovery experiment for each spin.
Figure 8. The deviation from initial value (at t=0) of the coefficient $`๐`$ of the PPS term calculated from Proton (left column) and Fluorine (right column) at four different sample temperature. The empty ($``$) and filled ($``$) circles correspond to the $`|00`$ and $`|11`$ PPS respectively.
Figure 9. The growth of the coefficients $``$ and $`๐`$ at different sample temperatures. $``$ was calculated from Fluorine spectrum while $`๐`$ was calculated from the Proton spectrum. The empty ($``$) and filled ($``$) circles correspond to the $`|00`$ and $`|11`$ PPS respectively.
|
warning/0506/hep-ph0506091.html
|
ar5iv
|
text
|
# Lorentz and CPT Invariance Violation In High-Energy Neutrinos
## I Introduction
The invariance of the product of Charge conjugation, Parity and Time reversal (CPT) is one of the most fundamental symmetries found in physics PCT . The invariance of CPT is also intimately related with Lorentz invariance, another symmetry of deeply embedded within our current picture of the physical world.
It is not completely clear that these symmetries will remain perfectly unbroken under all conditions, however. In particular, many efforts to describe the force of gravity within the context of a quantum theory imply the breaking of CPT and Lorentz symmetries. These effects are so minuscule, however, that any indication of them will likely require observations either at extremely high energies, or over incredibly long baselines. High energy neutrino astronomy provides precisely such a laboratory.
Most varieties of particles cannot travel undisturbed over cosmological distances. Particles which scatter off of background radiation or other targets are of limited use to very long baseline experiments, particularly at very high energies. High-energy neutrinos, on the other hand, can travel thousands of mega-parsecs without undergoing any energy loses beyond those from Hubble expansion.
Experimentally speaking, the field of high-energy neutrino astronomy is developing very rapidly. Current technology, such as the AMANDA II amanda2 and Rice rice telescopes at the South Pole, have been operating successfully for several years. The construction of next generation optical Cerenkov neutrino experiments such as IceCube icecube and Antares antares is also going ahead. Additionally, ultra-high cosmic ray experiments such as Auger auger , EUSO euso and OWL owl will be very sensitive to the highest energy range of the cosmic neutrino spectrum. The prospects for other technologies which incorporate radio anita and acoustic acoustic detectors appear to be very promising as well. With currently existing experiments collecting data and new experiments being developed and constructed, a new window into the Universe is beginning to open.
Over very long distances, the effects of CPT and Lorentz violation can result in modifications to the standard neutrino oscillation phenomenology. By using high-energy neutrino telescopes to measure the ratios of neutrinos flavors coming from distant sources, the possible effects of CPT and Lorentz violation can be constrained far beyond the levels which are currently experimentally accessible.
In this article, we discuss how high-energy cosmic neutrinos can be used to test for the violation of CPT and Lorentz symmetries. In section II, we review the present status of standard neutrino oscillation phenomenology and discuss the violation of Lorentz and CPT symmetries in section III. In section IV we discuss these effects in the context of neutrino oscillation phenomenology, and describe potential sources and experimental prospects for the observations of these effects in section V. We summarize our conclusions in section VII.
## II Standard Neutrino Oscillations: Present Status
### II.1 Formalism
#### II.1.1 Vacuum oscillations
Normally, neutrinos are identified by their flavor ($`e`$, $`\mu `$, $`\tau `$) rather than their mass. Neutrinos which have definite flavor need not be in states of definite mass, however. If this is the case, we consider the flavor states, $`|\nu _\alpha `$, to be a linear combination of mass states, $`|\nu _i`$:
$$|\nu _\alpha =\underset{i}{}U_{\alpha i}|\nu _i,$$
(1)
where $`U_{\alpha i}`$ are the components of the unitarity leptonic mixing matrix. In the Schrรถdinger representation, the time evolution of the mass eigenstates has the form
$$i\frac{d}{d\tau }|\nu _i(\tau )=m_i|\nu _i(\tau ),$$
(2)
where $`\tau `$ is the time in the mass frame and $`m_i`$ the mass eigenvalues. Using Eq. (2), the probability of oscillation from $`\nu _\alpha `$ to $`\nu _\beta `$ can be calculated to be
$`P(\nu _\alpha \nu _\beta )`$ $`=`$ $`\delta _{\alpha \beta }4{\displaystyle \underset{j>i}{}}\mathrm{}(U_{\alpha j}^{}U_{\beta j}U_{\alpha i}U_{\beta i}^{})\mathrm{sin}^2[\mathrm{\Delta }m_{ji}^2(L/4E)]`$ (3)
$`+2{\displaystyle \underset{j>i}{}}\mathrm{}(U_{\alpha j}^{}U_{\beta j}U_{\alpha i}U_{\beta i}^{})\mathrm{sin}[\mathrm{\Delta }m_{ji}^2(L/2E)],`$
where $`\mathrm{\Delta }m_{ji}^2m_j^2m_i^2`$, $`E`$ is the energy of the neutrino and $`L`$ is the path length. We have assumed the neutrino to be relativistic. If we consider three flavors of Majorana neutrinos, then the $`3\times 3`$ mixing matrix is given by $`U=VM`$, where
$$V=\left(\begin{array}{ccc}c_{12}c_{13}& s_{12}c_{13}& s_{13}e^{i\delta _{CP}}\\ s_{12}c_{23}c_{12}s_{23}s_{13}e^{i\delta _{CP}}& c_{12}c_{23}s_{12}s_{23}s_{13}e^{i\delta _{CP}}& s_{23}c_{13}\\ s_{12}s_{23}c_{12}c_{23}s_{13}e^{i\delta _{CP}}& c_{12}s_{23}s_{12}c_{23}s_{13}e^{i\delta _{CP}}& c_{23}c_{13}\end{array}\right),$$
$$M=\left(\begin{array}{ccc}e^{i\varphi _1}& 0& 0\\ 0& e^{i\varphi _2}& 0\\ 0& 0& 1\end{array}\right).$$
(4)
Here, $`c_{ij}`$ and $`s_{ij}`$ represent the cosine and sine of the mixing angle $`\theta _{ij}`$ respectively, $`\delta _{CP}`$ is a CP violating phase and the phases in $`M`$ are the Majorana phases. For Dirac neutrinos, the situation is similar but the Majorana phases may be absorbed into the phases of the mass eigenstates. In practice, since the mixing angle $`\theta _{13}`$ has been found to be small, systems involving just two neutrinos are often considered. Then, replacing $`c`$ and $`\mathrm{}`$, the vacuum oscillation probability reduces to
$$P(\nu _\alpha \nu _\beta )=\mathrm{sin}^22\theta \mathrm{sin}^2\left[1.27\mathrm{\Delta }m^2\frac{L}{E}\right],$$
(5)
where $`\mathrm{\Delta }m^2`$ is measured in eV<sup>2</sup>, $`L`$ is measured in kilometers and $`E`$ is measured in GeV. By convention, $`\theta _{12}`$ and $`\mathrm{\Delta }m_{21}^2`$ are solar oscillation parameters describing $`\nu _e\nu _{\mu ,\tau }`$ whilst $`\theta _{23}`$ and $`\mathrm{\Delta }m_{32}^2`$ are atmospheric neutrino oscillation parameters, describing $`\nu _\mu \nu _\tau `$.
#### II.1.2 Oscillations in matter
The situation becomes somewhat more complicated if neutrinos are passing through dense matter, for example in the Sun. In this case, the Hamiltonian in the mass basis is no longer diagonal, resulting in an energy dependence in the mixing angles. For simplicity, we will consider a two neutrino system. Incorporating matter effects, the Hamiltonian in the mass basis is zuber
$$H_m^i=\frac{1}{2E}\left(\begin{array}{cc}m_1^2+A\mathrm{cos}^2\theta & A\mathrm{sin}\theta \mathrm{cos}\theta \\ A\mathrm{sin}\theta \mathrm{cos}\theta & m_2^2+A\mathrm{sin}^2\theta \end{array}\right),$$
(6)
where $`A`$, embodying the matter effects, is $`A=2\sqrt{2}EG_FN_e`$ with $`G_F`$ the Fermi constant and $`N_e`$ the electron number density. The oscillation probability therefore becomes
$$P(\nu _e\nu _\beta )=\mathrm{sin}^22\theta _m\mathrm{sin}^2\left[\mathrm{\Omega }_ML\right].$$
(7)
where
$$\mathrm{sin}2\theta _m=\frac{\mathrm{\Delta }m^2}{4E}\frac{\mathrm{sin}2\theta }{\mathrm{\Omega }_M},$$
(8)
with
$$\mathrm{\Omega }_M=\frac{\mathrm{\Delta }m^2}{4E}\sqrt{[(\frac{A}{\mathrm{\Delta }m^2}\mathrm{cos}2\theta )^2+\mathrm{sin}^22\theta ].}$$
(9)
For a comprehensive overview of neutrino physics, see for example Ref. zuber .
### II.2 Where we stand - the status of neutrino oscillations
#### II.2.1 Solar neutrino oscillations
The first indirect evidence for neutrino oscillations came from those neutrinos created in the Sun. The standard solar model bahcall describes the complex nuclear processes occurring in the Sun and from this, predictions of the solar neutrino flux may be extracted. Early experiments measuring the solar neutrino flux, such as the Chlorine chlorine and Gallium gallium , were sensitive to electron neutrinos only and reported a neutrino flux significantly less than those predicted by the standard solar model. More recently, the SNO sno experiment, which is sensitive to all neutrino flavors, showed the total solar neutrino flux agrees well with the predictions of the standard solar model but with an appreciable suppression of the electron neutrino flux. This suppression is definitive evidence of neutrino oscillations from $`\nu _e\nu _{\mu ,\tau }`$. The KamLAND experiment kamland , which detects electron anti-neutrinos from nuclear reactors, reported a significant suppression in event rates thus corroborating the SNO results. Using data from these experiments, the solar neutrino oscillation parameters, $`\mathrm{\Delta }m_{21}^2=\mathrm{\Delta }m_{}^2`$ and $`\mathrm{sin}^2\theta _{21}=\mathrm{sin}^2\theta _{}`$, have been measured with high precision. These values are shown in table 1.
#### II.2.2 Atmospheric neutrino oscillations
Whilst the first indirect evidence for neutrino oscillations came from solar neutrinos, the first direct evidence came from atmospheric neutrinos. Atmospheric neutrinos are created by interactions of cosmic rays with atmospheric atomic nuclei. Pions, created in this interaction decay by $`\pi ^{}\mu ^{}+\overline{\nu }_\mu \overline{\nu }_\mu +\nu _\mu +\overline{\nu }_e+e^{}`$ (with an analogous decay chain for $`\pi ^+`$), thus indicating that the muon neutrino flux should be roughly twice that of electron neutrinos. However, the Kamiokande kamio , IMB imb and Soudan soudan experiments reported a significant deficit in the expected $`\nu _\mu :\nu _e`$ ratio kamiores ; imbres ; soudanres which suggested oscillations from $`\nu _\mu `$ to $`\nu _\tau `$. These results were somewhat inconclusive, however. In 1998, this changed when the Super-Kamiokande collaboration showed the muon neutrino flux had a zenith angle dependence, which implied a dependence upon path length newsuperk . This was the first direct evidence that neutrinos oscillate from one flavor to another. A summary of the present values of the atmospheric neutrino oscillation parameters is given in table 1.
#### II.2.3 Three neutrino oscillations
The standard analysis of solar and atmospheric neutrino flux is done within the two neutrino approximation. However, in order to place values on $`\mathrm{\Delta }m_{31}^2`$ and $`\mathrm{sin}^2\theta _{31}`$, the data must be analyzed taking all three neutrino flavors into account. Table 1 again shows the present status.
#### II.2.4 The LSND result
It would seem from the discussion above that, short of higher precision measurements of the oscillation parameters being performed, the phenomenology of neutrino oscillations is well understood. However, this is not entirely the case. The results of the LSND experiment lsnd , which produces a beam of $`\nu _e`$, $`\nu _\mu `$ and $`\overline{\nu }_\mu `$ and then searches for the appearance of $`\overline{\nu }_e`$ that have oscillated from $`\overline{\nu }_\mu `$, cannot be reconciled with the solar and atmospheric neutrino data. The LSND results imply a mass difference which lies in the range of $`0.2<\mathrm{\Delta }m_{LSND}^2<10`$ $`\mathrm{eV}^2`$ lsndres . If this result were corroborated by the miniBoone experiment miniboone , then it would provide indications of new physics. In particular, in order to combine in a compatible way the LSND result with the atmospheric and solar oscillation data, one would have to invoke oscillations into sterile neutrinos or break CPT invariance. This latter possibility is discussed below.
## III Violating Lorentz and CPT invariance
The issues of CPT invariance violation (CPTV) and Lorentz invariance violation (LV) are intimately related. The CPT theorem is a fundamental ingredient of quantum field theory ensuring that quantities appearing in the theories, such as the Hamiltonian and Lagrangian density, are invariant under the combined operations of charge conjugation (C), parity reflection (P) and time reversal (T). The CPT theorem holds in flat space-times provided the theory obeys
* locality,
* unitarity,
* Lorentz invariance.
Deviation from any one of these requirements leads to CPTV. It has also been shown recently that CPTV leads to LV greenberg .
### III.1 Violations of Lorentz invariance
The breaking of Lorentz symmetry may arise as a consequence of quantum gravity from non-trivial effects at the Planck scale, such as the existence of a fundamental length scale. Naively, one would expect this length to be the same in all reference frames due to its fundamental nature. This invariance between frames is in direct disagreement with special relativity, the Lorentz transformation predicting length contraction. Thus, there are three possibilities for the fate of LV. Firstly, Lorentz invariance may hold at the Planck scale, so that the flat space-time picture is valid. In this case, it is our naive thought experiment which is wrong. Secondly, Lorentz symmetry may be broken at the Planck scale suggesting a class of preferred inertial frames. Very often, this preferred frame is assumed to be the related to the cosmic microwave background. In this case, the dispersion relation for energy and momentum is modified and depends upon the quantum gravity environment. Thirdly, Lorentz symmetry may be deformed camelia . Again, this leads to a modified dispersion relation but with the Lorentz transformations now containing a second observer independent scale. For example, the Planck length could be an independent scale in addition to the speed of light.
#### III.1.1 Lorentz invariance violation in string theory
String theory approaches the quantum gravity problem from a particle physics perspective. Although string theory does not quantize space-time since it describes the background space-time entirely classically, the issues related to this are still far from being resolved. At this moment in time, since the background is classical, there are no indications that Lorentz invariance is broken within string theory. Of course, if it is found that the background space-time needs to be quantized, then this could lead to Lorentz invariance violating string theories. Having said that, two particular theories which are considered to be low energy limits of string theory, namely flat non-commutative space-times madore and the Standard Model Extensions (SMEs) kostelecky , indicate the presence of LV. The non-commutative space-time approach assumes that space-time coordinates do not commute, leading to the breaking of Lorentz symmetry and various forms of the dispersion relation minwalla ; matusis . The SMEs extend the Standard Model Lagrangian to include all LV and CPTV operators which are of dimension 4 or less, in order to be renormalizable. Again, this phenomenological model modifies the dispersion relation.
#### III.1.2 Lorentz invariance violation in loop quantum gravity
In contrast to string theory, loop quantum gravity (also known as canonical quantum gravity) approaches the quantum gravity problem from a general relativity perspective. The theory is fully background independent, as is general relativity, which leads to the prediction of discrete space-times rovelli . Initially, it was thought that loop quantum gravity would preserve Lorentz symmetry, however it is now thought that this theory could break gambini ; alfaro ; thiemann or deform camelia2 ; freidel Lorentz symmetry, thus leading to modified dispersion relations. The issue of whether Lorentz symmetry is preserved, broken or deformed is still unresolved in loop quantum gravity.
### III.2 Violations of CPT
The breaking of CPT invariance may occur independently of LV effects from the loss of unitarity leading to quantum decoherence. Again, this loss of unitarity could occur because of the discrete and topologically non-trivial nature of space-time resulting in the vacuum creation of quantum black holes with event horizons having radii of order the Planck length. This continuous creation and evaporation of these quantum singularities results in space-time having a foamy nature. When particles pass by these quantum black-holes, some of the particlesโ quantum numbers could be captured by the space-time fluctuations. With the evaporation of the black holes, the information captured would be lost to the vacuum, inaccessible to low energy experiments. This loss of information would imply that initially pure quantum states may evolve into mixed quantum states; a process forbidden within standard quantum mechanics. Since we are now dealing with mixed quantum states, we use the density matrix formalism. Including decoherence linearly, the time evolution of the density matrix, $`\rho `$, is modified:
$$\frac{d\rho }{dt}=i[H,\rho ]+๐[\rho ],$$
(10)
where $`H`$ is the Hamiltonian of the system interacting with the environment through the operators $`D_j,D_j^{}`$. We would expect the the CPTV term, $`๐[\rho ]`$, to take the Lindblad form lindblad
$$๐[\rho ]=\underset{j}{}\left(\{\rho ,D_j^{}D_j\}2D_jD_j^{}\right)$$
(11)
where $`\{\mathrm{}\}`$ represents an anti-commutator. From a physical point of view, we require energy conservation and monotonic increase in the von-Neumann entropy. In this case, we find we have to specify the operators, $`D`$, to be self-adjoint and that they commute with the Hamiltonian. We therefore find
$$๐[\rho ]=\underset{j}{}[D_j,[D_j,\rho ]].$$
(12)
From a quantum gravity perspective, we would expect the operators, $`D`$, to be proportional to the inverse of the Planck mass and thus $`๐[\rho ]M_p^2`$.
#### III.2.1 Quantum decoherence in string theory
The evolution of mixed states into pure states, as described above, results in problems with defining an $`S`$ matrix. Since string theory relies on the defining of $`S`$ matrices, quantum decoherence is generally not expected within string theory. However, one class of string theories, namely non-critical string theories, may allow decoherence. This theory may be viewed as a type of non-equilibrium string theory with the so called critical strings corresponding to equilibrium points within the theory (for more details see, for example, Ref. noncritical ). In this case, we find an analogous expression for the time evolution of the string matter density matrix as with Eq. (12) mavromatos2 .
#### III.2.2 Quantum decoherence in loop quantum gravity
Whilst loop quantum gravity implies that space-time is discrete, there is no a priori reason to expect quantum decoherence. However, there have been proposals gambini2 suggesting that the discreteness of space-time may induce decoherence having the Lindblad form outlined in Eq. (10). However, it seems that there is still much work needed in order to clarify this.
#### III.2.3 Cosmological decoherence
In addition to quantum decoherence induced by space-time foam effects, it may be that decoherence arises from cosmological considerations. It is now established that the universe has entered a period of acceleration supernova1 ; supernova2 driven by some exotic dark energy. If this expansion continues, the universe will evolve into a de-Sitter universe, expanding at an exponential rate. This would imply the existence of a cosmological horizon. This situation can be considered in the same way as that of the space-time foam situation except we, as the observers, now inhabit the space within the horizon instead of outside. The existence of this horizon would again lead to the inability to define $`S`$ matrices leading to decoherence. It has been argued in non-critical string theory mavromatos3 that this cosmological decoherence may be intimately linked with quantum gravity. Considering a two level neutrino system, the cosmological decoherence parameter, $`\gamma _{cosmo}`$, is related to the cosmological constant, $`\mathrm{\Lambda }`$, the weak string coupling, $`g_s`$, the difference of the squares of the mass eigenstates, $`\mathrm{\Delta }m^2`$, the energy of the neutrino, $`E`$, and the string mass scale, $`M_s`$:
$$\gamma _{cosmo}\frac{\mathrm{\Lambda }g_s^2(\mathrm{\Delta }m^2)^2}{E^2M_s}.$$
(13)
## IV Neutrino oscillation phenomenology with CPTV and LV
### IV.1 Neutrino oscillations and LV effects
#### IV.1.1 Modified dispersion relations
As discussed above, if we allow LV, then this leads to modified dispersion relations (MDR). From the discussions above, we find it useful to parameterize the MDRโs, to leading order in the Planck energy, $`E_p`$, as
$$E^2=p^2+m^2+\eta p^2\left(\frac{E}{E_p}\right)^\alpha $$
(14)
where $`E`$ is the energy of the neutrino, $`p`$ is its momentum, $`m`$ is the mass eigenstate and $`\eta `$ and $`\alpha `$ are LV parameters. Assuming that that the parameter $`\eta `$ is not universal and depends upon the mass eigenstate, we may write the two neutrino Hamiltonian in the mass basis as
$$H=\left(\begin{array}{cc}\frac{m_1^2}{2E}+\frac{\eta _1E^{\alpha +1}}{2}& 0\\ 0& \frac{m_2^2}{2E}+\frac{\eta _2E^{\alpha +1}}{2}\end{array}\right),$$
(15)
where we have neglected the kinetic term and identified $`p`$ with $`E`$ since the mass eigenstates and the LV are terms are much smaller than the momentum of the neutrinos. For simplicity, we have absorbed the Planck energy into $`\eta `$. This leads to the neutrino oscillation probability,
$$P[\nu _\alpha \nu _\beta ]=\mathrm{sin}^22\theta \mathrm{sin}^2\left[\frac{\mathrm{\Delta }m^2L}{4E}+\frac{\mathrm{\Delta }\eta E^nL}{4}\right],$$
(16)
where $`n=\alpha +1`$ and $`\mathrm{\Delta }\eta `$ is the difference between the two values of $`\eta `$. If there are no LV effects, we recover the standard neutrino oscillation probability (5) (replacing $`c`$ and $`\mathrm{}`$). We also assumed that the LV parameter, $`\eta `$, had a dependence on the mass eigenstate. If this is not the case, then the neutrino probability remains invariant even if Lorentz invariance is violated or deformed. Assuming that these effects take place in atmospheric neutrino oscillations, the LV effects become significant when
$$1.27\frac{\mathrm{\Delta }m^2L}{E}1.27\times 10^{27}\mathrm{\Delta }\eta E^2L$$
(17)
where we have set $`\alpha =1`$ for simplicity. We therefore find
$$\mathrm{\Delta }\eta \frac{\mathrm{\Delta }m^2}{10^{27}E^2}10^{30}eV^1$$
(18)
using the value for $`\mathrm{\Delta }m^2`$ from table 1 and $`E=1\mathrm{GeV}`$, the peak in the atmospheric neutrino flux.
#### IV.1.2 Neutrino oscillations and the SME
As we outlined above, the Standard Model Extension (SME) can be considered as a low energy phenomenological model of string theory. The effective SME Hamiltonian describing flavor neutrino propagation, to first order, is kostelecky2
$$H_{\alpha \beta }^{eff}=|\stackrel{}{p}|\delta _{\alpha \beta }+\frac{1}{2|\stackrel{}{p}|}[\stackrel{~}{m}^2+2(a_L^\mu p_\mu (c_L)^{\mu \nu }p_\mu p_\nu )]_{\alpha \beta }$$
(19)
where $`\stackrel{~}{m}`$ is related to the standard neutrino mass, $`\alpha `$, $`\beta `$ are flavor indices and $`a_L`$, $`c_L`$ violate Lorentz invariance and CPT invariance. One of the main differences between this model and the LV effects described in the last section is that the Hamiltonian need not be diagonal. In Ref. kostelecky2 , various assumptions are made in order to simplify the model. Here, instead, we will adopt a general off-diagonal formalism.
In the two neutrino case, we will assume a Hamiltonian in the mass basis of the form
$$H_{eff}=\left(\begin{array}{cc}\frac{m_1^2}{2E}& a_1ia_2\\ a_1+ia_2& \frac{m_2^2}{2E}\end{array}\right),$$
(20)
where $`a_1`$ and $`a_2`$ are real, off-diagonal, LV parameters (the $`a`$โs here are independent of the $`a`$ in Eq. (19)). In order to calculate the probability, we will use the density matrix formalism. Writing the Hamiltonian in terms of the Pauli matices gives
$$h_{ij}=2\left(\begin{array}{ccc}0& \frac{\mathrm{\Delta }m^2}{4E}& a_2\\ \frac{\mathrm{\Delta }m^2}{4E}& 0& a_1\\ a_2& a_1& 0\end{array}\right)$$
(21)
where we have omitted the zeroth components for simplicity as they are all identically zero. This matrix has eigenvalues, $`\lambda _i`$, given by $`\{\pm i\mathrm{\Omega },0\}`$ where $`\mathrm{\Omega }=\sqrt{\omega ^2+a_1^2+a_2^2}`$ and $`\omega =\mathrm{\Delta }m^2/4E`$ and the matrix in Eq. (21) is diagonalized by the unitary matrix
$$U=\frac{1}{\sqrt{2a_1^2+2a_2^2}\mathrm{\Omega }}\left(\begin{array}{ccc}\omega a_1ia_2\mathrm{\Omega }& \omega a_1+ia_2\mathrm{\Omega }& a_1\sqrt{2a_1^2+a_2^2}\\ \omega a_2+ia_1\mathrm{\Omega }& \omega a_2ia_1\mathrm{\Omega }& a_2\sqrt{2a_1^2+a_2^2}\\ a_1^2+a_2^2& a_1^2+a_2^2& \omega \sqrt{2a_1^2+a_2^2}\end{array}\right).$$
(22)
The components of the density matrix are given by
$$\rho _i(L)=\underset{j,k}{}U_{ij}e^{\lambda _jL}U_{jk}^1\rho _k(0),$$
(23)
where $`U_{ij}`$ are components of the matrix in equation (22) and $`\rho (0)`$ is the density matrix initially. Assuming we have a muon neutrino which oscillates into a tau neutrino, the probability of oscillation is given by
$$P=Tr[\rho _\mu (L)\rho _\tau (0)]$$
(24)
with
$$\rho _\mu (0)=\left(\begin{array}{cc}\mathrm{cos}^2\theta & \mathrm{sin}\theta \mathrm{cos}\theta \\ \mathrm{sin}\theta \mathrm{cos}\theta & \mathrm{sin}^2\theta \end{array}\right),\rho _\tau (0)=\left(\begin{array}{cc}\mathrm{sin}^2\theta & \mathrm{sin}\theta \mathrm{cos}\theta \\ \mathrm{sin}\theta \mathrm{cos}\theta & \mathrm{cos}^2\theta \end{array}\right).$$
(25)
Thus the probability of oscillation is
$`P[\nu _\mu \nu _\tau ]`$ $`=`$ $`{\displaystyle \frac{1}{2}}[\mathrm{cos}^22\theta (1{\displaystyle \frac{\omega ^2}{\mathrm{\Omega }^2}}{\displaystyle \frac{|a|^2}{\mathrm{\Omega }^2}}\mathrm{cos}(2\mathrm{\Omega }L))`$ (26)
$`+\mathrm{sin}^22\theta \left(1{\displaystyle \frac{a_1^2}{\mathrm{\Omega }^2}}{\displaystyle \frac{(\omega ^2+a_2^2)}{\mathrm{\Omega }^2}}\mathrm{cos}(2\mathrm{\Omega }L)\right)`$
$`{\displaystyle \frac{1}{2}}\mathrm{sin}4\theta ({\displaystyle \frac{4\omega a_1}{\mathrm{\Omega }^2}}\mathrm{sin}^2(\mathrm{\Omega }L))],`$
with $`a=a_1+ia_2`$. In an analogous way to the diagonal case, the LV parameter, $`a`$, may have an explicit dependence on the neutrino energy. To examine this energy dependence, we let $`aaE^n`$ where $`n`$ is an extra parameter of the theory. It is also particularly interesting to note that if we let the quantities in Eq. (20) go to
$`m_1^2`$ $``$ $`m_1^2+A\mathrm{cos}^2\theta ,`$
$`m_2^2`$ $``$ $`m_1^2+A\mathrm{sin}^2\theta ,`$
$`a_1`$ $``$ $`{\displaystyle \frac{A}{2E}}\mathrm{sin}\theta \mathrm{cos}\theta ,`$
$`a_2`$ $`=`$ $`0,`$ (27)
then we recover the Lorentz invariant matter effects situation as described in section II.
If we wish to include these off-diagonal LV effects in the case for three neutrinos, the situation becomes very difficult as, we have three mixing angles, three mass differences and three LV parameters. In order to examine how the LV effects manifest themselves, we consider only a first order approximation in the LV parameters. In the standard oscillation case, the time evolution of the density matrix is given by
$$\frac{d\rho }{dt}=B\rho $$
(28)
where $`B`$ is the matrix representing the Hamiltonian in the Pauli basis. Perturbing the density matrix and the matrix $`B`$:
$`\rho `$ $``$ $`\rho _0+\delta \rho _1`$
$`B`$ $``$ $`B+\delta C,`$ (29)
where the $`\delta `$ quantities contain the LV effects. Substituting (IV.1.2) into (28) and equating coefficients gives
$$\delta \dot{\rho }_1=B\delta \rho _1+\delta C\rho _0.$$
(30)
Defining the vectors $`\text{x},\text{y}`$ as
$$๐_\mathrm{๐}=U๐ฑ,\delta ๐_\mathrm{๐}=U๐ฒ,$$
(31)
where the components of the $`\rho `$ vectors are the components of the density matrix and $`U`$ is the unitary matrix diagonalizing $`B`$, we may rewrite Eq. (30) as
$$\dot{๐ฒ}U^1BU๐ฒ=U^1\delta CU๐ฑ.$$
(32)
Since we know $`U,B,\delta C`$ and can evaluate $`๐ฑ`$, then solving this equation will give us the perturbation to the density matrix from which we may calculate oscillation probabilities. In reality, this calculation still results in complicated expressions for the probabilities. However, the expressions are greatly simplified if we assume very long path lengths. This is entirely reasonable since we need only consider the three neutrino system when considering neutrinos from astrophysical sources. Using the values of the mixing parameters from table 1, assuming a normal mass hierarchy and that the LV parameter, $`a`$, is real, we find the oscillation probabilities to be
$`P[\nu _e\nu _e]`$ $`=`$ $`0.5644.39\times 10^{9n+12}aE^{n+1},`$
$`P[\nu _e\nu _\mu ]`$ $`=`$ $`0.264+1.54\times 10^{9n+12}aE^{n+1},`$
$`P[\nu _e\nu _\tau ]`$ $`=`$ $`0.180+2.93\times 10^{9n+12}aE^{n+1},`$
$`P[\nu _\mu \nu _\mu ]`$ $`=`$ $`0.3651.30\times 10^{9n+11}aE^{n+1},`$
$`P[\nu _\mu \nu _\tau ]`$ $`=`$ $`0.3671.16\times 10^{9n+12}aE^{n+1},`$
$`P[\nu _\tau \nu _\tau ]`$ $`=`$ $`0.4491.56\times 10^{9n+12}aE^{n+1},`$ (33)
where we have written out in full the explicit dependence of the LV parameter on the neutrino energy and replaced $`c`$ and $`\mathrm{}`$. These results are only valid near the threshold of LV effects setting in. Using these probabilities, it is possible to find expressions describing the flux of neutrinos originating in astrophysical sources. If we assume that only electron and muon neutrinos are created, we parameterize the initial flux as
$`\mathrm{\Phi }_e`$ $`=`$ $`\epsilon \mathrm{\Phi }_{tot},`$
$`\mathrm{\Phi }_\mu `$ $`=`$ $`(1\epsilon )\mathrm{\Phi }_{tot}`$ (34)
where $`\epsilon [0,1]`$ and $`\mathrm{\Phi }_{tot}`$ is the total flux. In terms of the neutrino probabilities of Eq. (IV.1.2), the neutrino flavor composition at the detector is given by
$`R_{\nu _e}`$ $`=`$ $`(P[\nu _e\nu _e]\mathrm{\Phi }_{\nu _e}+P[\nu _\mu \nu _e]\mathrm{\Phi }_{\nu _\mu }`$
$`+`$ $`P[\nu _\tau \nu _e]\mathrm{\Phi }_{\nu _\tau })/\mathrm{\Phi }_{\mathrm{tot}},`$
$`R_{\nu _\mu }`$ $`=`$ $`(P[\nu _e\nu _\mu ]\mathrm{\Phi }_{\nu _e}+P[\nu _\mu \nu _\mu ]\mathrm{\Phi }_{\nu _\mu }`$
$`+`$ $`P[\nu _\tau \nu _\mu ]\mathrm{\Phi }_{\nu _\tau })/\mathrm{\Phi }_{\mathrm{tot}},`$
$`R_{\nu _\tau }`$ $`=`$ $`(P[\nu _e\nu _\tau ]\mathrm{\Phi }_{\nu _e}+P[\nu _\mu \nu _\tau ]\mathrm{\Phi }_{\nu _\mu }`$ (35)
$`+`$ $`P[\nu _\tau \nu _\tau ]\mathrm{\Phi }_{\nu _\tau })/\mathrm{\Phi }_{\mathrm{tot}}`$
and so we find
$`R_{\nu _e}`$ $`=`$ $`0.264+0.300\epsilon aE^{n+1}[0.593\epsilon 0.154]\times 10^{9n+13},`$
$`R_{\nu _\mu }`$ $`=`$ $`0.3650.101\epsilon +aE^{n+1}[0.167\epsilon 0.013]\times 10^{9n+13},`$
$`R_{\nu _\tau }`$ $`=`$ $`0.3670.187\epsilon +aE^{n+1}[0.409\epsilon 0.116]\times 10^{9n+13}.`$ (36)
Again, this result is only valid near the threshold of LV effects. We have also calculated the ratios in the large $`a`$ limit and find: $`R_{\nu _e}:R_{\nu _\mu }:R_{\nu _\tau }0.42:0.57:0.013`$ for $`\epsilon =1/3`$ and $`R_{\nu _e}:R_{\nu _\mu }:R_{\nu _\tau }0.70:0.27:0.027`$ for $`\epsilon =1`$. Our numerical calculations indicate that the transition between standard oscillation phenomenology and the phenomenology of the large $`a`$ limit takes place suddenly. If we assume what is perhaps the most natural choice of $`a=M_{\mathrm{Pl}}^1`$ ($`n=1`$) or $`a=M_{\mathrm{Pl}}^2`$ ($`n=2`$), the thresholds for these effects take place at $`1`$ TeV and $`10`$ PeV, respectively. Approximate numerical results are illustrated in figure 1.
### IV.2 Quantum decoherence and neutrino oscillations
We now turn to the violation of CPT without explicit Lorentz violation and consider how quantum decoherence affects neutrino oscillations. As we discussed in the previous section, quantum decoherence causes the time evolution of the density matrix to be altered:
$$\frac{d\rho }{dt}=i[H,\rho ]+\delta H/\rho $$
(37)
where $`\delta H/`$ arises due to the loss of coherence. Considering two neutrinos and expressing this equation in the Pauli matrices basis, we find the time evolution of the density matrix to take the form
$$\frac{d\rho _\mu }{dt}=(h_{\mu \nu }+h_{\mu \nu }^{})\rho _\nu $$
(38)
where $`h`$ represents standard oscillations and $`h^{}`$ the decoherence effects. Following Ref. ellis , we parameterize $`h^{}`$ as
$$h^{}=2\left(\begin{array}{cccc}0& 0& 0& 0\\ 0& a& b& d\\ 0& b& \alpha & \beta \\ 0& d& \beta & \delta \end{array}\right).$$
(39)
The first row and column contain only zeros in order to conserve probability and obey the second law of thermodynamics. We may further constrain the theory if we assume that energy is conserved within the neutrino system. In order for this to be the case, the parameters $`d,\beta ,\delta `$ must all be identically zero. However, it is not clear what assumptions we should make from a quantum gravity perspective and so we will also examine the possibilities that these parameters have non-zero values. Using Eqs. (38) and (39), we obtain the equations,
$`\dot{\rho }_0`$ $`=`$ $`0,`$
$`\dot{\rho }_1`$ $`=`$ $`2a\rho _12\left(b{\displaystyle \frac{\mathrm{\Delta }m^2}{4E}}\right)\rho _22d\rho _3,`$
$`\dot{\rho }_2`$ $`=`$ $`2\left(b+{\displaystyle \frac{\mathrm{\Delta }m^2}{4E}}\right)\rho _12\alpha \rho _22\beta \rho _3,`$
$`\dot{\rho }_3`$ $`=`$ $`2d\rho _12\beta \rho _22\delta \rho _3.`$ (40)
In order to find the oscillation probability, we solve these equations with suitable initial conditions. There does not exist, however, a simple closed form for the solution to these equations and so, for illustrative purposes, we consider two limiting cases. The first is that the parameters $`d`$ and $`\beta `$ are zero and the second is that all the decoherence parameters are zero except $`d`$. In the first case, the oscillation probability takes the form
$`P[\nu _\alpha \nu _\beta ]`$ $`=`$ $`{\displaystyle \frac{1}{2}}\{\mathrm{cos}^22\theta (1e^{2\delta L})`$ (41)
$`+\mathrm{sin}^22\theta [1e^{(a+\alpha )L}\mathrm{cos}\left(2L[\left({\displaystyle \frac{\mathrm{\Delta }m^2}{4E}}\right)^2{\displaystyle \frac{1}{4}}(\alpha a)^2b^2]^{\frac{1}{2}}\right)]\}.`$
If we assume complete positivity benatti , necessary for the Lindblad form, (as described in section III) with energy conservation, we find the simplest possible extension to standard neutrino oscillations which includes dehoherence. In this case $`a=\alpha `$ with all other parameters equal to zero. Considering the second approximation, with non-zero $`d`$ only, gives the probability of oscillation to be
$`P[\nu _\alpha \nu _\beta ]`$ $`=`$ $`{\displaystyle \frac{1}{2}}\{\mathrm{cos}^22\theta [1{\displaystyle \frac{\omega ^2}{\mathrm{\Omega }_d^2}}+{\displaystyle \frac{d^2}{\mathrm{\Omega }_d^2}}\mathrm{cos}(2\mathrm{\Omega }_dL)]`$ (42)
$`+\mathrm{sin}^22\theta \left[1+{\displaystyle \frac{d^2\omega ^2}{\mathrm{\Omega }_d^2}}\mathrm{cos}(2\mathrm{\Omega }_dL)\right]`$
$`+\mathrm{sin}4\theta \left[{\displaystyle \frac{d}{\mathrm{\Omega }_d}}\mathrm{sin}(2\mathrm{\Omega }_dL)\right]\},`$
where $`\mathrm{\Omega }_d=\sqrt{w^2+d^2}`$ and $`\omega =\mathrm{\Delta }m^2/4E`$. Note the similarity to the probability given in Eq. (26).
So far, we have said nothing about the form of the decoherence parameters. It is clear that the decoherence parameters must have dimensions of energy but it is also possible that they have an explicit dependence on the neutrino energy. In the literature decolit1 ; decolit2 ; decolit3 , three models have received significant attention, specifically, with energy dependences $`E^0,E^1`$ and $`E^2`$. From a quantum gravity perspective, an energy dependence of $`E^2`$ is particularly interesting decostring1 ; decostring2 .
In a similar way to the LV case, it is also worthwhile studying the three neutrino case. Again, the situation is made somewhat more complicated by the existence of 3 mixing angles, two independent mass differences and many decoherence parameters. We follow an analogous method to that described above but now, instead of using the Pauli matrices as a basis, we choose the generators of SU(3). In order to derive analytical results, we choose the form of the additional decoherence matrix so we may solve the differential equations straightforwardly but keep as many decoherence parameters as possible. Including standard oscillations, we therefore find the analogous sum of $`h+h^{}`$ in Eq. (38) to be given by
$$=\left(\begin{array}{ccccccccc}0& 0& 0& 0& 0& 0& 0& 0& 0\\ 0& A& B+\omega _{21}& 0& 0& 0& 0& 0& 0\\ 0& B\omega _{21}& \mathrm{\Lambda }& 0& 0& 0& 0& 0& 0\\ 0& 0& 0& \mathrm{\Psi }& 0& 0& 0& 0& 0\\ 0& 0& 0& 0& x& y+\omega _{31}& 0& 0& 0\\ 0& 0& 0& 0& y\omega _{31}& z& 0& 0& 0\\ 0& 0& 0& 0& 0& 0& a& b+\omega _{32}& 0\\ 0& 0& 0& 0& 0& 0& b\omega _{32}& \alpha & 0\\ 0& 0& 0& 0& 0& 0& 0& 0& \delta \end{array}\right),$$
(43)
where $`\omega _{ji}=\mathrm{\Delta }m_{ji}^2/4E`$. Solving these equations and using Eq. (24) gives the oscillation probability from flavor $`p`$ to $`q`$ as
$`P[\nu _p\nu _q]`$ $`=`$ $`{\displaystyle \frac{1}{3}}+{\displaystyle \frac{1}{2}}(U_{p1}^2U_{p2}^2)(U_{q1}^2U_{q2}^2)e^{2\mathrm{\Psi }L}`$ (44)
$`+{\displaystyle \frac{1}{6}}(U_{p1}^2+U_{p2}^22U_{p3}^2)(U_{q1}^2+U_{q2}^22U_{q3}^2)e^{2\delta L}`$
$`+2U_{p1}U_{p2}U_{q1}U_{q2}e^{(A+\mathrm{\Lambda })L}\left[\mathrm{cos}(2\mathrm{\Omega }_{21}L)+{\displaystyle \frac{(\mathrm{\Lambda }A)}{2\mathrm{\Omega }_{21}}}\mathrm{sin}(2\mathrm{\Omega }_{21}L)\right]`$
$`+2U_{p1}U_{p3}U_{q1}U_{q3}e^{(x+z)L}\left[\mathrm{cos}(2\mathrm{\Omega }_{31}L)+{\displaystyle \frac{(zx)}{2\mathrm{\Omega }_{31}}}\mathrm{sin}(2\mathrm{\Omega }_{31}L)\right]`$
$`+2U_{p2}U_{p3}U_{q2}U_{q3}e^{(a+\alpha )L}\left[\mathrm{cos}(2\mathrm{\Omega }_{32}L)+{\displaystyle \frac{(\alpha a)}{2\mathrm{\Omega }_{32}}}\mathrm{sin}(2\mathrm{\Omega }_{32}L)\right],`$
where the $`U`$โs denote the entries in the standard mixing matrix introduced in Eq. (II.1.1). If we again consider neutrinos travelling large distances from astrophysical objects, we can average the $`\mathrm{sin}`$ and $`\mathrm{cos}`$ terms to zero. If, for simplicity, we assume $`\mathrm{\Psi }=\delta `$, the probability reduces to
$`P[\nu _p\nu _q]`$ $`=`$ $`{\displaystyle \frac{1}{3}}+{\displaystyle \frac{1}{6}}e^{2\delta L}[3(U_{p1}^2U_{p2}^2)(U_{q1}^2U_{q2}^2)`$
$`+(U_{p1}^2+U_{p2}^22U_{p3}^2)(U_{q1}^2+U_{q2}^22U_{q3}^2)].`$
We can now derive equations describing the flavor composition at the detector in the same way as we did in the LV case:
$`R_{\nu _e}`$ $`=`$ $`{\displaystyle \frac{1}{3}}+e^{2\delta L}[0.287\epsilon 0.065],`$
$`R_{\nu _\mu }`$ $`=`$ $`{\displaystyle \frac{1}{3}}e^{2\delta L}[0.096\epsilon 0.03],`$
$`R_{\nu _\tau }`$ $`=`$ $`{\displaystyle \frac{1}{3}}e^{2\delta L}[0.189\epsilon 0.034].`$ (45)
### IV.3 CPTV and the LSND anomaly
Having discussed how LV and CPTV may alter the phenomenology of neutrino oscillations, we are now in a position to discuss these effects within the context of the anomalous LSND result. As we discussed in section II, the LSND experiment found a mass difference for anti-neutrinos which is incompatible with that of standard oscillations with three neutrinos. Assuming that these results are indeed correct, there are two ways to explain this result. The first is to assume that there are more than three neutrinos, with the additional neutrino(s) being sterile and not directly detectable. The second is to violate CPT invariance.
#### IV.3.1 Direct CPTV and the LSND anomaly
In order to reconcile the LSND result with the rest of the neutrino oscillation data, it seems necessary to modify the neutrino sector to include different independent mass splittings for neutrinos and anti-neutrinos. In this way, we preserve the number of neutrinos but allow the mass differences to differ between the neutrino and anti-neutrino sectors. The present situation, taking in to account solar, atmospheric and KamLAND data, disfavours this scenario global , however, in both two and three neutrino models lsnd2 ; lsnd3 , leaving room only for CPTV in four neutrino models where the mixing parameters may be different in neutrino and anti-neutrino sectors lsnd4 .
#### IV.3.2 CPTV from quantum decoherence and the LSND anomaly
A second possibility for explaining the LSND result without enlarging the neutrino sector is to consider quantum decoherence in the anti-neutrino sector only lsndnick . The oscillation probability for three anti-neutrinos now takes the form of Eq. (44), whilst neutrinos experience no quantum decoherence. Since decoherence parameters are present in the arguments of the sine and cosine terms, they alter the effective mass differences leading to an apparent difference between the measured mass differences in the neutrino and anti-neutrino sectors. It is therefore possible to reconcile the LSND results with other existing oscillation data lsndnick . However, this particular model fails to fit the spectral distortions observed in the KamLAND experiment KLspec . Having said that, the authors of Ref. lsndnick chose only one set of quantum decoherence parameters and so there is still much scope for further investigation.
## V Sources and Detection of High-Energy Neutrinos
### V.1 Sources of high-energy neutrinos
High-energy neutrinos are thought to be generated in a wide range of astrophysical sources. Such sources may include Active Galactic Nuclei (AGN) agn , Gamma-Ray Bursts (GRB) grb , microquasars microquasars , supernova remnants, star clusters and X-ray binaries. Also, ultra-high protons or nuclei travelling over cosmological distances can interact with the Cosmic Microwave Background (CMB) and/or Cosmic Infra-Red Background (CIRB), generating what is often called the cosmogenic neutrino flux cosmogenic .
In any of these sources, there are basically two mechanisms by which high-energy neutrinos are generated. Firstly, Fermi accelerated protons (or charged nuclei) can collide with hadronic or photonic targets generating charged and neutral pions. These charged pions then decay, $`\pi ^+\mu ^+\nu _\mu e^+\overline{\nu }_e\nu _e\nu _\mu `$, $`\pi ^{}\mu ^{}\overline{\nu }_\mu e^{}\nu _e\overline{\nu }_e\overline{\nu }_\mu `$, generating electron and muon neutrinos and anti-neutrinos. Secondly, atomic nuclei which undergo Fermi acceleration can be disintegrated by interacting with infra-red photons surrounding their source. Neutrons broken off of such a nucleus can then decay, $`np^+e^{}\overline{\nu }_e`$, generating a flux of electron anti-neutrinos.
The important thing to keep in mind regarding these two mechanisms for high-energy neutrino generation is the quantity of neutrinos produced of various flavors. Neutrinos produced in charged pion decay follow the ratio: $`\nu _e:\nu _\mu :\nu _\tau =1/3:2/3:0`$, while those produced in neutron decay follow: $`\nu _e:\nu _\mu :\nu _\tau =1:0:0`$. It is also important to note that pion decay can generate both neutrinos and anti-neutrinos, while neutron decay generates only anti-neutrinos.
For the purposes of the measurement of flavor ratios, identified point sources of high-energy neutrinos are considerably more useful than diffuse fluxes. With such a source (or sum of sources), the distance the neutrinos have propagated will likely be known. Furthermore, the background from atmospheric neutrinos can be controlled by only considering events from one direction in the sky. In some sources which emit neutrinos only for short lengths of time, GRB and AGN flares for example, the background can be further reduced by only considering events in particular time windows. For these reasons, GRB and AGN are likely to be among the most useful source of high-energy neutrinos for the purposes of flavor identification although bright and nearby (galactic) sources, if present, could also be very useful.
In addition to these theoretical arguments, there is some limited experimental evidence that might suggest the existance of bright point sources of high-energy neutrinos. Firstly, it has been argued recently that anisotropies observed in the cosmic ray spectrum at EeV energies is the result of neutrons propagating from galactic sources. If this is the case, then large fluxes of high-energy (anti-)neutrinos will also be generated neutrondecay . Secondly, the AMANDA-II experiment has recently reported the detection of two neutrinos coincident with TeV flares seen by the Whipple gamma-ray telescope from the blazar (AGN) 1ES 1959+650 1959 . These events were not found in a blind analysis, however, so their statistical significance cannot be determined. If these neutrinos are the product of this TeV blazar, it would suggest the existance of very bright point sources of high-energy neutrinos.
More exotic processes which do not fall into these two descriptions may also be capable of generating high-energy cosmic neutrinos. Such possibilities include annihilating or decaying dark matter or topological defects darkmatter , Hawking radiating primordial black holes hawkbh , or the interactions of ultra-high energy neutrinos with the cosmic neutrino background via the $`Z`$-burst mechanism zburst . For a review of sources of high-energy neutrinos and other aspects of high-energy neutrino astronomy, see Ref. neutrinoreview .
### V.2 High-energy neutrino detection
Once such neutrinos are generated and propagate to Earth, they can be detected in one of several ways. Neutral current interactions of neutrinos of all flavors with nucleons generates hadronic showers which can be observed. Charged current interactions of electron and muon neutrinos generate, in addition to hadronic showers, potentially observable electromagnetic showers and muons, respectively. The tau leptons generated in the charged current interactions of tau neutrinos can produce a class of events unique to tau neutrinos: double bangs and lollipops.
Many of the experimental techniques being developed and deployed are only capable of detecting showers generated in high-energy neutrino interactions. Observing shower events alone will not enable the flavor ratios of a flux of cosmic neutrinos to be identified, however. For this reason we focus on experiments capable of observing high-energy neutrinos in the form of showers, muon tracks and tau-unique events. In particular, we focus on next generation, kilometer-scale, optical Cerenkov detectors. These include IceCube, currently under construction at the South Pole, and possibly a future kilometer-scale neutrino telescope in the Mediterranean Sea, sometimes called KM3.
#### V.2.1 Shower events
All high-energy neutrino interactions produce an electromagnetic and/or hadronic shower. The probability of detecting a hadronic shower produced in a neutral current interaction as a neutrino travels through the effective area of the detector is given by
$$P_{\nu \mathrm{shower}}=\rho N_AL_{E_{\mathrm{sh}}^{\mathrm{thr}}/E_\nu }^1\frac{d\sigma }{dy}๐y,$$
(46)
where $`\rho `$ is the target nucleon density, $`N_A`$ is Avogadroโs number, $`L1`$ km is the length of the detector, $`d\sigma /dy`$ is the differential neutrino-nucleon neutral current cross section crosssection , $`y`$ is the fraction of energy which is transferred from the neutrino (and therefore the fraction of energy which goes into the shower) and $`E_{\mathrm{sh}}^{\mathrm{thr}}3`$ TeV is the threshold energy for the experiment detecting a shower event.
In the charged current interactions of electron neutrinos, all of the neutrinoโs energy goes into a combination of electromagnetic and hadronic showers. In this case, the probability of detecting a shower reduces to
$$P_{\nu \mathrm{shower}}=\rho N_AL\sigma ,$$
(47)
if $`E_\nu >E_{\mathrm{sh}}^{\mathrm{thr}}`$, and zero otherwise. $`\sigma `$ in this expression is the total charged current neutrino-nucleon cross section crosssection .
In these expressions, we have made the simplifying assumption that only showers generated inside of the detector volume can be detected. This is not always true, particularly at very high energies. Very energetic showers which are initiated outside of the experimentโs instrumented volume can expand into the experiment. Another way of saying this is that the effective volume of such an experiment is generally larger than its instrumented volume at very high energies.
#### V.2.2 Muon events
Charged current interactions of high-energy muon neutrinos produce muons which can travel through the medium of the experiment (ice or water) and potentially into the detector volume. The energy loss rate of such a muon is given by
$$\frac{dE}{dX}\alpha \beta E,$$
(48)
where the parameters in ice or water are given by $`\alpha 2.0`$ MeV cm<sup>2</sup>/g and $`\beta 4.2\times 10^6`$ cm<sup>2</sup>/g dutta . The distance a muon travels before its energy drops below the threshold, $`E_\mu ^{\mathrm{thr}}`$, is given then by
$$R_\mu =\frac{1}{\beta }\mathrm{ln}\left[\frac{\alpha +\beta E_\mu }{\alpha +\beta E_\mu ^{\mathrm{thr}}}\right].$$
(49)
This quantity if often referred to as the muon range. $`E_\mu ^{\mathrm{thr}}`$ 50-100 GeV are typical for high-energy neutrino telescopes. The muon range can extend for many kilometers for very high energy muons, dramatically increasing the number of muon tracks that are observed.
The probability of detecting a muon produced in a charged current interaction is given by
$$P_{\nu _\mu \mu }\rho N_A\sigma R_\mu ,$$
(50)
where, here, $`\sigma `$ is the total charged current neutrino-nucleon cross section crosssection . In cases with the muon neutrino comes from above the ice or water, the value of $`R_\mu `$ used in this equation should not exceed the depth of the experiment. Similarly, if the neutrino comes from below the detector, the rock or other material below the ice or water should be accounted for.
#### V.2.3 Events involving tau neutrinos
For tau neutrinos with energies less than $``$PeV, their interactions generate only shower events, as any tau leptons generated decay before they can be identified. Higher energy tau neutrinos, on the other hand, can be identified by the combined signatures of tau tracks and showers.
Double bang doublebang ; measure events are produced with a tau neutrino interacts via charged current inside of the detector, producing a tau lepton which travels across the detector to decay, still inside of the detector volume. The hadronic shower produced in the initial interaction constitutes the first โbangโ while the second shower is generated in the tauโs decay. A tau lepton can travel a distance
$$R_\tau =\frac{E_\tau c\tau _\tau }{m_\tau }\frac{E_\tau }{1\mathrm{PeV}}\mathrm{\hspace{0.17em}50}\mathrm{meters}$$
(51)
before decaying. To resolve two showers in a high-energy neutrino telescope, they must be separated by a distance of roughly 100-400 meters. Furthermore, they cannot be separated by more than about 1000 meters and both be within the detector volume. Therefore double bang events are most useful in the energy range between a few PeV and one hundred PeV.
At higher energies, lollipop events become very useful measure . A lollipop event is observed when the first shower of a double bang event occurs outside of the detector (without being observed) with the tau track extending into the detector and decaying. You might also think that an observation of the first shower with a tau track would be a useful signature, but muons and taus produced in ordinary hadronic or electromagnetic showers could mimic such an event.
The probability of observing a double bang event from a given incident tau neutrino is given by
$$P_{\mathrm{DB}}\rho N_A_0^1๐y\frac{d\sigma }{dy}_{x_{\mathrm{min}}}^L๐x\frac{(Lx)}{R_\tau }e^{x/R_\tau },$$
(52)
where as the probability for a lollipop event is
$$P_{\mathrm{LP}}\rho N_A(Lx_{\mathrm{min}})_0^1๐y\frac{d\sigma }{dy}e^{x/R_\tau }.$$
(53)
In each of these expressions, $`x_{\mathrm{min}}100400`$ meters, is the minimum shower seperation required to seperate the showers.
For a detailed description of both double bang and lollipop tau events, see Ref. measure .
#### V.2.4 Neutrino events in cosmic ray experiments
Although in this article we focus on next generation optical Cerenkov neutrino telescopes, it is interesting to point out that ultra-high energy cosmic ray experiments may also have a limited ability to resolve neutrino flavors. These experiments can detect ultra-high energy neutrinos in essentially two ways. First, quasi-horizontal neutrinos can penetrate deeply into the atmosphere before interacting, producing showers which are distinguishable from those initiated by cosmic rays. Second, tau neutrinos which skim the Earth can, through charged current interactions, produce tau leptons which escape the Earth before decaying and generating a shower neutrinocr . While the former signature can be produced by neutrinos of all three flavors, the latter are uniquely generated by tau neutrinos. This, in principle, could be used to measure the fraction of the ultra-high energy neutrino flux which consists of tau neutrinos.
## VI Signatures of CPT and Lorentz Violation
There are two primary methods of probing physics beyond the Standard Model with high-energy neutrinos. First, neutrino-nucleon interactions can be observed measuring the cross section nucross and other characteristics in hope of identifying new interactions: microscopic black hole or P-brane production nublackhole , processes resulting from low-scale gravity nugravity , string effects nustring or electoweak instantons nuinstanton . Second, the ratios of neutrino flavors can be measured, potentially identifying the effects of neutrino decay nudecay , pseudo-Dirac states nupseudodirac or quantum decoherence decolit1 . In this article, we focus on using this latter technique to constrain or discover the effects of CPT and Lorentz violation.
### VI.1 Flavor ratio predictions
To assess the prospects of identifying the violation of CPT and Lorentz invariance in the high-energy neutrino sector, we must first determine the flavor ratios predicted in various scenarios. First of all, we consider the case with only known physics, and no CPT or Lorentz violation. For an initial set of ratios corresponding to neutrinos from pion decay, $`\nu _e:\nu _\mu :\nu _\tau =1/3:2/3:0`$, after oscillations over a long distance (which will always be the case for high-energy neutrino astronomy), these ratios become $`\nu _e:\nu _\mu :\nu _\tau 0.36:0.33:0.30`$. (Anti)-neutrinos coming from neutron decay, on the other hand, are generated in the ratio $`\nu _e:\nu _\mu :\nu _\tau =1:0:0`$, which oscillates to $`\nu _e:\nu _\mu :\nu _\tau 0.56:0.26:0.18`$.
These ratios can be changed dramatically if the effects of CPT or Lorentz violation are present. In the case of LV, above the energy threshold for such effects, the neutrino flavor ratios from pion decay are modified as $`\nu _e:\nu _\mu :\nu _\tau =1/3:2/3:00.42:0.57:0.013`$. The flavor ratios from neutron decay are modified as $`\nu _e:\nu _\mu :\nu _\tau =1:0:00.70:0.27:0.027`$. In the case of quantum decoherence, on the other hand, all ratios shift toward $`\nu _e:\nu _\mu :\nu _\tau =1/3:1/3:1/3`$, regardless of their source (see figure 2).
### VI.2 Lorentz Violation
There are two potentially identifiable features in the neutrino flavor ratios predicted as a result of LV. First, in the case of neutrinos generated via pion decay, a particularly large fraction of these neutrinos will be of muon flavor. Secondly, regardless of whether the neutrinos are generated in pion or neutron decay, only a very small fraction will appear with tau flavor.
#### VI.2.1 A large muon neutrino fraction
If the effects of LV are considered in neutrinos generated via pion decay, the result can be a neutrino flux which is of nearly 60% muon flavor. A large muon to shower ratio could, therefore, be seen as a signature of LV. The ability of high-energy neutrino telescopes to measure the ratio of muon to shower events and corresponding flavor ratio is discussed in detail in Ref. measure .
#### VI.2.2 A tau neutrino deficit
Looking for a deficit of high-energy cosmic tau neutrinos may also be useful in identifying the effects of LV. Events that are identifiable as tau neutrinos (double bang and lollipop events) are somewhat rare, however. Assuming a spectrum proportional to $`E_\nu ^2`$, a flux of $`E_{\nu _\tau }^2dN_{\nu _\tau }/dE_{\nu _\tau }=10^7`$ GeV cm<sup>-2</sup> s<sup>-1</sup> would yield only about 0.35 tau-unique events per year in a cubic kilometer experiment. Over several years of observation, observing a deficit of tau neutrinos may indeed become possible if bright point sources of PeV-EeV neutrinos are discovered.
The absence of tau neutrino induced Earth-skimming showers at cosmic ray experiments, such as Auger, could also be an anticipated signature of LV effects. As experiments such as Auger most efficiently detect showers with energies above $`10^8`$ GeV, or so, they are particularly well suited for testing the effects of LV.
### VI.3 Quantum Decoherence
The signature of quantum decoherence in cosmic neutrinos, is the presence of an equal fraction of neutrinos of each flavor. Since sources which produce neutrinos through pion decay generate nearly this ratio (after oscillations) without the effect of quantum decoherence, they are not very useful in probing for these effects. Sources of high-energy (anti-)neutrinos produced through neutron decay, on the other hand, can be used to potentially identify these effects decolit1 .
In figure 3, we show the effects of quantum decoherence on the events observed in a next generation high-energy neutrino telescope. In the left frame, we consider a model with a $`\delta E^2`$ dependence, normalized such that quantum decoherence sets in at the 10 TeV scale ($`\delta =(E/10,000\mathrm{GeV})^2/\mathrm{L}`$). In the right frame, the source is sufficiently distant that the effects of quantum decohence have set in fully ($`\nu _e:\nu _\mu :\nu _\tau =1/3:1/3:1/3`$). In either case, the number of muon tracks detected are increased and the number of showers detected decreases. Since the normalization of this flux will likely be unknown, it is the ratio of these event types that is of the most interested to us.
In figure 3, we plot the total rate of muon events and the rate of contained muon events (i.e. with a vertex contained within the detector volume) seperately. We do this to illustrate that energy dependent effects, such as those shown in the left frame of this figure, are more clearly identified by using only contained events. The statistics are considerably better when using all of the events, however. Depending on the flux of high-energy neutrinos present, a strategy to use these different types of events most effectively will have to be devised.
The prospects for detecting the effects of quantum decoherence or other signatures of CPT or Lorentz violation ultimately depend on the variety of high-energy neutrino sources which exist in nature. A detailed study of IceCubeโs sensitivity to decoherence effects from observing anti-neutrinos from the Cygnus spiral arm is currently underway cygnus . This is likely to be one of the most useful sources for constraining the effects of decoherence.
## VII Conclusions
High-energy neutrino astronomy provides an opportunity to observe particles at extremely high energies and over extremely long baselines. Both of these characteristics make such experiments particularly adept at testing for the effects of CPT and Lorentz violation.
In this article, we have discussed in detail the effects that CPT and Lorentz violation can have on the flavors of high-energy cosmic neutrinos observed at Earth. After discussing the theoretical basis and motivations for such effects, we calculated the ratios of neutrino flavors predicted to be observed from astrophysical sources of high-energy neutrinos in various CPT and Lorentz violating scenarios.
The effects of Lorentz violation may potentially be detected or constrained by the observation, or lack thereof, of anomalously large fractions of the cosmic neutrino spectrum consisting of muon neutrinos, accompanied by very few tau neutrinos. This deviation from the standard neutrino oscillation phenomenology will occur above a model-dependent energy threshold, and thus experiments capable of detecting extremely high energy neutrinos are particularly useful in such measurements.
The effects of quantum decoherence may also be observable in cosmic neutrinos. In particular, anti-neutrinos generated in the decays of neutrons produced in the photo-disintegration of ultra-high energy cosmic nuclei will have their flavor ratios significantly modified if quantum decoherence effects are significant. The very long baselines over which such neutrinos travel before reaching Earth provide an opportunity to test for these effects with much greater precision than other techniques can achieve.
Acknowledgements:
The work of DM is supported by a studentship from the University of Sheffield. The work of EW is supported by PPARC, grant reference number PPA/G/S/2003/00082, the Royal Society and the London Mathematical Society. EW would like to thank the Universities of Durham and Newcastle-upon-Tyne and University College Dublin for hospitality while this work was completed. DH is supported by the Leverhulme trust.
|
warning/0506/math-ph0506056.html
|
ar5iv
|
text
|
# Fractal Weyl Law for Open Chaotic Maps
## 1 Introduction
We summarize our work in collaboration with Maciej Zworski nz , on the semiclassical density of resonances for a quantum open system, in the case when the associated classical dynamics is uniformly hyperbolic, and the set of trapped trajectories is fractal repeller. The system we consider is not a Hamiltonian flow, but rather a โsymplectic map with a holeโ on a compact phase space (the 2-torus). Such a map can be considered as a model for the Poincarรฉ section associated with a scattering Hamiltonian on $`^2`$, at some positive energy; the โholeโ represents the points which never return to the Poincarรฉ section, that is, which are scattered to infinity. We then quantize this open map, obtaining a sequence of subunitary operators, the eigenvalues of which are interpreted as resonances.
We are especially interested in the asymptotic density of โlong-living resonancesโ, representing metastable states which decay in a time bounded away from zero (as opposed to โshort resonancesโ, associated with states decaying instantaneously). Our results (both numerical and analytical) support the conjectured fractal Weyl law, according to which the number of long-living resonances scales as $`\mathrm{}^d`$, where $`d`$ is the (partial) fractal dimension of the trapped set.
### 1.1 Generalities on resonances
A Hamiltonian dynamical system (say, $`H(q,p)=p^2+V(q)`$ on $`^{2n}`$) is said to be โclosedโ at the energy $`E`$ when the energy surface $`\mathrm{\Sigma }_E`$ is a compact subset of the phase space. The associated quantum operator $`H_{\mathrm{}}=\mathrm{}^2\mathrm{\Delta }+V(q)`$ then admits discrete spectrum near the energy $`E`$ (for small enough $`\mathrm{}`$). Furthermore, if $`E`$ is nondegenerate (meaning that the flow of $`H`$ has no fixed point on $`\mathrm{\Sigma }_E`$), then the semiclassical density of eigenvalues is given by the celebrated Weylโs law Ivr :
$$\mathrm{\#}\left\{\mathrm{Spec}(H)[E\delta ,E+\delta ]\right\}=\frac{1}{(2\pi \mathrm{})^n}_{|H(q,p)E|<\delta }dqdp+๐ช(\mathrm{}^{1n}).$$
(1)
This formula connects the density of quantum eigenvalues with the geometry of the classical energy surface $`\mathrm{\Sigma }_E`$. It shows that the number of resonances in an interval of type $`[E+C\mathrm{},EC\mathrm{}]`$ is of order $`๐ช(\mathrm{}^{1n})`$. Intuitively, this Weyl law means that one quantum state is associated with each phase space cell of volume $`(2\pi \mathrm{})^n`$.
When $`\mathrm{\Sigma }_E`$ is non-compact, or even of infinite volume, the spectral properties of $`H_{\mathrm{}}`$ are different. Consider the case of a scattering situation, when the potential $`V(q)`$ is of compact support: for any $`E>0`$, $`\mathrm{\Sigma }_E`$ is unbounded, and $`H_{\mathrm{}}`$ admits absolutely continuous spectrum on $`[0,\mathrm{})`$. However, one can meromorphically continue the resolvent $`(zH_{\mathrm{}})^1`$ across the real axis from the upper half-plane into the lower half-plane. In general, this continuation will have discrete poles $`\{z_j=E_j\mathrm{i}\gamma _j\}`$ with โwidthsโ $`\gamma _j>0`$, which are the resonances of $`H_{\mathrm{}}`$.
Physically, each resonance is associated with a metastable state: a (not square-integrable) solution of the Schrรถdinger equation at the energy $`z_j`$, which decays like $`\mathrm{e}^{t\gamma _j/\mathrm{}}`$ when $`t+\mathrm{}`$. In spectroscopy experiments, one measures the energy dependence of some scattering cross-section $`\sigma (E)`$. Each resonance $`z_j`$ imposes a Lorentzian component $`\frac{\gamma _j}{(EE_j)^2+\gamma _j^2}`$ on $`\sigma (E)`$; a resonance $`z_j`$ will be detectable on the signal $`\sigma (E)`$ only if its Lorentzian is well-separated from the ones associated with nearby resonances of comparable widths, therefore iff $`|E_j^{}E_j|\gamma _j`$. This condition of โwell-separabilityโ is NOT the one we will be interested in here. We will rather consider the order of magnitude of each resonance lifetime $`\mathrm{}/\gamma _j`$, independently of the nearby ones, in the semiclassical rรฉgime: a resonant state will be โvisibleโ, or โlong-livingโ, if $`\gamma _j=๐ช(\mathrm{})`$. Our objective will be to count the number of resonances $`z_j`$ in boxes of the type $`\{|E_jE|C\mathrm{},\gamma _jC\mathrm{}\}`$, or equivalently $`\{|z_jE|C\mathrm{}\}`$.
### 1.2 Trapped sets
Since resonant states are โinvariant up to rescalingโ, it is natural to relate them, in the semiclassical spirit, to invariant structures of the classical dynamics. For a scattering system, the set of points (of energy $`E`$) which donโt escape to infinity (either in past future) is called the *trapped set* at energy $`E`$, and denoted by $`K(E)`$. The textbook example of a radially-symmetric potential shows that this set may be empty (if $`V(r)`$ decreases monotonically from $`r=0`$ to $`r\mathrm{}`$), or have the same dimension as $`\mathrm{\Sigma }_E`$ (if $`V(r)`$ has a maximum $`V(r_0)>0`$ before decreasing as $`r\mathrm{}`$).
For $`n=2`$ degrees of freedom, the geometry of the trapped set can be more complex. Let us consider the well-known example of $`2`$-dimensional scattering by a set of non-overlapping disks GasRic ; Cv-E (a similar model was studied in Troll ; BluSmi ).
When the scatterer is a single disk, the trapped set is obviously empty.
The scattering by two disks admits a single trapped periodic orbit, bouncing back and forth between the disks. Since the evolution between two bounces is โtrivialโ, it is convenient to represent the scattering system through the *bounce map* on the reduced phase space (position along the boundaries $`\times `$ velocity angle). This map is actually defined only on a fraction of this phase space, namely on those points which will bounce again at least once. For the 2-disk system, this map has a unique periodic point (of period $`2`$), which is of hyperbolic nature due to the curvature of the disks. The trapped set $`K`$ of the map (โreducedโ trapped set) reduces to this pair of points; it lies at the intersection of the forward trapped set $`\mathrm{\Gamma }_{}`$ (points trapped as $`t+\mathrm{}`$) and the backward trapped set $`\mathrm{\Gamma }_+`$ (points trapped as $`t\mathrm{}`$).
The addition of a third disk generates a complex bouncing dynamics, for which the trapped set is a *fractal repeller* GasRic . We will explain in the next section how such a structure arises in the case of the open bakerโs map. As in the $`2`$-disk case, the bounce map is uniformly hyperbolic; each forward trapped point $`x\mathrm{\Gamma }_{}`$ admits a stable manifold $`W_{}(x)`$ (and vice-versa for $`x\mathrm{\Gamma }_+`$). One can show that $`\mathrm{\Gamma }_{}`$ is fractal along the unstable direction $`W_+`$: $`\mathrm{\Gamma }_{}W_+`$ has a Hausdorff dimension $`0<d<1`$ which depends on the positions and sizes of the disks. Due to time-reversal symmetry, $`\mathrm{\Gamma }_+W_{}`$ has the same Hausdorff dimension. Finally, the reduced trapped set $`K=\mathrm{\Gamma }_+\mathrm{\Gamma }_{}`$ is a fractal of dimension $`2d`$, which contains infinitely many periodic orbits. The unreduced trapped set $`K(E)\mathrm{\Sigma }_E`$ has one more dimension corresponding to the direction of the flow, so it is of dimension $`D=2d+1`$.
### 1.3 Fractal Weyl law
We now relate the geometry of the trapped set $`K(E)`$, to the density of resonances of the quantized Hamiltonian $`H_h`$ in boxes $`\left\{|zE|C\mathrm{}\right\}`$. The following conjecture (which dates back at least to the work of Sjรถstrand SjDuke ) relates this density with the โthicknessโ of the trapped set.
###### Conjecture 1
Assume that the trapped set $`K(E)`$ at energy $`E`$ has dimension $`2d_E+1`$. Then, the density of resonances near $`E`$ grows as follows in the semiclassical limit:
$$r>0,\frac{\mathrm{\#}\left\{\mathrm{Res}(H_{\mathrm{}})\{z:|zE|<r\mathrm{}\}\right\}}{\mathrm{}^{d_E}}\stackrel{h0}{}c_E(r),$$
(2)
for a certain โshape functionโ $`0c_E(r)<\mathrm{}`$.
We were voluntarily rather vague on the concept of โdimensionโ (a fractal set can be characterized by many different dimensions). In the case of a closed system, $`K(E)`$ has dimension $`2n1`$, so we recover the Weyl law (1). If $`K(E)`$ consists in one unstable periodic orbit, the resonances form a (slightly deformed) rectangular lattice of sides $`\mathrm{}`$, so each $`\mathrm{}`$-box contains at most finitely many resonances Sj2 .
For intermediate situations ($`0<d_E<n1`$), one has only been able to prove one half of the above estimate, namely the *upper bound* for this resonance counting SjDuke ; ZwIn ; GLZ ; SjZw04 . The dimension appearing in these upper bounds is the *Minkowski dimension* defined by measuring $`ฯต`$-neighborhoods of $`K(E)`$. In the case we will study, this dimension is equal to the Hausdorff one. Some lower bounds for the resonance density have been obtained as well SjZw-lower , but are far below the conjectured estimate.
Several numerical studies have attempted to confirm the above estimate for a variety of scattering Hamiltonians GLZ ; L ; LZ ; LSZ , but with rather inconclusive results. Indeed, it is numerically demanding to compute resonances. One method is to โcomplex rotateโ the original Hamiltonian into a non-Hermitian operator, the eigenvalues of which are the resonances. Another method uses the (approximate) relationship between, on one side, the resonance spectrum of $`H_{\mathrm{}}`$, one the other side, the set of zeros of some semiclassical zeta function, which is computed from the knowledge of classical periodic orbits Cv-E ; LSZ . In the case of the geodesic flow on a convex co-compact quotient of the Poincarรฉ disk (which has a fractal trapped set), the resonances of the Laplace operator are *exactly* given by the zeros of Selbergโs zeta function. Even in that case, it has been difficult to check the asymptotic Weyl law (2), due to the necessity to reach sufficiently high values of the energy GLZ .
### 1.4 Open maps
Confronted with these difficulties to deal with open Hamiltonian systems, we decided to study semiclassical resonance distributions for toy models which have already proven efficient to modelize closed systems. In the above example of obstacle scattering, the bounce map emerged as a way to simplify the description of the classical dynamics. It acts on a reduced phase space, and gets rid of the โtrivialโ evolution between bounces. The exact quantum problem also reduces to analyzing an operator acting on wavefunctions on the disk boundaries, but this operator is infinite-dimensional, and extracting its resonances is not a simple task GasRic ; BluSmi .
Canonical maps on the $`2`$-torus were often used to mimic closed Hamiltonian systems; they can be quantized into unitary matrices, the eigenphases of which are to be compared with the eigenvalues $`\mathrm{e}^{\mathrm{i}E_j/\mathrm{}}`$ of the propagator $`\mathrm{e}^{\mathrm{i}H_{\mathrm{}}/\mathrm{}}`$ (see e.g. DEGra and references therein for a mathematical introduction on quantum maps).
We therefore decided to construct a โtoy bounce mapโ on $`๐^2`$, with dynamics similar to the original bounce map, and which can be easily quantized into an $`N\times N`$ subunitary matrix (where $`N=(2\pi \mathrm{})^1`$). This matrix is then easily diagonalized, and its subunitary eigenvalues $`\{\lambda _j\}`$ should be compared with the set $`\{\mathrm{e}^{\mathrm{i}z_j/\mathrm{}}\}`$, where the $`z_j`$ are the resonances of $`H_{\mathrm{}}`$ near some positive energy $`E`$. We cannot prove any direct correspondence between, on one side the eigenvalues of our quantized map, on the other side resonances of a bona fide scattering Hamiltonian. However, we expect a semiclassical property like the fractal Weyl law to be robust, in the sense that it should be shared by all types of โquantum modelsโ. To support this claim, we notice that the usual, โclosedโ Weyl law is already (trivially) satisfied by quantized maps: the number of eigenphases $`\theta _j`$ on the unit circle (corresponding to an energy range $`\mathrm{\Delta }E=2\pi \mathrm{}`$) is exactly $`N=(2\pi \mathrm{})^1`$, which agrees with the Weyl law (1) for $`n=2`$ degrees of freedom. Testing Conjecture 1 in the framework of quantum maps should therefore give a reliable hint on its validity for more realistic Hamiltonian systems.
Schomerus and Tworzydลo recently studied the quantum spectrum of an open chaotic map on the torus, namely the open kicked rotator schomerus ; they obtain a good agreement with a fractal Weyl law for the resonances (despite the fact that the geometry of the trapped set is not completely understood for that map). The authors also provide a heuristic argument to explain this Weyl law. We believe that this argument, upon some technical improvement, could yield a rigorous proof of the upper bound for the fractal Weyl law in case of maps.
We preferred to investigate that problem using one of the best understood chaotic maps on $`๐^2`$, namely the bakerโs map.
## 2 The open bakerโs map and its quantization
### 2.1 Classical closed baker
The (closed) bakerโs map is one of the simplest examples of uniformly hyperbolic, strongly chaotic systems (it is a perfect model of Smaleโs horseshoe). The โ3-bakerโs mapโ $`B`$ on $`๐^2[0,1)\times [0,1)`$ is defined as follows:
$$๐^2(q,p)B(q,p)=\{\begin{array}{cc}(3q,\frac{p}{3})\hfill & \mathrm{if}0q<1/3,\hfill \\ (3q1,\frac{p+1}{3})\hfill & \mathrm{if}1/3q<2/3,\hfill \\ (3q2,\frac{p+2}{3})\hfill & \mathrm{if}2/3q<1.\hfill \end{array}$$
(3)
This map preserves the symplectic form $`dqdp`$ on $`๐^2`$, and is invertible. Compared with a generic Anosov map, it has the particularity to be linear by parts, and its linearized dynamics (well-defined away from its lines of discontinuities) is independent of the point $`x๐^2`$. As a consequence, the stretching exponent is constant on $`๐^2`$, as well as the unstable/stable directions (horizontal/vertical).
This map admits a very simple Markov partition, made of the three vertical rectangles $`R_j=\{q[j/3,(j+1)/3),p[0,1)\}`$, $`j=0,1,2`$ (see Fig. 1). Any bi-infinite sequence of symbols $`\mathrm{}ฯต_2ฯต_1ฯต_0ฯต_1ฯต_2\mathrm{}`$ (where each $`ฯต_i\{0,1,2\}`$) will be associated with the *unique* point $`x`$ s.t. $`B^t(x)R_{ฯต_t}`$ for all $`t`$. This is the point of coordinates $`(q,p)`$, where $`q`$ and $`p`$ admit the ternary decompositions
$$q=0ฯต_0ฯต_1\mathrm{}\stackrel{\mathrm{def}}{=}\underset{i1}{}\frac{ฯต_{i1}}{3^i},p=0ฯต_1ฯต_2\mathrm{}.$$
The bakerโs map $`B`$ simply acts as a shift on this symbolic sequence:
$$B(x=\mathrm{}ฯต_2ฯต_1ฯต_0ฯต_1ฯต_2\mathrm{})=\mathrm{}ฯต_2ฯต_1ฯต_0ฯต_1ฯต_2\mathrm{}.$$
(4)
### 2.2 Opening the classical map
We explained above that the bounce maps associated with the $`2`$\- or $`3`$-disk systems were defined only on parts of the reduced phase space, namely on those points which bounce at least one more time. The remaining points, which escape to infinity right after the bounce, have no image through the map.
Hence, to open our bakerโs map $`B`$, we just decide to restrict it on a subset $`S๐^2`$, or equivalently we send points in $`๐^2S`$ to infinity. We obtain an Anosov map โwith a holeโ, a class of dynamical systems recently studied in the literature Cher2 . The study is simpler when the hole corresponds to a Markov rectangle Cher1 , so this is the choice we will make (we expect the fractal Weyl law to hold for an arbitrary hole as well). Let us choose for the hole the second Markov rectangle $`R_1`$, so that $`S=R_0R_2`$. Our open map $`C=B_S`$ reads (see Fig. 1):
$$C(q,p)=\{\begin{array}{cc}(3q,\frac{p}{3})\hfill & \mathrm{if}qR_0,\hfill \\ (3q2,\frac{p+2}{3})\hfill & \mathrm{if}qR_2.\hfill \end{array}$$
(5)
This map is canonical on $`S`$, and its inverse $`C^1`$ is defined on the set $`C(S)`$. Our choice for $`S`$ coincides with the points $`x=(q,p)`$ satisfying $`ฯต_0(x)\{0,2\}`$ (equivalently, points s.t. $`ฯต_0(x)=1`$ are sent to infinity through $`C`$). This allows us to characterize the trapped sets very easily:
* the forward trapped set $`\mathrm{\Gamma }_{}`$ (see fig. 2) is made of the points $`x`$ which will never fall in the strip $`R_1`$ for times $`t0`$: these are the points s.t. $`ฯต_i\{0,2\}`$ for all $`i0`$, with no constraint on the $`ฯต_i`$ for $`i<0`$. This set is of the form $`\mathrm{\Gamma }_{}=Can\times [0,1)`$, where $`Can`$ is the standard $`1/3`$-Cantor set on the unit interval. As a result, the intersection $`\mathrm{\Gamma }_{}W_+Can`$ has the Hausdorff (or Minkowski) dimension $`d=\frac{\mathrm{log}2}{\mathrm{log}3}`$.
* the backward trapped set $`\mathrm{\Gamma }_+`$ is made of the points satisfying $`ฯต_i\{0,2\}`$ for all $`i<0`$, and is given by $`[0,1)\times Can`$.
* the full trapped set $`K=Can\times Can`$.
### 2.3 Quantum bakerโs map
We now describe in some detail the quantization of the above maps. We recall DEGra ; DBgiens that a nontrivial quantum Hilbert space can be associated with the phase space $`๐^2`$ only for discrete values of Planckโs constant, namely $`\mathrm{}=(2\pi N)^1`$, $`N_0`$. In that case (the only one we will consider), this space $`_N`$ is of dimension $`N`$. It admits the โpositionโ basis $`\{Q_j,j=0,\mathrm{},N1\}`$ made of the โDirac combsโ
$$Q_j(q)=\frac{1}{\sqrt{N}}\underset{\nu }{}\delta (q\frac{j}{N}\nu ).$$
This basis is connected to the โmomentumโ basis $`\{P_k,k=0,\mathrm{},N1\}`$ through the discrete Fourier transform:
$$P_k|Q_j=(_N)_{kj}=\frac{\mathrm{e}^{2\mathrm{i}\pi Nkj}}{\sqrt{N}},j,k\{0,\mathrm{},N1\},$$
(6)
where the Fourier matrix $`F_N`$ is unitary. Balazs and Voros BaVo proposed to quantize the closed bakerโs map $`B`$ as follows, when $`N`$ is a multiple of $`3`$ (a condition we will always assume): in the position basis, it takes the block form
$$B_N=_N^1\left(\begin{array}{ccc}_{N/3}& & \\ & _{N/3}& \\ & & _{N/3}\end{array}\right).$$
(7)
This matrix is obviously unitary, and exactly satisfies the Van Vleck formula (the semiclassical expression for a quantum propagator, in terms of the classical generating function). In the semiclassical limit $`N\mathrm{}`$, it was shown DENW that these matrices classically propagate Gaussian coherent states supported far enough from the lines of discontinuities. As usual, discontinuities of the classical dynamics induce diffraction effects at the quantum level, which have been partially analyzed for the bakerโs map ToVaSa (in particular, diffractive orbits have to be taken into account in the Gutzwiller formula for $`\mathrm{tr}(B_N^t)`$). We believe that these diffractive effects should only induce lower-order corrections to the Weyl law (9).
We are now ready to quantize our open bakerโs map $`C`$ of (5): since the classical map sends points in $`R_1`$ to infinity and acts through $`B`$ on $`S=R_0R_2`$, the quantum propagator should kill states microsupported on $`R_1`$, and act as $`B_N`$ on states microsupported on $`S`$. Therefore, in the position basis we get the subunitary matrix
$$C_N=_N^1\left(\begin{array}{ccc}_{N/3}& & \\ & 0& \\ & & _{N/3}\end{array}\right).$$
(8)
A very similar open quantum baker was constructed in SaVa , as a quantization of Smaleโs horseshoe. In Figure 3 (left) we represent the moduli of the matrix elements $`(C_N)_{nm}`$. The largest elements are situated along the โtilted diagonalsโ $`n=3m`$, $`n=3(m2N/3)`$, which correspond to the projection on the $`q`$-axis of the graph of $`C`$. Away from these โdiagonalsโ, the amplitudes of the elements decrease relatively slowly (namely, like $`1/|n3m|`$). This slow decrease is due to the diffraction effects associated with the discontinuities of the map.
### 2.4 Resonances of the open bakerโs map
We numerically diagonalized the matrices $`C_N`$, for larger and larger Planckโs constants $`N`$. First of all, we notice that the subspace $`\mathrm{Span}\{Q_j,j=N/3,\mathrm{},2N/31\}`$, made of position states in the โholeโ, is in the kernel of $`C_N`$. Therefore, it is sufficient to diagonalize the matrix obtained by removing the corresponding lines and columns. Upon a slight modification of the quantization procedure Sa , one obtains for $`C_N`$ a matrix covariant w.r.to parity, allowing for a separation of the even and odd eigenstates, and therefore reducing the dimension of each part by $`2`$. This is the quantization we used for our numerics: we only plot the even-parity resonances (the distribution of the odd-dimensional ones is very similar).
In figure 4 we show the even-parity spectra of the matrix $`C_N`$ for $`N=3^5`$ and $`N=3^8`$. Although we could not detect exact null states for the reduced matrix, many among the $`N/3`$ eigenvalues had very small moduli: for large values of $`N`$, the spectrum of $`C_N`$ accumulates near the origin. This accumulation is an obvious consequence of the fractal Weyl law we want to test:
###### Conjecture 2
For any radius $`1>r>0`$ and $`N_0`$, $`3|N`$, let us denote
$$n(N,r)\stackrel{\mathrm{def}}{=}\mathrm{\#}\left\{\lambda \mathrm{Spec}(C_N)\left\{|\lambda |r\right\}\right\}.$$
In the semiclassical limit, this counting function behaves as
$$\frac{n(N,r)}{N^{\frac{\mathrm{log}2}{\mathrm{log}3}}}\stackrel{N\mathrm{}}{}c(r),$$
(9)
with a โshape functionโ $`0c(r)<\mathrm{}`$.
To test this conjecture, we proceed in two ways:
$``$ In a first step, we select some discrete values for $`r`$, and plot $`n(N,r)`$ for an arbitrary sequence of $`N`$, in a log-log plot (see Fig. 5).
We observe that the slope of the data nicely converges towards the theoretical one $`\frac{\mathrm{log}2}{\mathrm{log}3}`$ (thick line), all the more so along geometric subsequences $`N=3^kN_o`$, and for relatively large values of the radius ($`r=0.5`$). For the smaller value $`r=0.03`$, the annulus $`\{|z|r\}`$ still contains โtoo many resonancesโ and the asymptotic rรฉgime is not yet reached.
$``$ In a second step, confident that $`n(N,r)`$ scales like $`N^{\frac{\mathrm{log}2}{\mathrm{log}3}}`$, we try to extract the shape function $`c(r)`$. For an arbitrary sequence of values of $`N`$, we plot the function $`n(N,r)`$ (Fig. 6, left), and then rescale the vertical coordinate by a factor $`N^{\frac{\mathrm{log}2}{\mathrm{log}3}}`$ (right). The rescaled curves do roughly superpose on one another, supporting the conjecture. However, there remains relatively large fluctuations, even for large values of $`N`$. The curves corresponding to a geometric sequence $`N=3^kN_o`$, $`k=0,1,\mathrm{}`$ tend to be nicely superposed to one another, but slightly differ from one sequence to another. Similar plots were given in schomerus in the case of the kicked rotator; the shape function $`c(r)`$ is conjectured there to correspond to some ensemble of random subunitary matrices. Our data are too unprecise to perform such a check.
The fact that the spectra of the matrices $`C_N`$ โbehave nicelyโ along geometric sequences, while they fluctuate more strongly between successive values of $`N`$, is not totally unexpected (similar phenomena had been noticed for the quantizations $`B_N`$ of the closed baker BaVo ). In view of Fig. 6, our conjecture (9) may be too strict if we apply it to a general sequence of $`N`$. At least, it seems to be satisfied along geometric sequences $`\left\{\mathrm{\hspace{0.17em}3}^kN_o,k\right\}`$, with shape functions $`c_{N_o}(r)`$ slightly depending on the sequence.
## 3 A solvable toy model for the quantum baker
### 3.1 Description of the toy model
In an attempt to get some analytical grip on the resonances, we tried to simplify the quantum matrix $`C_N`$, keeping only its โbackboneโ along the tilted diagonals and removing the off-diagonal components. We obtained the โtoy-of-the-toy modelโ given by the following matrices (the moduli of the components are shown on right plot of Fig. 3):
$$\stackrel{~}{C}_{N=9}=\frac{1}{\sqrt{3}}\left(\begin{array}{ccccccccc}1\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill & 1\hfill & 0\hfill & 0\hfill \\ 1\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill & \omega ^2\hfill & 0\hfill & 0\hfill \\ 1\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill & \omega \hfill & 0\hfill & 0\hfill \\ 0\hfill & 1\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill & 1\hfill & 0\hfill \\ 0\hfill & 1\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill & \omega ^2\hfill & 0\hfill \\ 0\hfill & 1\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill & \omega \hfill & 0\hfill \\ 0\hfill & 0\hfill & 1\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill & 1\hfill \\ 0\hfill & 0\hfill & 1\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill & \omega ^2\hfill \\ 0\hfill & 0\hfill & 1\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill & \omega \hfill \end{array}\right),\omega =\mathrm{e}^{2\pi \mathrm{i}/3}.$$
(10)
From this example, it is pretty clear how one constructs $`\stackrel{~}{C}_N`$ for $`N`$ an arbitrary multiple of $`3`$. A similar quantization of the closed $`2`$-baker was introduced in schack .
Before describing the spectra of these matrices, we describe their propagation properties. Removing the โoff-diagonalโ elements, we have eliminated the effects of diffraction due to the discontinuities of $`C`$. However, this elimination is so abrupt that it modifies the semiclassical transport. Indeed, a coherent state situated at a point $`x`$ away from the discontinuities will not be transformed by $`\stackrel{~}{C}_N`$ into a single coherent state (as does $`C_N`$), but rather into a linear combination of $`3`$ coherent states, shifted vertically by $`1/3`$ from one another. Therefore, the matrices $`\stackrel{~}{C}_N`$ do not quantize the open baker $`C`$ of (5), but rather the following multivalued (โray-splittingโ) map:
$$\stackrel{~}{C}(q,p)=\{\begin{array}{cc}(3q,\frac{p}{3})(3q,\frac{p+1}{3})(3q,\frac{p+2}{3})\hfill & \mathrm{if}qR_0,\hfill \\ (3q2,\frac{p}{3})(3q2,\frac{p+1}{3})(3q2,\frac{p+2}{3})\hfill & \mathrm{if}qR_2.\hfill \end{array}$$
(11)
This modification of the classical dynamics is rather annoying. Still, the dynamics $`\stackrel{~}{C}`$ shares some common features with that of $`C`$: the forward trapped set for $`\stackrel{~}{C}`$ is the same as for $`C`$, that is the set $`\mathrm{\Gamma }_{}`$ described in Fig. 2. On the other hand, the backward trapped set is now the full torus $`๐^2`$.
### 3.2 Interpretation of $`\stackrel{~}{C}_N`$ as a Walsh-quantized baker
A possible way to avoid this modified classical dynamics is to interpret $`\stackrel{~}{C}_N`$ as a โWalsh-quantized mapโ (this interpretation makes sense when $`N=3^k`$, $`k`$). To introduce this Walsh formalism, let us first write the Hilbert space as a tensor product $`_N=(^3)^k`$, where we take the ternary decomposition of discrete positions $`\frac{j}{N}=0ฯต_0ฯต_1\mathrm{}ฯต_{k1}`$ into account. If we call $`\{e_0,e_1,e_2\}`$ the canonical basis of $`^3`$, each position state $`Q_j_N`$ can be represented as the tensor product state
$$Q_j=e_{ฯต_0}e_{ฯต_1}\mathrm{}e_{ฯต_{k1}}.$$
In the language of quantum computing, each tensor factor $`^3`$ is the Hilbert space of a โqutritโ associated with a certain scale schack .
The Walsh Fourier transform is a modification of the discrete Fourier transform (6), which first appeared in signal theory, and has been recently used as a toy model for harmonic analysis muscalu . Its major advantage is the possibility to construct states compactly supported in both position and โWalsh momentumโ. In our finite-dimensional framework, this Walsh transform is the matrix
$$(W_N)_{jj^{}}=3^{k/2}\mathrm{exp}\left(\frac{2i\pi }{3}\underset{\mathrm{}+\mathrm{}^{}=k1}{}ฯต_{\mathrm{}}(Q_j)ฯต_{\mathrm{}^{}}(Q_j^{})\right),j,j^{}=0,\mathrm{},N1,$$
and acts as follows on tensor product states:
$$W_N(v_0v_1\mathrm{}v_{k1})=_3v_{k1}\mathrm{}_3v_1_3v_0,v_{\mathrm{}}^3,\mathrm{}=0,\mathrm{},k1.$$
Now, in the case $`N=3^k`$, our toy model $`\stackrel{~}{C}_N`$ can be expressed as
$$\stackrel{~}{C}_N=W_N^1\left(\begin{array}{ccc}W_{N/3}& & \\ & 0& \\ & & W_{N/3}\end{array}\right).$$
One can show that โWalsh coherent statesโ are propagated through $`\stackrel{~}{C}_N`$ according to the map $`C`$. Hence, as opposed to what happens in โstandardโ quantum mechanics, $`\stackrel{~}{C}_N`$ Walsh-quantizes the open baker $`C`$.
### 3.3 Resonances of $`\stackrel{~}{C}_{N=3^k}`$
We now use the very peculiar properties of the matrices $`\stackrel{~}{C}_{3^k}`$ to analytically compute their spectra.
From the expressions in last section, one can see that the toy model $`\stackrel{~}{C}_N`$ acts very simply on tensor product states:
$$\stackrel{~}{C}_Nv_0v_1\mathrm{}v_{k1}=v_1\mathrm{}v_{k1}_3^1\pi _{02}v_0,$$
(12)
where $`\pi _{02}`$ projects $`^3`$ orthogonally onto $`\mathrm{Span}\{e_0,e_2\}`$. Like its classical counterpart, $`\stackrel{~}{C}_N`$ realizes a symbolic shift between the different scales. It also sends the first symbol $`ฯต_0`$ to the โend of the queueโ, after a projection and a Fourier transform. The projection $`\pi _{02}`$ kills the states $`Q_j`$ localized in the rectangle $`R_1`$. The vector $`_3^1e_{ฯต_0}`$ in the last qutrit induces a localization in the momentum direction, near the momentum $`p=0ฯต_0`$.
By iterating this expression $`k`$ times, we see that the operator $`(\stackrel{~}{C}_N)^k`$ acts independently on each tensor factor $`^3`$, through the matrix $`_3^1\pi _{02}`$. The latter has three eigenvalues:
* it kills the state $`e_1`$, implying that $`(\stackrel{~}{C}_N)^k`$ kills any state $`Q_j`$ for which at least one of the symbols $`ฯต_{\mathrm{}}(Q_j)`$ is equal to $`1`$. These $`3^k2^k`$ position states are localized โoutsideโ of the trapped set $`\mathrm{\Gamma }_{}`$, which explains why they are killed by the dynamics.
* its two remaining eigenvalues $`\lambda _\pm `$ have moduli $`|\lambda _+|0.8443`$, $`|\lambda _{}|0.6838`$. They build up the ($`2^k`$-dimensional) nontrivial spectrum of $`\stackrel{~}{C}_N`$, which has the form of a โlatticeโ (see Fig. 7):
###### Proposition 1
For $`N=3^k`$, the nonzero spectrum of $`\stackrel{~}{C}_N`$ is the set
$$\left\{\lambda _+\right\}\left\{\lambda _{}\right\}\{\mathrm{e}^{2\mathrm{i}\pi \frac{j}{k}}\lambda _+^{1p/k}\lambda _{}^{p/k}:\mathrm{\hspace{0.33em}1}pk1,0jk1\}.$$
Most of these eigenvalues are highly degenerate (they span a subspace of dimension $`2^k`$). When $`k\mathrm{}`$, the highest degeneracies occur when $`p/k2`$, which results in the following asymptotic distribution:
$$fC(^2),\underset{k\mathrm{}}{lim}\frac{1}{2^k}\underset{\lambda \mathrm{Spec}(\stackrel{~}{C}_{3^k})0}{}\mathrm{mult}(\lambda )f(\lambda )=_0^{2\pi }f(|\lambda _{}\lambda _+|^{1/2},\theta )\frac{\mathrm{d}\theta }{2\pi }.$$
The last formula shows that the spectrum of $`\stackrel{~}{C}_N`$ along the geometric sequence $`\{N=3^k,k\}`$ satisfies the fractal Weyl law (9), with a shape function in form of an abrupt step: $`c(r)=\mathrm{\Theta }(|\lambda _+\lambda _{}|^{1/2}r)`$. Although the above spectrum seems very nongeneric (lattice structure, singular shape function), it is the first example (to our knowledge) of a quantum open system proven to satisfy the fractal Weyl law.
Acknowledgments. We benefited from insightful discussions with Marcos Saraceno, Andrรฉ Voros, Uzy Smilansky, Christof Thiele and Terry Tao. Part of the work was done while I was visiting M. Zworski in UC Berkeley, supported by the grant DMS-0200732 of the National Science Foundation.
|
warning/0506/cond-mat0506795.html
|
ar5iv
|
text
|
# Dynamic critical behaviour in Ising spin glasses
## I Introduction
At a continuous transition only a few static exponents are enough to completely describe equilibrium critical behaviour. When dynamic measurements are considered the critical behaviour becomes much richer, with further independent critical quantities having non-trivial critical exponents.
It has been generally assumed that in a given dimension all Ising Spin Glasses (ISGs) lie in a single universality class. The Hamiltonian of the ISGs is given by
$$=\underset{i,j}{}J_{ij}S_iS_j$$
(1)
where $`S_i=\pm 1`$ are the usual Ising spins and the nearest-neighbour couplings $`J_{ij}`$ are random variables. We have studied numerically critical dynamic behaviour for ISGs having near neighbour interactions in dimensions $`d=3`$ and $`4`$ with Bimodal, Gaussian, or Laplacian distributions of the couplings. bernardi:97 We find that the dynamic exponents vary strongly and systematically from one distribution to another. One possible explanation for this could be that for ISGs the universality class depends on the form of the interaction distribution.
In Section II we introduce the dynamical quantities, whereas in Section III we treat the important point of the reliable determination of critical temperatures. We then present our results for the spin glasses in four (Section IV) and three (Section V) space dimensions. Finally, Section VI contains our conclusions.
## II Definitions
In the field of dynamic critical measurements there is not complete consensus as to a universal convention for indicating protocols or exponents, particularly for the case of spin glasses. We will first define the convention that we will use, following godreche:02 ; henkel:04 , and relate it to other standard conventions. We will then summarize dynamic scaling properties. This section leans heavily on the review calabrese:05 .
Throughout, we will implicitly consider only model A dynamics hohenberg:77 with single spin Glauber (or heat bath) updates in Ising spin glasses; the total time after quench is refered to as $`t`$. The basic dynamic protocol (there are potentially many others) consists in quenching the sample at time $`t=0`$ from $`T=\mathrm{}`$ to a temperature $`T`$, waiting (carrying out updates) for a time that will be called $`s`$; then at $`t=s`$ and still at fixed $`T`$, either the physical conditions are changed in some way (typically by switching on or switching off a small magnetic field $`h`$) or the instantaneous spin configuration at time $`s`$ is simply registered to provide a reference state, without physical conditions being changed. There is then an observation period of further updates during which physical parameters are measured as functions of $`t`$. For convenience the time difference $`ts`$ is also denoted $`\tau `$. In alternative conventions the waiting time $`s`$ is labeled $`t_\mathrm{w}`$, and $`\tau `$ is labeled $`t`$.
Among limiting conditions that can be profitably studied for large samples are the condition $`s=0`$ (i.e. measurements start immediately on quenching), or alternatively $`s\tau `$, a long waiting time condition after which the sample is in โquasi-equilibriumโ. True equilibrium can be achieved if $`s`$ is โlong enoughโ, a criterion that depends in a non-trivial way on the system, on the temperature $`T`$ and on the sample size $`L`$.
Many dynamic observables can be measured. Here we will concentrate on observations at criticality, $`T=T_\mathrm{c}`$, for the moment ignoring the question of how $`T_\mathrm{c}`$ is to be estimated.
A first fundamental definition is that of the dynamic critical exponent $`z_\mathrm{c}`$ . At $`T_\mathrm{c}`$, the equilibrium autocorrelation relaxation time (with standard single spin updates) increases with sample size $`L`$ as
$$\tau _{auto}(L)L^{z_\mathrm{c}}$$
(2)
where $`z_\mathrm{c}`$ is the dynamical critical exponent.
The two-time autocorrelation function is defined as
$$C(t,s)=\frac{1}{N}[\underset{j=1}{\overset{N}{}}S_j(s)S_j(t)]$$
(3)
where $`\mathrm{}`$ indicates the average over the thermal noise and $`[\mathrm{}]`$ the average over the disorder (in alternative conventions $`C`$ may also be written as $`q`$). The critical scaling relation for $`C`$ is
$$C(t,s)=s^bf_c(t/s).$$
(4)
In the quasi-equilibrium limit where $`s\tau `$ the critical scaling function $`f_c(t/s)`$ should follow the asymptotic behaviour
$$f_c(t/s)[(t/s)1)]^b.$$
(5)
Thus in this limit we have
$$C(t,s)\tau ^b$$
(6)
(in the alternative convention this is written as $`q(t)t^x`$ with $`xb`$). For spin glasses the dynamic scaling relation governing $`b`$ is ogielski:85
$$b=(d2+\eta )/2z_\mathrm{c}$$
(7)
where $`\eta `$ is the static critical exponent. In the opposite limit when $`(t/s)\mathrm{}`$
$$f_c(t/s)(t/s)^{\lambda _c/z_\mathrm{c}}$$
(8)
or for $`s=0`$
$$f_c(t)t^{\lambda _c/z_\mathrm{c}}.$$
(9)
$`\lambda _c/z_\mathrm{c}`$ is related to the โinitial slipโ exponent $`\theta _c`$ (an independent critical exponent janssen:89 ; godreche:02 ) through
$$\theta _c=d/z_\mathrm{c}\lambda _c/z_\mathrm{c}.$$
(10)
The two-time linear autoresponse function is
$$R(t,s)=[\delta S_i(t)/\delta h(s)]_{h=0}(t>s)$$
(11)
where $`h(s)`$ is a time dependent conjugate magnetic field. The scaling equation for $`R`$ is
$$TR(t,s)=s^{(1+a)}f_R(t/s)$$
(12)
with $`a=b`$ for critical systems. The critical scaling function $`f_R(t/s)`$ should follow the asymptotic behaviour
$$f_R(t/s)(t/s)^{\lambda _R/z_\mathrm{c}}$$
(13)
when $`t/s\mathrm{}`$. For short range inital correlations $`\lambda _R=\lambda _c`$.
In simulations where the field $`h`$ is applied at $`t=0`$ and switched off at $`t=s`$, the following integrated response is measured at times $`t>s`$:
$$\rho (t,s)=T\underset{0}{\overset{s}{}}๐uR(t,u).$$
(14)
This integrated response is directly related to the commonly studied thermoremanent magnetization:
$$M_{TRM}(t,s)=h\rho (t,s).$$
(15)
Of further interest is the fluctuation-dissipation ratio $`X`$ defined by
$$X(t,s)=TR(t,s)/(\delta C(t,s)/\delta s)=\widehat{X}(t/s).$$
(16)
In the quasi-equilibrium condition $`s\tau `$ the fluctuation-dissipation theorem holds and $`X=1`$.calabrese:05 When $`t/s\mathrm{}`$, $`X`$ takes a limiting value $`X_{\mathrm{}}`$. For $`ts`$ the ratio $`\rho (t,s)/C(t,s)`$ also converges to this limit value. If the amplitudes $`A_c`$ and $`A_R`$ are defined in the same limit $`ts`$ by
$$f_c(t/s)=A_c(t/s)^{\lambda _c/z_\mathrm{c}}$$
(17)
and
$$f_R(t/s)=A_R(t/s)^{\lambda _c/z_\mathrm{c}}$$
(18)
then chatelain:04
$$X_{\mathrm{}}=(A_R/A_c)(\lambda _c/z_\mathrm{c}b)^1.$$
(19)
Finally, the dynamic spin glass susceptibility is measured through
$$\chi _{ne}(t)=\frac{1}{N}[\underset{j=1}{\overset{N}{}}S_j^\alpha (t)S_j^\beta (t)^2]$$
(20)
where $`\alpha `$ and $`\beta `$ are two replicas of the same system relaxing independently. The infinite time limit to the dynamic SG susceptibility is the equilibrium SG susceptibility for each size $`L`$, which at criticality increases with $`L`$ as
$$\chi _{eq}(L)L^{(2\eta )}$$
(21)
where $`\eta `$ is the static critical exponent. The critical time dependence of the non-equilibrium spin-glass susceptibility for large samples after a quench to $`T_\mathrm{c}`$ and with no anneal ($`s=0`$) is huse:89
$$\chi _{\mathrm{ne}}(t)t^{(2\eta )/z_\mathrm{c}}=t^h^{}$$
(22)
where $`t`$ is the time after quench and $`z_\mathrm{c}`$ is again the dynamical critical exponent. For convenience we have introduced an exponent $`h^{}=(2\eta )/z_\mathrm{c}`$.
Even for a canonical continuous transition such as that of the $`2d`$ Ising ferromagnet, where the static critical exponents are all known analytically and are rational numbers, the dynamic critical exponents can only be established numerically and have non-trivial values.godreche:02 ; calabrese:05 For Ising ferromagnets however, field theory (FT) epsilon expansion estimates give reasonable agreement with numerical dynamic exponent estimates in dimensions $`3`$ and $`2`$.calabrese:05
It is now well established that for standard systems that are in the same universality class not only the static exponents $`\nu ,\eta `$ etc. but also the dynamic exponents $`z_\mathrm{c}`$, $`\theta _c`$ and $`X_{\mathrm{}}`$ are all universal. The numerical data discussed in the following show that in each dimension for ISGs expected to lie in the same universality class the dynamic exponents vary strongly with the form of the interaction distribution.
## III Ordering temperatures
In order to obtain accurate and reliable simulation values for the dynamic exponents an a priori requisite is to have reliable estimates for the ordering temperatures $`T_c`$.
High temperature series calculations give $`T_c`$ estimates which are not subject to finite size corrections and are thus intrinsically reliable, but whose accuracy is limited by the number of known terms in the series. Series estimates can be extremely precise at high dimension but unfortunately they become progressively more inaccurate as the system dimension drops.singh:87 ; klein:91 ; daboul:04
There are a number of different ways in which to obtain estimates of ordering temperatures in ISGs through simulations; for most of them one or more critical exponent estimates are also obtained simultaneously. Equilibrium simulations can provide accurate data but necessarily on samples which are of small or moderate size $`L`$, and the reliability and precision of the $`T_c`$ estimates are finally limited by the need to extrapolate to large $`L`$ to eliminate corrections to finite size scaling whose importance depends on the system being studied, on the parameter being measured, and on the maximum range of sample sizes that can be equilibrated with the computing facilities available. Corrections to scaling are subtle even for the canonical Ising ferromagnets hasenbusch:99 ; salas:00 ; caselle:02 where the leading corrections include both โirrelevant operatorโ and โanalyticโ contributions, while for ISGs basic guide-lines are lacking so that one must rely on empirical observations. In principle it should be possible to use the onset of deviations from strict critical behaviour to monitor $`T_c`$; for instance $`\mathrm{ln}(\chi (L,T))`$ against $`\mathrm{ln}(L)`$ curves bend downwards/upwards for $`T`$ greater/less than $`T_c`$. In the SG context this approach has been rarely used as the upbending below $`T_c`$ is weak, at least in 3d. The equilibrium finite size scaling simulation techniques, which are most often relied on to estimate $`T_c`$ in ISGs, are based on measurements of parameters that are dimensionless and take on an $`L`$-independent value at $`T_c`$ for large $`L`$. Well known examples are the Binder moment ratio kawashima:96 $`g(L,T)`$ and the second moment correlation length ratio palassini:99 $`\xi (L,T)/L`$. Plots of $`g(L,T)`$ or $`\xi (L,T)/L`$ as functions of $`T`$ for fixed $`L`$ have a unique โcrossing pointโ at a temperature which is equal to $`T_c`$ in the limit of large $`L`$. These methods require strict thermal equilibration at each size $`L`$; also the exact position of the large $`L`$ limit crossing point may be masked up to quite large $`L`$ by corrections to scaling (see for instance beach:05 for the case of the canonical Ising ferromagnet in three dimensions).
Simulations become heavier with increasing $`d`$ simply because for given $`L`$ the number of spins is $`L^d`$, but this effect is compensated by the fact that $`z_\mathrm{c}`$ tends to drop with increasing $`d`$. Furthermore crossing points become better defined at higher $`d`$, and it turns out that corrections to finite size scaling become weaker as $`d`$ increases. On balance it is in fact easier to estimate $`T_c`$ reliably by equilibrium simulations at $`d=4`$ (and above) than at $`d=3`$.
An alternative simulation technique which we will rely on below is to combine static and dynamic measurements to estimate $`T_c`$ by consistency.bernardi:96 This has the advantage of using two dynamic measurements which do not require equilibration and which have negligible corrections to scaling, together with equilibrium spin glass susceptibility measurements which do require equilibration but which are less sensitive to corrections to scaling than are Binder moment ratio or correlation length ratio measurements. A range of putative $`T_c`$ values $`T^{}`$ are chosen, and three measurements are made at each $`T^{}`$ :
\- the effective dynamic exponent $`b(T^{})`$ from large $`L`$ quasi-equilibrium measurements with $`s\tau `$ using Eq. (6),
\- the effective dynamic exponent $`h^{}(T^{})`$ from large $`L`$ measurements of the dynamic SG susceptibility Eq. (22), and
\- the effective static exponent $`\eta (T^{})`$ from equilibrium SG susceptibility finite size scaling measurements, Eq. (21).
As these three parameters are linked at $`T_c`$ through the two exponents $`\eta `$ and $`z_\mathrm{c}`$, Eq. (7), Eq. (21) and Eq. (22), there is a consistency condition which holds at and only at $`T^{}=T_c`$. There are different ways to implement this condition. We can first use the equilibrium and dynamic SG susceptibility results together to obtain a set of values of $`z(T^{})`$ at each $`T^{}`$. katzgraber:05 These values are to good precision independent of corrections to scaling, and the set of $`z(T^{})`$ extends from above to below $`T_c`$. Secondly, from the effective $`\eta (T^{})`$ and $`b(T^{})`$ one can derive a second effective $`z`$, $`z^+(T^{})=(d2+\eta (T^{}))/2b(T^{})`$. At $`T=T_c`$ the consistency condition is simply $`z(T^{})=z^+(T^{})`$. The measured values of the parameters $`\eta (T^{})`$ and $`z(T^{})`$ at this unique temperature correspond to the true critical exponents $`\eta `$ and $`z_\mathrm{c}`$. This method is rather insensitive to corrections to finite size scaling.
The three ISG distributions which will be considered here are the random Bimodal, Gaussian, and Laplacian near neighbour interaction distributions on (hyper)cubic lattices. The explicit normalized distributions are
$$P_B(J_{ij})=[\delta (J_{ij}J)+\delta (J_{ij}+J)]/2,$$
(23)
$$P_G(J_{ij})=\mathrm{exp}(J_{ij}^2/2J^2)/(J\sqrt{2\pi })$$
(24)
and
$$P_L(J_{ij})=\mathrm{exp}(\sqrt{2}J_{ij}/J)/(J\sqrt{2})$$
(25)
respectively. The distributions are symmetric about zero and are normalized in such a way that $`<J_{ij}^2>/J^2=1`$.
## IV Dimension 4
We will first consider explicitly dimension 4. $`T_c`$ values for a range of interaction distributions including the three cases that concern us here were obtained by simulations using the consistency method.bernardi:97 For the Bimodal and Gaussian distributions the values were fully consistent with and were as accurate or more accurate than other simulation estimates using alternative simulation techniques.hukushima:99 ; young ; ney-nifle:98 ; parisi:96 ; marinari:99 No other result appears to have been reported for the Laplacian distribution. Essentially negligible corrections to scaling can be seen in the data for any of the simulation techniques at this dimension; for instance, the Binder parameter crossing points are well defined and appear to be independent of $`L`$ to within high numerical precision. High temperature series estimates for the Bimodal case singh:87 ; klein:91 ; daboul:04 and for other interaction distributions daboul:04 are in excellent agreement with the simulation estimates; an overview of the data is given in campbell:05 . The overall agreement between the complementary approaches means that in $`d=4`$ the consistency simulation technique bernardi:96 is validated. Hence the $`T_c`$ values, together with the associated $`\eta `$ and $`z_\mathrm{c}`$ critical exponent values from the consistency method, can be taken as reliable.
Non-equilibrium measurements of the two-time autocorrelation function and the two-time linear autoresponse function were made at the temperatures corresponding to the $`T_c`$ values estimated from the consistency method. Large systems containing $`20^4`$ spins were simulated using the standard heat-bath algorithm. The systems were prepared initially in a completely disordered state and then quenched down to $`T_c`$ at time $`t=0`$. For the computation of the thermoremanent magnetization an external field with strength $`h=0.05`$ was applied between $`t=0`$ and $`t=s`$ with $`s`$ varying from 25 to 400.
Figure 1a and Figure 2 summarize our findings for the four-dimensional systems. The expected dynamical scaling behaviour (4) of the autocorrelation function is illustrated in Figure 1a for the case of a Laplacian distribution of the couplings. Plotting $`C(t,s)`$ as a function of $`t/s`$ for various values of the waiting time $`s`$, an excellent data collapse is achieved for the value $`b=0.140(3)`$. Deviations from this scaling behaviour are only obvious in the regime $`tss`$, i.e. outside of the dynamical scaling regime. A similar good data collapse footnote is obtained for the other distributions, see Table 1 and Ref. henkel:04 . It is worth noting that the values of $`b`$ obtained in these non-equilibrium simulations agree with those derived from the quantities $`\eta `$ and $`z_c`$ via Eq. (7). Equilibrium and non-equilibrium simulations therefore consistently yield for ISGs critical quantities depending on the form of the distribution of the couplings.
In Figure 2 we discuss truly non-equilibrium quantities which can not be expressed solely by equilibrium quantities. As shown in Figure 2a plotting $`\mathrm{ln}C(t,0)`$ versus $`\mathrm{ln}t`$ results in straight lines in the long time limit, in agreement with the expected power-law behaviour (9). The slopes of these lines yield the exponent $`\lambda _c/z_c`$. Again, this quantity, supposed to be universal, shows a clear dependence on the chosen distribution, see Table 1. Finally, Figure 2b displays the temporal evolution of the fluctuation-dissipation ratio (16) which in the limit $`t/s\mathrm{}`$ yields the limit value $`X_{\mathrm{}}`$, again supposed to be universal. It is obvious from this plot that the value of $`X_{\mathrm{}}`$ is different for the three distributions considered in this work.
Values for the various parameters corresponding to the present three distributions are shown in Table 1. The amplitude ratio $`A_R/A_c`$ has been derived from Eq. (19).
By inspection of the results in Table I it can be seen that the equilibrium critical exponent $`\eta `$ together with all the dynamic critical exponents vary strongly from one distribution to another. Apparent non-universality of critical exponents obtained from simulation data has in the past been ascribed to a consequence of errors in the estimation of critical temperatures or to a lack of care in allowing for corrections to finite size scaling. In the present case, the values of the ordering temperatures bernardi:97 have been validated by high temperature series calculations daboul:04 and internal evidence, inherent to the consistency method described in Section III, shows that corrections to finite size scaling are negligible in this dimension. In addition, the dynamic parameters are obtained from simulations on large samples and can be considered virtually free of finite-size corrections to scaling. The values of each of the dynamical critical parameters are insensitive to the precise value of the ordering temperature so even if there were small errors in the assumed values of the ordering temperatures the effects on the dynamic critical parameter estimates would be negligible.
At this point some remarks on the reliability of the extraction of critical exponents from out-of-equilibrium simulations are in order. The whole approach is based on the assumption that, in the time range in which the exponents are determined, the dynamical correlation length $`\xi (t)`$ increases as a simple power-law
$$\xi (t)t^{1/z_c}$$
(26)
where $`z_c`$ is the dynamical critical exponent. For critical ferromagnets this growth law and the resulting dynamical scaling set in already after a few time steps. However, as spin glasses are characterized by a large value of $`z_c`$ one may wonder whether this simple growth low prevails for the times we have accessed in our simulations or whether a more general growth law of the form
$$\xi (t)=at^{1/z_c}+bt^{1/z^{}}$$
(27)
with a sizeable finite-time correction is observed.
In fact, the growth of the dynamical correlation length at the critical point of different three- and four-dimensional spin glasses has been intensively investigated in the recent past. We mention here the Ising spin glass with a Bimodalyoshino:02 ; berthier:02 or a Gaussianberthier:02 ; katzgraber:05 distribution of the couplings, the gauge glass with a Gaussian distributionkatzgraber:05 , the $`XY`$ spin glass with a Bimodal distributionyamamoto:04 or the Heisenberg spin glass with a Gaussian distributionberthier:04 . All these studies reveal that for times $`t20`$ the increase of the dynamical correlation length in various spin glasses (including some of the cases we consider in this work) is given by the simple power-law (26). No finite-time corrections of the form (27) have been observed. Additional support for the growth law (26) comes from the perfect dynamical scaling behaviour of two-time quantities (as for example the autocorrelation function shown in Figure 1), as a sizeable finite-time correction would completely spoil the observed data collapse.
## V Dimension 3
In dimension $`d=3`$ the overall situation is rather less satisfactory; high temperature series values singh:87 ; klein:91 become imprecise for the Bimodal case and none have been reported for the other distributions. The Binder ratio method becomes delicate because the $`g_L(T)`$ curves only fan out weakly at low temperatures making the limiting crossing point difficult to identify and very sensitive to corrections to scaling.kawashima:96 The correlation length ratio appears to suffer from strong corrections to scaling, especially at low $`L`$.katzgraber:05b For the Gaussian distribution there is a general consensus as to the value of $`T_c`$ from different simulation estimates.marinari:98 ; mari:01 ; katzgraber:05b For the Bimodal distribution, published $`T_c`$ estimates are much more scattered ogielski:85 ; kawashima:96 ; palassini:99 ; ballesteros:00 ; mari:02 which we ascribe to difficulties related to corrections to finite size scaling. For the Laplacian distribution we are not aware of other published estimates.
We will rely on the value from the consistency method because, for the reasons outlined above, this technique is much less sensitive to problems of corrections and because it has given excellent agreement with the high temperature series values in $`d=4`$. Data from the consistency method are presented in Figure 3. With sets of trial temperatures $`T^{}`$ we plot for each system $`z(T^{})`$ and $`z^+(T^{})`$ against $`b(T^{})`$. $`z(T^{})`$ is derived from a comparison of the equilibrium and dynamic SG susceptibility results katzgraber:05 at each $`T^{}`$, and $`z^+(T^{})=(d2+\eta (T^{}))/2b(T^{})`$. $`b(T^{})`$ is measured from the autocorrelation function decay in quasi-equilibrium as defined above, and the effective $`\eta (T^{})`$ is obtained from equilibrium finite size SG susceptibility measurements. At $`T_c`$ consistency of the various exponents dictates that $`z(T^{})z^+(T^{})`$. The values of $`T_c`$ together with the exponents $`z_\mathrm{c}`$ and $`\eta `$ obtained are given in Table II (the values are more precise than those given using the same method in bernardi:96 because of improved equilibrium susceptibility datapalassini:05 ; katzgraber:05b ). It is interesting to note that Migdal-Kadanoff estimates of $`T_c`$ for the different distributions with $`d=3`$ and the MK parameter $`b=2`$ prakash:97 ; nogueira:99 are strikingly similar to the values given here.
For the non-equilibrium simulations in three dimensions we considered systems with $`50^3`$ spins and waiting times $`s1600`$. The expected scaling behaviour of the two-time quantities is again observed, see Figure 1b for the autocorrelation of the Laplacian distribution. As shown in Figure 4 (see Table 2) one observes that the values of $`\lambda _c/z_c`$ and $`X_{\mathrm{}}`$ also depend in three dimensions on the form of the distribution function of the couplings. The insert in Figure 4a displays the effective exponent
$$(\lambda _c/z_c)_{eff}=(\mathrm{ln}(C(20t,0)\mathrm{ln}(C(t,0))/(\mathrm{ln}(20t)\mathrm{ln}(t))$$
(28)
as a function of $`1/t`$. In all cases this effective exponent rapidly reaches a constant, distribution dependent, value.
Once again the estimates of the various static and dynamic critical exponents vary considerably from distribution to distribution. The sense of the variations is systematic and is the same as in dimension $`4`$ : with increasing kurtosis of the distribution, $`T_c`$ drops, $`z_\mathrm{c}`$ increases, $`\eta `$ becomes more negative, and the dynamic exponents either expressed as $`\lambda _c/z_\mathrm{c}`$ and $`X_{\mathrm{}}`$ or as $`\theta _c`$ and $`A_R/A_c`$ all drop.
## VI Discussion
For standard continuous transitions the renormalization group theory provides a comprehensive explanation of critical behaviour and in particular of the strict identity of exponents of all systems within each universality class, the class being defined by a restricted list of parameters which includes the physical dimension $`d`$ and the number of order parameter components $`n`$. The universality covers not only equilibrium exponents but extends to the whole family of dynamic exponents. This universality reflects the fundamental principle that within each class, the details of the physics at the local level do not affect the large scale behaviour which determines the critical exponents.
It has been widely assumed that in a given dimension all ISGs fall in the same universality class. However it should be noted that the critical behaviour of spin glasses is qualitatively very different from that of standard systems such as ferromagnets. The upper critical dimension is 6 rather than 4, and below the upper critical dimension the specific heat exponents are strongly negative so there is no specific heat peak or cusp. Field theory (which provides the well known $`ฯต`$ expansion development at standard continuous transitions) has proved intractable in the ISG context below the upper critical dimension $`d=6`$. dedominicis:98 Already at $`d=5`$ and $`d=4`$, numerical values of the equilibrium exponents obtained from summing the known leading terms to order three in the ISG $`ฯต`$ expansion yeo:05 are very different from estimates using the high temperature series method klein:91 ; daboul:04 or simulations. For the Ising ferromagnet at $`d=3`$ and so $`ฯต=1`$, the FT development in $`ฯต`$ to third order is accurate to better than $`0.001`$.zinn-justin:89 This is in total contrast to the situation for the Bimodal ISG where at $`d=5`$ (so again at $`ฯต=1`$) the FT sum to third order in $`ฯต`$ gives $`\eta (d=5)=1.6897`$, strikingly different from the high temperature series value klein:91 $`\eta (d=5)=0.38(7)`$ and the simulation estimate $`\eta (d=5)=0.39(2)`$.bernardi:97 This implies that a sum including many further terms (oscillating in sign) would be needed to finally obtain stable and accurate FT predictions. In practice, establishing such a sum seems entirely ruled out, but the question remains open as to whether the necessary quasi-cancellations among the unknown higher order FT terms could depend on parameters such as the lattice structure or the form of the interactions.
In our simulations of the Ising spin glasses we have found that, in most cases, the differences between the exponent estimates for the different systems are much larger than the statistical error bars. This is especially obvious for the amplitude ratio $`A_R/A_c`$ which is supposed to yield clearly distinct values for different universality classes, similar to what is observed for static ampltude ratios at equilibrium. Extrapolations in each dimension suggest that if the interaction distribution is modified and $`T_c`$ decreases, all the exponents studied vary systematically in such a way that $`z_\mathrm{c}`$ increases strongly, $`\eta `$ tends towards a value near $`2d`$ (which is the strict limiting value for $`\eta `$ in each dimension when $`T_c=0`$) while $`X_{\mathrm{}}`$ and $`\theta _c`$ tend to near zero. The data as they stand are thus compatible with exponents each varying continuously towards a $`T_c=0`$ limit as $`T_c`$ is driven lower by a widening of the interaction distribution (increasing kurtosis).
Our non-equilibrium simulations can in principle be subjected to only three different sources of systematic errors: (1) the values of the critical temperatures are erroneous, (2) the sizes of the samples are too small, or (3) the time range of our runs is insufficient. Let us address these three different points.
The values of the critical temperatures we use have been estimated with a technique combining equilibrium and non-equilibrium measurementsbernardi:97 and are in excellent agreement with independent simulation estimates. In addition, $`T_c`$ in the four-dimensional systems have been confirmed recentlydaboul:04 by high temperature series estimates. This agreement validates the technique used for the evaluation of $`T_c`$ and also, as the method relies on a consistency argument, indirectly the values of $`z_c`$ and $`\eta `$ obtained in the same simulations.
The data we have discussed in this paper concern principally dynamic exponents in ISGs. These measurements have the advantage of not requiring strict equilibration, which in turn permits studies on samples so large that they cannot be conveniently equilibrated at or near criticality. As the sample sizes $`L`$ are much larger than the maximum of the correlation length $`\xi (t)`$ attained during the simulations, the measurements are always taken in the infinite sample size limit and are not hampered by finite-size effects. We have thus been able to minimize one of the major sources of systematic errors in numerical simulations, namely corrections to finite size scaling. This is very similar to what is observed when studying critical ferromagnets, as for example the three-dimensional Ising model jaster:99 , where no notable finite-size corrections to scaling are encountered in non-equilibrium simulations, in contrast to equilibrium simulations where these corrections to scaling can be very strong.
Whereas finite-size effects are well controlled in non-equilibrium simulations of large systems, this is not immediately obvious for finite-time corrections. Indeed, for spin glasses the critical dynamical exponent $`z_c`$ takes on very large values, whch might raise some doubts whether the simple power-law increase (26) of the dynamical correlation length is valid in the time range we accessed. However, various investigations of spin glasses in different dimensions and with different distributions of the couplings in the recent past have found that in general the increase of $`\xi (t)`$ is completely described by (26) for times $`t20`$. We can therefore be confident that the dynamical scaling approach, underlying our estimates of the critical quantities, is also valid for spin glasses and that the values of the equilibrium and non-equilibrium critical quantities we obtain are reliable. Furthermore, as shown in Figure 4, the run times of our simulations are clearly sufficient for the dynamic parameters to take up their limiting values.
Having now excluded the most probable sources of systematic errors we interpret our numerical data as strong evidence that in spin glasses critical quantities do depend on the exact form of the distribution of the couplings. Of course, as we only provide numerical evidence, we can not completely exclude that corrections coming from other sources could have some impact on the values of the critical exponents.
###### Acknowledgements.
We would like to thank H.G. Katzgraber for useful discussions and for making his data available, and M. Palassini for kindly providing equilibrium susceptibility data. MP acknowledges the support by the Deutsche Forschungsgemeinschaft through grant no. PL 323/2. The non-equilibrium simulations were done on the IBM supercomputer Jump at the NIC Jรผlich (project Her10). Complementary simulations were done by J. Kirmair.
|
warning/0506/hep-ph0506176.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
The quest for Higgs bosons is of utmost importance at high energy colliders ,, . In the Standard Model (SM), one isospin $`I=1/2`$, hypercharge $`Y=1`$ complex scalar doublet breaks the Electroweak Symmetry (EW) and provides mass for the fermions, $`W^\pm `$ and $`Z`$. One neutral scalar, $`\varphi ^0`$, remains as a physical degree of freedom โ โthe SM Higgs bosonโ. Such a framework predicts $`\rho (=M_W^2/M_Z^2\mathrm{cos}^2\theta _W)=1`$ at tree-level, a result which is in impressive agreement with the experimental measurement of $`\rho 1`$ . More generally, any Higgs sector composed solely of $`I=1/2,Y=1`$ doublets assures $`\rho =1`$ at tree-level, with calculable 1-loop corrections .
Predicting $`\rho =1`$ at tree-level is certainly an attractive feature of $`I=1/2,Y=1`$ doublet representations, although models with isospin triplets ($`I=1`$) can also be considered . Such models have various virtues and deficiencies. If the neutral member of the triplet acquires a vacuum expectation value (VEV) then $`\rho =1`$ at tree-level is no longer guaranteed, and the triplet VEV must be very small in order to comply with the measured value $`\rho 1`$. However, unlike doublets, $`Y=2`$ triplets can give rise to neutrino masses and mixings whose magnitude is proportional to the triplet vacuum expectation value multiplied by an arbitrary Yukawa coupling ($`h_{ij}`$) without invoking a right handed neutrino , .
A clear phenomenological signature of $`Y=2`$ triplets would be the observation of a doubly charged Higgs boson $`H^{\pm \pm }`$. Such $`H^{\pm \pm }`$ have been searched for at the $`e^+e^{}`$ collider LEP, resulting in mass limits of the order $`m_{H^{\pm \pm }}>100`$ GeV ,, ,. Their existence can also affect a wide variety of processes, such as Bhabha scattering, the anomalous magnetic moment of the muon $`(g2)_\mu `$, and lepton flavour violating $`\mu ^\pm `$ and $`\tau ^\pm `$ decays ,,, ,,. The Fermilab Tevatron recently performed the first search for $`H^{\pm \pm }`$ at hadron colliders. The production process $`p\overline{p}\gamma ,ZH^{++}H^{}`$ was assumed, with subsequent decay $`H^{\pm \pm }l^\pm l^\pm `$. D0 searched for $`H^{\pm \pm }\mu ^\pm \mu ^\pm `$ while CDF searched for 3 final states $`H^{\pm \pm }\mu ^\pm \mu ^\pm ,\mu ^\pm e^\pm `$ and $`e^\pm e^\pm `$. Mass limits of the order $`m_{H^{\pm \pm }}>130`$ GeV were obtained with an integrated luminosity of 240 pb<sup>-1</sup>, assuming BR($`H^{\pm \pm }l_i^\pm l_j^\pm )=100\%`$ in a given channel. These are the strongest direct mass limits on any type of Higgs boson, which shows the strong search capability of hadron colliders in the channel $`H^{\pm \pm }l^\pm l^\pm `$.
Given this strong search potential, in this paper we consider the phenomenological effect of relaxing these simplifying assumptions for the dominant production mechanism and decay modes of $`H^{\pm \pm }`$. Although work along these lines has appeared previously , ,,, , we develop and expand the preceding analyses. For example, if $`h_{ij}`$ are solely responsible for the currently favoured form of the neutrino mass matrix then BR($`H^{\pm \pm }l_i^\pm l_j^\pm )<100\%`$ in a given channel . In this paper we study in detail the alternative production mechanism $`q^{}\overline{q}W^{}H^{\pm \pm }H^{}`$ , which can be as large as $`q\overline{q}\gamma ,ZH^{++}H^{}`$. Since the current search strategy at the Tevatron is in fact sensitive to single production of $`H^{\pm \pm }`$, we introduce the inclusive single production cross-section $`(\sigma _{H^{\pm \pm }}`$) as the sum of the single and pair production cross-sections. We point out that the contribution of $`q^{}\overline{q}W^{}H^{\pm \pm }H^{}`$ to $`\sigma _{H^{\pm \pm }}`$ strengthens the Tevatron mass limit on $`H^{\pm \pm }`$, which in general has a dependence on $`m_{H^\pm }`$. Moreover, we quantify the impact of the potentially important decay mode $`H^{\pm \pm }H^\pm W^{}`$ in the light of recent neutrino data. Although such a decay can weaken the $`H^{\pm \pm }`$ search capability in the leptonic channel, observation of $`H^{\pm \pm }H^\pm W^{}`$ together with one or more leptonic channels might permit an order of magnitude estimate of $`h_{ij}`$ , .
Our work is organized as follows. In Section 2 we introduce the Higgs Triplet Model. In Section 3 we study the production mechanism $`q^{}\overline{q}H^{\pm \pm }H^{}`$ and its phenomenological effect on the $`H^{\pm \pm }`$ search at the Tevatron and LHC. In Section 4 we quantify the impact of the decay $`H^{\pm \pm }H^\pm W^{}`$, while Section 5 considers the search potential of the Tevatron in the generalized scenario. Finally, in Section 6 we present our conclusions.
## 2 The Higgs Triplet Model
Higgs $`I=1`$ triplet representations arise in several well motivated models of physics beyond the SM ,. For example, Left-Right (L-R) symmetric models built on the gauge group $`SU(2)_R\times SU(2)_L\times U(1)`$ contain both left- and right-handed $`I=1,Y=2`$ triplet representations. Such models also require extra gauge bosons and can provide naturally light neutrino masses via the seesaw mechanism. Little Higgs models also require $`I=1,Y=2`$ triplet representations, as well as new gauge bosons and fermions. However, Higgs triplets can be considered as a minimal addition to the SM \- for a review see . We will focus on a particularly simple model , which merely adds a $`I=1,Y=2`$ complex (left-handed) Higgs triplet to the SM Lagrangian, hereafter referred to as the โHiggs Triplet Modelโ or โHTMโ. Such a model can provide a Majorana mass for the observed neutrinos without the need for a right handed neutrino via the gauge invariant Yukawa interaction: <sup>1 </sup><sup>1 </sup>1 Note that the analogous term for a $`Y`$=0 triplet is forbidden by gauge invariance
$$=h_{ij}\psi _{iL}^TCi\tau _2\mathrm{\Delta }\psi _{jL}+h.c$$
(1)
Here $`h_{ij}(i,j=1,2,3)`$ is an arbitrary coupling, $`C`$ is the Dirac charge conjugation operator, $`\psi _{iL}=(\nu _i,l_i)_L^T`$ is a left-handed lepton doublet, and $`\mathrm{\Delta }`$ is a $`2\times 2`$ representation of the $`Y=2`$ complex (left-handed) triplet fields:
$$\mathrm{\Delta }=\left(\begin{array}{cc}\mathrm{\Delta }^+/\sqrt{2}& \mathrm{\Delta }^{++}\\ \mathrm{\Delta }^0& \mathrm{\Delta }^+/\sqrt{2}\end{array}\right)$$
(2)
The Higgs potential is as follows, with $`\mathrm{\Phi }=(\varphi ^+,\varphi ^0)^T`$:
$`V`$ $`=`$ $`m^2(\mathrm{\Phi }^{}\mathrm{\Phi })+\lambda _1(\mathrm{\Phi }^{}\mathrm{\Phi })^2+M^2\mathrm{Tr}(\mathrm{\Delta }^{}\mathrm{\Delta })+\lambda _2[\mathrm{Tr}(\mathrm{\Delta }^{}\mathrm{\Delta })]^2+\lambda _3\mathrm{Det}(\mathrm{\Delta }^{}\mathrm{\Delta })`$ (3)
$`+\lambda _4(\mathrm{\Phi }^{}\mathrm{\Phi })\mathrm{Tr}(\mathrm{\Delta }^{}\mathrm{\Delta })+\lambda _5(\mathrm{\Phi }^{}\tau _i\mathrm{\Phi })\mathrm{Tr}(\mathrm{\Delta }^{}\tau _i\mathrm{\Delta })+({\displaystyle \frac{1}{\sqrt{2}}}\mu (\mathrm{\Phi }^Ti\tau _2\mathrm{\Delta }^{}\mathrm{\Phi })+h.c)`$
The term $`\mu \mathrm{\Phi }\mathrm{\Delta }\mathrm{\Phi }`$, where $`\mu `$ is a dimensionful trilinear coupling, gives rise to a VEV $`v_\mathrm{\Delta }`$ for the neutral member of the triplet $`\mathrm{\Delta }^0`$:
$$v_\mathrm{\Delta }\mu v^2/2M^2$$
(4)
Here $`M`$ is the common triplet mass ($`M^2\mathrm{\Delta }^{}\mathrm{\Delta }`$). Since we are interested in the case of light triplets we take $`Mv`$, and so $`v_\mathrm{\Delta }\mu `$. A non-zero $`v_\mathrm{\Delta }`$ gives rise to the following mass matrix for neutrinos:
$$m_{ij}=2h_{ij}\mathrm{\Delta }^0=\sqrt{2}h_{ij}v_\mathrm{\Delta }$$
(5)
Note that the HTM is free from a massless Goldstone boson (Majoron) arising from the violation of the lepton number ($`L`$) global symmetry, because the Higgs potential contains the term $`\mu \mathrm{\Phi }\mathrm{\Delta }\mathrm{\Phi }`$ term which explicitly violates lepton number when $`\mathrm{\Delta }`$ is assigned $`L=2`$. Cosmological data provides a constraint on the neutrino masses $`m_i`$, $`\mathrm{\Sigma }m_i\mathrm{\Gamma }<\mathrm{\hspace{0.17em}0.75}`$ eV . Lepton flavour violating (LFV) processes involving $`\mu `$ and $`\tau `$ provide the strongest upper limits on $`h_{ij}`$ and hence $`v_\mathrm{\Delta }`$ cannot be arbitrarily small if the HTM is to accommodate the currently favoured form of the neutrino mass matrix. A rough lower bound $`v_\mathrm{\Delta }\mathrm{\Gamma }>\mathrm{\hspace{0.17em}10}`$ eV can be derived. An upper limit on $`v_\mathrm{\Delta }`$ can be obtained from considering its effect on $`\rho `$. In the HTM $`\rho `$ is given by (where $`x=v_\mathrm{\Delta }/v`$):
$$\rho 1+\delta \rho =\frac{1+2x^2}{1+4x^2}$$
(6)
From the measurement of $`\rho 1`$ a purely tree-level analysis gives the bound $`v_\mathrm{\Delta }/v\mathrm{\Gamma }<\mathrm{\hspace{0.17em}0.03}`$. We will comment on the 1-loop expression for $`\delta \rho `$ below ,,. In this paper we will assume
$$10\mathrm{eV}\mathrm{\Gamma }<v_\mathrm{\Delta }\mathrm{\Gamma }<\mathrm{\hspace{0.17em}10000}\mathrm{eV}$$
(7)
Hence the tree-level value of $`\rho `$ is essentially equal to 1, thus easily satisfying the experimental constraint on $`\delta \rho `$. Such small values of $`v_\mathrm{\Delta }`$ can be explained by a 2 loop mechanism or in the context of extra dimensions ,. Moreover, such values of $`v_\mathrm{\Delta }`$ would permit some $`h_{ij}`$ to be sufficiently large to enhance various LFV $`\mu `$ and $`\tau `$ decays to the sensitivity of current and forthcoming experiments , ,, and are also consistent with the requirement that any primordially generated baryon asymmetry is not erased by the lepton number violating triplet interactions .
The HTM has 7 Higgs bosons $`(H^{++},H^{},H^+,H^{},H^0,A^0,h^0)`$. While $`H^{\pm \pm }`$ is purely triplet ($`=\mathrm{\Delta }^{\pm \pm }`$), the remaining eigenstates would in general be mixtures of the doublet and triplet fields. Such mixing is proportional to the triplet VEV, and hence small even if $`v_\mathrm{\Delta }`$ assumes its largest value of a few GeV. Therefore the first six eigenstates are essentially composed of triplet fields, while the $`I=1/2`$ doublet gives rise to a SM like $`h^0`$ and the Goldstone bosons $`G^\pm ,G^0`$. The most striking signature of the HTM would be the observation of $`H^{\pm \pm }`$. <sup>2 </sup><sup>2 </sup>2 The dominantly triplet eigenstates $`H^\pm `$, $`H^0`$ and $`A^0`$ can have a different phenomenology to the analogous Higgs bosons in doublet ($`I=1/2,Y=1)`$ representations. In the HTM there exists the following relationships among the masses of the physical Higgs bosons:
$`m_{H^{\pm \pm }}^2M^2+2{\displaystyle \frac{(\lambda _4\lambda _5)}{g^2}}M_W^2`$ (8)
$`m_{H^\pm }^2m_{H^{\pm \pm }}^2+2{\displaystyle \frac{\lambda _5}{g^2}}M_W^2`$
$`m_{H^0,A^0}^2m_{H^\pm }^2+2{\displaystyle \frac{\lambda _5}{g^2}}M_W^2`$
Here $`M`$ is the triplet mass term, while $`\lambda _4,\lambda _5`$ are dimensionless quartic couplings. For $`\lambda _5>0`$ ($`\lambda _5<0`$) one has the following hierarchy $`m_{H^{\pm \pm }}<m_{H^\pm }<m_{H^0,A^0}`$ ($`m_{H^{\pm \pm }}>m_{H^\pm }>m_{H^0,A^0}`$). Clearly $`M`$ sets the scale for the mass of the triplet fields, while the mass splitting among the eigenstates is determined by the quartic couplings and can be $`๐ช(M_W)`$. We will focus on Higgs boson masses of interest for the Tevatron and LHC, and hence we assume $`M\mathrm{\Gamma }>\mathrm{\hspace{0.17em}1}`$ TeV.
At the 1-loop level the Higgs sector contribution to $`\delta \rho `$ is a function of $`v_\mathrm{\Delta }`$ and the Higgs boson masses. Although a quantitative analysis in the context of the HTM is still lacking, explicit formulae for the contributions of $`Y=2`$ triplets to the self-energies of the $`W`$ and $`Z`$ in the context of L-R symmetric models and Little Higgs Models can be found in , . In particular, such contributions are sensitive to the mass splittings of the Higgs bosons. In the HTM the triplet Higgs boson mass splitting is determined by the quartic coupling $`\lambda _5`$, with $`\lambda _5=0`$ giving rise to degenerate triplet scalars of mass $`M`$. We will present results for both the degenerate case and for mild splittings of up to 20 GeV in our discussion of $`H^{\pm \pm }`$ phenomenology at the Tevatron.
We now briefly discuss present mass bounds on the Higgs bosons of the HTM, which differ in some cases from the commonly quoted mass bounds in the 2HDM. If $`H^0`$ and $`A^0`$ were the lightest, they could have been produced at LEP via the mechanism $`e^+e^{}A^0H^0`$ (note that $`e^+e^{}ZH^0`$ is proportional to $`v_\mathrm{\Delta }`$ and hence negligible). However, since $`A^0`$ and $`H^0`$ would both decay invisibly to $`\nu \overline{\nu }`$, Ref. suggested using LEP data on $`\gamma \nu \overline{\nu }`$ (where $`\gamma `$ arises from bremsstrahlung from $`e^+`$ or $`e^{}`$) and derived the mass limit $`m_{H^0,A^0}\mathrm{\Gamma }>\mathrm{\hspace{0.17em}55}`$ GeV. Concerning $`H^\pm `$, LEP searched for $`H^\pm cs`$ or $`\tau \nu _\tau `$ which are expected to be the dominant decays in doublet models, and obtained mass limits around $`m_{H^\pm }\mathrm{\Gamma }>\mathrm{\hspace{0.17em}80}`$ GeV. For the triplet $`H^\pm `$ the decays $`H^\pm e^\pm \nu ,\mu ^\pm \nu `$ may have large BRs. However, in this scenario one could presumably use data from slepton searches $`e^+e^{}\stackrel{~}{l}^+\stackrel{~}{l}^{}l^+l^{}\chi ^0\overline{\chi }^0`$ to derive similar mass limits ($`\mathrm{\Gamma }>\mathrm{\hspace{0.17em}80}`$ GeV) . A recent quantitative analysis of the above decays in the context of a Little Higgs Model can be found in .
Concerning $`H^{\pm \pm }`$, LEP searched for both left-handed $`H_L^{\pm \pm }`$ and right-handed $`H_R^{\pm \pm }`$ (which we will not consider in this paper) via several mechanisms:
* Pair production via $`e^+e^{}\gamma ^{},Z^{}H^{++}H^{}`$ followed by decay to $`l^+l^+l^{}l^{}`$ ($`l^\pm =e^\pm ,\mu ^\pm ,\tau ^\pm `$); the cross-section is determined by gauge couplings and leads to mass limits of $`m_{H^{\pm \pm }}>100`$ GeV , ,.
* Single production of $`H^{\pm \pm }`$ via $`e^+e^{}H^{\pm \pm }e^{}e^{}`$; the rate is determined by the coupling $`h_{11}`$ and leads to excluded regions in the plane ($`h_{11},m_{H^{\pm \pm }}`$), with sensitivity up to $`m_{H^{\pm \pm }}\mathrm{\Gamma }<\mathrm{\hspace{0.17em}180}`$ GeV. Limits of $`10^210^1`$ were set on $`h_{11}`$ .
* The effect of $`H^{\pm \pm }`$ on Bhabha scattering $`e^+e^{}e^+e^{}`$; as in (ii) above this leads to excluded regions in the plane ($`h_{11},m_{H^{\pm \pm }}`$) , with sensitivity up to $`m_{H^{\pm \pm }}\mathrm{\Gamma }<\mathrm{\hspace{0.17em}2}`$ TeV. Limits of $`10^210^1`$ were set on $`h_{11}`$.
The direct searches for $`H^{\pm \pm }`$ will continue at the hadron colliders, Tevatron and LHC.
## 3 Production of $`H^{\pm \pm }`$ at the Tevatron
A distinct signature of $`H^{\pm \pm }`$ would be a pair of same sign charged leptons ($`e^\pm `$ or $`\mu ^\pm `$) with high invariant mass. At hadron colliders such a signal has a relatively high detection efficiency and enjoys essentially negligible background from Standard Model processes. Earlier theoretical studies of the search potential for $`H^{\pm \pm }`$ at such colliders can be found in ,, with a recent analysis at the LHC in . The decays of $`H^{\pm \pm }`$ to states involving $`\tau ^\pm `$ are more problematic at hadron colliders, although simulations in these channels , promise sensitivity to values of $`m_{H^{\pm \pm }}`$ beyond the LEP limits. The decays $`H^{\pm \pm }W^\pm W^\pm `$ are proportional to $`v_\mathrm{\Delta }`$, and can be neglected in the case of very small $`v_\mathrm{\Delta }`$ of interest to us.
In 2003 the Tevatron performed the first search for $`H^{\pm \pm }`$ at a hadron collider. D0 have searched for $`H^{\pm \pm }\mu ^+\mu ^{}`$ while CDF searched for 3 final states: $`H^{\pm \pm }e^\pm e^\pm ,e^\pm \mu ^\pm ,\mu ^\pm \mu ^\pm `$. The assumed production mechanism for $`H^{\pm \pm }`$ is $`q\overline{q}\gamma ^{},Z^{}H^{++}H^{}`$. <sup>3 </sup><sup>3 </sup>3 The model-dependent contribution from any $`Z^{}`$ (which can enhance the cross-section ,) is currently not considered. This cross-section depends on only one unknown parameter, $`m_{H^{\pm \pm }}`$, and importantly is not suppressed by any small factor such as a Yukawa coupling $`h_{ij}`$ or a triplet VEV. The search assumes that $`H^{\pm \pm }`$ is sufficiently long-lived to decay in the detector, which corresponds to $`h_{ll}>10^5`$. A search for a long lived $`H^{\pm \pm }`$ decaying outside the detector has been performed in . The cross-section also depends on the hypercharge of the Higgs representation, which is $`Y=2`$ in the HTM. This value of $`Y`$ is also assumed in the experimental searches. The explicit partonic cross-section at leading order (LO) is as follows (where $`q=u,d`$):
$$\sigma _{LO}(q\overline{q}H^{++}H^{})=\frac{\pi \alpha ^2}{9Q^2}\beta _1^3\left[e_q^2e_H^2+\frac{e_qe_H\mathrm{v}_q\mathrm{v}_H(1M_Z^2/Q^2)+(\mathrm{v}_q^2+\mathrm{a}_q^2)\mathrm{v}_H^2}{(1M_Z^2/Q^2)^2+M_Z^2\mathrm{\Gamma }_Z^2/Q^4}\right]$$
(9)
Here $`\mathrm{v}_q=(I_{3q}2e_qs_W^2)/(s_Wc_W)`$, $`\mathrm{a}_q=I_{3q}/(s_Wc_W)`$, and $`\mathrm{v}_H=(I_{3H}e_Hs_W^2)/(s_Wc_W)`$. The third isospin component is denoted by $`I_{3q}`$ ($`I_{3H}`$) and $`e_q(e_H)`$ is the electric charge of the quark $`q`$ $`(H^{\pm \pm })`$. $`s_W`$ and $`c_W`$ are $`\mathrm{sin}\theta _W`$ and $`\mathrm{cos}\theta _W`$ respectively. $`Q^2`$ is the partonic centre-of-mass (COM) energy. $`\alpha `$ is the QED coupling evaluated at the scale $`Q`$, $`M_Z`$ is the $`Z`$ boson mass, $`\mathrm{\Gamma }_Z`$ is the $`Z`$ boson width, and $`\beta _1=\sqrt{14m_{H^{\pm \pm }}^2/Q^2}`$. Order $`\alpha _s`$ QCD corrections modify the LO cross-section by a factor $`K1.3`$ at the Tevatron for $`m_{H^{\pm \pm }}<200`$ GeV, and $`K1.25`$ at the LHC for $`m_{H^{\pm \pm }}<1000`$ GeV . We neglect the gluon-gluon fusion ($`\alpha _s^2`$) contribution to $`H^{++}H^{}`$ production, which has no compensatory enhancement factor analogous to the $`\mathrm{tan}^4\beta `$ term for doublet $`H^\pm `$ production via $`ggH^+H^{}`$ .
Assuming that $`H^{\pm \pm }`$ production proceeds via this pair production process, the absence of signal enables a limit to be set on the product:
$$\sigma (p\overline{p}H^{++}H^{})\times BR(H^{\pm \pm }l_i^\pm l_j^\pm )$$
(10)
Clearly the strongest constraints on $`m_{H^{\pm \pm }}`$ are obtained assuming BR$`(H^{\pm \pm }l_i^\pm l_j^\pm )=100\%`$. Currently these mass limits stand at: 133,115,136 GeV for the $`e^\pm e^\pm ,e^\pm \mu ^\pm ,\mu ^\pm \mu ^\pm `$ channels respectively . In the HTM one expects $`BR(H^{\pm \pm }l_i^\pm l_j^\pm )100\%`$ if Eqn.(5) is required to explain the currently favoured form of the neutrino mass matrix .
The current search strategy is in fact sensitive to any singly produced $`H^{\pm \pm }`$, i.e. signal candidates are events with one pair of same sign leptons reconstructing to $`m_{H^{\pm \pm }}`$. This requirement is sufficient to reduce the SM background to negligible proportions. Hence the search potential of the Tevatron merely depends on the signal efficiencies for the signal (currently $`34\%,34\%,18\%`$ for $`\mu \mu ,ee,e\mu `$) and the integrated luminosity. With these relatively high efficiencies and an expected $`=48fb^1`$ by the year 2009, discovery with $`>5`$ events will be possible for $`\sigma _{H^{++}H^{}}`$ of a few fb, which corresponds to a mass reach $`m_{H^{\pm \pm }}<200`$ GeV.
Although single $`H^{\pm \pm }`$ production processes such as $`p\overline{p}W^\pm W^{}H^{\pm \pm }`$ can be neglected <sup>4 </sup><sup>4 </sup>4 Single production of a right-handed triplet via $`q^{}\overline{q}W_R^\pm W_R^{}H^{\pm \pm }`$ and $`W_R^\pm W_R^\pm `$ fusion can be sizeable at the LHC. due to the strong triplet VEV suppression, the mechanism $`p\overline{p}W^{}H^{\pm \pm }H^{}`$ is potentially sizeable. This latter process proceeds via a gauge coupling constant and is not suppressed by any small factor. The LO partonic cross-section is as follows:
$$\sigma _{LO}(q^{}\overline{q}H^{++}H^{})=\frac{\pi \alpha ^2}{144s_W^4Q^2}C_T^2p_W^2\beta _2^3$$
(11)
Here $`C_T`$ arises from the $`H^{\pm \pm }H^{}W^{}`$ vertex and $`C_T=2`$ for $`I=1`$,$`Y=2`$ triplet fields (the doublet component of $`H^\pm `$ is negligible); $`\beta _2=\sqrt{(1(m_{H^\pm }+m_{H^{\pm \pm }})^2/Q^2)(1(m_{H^\pm }m_{H^{\pm \pm }})^2/Q^2)}`$ and $`p_W=Q^2/(Q^2M_W^2)`$. For simplicity, we take the same $`K=1.3`$ as for $`\sigma (q\overline{q}H^{++}H^{})`$ at the Tevatron and $`K=1.25`$ at the LHC. Explicit calculations for the $`K`$ factor for the process $`\sigma (q^{}\overline{q}H^\pm A^0)`$ in the MSSM (which shares the same $`K`$ factor as $`q^{}\overline{q}H^{++}H^{}`$) give $`K1.2`$. In this paper we will study in detail the magnitude and relative importance of $`\sigma (q^{}\overline{q}H^{\pm \pm }H^{})`$. Although we work in the HTM, our numerical analysis is relevant for other models which possess a $`I=1`$,$`Y=2`$ Higgs triplet (e.g. L-R symmetric models and Little Higgs Models).
A previous quantitative study of this mechanism can be found in . Cross-sections were given at both LHC and Tevatron energies for $`m_{H^{\pm \pm }}>200`$ GeV with the simplifying assumption $`m_{H^{\pm \pm }}=m_{H^\pm }`$. It was shown that $`\sigma (q^{}\overline{q}H^{\pm \pm }H^{})`$ can of comparable size to $`\sigma (q\overline{q}H^{++}H^{}`$).
In this paper we first generalize the work of as follows:
* In our discussion at the Tevatron we consider masses in the range 100 GeV $`<m_{H^{\pm \pm }}<200`$ GeV which will be probed during Run II, and allow mild mass splittings $`|m_{H^{\pm \pm }}m_{H^\pm }|20`$ GeV.
* In our discussion at the LHC we consider larger mass splittings $`|m_{H^{\pm \pm }}m_{H^\pm }|80`$ GeV.
* For both the Tevatron and LHC we study in detail the relative magnitude of $`\sigma (q^{}\overline{q}H^{\pm \pm }H^{})`$ and $`\sigma (q\overline{q}H^{++}H^{}`$).
Moreover, motivated by the fact that the currently employed Tevatron search strategy is sensitive to single production of $`H^{\pm \pm }`$, we advocate the use of the inclusive single production cross-section ($`\sigma _{H^{\pm \pm }}`$) when comparing the experimentally excluded region with the theoretical cross-section. This leads to a strengthening of the mass bound for $`m_{H^{\pm \pm }}`$ which now carries a dependence on $`m_{H^\pm }`$. We introduce the single production cross-section as follows:
$$\sigma _{H^{\pm \pm }}=\sigma (p\overline{p},ppH^{++}H^{})+\sigma (p\overline{p},ppH^{++}H^{})+\sigma (p\overline{p},ppH^{}H^+)$$
(12)
At the Tevatron $`\sigma (p\overline{p}H^{++}H^{})=\sigma (p\overline{p}H^{}H^+)`$ while at the LHC $`\sigma (ppH^{++}H^{})>\sigma (ppH^{}H^+)`$. If a signal for $`H^{\pm \pm }`$ were found in the 2 lepton channel, subsequent searches could select signal events with 3 or 4 leptons, in order to disentangle $`q\overline{q}H^{++}H^{}`$ and $`q^{}\overline{q}H^{\pm \pm }H^{}`$. In our numerical analysis we utilize the CTEQ6L1 parton distribution functions (pdfs) . We take the factorization scale ($`Q`$) as the partonic COM energy ($`\sqrt{s}`$). Our results for $`\sigma (q\overline{q}H^{++}H^{})`$ agree with those in ,. Our results for $`\sigma (q^{}\overline{q}H^{\pm \pm }H^{})`$ agree with those in (and taking $`C_T=1`$ agree with $`\sigma (q^{}\overline{q}H^\pm A^0)`$ in the 2HDM/MSSM ). The above cross-sections evaluated with MRST02 pdfs agree with those evaluated with CTEQ6L1 to within $`10\%15\%`$.
In Fig. 1 (a) we plot $`\sigma _{H^{\pm \pm }}`$ as a function of $`m_{H^{\pm \pm }}`$ at the Tevatron for three different values of $`m_{H^\pm }`$. We take $`K=1.3`$. The current excluded regions from the $`e^\pm e^\pm ,e^\pm \mu ^\pm ,\mu ^\pm \mu ^\pm `$ searches correspond to the area above horizontal lines at roughly 40, 70, 35 fb respectively. The present mass limits for $`m_{H^{\pm \pm }}`$ are where the curve for $`H^{++}H^{}`$ intersects with the above horizontal lines, and read as $`133,115,136`$ GeV respectively for BR$`(H^{\pm \pm }l_i^\pm l_j^\pm )=100\%`$ . With the inclusion of the $`H^{\pm \pm }H^{}`$ channel, these mass limits increase to 150, 130, 150 for $`m_{H^\pm }=m_{H^{\pm \pm }}+20`$ GeV, strengthening to 160,140,160 for $`m_{H^\pm }=m_{H^{\pm \pm }}20`$ GeV. Clearly the search potential of the Tevatron (i.e. the mass limit on $`m_{H^{\pm \pm }}`$) increases significantly when one includes the contribution to $`\sigma _{H^{\pm \pm }}`$ from $`p\overline{p}H^{\pm \pm }H^{}`$. Note that the above mass limits strictly apply to the case when $`H^{\pm \pm }`$ decays leptonically, and with BR=$`100\%`$ in a given channel. However, if $`h_{ij}`$ are to provide the currently favoured form of the neutrino mass matrix then BR$`(H^{\pm \pm }l_i^\pm l_j^\pm )<100\%`$ in a given channel. Moreover, if $`m_{H^{\pm \pm }}>m_{H^\pm }`$ then the decay channel $`H^{\pm \pm }H^\pm W^{}`$ would be open. As shown in , this decay can be sizeable and thus reduces BR$`(H^{\pm \pm }l_i^\pm l_j^\pm `$). We will return to these issues in Section 5.
In Fig. 1 (b) we plot the ratio of cross-sections $`R`$ at the Tevatron as a function of $`m_{H^{\pm \pm }}`$, where $`R`$ is defined as follows:
$$R\frac{\sigma (p\overline{p},ppH^{++}H^{})+\sigma (p\overline{p},ppH^{}H^+)}{\sigma (p\overline{p},ppH^{++}H^{})}$$
(13)
The $`m_{H^{\pm \pm }}`$ dependence arises from the phase space functions $`\beta _1`$ and $`\beta _2`$ in Eqns.9 and 11. As can be seen, $`0.8<R<2.2`$ and thus $`q^{}\overline{q}H^{\pm \pm }H^{}`$ contributes significantly to $`\sigma _{H^{\pm \pm }}`$.
In Fig.2 we plot the analogies of Fig.1 for the LHC. In Fig. 2 (a) we plot $`\sigma _{H^{\pm \pm }}`$ for 3 values of $`m_{H^{\pm \pm }}`$ and for larger mass splittings ($`|m_{H^{\pm \pm }}m_{H^\pm }|80`$ GeV) than in Fig. 2. We take $`K=1.25`$. As before, the inclusion of $`q^{}\overline{q}H^{\pm \pm }H^{}`$ significantly increases the search potential e.g. if sensitivity to $`\sigma _{H^{\pm \pm }}=1`$ fb is attained, the mass reach extends from $`m_{H^{\pm \pm }}<600`$ GeV ($`H^{++}H^{}`$ only) to 750 GeV for ($`m_{H^\pm }=m_{H^{\pm \pm }}80`$ GeV). Recently performed a simulation of the detection prospects at the LHC for $`q\overline{q}H^{++}H^{}`$ for the cases where 3 and 4 leptons are detected. With 100 fb<sup>-1</sup>, sensitivity to $`m_{H^{\pm \pm }}\mathrm{\Gamma }<\mathrm{\hspace{0.17em}800}`$ GeV (3 leptons) and $`m_{H^{\pm \pm }}\mathrm{\Gamma }<\mathrm{\hspace{0.17em}700}`$ GeV (4 leptons) is expected. We are not aware of a simulation for the case where only 2 leptons are detected. Presumably even larger values of $`m_{H^{\pm \pm }}`$ ($`\mathrm{\Gamma }>\mathrm{\hspace{0.17em}800}`$ GeV) could be probed. In Fig. 2 (b) we plot $`R`$ as a function of $`m_{H^{\pm \pm }}`$. One can see that $`R>1`$ for the upper two curves for all $`m_{H^{\pm \pm }}`$, while for the lower curve $`R>1`$ for $`m_{H^{\pm \pm }}>260`$ GeV. Note that the dependence of $`R`$ on $`m_{H^{\pm \pm }}`$ differs from that observed in Fig. 1 (b), which can be attributed to the different parton luminosity functions at the Tevatron and LHC.
## 4 Neutrino mass hierarchy and the decay $`H^{\pm \pm }H^\pm W^{}`$
The current experimental searches assume that the sole decay mode of $`H^{\pm \pm }`$ is $`H^{\pm \pm }l_i^\pm l_j^\pm `$ mediated by the arbitrary Yukawa couplings $`h_{ij}`$. The decay rate for $`H^{\pm \pm }l_i^\pm l_j^\pm `$ is given by:
$$\mathrm{\Gamma }(H^{\pm \pm }l_i^\pm l_j^\pm )=S\frac{m_{H^{\pm \pm }}}{8\pi }|h_{ij}|^2$$
(14)
where $`S=1(2)`$ for $`i=j`$ ($`ij`$). Clearly $`\mathrm{\Gamma }(H^{\pm \pm }l_i^\pm l_j^\pm )`$ depends crucially on the absolute value of the $`h_{ij}`$, although the leptonic BRs are determined by the relative values. In this section we consider the impact of the decay mode $`H^{\pm \pm }H^\pm W^{}`$ on the BRs of the leptonic channels. It has been known for some time that BR($`H^{\pm \pm }H^\pm W^{}`$) is potentially sizeable and a quantitative analysis can be found in . The decay rate for $`H^{\pm \pm }H^\pm W^{}`$ (summing over all fermion states for $`W^{}ff`$ excluding the $`t`$ quark) is given by:
$$\mathrm{\Gamma }(H^{\pm \pm }H^\pm W^{})=9G_F^2M_W^4m_{H^{\pm \pm }}C_T^2P/(16\pi ^3)$$
(15)
where $`P`$ is the phase space term (which we calculate by numerical integration) and $`C_T(=2)`$ is from the coupling $`H^{\pm \pm }H^\pm W`$. $`P`$ depends on the mass difference $`\mathrm{\Delta }m`$ defined by $`\mathrm{\Delta }m=m_{H^{\pm \pm }}m_{H^\pm }`$, and $`P=0`$ for $`\mathrm{\Delta }m=0`$. If $`m_{H^\pm }<m_{H^{\pm \pm }}`$ this decay can compete with $`H^{\pm \pm }l_i^\pm l_j^\pm `$ since the phase space suppression of the virtual $`W^{}`$ is compensated by the gauge strength coupling . Ref. showed that $`H^{\pm \pm }H^\pm W^{}`$ can dominate over $`H^{\pm \pm }l_i^\pm l_j^\pm `$ if $`\mathrm{\Delta }m`$ is sizeable ($`>40`$ GeV) and $`h_{ij}`$ are of order $`10^3`$ or less. A large BR($`H^{\pm \pm }H^\pm W^{}`$) would debilitate the $`H^{\pm \pm }`$ search potential in the leptonic channel. However, as emphasized in , observation of $`H^{\pm \pm }H^\pm W^{}`$ together with one or more of the leptonic channels could provide information on the absolute values of $`h_{ij}`$. If only BR($`H^{\pm \pm }l_i^\pm l_j^\pm `$) are measured then only the relative values of the $`h_{ij}`$ can be evaluated. The decay rate for $`H^{\pm \pm }H^\pm W^{}`$ is theoretically calculable once $`m_{H^\pm }`$ and $`m_{H^{\pm \pm }}`$ are known experimentally, and thus it can be used as a benchmark decay with which to estimate the total width of $`H^{\pm \pm }`$. It is known that the BRs of the leptonic channels depend on which solution to the neutrino mass matrix is realized . However, a quantitative analysis of the impact of $`H^{\pm \pm }H^\pm W^{}`$ in the various allowed scenarios is still lacking and will be presented below. We are not aware of any experimental simulation of $`H^{\pm \pm }H^\pm W^{}`$. The signature would depend crucially on the decay products of $`H^\pm `$, which are are either $`H^\pm l^\pm \nu _l`$ (driven by $`h_{ij}`$), or possibly $`H^\pm H^0W^{},A^0W^{}`$.
We now briefly review relevant results and formulae from neutrino physics. The neutrino mass matrix is diagonalized by the MNS (Maki-Nakagawa-Sakata) matrix $`V_{_{\mathrm{MNS}}}`$ . Using Eq.(5) one can write the couplings $`h_{ij}`$ as follows:
$$h_{ij}=\frac{1}{\sqrt{2}v_\mathrm{\Delta }}V_{_{\mathrm{MNS}}}diag(m_1,m_2,m_3)V_{_{\mathrm{MNS}}}^T$$
(16)
Here we take the basis in which the unitary matrix responsible for diagonalizing the charged-lepton mass matrix is a unit matrix. The MNS matrix in the standard parametrization is as follows:
$$V_{_{\mathrm{MNS}}}=\left(\begin{array}{ccc}c_1c_3& s_1c_3& s_3e^{i\delta }\\ s_1c_2c_1s_2s_3e^{i\delta }& c_1c_2s_1s_2s_3e^{i\delta }& s_2c_3\\ s_1s_2c_1c_2s_3e^{i\delta }& c_1s_2s_1c_2s_3e^{i\delta }& c_2c_3\end{array}\right)\left(\begin{array}{ccc}1& 0& 0\\ 0& e^{i\phi _1/2}& 0\\ 0& 0& e^{i\phi _2/2}\end{array}\right),$$
(17)
where $`s_i\mathrm{sin}\theta _i`$ and $`c_i\mathrm{cos}\theta _i`$, $`\delta `$ is the Dirac phase and $`\phi _1`$ and $`\phi _2`$ are the Majorana phases.
Neutrino oscillation experiments involving solar , atmospheric and reactor neutrinos are sensitive to the mass-squared differences and the mixing angles. and give the following preferred values:
$`\mathrm{\Delta }m_{12}^2m_2^2m_1^28.0\times 10^5\mathrm{eV}^2,|\mathrm{\Delta }m_{13}^2||m_3^2m_1^2|2.1\times 10^3\mathrm{eV}^2,`$ (18)
$`\mathrm{sin}^22\theta _10.8,\mathrm{sin}^22\theta _21,\mathrm{sin}^22\theta _3\mathrm{\Gamma }<\mathrm{\hspace{0.17em}0.16}.`$ (19)
Since the sign of $`\mathrm{\Delta }m_{13}^2`$ and the mass of the lightest neutrino are both undetermined at present, distinct neutrino mass hierarchy patterns are classified as follows: Normal hierarchy (NH) ($`m_1<m_2m_3`$), Inverted hierarchy (IH) ($`m_2>m_1m_3`$), Quasi-degenerate (DG) ($`m_1m_2m_3\sqrt{|\mathrm{\Delta }m_{13}^2|}`$). From Eq.(5) and Eq.(16) it can be shown that:
$$\underset{i,j}{}h_{ij}^2v_\mathrm{\Delta }^2\underset{i}{}m_i^2,$$
(20)
Hence the total leptonic decay width depends on the absolute mass of the neutrinos, and the value of $`_im_i^2`$ depends on which solution to the neutrino mass matrix (NH,IH,DG) is realized. The minimum value of $`_im_i^2`$ is $`|\mathrm{\Delta }m_{13}^2|`$ while the maximum is given by the cosmological constraint.
In Fig.3 we show contours of BR$`(H^{\pm \pm }H^\pm W^{})`$ in the plane ($`m_{H^{\pm \pm }},v_\mathrm{\Delta }`$), for three different solutions to the neutrino mass matrix. We assume that $`m_{1(3)}=0`$ for NH (IH) and $`m_1=0.2`$ eV for DG. We take $`m_{H^\pm }=m_{H^{\pm \pm }}20`$ GeV. From Eq.16, all $`h_{ij}`$ are determined once $`v_\mathrm{\Delta }`$ is specified. In order to comply with current experimental upper limits on LFV decays of $`\mu ^\pm `$ and $`\tau ^\pm `$, one can derive the bound $`v_\mathrm{\Delta }>10`$ eV for NH and IH, and $`v_\mathrm{\Delta }>100`$ eV for DG. The stronger constraint on $`v_\mathrm{\Delta }`$ in DG arises because $`_im_i`$ in DG is larger than those in NH and IH.
From Fig.3 it is clear that BR$`(H^{\pm \pm }H^\pm W^{})`$ can be sizeable, and approaches $`100\%`$ for larger $`v_\mathrm{\Delta }`$. For a fixed value of $`v_\mathrm{\Delta }`$, one can see that BR$`(H^{\pm \pm }H^\pm W^{})`$ is relatively more important in NH and IH than in DG. This can be understood from Eq. 20, since DG requires heavier neutrinos (and thus larger $`h_{ij}`$) which in turn reduces BR$`(H^{\pm \pm }H^\pm W^{})`$. One can consider three distinct scenarios with very different magnitudes for BR$`(H^{\pm \pm }H^\pm W^{})`$ and BR$`(H^{\pm \pm }l^\pm l^\pm `$):
* BR$`(H^{\pm \pm }H^\pm W^{})`$ BR$`(H^{\pm \pm }l^\pm l^\pm `$):
In this case the current search strategy (which requires $`H^{\pm \pm }l^\pm l^\pm `$ decay) is ineffective. Simulations have not been carried out for the decay $`H^{\pm \pm }H^\pm W^{}`$ although one might naively expect sensitivity comparable to that for the decay $`H^{\pm \pm }\tau ^\pm \tau ^\pm `$, as suggested in ,.
* BR$`(H^{\pm \pm }H^\pm W^{})`$ BR$`(H^{\pm \pm }l^\pm l^\pm `$):
The search for $`H^{\pm \pm }l^\pm l^\pm `$ would be effective and $`H^{\pm \pm }`$ could be discovered in one or more leptonic channels. If $`H^{\pm \pm }H^\pm W^{}`$ is also observed then information on the absolute value of $`h_{ij}`$ might be possible: Using Eqs.14 and 15, the ratio of leptonic events ($`N_{l_il_j}`$) to $`H^\pm W^{}`$ events ($`N_{H^\pm W^{}}`$) is given as follows:
$$\frac{N_{l_il_j}}{N_{H^\pm W^{}}}\frac{h_{ij}^2}{P}$$
(21)
Observation of the leptonic channel provides $`m_{H^{\pm \pm }}`$. If $`m_{H^\pm }`$ can be roughly measured then $`P`$ (and hence the partial width for $`H^{\pm \pm }H^\pm W^{}`$) can be calculated. From the above equation one can obtain an order of magnitude estimate of $`h_{ij}`$.
* BR$`(H^{\pm \pm }H^\pm W^{})`$ BR$`(H^{\pm \pm }l^\pm l^\pm `$):
In this case the current search strategy ($`H^{\pm \pm }l^\pm l^\pm `$) is effective. If BR$`(H^{\pm \pm }l_i^\pm l_j^\pm `$) are measured then the ratios of $`h_{ij}`$ can be evaluated. This can be compared with Eqn.5 in order to see which neutrino solution is realized . The absolute values of $`h_{ij}`$ cannot be measured unless a LFV decay of $`\mu `$ and/or $`\tau `$ is observed.
## 5 Tevatron search potential in HTM
We now study the search potential of the Tevatron for the generalized case in the HTM where $`p\overline{p}H^{\pm \pm }H^{}`$ is included, BR$`(H^{\pm \pm }H^\pm W^{})0\%`$ and $`h_{ij}`$ are required to reproduce a phenomenologically acceptable neutrino mass matrix. We relax the assumptions for the Majorana phases and take $`\phi _1,\phi _2=0`$ or $`\pi `$, which leads the 7 distinct solutions:
| NH: | $`m_1<m_2m_3`$, | | | |
| --- | --- | --- | --- | --- |
| IH1: | $`m_2>m_1m_3`$, | | IH2: | $`m_2>m_1m_3`$, |
| DG1: | $`m_1m_2m_3`$, | | DG2: | $`m_1m_2m_3`$, |
| DG3: | $`m_1m_2m_3`$, | | DG4: | $`m_1m_2m_3`$. |
In the HTM, BR$`(H^{\pm \pm }l^\pm l^\pm )`$ are predicted and different in each of the 7 distinct solutions (NH,IH1,IH2,DG1$``$DG4), and their ratios were evaluated in . Note that such predictions of BR$`(H^{\pm \pm }l^\pm l^\pm )`$ are a feature of the HTM in which the couplings $`h_{ij}`$ are the sole origin of neutrino mass. This direct correlation between BR$`(H^{\pm \pm }l^\pm l^\pm )`$ and the neutrino mass matrix may not extend to $`H^{\pm \pm }`$ of other models in which neutrinos can acquire mass by other means e.g. the seesaw mechanism in L-R models or by a combination of mechanisms which may or may not include the $`h_{ij}`$ couplings ,. In contrast, the production process $`\sigma (p\overline{p}H^{\pm \pm }H^{}`$) is certainly relevant in any model with $`Y=2`$ triplets.
In Figs.4$``$ 6 we plot $`\sigma _{ll}`$ as a function of $`m_{H^{\pm \pm }}`$, where $`\sigma _{ll}`$ is the total leptonic ($`l=e,\mu ,\tau `$) cross-section defined as:
$$\sigma _{ll}=\sigma (p\overline{p}H^{++}H^{})\times B_{ll}(2B_{ll})+2\sigma (p\overline{p}H^{++}H^{})\times B_{ll}$$
(22)
The contribution to $`\sigma _{ll}`$ from $`\sigma (p\overline{p}H^{++}H^{})`$ falls more slowly with decreasing $`B_{ll}`$ since signal candidates are events with at least 2 leptons. Eq.(22) simplifies to Eq.10 in the limit where $`\sigma (p\overline{p}H^{\pm \pm }H^{})=0`$ and $`B_{ll}=1`$. Figs.4(a) shows $`\sigma _{ll}`$ for the NH with $`m_{H^\pm }=m_{H^{\pm \pm }}`$, which leads to $`B_{ll}=1`$. In this case $`\sigma _{ll}=\sigma _{H^{\pm \pm }}`$. For the other figures we take $`m_{H^\pm }=m_{H^{\pm \pm }}20`$ GeV, which induces a sizeable (but not dominant) BR($`H^{\pm \pm }H^\pm W^{})`$, and hence $`\sigma _{ll}<\sigma _{H^{\pm \pm }}`$. We set $`v_\mathrm{\Delta }=10`$ eV in Figs.4 and 5 and $`v_\mathrm{\Delta }=100`$ eV in Figs.6. We only plot $`\sigma _{ll}`$ for $`ee,e\mu ,\mu \mu `$ since the Tevatron has already performed searches in these channels. Sensitivity to $`\sigma _{ll}`$ of a few fb will be possible with the anticipated integrated luminosities of $`48`$ fb<sup>-1</sup>. There are plans to search for the 3 leptonic decays involving $`\tau `$ ($`e\tau ,\mu \tau ,\tau \tau `$) although the discovery reach in $`m_{H^{\pm \pm }}`$ is expected to be inferior to that for the $`ee,e\mu ,\mu \mu `$ channels. In all figures we take $`\theta _3=0^{}`$. From the figures it is clear that $`\sigma _{ee,e\mu ,\mu \mu }`$ differ considerably in each of the 7 scenarios. Optimal coverage is for cases DG1 and DG4, which have $`\sigma _{ee,\mu \mu }5`$ fb and $`\sigma _{e\mu ,\mu \mu }5`$ fb respectively for $`m_{H^{\pm \pm }}\mathrm{\Gamma }<\mathrm{\hspace{0.17em}180}`$ GeV. For NH, $`\sigma _{\mu \mu }5`$ fb for $`m_{H^{\pm \pm }}\mathrm{\Gamma }<\mathrm{\hspace{0.17em}190}`$ GeV but $`\sigma _{ee}`$ and $`\sigma _{e\mu }`$ are both unobservable. Taking $`\theta _3`$ at its largest experimentally allowed value results in minor changes to all figures, with the most noticeable effect being a significant reduction of $`\sigma _{\mu \mu }`$ in DG4. Clearly the Tevatron Run II not only has strong search potential for $`H^{\pm \pm }`$, but is also capable of distinguishing between the various allowed scenarios for the neutrino mass matrix.
## 6 Conclusions
We have studied the production of doubly charged Higgs bosons ($`H^{\pm \pm }`$) at hadron colliders in the Higgs Triplet Model (HTM), in which a complex $`Y=2`$ scalar triplet is added to the Standard Model. The HTM can explain the observed neutrino mass matrix by invoking Yukawa couplings $`h_{ij}`$ of the triplet fields to leptons. A definitive signal of the HTM would be the observation of the decay $`H^{\pm \pm }l^\pm l^\pm `$, which enjoys almost negligible background at hadron colliders, and whose branching ratios are correlated with the neutrino mass matrix. We studied the production mechanism $`q^{}\overline{q}H^{\pm \pm }H^{}`$ which can be as large as the mechanism $`q\overline{q}H^{++}H^{}`$ assumed in the current searches at the Tevatron. Since the present search strategy is sensitive to single production of $`H^{\pm \pm }`$, we advocated the use of the inclusive single production cross-section ($`\sigma _{H^{\pm \pm }}`$) when comparing the experimentally excluded region with the theoretical cross-section. This leads to a strengthening of the mass bound for $`m_{H^{\pm \pm }}`$ which now carries a dependence on $`m_{H^\pm }`$, and significantly improves the $`H^{\pm \pm }`$ search potential at the Tevatron and LHC. Although we performed our numerical analysis in the HTM, we emphasized that the introduction of $`\sigma _{H^{\pm \pm }}`$ is also relevant for any model which contains a $`Y=2`$ Higgs triplet (e.g. L-R symmetric models and Little Higgs Models).
Moreover, we quantified the impact of the decay mode $`H^{\pm \pm }H^\pm W^{}`$ for the case of a hierarchical, inverted hierarchical and degenerate neutrino mass spectrum. On discovering a $`H^{\pm \pm }`$ it would be imperative to measure the absolute value of $`h_{ij}`$ (and hence $`v_\mathrm{\Delta }`$) in order to reconstruct the low energy Higgs triplet Lagrangian. We stressed that an order of magnitude estimate of $`h_{ij}`$ could be obtained if the channel $`H^{\pm \pm }H^\pm W^{}`$ is observed and $`m_{H^\pm }`$ is roughly measured. We encourage a detailed experimental simulation of this decay mode at both the Tevatron and LHC.
|
warning/0506/math0506585.html
|
ar5iv
|
text
|
# Greatest least eigenvalue of the Laplacian on the Klein bottle
## 1. Introduction and statement of main results
Among all the possible Riemannian metrics on a compact differentiable manifold $`M`$, the most interesting ones are those which extremize a given Riemannian invariant. In particular, many recent works have been devoted to the metrics which maximize the fundamental eigenvalue $`\lambda _1(M,g)`$ of the Laplace-Beltrami operator $`\mathrm{\Delta }_g`$ under various constraints (see, for instance, ). Notice that, since $`\lambda _1(M,g)`$ is not invariant under scaling ($`\lambda _1(M,kg)=k^1\lambda _1(M,g)`$), such constraints are necessary.
In , Yang and Yau proved that, on any compact orientable surface $`M`$, the first eigenvalue $`\lambda _1(M,g)`$ is uniformly bounded over the set of Riemannian metrics of fixed area. More precisely, one has, for any Riemannian metric $`g`$ on $`M`$,
$$\lambda _1(M,g)A(M,g)8\pi (\text{genus}(M)+1),$$
where $`A(M,g)`$ stands for the Riemannian area of $`(M,g)`$ (see for an improvement of the upper bound). In the non-orientable case, the following upper bound follows from Li and Yauโs work : $`\lambda _1(M,g)A(M,g)24\pi (\text{genus}(M)+1)`$. On the other hand, if the dimension of $`M`$ is greater than 2, then $`\lambda _1(M,g)`$ is never bounded above over the set of Riemannian metrics of fixed volume, see .
Hence, one obtains a relevant topological invariant of surfaces by setting, for any compact 2-dimensional manifold $`M`$,
$$\mathrm{\Lambda }(M)=\underset{g}{sup}\lambda _1(M,g)A(M,g)=\underset{g(M)}{sup}\lambda _1(M,g),$$
where $`(M)`$ denotes the set of Riemannian metrics of area 1 on $`M`$. The natural questions related to this invariant are :
1. How does $`\mathrm{\Lambda }(M)`$ depend on (the genus of) $`M`$?
2. Can one determine $`\mathrm{\Lambda }(M)`$ for some $`M`$?
3. Is the supremum $`\mathrm{\Lambda }(M)`$ achieved, and, if so, by what metrics?
Concerning the first question, it follows from that $`\mathrm{\Lambda }(M)`$ is an increasing function of the genus with a linear growth rate. Explicit answers to questions (2) and (3) are only known for the three following surfaces : the sphere $`๐^2`$, the real projective plane $`P^2`$ and the Torus $`๐^2`$. Indeed, according to the results of Hersch , Li and Yau and Nadirashvili , one has
$$\mathrm{\Lambda }(๐^2)=\lambda _1(๐^2,g_{๐^2})A(๐^2,g_{๐^2})=8\pi ,$$
where $`g_{๐^2}`$ is the standard metric of $`๐^2`$ (see ),
$$\mathrm{\Lambda }(P^2)=\lambda _1(P^2,g_{P^2})A(P^2,g_{P^2})=12\pi ,$$
where $`g_{P^2}`$ is the standard metric of $`P^2`$ (see ), and
$$\mathrm{\Lambda }(๐^2)=\lambda _1(๐^2,g_{eq})A(๐^2,g_{eq})=\frac{8\pi ^2}{\sqrt{3}},$$
where $`g_{eq}`$ is the flat metric on $`๐^2^2/\mathrm{\Gamma }_{eq}`$ corresponding to the equilateral lattice $`\mathrm{\Gamma }_{eq}=(1,0)(\frac{1}{2},\frac{\sqrt{3}}{2})`$ (see ). Moreover, on each one of these three surfaces, the maximizing metric is unique, up to a dilatation.
What about the Klein bottle $`๐`$?
Nadirashvili proved that the supremum $`\mathrm{\Lambda }(๐)`$ is necessarily achieved by a regular (real analytic) Riemannian metric. Recently, Jakobson, Nadirashvili and Polterovich conjectured that the exact value of $`\mathrm{\Lambda }(๐)`$ is given by
$$\mathrm{\Lambda }(๐)=12\pi E(2\sqrt{2}/3)13.365\pi ,$$
where $`E(2\sqrt{2}/3)`$ is the complete elliptic integral of the second kind evaluated at $`\frac{2\sqrt{2}}{3}`$. They also conjectured that this value is uniquely achieved, up to a dilatation, by the metric of revolution
$$g_0=\frac{9+(1+8\mathrm{cos}^2v)^2}{1+8\mathrm{cos}^2v}\left(du^2+\frac{dv^2}{1+8\mathrm{cos}^2v}\right),$$
with $`0u,v<\pi `$.
The main purpose of this paper is to prove this conjecture. Indeed, we will prove the following
###### Theorem 1.1.
For any Riemannian metric $`g`$ on $`๐`$ one has
$$\lambda _1(๐,g)A(๐,g)\lambda _1(๐,g_0)A(๐,g_0)=12\pi E(2\sqrt{2}/3).$$
Moreover, the equality holds for a metric $`g`$ if, and only if, $`g`$ is homothetic to $`g_0`$.
As noticed in , if we denote by $`(๐^2,\overline{g}_0)`$ the double cover of $`(๐,g_0)`$, then $`\overline{g}_0`$ is nothing but the Riemannian metric induced on $`๐^2`$ from the bipolar surface of Lawsonโs minimal torus $`\tau _{3,1}`$ defined as the image in $`๐^3`$ of the map
$$(u,v)(\mathrm{cos}v\mathrm{exp}(3iu),\mathrm{sin}v\mathrm{exp}(iu)).$$
It is worth noticing that the metric $`g_0`$ does not maximize the systole functional $`g\text{sys}(g)`$ (where $`\text{sys}(g)`$ denotes the length of the shortest noncontractible loop) over the set of metrics of fixed area on the Klein bottle (see ), while on $`P^2`$ and $`๐^2`$, the functionals $`\lambda _1`$ and sys are maximized by the same Riemannian metrics.
The proof of Theorem 1.1 relies on the characterization of critical metrics of the functional $`\lambda _1`$ in terms of minimal immersions into spheres by the first eigenfunctions. Note that, in spite of the non-differentiability of this functional with respect to metric deformations, a natural notion of criticality can be introduced (see ). Moreover, the criticality of a metric $`g`$ for $`\lambda _1`$ with respect to area preserving deformations is characterized by the existence of a family $`h_1,\mathrm{},h_d`$ of first eigenfunctions of $`\mathrm{\Delta }_g`$ satisfying $`_{id}dh_idh_i=g`$. This last condition actually means that the map $`(h_1,\mathrm{},h_d)`$ is an isometric immersion from $`(M,g)`$ into $`^d`$ whose image is a minimal immersed submanifold of a sphere. Theorem 1.1 then follows from Nadirashviliโs existence result and the following
###### Theorem 1.2.
The Riemannian metric $`g_0`$ is, up to a dilatation, the only critical metric of the functional $`\lambda _1`$ under area preserving deformations of metrics on $`๐`$.
Equivalently, the metric $`g_0`$ is, up to a dilatation, the only metric on $`๐`$ such that $`(๐,g_0)`$ admits a minimal isometric immersion into a sphere by its first eigenfunctions.
In , Ilias and the first author gave a necessary condition of symmetry for a Riemannian metric to admit isometric immersions into spheres by the first eigenfunctions. On the Klein bottle, this condition amounts to the invariance of the metric under the natural $`๐^1`$-action on $`๐`$. Taking into account this symmetry property and the fact that any metric $`g`$ is conformally equivalent to a flat one, for which the eigenvalues and the eigenfunctions of the Laplacian are explicitly known, it is of course expected that the existence problem of minimal isometric immersions into spheres by the first eigenfunctions reduces to a second order system of ODEs (see Proposition 2.1). Actually, the substantial part of this paper is devoted to the study of the following second order nonlinear system:
(1)
$$\{\begin{array}{ccc}\phi _1^{\prime \prime }=(12\phi _1^28\phi _2^2)\phi _1,\hfill & & \\ & & \\ \phi _2^{\prime \prime }=(42\phi _1^28\phi _2^2)\phi _2,\hfill & & \end{array}$$
for which we look for periodic solutions satisfying
(2)
$$\{\begin{array}{c}\phi _1\text{ is odd and has exactly two zeros in a period,}\hfill \\ \phi _2\text{ is even and positive everywhere;}\hfill \end{array}$$
and the initial conditions
(3)
$$\{\begin{array}{c}\phi _1(0)=\phi _2^{}(0)=0\text{(from parity conditions),}\hfill \\ \phi _2(0)=\frac{1}{2}\phi _1^{}(0)=:p(0,1].\hfill \end{array}$$
In , Jakobson, Nadirashvili and Polterovich proved that the initial value $`p=\phi _2(0)=\sqrt{3/8}`$ corresponds to a periodic solution of (1)-(3) satisfying (2). Based on numerical evidence, they conjectured that this value of $`p`$ is the only one corresponding to a periodic solution satisfying (2). As mentioned by them, a computer-assisted proof of this conjecture is extremely difficult, due to the lack of stability of the system.
In Section 3, we provide a complete analytic study of System (1). First, we show that this system admits two independent first integrals (one of them has been already found in ). Using a suitable linear change of variables, we show that the system becomes Hamiltonian and, hence, integrable. The general theory of integrable Hamiltonian systems tells us that bounded orbits correspond to periodic or quasi-periodic solutions (see ). However, to distinguish periodic solutions from non-periodic ones is not easy in general. Fortunately, our first integrals turn out to be quadratic in the momenta, which enables us to apply the classical Bertrand-Darboux-Whittaker Theorem and, therefore, to completely decouple the system by means of a parabolic type change of coordinates $`(\phi _1,\phi _2)(u,v)`$. We show that, for any $`p\sqrt{3}/2`$, the solutions $`u`$ and $`v`$ of the decoupled system are periodic. The couple $`(u,v)`$ is then periodic if and only if the periods of $`u`$ and $`v`$ are commensurable. We express the periods of $`u`$ and $`v`$ in terms of hyper-elliptic integrals and study their ratio as a function of $`p`$. The following fact (Proposition 3.1) gives an idea about the complexity of the situation: there exists a countable dense subset $`๐ซ(0,\sqrt{3}/2)`$ such that the solution of (1)-(3) corresponding to $`p(0,\sqrt{3}/2)`$ is periodic if and only if $`p๐ซ`$.
In conclusion, we show that the solution associated with $`p=\sqrt{3/8}`$ is the only periodic one to satisfy Condition (2).
## 2. Preliminaries: reduction of the problem
Let us first recall that, if $`g_\epsilon `$ is a smooth deformation of a metric $`g`$ on a compact surface $`M`$, then the function $`\epsilon \lambda _1(M,g_\epsilon )`$, which is continuous but not necessarily differentiable at $`\epsilon =0`$, always admits left and right derivatives at $`\epsilon =0`$ with
$$\frac{d}{d\epsilon }\lambda _1(M,g_\epsilon )|_{\epsilon =0^+}\frac{d}{d\epsilon }\lambda _1(M,g_\epsilon )|_{\epsilon =0^{}}$$
(see for details). The metric $`g`$ is then said to be critical for the functional $`\lambda _1`$ under area preserving deformations if, for any deformation $`g_\epsilon `$ with $`g_0=g`$ and $`A(M,g_\epsilon )=A(M,g)`$, one has
$$\frac{d}{d\epsilon }\lambda _1(M,g_\epsilon )|_{\epsilon =0^+}0\frac{d}{d\epsilon }\lambda _1(M,g_\epsilon )|_{\epsilon =0^{}}.$$
It is easy to check that this last condition is equivalent to:
$$\lambda _1(M,g_\epsilon )\lambda _1(M,g)+o(\epsilon )\text{ as }\epsilon 0.$$
Of course, the metric $`g`$ is critical for the functional $`\lambda _1`$ under area preserving deformations if, and only if, it is critical for the functional $`\lambda _1A`$.
Following and , a necessary and sufficient condition for the metric $`g`$ to be critical for $`\lambda _1`$ under area-preserving metric deformations is that there exists a family $`h_1,\mathrm{},h_d`$ of first eigenfunctions of $`\mathrm{\Delta }_g`$ satisfying
(4)
$$\underset{id}{}dh_idh_i=g,$$
which means that the map $`h=(h_1,\mathrm{},h_d)`$ is an isometric immersion from $`(M,g)`$ to $`^d`$. Since $`h_1,\mathrm{},h_d`$ are eigenfunctions of $`\mathrm{\Delta }_g`$, the image of $`h`$ is a minimal immersed submanifold of the Euclidean sphere $`๐^{d1}\left(\sqrt{\frac{2}{\lambda _1(M,g)}}\right)`$ of radius $`\sqrt{2/\lambda _1(M,g)}`$ (Takahashiโs theorem ). In particular, we have
(5)
$$\underset{id}{}h_i^2=\frac{2}{\lambda _1(M,g)}.$$
In , Ilias and the first author have studied conformal properties of Riemannian manifolds $`(M,g)`$ admitting such minimal isometric immersions into spheres. It follows from their results that, if $`g`$ is a critical metric of $`\lambda _1`$ under area preserving deformations, then
* $`g`$ is, up to a dilatation, the unique critical metric in its conformal class,
* $`g`$ maximizes the restriction of $`\lambda _1`$ to the set of metrics conformal to $`g`$ and having the same volume,
* the isometry group of $`(M,g)`$ contains the isometry groups of all the metrics $`g^{}`$ conformal to $`g`$.
For any positive real number $`a`$, we denote by $`\mathrm{\Gamma }_a`$ the rectangular lattice of $`^2`$ generated by the vectors $`(2\pi ,0)`$ and $`(0,a)`$ and by $`\stackrel{~}{g}_a`$ the flat Riemannian metric of the torus $`๐_a^2^2/\mathrm{\Gamma }_a`$ associated with the rectangular lattice $`\mathrm{\Gamma }_a`$. The Klein bottle $`๐`$ is then diffeomorphic to the quotient of $`๐_a^2`$ by the involution $`s:(x,y)(x+\pi ,y)`$. We denote by $`g_a`$ the flat metric induced on $`๐`$ by such a diffeomorphism. It is well known that any Riemannian metric on $`๐`$ is conformally equivalent to one of the flat metrics $`g_a`$.
Let $`g=fg_a`$ be a Riemannian metric on $`๐`$. From the property (iii) above, if $`g`$ is a critical metric of $`\lambda _1`$ under area preserving deformations, then $`\text{Isom}(๐,g_a)\text{Isom}(๐,g)`$, which implies that the function $`f`$ is invariant under the $`๐^1`$-action $`(x,y)(x+t,y)`$, $`t[0,\pi ]`$, on $`๐`$, and then, $`f`$ (or its lift to $`^2`$) does not depend on the variable $`x`$.
###### Proposition 2.1.
Let $`a`$ be a positive real number and $`f`$ a positive periodic function of period $`a`$. The following assertions are equivalent
* The Riemannian metric $`g=f(y)g_a`$ on $`๐`$ is a critical metric of $`\lambda _1`$ under area preserving deformations.
* The function $`f`$ is proportional to $`\phi _1^2+4\phi _2^2`$, where $`\phi _1`$ and $`\phi _2`$ are two periodic functions of period $`a`$ satisfying the following conditions:
1. $`(\phi _1,\phi _2)`$ is a solution of the equations
$$\{\begin{array}{ccc}\phi _1^{\prime \prime }=(12\phi _1^28\phi _2^2)\phi _1,\hfill & & \\ & & \\ \phi _2^{\prime \prime }=(42\phi _1^28\phi _2^2)\phi _2;\hfill & & \end{array}$$
2. $`\phi _1`$ is odd, $`\phi _2`$ is even and $`\phi _1^{}(0)=2\phi _2(0)`$;
3. $`\phi _1`$ admits two zeros in a period and $`\phi _2`$ is positive everywhere;
4. $`\phi _1^2+\phi _2^21`$ and the equality holds at exactly two points in a period.
Most of the arguments of the proof of โ(I) implies (II)โ can be found in . For the sake of completeness, we will recall the main steps. The proof of the converse relies on the fact that the system (1) admits two independent first integrals.
###### Proof.
The Laplacian $`\mathrm{\Delta }_g`$ associated with the Riemannian metric $`g=f(y)g_a`$ on $`๐`$ can be identified with the operator $`\frac{1}{f(y)}\left(_x^2+_y^2\right)`$ acting on
$`\mathrm{\Gamma }_a`$-periodic and $`s`$-invariant functions on $`^2`$. Using separation of variables and Fourier expansions, one can easily show that any eigenfunction of $`\mathrm{\Delta }_g`$ is a linear combination of functions of the form $`\phi _k(y)\mathrm{cos}kx`$ and $`\phi _k(y)\mathrm{sin}kx`$, where, $`k`$, $`\phi _k`$ is a periodic function with period $`a`$ satisfying $`\phi _k(y)=(1)^k\phi _k(y)`$ and $`\phi _k^{\prime \prime }=(k^2\lambda f)\phi _k`$. Since a first eigenfunction always admits exactly two nodal domains, the first eigenspace of $`\mathrm{\Delta }_g`$ is spanned by
$$\{\phi _0(y),\phi _1(y)\mathrm{cos}x,\phi _1(y)\mathrm{sin}x,\phi _2(y)\mathrm{cos}2x,\phi _2(y)\mathrm{sin}2x\},$$
where, unless they are identically zero, $`\phi _2`$ does not vanish while $`\phi _0`$ and $`\phi _1`$ admit exactly two zeros in $`[0,a)`$. In particular, the multiplicity of $`\lambda _1(๐,g)`$ is at most 5.
Let us suppose that $`g`$ is a critical metric of $`\lambda _1`$ under area preserving deformations and let $`h_1,\mathrm{},h_d`$ be a family of first eigenfunctions satisfying the equations (4) and (5) above. Without loss of generality, we may assume that $`\lambda _1(๐,g)=2`$ and that $`h_1,\mathrm{},h_d`$ are linearly independent, which implies that $`d5`$. Since $`h=(h_1,\mathrm{},h_d):๐๐^{d1}`$ is an immersion, one has $`d4`$. If $`d=4`$, then using elementary algebraic arguments like in the proof of Proposition 5 of , one can see that there exists an isometry $`\rho O(4)`$ such that $`\rho h=(\phi _1(y)e^{ix},\phi _2(y)e^{2ix})`$ with $`\phi _1^2+\phi _2^2=1`$ (eq. (5)) and $`\phi _{}^{}{}_{1}{}^{2}+\phi _{}^{}{}_{2}{}^{2}=\phi _1^2+4\phi _2^2=f`$ (eq. (4)) which is impossible since $`\phi _1^2+\phi _2^2=1`$ implies that $`\phi _1`$ and $`\phi _2`$ admit a common critical point. Therefore, $`d=`$ multiplicity of $`\lambda _1(๐,g)=5`$ and there exists $`\rho O(5)`$ such that $`\rho h=(\phi _0(y),\phi _1(y)e^{ix},\phi _2(y)e^{2ix})`$, with $`\phi _0^2+\phi _1^2+\phi _2^2=1`$ and $`\phi _{}^{}{}_{0}{}^{2}+\phi _{}^{}{}_{1}{}^{2}+\phi _{}^{}{}_{2}{}^{2}=\phi _1^2+4\phi _2^2=f`$. Since the linear components of $`\rho h`$ are first eigenfunctions of $`(๐,g)`$, one should has, $`k=0,1,2`$, $`\phi _k^{\prime \prime }=(k^2\lambda _1(๐,g)f)\phi _k=(k^22\phi _1^28\phi _2^2)\phi _k`$. Now, it is immediate to check that one of the couples of functions $`(\pm \phi _1,\pm \phi _2)`$ satisfies the Conditions (a), $`\mathrm{}`$, (d) of the statement. Indeed, the parity condition $`\phi _k(y)=(1)^k\phi _k(y)`$ implies that $`\phi _1(0)=\phi _0^{}(0)=\phi _2^{}(0)=0`$ and, then, $`\phi _{}^{}{}_{1}{}^{2}(0)=4\phi _2^2(0)`$. Conditions (c) and (d) follow from the fact that a first eigenfunction has exactly two nodal domains in $`๐`$.
Conversely, let $`\phi _1`$ and $`\phi _2`$ be two periodic functions of period $`a`$ satisfying Conditions (a), $`\mathrm{}`$, (d) and let us show that the metric $`g=f(y)g_a`$, where $`f=\phi _1^2+4\phi _2^2`$, is a critical metric of $`\lambda _1`$ under area preserving deformations. For this, we set $`\phi _0=\sqrt{1\phi _1^2\phi _2^2}`$ and define the map $`h:๐๐^4`$ by $`h=(\phi _0(y),\phi _1(y)e^{ix},\phi _2(y)e^{2ix})`$. It suffices to check that the components of $`h`$ are first eigenfunctions of $`\mathrm{\Delta }_g`$ satisfying (4).
Indeed, in the next section we will see that the second order differential system satisfied by $`\phi _1`$ and $`\phi _2`$ (Condition (a)) admits the two following first integrals:
(6)
$$\{\begin{array}{c}(\phi _1^2+4\phi _2^2)^2\phi _1^216\phi _2^2+\phi _{1}^{}{}_{}{}^{2}+4\phi _{2}^{}{}_{}{}^{2}=C,\hfill \\ \\ 12\phi _2^2(\phi _2^21)+3\phi _1^2\phi _2^2+\phi _2^2\phi _{1}^{}{}_{}{}^{2}2\phi _1\phi _1^{}\phi _2\phi _2^{}+(3+\phi _1^2)\phi _{2}^{}{}_{}{}^{2}=C,\hfill \end{array}$$
with $`C=4\phi _2(0)^2(4\phi _2(0)^23)`$ (note that Condition (b) implies that $`\phi _1(0)=\phi _2^{}(0)=0`$). Differentiating $`\phi _0^2+\phi _1^2+\phi _2^2=1`$ and using the second equation in (6), we get
$`\phi _0^2\phi _{0}^{}{}_{}{}^{2}`$ $`=`$ $`\phi _1^2\phi _{1}^{}{}_{}{}^{2}+\phi _2^2\phi _{2}^{}{}_{}{}^{2}+2\phi _1\phi _1^{}\phi _2\phi _2^{}`$
$`=`$ $`\phi _1^2\phi _{1}^{}{}_{}{}^{2}+\phi _2^2\phi _{2}^{}{}_{}{}^{2}+12\phi _2^2(\phi _2^21)+3\phi _1^2\phi _2^2+\phi _2^2\phi _{1}^{}{}_{}{}^{2}+(3+\phi _1^2)\phi _{2}^{}{}_{}{}^{2}C`$
$`=`$ $`(\phi _1^2+\phi _2^2)\phi _{1}^{}{}_{}{}^{2}+(3+\phi _1^2+\phi _2^2)\phi _{2}^{}{}_{}{}^{2}+12\phi _2^2(\phi _2^21)+3\phi _1^2\phi _2^2C`$
$`=`$ $`(1\phi _0^2)\phi _{1}^{}{}_{}{}^{2}+(4\phi _0^2)\phi _{2}^{}{}_{}{}^{2}+12\phi _2^2(\phi _2^21)+3\phi _1^2\phi _2^2C.`$
Therefore
$`\phi _0^2\left(\phi _{0}^{}{}_{}{}^{2}+\phi _{1}^{}{}_{}{}^{2}+\phi _{2}^{}{}_{}{}^{2}\right)`$ $`=`$ $`\phi _{1}^{}{}_{}{}^{2}+4\phi _{2}^{}{}_{}{}^{2}+12\phi _2^2(\phi _2^21)+3\phi _1^2\phi _2^2C`$
$`=`$ $`\left(1\phi _1^2\phi _2^2\right)\left(\phi _1^2+4\phi _2^2\right),`$
where the last equality follows from the first equation of (6). Hence,
$$|_yh|^2=\phi _{0}^{}{}_{}{}^{2}+\phi _{1}^{}{}_{}{}^{2}+\phi _{2}^{}{}_{}{}^{2}=\phi _1^2+4\phi _2^2=|_xh|^2$$
and, since $`_xh`$ and $`_yh`$ are orthogonal, the map $`h`$ is isometric, which means that Equation (4) is satisfied.
From Condition (a) one has $`\phi _1^{\prime \prime }=(12f)\phi _1`$ and $`\phi _2^{\prime \prime }=(42f)\phi _2`$, which implies that the functions $`h_1=\phi _1(y)\mathrm{cos}x`$, $`h_2=\phi _1(y)\mathrm{sin}x`$, $`h_3=\phi _2(y)\mathrm{cos}2x`$ and $`h_4=\phi _2(y)\mathrm{sin}2x`$ are eigenfunctions of $`\mathrm{\Delta }_g`$ associated with the eigenvalue $`\lambda =2`$. Moreover, differentiating twice the identity $`\phi _0^2+\phi _1^2+\phi _2^2=1`$ and using Condition (a) and the identity $`\phi _{0}^{}{}_{}{}^{2}+\phi _{1}^{}{}_{}{}^{2}+\phi _{2}^{}{}_{}{}^{2}=\phi _1^2+4\phi _2^2=f`$, one obtains after an elementary computation, $`\phi _0^{\prime \prime }=2f\phi _0`$. Hence, all the components of $`h`$ are eigenfunctions of $`\mathrm{\Delta }_g`$ associated with the eigenvalue $`\lambda =2`$. It remains to prove that $`2`$ is the first positive eigenvalue of $`\mathrm{\Delta }_g`$ or, equivalently, for each $`k=0,1,2`$, the function $`\phi _k`$ corresponds to the lowest positive eigenvalue of the Sturm-Liouville problem $`\phi ^{\prime \prime }=(k^2\lambda f)\phi `$ subject to the parity condition $`\phi (y)=(1)^k\phi (y)`$. As explained in the proof of Proposition 3.4.1 of , this follows from conditions (c) and (d) giving the number of zeros of $`\phi _k`$, and the special properties of the zero sets of solutions of Sturm-Liouville equations (oscillation theorems of Haupt and Sturm). โ
## 3. Study of the dynamical system: proof of results
According to Proposition 2.1, one needs to deal with the following system of second order differential equations (Condition (a) of Prop. 2.1)
(7)
$$\{\begin{array}{c}\phi _1^{\prime \prime }=(12\phi _1^28\phi _2^2)\phi _1,\hfill \\ \phi _2^{\prime \prime }=(42\phi _1^28\phi _2^2)\phi _2,\hfill \end{array}$$
subject to the initial conditions (Condition (b) of Prop. 2.1)
(8)
$$\{\begin{array}{c}\phi _1(0)=0,\phi _2(0)=p,\hfill \\ \phi _1^{}(0)=2p,\phi _2^{}(0)=0,\hfill \end{array}$$
where $`p(0,1]`$ (Condition (d) of Prop. 2.1).
Notice that the system (7)-(8) is invariant under the transform
$$(\phi _1(y),\phi _2(y))(\phi _1(y),\phi _2(y)).$$
Consequently, the solution $`(\phi _1,\phi _2)`$ of (7)-(8) is such that $`\phi _1`$ is odd and $`\phi _2`$ is even.
We are looking for periodic solutions satisfying the following condition (Condition (c) of Prop. 2.1):
(9)
$$\{\begin{array}{c}\phi _1\text{ has exactly two zeros in a period,}\hfill \\ \phi _2\text{ is positive everywhere.}\hfill \end{array}$$
Our aim is to prove the following
###### Theorem 3.1.
There exists only one periodic solution of (7)-(8) satisfying Condition (9). It corresponds to the initial value $`\phi _2(0)=p=\sqrt{3/8}`$.
In fact, this theorem follows from the qualitative behavior of solutions, in terms of $`p`$, given in the following
###### Proposition 3.1.
Let $`(\phi _1,\phi _2)`$ be the solution of (7)-(8).
1. For all $`p(0,1]`$, $`p\sqrt{3}/2`$, $`(\phi _1,\phi _2)`$ is periodic or quasi-periodic.
2. For $`p=\frac{\sqrt{3}}{2}`$, $`(\phi _1,\phi _2)`$ tends to the origin as $`y\mathrm{}`$ (hence, it is not periodic).
3. For all $`p(\sqrt{3}/2,1]`$, $`\phi _2`$ vanishes at least once in each period (of $`\phi _2`$). Hence, Condition (9) is not satisfied.
4. There exists a countable dense subset $`๐ซ(0,\sqrt{3}/2)`$, with $`\sqrt{3/8}๐ซ`$, such that the solution $`(\phi _1,\phi _2)`$ corresponding to $`p(0,\sqrt{3}/2)`$ is periodic if and only if $`p๐ซ`$.
5. For $`p=\sqrt{3/8}`$, $`(\phi _1,\phi _2)`$ satisfies (9) and, for any $`p๐ซ`$, $`p\sqrt{3/8}`$, $`\phi _1`$ admits at least 6 zeros in a period.
The first fundamental step in the study of the system above is the existence of the following two independent first integrals.
### 3.1. First integrals
The functions
(10)
$$\{\begin{array}{c}H_1(\phi _1,\phi _2,\phi _1^{},\phi _2^{}):=(\phi _1^2+4\phi _2^2)^2\phi _1^216\phi _2^2+(\phi _1^{})^2+4(\phi _2^{})^2,\hfill \\ \\ H_2(\phi _1,\phi _2,\phi _1^{},\phi _2^{}):=12\phi _2^2(\phi _2^21)+3\phi _1^2\phi _2^2+\phi _2^2(\phi _1^{})^2\hfill \\ \\ 2\phi _1\phi _1^{}\phi _2\phi _2^{}+(3+\phi _1^2)(\phi _2^{})^2,\hfill \end{array}$$
are two independent first integrals of (7), i.e. they satisfy the equation
$$\phi _1^{}\frac{H_i}{\phi _1}+\phi _2^{}\frac{H_i}{\phi _2}+\phi _1^{\prime \prime }\frac{H_i}{\phi _1^{}}+\phi _2^{\prime \prime }\frac{H_i}{\phi _2^{}}0.$$
The first one, $`H_1`$, has been obtained by Jakobson et al. . The orbit of a solution of (7) is then contained in an algebraic variety defined by
(11)
$$\{\begin{array}{c}H_1(\phi _1,\phi _2,\phi _1^{},\phi _2^{})=K_1,\hfill \\ \\ H_2(\phi _1,\phi _2,\phi _1^{},\phi _2^{})=K_2,\hfill \end{array}$$
where $`K_1`$ and $`K_2`$ are two constants. Taking into account the initial conditions (8), one has $`K_1=K_2=4p^2(34p^2)`$. In other words, the solution of (7)-(8) is also solution of
(12)
$$\{\begin{array}{c}(\phi _1^2+4\phi _2^2)^2\phi _1^216\phi _2^2+(\phi _1^{})^2+4(\phi _2^{})^2+4p^2(34p^2)=0,\hfill \\ \\ 12\phi _2^2(\phi _2^21)+3\phi _1^2\phi _2^2+\phi _2^2(\phi _1^{})^22\phi _1\phi _1^{}\phi _2\phi _2^{}\hfill \\ \\ +(3+\phi _1^2)(\phi _2^{})^2+4p^2(34p^2)=0,\hfill \end{array}$$
with the initial conditions
(13)
$$\{\begin{array}{c}\phi _1(0)=0,\hfill \\ \\ \phi _2(0)=p.\hfill \end{array}$$
Notice that the parameter $`p`$ appears in both the equations (12) and the initial conditions (13). The system (12) gives rise to a โmulti-valuedโ 2-dimensional dynamical system in the following way.
### 3.2. 2-dimensional dynamical systems
From (12) one can extract explicit expressions of $`\phi _1^{}`$ and $`\phi _2^{}`$ in terms of $`\phi _1`$ and $`\phi _2`$. For instance, eliminating $`\phi _1^{}`$, one obtains the following fourth degree equation in $`\phi _2^{}`$
(14)
$$d_4(\phi _1,\phi _2)(\phi _2^{})^42d_2(\phi _1,\phi _2)(\phi _2^{})^2+d_0(\phi _1,\phi _2)=0,$$
where $`d_0`$, $`d_2`$ and $`d_4`$ are polynomials in $`\phi _1`$, $`\phi _2`$ and $`p`$. The discriminant of (14) is given by
$$\mathrm{\Delta }:=64\phi _1^2\phi _2^2w_1w_2w_3,$$
with
$$\begin{array}{c}w_1(\phi _1,\phi _2)=\phi _1^2+\phi _2^21,\hfill \\ \\ w_2(\phi _1,\phi _2)=p^2\phi _1^2(34p^2)\phi _2^2+p^2(34p^2),\hfill \\ \\ w_3(\phi _1,\phi _2)=(34p^2)\phi _1^2+16p^2\phi _2^24p^2(34p^2).\hfill \end{array}$$
It is quite easy to show that, for any $`p`$, each one of the curves $`(w_i=0)`$ contains the orbit of a particular solution of (12). Moreover, the unit circle $`(w_1=0)`$ represents the orbit of the solution of (12) satisfying the initial conditions (13) with $`p=1`$. For $`p=\sqrt{3/8}`$, we have $`w_34w_2`$ and the curve $`(w_2=0)`$ contains the orbit of the solution of (12)-(13).
These particular algebraic orbits suggest us searching solutions $`(\phi _1,\phi _2)`$ defined by algebraic relations of the form $`w_4(\phi _1,\phi _2)=F(\phi _1^2,\phi _2^2)=0`$, where $`F`$ is a polynomial of degree $`4`$. Apart the three quadrics above, the only additional solution of this type we found is
$$w_4=(\phi _1^2+4\phi _2^2)^212\phi _2^2=0.$$
Like $`(w_1=0)`$, the curve $`(w_4=0)`$ is independent of $`p`$ and represents the orbit of a particular solution of (12) for arbitrary values of $`p`$. Since
$$w_4(\phi _1,\phi _2)=(\phi _1^2+4\phi _2^22\sqrt{3}\phi _2)(\phi _1^2+4\phi _2^2+2\sqrt{3}\phi _2),$$
the set $`(w_4=0)`$ is the union of two ellipses passing through the origin, each one being symmetric to the other with respect to the $`\phi _1`$-axis. The upper ellipse
(15)
$$\phi _1^2+4\phi _2^22\sqrt{3}\phi _2=0$$
corresponds to the orbit of the solution of (12)-(13) associated with $`p=\frac{\sqrt{3}}{2}`$.
### 3.3. Proof of Proposition 3.1(2): case $`๐ฉ=\frac{\sqrt{\mathrm{๐}}}{\mathrm{๐}}`$.
In this case, the orbit of the solution of (12)-(13) is given by (15). The only critical point of (12) lying on this ellipse is the origin, which is also a critical point of the system (7). Therefore, $`(\phi _1(y),\phi _2(y))`$ tends to the origin as $`y`$ goes to infinity, and, hence, it is not periodic (see also ).
*From now on, we will assume that $`p\frac{\sqrt{3}}{2}`$*.
### 3.4. A bounded region for the orbit
The orbit of the solution of (12)-(13) must lie in the region of the $`(\phi _1,\phi _2)`$-plane where the discriminant $`\mathrm{\Delta }`$ of (14) is nonnegative. This region, $`(\mathrm{\Delta }0)`$, is a bounded domain delimited by the unit circle $`(w_1=0)`$ and the quadrics $`(w_2=0)`$ and $`(w_3=0)`$. Its shape depends on the values of $`p`$.
* For $`p(0,\sqrt{3}/2)`$, $`(w_2=0)`$ and $`(w_3=0)`$ are hyperbolae.
* The case $`p=\sqrt{3/8}`$ is a special one since then, $`w_34w_2`$, and the region $`(\mathrm{\Delta }0)`$ shrinks to the arc of the hyperbola $`(w_2=0)`$ lying inside the unit disk.
* For $`p(\sqrt{3}/2,1]`$, $`w_3`$ is positive and $`(w_2=0)`$ is an ellipse.
From (14) and (12) one can express $`\phi _1^{}`$ and $`\phi _2^{}`$ in terms of $`\phi _1`$, $`\phi _2`$ and $`p`$. Thus, we obtain a multi-valued 2-dimensional dynamical system parameterized by $`p`$ with the initial conditions $`\phi _1(0)=0`$ and $`\phi _2(0)=p`$. However, the dynamics of such a multi-valued system is very complex to study. Fortunately, as we will see in the next subsections, the system (7) can be transformed, by means of a suitable change of variables, into a Hamiltonian system, completely integrable by quadratures.
### 3.5. Hamiltonian dynamical system
Let us introduce the new variables $`q_1`$ and $`q_2`$ defined by
(16)
$$q_1:=\frac{1}{\sqrt{2}}\phi _1,q_2:=\sqrt{2}\phi _2.$$
The system (7) becomes
(17)
$$\{\begin{array}{c}q_1^{\prime \prime }=[14(q_1^2+q_2^2)]q_1=\frac{V}{q_1},\hfill \\ \\ q_2^{\prime \prime }=4[1q_1^2q_2^2]q_2=\frac{V}{q_2},\hfill \end{array}$$
with
$$V(q_1,q_2):=(q_1^2+q_2^2)^2\frac{1}{2}q_1^22q_2^2.$$
Therefore, one has a Hamiltonian system with two degrees of freedom. The Hamiltonian $`H`$ is given by
$$H(q_1,q_2,q_1^{},q_2^{}):=\frac{1}{2}[(q_1^{})^2+(q_2^{})^2]+V(q_1,q_2).$$
This Hamiltonian is a first integral of (17) (notice that $`H=\frac{1}{4}H_1`$). A second independent first integral of (17) can be obtained from $`H_2`$. Consequently, the Hamiltonian system (17) is integrable and all its bounded orbits in phase space $`(q_1,q_2,q_1^{},q_2^{})`$ are contained in a 2-dimensional topological torus (see ), which means that the corresponding solutions are periodic or quasi-periodic, provided that there is no critical point in the closure of the orbit. However, it is in general difficult to decide whether such a solution is periodic or not. The corresponding topological torus obtained from (12) is given by:
(18)
$$\{\begin{array}{c}\frac{1}{2}[(q_1^{})^2+(q_2^{})^2]+(q_1^2+q_2^2)^2\frac{1}{2}q_1^22q_2^2+p^2(34p^2)=0,\hfill \\ \\ 3q_2^2(q_2^22)+3q_1^2q_2^2+(q_1^{})^2q_2^22q_1q_1^{}q_2q_2^{}\hfill \\ \\ +\frac{1}{2}(3+2q_1^2)(q_2^{})^2+4p^2(34p^2)=0.\hfill \end{array}$$
It is important to notice that the second first integral is also quadratic in the $`q_1^{}`$ and $`q_2^{}`$ variables. Indeed, this enables us to apply the Bertrand-Darboux-Whittaker theorem: Given a Hamiltonian system defined by
$$H=\frac{1}{2}[(q_1)^2+(q_2^{})^2]+V(q_1,q_2),$$
the system admits an additional independent first integral, quadratic in $`q_1^{}`$ and $`q_2^{}`$, if and only if the system is separable in cartesian, polar, parabolic, or elliptic-hyperbolic coordinates. (see for details).
In our case, an adequate change of variables is a parabolic one, given by
(19)
$$\{\begin{array}{c}q_1^2=\frac{2}{3}uv,\hfill \\ \\ q_2^2=\frac{1}{6}(3+2u)(3+2v).\hfill \end{array}$$
Indeed, from (18), one obtains after an elementary computation
(20)
$$\{\begin{array}{c}(u^{})^2=\frac{P(u)}{(uv)^2},\hfill \\ \\ (v^{})^2=\frac{P(v)}{(uv)^2},\hfill \end{array}$$
where $`P(s):=s(12s)(3+2s)(2p^2+s)(34p^2+2s)`$. Observe that (20) is not completely decoupled yet; this can be done by means of a suitable change of the independent variable (see Subsection 3.6). Each one of the quadrics $`(w_1=0)`$, $`(w_2=0)`$ and $`(w_3=0)`$ is transformed into two parallel lines. Indeed, we have $`w_1(u,v)=\frac{1}{4}(12u)(12v)`$, $`w_2(u,v)=(\frac{3}{2}2p^2+u)(\frac{3}{2}2p^2+v)`$ and $`w_3(u,v)=4(2p^2+u)(2p^2+v)`$. Also, we have $`\mathrm{\Delta }=\frac{16}{9}P(u)P(v)`$. Thus, the region $`(\mathrm{\Delta }0)`$ is transformed into the region $`(P(u)P(v)0)`$.
Observe that the system (20) is symmetric in $`u`$ and $`v`$. As the change of variables (19) is also symmetric in $`u`$ and $`v`$, and since $`uv`$ must be non positive, one can assume, without loss of generality, that $`u0`$ and, hence, $`\frac{3}{2}v0`$. Now, the condition $`(\mathrm{\Delta }0)`$ implies that $`(u,v)I_1\times I_2`$, where
(21)
$$I_1:=[\alpha _0,1/2]:=\{\begin{array}{cc}[0,\frac{1}{2}]\hfill & \text{if }p^2<\frac{3}{4},\hfill \\ & \\ [2p^2\frac{3}{2},\frac{1}{2}]\hfill & \text{if }\frac{3}{4}<p^21,\hfill \end{array}$$
and
(22)
$$I_2:=[a_0,a_1]:=\{\begin{array}{cc}[2p^2\frac{3}{2},2p^2]\hfill & \text{if }p^2\frac{3}{8},\hfill \\ & \\ [2p^2,2p^2\frac{3}{2}]\hfill & \text{if }\frac{3}{8}p^2<\frac{3}{4},\hfill \\ & \\ [\frac{3}{2},0]\hfill & \text{for }\frac{3}{4}<p^21.\hfill \end{array}$$
The initial conditions (13) become
$$(u(0),v(0))=\{\begin{array}{cc}(0,2p^2\frac{3}{2})\hfill & \text{if }p^2<\frac{3}{4},\hfill \\ & \\ (2p^2\frac{3}{2},0)\hfill & \text{if }\frac{3}{4}<p^21.\hfill \end{array}$$
In all cases, $`u(0)`$ and $`v(0)`$ are zeros of $`P`$. Hence
$$(u^{}(0),v^{}(0))=(0,0).$$
The behavior of $`(u,v)`$ near $`y=0`$ is then determined by the acceleration vector
$$(u^{\prime \prime }(0),v^{\prime \prime }(0))=\{\begin{array}{cc}(\frac{12p^2}{34p^2},\frac{16p^2(1p^2)(38p^2)}{(34p^2)})\hfill & \text{if }p^2<\frac{3}{4},\hfill \\ & \\ (\frac{16p^2(1p^2)(38p^2)}{(34p^2)},\frac{12p^2}{34p^2})\hfill & \text{if }\frac{3}{4}<p^21.\hfill \end{array}$$
Notice that for $`p=\sqrt{\frac{3}{8}}`$, $`v`$ is constant, namely $`v(y)=\frac{3}{4}`$ for all $`y`$, while for $`p=1`$, $`u`$ is constant with $`u(y)=\frac{1}{2}`$ for all $`y`$.
### 3.6. Proof of Proposition 3.1(1): Decoupling the system
In order to completely decouple the previous system we introduce a change of the independent variable $`y\tau `$ defined by:
$$\frac{d\tau }{dy}=\frac{1}{uv}.$$
Notice that this change of variable is one-to-one since $`uv0`$ (indeed, $`I_1I_2=\mathrm{}`$). In this new variable, the system splits into two independent equations:
(23)
$$(\dot{u})^2=P(u),$$
(24)
$$(\dot{v})^2=P(v),$$
where $`\dot{u}:=du/d\tau `$ and $`\dot{v}:=dv/d\tau `$. The solution $`\tau u(\tau )`$ of (23) is also a solution of the second order ODE
(25)
$$\ddot{u}=\frac{1}{2}P^{}(u),$$
with the initial conditions $`u(0)=\alpha _0`$ (see (21) for the definition of $`\alpha _0`$) and $`\dot{u}(0)=0`$, where $`P^{}:=dP/du`$. This solution lies on the curve
(26)
$$(\dot{u})^2P(u)=0$$
in the $`(u,\dot{u})`$-phase plane of (25). Since $`\alpha _0`$ and $`\frac{1}{2}`$ are two consecutive zeros of $`P`$, the equation (26) in the region $`\alpha _0u\frac{1}{2}`$ represents a closed curve. On the other hand, it is easy to check that $`P`$ and $`P^{}`$ admit no common zero in the interval $`[\alpha _0,\frac{1}{2}]`$. Hence, there exists no critical point for (25) on the orbit defined by (26) inside the region $`\alpha _0u\frac{1}{2}`$. Consequently, this closed orbit corresponds to a periodic solution of (25) and, therefore, $`\tau u(\tau )`$ oscillates between $`\alpha _0`$ and $`\frac{1}{2}`$.
A similar analysis for $`\tau v(\tau )`$ implies that it is a periodic solution of
(27)
$$\ddot{v}=\frac{1}{2}P^{}(v),$$
with the initial conditions $`v(0)=a_0`$ (see (22) for the definition of $`a_0`$) and $`\dot{v}(0)=0`$. Consequently, $`\tau v(\tau )`$ oscillates between $`a_0`$ and $`a_1`$. This proves Assertion (1) of Proposition 3.1(1).
### 3.7. Proof of Proposition 3.1(3): case $`๐ฉ>\frac{\sqrt{\mathrm{๐}}}{\mathrm{๐}}`$
We have just seen that $`v()=[a_0,a_1]`$, with $`a_0=\frac{3}{2}`$ for $`p(\frac{\sqrt{3}}{2},1]`$ (see (22)). This implies that $`q_2`$, and then $`\phi _2`$, vanishes at least once in a period (see (19)).
### 3.8. About the periods of $`u`$ and $`v`$: case $`๐ฉ<\frac{\sqrt{\mathrm{๐}}}{\mathrm{๐}}`$
Let us denote $`๐ฏ_u(p)`$ the period of $`u`$. The function $`\tau u(\tau )`$ oscillates between $`\alpha _0=0`$ and $`\frac{1}{2}`$ with velocity $`(\dot{u})^2=P(u)0`$ if $`u(0,\frac{1}{2})`$. Hence, $`u(\tau )`$ increases from $`0`$ to $`\frac{1}{2}`$ when $`\tau `$ goes from 0 to $`\frac{๐ฏ_u(p)}{2}`$. It follows that
(28)
$$๐ฏ_u(p)=2_0^{\frac{1}{2}}\frac{ds}{\sqrt{P(s)}}.$$
Similarly, for $`p\sqrt{3/8}`$, the period $`๐ฏ_v(p)`$ of $`\tau v(\tau )`$ is given by
$$๐ฏ_v(p)=2_{a_0}^{a_1}\frac{ds}{\sqrt{P(s)}}.$$
Setting, for $`p\sqrt{3/8}`$, $`s=(38p^2)r\frac{3}{2}+2p^2`$, one can write
(29)
$$๐ฏ_v(p)=2_0^{\frac{1}{2}}\frac{dr}{\sqrt{Q(r)}},$$
where
$$Q(r):=2r(12r)[2p^2+(38p^2)r][22p^2(38p^2)r][34p^22(38p^2)r].$$
Hence, the functions $`๐ฏ_u(p)`$ and $`๐ฏ_v(p)`$ are explicitly given by complete hyper-elliptic integrals. Although the function $`๐ฏ_v`$ is not defined at $`p=\sqrt{3/8}`$, its limit exists. Indeed, setting $`t=\frac{3}{4}+r`$ and $`\alpha :=\frac{3}{4}2p^2`$, we get
$$๐ฏ_v(p)=2_\alpha ^\alpha \frac{dt}{\sqrt{(\frac{3}{2}2t)(\frac{5}{2}2t)(\frac{3}{2}+2t)}\sqrt{\alpha ^2t^2}}.$$
As $`\alpha 0`$, we have
$$๐ฏ_v(p)2_\alpha ^\alpha \frac{4dt}{3\sqrt{10}\sqrt{\alpha ^2t^2}}.$$
Thus
$$\underset{p\sqrt{\frac{3}{8}}}{lim}๐ฏ_v(p)=\frac{8\pi }{3\sqrt{10}}.$$
On the other hand, we have
$$๐ฏ_u(\sqrt{3/8})=\frac{4}{5}\mathrm{\Pi }(2/5,1/4),$$
where $`\mathrm{\Pi }`$ is the complete elliptic integral of the third kind given by
$$\mathrm{\Pi }(n,m):=_0^{\frac{\pi }{2}}\frac{d\theta }{(1n\mathrm{sin}^2\theta )\sqrt{1m\mathrm{sin}^2\theta }}.$$
Since for $`p=\sqrt{3/8}`$, $`v`$ is constant, the couple $`(u,v)`$ is periodic of period $`๐ฏ_u(\sqrt{3/8})`$.
Behavior of $`๐ฏ_v๐ฏ_u`$ and $`๐ฏ_v/๐ฏ_u`$ near $`p=0`$: One has
$$๐ฏ_v(p)๐ฏ_u(p)=2_0^{\frac{1}{2}}\left[\frac{1}{\sqrt{Q(s)}}\frac{1}{\sqrt{P(s)}}\right]๐s.$$
The integral of $`\frac{1}{\sqrt{P(s)}}`$ is singular only at $`p=0`$. A direct computation gives
$$_0^{\frac{1}{2}}\frac{ds}{\sqrt{P(s)}}_0^{\frac{1}{2}}\frac{ds}{3\sqrt{s^2+2p^2s}}\frac{2}{3}\mathrm{ln}(p).$$
Similarly, we get
$$_0^{\frac{1}{2}}\frac{ds}{\sqrt{Q(s)}}\mathrm{ln}p.$$
In other words $`๐ฏ_v(p)๐ฏ_u(p)+\mathrm{}`$ as $`p0`$ while the ratio $`๐ฏ_v(p)/๐ฏ_u(p)`$ goes to $`\frac{3}{2}`$ (see the figure below).
Behavior of $`๐ฏ_v๐ฏ_u`$ and $`๐ฏ_v/๐ฏ_u`$ near $`p=\sqrt{3}/2`$: As $`p\sqrt{3}/2`$, $`๐ฏ_u(p)\frac{2}{3}\mathrm{ln}(\sqrt{3}/2p)`$ and $`๐ฏ_v(p)\mathrm{ln}(\sqrt{3}/2p)`$. Hence, $`๐ฏ_v๐ฏ_u+\mathrm{}`$ as $`p\sqrt{3}/2`$ and $`๐ฏ_v(p)/๐ฏ_u(p)`$ goes again to $`\frac{3}{2}`$.
Thus, $`๐ฏ_v(p)>๐ฏ_u(p)`$ near $`p=0`$ and $`p=\sqrt{3}/2`$. Actually, one has $`๐ฏ_v(p)>๐ฏ_u(p)`$ for all $`p(0,\sqrt{3}/2)`$ as shown by the graphic representation of $`๐ฏ_v`$ and $`๐ฏ_u`$ given below.
The functions $`p๐ฏ_v(p)`$ (the upper one) and $`p๐ฏ_u(p)`$.
### 3.9. Proof of Proposition 3.1(4).
The couple $`(u,v)`$ is periodic if and only if the ratio $`R(p):=\frac{๐ฏ_v(p)}{๐ฏ_u(p)}`$ is a rational number. From the previous subsection, $`R`$ is a nonconstant continuous function on $`(0,\sqrt{3}/2)`$ with $`lim_{p0}R(p)=lim_{p\sqrt{3}/2}R(p)=3/2`$. The range of $`R`$ is a closed interval $`[r_1,r_2][1.480473,1.507784]`$ (see the figure below).
The ratio $`\frac{๐ฏ_v}{๐ฏ_u}`$
To end the proof of Assertion (4), we only need to define $`๐ซ`$ to be the set of $`p(0,\sqrt{3}/2)`$ such that $`R(p)`$ is a rational number.
### 3.10. Proof of Proposition 3.1(5).
In the case $`p=\sqrt{3/8}`$ we have, $`y`$, $`v(y)=\frac{3}{4}`$ and, then, $`\phi _1^2=u`$ and $`\phi _2^2=\frac{1}{8}(3+2u)`$. The couple of periodic functions $`(\phi _1,\phi _2)=(\sqrt{u},\sqrt{\frac{1}{8}(3+2u)})`$ on $`[0,๐ฏ_u(\sqrt{3/8})]`$, such that $`\phi _1`$ is odd and $`\phi _2`$ is even, solves the original system and satisfies Condition (9).
Let $`p๐ซ`$, $`p\sqrt{3/8}`$, and let $`\frac{q}{m}`$ be an irreducible fraction, with $`q`$, $`m`$, such that $`R(p)=\frac{๐ฏ_v(p)}{๐ฏ_u(p)}=\frac{q}{m}`$. The period of the couple $`(u,v)`$ is given by
$$๐ฏ(p)=q๐ฏ_u(p)=m๐ฏ_v(p).$$
The number of zeros of $`u`$ in a period, for instance $`[0,๐ฏ(p))`$, of $`(u,v)`$ is equal to $`q`$ times the number of zeros of $`u`$ in $`[0,๐ฏ_u(p))`$. As we saw above, $`p(0,\sqrt{3}/2)`$, one has $`1<R(p)<2`$. Hence, $`m2`$ and $`q>m`$, which implies $`q3`$. Since $`u(0)=0`$, the number of zeros of $`u`$ in a period of $`(u,v)`$ is at least 3. Since $`\phi _1`$ is odd, the period of $`(\phi _1,\phi _2)`$ is twice the period of $`(u,v)`$. From $`\phi _1=2\sqrt{\frac{1}{3}uv}`$ on $`[0,๐ฏ(p))`$, one deduces that $`\phi _1`$ admits at least 6 zeros in a period of $`(\phi _1,\phi _2)`$. Notice that the case $`p=\sqrt{3/8}`$ is special since, for this value of $`p`$, $`v`$ is constant and the couple $`(u,v)`$ is periodic whose period is equal to that of $`u`$.
### 3.11. On the shape of solutions
Although it is not necessary for the proof of our main result, one can obtain as a by product of our study, some properties concerning the shape of solutions. First, notice that, for $`p(0,\frac{\sqrt{3}}{2})`$, the 2-dimensional dynamical system admits four critical points in the region $`(\mathrm{\Delta }0)(\phi _20)`$: $`A=(2p/\sqrt{3},\sqrt{14p^2/3})`$, $`B=(\sqrt{14p^2/3},2p/\sqrt{3})`$ and their symmetric with respect to the $`\phi _2`$-axis that we denote $`A^{}`$ and $`B^{}`$. Notice that these critical points are on the boundary of the region $`(\mathrm{\Delta }0)`$.
Non-periodic solutions. They correspond to the case where $`๐ฏ_v(p)/๐ฏ_u(p)`$ is irrational. In this case, the orbit $`(u,v)`$ fill the rectangle $`I_1\times I_2`$, and, then, in the $`(\phi _1,\phi _2)`$-plane, the orbit fill the region $`(\mathrm{\Delta }0)`$.
Periodic solutions. For $`p=\sqrt{3/8}`$, the solution $`(\phi _1,\phi _2)`$ lies on the hyperbola of equation $`\phi _1^24\phi _2^2+3/2=0,`$ oscillating between the points $`A=(\frac{1}{\sqrt{2}},\frac{1}{\sqrt{2}})`$ and $`A^{}=(\frac{1}{\sqrt{2}},\frac{1}{\sqrt{2}})`$. For $`p\sqrt{3/8}`$, let $`q/m=๐ฏ_v(p)/๐ฏ_u(p)`$ be an irreducible fraction, with $`q,m`$, and set $`๐ฏ:=q๐ฏ_u(p)=m๐ฏ_v(p)`$. Geometrically, this means that $`u`$ makes $`q`$ round trips in a period while $`v`$ makes $`m`$ round trips. We distinguish three cases:
* If $`q`$ and $`m`$ are both odd, then $`u(๐ฏ/2)=\frac{1}{2}`$ and $`v(๐ฏ/2)=a_1`$. This corresponds to the point $`A`$ for $`p^2<\frac{3}{8}`$ and to the point $`B`$ for $`\frac{3}{8}<p^2<\frac{3}{4}`$. The orbit is not closed and $`(\phi _1,\phi _2)`$ oscillates between $`A`$ and $`A^{}`$ or $`B`$ and $`B^{}`$.
* If $`q`$ is odd and $`m`$ is even, then $`u(๐ฏ/2)=\frac{1}{2}`$ and $`v(๐ฏ/2)=a_0`$. This corresponds to the point $`B`$ for $`p^2<\frac{3}{8}`$ and to the point $`A`$ for $`\frac{3}{8}<p^2<\frac{3}{4}`$. Again, $`(\phi _1,\phi _2)`$ oscillates between $`A`$ and $`A^{}`$ or $`B`$ and $`B^{}`$.
* If $`q`$ is even and $`m`$ is odd, then $`u(๐ฏ/2)=0`$ and $`v(๐ฏ/2)=a_1`$. This corresponds in the $`(\phi _1,\phi _2)`$-plane to the point $`(0,\frac{3}{4}p^2)`$. This point is the intersection between the quadric $`(w_3=0)`$ with the $`\phi _2`$-axis. In this case the orbit is closed.
|
warning/0506/math0506348.html
|
ar5iv
|
text
|
# On compactifications of the Steinberg zero-fiber
## 1. Introduction
Let $`G`$ be a connected semisimple linear algebraic group over an algebraically closed field $`k`$ of positive characteristic. The set of elements in $`G`$ with semisimple part within a fixed $`G`$-conjugacy class is called a Steinberg fiber. Examples of Steinberg fibers include the unipotent variety and the conjugacy class of a regular semisimple element. Lately there has been some interest in describing the closure of Steinberg fibers within equivariant embeddings of the group $`G`$ (see \[He\], \[H-T\], \[Spr\], \[T\]).
In this paper we study the closure of a distinguished Steinberg fiber $`N`$ called the Steinberg zero-fiber (see Section 3 for the precise definition of $`N`$). We will prove that the closure $`\overline{N}`$ of $`N`$ within any equivariant embedding $`X`$ of $`G`$ will be normal and Cohen-Macaulay. Moreover, when $`X`$ is smooth we will prove that $`\overline{N}`$ is a local complete intersection. These results will all be proved by Frobenius splitting techniques. As a byproduct we find that $`\overline{N}`$ has a canonical Frobenius splitting and hence the set of global sections of any $`G`$-linearized line bundle on $`\overline{N}`$ will admit a good filtration.
The presentation in this paper is close to \[T\], but the setup is somehow opposite. More precisely, in loc.cit. the group $`G`$ was fixed to be of simply connected type while the Steinberg fiber was arbitrary; in the present paper the Steinberg fiber is fixed but the semisimple group $`G`$ is of arbitrary type.
It is worth noticing (see \[T\]) that for a fixed equivariant embedding $`X`$ of $`G`$ the boundary $`\overline{F}F`$ of the closure of a Steinberg fiber $`F`$ in $`G`$ is independent of $`F`$. This shows that the results obtained in this paper for the rather special Steinberg zero-fiber will provide some knowledge about closures of Steinberg fibers in general. E.g. the boundary $`\overline{F}F`$ will always be Frobenius split (see Theorem 8.1). This suggests, that the results in this paper may be generalized to arbitrary Steinberg fibers. However, we give an example (see Example 8.2) showing that this is not always the case.
## 2. Notation
Let $`G`$ denote a connected semisimple linear algebraic group over an algebraically closed field $`k`$. The associated groups of simply connected and adjoint type will be denoted by $`G_{\mathrm{sc}}`$ and $`G_{\mathrm{ad}}`$ respectively. Let $`T`$ denote a maximal torus in $`G`$ and let $`B`$ denote a Borel subgroup of $`G`$ containing $`T`$. The associated maximal torus and Borel subgroup in $`G_{\mathrm{ad}}`$ (resp. $`G_{\mathrm{sc}}`$) will be denoted by $`T_{\mathrm{ad}}`$ and $`B_{\mathrm{ad}}`$ (resp. $`T_{\mathrm{sc}}`$ and $`B_{\mathrm{sc}}`$).
The set of roots associated to $`T`$ is denoted by $`R`$. We define the set of positive roots $`R^+`$ to be the nonzero $`T`$-weights of the Lie algebra of $`B`$. The set of simple roots $`(\alpha _i)_{iI}`$ will be indexed by $`I`$ and will have cardinality $`l`$. To each simple root $`\alpha _i`$ we let $`s_i`$ denote the associated simple reflection in the Weyl group $`W`$ defined by $`T`$. The length of an element $`w`$ in $`W`$ is defined as the number of simple reflections in a reduced expression of $`w`$. The unique element in $`W`$ of maximal length will be denoted by $`w_0`$.
The weight lattice $`\mathrm{\Lambda }(R)`$ of the root system associated to $`G`$ is identified with the character group $`X^{}(T_{\mathrm{sc}})`$ of $`T_{\mathrm{sc}}`$ and contains the set of $`T`$-characters $`X^{}(T)`$. We let $`\alpha _i^{}`$, $`iI`$, be the set of simple coroots and let $`,`$ denote the pairing between coweights and weights in the root system $`R`$. A weight $`\lambda `$ is then dominant if $`\lambda ,\alpha _i^{}`$ for all $`iI`$. The fundamental dominant weight associated to $`\alpha _i`$, $`iI`$, will be denote by $`\omega _i`$.
For a dominant weight $`\lambda \mathrm{\Lambda }(R)`$ we let $`\mathrm{H}(\lambda )`$ denote the dual Weyl $`G_{\mathrm{sc}}`$-module with heighest weight $`\lambda `$, i.e. containing a $`B`$-semiinvariant element $`v_\lambda ^+`$ of weight $`\lambda `$. The Picard group of $`G/B`$ may be identified with the weight lattice $`\mathrm{\Lambda }(R)`$ and we let $`(\lambda )`$ denote the line bundle associated to $`\lambda \mathrm{\Lambda }(R)`$. The line bundle $`(\lambda )`$ has a unique $`G_{\mathrm{sc}}`$-linearization so we may regard the set of global sections of $`(\lambda )`$ as a $`G_{\mathrm{sc}}`$-module. We assume that the notation is chosen such that the set of global sections of $`(\lambda )`$ is isomorphic to $`\mathrm{H}(w_0\lambda )`$. For $`\lambda ,\mu \mathrm{\Lambda }(R)`$ we denote by $`(\lambda ,\mu )`$ the line bundle $`(\lambda )(\mu )`$ on the variety $`G/B\times G/B`$.
## 3. Steinberg fibers
The set of elements $`g`$ in $`G`$ with semisimple part $`g_s`$ in a fixed $`G`$-conjugacy class is called a Steinberg fiber. Any Steinberg fiber is a closed irreducible subset of $`G`$ of codimension $`l`$ (see \[St, Thm.6.11\]). Examples of Steinberg fibers include the conjugacy classes of regular semisimple elements and the unipotent variety; i.e. the set of elements with $`g_s`$ equal to the identity element $`e`$.
When $`G=G_{\mathrm{sc}}`$ is simply connected the Steinberg fibers may also be described as genuine fibers of a morphism $`G_{\mathrm{sc}}k^l`$. Here the $`i`$-th coordinate map is given by the $`G_{\mathrm{sc}}`$-character of the representation $`\mathrm{H}(\omega _i)`$. In this formulation the fiber $`N_{\mathrm{sc}}`$ above $`(0,0,\mathrm{},0)`$ is called the Steinberg zero-fiber. When $`G`$ is arbitrary we define the Steinberg zero-fiber $`N`$ of $`G`$ to be the image of $`N_{\mathrm{sc}}`$ under the natural morphism $`\pi :G_{\mathrm{sc}}G`$.
The structure of the Steinberg zero-fiber is very dependent on the characteristic of $`k`$. In most cases $`N`$ is just the conjugacy class of a regular semisimple element. At the other extreme we can have that $`N`$ coincides with the unipotent variety of $`G`$.
###### Remark 3.1.
In the following cases the Steinberg zero-fiber and the unipotent variety of $`G`$ coincide :
Type $`๐ _n`$: when $`n=p^m1`$ ($`m`$) and $`p=\mathrm{char}(k)>0`$.
Type $`๐ข_n`$: when $`n=2^m1`$ ($`m`$) and $`\mathrm{char}(k)=2`$.
Type $`๐ฃ_n`$: when $`n=2^m`$ ($`m`$) and $`\mathrm{char}(k)=2`$.
Type $`๐ค_6`$: when $`\mathrm{char}(k)=3`$.
Type $`๐ค_8`$: when $`\mathrm{char}(k)=31`$.
Type $`๐ฅ_4`$: when $`\mathrm{char}(k)=13`$.
Type $`๐ฆ_2`$: when $`\mathrm{char}(k)=7`$.
## 4. Equivariant Embeddings
An *equivariant embedding* of $`G`$ is a normal $`G\times G`$-variety containing a $`G\times G`$-invariant open dense subset isomorphic to $`G`$ in such a way that the induced $`G\times G`$-action on $`G`$ is by left and right translation.
### 4.1. The wonderful compactification
When $`G=G_{\mathrm{ad}}`$ is of adjoint type there exists a distinguished equivariant embedding $`๐ฟ`$ of $`G`$ called the wonderful compactification (see e.g. \[DC-S\]). The wonderful compactification of $`G`$ is a smooth projective variety with finitely many $`G\times G`$-obits $`O_J`$ indexed by the subsets $`J`$ of the simple roots $`I`$. We let $`๐ฟ_J`$ denote the closure of $`O_J`$ and assume that the index set is chosen such that $`๐ฟ_J^{}๐ฟ_J=๐ฟ_{J^{}J}`$ for all $`J,J^{}I`$. Then $`๐:=๐ฟ_{\mathrm{}}`$ is the unique closed orbit in $`๐ฟ`$. It is known that $`๐`$ is isomorphic to $`G/B\times G/B`$ as a $`G\times G`$-variety.
To each dominant element $`\lambda `$ in the weight lattice $`\mathrm{\Lambda }(R)`$ we let
$$\rho _\lambda ^{\mathrm{ad}}:G(\mathrm{End}(\mathrm{H}(\lambda ))),$$
denote the $`G\times G`$-equivariant morphism defined by letting $`\rho _\lambda ^{\mathrm{ad}}(e)`$ be the element in $`(\mathrm{End}(\mathrm{H}(\lambda )))`$ represented by the identity map on $`\mathrm{H}(\lambda )`$. Then it is known (see \[DC-S\]) that $`\rho _\lambda ^{\mathrm{ad}}`$ extends to a morphism $`๐ฟ(\mathrm{End}(\mathrm{H}(\lambda )))`$ which we also denote by $`\rho _\lambda ^{\mathrm{ad}}`$.
###### Lemma 4.1.
Let $`v_\lambda ^+`$ (resp. $`u_\lambda ^+`$) denote a nonzero $`B`$-stable element in $`\mathrm{H}(\lambda )`$ (resp. $`\mathrm{H}(\lambda )^{}`$) of weight $`\lambda `$ (resp. $`w_0\lambda `$). Identify $`\mathrm{End}(\mathrm{H}(\lambda ))`$ with $`\mathrm{H}(\lambda )\mathrm{H}(\lambda )^{}`$. Then the restriction of $`\rho _\lambda ^{\mathrm{ad}}`$ to $`๐G/B\times G/B`$ is given by
$$\rho _\lambda ^{\mathrm{ad}}(gB,g^{}B)=(g,g^{})[v_\lambda ^+u_\lambda ^+].$$
###### Proof.
It suffices to prove that the only $`B\times B`$-invariant element of the image of $`\rho _\lambda ^{\mathrm{ad}}`$ is $`[v_\lambda ^+u_\lambda ^+]`$. So let $`x=[f]`$ denote a $`B\times B`$-invariant element of the image of $`\rho _\lambda ^{\mathrm{ad}}`$ represented by an element $`f\mathrm{End}(\mathrm{H}(\lambda ))`$. Then $`f`$ is $`B\times B`$-semiinvariant and thus when writing $`f`$ as an element of $`\mathrm{H}(\lambda )\mathrm{H}(\lambda )^{}`$ it will be equal to $`v_\lambda ^+v`$ for some $`B`$-semiinvariant element $`v`$ of $`\mathrm{H}(\lambda )^{}`$.
Let $`L`$ denote the unique simple submodule of $`\mathrm{H}(\lambda )`$ and let $`M`$ denote the kernel of the associated morphism $`\mathrm{H}(\lambda )^{}L^{}`$. Assume that $`v`$ is contained in $`M`$. Then every $`G\times G`$-translate of $`f`$ is contained in the subset $`LM`$ consisting of nilpotent endomorphisms. Now we apply \[B-K, Lemma.6.1.4\]. It follows that the closure $`C`$ of the $`T\times T`$-orbit through $`\rho _\lambda ^{\mathrm{ad}}(e)`$ will contain an element represented by a nilpotent endomorphism. But clearly every element in $`C`$ will be represented by a semisimple endomorphism of $`\mathrm{H}(\lambda )`$. This is a contradiction.
As a consequence $`v`$ is not contained in $`M`$ and therefore its image in $`L^{}`$ will be a nonzero $`B`$-semiinvariant vector. As a consequence $`v`$ must be a nonzero multiple of $`u_\lambda ^+`$. โ
### 4.2. Toroidal embeddings
Now let $`G`$ be an arbitrary connected semisimple group. An equivariant embedding $`X`$ of $`G`$ is called toroidal if the natural map $`\pi _{\mathrm{ad}}:GG_{\mathrm{ad}}`$ extends to a morphism $`X๐ฟ`$. In the present paper toroidal embeddings will play a central role. This is due to the following fact (see \[Rit, Prop.3\])
###### Theorem 4.2.
Let $`X`$ be an arbitrary equivariant embedding of $`G`$. Then there exists a smooth toroidal embedding $`X^{}`$ of $`G`$ and a birational projective morphism $`X^{}X`$ extending the identity map on $`G`$.
An important property of toroidal embeddings is that for each dominant weight $`\lambda `$ there exists a $`G\times G`$-equivariant morphism
$$\rho _\lambda :X(\mathrm{End}(\mathrm{H}(\lambda )))$$
induced by $`\rho _\lambda ^{\mathrm{ad}}`$. When $`X`$ is a complete toroidal embedding of $`G`$ and $`Y`$ is a closed $`G\times G`$-orbit in $`X`$ we may describe the restricted mapping $`Y(\mathrm{End}(\mathrm{H}(\lambda )))`$ using Lemma 4.1. Notice that $`Y`$ maps surjectively to $`๐G/B\times G/B`$ and that $`Y`$ is a quotient of $`G\times G`$. Consequently $`Y`$ maps bijectively onto $`๐`$ and hence $`Y`$ must be $`G\times G`$-equivariantly isomorphic to $`G/B\times G/B`$. Moreover
###### Lemma 4.3.
Let $`X`$ be a complete toroidal embedding of a connected semisimple group $`G`$ and let $`Y`$ be a closed $`G\times G`$-orbit in $`X`$. Then $`Y`$ is $`G\times G`$-equivariantly isomorphic to $`G/B\times G/B`$ and
$$\rho _\lambda (gB,g^{}B)(g,g^{})[v_\lambda ^+u_\lambda ^+].$$
Consequently, the pull back of the ample generator $`๐ช_\lambda (1)`$ of the Picard group of $`(\mathrm{End}(\mathrm{H}(\lambda )))`$ to $`Y`$ is isomorphic to $`(\lambda ,w_0\lambda )`$.
### 4.3. The dualizing sheaf of equivariant embeddings
Let $`X`$ be a smooth equivariant embedding of $`G`$. As $`G`$ is an affine variety the complement $`XG`$ of $`G`$ is of pure codimension 1 in $`X`$. Let $`X_1,\mathrm{},X_n`$ denote the irreducible components of $`XG`$ which are then divisors in $`X`$. Let $`D_i`$, $`i=1,\mathrm{},l`$, denote the closures of the Bruhat cells $`Bw_0s_iB`$ within $`X`$. Then also $`D_i`$ is a divisor in $`X`$. When $`X`$ is the wonderful compactification $`๐ฟ`$ of a group of adjoint type we will also denote $`X_j`$ and $`D_i`$ by $`๐ฟ_j`$ and $`๐ซ_i`$ respectively.
###### Proposition 4.4.
\[B-K, Prop.6.2.6\] The canonical divisor of the smooth equivariant embedding $`X`$ is
$$K_X=2\underset{iI}{}D_i\underset{j=1}{\overset{n}{}}X_j.$$
The line bundle $`(D_i)`$ associated to the divisor $`D_i`$ is connected to the morphisms $`\rho _\lambda `$ as explained by
###### Lemma 4.5.
Assume that $`X`$ is a toroidal embedding. Then there exists an isomorphism
$$(D_i)\rho _{\omega _i}^{}(๐ช_{\omega _i}(1)),$$
such that $`\rho _{\omega _i}^{}(u_{\omega _i}^+v_{\omega _i}^+)`$ is the canonical section of $`(D_i)`$.
###### Proof.
Consider first the case when $`X`$ is the wonderful compactification of $`G_{\mathrm{ad}}`$. Consider the pull back $`s_{\mathrm{ad}}:=(\rho _{\omega _i}^{\mathrm{ad}})^{}(u_{\omega _i}^+v_{\omega _i}^+)`$ of the global section $`u_{\omega _i}^+v_{\omega _i}^+`$ of $`๐ช_{\omega _i}(1)`$. Then the zero divisor $`(s_{\mathrm{ad}})_0`$ of $`s_{\mathrm{ad}}`$ is $`B\times B`$-invariant. Thus there exist nonnegative integers $`a_r`$ and $`b_j`$ , for $`r,j=1,\mathrm{},l`$, such that
$$(s_{\mathrm{ad}})_0=\underset{r=1}{\overset{l}{}}a_r๐ซ_r+\underset{j=1}{\overset{l}{}}b_j๐ฟ_j.$$
If $`b_j>0`$ for some $`j`$ then $`s_{\mathrm{ad}}`$ vanishes on $`๐`$ which by Lemma 4.1 is a contradiction. Hence, $`(s_{\mathrm{ad}})_0=_{r=1}^la_r๐ซ_r`$. It is known (see e.g. \[B-K, Prop. 6.1.11\]) that the restriction of $`(๐ซ_i)`$ to $`๐`$ is isomorphic to $`(\omega _i,w_0\omega _i)`$, so using using Lemma 4.3 it follows that $`a_i=1`$ and $`a_r=0`$ for $`ri`$. This proves the statement when $`X`$ is the wonderful compactification of $`G_{\mathrm{ad}}`$.
Consider now an arbitrary toroidal equivariant embedding $`X`$ of $`G`$. Let $`s:=\rho _{\omega _i}^{}(u_{\omega _i}^+v_{\omega _i}^+)`$ be the pull back of the the global section $`u_{\omega _i}^+v_{\omega _i}^+`$ of $`๐ช_{\omega _i}(1)`$. Then the zero divisor $`(s)_0`$ of $`s`$ is $`B\times B`$-invariant and by the already proved case above we conclude that
$$(s)_0=cD_i+\underset{j=1}{\overset{n}{}}d_jX_j,$$
for certain nonnegative integers $`c>0`$ and $`d_j`$, $`j=1,\mathrm{},n`$. Assume $`d_j>0`$ for some $`j`$. Then $`s`$ vanishes on a $`G\times G`$-stable subset and as $`s`$ is the pull back of $`s_{\mathrm{ad}}`$ from $`๐ฟ`$ to $`X`$, we conclude that $`s_{\mathrm{ad}}`$ also vanishes on a $`G\times G`$-stable subset $`V`$. But then $`s_{\mathrm{ad}}`$ vanishes on a closed $`G\times G`$-orbit in the closure of $`V`$ which can only be $`๐`$. As above this is a contradiction and we conclude that $`(s)_0=cD_i`$.
In order to prove that $`c=1`$ we may assume that $`G=G_{\mathrm{sc}}`$ is simply connected and that $`X=G`$. In this case the statement is well known (cf. proof of 6.1.11 in \[B-K\]). โ
### 4.4. Sections of the dualizing sheaf
Let again $`X`$ be a smooth equivariant embedding of $`G`$. For each $`i=1,\mathrm{},l`$, there exists a unique $`G_{\mathrm{sc}}\times G_{\mathrm{sc}}`$-linearization of the line bundle $`(D_i)`$. The set of global sections of $`(D_i)`$ may then be regarded as a $`G_{\mathrm{sc}}\times G_{\mathrm{sc}}`$-module. We claim
###### Proposition 4.6.
There exists a $`G_{\mathrm{sc}}\times G_{\mathrm{sc}}`$-equivariant morphism
$$\psi _i:\mathrm{H}(\omega _i)^{}\mathrm{H}(\omega _i)\mathrm{H}^0(X,(D_i)),$$
such that $`\psi _i(u_{\omega _i}^+v_{\omega _j}^+)`$ is the canonical section of $`(D_i)`$.
###### Proof.
When $`X`$ is toroidal this follows by Lemma 4.5. For a general smooth embedding $`X`$ there exists by Zariskiโs main theorem (cf. proof of Prop.6.2.6 \[B-K\]) an open subset $`X^{}`$ of $`X`$ such that $`X^{}`$ is a toroidal embedding of $`G`$ and such that the complement $`XX^{}`$ has codimension $`2`$ in $`X`$. As the statement is invariant under replacing $`X`$ with $`X^{}`$ the result now follows. โ
#### 4.4.1. The complete toroidal case
When $`X`$ is a complete toroidal embedding of $`G`$ we may even give more structure to the map $`\psi _i`$ given in Proposition 4.6. To this end, let $`Y`$ denote a closed $`G\times G`$-orbit in $`X`$ and consider the restriction map
$$i_{|Y}^{}:\mathrm{H}^0(X,(D_i))\mathrm{H}^0(Y,(D_i)_{|Y}).$$
By Lemma 4.3 and Lemma 4.5 it follows that $`(D_i)_{|Y}(\omega _i,w_0\omega _i)`$. Consequently, there exists a $`G_{\mathrm{sc}}\times G_{\mathrm{sc}}`$-equivariant isomorphism
$$\mathrm{H}^0(Y,(D_i)_{|Y})\mathrm{H}(w_0\omega _i)\mathrm{H}(\omega _i).$$
By Lemma 4.3 the composition of $`i_{|Y}^{}`$ with $`\psi _i`$ is nonzero and hence we obtain a commutative $`G_{\mathrm{sc}}\times G_{\mathrm{sc}}`$-equivariant diagram
where the left vertical map is defined by some nonzero $`G_{\mathrm{sc}}`$-equivariant map $`\mathrm{H}(\omega _i)^{}\mathrm{H}(w_0\omega _i)`$. Notice that by Frobenius reciprocity the map $`\mathrm{H}(\omega _i)^{}\mathrm{H}(w_0\omega _i)`$ is defined uniquely up to a nonzero constant.
## 5. Frobenius Splitting
In this section we collect a number of facts from the theory of Frobenius splitting. The presentation will be sketchy only stating the results which we will need. For a more thorough, and closely related, presentation we refer to \[T\].
### 5.1. Frobenius splitting
Let $`X`$ denote a scheme of finite type over an algebraically closed field $`k`$ of characteristic $`p>0`$. The *absolute Frobenius morphism* on $`X`$ is the morphism $`F:XX`$ of schemes, which is the identity on the set of points and where the associated map of sheaves
$$F^{\mathrm{}}:๐ช_XF_{}๐ช_X$$
is the $`p`$-th power map. We say that $`X`$ is *Frobenius split* (or just F-split) if there exists a morphism $`s\mathrm{Hom}_{๐ช_X}(F_{}๐ช_X,๐ช_X)`$ such that the composition $`sF^{\mathrm{}}`$ is the identity map on $`๐ช_X`$.
### 5.2. Stable Frobenius splittings along divisors
Let $`D`$ denote an effective Cartier divisor on $`X`$ with associated line bundle $`๐ช_X(D)`$ and canonical section $`\sigma _D`$. We say that $`X`$ is *stably Frobenius split along $`D`$* if there exists a positive integer $`e`$ and a morphism
$$s\mathrm{Hom}_{๐ช_X}(F_{}^e๐ช_X(D),๐ช_X),$$
such that $`s(\sigma _D)=1`$. In this case we say that $`s`$ is a stable Frobenius splitting of $`X`$ along $`D`$ of degree $`e`$. Notice that $`X`$ is Frobenius split exactly when there exists a stable Frobenius splitting of $`X`$ along the zero divisor $`D=0`$.
###### Remark 5.1.
Consider an element $`s\mathrm{Hom}_{๐ช_X}(F_{}^e๐ช_X(D),๐ช_X)`$. Then the condition $`s(\sigma _D)=1`$ on $`s`$ for it to define a stable Frobenius splitting of $`X`$, may be checked on any open dense subset of $`X`$.
### 5.3. Subdivisors
Let $`D^{}D`$ denote an effective Cartier subdivisor and let $`s`$ be a stable Frobenius splitting of $`X`$ along $`D`$ of degree $`e`$. The composition of $`s`$ with the map
$$F_{}^e๐ช_X(D^{})F_{}^e๐ช_X(D),$$
defined by the canonical section of the divisor $`DD^{}`$, is then a stable Frobenius splitting of $`X`$ along $`D^{}`$ of degree $`e`$. Applying this to the case $`D^{}=0`$ it follows that if $`X`$ is stably Frobenius split along any effective divisor $`D`$ then $`X`$ is also Frobenius split.
### 5.4. Compatibly split subschemes
Let $`Y`$ denote a closed subscheme of $`X`$ with sheaf of ideals $`_Y`$. When
$$s\mathrm{Hom}_{๐ช_X}(F_{}^e๐ช_X(D),๐ช_X)$$
is a stable Frobenius splitting of $`X`$ along $`D`$ we say that $`s`$ *compatibly Frobenius splits* $`Y`$ if the following conditions are satisfied
1. The support of $`D`$ does not contain any of the irreducible components of $`Y`$.
2. $`s\left(F_{}^e(_Y๐ช_X(D))\right)_Y`$.
When $`s`$ compatibly Frobenius splits $`Y`$ there exists an induced stable Frobenius splitting of $`Y`$ along $`DY`$ of degree $`e`$.
###### Lemma 5.2.
Let $`s`$ denote a stable Frobenius splitting of $`X`$ along $`D`$ which compatibly Frobenius splits a closed subscheme $`Y`$ of $`X`$. If $`D^{}D`$ then the induced stable Frobenius splitting of $`X`$ along $`D^{}`$, defined in Section 5.3, compatibly Frobenius splits $`Y`$.
###### Lemma 5.3.
Let $`D_1`$ and $`D_2`$ denote effective Cartier divisors. If $`s_1`$ (resp. $`s_2`$) is a stable Frobenius splitting of $`X`$ along $`D_1`$ (resp. $`D_2`$) of degree $`e_1`$ (resp. $`e_2`$) which compatibly splits a closed subscheme $`Y`$ of $`X`$, then there exists a stable Frobenius splitting of $`X`$ along $`D_1+D_2`$ of degree $`e_1+e_2`$ which compatibly splits $`Y`$.
###### Lemma 5.4.
Let $`s`$ denote a stable Frobenius splitting of $`X`$ along an effective divisor $`D`$. Then
1. If $`s`$ compatibly Frobenius splits a closed subscheme $`Y`$ of $`X`$ then $`Y`$ is reduced and each irreducible component of $`Y`$ is also compatibly Frobenius split by $`s`$.
2. Assume that $`s`$ compatibly Frobenius splits closed subschemes $`Y_1`$ and $`Y_2`$ and that the support of $`D`$ does not contain any of the irreducible components of the scheme theoretic intersection $`Y_1Y_2`$. Then $`s`$ compatibly Frobenius splits $`Y_1Y_2`$.
The following statement relates stable Frobenius splitting along divisors to compatibly Frobenius splitting.
###### Lemma 5.5.
Let $`D`$ and $`D^{}`$ denote effective Cartier divisors and let $`s`$ denote a stable Frobenius splitting of $`X`$ along $`(p1)D+D^{}`$ of degree $`1`$. Then there exists a stable Frobenius splitting of $`X`$ along $`D^{}`$ of degree $`1`$ which compatibly splits the closed subscheme defined by $`D`$.
### 5.5. Cohomology and Frobenius splitting
The notion of Frobenius splitting is particular useful in connection with proving higher cohomology vanishing for line bundles. We will need
###### Lemma 5.6.
Let $`s`$ denote a stable Frobenius splitting of $`X`$ along $`D`$ of degree $`e`$ and let $`Y`$ denote a closed compatibly Frobenius split subscheme of $`X`$. Then for every line bundle $``$ on $`X`$ and every integer $`i`$ there exists an inclusion
$$\mathrm{H}^i(X,_Y)\mathrm{H}^i(X,_Y^{p^e}๐ช_X(D)),$$
of abelian groups. In particular, when $`X`$ is projective, $``$ is globally generated and $`D`$ is ample then the group $`\mathrm{H}^i(X,_Y)`$ is zero for $`i>0`$.
### 5.6. Push forward
Let $`f:XX^{}`$ denote a proper morphism of schemes and assume that the induced map $`๐ช_X^{}f_{}๐ช_X`$ is an isomorphism. Then every Frobenius splitting of $`X`$ induces, by application of the functor $`f_{}`$, a Frobenius splitting of $`X^{}`$. Moreover, when $`Y`$ is a compatibly Frobenius split subscheme of $`X`$ then the induced Frobenius splitting of $`X^{}`$ compatibly splits the scheme theoretic image $`f(Y)`$ (see \[M-R, Prop.4\]). We will need the following connected statement.
###### Lemma 5.7.
Let $`f:XX^{}`$ denote a morphism of projective schemes such that $`๐ช_X^{}f_{}๐ช_X`$ is an isomorphism. Let $`Y`$ be a closed subscheme of $`X`$ and denote by $`Y^{}`$ the scheme theoretic image $`f(Y)`$. Assume that there exists a stable Frobenius splitting of $`X`$ along an ample divisor $`D`$ which compatibly splits $`Y`$. Then $`f_{}๐ช_Y=๐ช_Y^{}`$ and $`\mathrm{R}^if_{}๐ช_Y=0`$ for $`i>0`$.
### 5.7. Frobenius splitting of smooth varieties
When $`X`$ is a smooth variety there exists a canonical $`๐ช_X`$-linear identification (see e.g. \[B-K, ยง1.3.7\])
$$F_{}\omega _X^{1p}\mathrm{Hom}_{๐ช_X}(F_{}๐ช_X,๐ช_X).$$
Hence, a Frobenius splitting of $`X`$ may be identified with a global section of $`\omega _X^{1p}`$ with certain properties. A global section $`\tau `$ of $`\omega _X^{1p}`$ which corresponds to a Frobenius splitting up to a nonzero constant will be called a *Frobenius splitting section*.
###### Lemma 5.8.
Let $`\tau `$ be a Frobenius splitting section of a smooth variety $`X`$. Then there exists a stable Frobenius splitting of $`X`$ of degree $`1`$ along the Cartier divisor defined by $`\tau `$. In particular, if $`\tau =\stackrel{~}{\tau }^{p1}`$ is a $`(p1)`$-th power of a global section $`\stackrel{~}{\tau }`$ of $`\omega _X^1`$, then $`X`$ is Frobenius split compatibly with the zero divisor of $`\stackrel{~}{\tau }`$.
### 5.8. Frobenius splitting of $`๐ฎ/๐ฉ`$
The flag variety $`X=G/B`$ is a smooth variety with dualizing sheaf $`\omega _X=(2\rho )`$ where $`\rho `$ is a dominant weight defined as half of the sum of the positive roots. Let $`\mathrm{St}:=\mathrm{H}((p1)\rho )`$ denote the Steinberg module of $`G_{\mathrm{sc}}`$ and consider the multiplication map
$$m_{G/B}:\mathrm{St}\mathrm{St}\mathrm{H}(2(p1)\rho )\mathrm{H}^0(G/B,\omega _{G/B}^{1p}).$$
The Steinberg module $`\mathrm{St}`$ is an irreducible selfdual $`G_{\mathrm{sc}}`$-module and hence there exists a unique (up to nonzero scalars) nondegenerate $`G_{\mathrm{sc}}`$-invariant bilinear form
$$\varphi _{G/B}:\mathrm{St}\mathrm{St}k.$$
Then (see \[L-T, Thm.2.3\])
###### Theorem 5.9.
Let $`t`$ be an element in $`\mathrm{St}\mathrm{St}`$. Then $`\varphi _{G/B}(t)`$ is a Frobenius splitting section of $`G/B`$ if and only if $`\varphi _{G/B}(t)`$ is nonzero.
## 6. F-splitting of smooth equivariant embeddings
Let $`X`$ denote a smooth equivariant embedding of $`G`$. Define $`S`$ to be the $`G_{\mathrm{sc}}\times G_{\mathrm{sc}}`$-module
$$S=\underset{i=1}{\overset{l}{}}\left(\mathrm{H}(\omega _i)^{}\mathrm{H}(\omega _i)\right)^{(p1)}.$$
By Proposition 4.6 there exists a $`G_{\mathrm{sc}}\times G_{\mathrm{sc}}`$-equivariant morphism
$$\psi _X:S\mathrm{H}^0(X,\left((p1)\underset{i=1}{\overset{l}{}}D_i\right)).$$
defined as the $`(p1)`$-th product of the $`\psi _i`$โs. Let $`\sigma _j`$ denote the canonical section of $`(X_j)`$, for $`j=1,\mathrm{},n`$, and define for $`s,tS`$ the section
$$\mathrm{\Psi }_X(s,t)=\psi _X(s)\psi _X(t)\underset{i=1}{\overset{n}{}}\sigma _i^{p1},$$
of the line bundle $`\omega _X^{1p}`$ on $`X`$. Notice that if $`X^{}`$ is an equivariant embedding of $`G`$ which moreover is an open subset of $`X`$, then the restriction of $`\mathrm{\Psi }_X(s,t)`$ to $`X^{}`$ is equal to $`\mathrm{\Psi }_X^{}(s,t)`$. The main result Theorem 6.4 in this section describes when $`\mathrm{\Psi }_X(s,t)`$ is a Frobenius splitting section of the smooth embedding $`X`$.
### 6.1. F-splitting smooth complete toroidal embeddings
Consider a smooth complete toroidal embedding $`X`$ of $`G`$ and choose a closed $`G\times G`$-orbit $`Y`$ in $`X`$. By Lemma 4.3 we may identify $`Y`$ with $`G/B\times G/B`$ and under this isomorphism the restriction of $`(D_i)`$ to $`Y`$ corresponds to $`(\omega _i,w_0\omega _i)`$. In particular, restricting to $`Y`$ induces a map
$$i_{|Y}^{}:\mathrm{H}^0(X,\left(2(p1)\underset{i=1}{\overset{l}{}}D_i\right))\mathrm{H}^0(Y,\omega _Y^{1p}).$$
This leads to the following result which also explains the standard way of Frobenius splitting $`X`$ (cf. proof of Thm.6.2.7 \[B-K\])
###### Lemma 6.1.
Let $`X`$ denote a smooth complete toroidal embedding of $`G`$ and let $`Y`$ denote a closed $`G\times G`$-orbit in $`X`$. Let $`s`$ and $`t`$ be elements of $`S`$. Then $`\mathrm{\Psi }_X(s,t)`$ is a Frobenius splitting section of $`X`$ if and only if the restriction of $`\psi _X(s)\psi _X(t)`$ to $`Y`$ is a Frobenius splitting section of $`Y`$.
In order to control the restriction of $`\psi _X(s)\psi _X(t)`$ to $`Y`$ we use Section 4.4.1. It follows that we have a commutative $`G_{\mathrm{sc}}\times G_{\mathrm{sc}}`$-equivariant diagram with nonzero maps
where $`\mathrm{St}=\mathrm{H}\left((p1)\rho \right)`$ denotes the Steinberg module of $`G_{\mathrm{sc}}`$. Using the $`G_{\mathrm{sc}}\times G_{\mathrm{sc}}`$-invariant form $`\varphi _{G/B\times G/B}`$ on $`\mathrm{St}\mathrm{St}`$ we may define a similar form on $`S`$ by
$$\varphi :SSk,$$
$$st\varphi _{G/B\times G/B}(\eta (s)\eta (t))$$
Notice that $`S`$ and the $`G_{\mathrm{sc}}\times G_{\mathrm{sc}}`$-invariant form $`\varphi `$ is defined without the help of $`X`$. In particular, $`S`$ and $`\varphi `$ does not depend on $`X`$. Now by Lemma 6.1 and Theorem 5.9 we find
###### Proposition 6.2.
Let the notation be as above and let $`s`$ and $`t`$ be elements of $`S`$. Then $`\mathrm{\Psi }_X(s,t)`$ is a Frobenius splitting section of $`X`$ if and only if $`\varphi (st)`$ is nonzero.
### 6.2. Frobenius splitting $`G`$
By restricting the statement of Proposition 6.2 to $`G`$ we find
###### Corollary 6.3.
Let $`s`$ and $`t`$ be elements of $`S`$. Then
$$\mathrm{\Psi }_G(s,t):=\psi _G(s)\psi _G(t),$$
is a Frobenius splitting section of $`G`$ if and only if $`\varphi (st)`$ is nonzero.
###### Proof.
Choose a smooth complete toroidal embedding $`X`$ of $`G`$ and consider $`\mathrm{\Psi }_X(s,t)`$. Remember that checking whether $`\mathrm{\Psi }_X(s,t)`$ is a Frobenius splitting section of $`X`$ may be done on the open subset $`G`$ (see Remark 5.1). Now apply Proposition 6.2. โ
### 6.3. Frobenius splitting smooth equivariant embeddings
We can now prove that main result of this section.
###### Theorem 6.4.
Let $`X`$ denote an arbitrary smooth embedding of $`G`$ and let $`s`$ and $`t`$ be elements of $`S`$. Then $`\mathrm{\Psi }_X(s,t)`$ is a Frobenius splitting section of $`X`$ if and only if $`\varphi (st)`$ is nonzero.
###### Proof.
That $`\mathrm{\Psi }_X(s,t)`$ is a Frobenius splitting section may be checked on the open subset $`G`$. Now apply Corollary 6.3. โ
In the following statement $`t_i`$, $`i=1,\mathrm{},l`$, denotes the identity map in $`\mathrm{End}(\mathrm{H}(\omega _i)^{})\mathrm{H}(\omega _i)^{}\mathrm{H}(\omega _i)`$. Notice that as an element of $`\mathrm{End}(\mathrm{H}(\omega _i))^{}`$ the element $`t_i`$ is just the trace map on $`\mathrm{End}(\mathrm{H}(\omega _i))`$. We also fix a nonzero weight vector $`u_{\omega _i}^{}`$ of $`\mathrm{H}(\omega _i)^{}`$ of weight $`\omega _i`$.
###### Corollary 6.5.
The global section
$$\underset{i=1}{\overset{l}{}}\psi _i(t_i)^{p1}\underset{i=1}{\overset{l}{}}\psi _i(u_{\omega _i}^{}v_{\omega _i}^+)^{p1}\underset{j=1}{\overset{n}{}}\sigma _j^{p1},$$
of $`\omega _x^{1p}`$ is a Frobenius splitting section of $`X`$.
###### Proof.
It suffices by Theorem 6.4 to prove that
$$\varphi \left(\underset{i=1}{\overset{l}{}}t_i^{(p1)}\underset{i=1}{\overset{l}{}}(u_{\omega _i}^{}v_{\omega _i}^+)^{(p1)}\right)$$
is nonzero. The image of $`_{i=1}^lt_i^{(p1)}`$ in $`\mathrm{St}\mathrm{St}`$ coincides with a nonzero diag$`(G)`$-invariant element $`v_\mathrm{\Delta }`$. Moreover, the image of the element $`_{i=1}^l(u_{\omega _i}^{}v_{\omega _i}^+)^{(p1)}`$ in $`\mathrm{St}\mathrm{St}`$ equals $`v_{}v_+`$ for some nonzero weight vectors $`v_+`$ and $`v_{}`$ in $`\mathrm{St}`$ of weight $`(p1)\rho `$ and $`(p1)\rho `$ respectively. Thus, we have to show that $`\varphi _{G/B\times G/B}\left(v_\mathrm{\Delta }(v_{}v_+)\right)`$ is nonzero. But by weight consideration this is clearly the case. โ
## 7. Consequences in the smooth case
In this section we collect a number of consequences of the results in Section 6 and the following Lemma 7.1. Notice that when $`f`$ is a global section of a line bundle $``$ on a variety $`X`$, then we may regard $`f`$ as an element in the local rings $`๐ช_{X,x}`$ at points $`xX`$. This identification is unique up to units in $`๐ช_{X,x}`$. Using this identification we may now state
###### Lemma 7.1.
Let $`X`$ denote a smooth variety with dualizing sheaf $`\omega _X`$ and let $`_1,\mathrm{},_N`$ denote a collection of line bundles on $`X`$ such that $`_{i=1}^N_i\omega _X^1`$. Let $`f_i`$, $`i=1,\mathrm{},N`$, denote a global section of $`_i`$ and assume that $`_{i=1}^Nf_i^{p1}`$, considered as a global section of $`\omega _X^{1p}`$, is a Frobenius splitting section of $`X`$. Choose a sequence $`1i_1,\mathrm{},i_rN`$ of pairwise distinct integers. Then
1. The sequence $`f_{i_1},\mathrm{},f_{i_r}`$ forms a regular sequence in the local ring $`๐ช_{X,x}`$ at a point $`x`$ contained in the common zero set of $`f_{i_1},\mathrm{},f_{i_r}`$.
2. The common zero set of $`f_{i_1},\mathrm{},f_{i_r}`$ has pure codimension $`r`$.
###### Proof.
As all statements are local we may assume that $`X`$ is affine and that $`\omega _X`$ and $`_1,\mathrm{},_N`$ are all trivial. Hence, the elements $`f_1,\mathrm{},f_N`$ are just regular global functions on $`X`$. Moreover, by assumption there exists a function $`\tau :F_{}k[X]k[X]`$ such that $`\tau \left(a^p(f_1\mathrm{}f_N)^{p1}\right)=a`$ for all global regular functions $`ak[X]`$.
Let $`x`$ be a common zero of $`f_{i_1},\mathrm{},f_{i_r}`$ and assume that we have a relation of the form $`_{s=1}^ja_sf_{i_s}=0`$ for certain elements $`a_s`$ in $`๐ช_{X,x}`$. In particular, the product $`a_j^p(f_1\mathrm{}f_N)^{p1}`$ is contained in the ideal $`(f_{i_1}^p,\mathrm{},f_{i_{j1}}^p)`$ of $`๐ช_{X,x}`$ and hence
$$a_j=\tau \left(a_j^p(f_1\mathrm{}f_N)^{p1}\right)(f_{i_1},\mathrm{},f_{i_{j1}}).$$
This proves (1). Now (2) follows as a direct consequence of (1). โ
We can now prove the first of our main results
###### Corollary 7.2.
Let $`X`$ be a smooth equivariant embedding of $`G`$ and let $`\overline{N}`$ denote the closure of the Steinberg zero-fiber in $`X`$. Then
1. $`\overline{N}`$ coincides with the scheme theoretic intersection of the zero sets of $`t_i`$, $`i=1,\mathrm{},l`$. In particular, $`\overline{N}`$ is a local complete intersection.
2. $`\overline{N}`$ is normal, Gorenstein and Cohen-Macaulay.
3. The dualizing sheaf of $`\overline{N}`$ is isomorphic to the restriction of the line bundle $`_{\overline{N}}:=(_{i=1}^l(w_0,1)D_i_{j=1}^nX_j)`$ to $`\overline{N}`$.
###### Proof.
Consider the Frobenius splitting section of $`X`$ defined in Corollary 6.5. By Lemma 5.8 and Lemma 5.4 the scheme theoretic intersection $`C`$ of $`t_i`$, $`i=1,\mathrm{},l`$, is a reduced scheme. Moreover, by Lemma 7.1 each component of $`C`$ has codimension $`l`$ and will intersect the open locus $`G`$ (else, by Lemma 7.1, such a component would have codimension $`l+1`$). We conclude that $`CG`$ is dense in $`C`$ and that $`C`$ is a local complete intersection. But clearly (see remark above Corollary 6.5) $`CG`$ coincides with the Steinberg zero-fiber $`N`$, and thus $`C`$ must be equal to the closure $`\overline{N}`$. This proves (1).
To prove (2) it then suffices to show that $`\overline{N}`$ is regular in codimension 1. Let $`Z`$ denote a component of the singular locus of $`\overline{N}`$. If $`ZG\mathrm{}`$ then the codimension of $`Z`$ is $`2`$ as $`N`$ is normal by \[St, Thm.6.11\]. So assume that $`Z`$ is contained in a boundary component $`X_j`$. Now, by Lemma 5.8 the scheme theoretic intersection $`\overline{N}X_j`$ is reduced. Hence, as $`X_j`$ is a Cartier divisor, every smooth point of $`\overline{N}X_j`$ is also a smooth point of $`\overline{N}`$. In particular, $`Z`$ is properly contained in a component of $`\overline{N}X_j`$. But the variety $`\overline{N}X_j`$ has pure codimension 1 in $`\overline{N}`$ which ends the proof of (2).
Statement (3) follows by (1) and the description of the dualizing sheaf of $`X`$ in Proposition 4.4. โ
### 7.1. Stable Frobenius splittings along divisors
###### Proposition 7.3.
Let $`X`$ be a smooth equivariant embedding of $`G`$. Then there exists a stable Frobenius splitting of $`X`$ along the divisor
$$(p1)\left(\underset{j=1}{\overset{n}{}}X_j+\underset{i=1}{\overset{\mathrm{}}{}}(w_0,1)D_i\right)$$
of degree 1 which compatibly Frobenius splits the closure $`\overline{N}`$ of the Steinberg zero-fiber.
###### Proof.
Let $`\tau `$ denote the Frobenius splitting section of Corollary 6.5. By Lemma 5.8, Lemma 5.5 and Lemma 4.6 we know that $`\tau `$ defines a degree 1 stable Frobenius splitting of $`X`$ along the divisor
$$(p1)\left(\underset{j=1}{\overset{n}{}}X_j+\underset{i=1}{\overset{\mathrm{}}{}}(w_0,1)D_i\right),$$
which compatibly Frobenius splits the zero divisor of $`_{i=1}^l\psi _i(t_i)`$. Now apply Lemma 5.4(2), Corollary 7.2 and Lemma 7.1. โ
###### Corollary 7.4.
Let $`X`$ denote a projective smooth equivariant embedding of $`G`$. Then there exists a stable Frobenius splitting of $`X`$ along an ample divisor with support $`XG`$ which compatibly Frobenius splits the subvariety $`\overline{N}`$.
###### Proof.
By Proposition 7.3 and Lemma 5.2 there exists a stable Frobenius splitting of $`X`$ along the divisor $`_{j=1}^nX_j`$ which compatibly splits $`\overline{N}`$. Applying Lemma 5.2 and Lemma 5.3 it suffices to show that there exist positive integers $`c_j>0`$ such that $`_{j=1}^nc_jX_j`$ is ample. This follows from \[B-T, Prop.4.1(2)\]. โ
This has the following implications for resolutions
###### Corollary 7.5.
Let $`X`$ be a projective equivariant embedding of $`G`$ and let $`f:X^{}X`$ be a projective resolution of $`X`$ by a smooth projective equivariant $`G`$-embedding $`X^{}`$. Denote by $`\overline{N}^{}`$ (resp. $`\overline{N}`$) the closure of the Steinberg zero-fiber within $`X^{}`$ (resp. $`X`$). Then
(i) $`f_{}๐ช_X^{}=๐ช_X`$ and $`\mathrm{R}^if_{}๐ช_X^{}=0`$ for $`i>0`$. (cf. \[Rit, pf. of Cor.2\])
(ii) $`f_{}๐ช_{\overline{N}^{}}=๐ช_{\overline{N}}`$ and $`\mathrm{R}^if_{}๐ช_{\overline{N}^{}}=0`$ for $`i>0`$.
###### Proof.
As $`X^{}`$ is normal and $`f`$ is birational it follows from Zariskiโs main theorem that $`f_{}๐ช_X=๐ช_X^{}`$. Hence, by Lemma 5.7 it suffices to prove that there exists a stable Frobenius splitting of $`X`$ along an ample divisor which compatibly Frobenius splits $`\overline{N}`$. Now apply Corollary 7.4. โ
## 8. Frobenius splitting $`\overline{N}`$ for general embeddings
In this section $`X`$ will denote an arbitrary equivariant embedding of $`G`$ and $`\overline{N}`$ will denote the closure of the Steinberg zero-fiber in $`X`$.
###### Theorem 8.1.
There exists a Frobenius splitting of $`X`$ which simultaneously compatibly splits the closed subvarieties $`\overline{N},(w_0,1)D_i,X_j`$, for $`i=1,\mathrm{},l`$ and $`j=1,\mathrm{},n`$.
###### Proof.
By Theorem 4.2 we may find a projective resolution $`f:X^{}X`$ by a smooth toroidal embedding $`X^{}`$ of $`G`$. By Zariskiโs Main Theorem, $`f_{}๐ช_X^{}=๐ช_X`$. Thus, by \[M-R, Prop.4\] (cf. section 5.6) we can reduce to the case where $`X`$ is smooth. Now apply Corollary 6.5, Lemma 5.8, Lemma 5.4, Lemma 4.6 and Corollary 7.2 in the given order. โ
###### Example 8.2.
Consider the group $`G=\mathrm{PSL}_2(k)`$ over a field $`k`$ of positive characteristic different from $`2`$. Then the wonderful compactification $`๐ฟ`$ of $`G`$ may be identified with the projectivization of the set of $`2\times 2`$-matrices with entries in $`k`$. Denote the homogeneous coordinates in $`๐ฟ`$ by $`a,b,c`$ and $`d`$. Then the closure $`\overline{๐ฐ}`$ of the unipotent variety $`๐ฐ`$ of $`G`$ within $`๐ฟ`$, is defined by the polynomial $`f=(a+d)^24(adbc)`$. Moreover, the boundary is defined by the polynomial $`g=(adbc)`$. In particular, the ideal generated by $`f`$ and $`g`$ is not reduced and, as a consequence, the boundary $`๐ฟG`$ and the closure $`\overline{๐ฐ}`$ cannot be compatibly Frobenius split at the same time. When $`k`$ has characteristic $`2`$ the unipotent variety $`๐ฐ`$ coincides with the Steinberg zero-fiber. In this case the polynomial defining $`\overline{๐ฐ}`$ is given by $`f=a+d`$ and we do not see a similar problem.
###### Remark 8.3.
W. van der Kallen and T. Springer has informed us that they have proved Theorem 8.1 in case $`X`$ is the wonderful compactification of a group of adjoint type. Their proof proceeds by descending the Frobenius splitting results in \[T\] to the wonderful compactification.
We can also prove a vanishing result for line bundles on $`\overline{N}`$ :
###### Proposition 8.4.
Let $`X`$ denote a projective equivariant $`G`$-embedding and let $``$ (resp. $``$) denote a globally generated line bundle on $`X`$ (resp. $`\overline{N}`$). Then
$$\mathrm{H}^i(X,)=\mathrm{H}^i(\overline{N},)=0,i>0.$$
Moreover, the restriction map
$$\text{H}^0(X,)\text{H}^0(\overline{N},),$$
is surjective.
###### Proof.
By Corollary 7.5 we may assume that $`X`$ is smooth. Now apply Corollary 7.4 and the โin particularโ part of Lemma 5.6. โ
### 8.1. Canonical Frobenius splittings of $`X`$
A Frobenius splitting $`s:F_{}๐ช_Z๐ช_Z`$ of a $`B`$-variety $`Z`$ is a $`T`$-invariant Frobenius splitting such that the action of a root subgroup of $`G`$ associated to the simple root $`\alpha _i`$, is of the form
$$x_{\alpha _i}(c)s=\underset{j=1}{\overset{p1}{}}c^js_j,$$
for certain morphisms $`s_j:F_{}๐ช_Z๐ช_Z`$ and all $`ck`$.
As a subset of $`X`$ the closure $`\overline{N}`$ is invariant under the diagonal action of $`G`$. In particular, $`\overline{N}`$ is invariant under diag$`(B)`$ and we claim
###### Lemma 8.5.
The variety $`\overline{N}`$ is canonical Frobenius split with respect to the action of $`\mathrm{diag}(B)`$.
###### Proof.
It suffices to prove that $`X`$ has a diag$`(B)`$-canonical splitting which compatibly splits $`\overline{N}`$. Moreover, by Theorem 4.2 we may assume that $`X`$ is smooth. By the proof of Corollary 7.2 it then suffices to prove that the Frobenius splitting section of Corollary 6.5 is canonical.
As $`\psi _i^{}(t_i)`$ and $`\sigma _j`$ are diag$`(G)`$-invariant we may concentrate on the diag$`(T)`$-invariant factors $`\psi _i^{}(u_{\omega _i}^{}v_{\omega _i}^+)`$. The statement follows now as
$`x_{\alpha _j}(c).v_{\omega _i}^+=v_{\omega _i}^+,`$
$`x_{\alpha _j}(c).u_{\omega _i}^{}=u_{\omega _i}^{}+cu_{i,j},`$
for certain elements $`u_{i,j}\mathrm{H}(w_o\omega _i)^{}`$. โ
As a consequence we have (see \[B-K, Thm.4.2.13\])
###### Proposition 8.6.
Let $``$ denote a $`G_{\mathrm{sc}}`$-linearized line bundle on $`\overline{N}`$. Then the $`G_{\mathrm{sc}}`$-module $`\mathrm{H}^0(\overline{N},)`$ admits a good filtration, i.e. there exists a filtration by $`G_{\mathrm{sc}}`$-modules
$$0=M^0M^1M^2\mathrm{}\mathrm{H}^0(\overline{N},),$$
such that $`\mathrm{H}^0(\overline{N},)=_iM^i`$ and satisfying that the successive quotients $`M^{j+1}/M^j`$ are isomorphic to modules of the form $`\mathrm{H}(\lambda _j)`$ for certain dominant weights $`\lambda _j`$.
## 9. Geometric properties of $`\overline{N}`$
Let $`X`$ be an arbitrary equivariant $`G`$-embedding. When $`X`$ is smooth we have seen that $`\overline{N}`$ is normal and Cohen-Macaulay. In this section we will extend these two properties to arbitrary equivariant embeddings.
The following result is due to G. Kempf although the version stated here is taken from \[B-P, ยง7\] :
###### Lemma 9.1.
Let $`f:Z^{}Z`$ denote a proper map of algebraic schemes satisfying that $`f_{}๐ช_Z^{}๐ช_Z`$ and $`\mathrm{R}^if_{}๐ช_Z^{}=0`$ for $`i>0`$. If $`Z^{}`$ is Cohen-Macaulay with dualizing sheaf $`\omega _Z^{}`$ and if $`\mathrm{R}^if_{}\omega _Z^{}=0`$ for $`i>0`$, then $`Z`$ is Cohen-Macaulay with dualizing sheaf $`f_{}\omega _Z^{}`$.
We will also need the following result due to V. Mehta and W. van der Kallen (\[M-vdK, Thm.1.1\]):
###### Lemma 9.2.
Let $`f:Z^{}Z`$ denote a proper morphism of schemes and let $`V^{}`$ (resp. $`V`$) denote a closed subscheme of $`Z^{}`$ (resp. $`Z`$). By $`_V^{}`$ we denote the sheaf of ideals of $`V^{}`$. Fix an integer $`i`$ and assume
1. $`f^1(V)V^{}`$.
2. $`\mathrm{R}^if_{}_V^{}`$ vanishes outside $`V`$.
3. $`V^{}`$ is compatibly F-split in $`Z^{}`$.
Then $`\mathrm{R}^if_{}_V^{}=0`$.
We are ready to prove
###### Theorem 9.3.
Let $`X`$ denote an arbitrary equivariant $`G`$-embedding. Then the closure $`\overline{N}`$ of the Steinberg zero-fiber in $`X`$ is normal and Cohen-Macaulay.
###### Proof.
Any equivariant embedding has an open cover by open equivariant subsets of projective equivariant embeddings (see e.g. proof of \[B-K\] Corollary 6.2.8). This reduces the statement to the case where $`X`$ is projective. Choose a projective resolution $`f:X^{}X`$ of $`X`$ by a smooth equivariant embedding $`X^{}`$. By Corollary 7.5 we know that $`f_{}๐ช_{\overline{N}^{}}=๐ช_{\overline{N}}`$ and applying Corollary 7.2 this implies that $`\overline{N}`$ is normal.
In order to show that $`\overline{N}`$ is Cohen-Macaulay we apply the above Lemma 9.1 and Lemma 9.2. By Corollary 7.5 it suffices to prove that $`\mathrm{R}^if_{}\omega _{\overline{N}^{}}=0`$, $`i>0`$, where $`\omega _{\overline{N}^{}}`$ is the dualizing sheaf of $`\overline{N}^{}`$. By Corollary 7.2 the dualizing sheaf $`\omega _{\overline{N}^{}}`$ is isomorphic to the restriction of $`_{\overline{N}^{}}`$ to $`\overline{N}^{}`$. Let $`s^{}`$ denote the canonical section of the line bundle $`_{\overline{N}^{}}^1`$ on $`X^{}`$ and let $`V^{}`$ denote the intersection of $`\overline{N}^{}`$ with the zero divisor of $`s^{}`$. Combining Proposition 7.3 and Lemma 5.5 we find that $`\overline{N}^{}`$ is Frobenius split compatibly with the closed subscheme $`V^{}`$. Moreover, $`f:\overline{N}^{}\overline{N}`$ is an isomorphism above the open subset $`N`$ and
$$f^1(\overline{N}N)V^{}.$$
Hence, by Lemma 9.2 we conclude $`\mathrm{R}^if_{}_V^{}=0`$ for $`i>0`$. But $`_V^{}`$ is isomorphic to the restriction of $`_{\overline{N}^{}}`$ to $`\overline{N}^{}`$. โ
|
warning/0506/hep-th0506250.html
|
ar5iv
|
text
|
# Note About Tachyon Kink In Nontrivial Background
## 1 Introduction
Study of various aspects of the tachyon dynamics on a non-BPS Dp-brane in type IIA or IIB theories has led to some understanding of the tachyon dynamics near the tachyon vacuum <sup>1</sup><sup>1</sup>1For review of the open string tachyon condensation, see .. The tachyon effective action (2), describing the dynamics of the tachyon field on a non-BPS Dp-brane of type IIA and IIB theory was proposed in . It was argued in many papers that the tachyon effective action (2) gives a good description of the system under condition that tachyon is large and the second and higher derivatives of the tachyon are small <sup>2</sup><sup>2</sup>2For discussion of the effective field theory description of the tachyon condensation, see .. A kink solution in the full tachyon effective field theory, which is supposed to describe a BPS D(p-1)-brane was also constructed in . A kink solution, that by definition interpolates between the vacuua at $`T=\pm \mathrm{}`$ has to pass through $`0`$. Then we could expect that higher derivative corrections to the tachyon effective action will be needed to provide a good description of the D(p-1)-brane as a kink solution. This issue was carefully analysed in paper where it was shown that the energy density of the kink in the effective field theory is localised on codimension one surface as in the case of a BPS D(p-1)-brane. It was then also shown that the worldvolume theory of the kink solution is also given by the Dirac-Born-Infeld (DBI) action on a BPS D(p-1)-brane. Thus result shows that the kink solution of the effective field theory does provide a good description of the D(p-1)-brane even without taking into account higher derivative corrections. In other words, the tachyon effective action reproduces the low energy effective action on the world-volume of the soliton without any correction terms.
Since these results are very impressive it would be certainly useful to test the effective field theory description of the tachyon condensation in other, more general situations. In fact, since the DBI action describes the low energy dynamics of the BPS Dp-brane in general curved background we can ask the question whether we can construct the tachyon kink on the worldvolume of a non-BPS Dp-brane embedded in curved background and whether this kink has the interpretation as a lower dimensional D(p-1)-brane <sup>3</sup><sup>3</sup>3Some partial results considering tachyon condensation on non-BPS Dp-brane in curved background were presented in .. To answer this question we begin with common presumption that the tachyon effective action for Dp-brane (2) can be applied for the description of the tachyon dynamics in the nontrivial background <sup>4</sup><sup>4</sup>4The case of more general background, including NS $`B`$ field and Ramond-Ramond forms will be discussed in forthcoming publication.as well. Then we will study the equation of motion for the tachyon and for the modes that parametrise the embedding of the unstable Dp-brane in given spacetime. We will solve these equations with the field configuration similar to the ansatz that was given in . We will show that this ansatz solves the equation of motion for tachyon on condition that the mode $`t`$ that characterises the core of the kink (The precise meaning of this claim will be given bellow.) satisfies the equation of motion of the scalar field that describes the embedding of D(p-1)-brane in given background. This result shows that the spatial dependent tachyon condensation leads to the emergence of a D(p-1)-brane where the scalar modes that propagate on its worldvolume solve the equation of motion that arise from the DBI action for D(p-1)-brane that is moving in the same background.
The structure of this paper is as follows. In the next section (2) we will analyse the equation of motion for non-BPS Dp-brane in curved background. We will find the spatial dependent tachyon solution that has interpretation as a lower dimensional D(p-1)-brane whose dynamics is governed by DBI action. In section (3) we will study some examples of the nontrivial background. The first one corresponds to the stack of $`N`$ NS5-branes and the second one corresponds to the background generated by the collection of $`N`$ coincident Dk-branes. Finally, in conclusion (4) we will outline our results and suggest possible extension of this work.
## 2 Non-BPS Dp-brane in general background
The starting point for the analysis of the dynamics of a non-BPS Dp-brane in general background is the Dirac-Born-Infeld like tachyon effective action <sup>5</sup><sup>5</sup>5We use the convention where the fundamental string tension has been set equal to $`(2\pi )^1`$ (i.e. $`\alpha ^{}=1`$).
$`S={\displaystyle d^{p+1}\xi e^\mathrm{\Phi }V(T)\sqrt{det๐}},`$
$`๐_{\mu \nu }=g_{MN}_\mu Y^M_\nu Y^N+F_{\mu \nu }+_\mu T_\nu T,\mu ,\nu =0,\mathrm{},p,`$
$`F_{\mu \nu }=_\mu A_\nu _\nu A_\mu ,`$
where $`A_\mu ,\mu ,\nu =0,\mathrm{},p`$ and $`Y^{M,N},M,N=0,\mathrm{},9`$ are gauge and the transverse scalar fields on the worldvolume of the non-BPS Dp-brane and $`T`$ is the tachyon field. Since in this paper we will restrict ourselves to the situations when the gauge fields can be consistently taken to zero we will not write $`F_{\mu \nu }`$ anymore. $`V(T)`$ is the tachyon potential that is symmetric under $`TT`$ has maximum at $`T=0`$ equal to the tension of a non-BPS Dp-brane $`\tau _p`$ and has its minimum at $`T=\pm \mathrm{}`$ where it vanishes.
Using the worldvolume diffeomorphism invariance we can presume that the worldvolume coordinates $`\xi ^\mu `$ are equal to the spacetime coordinates $`y^\mu `$. Explicitly, we can write
$$Y^\mu =\xi ^\mu .$$
(2)
Then the induced metric takes the form <sup>6</sup><sup>6</sup>6We restrict ourselves in this paper to the situations when the background metric is diagonal.
$$\gamma _{\mu \nu }g_{MN}_\mu X^M_\nu X^N=g_{\mu \nu }+g_{mn}_\mu Y^m_\nu Y^n,$$
(3)
where $`Y^m,m,n=p+1,\mathrm{},9`$ parametrise the embedding of Dp-brane in a space transverse to its worldvolume. We should also mention that generally the metric components and dilaton are functions of $`\xi ^\mu `$ and $`Y^m`$:
$$g_{MN}=g_{MN}(\xi ^\mu ,Y^m),\mathrm{\Phi }=\mathrm{\Phi }(\xi ^\mu ,Y^m).$$
(4)
Now the equation of motion for $`T`$ and $`Y^m`$ that follow from (2) take the form
$$\frac{\delta V}{\delta T}e^\mathrm{\Phi }\sqrt{det๐}_\mu \left[e^\mathrm{\Phi }V\sqrt{det๐}_\nu T(๐^1)^{\nu \mu }\right]=0$$
(5)
and
$`{\displaystyle \frac{\delta [e^\mathrm{\Phi }]}{\delta Y^m}}V\sqrt{det๐}+{\displaystyle \frac{e^\mathrm{\Phi }}{2}}V\left[{\displaystyle \frac{\delta g_{\mu \nu }}{\delta Y^m}}+{\displaystyle \frac{\delta g_{np}}{\delta Y^m}}_\mu Y^n_\nu Y^p\right](๐^1)^{\nu \mu }\sqrt{det๐}`$
$`_\mu \left[e^\mathrm{\Phi }Vg_{mn}_\nu Y^n(๐^1)^{\nu \mu }\sqrt{det๐}\right]=0.`$
Our goal is to find the solution of these equations of motions that can be interpreted as a lower dimensional D(p-1)-brane. In order to obtain such a solution we will closely follow the paper by A. Sen . Let us choice one particular worldvolume coordinate, say $`\xi ^px`$ and consider following ansatz for the fields living on the worldvolume of Dp-brane
$$T=f(a(xt(\xi ))),Y^m=Y^m(\xi ),$$
(7)
where $`\xi ^\alpha ,\alpha =0,\mathrm{},p1`$ are coordinates tangential to the kink worldvolume. As in we presume that $`f(u)`$ satisfies following properties
$$f(u)=f(u),f^{}(u)>0,u,f(\pm \mathrm{})=\pm \mathrm{}$$
(8)
but is otherwise an arbitrary function of its argument $`u`$. $`a`$ is a constant that we shall take to $`\mathrm{}`$ in the end. In this limit we have $`T=\mathrm{}`$ for $`x>t(\xi )`$ and $`T=\mathrm{}`$ for $`x<t(\xi )`$.
Our goal is to check that the ansatz (7) solves the equation of motion (5) and (2). Firstly, using (7) the matrix $`๐_{\mu \nu }`$ takes the form
$`๐_{xx}=g_{xx}+a^2f^2,๐_{x\alpha }=g_{x\alpha }a^2f^2_\alpha t,`$
$`๐_{\beta x}=g_{\beta x}a^2f^2_\beta t,๐_{\alpha \beta }=(a^2f^2g_{xx})_\alpha t_\beta t+\stackrel{~}{๐}_{\alpha \beta },`$
$`\stackrel{~}{๐}_{\alpha \beta }=g_{\alpha \beta }+g_{mn}_\alpha Y^m_\beta Y^n+_\alpha t_\beta t.`$
For next purposes it will be useful to know the form of the determinant $`det๐`$. Using the following identity
$$det๐=det(๐_{\alpha \beta }๐_{\alpha x}\frac{1}{๐_{xx}}๐_{x\beta })det๐_{xx}$$
(10)
we get
$$det๐=a^2f^2det(\stackrel{~}{๐}_{\alpha \beta })+O(1/a).$$
(11)
As a next step we should express $`(๐^1)`$ in terms of $`\stackrel{~}{๐}`$. After some calculations we find
$`(๐^1)^{xx}=(\stackrel{~}{๐}^1)^{\alpha \beta }_\alpha t_\beta t,(๐^1)^{x\beta }=_\alpha t(\stackrel{~}{๐}^1)^{\alpha \beta },`$
$`(๐^1)^{\alpha x}=(\stackrel{~}{๐}^1)^{\alpha \beta }_\beta t,(๐^1)^{\alpha \beta }=(\stackrel{~}{๐}^1)^{\alpha \beta }`$
up to corrections of order $`\frac{1}{a^2}`$. In the following calculation we will also need an exact relation
$$(๐^1)^{\mu x}(๐^1)^{\mu \alpha }_\alpha t=\frac{1}{a^2f^2}\left(\delta _x^\mu (๐^1)^{xx}g_{xx}\right).$$
(13)
Using this expression we can now write
$`_\mu \left[e^\mathrm{\Phi }V\sqrt{det๐}(๐^1)^{\mu \nu }_\nu T\right]=_\mu \left[e^\mathrm{\Phi }Vaf^{}{\displaystyle \frac{1}{a^2f^2}}(\delta _x^\mu (๐^1)^{\mu x}g_{xx})\sqrt{det๐}\right].`$
Following we can now argue that due to the explicit factor of $`a^2f^2`$ in the denominator the leading contribution from individual terms in this expression is now of order $`a`$ and hence we can use the approximative results of $`det๐`$ and $`(๐^1)`$ given in (11) and (2) to analyse the equation of motion for tachyon
$`_\mu \left[e^\mathrm{\Phi }V\sqrt{det๐}af^{}{\displaystyle \frac{1}{a^2f^2}}(\delta _x^\mu (๐^1)^{\mu x}g_{xx})\right]`$
$`e^\mathrm{\Phi }V^{}\sqrt{det๐}=`$
$`_x\left[e^\mathrm{\Phi }V\sqrt{det\stackrel{~}{๐}}(1(\stackrel{~}{๐}^1)^{\alpha \beta }g_{xx}_\alpha t_\beta t)\right]`$
$`_\alpha \left[e^\mathrm{\Phi }V\sqrt{det\stackrel{~}{๐}}(\stackrel{~}{๐}^1)^{\alpha \beta }g_{xx}_\beta t\right]af^{}e^\mathrm{\Phi }V^{}\sqrt{det\stackrel{~}{๐}}=`$
$`=V\{_x\left[e^\mathrm{\Phi }\sqrt{det\stackrel{~}{๐}}(1(\stackrel{~}{๐}^1)^{\alpha \beta }g_{xx}_\alpha t_\beta t)\right]`$
$`_\alpha \left[e^\mathrm{\Phi }\sqrt{det\stackrel{~}{๐}}((\stackrel{~}{๐}^1)^{\alpha \beta }g_{xx}_\beta t)\right]\}=0.`$
This is important result that deserves deeper explanation. Firstly, from the form of the tachyon potential in the limit $`a\mathrm{}`$ we know that $`V`$ is equal to zero for $`xt(\xi )0`$ while for $`xt(\xi )=0`$ we get $`V(0)=\tau _p`$. Then it is clear that the tachyon equation of motion is obeyed for $`xt(\xi )0`$ while for $`x=t(\xi )`$ we should demand that the expression in the bracket should be equal to zero. In other words, we obtain following equation
$`{\displaystyle \frac{\delta e^\mathrm{\Phi }}{\delta x}}\sqrt{det\stackrel{~}{๐}}+{\displaystyle \frac{e^\mathrm{\Phi }}{2}}\left({\displaystyle \frac{\delta g_{\alpha \beta }}{\delta x}}+{\displaystyle \frac{\delta g_{xx}}{\delta x}}_\alpha t_\beta t+{\displaystyle \frac{\delta g_{mn}}{\delta x}}_\alpha Y^m_\beta Y^n\right)(\stackrel{~}{๐}^1)^{\beta \alpha }\sqrt{det\stackrel{~}{๐}}`$
$`_\alpha \left[e^\mathrm{\Phi }\sqrt{det\stackrel{~}{๐}}((\stackrel{~}{๐}^1)^{\alpha \beta }g_{xx}_\beta t)\right]_x\left[e^\mathrm{\Phi }\sqrt{det\stackrel{~}{๐}}(\stackrel{~}{๐}^1)^{\alpha \beta }g_{xx}\right]_\alpha t_\beta t=0.`$
We must stress that in (2) we firstly perform the derivative with respect to $`x`$ and then we insert the value $`x=t(\xi )`$ back to the resulting equation of motion. For example, in the first term on the second line we should perform a derivative with respect to $`\xi ^\alpha `$ with in mind that $`x`$ is an independent variable. After doing this we should everywhere replace $`x`$ with $`t(\xi )`$. Then the presence of the second term on the second line is crucial for an interpretation of $`t(\xi )`$ as an additional scalar field that parametrises the position of D(p-1)-brane in $`x`$ direction.
Put differently, we expect that the tachyon condensation leads to an emergence of D(p-1)-brane that is localised at $`x=t(\xi )`$. For that reason we should compare the equation (2) with the equation of motion for D(p-1)-brane embedded in the same background. As we know the dynamics of such a Dp-brane is governed by the DBI action
$$S=T_{p1}d^p\xi e^\mathrm{\Phi }\sqrt{det๐_{\alpha \beta }^{BPS}},$$
(17)
where the matrix $`๐_{\alpha \beta }^{BPS}`$ takes the form
$$๐_{\alpha \beta }^{BPS}=g_{\alpha \beta }+g_{xx}_\alpha Y_\beta Y+g_{mn}_\alpha Y^m_\beta Y^n,m,n=p+1,\mathrm{},9,$$
(18)
and where $`T_{p1}`$ is the tension of BPS D(p-1)-brane. Recall that $`T_p`$ is is related to the tension of the non-BPS D(p-1)-brane $`\tau _p`$ as $`\tau _{p1}=\sqrt{2}T_{p1}`$. In (18) we have chosen one particular transverse mode $`Y`$ in order to have a contact with the mode $`t`$ defined in the equation (7). Finally, the scalar fields $`Y^m`$ have the same meaning as in the case of a non-BPS Dp-brane. Now the equations of motion that follow from (17) take the form
$`{\displaystyle \frac{\delta }{\delta Y^m}}\left[e^\mathrm{\Phi }\sqrt{det๐_{\alpha \beta }^{BPS}}\right]_\alpha \left[e^\mathrm{\Phi }\sqrt{det๐_{\alpha \beta }^{BPS}}(๐^1)_{BPS}^{\beta \alpha }g_{mn}_\beta Y^n\right]=0,`$
$`{\displaystyle \frac{\delta }{\delta Y}}\left[e^\mathrm{\Phi }\sqrt{det๐_{\alpha \beta }^{BPS}}\right]_\alpha \left[e^\mathrm{\Phi }\sqrt{det๐_{\alpha \beta }^{BPS}}(๐^1)_{BPS}^{\beta \alpha }g_{xx}_\beta Y\right]=0,`$
where the variation $`\frac{\delta }{\delta Y^m},\frac{\delta }{\delta Y}`$ means the variation of the metric, dilaton with respect to $`Y^M,Y`$ respectively. Explicitly, the equation of motion for $`Y`$ can be written as
$`{\displaystyle \frac{\delta e^\mathrm{\Phi }}{\delta Y}}\sqrt{det๐}_{BPS}+`$
$`{\displaystyle \frac{1}{2}}e^\mathrm{\Phi }\left({\displaystyle \frac{\delta g_{\alpha \beta }}{\delta Y}}+{\displaystyle \frac{\delta g_{xx}}{\delta Y}}_\alpha Y_\beta Y+{\displaystyle \frac{\delta g_{mn}}{\delta Y}}_\alpha Y^m_\beta Y^n\right)(๐^1)_{BPS}^{\beta \alpha }\sqrt{det๐}_{BPS}`$
$`_\alpha \left[e^\mathrm{\Phi }\sqrt{det๐}_{BPS}(๐^1)_{BPS}^{\alpha \beta }g_{xx}_\beta Y\right]=0.`$
To see more clearly the relation with the equation (2) note that the expression on the third line can be written as
$`_\alpha \left[e^\mathrm{\Phi }\sqrt{det๐}_{BPS}(๐^1)_{BPS}^{\alpha \beta }g_{xx}_\beta Y\right]=`$
$`=_\alpha \left[e^{\mathrm{\Phi }(\xi ,x)}\sqrt{det๐_{BPS}(\xi ,x)}(๐^1)_{BPS}^{\alpha \beta }(\xi ,x)_\beta Y\right]`$
$`+_x\left[e^{\mathrm{\Phi }(\xi ,x)}\sqrt{det๐_{BPS}(\xi ,x)}(๐^1)_{BPS}^{\beta \alpha }(\xi ,x)\right]_\alpha Y_\beta Y,`$
where on the second line the derivative with respect to $`\xi ^\alpha `$ treats $`x`$ as an independent variable so that we firstly perform derivative with respect to $`\xi ^\alpha `$ and then we replace $`x`$ with $`Y`$. We proceed in the same way with the expression on the third line where we firstly perform the variation with respect to $`x`$ and then we replace $`x`$ with $`Y`$. Now it is clear that this prescription is the same as the expression on the second line in (2). More precisely, if we compare the equation (2) with the the equation (2) we see that these two expressions coincide when we identify $`t`$ with $`Y`$. In other words, the location of the tachyon kink is completely determined by field $`t(\xi )`$ that obeys the equation of motion of the embedding mode of D(p-1)-brane. We mean that this is very satisfactory result that shows that the Senโs construction of the tachyon kink can be consistently performed in curved background as well.
Now we will discuss the equation of motion for $`Y^k`$. Again, we will proceed as in . We begin with the first term in (2) that for the ansatz (7) takes the form
$$\frac{\delta e^\mathrm{\Phi }}{\delta Y^k}V\sqrt{det๐}=af^{}V\frac{\delta e^\mathrm{\Phi }}{\delta Y^k}\sqrt{det\stackrel{~}{๐}}.$$
(22)
In the same way we can show that the second term in (2) can be written as
$`e^\mathrm{\Phi }V[{\displaystyle \frac{\delta g_{\mu \nu }}{\delta Y^k}}+{\displaystyle \frac{\delta g_{mn}}{\delta Y^k}}_\mu Y^m_\nu Y^n](๐^1)^{\nu \mu }\sqrt{det๐}=af^{}Ve^\mathrm{\Phi }\sqrt{det\stackrel{~}{๐}}\times `$
$`\left[{\displaystyle \frac{\delta g_{xx}}{\delta Y^k}}(\stackrel{~}{๐}^1)^{\alpha \beta }_\alpha t_\beta t+({\displaystyle \frac{\delta g_{\alpha \beta }}{\delta Y^k}}+{\displaystyle \frac{g_{mn}}{\delta Y^k}}_\alpha Y^m_\beta Y^n)(\stackrel{~}{๐}^1)^{\alpha \beta }\right].`$
Finally, the third term in (2) is equal to
$`_\mu \left[e^\mathrm{\Phi }Vg_{km}_\nu Y^m(๐^1)^{\nu \mu }\sqrt{det๐}\right]=`$
$`_x\left[Ve^\mathrm{\Phi }g_{km}_\alpha Y^m(๐^1)^{\alpha x}\sqrt{det๐}\right]_\alpha \left[Ve^\mathrm{\Phi }g_{km}_\beta Y^m(๐^1)^{\alpha \beta }\sqrt{det๐}\right]=`$
$`=af^{}_x\left[Ve^\mathrm{\Phi }g_{km}_\alpha Y^m(\stackrel{~}{๐}^1)^{\alpha \beta }_\beta t\sqrt{det\stackrel{~}{๐}}\right]`$
$`af^{}_\alpha \left[Ve^\mathrm{\Phi }g_{km}_\beta Y^m(\stackrel{~}{๐}^1)^{\alpha \beta }\sqrt{det\stackrel{~}{๐}}\right]=`$
$`af^{}V\left(_\alpha \left[e^\mathrm{\Phi }g_{km}_\beta Y^m(\stackrel{~}{๐}^1)^{\alpha \beta }\sqrt{det\stackrel{~}{๐}}\right]+_x\left[e^\mathrm{\Phi }g_{km}_\alpha Y^m(\stackrel{~}{๐}^1)^{\alpha \beta }_\beta t\sqrt{det\stackrel{~}{๐}}\right]\right)`$
using the fact that $`_\alpha f=af^{}_\alpha t`$. Then collecting (22), (2) and (2) together we obtain
$`V\{{\displaystyle \frac{\delta e^\mathrm{\Phi }}{\delta Y^k}}\sqrt{det\stackrel{~}{๐}}+{\displaystyle \frac{1}{2}}e^\mathrm{\Phi }\sqrt{det\stackrel{~}{๐}}\times `$
$`\times \left[{\displaystyle \frac{\delta g_{xx}}{\delta Y^k}}_\alpha t_\beta t+{\displaystyle \frac{\delta g_{\alpha \beta }}{\delta Y^k}}+{\displaystyle \frac{g_{mn}}{\delta Y^k}}_\alpha Y^m_\beta Y^n\right](\stackrel{~}{๐}^1)^{\alpha \beta }`$
$`_\alpha \left[e^\mathrm{\Phi }g_{km}_\beta Y^m(\stackrel{~}{๐}^1)^{\alpha \beta }\sqrt{det\stackrel{~}{๐}}\right]_x\left[e^\mathrm{\Phi }g_{km}_\alpha Y^m(\stackrel{~}{๐}^1)^{\alpha \beta }\sqrt{det\stackrel{~}{๐}}\right]_\beta t\}=0.`$
Again it is easy to see that for $`xt(\xi )`$ the potential vanishes for $`a\mathrm{}`$ while for $`x=t(\xi )`$ we have $`V(0)=\tau _p`$. Then in order to obey the equation of motion the expression in the bracket should be equal to zero for $`x=t(\xi )`$. Note also that in the second expression on the last line in (2) we firstly perform a derivative with respect to $`x`$ and then we replace $`x`$ with $`t(\xi )`$. In other words, we can rewrite the last line in (2) into the form
$$_\alpha \left[e^{\mathrm{\Phi }(t(\xi ))}\sqrt{det\stackrel{~}{๐}(t(\xi ))}(\stackrel{~}{๐}^1)^{\alpha \beta }(t(\xi ))g_{km}(t(\xi ))_\beta \beta Y^m\right],$$
(26)
where we have explicitly stressed the dependence of the action on the mode $`t(\xi )`$ that replaces in the action the dependence on $`x`$. This result again supports the claim that we should identify $`t(\xi )`$ with an additional embedding coordinate of the D(p-1)-brane. Then by comparing the expression in the bracket in (2) with the equation of motion for $`Y^m`$ given in (2) we see that these two expressions coincide. In summary, we have shown that the tachyon kink solution on a non-BPS Dp-brane in nontrivial background can be identified as a lower dimensional D(p-1)-brane that is localised at the core of the kink. We have also shown that the dynamics of this D(p-1)-brane is governed by DBI action.
### 2.1 Stress energy tensor
Further support for the interpretation of the tachyon kink as a lower dimensional D(p-1)-brane can be obtained from the analysis of the stress energy tensor for the non-BPS Dp-brane. In order to find its form recall that we can write the action (2) as
$$S_p=d^{10}yd^{(p+1)}\xi \delta (Y^M(\xi )y^M)e^\mathrm{\Phi }V(T)\sqrt{det๐},$$
(27)
where
$$๐_{\mu \nu }=G_{MN}_\mu Y^M_\nu Y^N+_\mu T_\nu T,$$
(28)
and where $`\xi ^\mu ,\mu =0,\mathrm{},p`$ are worldvolume coordinates on Dp-brane. The form of action (27) is useful for determining the stress energy tensor $`T_{MN}(y)`$ of an unstable D-brane. In fact, the stress energy tensor $`T_{MN}(y)`$ is defined as the variation of $`S_p`$ with respect to $`g_{MN}(y)`$
$`T_{MN}(y)=2{\displaystyle \frac{\delta S_p}{\sqrt{g(y)}\delta g^{MN}(y)}}=`$
$`={\displaystyle d^{(p+1)}\xi \frac{\delta (Y^M(\xi )y^M)}{\sqrt{g(y)}}e^\mathrm{\Phi }Vg_{MK}g_{NL}_\mu Y^K_\nu Y^L(๐^1)^{\nu \mu }\sqrt{det๐}}.`$
The form of the stress energy tensor for gauge fixed Dp-brane action can be obtained from (2.1) by imposing the condition
$$Y^\mu =\xi ^\mu ,\mu =0,1,\mathrm{},p.$$
(30)
Then the integration over $`\xi ^\mu `$ swallows up the delta function $`\delta (y^\mu Y^\mu (\xi ))=\delta (y^\mu \xi ^\mu )`$ so that the resulting stress energy tensor takes the form
$`T_{mn}={\displaystyle \frac{\delta (Y^m(\xi )y^m)}{\sqrt{g}}}e^\mathrm{\Phi }Vg_{mm}_\mu Y^mg_{nn}_\nu Y^n(๐^1)^{\nu \mu }\sqrt{det๐},`$
$`T_{\mu \nu }={\displaystyle \frac{\delta (Y^m(\xi )y^m)}{\sqrt{g}}}e^\mathrm{\Phi }Vg_{\mu \mu }g_{\nu \nu }(๐^1)^{\nu \mu }\sqrt{det๐},`$
$`T_{\mu n}={\displaystyle \frac{\delta (Y^m(\xi )y^m)}{\sqrt{g}}}e^\mathrm{\Phi }Vg_{\mu \mu }g_{nn}_\nu Y^n(๐^1)^{\nu \mu }\sqrt{det๐},`$
$`T_{m\nu }={\displaystyle \frac{\delta (Y^m(\xi )y^m)}{\sqrt{g}}}e^\mathrm{\Phi }Vg_{mm}_\mu Y^mg_{\nu \nu }(๐^1)^{\nu \mu }\sqrt{det๐},`$
using the fact that the metric is diagonal.
If we now insert the ansatz (7) into these expressions we get
$`T_{mn}={\displaystyle \frac{\delta (Y^m(\xi )x^m)}{\sqrt{g}}}Vaf^{}e^\mathrm{\Phi }g_{mm}_\alpha Y^mg_{nn}_\beta Y^n(\stackrel{~}{๐}^1)^{\alpha \beta }\sqrt{det\stackrel{~}{๐}},`$
$`T_{\alpha \beta }={\displaystyle \frac{\delta (Y^m(\xi )y^m)}{\sqrt{g}}}Vf^{}ae^\mathrm{\Phi }g_{\alpha \alpha }g_{\beta \beta }(\stackrel{~}{๐}^1)^{\alpha \beta }\sqrt{det\stackrel{~}{๐}},`$
$`T_{xx}={\displaystyle \frac{\delta (Y^my^m)}{\sqrt{g}}}Vf^{}ae^\mathrm{\Phi }g_{xx}_\alpha t_\beta t(\stackrel{~}{๐}^1)^{\alpha \beta }\sqrt{det\stackrel{~}{๐}},`$
$`T_{x\alpha }=T_{\alpha x}=0,`$
$`T_{mx}=T_{xm}={\displaystyle \frac{\delta (Y^my^m)}{\sqrt{g}}}Vf^{}ae^\mathrm{\Phi }g_{mm}_\alpha Y^mg_{xx}_\beta t(\stackrel{~}{๐}^1)^{\alpha \beta }\sqrt{det\stackrel{~}{๐}}.`$
From now on the notation $`Y^m(\xi ),t(\xi )`$ means that these fields are functions of the coordinates on the worldvolume of the kink $`\xi ^\alpha ,\alpha =0,\mathrm{},p1`$.
According to the components of the stress energy tensor of the lower dimensional D(p-1)-brane arise by integrating all $`T_{MN}`$ given above over the direction of the tachyon condensation that in our case is $`x`$. Now we should be more careful since metric components generally depend on $`x`$. Let us introduce the following notation for the components of the stress energy tensors (2.1)
$$T_{MN}=V(f(a(t(\xi )x)))af^{}\stackrel{~}{T}_{MN}(x),$$
(33)
where we have explicitly stressed the dependence of $`\stackrel{~}{T}_{MN}`$ on $`x`$. If we now integrate $`T_{MN}`$ over $`x`$ we get
$$T_{MN}^{kink}=_{\mathrm{}}^{\mathrm{}}๐xV(f(a(xt(\xi ))))f^{}a\stackrel{~}{T}_{MN}(x)=๐mV(m)\stackrel{~}{T}_{MN}\left(\frac{f^1(m)}{a}+t(\xi )\right).$$
(34)
In the limit $`a\mathrm{}`$ the term proportional to $`1/a`$ goes to zero and we get that the components $`\stackrel{~}{T}_{MN}`$ are functions of $`t(\xi )`$ in place of $`x`$. Further, we can argue, following that the exponential fall off in $`V(m)`$ implies that in the limit $`a\mathrm{}`$ the contribution to the stress energy tensor is localised at the point where $`V`$ is equal to $`V(0)=\tau _p`$ which happens for $`x=t(\xi )`$. In other words, when we presume that the tension of BPS D(p-1)-brane is given by the integral
$$T_{p1}=_{\mathrm{}}^{\mathrm{}}๐mV(m)$$
(35)
we obtain the result that the components of the stress energy tensor of the kink take the form
$`T_{mn}^{kink}={\displaystyle \frac{T_{p1}\delta (Y^m(\xi )y^m)\delta (t(\xi )x)}{\sqrt{g}}}e^\mathrm{\Phi }g_{mm}_\alpha Y^mg_{nn}_\beta Y^n(\stackrel{~}{๐}^1)^{\alpha \beta }\sqrt{det\stackrel{~}{๐}},`$
$`T_{\alpha \beta }^{kink}={\displaystyle \frac{T_{p1}\delta (Y^m(\xi )y^m)\delta (t(\xi )x)}{\sqrt{g}}}e^\mathrm{\Phi }g_{\alpha \alpha }g_{\beta \beta }(\stackrel{~}{๐}^1)^{\alpha \beta }\sqrt{det\stackrel{~}{๐}},`$
$`T_{xx}^{kink}={\displaystyle \frac{T_{p1}\delta (Y^my^m)\delta (t(\xi )x)}{\sqrt{g}}}e^\mathrm{\Phi }g_{xx}_\alpha t_\beta t(\stackrel{~}{๐}^1)^{\alpha \beta }\sqrt{det\stackrel{~}{๐}},`$
$`T_{x\alpha }^{kink}=T_{\alpha x}^{kink}=0,`$
$`T_{mx}^{kink}=T_{xm}^{kink}={\displaystyle \frac{T_{p1}\delta (Y^my^m)\delta (t(\xi )x)}{\sqrt{g}}}e^\mathrm{\Phi }g_{mm}_\alpha Y^mg_{xx}_\beta t(\stackrel{~}{๐}^1)^{\alpha \beta }\sqrt{det\stackrel{~}{๐}},`$
where it is understood that $`g_{MN}`$ and $`\mathrm{\Phi }`$ are functions of $`\xi ^\alpha ,Y^m(\xi ),t(\xi )`$. In other words, the components of the stress energy tensors (2.1) correspond to the components of the stress energy tensor of a D(p-1)-brane localised at the points $`Y^m(\xi ),t(\xi )`$.
## 3 Examples of the tachyon condensation on a non-BPS Dp-brane in nontrivial background
In this section we will briefly discuss some examples of the tachyon condensation on a non-BPS Dp-brane that is embedded in nontrivial backgrounds.
### 3.1 NS5-brane background
As the first example we will consider the background corresponding to the stack of $`N`$ coincident NS5-branes
$`ds^2=dx_\mu dx^\mu +H_{NS}dx^mdx^m,`$
$`e^{2\mathrm{\Phi }}=H_{NS},`$
$`H_{mnp}=ฯต_{mnp}^q_q\mathrm{\Phi },`$
where the harmonic function $`H_{NS}`$ for $`N`$ coincident NS5-branes is equal to
$$H_{NS}(y^m)=1+\frac{2\pi N}{y^my_m},$$
(38)
where $`y^m,m=6,\mathrm{},9`$ label directions transverse to the worldvolume of NS5-branes.
The most simple case occurs when Dp-brane is stretched in the direction parallel with the worldvolume of NS5-branes <sup>7</sup><sup>7</sup>7In what follows we will consider the situation when we can ignore the NS two form background.. Using (3.1) it is then easy to determine the worldvolume metric
$`g_{\mu \nu }=\eta _{\mu \nu },g_{m_1n_1}=\delta _{m_1n_1},m_1,n_1=p+1,\mathrm{},5,`$
$`g_{m_2n_2}=H_{NS}\delta _{m_2n_2},m_2,n_2=6,\mathrm{},9,`$
where now $`H_{NS}`$ is function of $`Y^{m_2}`$
$$H_{NS}=1+\frac{2\pi N}{Y^{m_2}Y_{m_2}}.$$
(40)
Thanks to the manifest $`SO(p)`$ symmetry of the worldvolume theory all spatial coordinates $`\xi ^i,i=1,\mathrm{},p`$ are equivalent. Then we choose the direction on which the tachyon depends to be $`\xi ^p=x`$. Now it is clear that the spatial dependent tachyon condensation studied in previous section leads to the emergence of a D(p-1)-brane that is stretched in the $`x^0,\mathrm{},x^{p1}`$ directions and and which transverse position is determined by the worldvolume fields $`t(\xi ),Y^m(\xi )`$. These fields also obey the equations of motion that arise from the DBI action for a D(p-1)-brane that moves in the background of $`N`$ NS5-branes.
Another possibility occurs when we consider a non-BPS Dp-brane stretched in some of the transverse directions to the worldvolume of NS5-branes. More precisely, let us consider an unstable Dp-brane that is stretched in $`x^0,x^1,\mathrm{},x^k`$ directions and in $`x^6,\mathrm{},x^{6+pk}`$ directions. Then the metric components that appear on the worldvolume of the Dp-brane take the form
$`g_{\mu _1\nu _1}=\eta _{\mu _1\nu _1},\mu _1,\nu _1=0,\mathrm{},k,`$
$`g_{\mu _2\nu _2}=H_{NS}\delta _{\mu _2\nu _2},\mu _2,\nu _2=6,\mathrm{},(6+pk),`$
$`g_{m_1n_1}=\delta _{m_1n_1},m_1,n_1=k+1,\mathrm{},5,`$
$`g_{m_2n_2}=H_{NS}\delta _{m_2n_2},m_2,n_2=(7+pk),\mathrm{},9,`$
where the function $`H_{NS}`$ has the form
$$H_{NS}=1+\frac{2\pi N}{(\xi ^{\mu _2}\xi _{\mu _2}+Y^{m_2}Y_{m_2})^2}.$$
(42)
Now there are many possibilities how to construct lower dimensional D(p-1)-brane. If we perform the spatial dependent tachyon condensation on the worldvolume of the non-BPS Dp-brane where the tachyon $`t(x)`$ depends on coordinate from the set $`\xi ^1,\mathrm{},\xi ^k`$ (again, we take $`x=\xi ^k`$) we obtain D(p-1)-brane that is localised in $`x^k`$ direction and that is stretched in $`x^0,\mathrm{},x^{k1}`$ and $`x^6,\mathrm{},x^{(6+pk)}`$ directions. It is important to stress that the resulting configuration of $`N`$ NS5-brane and BPS D(p-1)-brane is not in general stable. Rather the dynamics of the BPS D(p-1)-brane in the background of $`N`$ NS5-branes is governed the equation of motions (2). To find stable configuration we should perform the same analysis as in .
Another possibility is to consider the tachyon condensation in direction from the set $`\xi ^6,\mathrm{},\xi ^{6+pk}`$. Let us choose $`x\xi ^6`$. Then it is clear that the tachyon condensation leads to the emergence of D(p-1)-brane stretched in $`(x^0,\mathrm{},x^k,x^7,\mathrm{},x^{6+pk})`$ directions and where the scalar fields on its worldvolume $`t(\xi ),Y^{m_1},Y^{m_2}`$ describing embedding of this D(p-1)-brane in nontrivial background, obey the equations of motions that arise from DBI action for BPS D(p-1)-brane.
### 3.2 Non-BPS Dp-brane in Dk-brane background
The second example that we will consider in this paper, is the spatial dependent tachyon condensation on the worldvolume of a non-BPS Dp-brane that moves in the background of $`N`$ BPS Dk-branes. This background is characterised by following metric and dilaton in the form
$`ds^2=H_k^{1/2}\eta _{\alpha \beta }dx^\alpha dx^\beta +H_k^{1/2}\delta _{mn}dx^mdx^n,`$
$`\alpha ,\beta =0,\mathrm{},k,m,n=k+1,\mathrm{},9`$
$`e^{2\mathrm{\Phi }}=H_k^{\frac{k3}{2}},`$
where the harmonic function $`H_p`$ takes the form
$$H_k=1+\frac{Ng_s(2\pi )^{\frac{7k}{2}}}{(y^my_m)^{\frac{7k}{2}}},$$
(44)
where $`y^m,m=k+1,\mathrm{},9`$ label the directions transverse to the worldvolume of $`N`$ Dkโbranes.
There is again many possibilities how to put in a non-BPS Dp-brane in this background. As the first possibility let us consider a non-BPS Dp-brane that is stretched in $`x^0,\mathrm{},x^p`$ directions and that is localised in $`Y^{m_1},m_1=p+1,\mathrm{},k`$ directions (parallel with the worldvolume of Dk-branes). This Dp-brane is also localised in $`Y^{m_2},m_2=k+1,\mathrm{},9`$ directions transverse to Dk-branes worldvolume. Now the metric components on its worldvolume take the form
$$g_{\mu \nu }=H_k^{1/2}\eta _{\mu \nu },g_{m_1n_1}=H_k^{1/2}\delta _{m_1n_1},g_{m_2n_2}=H_k^{1/2}\delta _{m_2n_2},$$
(45)
where $`H_k`$ depends on $`Y^{m_2}Y_{m_2}`$. It is clear that the spatial dependent tachyon condensation (Let us choose $`x`$ that appears in the ansatz (7) to be equal to $`\xi ^p`$.) leads to an emergence of a D(p-1)-brane with the worldvolume fields $`Y^{m_1},Y^{m_2}`$ as well as with the mode $`t(\xi )`$ that parametrises the location of D(p-1)-brane in $`x^p`$ direction.
Another possibility occurs when we consider Dp-brane where some of its worldvolume directions are parallel with the worldvolume of Dk-branes and other ones are stretched in the directions transverse to Dk-brane. This situation can be described by following induced metric on the worldvolume of non-BPS Dp-brane:
$`g_{\mu _1\nu _1}=H_k^{1/2}\eta _{\mu _1\nu _1},\mu _1,\nu _1=0,\mathrm{},l,`$
$`g_{\mu _2\nu _2}=H_{NS}^{1/2}\delta _{\mu _2\nu _2},\mu _2,\nu _2=k+1,\mathrm{},(k+1+pl),`$
$`g_{m_1n_1}=H_k^{1/2}\delta _{m_1n_1},m_1,n_1=l+1,\mathrm{},k,`$
$`g_{m_2n_2}=H_{NS}^{1/2}\delta _{m_2n_2},m_2,n_2=(k+2+pl),\mathrm{},9,`$
where the function $`H_k`$ is equal to
$$H_k=1+\frac{Ng_s(2\pi )^{\frac{7k}{2}}}{(\xi ^{\mu _2}\xi _{\mu _2}+Y^{m_2}Y_{m_2})^{\frac{7k}{2}}}.$$
(47)
If now the tachyon depends on one of the coordinates from the set $`\xi ^1,\mathrm{},\xi ^l`$, say $`x=\xi ^l`$, we obtain a D(p-1)-brane that is localised in $`x^l`$ direction and that is stretched in $`x^0,\mathrm{},x^{l1}`$ and $`x^{k+1},\mathrm{},x^{(k+1+pl)}`$ directions.
The next possibility corresponds to the tachyon condensation in the direction transverse to Dk-branes, say $`\xi ^{k+1}x`$. Following the general recipe given in previous section it is clear that this spatial dependent tachyon condensation leads to an emergence of a D(p-1)-brane that is stretched in $`(x^0,\mathrm{},x^l,x^{k+2},\mathrm{},x^{k+2+pl})`$ directions and its positions in the transverse space are described with the worldvolume scalar fields $`Y^{m_1}(\xi ),Y^{m_2}(\xi )`$ and also with $`t(\xi )`$ that parametrises the position of D(p-1)-brane in $`x^{k+1}`$ direction. It is also clear that these modes obey the equations of motions that follow from the DBI action for probe D(p-1)-brane in the background of $`N`$ Dk-branes. Note also that the resulting configuration of $`N`$ Dk-branes and D(p-1)-brane is not generally stable .
## 4 Conclusion
This paper was devoted to the study of the spatial dependent tachyon condensation on the worldvolume of a non-BPS Dp-brane that is moving in nontrivial background. We have shown that this tachyon condensation leads to an emergence of the D(p-1)-brane that is moving in the same background and where the scalar mode that determines the location of the kink on a non-BPS Dp-brane worldvolume can be interpreted as a mode that describes the transverse position of D(p-1)-brane and that obeys the equation of motion that follows from DBI action for D(p-1)-brane. We hope that this result is a nice example of the efficiently of the effective field theory description of the tachyon condensation and it also gives strong support for the form of the Dirac-Born-Infeld form of the tachyon effective action (2).
The extension of this paper is obvious. First of all we would like to study the tachyon condensation when we take into account nontrivial NS $`B`$ field and also nontrivial Ramond-Ramond field. It would be also interesting to study the tachyon condensation on the supersymmetric form of the non-BPS Dp-brane action. We hope to return to these problems in future.
Acknowledgement
This work was supported by the Czech Ministry of Education under Contract No. MSM 0021622409.
|
warning/0506/hep-th0506176.html
|
ar5iv
|
text
|
# Untitled Document
Microscopic Black Hole Entropy in Theories with Higher Derivatives
Per Kraus<sup>1</sup> pkraus@physics.ucla.edu and Finn Larsen<sup>2</sup> larsenf@umich.edu
<sup>1</sup>Department of Physics and Astronomy, UCLA,
Los Angeles, CA 90095-1547, USA.
<sup>2</sup>Michigan Center for Theoretical Physics, Department of Physics
University of Michigan, Ann Arbor, MI 48109-1120, USA.
Abstract
We discuss higher derivative corrections to black hole entropy in theories that allow a near horizon $`AdS_3\times X`$ geometry. In arbitrary theories with diffeomorphism invariance we show how to obtain the spacetime central charge in a simple way. Black hole entropy then follows from the Euclidean partition function, and we show that this gives agreement with Waldโs formula. In string theory there are certain diffeomorphism anomalies that we exploit. We thereby reproduce some recent computations of corrected entropy formulas, and extend them to the nonextremal, nonsupersymmetric context. Examples include black holes in M-theory on $`K3\times T^2`$, whose entropy reproduces that of the perturbative heterotic string with both right and left movers excited and angular momentum included. Our anomaly based approach also sheds light on why exact results have been obtained in four dimensions while ignoring $`R^4`$ type corrections.
June, 2005
1. Introduction
The famous area law of Bekenstein and Hawking relates the entropy of a black hole to the area of its event horizon as
$$S=\frac{1}{4G_D}A_{D2}.$$
In string theory this law has been verified in examples where the entropy is interpreted statistically in terms of microstates and the area is that of a black hole with the same macroscopic charges as the statistical system. In such computations many details of the string spectrum are known, implying numerous corrections to the microscopic theory. Additionally, higher derivative terms in the action modify the classical geometry and also change the area law (1.1) into Waldโs entropy formula
$$S=\frac{1}{8G_D}_{\mathrm{hor}}d^{D2}x\sqrt{h}\frac{\delta _D}{\delta R_{\mu \nu \alpha \beta }}ฯต^{\mu \nu }ฯต^{\alpha \beta },$$
which takes into account arbitrary derivative terms in the action. Remarkably, agreement between microscopics and macroscopics is maintained also after all these corrections are taken into account, at least in some examples \[2,,3,,4\] .
Recently it was pointed out that there are special cases of this agreement where the area of the black hole vanishes at leading order: $`A_{D2}=0`$ . For example, the microstates of the heterotic string consist of the usual perturbative spectrum. The black hole with the same classical charges has vanishing area in the two-derivative approximation to the action, but after higher derivatives are taken into account the entropy found from (1.1) agrees with the microscopic result. This example is important because the microscopics is so simple, which should facilitate very detailed comparisons that can test the whole framework and its interpretation. In particular, this seems like an ideal setting for testing Mathurโs conjecture that all microstates can be realized as distinct geometries.
Ultimately one would like to understand which features of quantum gravity are responsible for these striking agreements between radically different representations of black hole physics. The purpose of this note is to emphasize the central role played by symmetries, particularly diffeomorphism invariance and its anomalies. Viewed in this light, some of the agreements between microscopic and macroscopic results seem less surprising.
The key assumption in our approach is the existence of a near horizon region that includes an $`AdS_3`$ factor, even after higher derivative terms have been included in the Lagrangian. This assumption is suggested by the central role played by such near horizon geometries in the microscopics of black holes with finite area . Additionally, in an appropriate duality frame, an $`AdS_3`$ factor appears in the corrected geometry in all examples where derivative corrections have successfully been taken into account, at least as far as we are aware. The power of the assumption is that it relates the Lagrangian to the radius of the $`AdS_3`$ space and so, via generalized Brown-Henneaux reasoning, to the central charges $`c_{L,R}`$ of the associated conformal field theory. As we will see, the saddlepoint approximation to the black hole entropy, including all higher derivative corrections, is then given by the Cardy formula
$$S=2\pi \left[\sqrt{\frac{c_Lh_L}{6}}+\sqrt{\frac{c_Rh_R}{6}}\right]$$
where $`h_{L,R}`$ are the left and right moving momenta of the near horizon solution. Although the detailed form of the central charges $`c_{L,R}`$ depends critically on the spacetime Lagrangian, it will turn out that the Cardy formula (1.1) agrees with Waldโs formula (1.1) for general theories with diffeomorphism invariance. Thus, computation of the corrected black hole entropy reduces to finding the central charges. We will present a novel method for achieving this โ c-extremization โ which just involves solving algebraic equations. Given a higher derivative Lagrangian it is then quite simple to compute the corrected entropy.
Recent work has shown that in favorable cases it is possible to reproduce microscopic degeneracies of states to all orders in an expansion in inverse powers of charges \[9,,10\]. This result emerges just as naturally in our approach. Knowledge of the central charge leads to an expression for the black hole partition function which, when inverse Laplace transformed as in yields the microscopic degeneracies including all power law corrections.
We stress that our considerations are independent of spacetime supersymmetry. This contrasts with the (much) more explicit approach of \[2,,5\] which relies on the full power of supergravity. In particular, the usual approach has so far been limited to four dimensions, where supergravity is best developed, while our results apply equally in five dimensions.
In string theory there is additional structure due to anomalies which affect diffeomorphism invariance. Some relevant aspects are discussed in \[3,,4\]. These anomalies ultimately arise from $`M5`$-branes on the compactification manifold but they can also be understood without reference to $`M5`$-branes, using standard AdS/CFT reasoning. In this way we recover formulae from \[3,,4\] using elementary methods.
A natural context for these considerations is M-theory on $`AdS_3\times S^2\times X`$ where $`X`$ is some Calabi-Yau three-fold. A particularly striking example arises when $`X=K3\times T^2`$, so that M-theory is dual to heterotic string theory on $`T^5`$. In this case we find $`c_L=12`$ and $`c_R=24`$ which are indeed the correct central charges for the heterotic string. The remarkable feature is that we are sensitive to both chiral sectors of the heterotic string, and that we thereby derive the entropy for the heterotic string with both sectors excited. This shows that agreement is possible even without supersymmetry.
The point we wish to emphasize is that the constraints of matching symmetries and anomalies are enough to explain the successful entropy comparisons, at least in the cases we have considered. One puzzle in existing work has been why exact results are obtained by keeping only $`R^2`$ corrections, and neglecting higher powers. Here we see that it is the $`R^2`$ terms which yield the relevant diffeomorphism anomalies, and they are uncorrected by additional higher derivative terms.
The conventional approach of \[2,,5\] involves near horizon geometries with an $`AdS_2`$ factor and uses results from topological string theory \[11,,10\]. These $`AdS_2`$ geometries are related to $`AdS_3`$ by compactification. The $`AdS_3`$ perspective is simpler because spacetime symmetries such as the Virasoro algebra become manifest. On the other hand, in our approach we have not yet exploited the effects that can be seen only after compactification. It would be interesting to analyze how these further constrain the black hole spectrum.
Another open question is to find a criterion that determines when a near horizon $`AdS_3`$ appears from a singular geometry after derivative corrections are taken into account. This would characterize any ultimate limitations of our approach.
The remainder of this paper is organized as follows. In section 2 we consider the higher derivative corrections in a rather general setting, assuming only that the Lagrangian is formed from the metric in a diffeomorphism invariant way. In section 3 we apply these considerations to the case of M-theory on $`CY_3`$. In section 4, we discuss modifications due to gravitational anomalies and the appearance of the perturbative heterotic string spectrum. Finally, in section 5, we conclude with a discussion of how power law corrections to the entropy are taken into account using our approach.
2. Central charge and black hole entropy
In this section we first derive an expression for the central charge in terms of the Lagrangian including higher derivative corrections. We then review the computation of BTZ black hole entropy from the central charge. Finally, we combine the results and write the entropy in a form that agrees with Waldโs formula.
2.1. Computation of the central charge
We focus on brane configurations that have a near horizon geometry $`=`$ AdS$`{}_{3}{}^{}\times S^p\times X`$, where $`X`$ is an arbitrary compact space. One familiar case is $`p=3`$, which arises from the D1-D5 system in IIB string theory, where $`X`$ is $`T^4`$ or $`K3`$. This system gives rise to black holes in $`D=5`$. Another important example is $`p=2`$ which corresponds to $`D=4`$ black holes made from wrapping M2-branes and M5-branes on $`X=CY_3`$. We will come back to particular examples later, for now remaining in a general setting.
We take the near horizon limit, so that we have a theory of (not necessarily super) gravity on $``$. In this section we will take the metric to have Euclidean signature. For our purposes it is most convenient to perform a Kaluza-Klein reduction on $`X`$, to obtain a theory on AdS$`{}_{3}{}^{}\times S^p`$. The action for this theory is
$$I=\frac{1}{16\pi G_{p+3}}d^{p+3}x\sqrt{g}_{p+3}+S_{\mathrm{bndy}}+S_{CS}.$$
At this stage, $`_{p+3}`$ is an arbitrary function of the gravitational and matter fields, which is diffeomorphism invariant up to total derivatives that are cancelled by $`S_{\mathrm{bndy}}`$. In particular, it can include arbitrary higher derivative terms. The boundary terms indicated in (2.1) are needed to have a well-defined variational principle and also to define the boundary stress tensor \[12,,13\]; but we will not need their explicit form. $`S_{CS}`$ denotes Chern-Simons terms built out of gauge fields; we isolate these for reasons that will become apparent as we proceed.
We will be assuming that this theory admits solutions of the form AdS$`{}_{3}{}^{}\times S^p`$, over which $`_{p+3}`$ takes a constant value. This is indeed the case for the examples mentioned above. The radii of the two spaces are taken to be $`\mathrm{}_{Ads}`$ and $`\mathrm{}_{S^p}`$. For a general action there is not necessarily a single preferred definition of the metric, and so the radii are defined with respect to some particular choice.
As originally shown by Brown and Henneaux , a theory of gravity on a space whose noncompact part is AdS<sub>3</sub> corresponds to a conformal field theory on the two dimensional boundary. The conformal field theory has left and right moving central charges, $`c_L`$ and $`c_R`$, which are not necessarily equal. In this section we will consider the case in which they are equal. This is true for the D1-D5 system; for the M2-M5-brane case mentioned above it is only true for the leading part in an expansion in charges. We will come back to the case of unequal central charges later, for now just remarking that it leads to non-diffeomorphism invariant theories (gravitational anomalies), and so requires special care.
Our first task is to compute $`\mathrm{}_{Ads}`$ and $`\mathrm{}_{S^p}`$. Suppose we consider a family of trial solutions with $`\mathrm{}_{Ads}`$ and $`\mathrm{}_{S^p}`$ left as free parameters. In particular, we write the metric as
$$ds^2=\mathrm{}_{Ads}^2\left(d\eta ^2+\mathrm{sinh}^2\eta d\mathrm{\Omega }_2^2\right)+\mathrm{}_{S^p}^2d\mathrm{\Omega }_p^2.$$
The first two terms give AdS<sub>3</sub> in a convenient, but perhaps slightly unfamiliar, form. The actual values of the radii can then be obtained by demanding that the combination $`\mathrm{}_{Ads}^3\mathrm{}_{S^p}^p_{p+3}`$ be stationary under variation of $`\mathrm{}_{Ads}`$ and $`\mathrm{}_{S^p}`$. Roughly speaking, this can be thought of as extremizing the bulk action. A little care is required to establish that this is the correct procedure. In particular, we should recall that when the equations of motion are satisfied the full action is stationary under variations which vanish at the boundary; but in our case variations of the radii lead to variations even at the boundary. Furthermore, we have the boundary terms in (2.1). A simple way to avoid these complications is to consider an analytic continuation so that our solutions take the form $`S^3\times S^p`$. Then both complications are absent, and the total action is clearly proportional to $`\mathrm{}_{Ads}^3\mathrm{}_{S^p}^p_{p+3}`$. Hence this combination must be stationary. Our result follows after continuation back to AdS$`{}_{3}{}^{}\times S^p`$. We note that in general $`_{p+3}`$ will be a complicated function of the radii, incorporating for example the contributions of the field strengths, whose values are fixed by their equations of motion.
This discussion makes it clear why we isolated the Chern-Simon terms. These are not necessarily constant over our solution. On the other hand, being topological they play no role in determining the radii, or the central charge, so we are free to neglect them at this stage.
With foresight, we define the central charge function
$$c(\mathrm{}_{Ads},\mathrm{}_{S^p})=\frac{3\mathrm{\Omega }_2\mathrm{\Omega }_p}{32\pi G_{p+3}}\mathrm{}_{Ads}^3\mathrm{}_{S^p}^p_{p+3},$$
and so the actual values of the radii are determined by solving
$$\frac{}{\mathrm{}_{Ads}}c(\mathrm{}_{Ads},\mathrm{}_{S^p})|_{\mathrm{}_{Ads}=\overline{\mathrm{}}_{Ads}}=\frac{}{\mathrm{}_{S^p}}c(\mathrm{}_{Ads},\mathrm{}_{S^p})|_{\mathrm{}_{S^p}=\overline{\mathrm{}}_{S^p}}=0.$$
We wish to establish that $`c=c(\overline{\mathrm{}}_{Ads},\overline{\mathrm{}}_{S^p})`$ (equal on left and right!) is indeed the central charge, as defined by the conformal anomaly
$$T_i^i=\frac{c}{12}^{(2)}R,$$
of the dual $`D=2`$ CFT.
To this end, put the $`2D`$ CFT on a sphere with metric
$$ds^2=e^{2\omega }d\mathrm{\Omega }_2^2,$$
and focus on the partition function, $`Z=e^I`$, as a function of $`\omega `$. Under constant shifts of $`\omega `$ we have
$$\delta I=\frac{1}{4\pi }d^2x\sqrt{g}T^{ij}\delta g_{ij}=\frac{\delta \omega }{2\pi }d^2x\sqrt{g}T^{ij}g_{ij}=\frac{c}{24\pi }\delta \omega d^2x\sqrt{g}{}_{}{}^{(2)}R=\frac{c}{3}\delta \omega .$$
This is to be compared with the action (2.1) evaluated on (2.1):
$$I=\frac{\mathrm{\Omega }_2\mathrm{\Omega }_p}{16\pi G_{p+3}}\mathrm{}_{Ads}^3\mathrm{}_{S^p}^p๐\eta \mathrm{sinh}^2\eta _{p+3}+S_{\mathrm{bndy}}.$$
To make sense of this we need to impose an upper cutoff on $`\eta `$. The integral gives
$$_0^{\eta _{\mathrm{max}}}๐\eta \mathrm{sinh}^2\eta _{p+3}=\left(\frac{1}{2}\eta _{\mathrm{max}}+\frac{1}{2}\mathrm{cosh}\eta _{\mathrm{max}}\mathrm{sinh}\eta _{\mathrm{max}}\right)_{p+3}.$$
Now, $`S_{\mathrm{bndy}}`$ is the integral of a expression defined locally on the AdS boundary. It is constructed out of the intrinsic and extrinsic curvature of the boundary. Such terms will never give a contribution linear in $`\eta _{\mathrm{max}}`$. Instead, they cancel the second term in (2.1), leaving the action
$$I=\frac{\mathrm{\Omega }_2\mathrm{\Omega }_p}{32\pi G_{p+3}}\mathrm{}_{Ads}^3\mathrm{}_{S^p}^p_{p+3}\eta _{\mathrm{max}}.$$
Comparing (2.1) with (2.1) we have
$$\delta \omega =\delta \eta _{\mathrm{max}},$$
which then yields
$$c=\frac{3\mathrm{\Omega }_2\mathrm{\Omega }_p}{32\pi G_{p+3}}\mathrm{}_{Ads}^3\mathrm{}_{S^p}^p_{p+3},$$
as we wanted to show.
To summarize, we have shown that the central charge of an AdS$`{}_{3}{}^{}\times S^p\times X`$ solution can be obtained simply by extremizing the central charge function (2.1) with respect to the AdS and sphere radii. For a given Lagrangian this just means solving two algebraic equations. We will refer to our procedure as c-extremization.
We would like to emphasize a couple of important points. First, our result applies to an arbitrary higher derivative Lagrangian including matter fields. The requirement is just that this Lagrangian admits a solution with the assumed properties. Second, although we used some language familiar from the AdS/CFT correspondence, our result is completely independent of the validity of the AdS/CFT conjecture. Essentially, we have derived a result about how the gravitational action behaves under Weyl transformations of the AdS boundary.
2.2. Black hole entropy
Once the central charge is known, results for black hole entropy follow with little additional effort. We now review how this works in the general case, allowing independent values of the left and right moving central charges.
We consider black holes of the form $`BTZ\times S^p\times X`$. One way to compute the black hole entropy is by computing the action of the Euclidean black hole. From there, one gets the free energy, and then thermodynamic quantities follow in the standard way. The Euclidean BTZ black hole is a solid torus which can be continued to Lorentzian signature in many different ways. Consider the cycles on the boundary of the torus which are noncontractible with respect to the boundary. There is clearly one such cycle which is contractible in the solid torus. If one calls the coordinate along the contractible cycle $`\varphi `$, and the other cycle coordinate $`t_E`$, then upon continuing $`t_Eit`$ one obtains the geometry of thermal AdS<sub>3</sub>; that is, global AdS<sub>3</sub> with compact imaginary time. On the other hand, the opposite assignment of $`t_E`$ and $`\varphi `$ leads, upon continuation to Lorentzian signature, to the BTZ black hole.<sup>3</sup> Other choices lead to the so-called โ$`SL(2,Z)`$โ family of black holes.
From this point of view it becomes clear that the black hole partition function is just a rewriting of the thermal partition function. But the result for the thermal partition function follows directly from the central charges. Hence, so too does the black hole entropy.
Let us illustrate this in more detail; see \[14\]. An asymptotically AdS<sub>3</sub> solution carries energy $`H`$ and angular momentum $`J`$. In the CFT on the boundary $`J`$ is the momentum. We can also define the zero modes of the Virasoro generators as
$$\begin{array}{cc}\hfill h_L& =L_0\frac{c}{24}=\frac{HJ}{2},\hfill \\ \hfill h_R& =\stackrel{~}{L}_0\frac{\stackrel{~}{c}}{24}=\frac{H+J}{2}.\hfill \end{array}$$
We can think of a bulk solution as a contribution to the partition function
$$\begin{array}{cc}\hfill Z(\beta ,\mu )=e^I& =\mathrm{Tr}e^{\beta H\mu J}\hfill \\ & =\mathrm{Tr}e^{2\pi i\tau h_L}e^{2\pi i\overline{\tau }h_R},\hfill \end{array}$$
where we defined
$$\tau =i\frac{\beta \mu }{2\pi },\overline{\tau }=i\frac{\beta +\mu }{2\pi }.$$
When we go to Euclidean signature $`\mu `$ becomes pure imaginary and $`\overline{\tau }`$ becomes the complex conjugate of $`\tau `$. Also, it follows from (2.1) that $`\tau `$ is precisely the modular parameter of the Euclidean boundary torus.
Now consider thermal AdS<sub>3</sub>. In Lorentzian signature thermal AdS<sub>3</sub> takes the same form as AdS<sub>3</sub> written in the usual global coordinates. On the other hand, we know that global AdS<sub>3</sub> corresponds to the NS-NS vacuum, and as such carries $`L_0=\stackrel{~}{L}_0=0`$. Therefore, we conclude that the action of thermal AdS<sub>3</sub> is
$$I_{\mathrm{thermal}}(\tau ,\overline{\tau })=\frac{i\pi }{12}(c\tau \stackrel{~}{c}\overline{\tau }).$$
There are in fact additional contributions due to quantum fluctuations of massless fields. (2.1) just takes into account all the local contributions. The extra nonlocal contributions are, by definition, suppressed for large $`\beta `$, and will give subleading contributions to the entropy compared to the local piece.
We already noted that BTZ is obtained by interchanging $`t_E`$ and $`\varphi `$. This is just a modular transformation of the boundary torus: $`\tau \frac{1}{\tau }`$. Recall that a modular transformation is a diffeomorphism combined with a Weyl transformation. The action is invariant since if we take a flat metric on the torus then all potential anomalies vanish. We therefore conclude that
$$I_{\mathrm{BTZ}}(\tau ,\overline{\tau })=\frac{i\pi }{12}(\frac{c}{\tau }\frac{\stackrel{~}{c}}{\overline{\tau }}).$$
From (2.1) it follows that
$$\begin{array}{cc}\hfill h_L& =\frac{1}{2\pi i}\frac{I}{\tau }=\frac{c}{24\tau ^2},\hfill \\ \hfill h_R& =\frac{1}{2\pi i}\frac{I}{\overline{\tau }}=\frac{\stackrel{~}{c}}{24\overline{\tau }^2}.\hfill \end{array}$$
From the thermodynamic relation $`I=\beta H+\mu JS`$ we compute the entropy $`S`$ to be
$$S_{BTZ}=2\pi \left(\sqrt{\frac{c}{6}h_L}+\sqrt{\frac{\stackrel{~}{c}}{6}h_R}\right).$$
Three facts about this computation are worth stressing. First, the result (2.1) holds for an arbitrary theory of gravity admitting a BTZ black hole (times an arbitrary compactification space). Second, the result is valid entirely independent of the AdS/CFT correspondence. One can just think of it as a result for computing the action of the Euclidean black hole. Finally, (2.1) gives the entropy in terms of the black hole mass and angular momentum, and with the central charges appearing as โundetermined parametersโ. This shows that once we can compute the central charges, the black hole entropy follows directly from (2.1). But we have seen in the last section how the central charge โ in the case of $`\stackrel{~}{c}=c`$ โ follows from a simple extremization principle. Altogether, we have arrived at an efficient method of computing black hole entropy.
2.3. Equivalence with Waldโs approach
The Wald formula (1.1) gives the black hole entropy in an arbitrary diffeomorphism invariant theory . In his approach, one integrates a certain expression over the black hole horizon. The power of this result is its complete generality. However, for black holes with a near horizon AdS$`{}_{3}{}^{}\times S^p\times X`$ structure, the method we have described above is actually much simpler to implement. In particular, one is not required to locate the horizon at all: c-extremization gives the entropy directly. In any case, it is worthwhile to check that our result agrees with Waldโs formula, as we do now.
The essential ideas for demonstrating this equivalence appear in the paper where it is shown that Waldโs approach leads to a black hole entropy in the form (2.1). We will follow a slightly different procedure from .
We first want to write the central charge in a form suitable for comparison with the Wald formula. It is convenient to work directly in the theory compactified all the way to $`D=3`$. By assumption, all matter fields take constant values, so that we can write the action purely in terms of the metric. In $`D=3`$ the Riemann tensor can be expressed in terms of the Ricci tensor; so the general action will be a function of the Ricci tensor and its covariant derivatives <sup>4</sup> Actually, one can also include a Chern-Simons term, $`S\mathrm{Tr}\omega R`$, but for now we exclude such a term. It would lead to $`\stackrel{~}{c}c`$ and associated subtleties, which we postpone till a later section.
$$S=\frac{1}{16\pi G_3}d^3x\sqrt{g}_3(g^{\mu \nu },R_{\mu \nu })+S_{\mathrm{bndy}}.$$
Schematically we have
$$_3(g^{\mu \nu },R_{\mu \nu })\underset{n}{}a_n(g^{\mu \nu })^n(R_{\mu \nu })^n,$$
where the $`a_n`$ include covariant derivatives and contractions are not written out explicitly. The central charge function (2.1) is
$$c(\mathrm{}_{Ads})=\frac{3\mathrm{\Omega }_2}{32\pi G_3}\mathrm{}_{Ads}^3_3.$$
If we write introduce rescaled variables through
$$g_{\mu \nu }=\mathrm{}_{Ads}^2\widehat{g}_{\mu \nu },g^{\mu \nu }=\frac{1}{\mathrm{}_{Ads}^2}\widehat{g}^{\mu \nu },R_{\mu \nu }=2\widehat{g}_{\mu \nu }=\widehat{R}_{\mu \nu },$$
then $`\mathrm{}_{Ads}`$ satisfies
$$3_3+2\mathrm{}_{Ads}^2\frac{_3}{\mathrm{}_{Ads}^2}=0.$$
Furthermore, in the rescaled variables (2.1) the action reads
$$S=\frac{1}{16\pi G_3}d^3x\sqrt{\widehat{g}}\mathrm{}_{Ads}^3_3(\frac{\widehat{g}^{\mu \nu }}{\mathrm{}_{Ads}^2},\widehat{R}_{\mu \nu }),$$
so the derivative in (2.1) can be evaluated as <sup>5</sup> We use the fact that all covariant derivatives vanish on the background.
$$\mathrm{}_{Ads}^2\frac{_3}{\mathrm{}_{Ads}^2}=\widehat{R}_{\mu \nu }\frac{_3}{\widehat{R}_{\mu \nu }}=\frac{2}{\mathrm{}_{Ads}^2}g_{\mu \nu }\frac{_3}{R_{\mu \nu }}.$$
Simplifying (2.1) using (2.1) and (2.1) we find
$$c=\frac{\mathrm{}_{Ads}}{2G_3}g_{\mu \nu }\frac{_3}{R_{\mu \nu }}.$$
This formula generalizes the usual Brown-Henneaux central charge
$$c_0=\frac{3\mathrm{}_{Ads}}{2G_3},$$
by taking higher derivative corrections into account. The net result amounts to a rescaling of the $`AdS_3`$ radius $`\mathrm{}_{Ads}\mathrm{}_{\mathrm{eff}}=\mathrm{\Omega }\mathrm{}_{Ads}`$ where
$$\mathrm{\Omega }=\frac{1}{3}g_{\mu \nu }\frac{_3}{R_{\mu \nu }}=\frac{2G_3}{3\mathrm{}_{Ads}}c.$$
We are now ready to make the connection with Waldโs approach, since the latter involves an integration over the horizon of the derivative of the Lagrangian with respect to the curvature. Presently, the black hole entropy takes the form (2.1) with the central charge (2.1). The BTZ black hole, as usually written, is expressed in terms of the parameters $`M_3`$ and $`J_3`$ which, for a 2-derivative action are identified with the mass and angular momentum of the black hole. However, in the presence of higher derivatives the relation is rescaled by the conformal factor (2.1) and we have instead
$$\begin{array}{cc}\hfill h_{L,R}& =\mathrm{\Omega }\frac{M_3J_3}{2}.\hfill \end{array}$$
We now find the entropy (2.1)
$$\begin{array}{cc}\hfill S& =\frac{\pi }{12G_3}g^{\mu \nu }\frac{_3}{R_{\mu \nu }}\left[\sqrt{8G_3\mathrm{}_{Ads}(M_3+J_3)}+\sqrt{8G_3\mathrm{}_{Ads}(M_3J_3)}\right]\hfill \\ & =\frac{A_{\mathrm{BTZ}}}{4G_3}\mathrm{\Omega },\hfill \end{array}$$
where $`A_{\mathrm{BTZ}}`$ is the standard expression for the area of the BTZ black hole, i.e. a specific function of $`M_3`$, $`J_3`$, $`\mathrm{}_{Ads}`$ and $`G_3`$; and $`\mathrm{\Omega }`$ is the rescaling factor (2.1) that encodes the correction due to higher derivative terms. It is now straightforward to show that Waldโs formula
$$\begin{array}{cc}\hfill S& =\frac{1}{8G_3}_{\mathrm{hor}}๐\varphi \sqrt{g_{\varphi \varphi }}\frac{_3}{R_{\mu \alpha }}g^{\nu \beta }ฯต_{\mu \nu }ฯต_{\alpha \beta }\hfill \end{array},$$
agrees precisely with (2.1), and so with Cardyโs formula (2.1).
3. Example: M-theory on $`CY_3`$
To illustrate our approach, we now consider the example of M-theory compactified on a Calabi-Yau 3-fold $`X=CY_3`$, yielding a supergravity theory in $`D=5`$. This is a rich example that includes includes black holes in both four and five noncompact dimensions and also BPS black ring solutions.
3.1. Two-derivative action
We will follow the conventions in , to which we refer for more details. In particular, in this section we set $`G_5=\frac{\pi }{4}`$, which is convenient since it leads to integrally quantized charges $`q^I`$. The hypermultiplets are assumed to be consistently set to constant values. Then the $`D=5`$ action for the metric and vectormultiplets is given by (as in (2.1) with $`p=2`$)
$$\begin{array}{cc}\hfill _5& =R+\frac{1}{2}G_{IJ}_\mu X^I^\mu X^J+\frac{1}{4}G_{IJ}F_{\mu \nu }^IF^{J\mu \nu }+(\mathrm{fermions})+(\mathrm{higher}\mathrm{derivs})\hfill \\ \hfill S_{CS}& =\frac{1}{96\pi ^2}d^5xC_{IJK}ฯต_{\mu _1\mathrm{}\mu _5}F^{I\mu _1\mu _2}F^{J\mu _3\mu _4}A^{K\mu _5}.\hfill \end{array}$$
At first we neglect the higher derivative terms.
We consider AdS$`{}_{3}{}^{}\times S^2`$ vacua of this theory supported by magnetic flux. The magnetic charges are given by
$$q^I=\frac{1}{2\pi }_{S^2}F^I$$
where
$$F^I=\frac{q^I}{2\mathrm{}_{S^2}^2}ฯต_{S^2}$$
is interpreted microscopically as $`q^I`$ M5-branes wrapped on the $`I`$th 4-cycle of $`X`$. The scalars $`X^I`$ are taken to have constant values, fixed by the attractor mechanism to be
$$X^I=\frac{q^I}{(\frac{1}{6}C_{IJK}q^Iq^Jq^K)^{1/3}}.$$
The central charge function (2.1) becomes
$$c(\mathrm{}_{Ads},\mathrm{}_{S^2})=6\mathrm{}_{Ads}^3\mathrm{}_{S^2}^2\left(\frac{6}{\mathrm{}_{Ads}^2}+\frac{2}{\mathrm{}_{S^2}^2}\frac{G_{IJ}q^Iq^J}{4\mathrm{}_{S^2}^4}\right).$$
Extremizing, we find
$$\mathrm{}_{Ads}=2\mathrm{}_{S^2}=\frac{1}{3}\sqrt{6G_{IJ}q^Iq^J},c=4\sqrt{\frac{2}{3}}(G_{IJ}q^Iq^J)^{3/2}.$$
Special geometry relations (reviewed in ) give
$$G_{IJ}q^Iq^J=\frac{3}{2}\left(\frac{1}{6}C_{IJK}q^Iq^Jq^K\right)^{2/3},$$
which then yields
$$\mathrm{}_{Ads}=2\mathrm{}_{S^2}=\left(\frac{1}{6}C_{IJK}q^Iq^Jq^K\right)^{1/3},c=C_{IJK}q^Iq^Jq^K.$$
In an appendix we review how these relations appear in the explicit solutions.
3.2. Higher derivative corrections
Our approach makes it simple to include the effects of higher derivatives. As an example we consider adding to the action the term
$$\mathrm{\Delta }_5=A\left(R^{\mu \nu \alpha \beta }R_{\mu \nu \alpha \beta }4R^{\mu \nu }R_{\mu \nu }+R^2\right),$$
for some constant $`A`$. If we were in $`D=4`$ this term would be the Euler invariant. It is one particular higher derivative term present in M-theory on $`CY_3`$. Since other terms are present as well, we donโt expect (3.1) to capture the complete microscopic correction to the central charge or the black hole entropy. Later, we will do better, but this example is a good illustration.
Evaluated on AdS$`{}_{3}{}^{}\times S^2`$ we have
$$\mathrm{\Delta }_5=\frac{24A}{\mathrm{}_{Ads}^2\mathrm{}_{S^2}^2},$$
and so the central charge function is now
$$c(\mathrm{}_{Ads},\mathrm{}_{S^2})=6\mathrm{}_{Ads}^3\mathrm{}_{S^2}^2\left(\frac{6}{\mathrm{}_{Ads}^2}+\frac{2}{\mathrm{}_{S^2}^2}\frac{G_{IJ}q^Iq^J}{4\mathrm{}_{S^2}^4}\frac{24A}{\mathrm{}_{Ads}^2\mathrm{}_{S^2}^2}\right).$$
Extremizing and using (3.1), we find that both the radii and the central charge are corrected:
$$\begin{array}{cc}\hfill \mathrm{}_{Ads}& =\left(\frac{1}{6}C_{IJK}q^Iq^Jq^K\right)^{1/3}+\frac{4A}{\left(\frac{1}{6}C_{IJK}q^Iq^Jq^K\right)^{1/3}}+O(A^2)\hfill \\ \hfill \mathrm{}_{S^2}& =\frac{1}{2}\left(\frac{1}{6}C_{IJK}q^Iq^Jq^K\right)^{1/3}+\frac{A}{\left(\frac{1}{6}C_{IJK}q^Iq^Jq^K\right)^{1/3}}+O(A^2)\hfill \\ \hfill c& =C_{IJK}q^Iq^Jq^K+144A\left(\frac{1}{6}C_{IJK}q^Iq^Jq^K\right)^{1/3}+O(A^2).\hfill \end{array}$$
3.3. Black hole entropy
The corrected central charge (3.1) gives the black hole entropy according to (2.1)
$$S=2\pi \sqrt{\frac{c}{6}h_L}+2\pi \sqrt{\frac{c}{6}h_R}.$$
This formula could refer to either asymptotically AdS or asymptotically flat black holes, with slightly different interpretations. In the AdS case, the Virasoro generators are related to the mass and angular momentum of the black hole as in (2.1). The entropy formula (3.1) then holds for an arbitrary (i.e. nonextremal, nonsupersymmetric) BTZ black hole in this theory.
In the asymptotically flat case (3.1) still holds, but additional work is required to relate $`h_L`$ and $`h_R`$ to the asymptotic charges of the black hole. A good example is the case of M5-branes and M2-branes in M-theory on $`CY_3\times S^1`$. This was the case considered in . As before, we consider the M5-branes (with charges $`q^I`$) to wrap 4-cycles in $`CY_3`$, and in addition we take the M2-branes (with charges $`Q_I`$) to wrap 2-cycles. The asymptotically flat solution, in the case of $`CY_3=T^6`$ compactification, was given in . After taking the near horizon limit (which was all that was needed for the analysis in ) we find that AdS<sub>3</sub> becomes a extremal rotating BTZ black hole, with
$$h_R=Q_0+\frac{1}{2}C^{IJ}Q_IQ_J.$$
Here $`Q_0`$ is momentum running around the asymptotic $`S^1`$, i.e. the Kaluza-Klein electric charge, and $`C^{IJ}`$ is related to the intersection matrix of the compactification manifold. The extra term in $`h_R`$ is due to the nonzero M2-brane charges. More discussion of this effect can be found in \[3,,18,,19\]. With this identification (3.1) gives the entropy in terms of the charges measured at asymptotic infinity. As we have already stressed, once higher derivatives are included (3.1) will still hold but the central charges will be corrected.
4. Anomalies, central charges, and entropy
Up until this point we have restricted attention to cases with $`\stackrel{~}{c}=c`$, and focussed on computing the central charge and black hole entropy from the conformal anomaly. The approach is quite powerful, but for certain cases one can do even better. Indeed, one potential disadvantage is that to compute the conformal anomaly one needs to know all the terms in the action which are nonzero in the given background. But all such terms are not necessarily known when one is considering higher derivative theories. Furthermore, when $`\stackrel{~}{c}c`$, this approach is clearly insufficient to determine both central charges.
It is wise to take advantage of any other anomalies in the problem, as well as the relations among them following from symmetries. For the M5-brane example considered in the previous section gravitational anomalies are especially powerful. As we will review, there are two anomalies โ the tangent and normal bundle anomalies โ which follow from knowledge of a single term in the action, and which suffice to determine the corrections to both central charges \[4\]. So from this point of view the corrected entropy formula for the M5-brane emerges rather easily.
In the absence of gravitational anomalies the on-shell bulk supergravity action is a diffeomorphism invariant function of the boundary geometry. By the AdS/CFT correspondence it is supposed to yield the partition function of the CFT on the boundary. In the presence of gravitational anomalies, one is still led to conjecture the correspondence, but with each side suffering a loss of diffeomorphism invariance. This manifests itself in the non-conservation of the boundary stress tensor.
4.1. Some higher derivative terms
Several higher derivative terms in the effective action of M-theory are known (some relevant references are \[20,,21,,22\]). Those involving $`R^4`$ terms take the schematic forms
$$\begin{array}{cc}& t_8t_8RRRR,\hfill \\ & ฯต_{11}ฯต_{11}RRRR,\hfill \\ & ฯต_{11}C_3\left[\mathrm{Tr}R^4\frac{1}{4}(\mathrm{Tr}R^2)^2\right].\hfill \end{array}$$
For the precise definitions of these, and their coefficients in the action, see, e.g. . Of most interest to us is the term given in the third line since this term yields corrections to central charges and black hole entropy. The coefficient of this term is determined by requiring that its anomalous variation under diffeomorphisms cancel anomalous terms on the M5-brane worldvolume. We will review this in the dimensionally reduced context below.
Dimensional reduction of these terms on $`CY_3`$ leads to various higher derivative terms in $`D=5`$ , as well as shifts in the coefficients of some two-derivative terms. One of the terms that appear this way is the dimensionally continued Euler invariant (3.1). Here we focus on
$$S_{\mathrm{anom}}=\frac{c_2P_0}{48}_{M_5}Ap_1,$$
which arises from reduction of the third term in (4.1). In (4.1) $`p_1`$ is the first Pontryagin class
$$p_1=\frac{1}{2}\left(\frac{1}{2\pi }\right)^2\mathrm{Tr}RR.$$
We take the M5-brane to wrap the cycle $`P_0=P_0^I\sigma _I`$, where $`\{\sigma _I\}`$ form a basis for $`H_4(X,Z)`$. The choice of 4-cycle then determines a particular linear combination of gauge fields in five dimensions, which was denoted by $`A`$ in (4.1). Finally, $`c_2`$ is the second Chern class of $`X`$, which has coefficients $`c_{2I}`$ in its expansion with respect to chosen basis for $`H^4(X,Z)`$.
After reduction on $`X`$, the wrapped M5-branes correspond to a string in five dimensions, on which lives a chiral CFT. As explained in , the term (4.1) cancels the gravitational anomalies of the CFT.<sup>6</sup> More precisely, it cancels the part of the anomaly linear in M5-brane charge. There are also cubic terms which weโll discuss momentarily.
4.2. Anomalies
Anomaly cancellation occurs via the inflow mechanism, as we now recall. First of all, since $`A`$ is ill-defined in the presence of a magnetic charge, (4.1) should really be written after performing an integration by parts and discarding the boundary term. So the actual term of interest is
$$S=\frac{1}{2}\left(\frac{1}{2\pi }\right)^2\frac{c_2P_0}{48}_{M_5}F\omega _3,$$
where $`\omega _3`$ is the Lorentz Chern-Simons 3-form:
$$\omega _3=\mathrm{Tr}(\omega d\omega +\frac{2}{3}\omega ^3),$$
with $`\omega `$ being the spin connection. Now under a local Lorentz transformation parameterized by $`\mathrm{\Theta }`$,
$$\delta \omega =d\mathrm{\Theta }+[\omega ,\mathrm{\Theta }],$$
the action changes as
$$\delta S_{\mathrm{bulk}}=\frac{1}{2}\left(\frac{1}{2\pi }\right)^2\frac{c_2P_0}{48}_{M_5}F\mathrm{Tr}(d\mathrm{\Theta }d\omega ).$$
At this point we encounter two distinct interpretations. The approach of was to consider the magnetic string essentially as a pointlike defect placed in an ambient space. The presence of the magnetic string corresponds to $`dF`$ having delta function support at the location of the string. In this approach, one integrates (4.1) by parts, and then uses the delta function to perform the integral over the directions transverse to the string. What remains is an integral over the string worldvolume, which cancels a term coming from the variation of the path integral over the string degrees of freedom.
The interpretation in our case is somewhat different. We are dealing with a smooth supergravity solution with geometry AdS$`{}_{3}{}^{}\times S^2\times X`$ and $`dF=0`$. The branes have been replaced by flux. Instead of cancelling the anomaly at the brane location, we get a contribution at the AdS boundary. It is clear that this contribution yields the anomalous variation of the CFT on the boundary. This mechanism is well known in AdS/CFT, going back to the treatment of the R-symmetry anomaly of $`๐ฉ=4`$ super-Yang-Mills in . In particular, (4.1) gives the boundary term
$$\delta S_{\mathrm{bulk}}=\frac{1}{2}\left(\frac{1}{2\pi }\right)^2\frac{c_2P_0}{48}_{M_5}F\mathrm{Tr}(\mathrm{\Theta }d\omega ).$$
We consider pure AdS$`{}_{3}{}^{}\times S^2`$ with the components of $`F`$ given by (3.1). Integrating (4.1) over the $`S^2`$ we obtain
$$\delta S_{\mathrm{bulk}}=\frac{1}{2}\frac{c_2q}{48}\frac{1}{2\pi }_{\mathrm{AdS}_3}\mathrm{Tr}(\mathrm{\Theta }d\omega ).$$
Importantly the matrices $`\mathrm{\Theta }`$ and $`d\omega `$ are still by $`5\times 5`$; they include indices along the $`AdS_3`$ boundary and also in the radial and $`S^2`$ directions. Accordingly, we can study two kinds of anomalies, associated with diffeomorphisms that map the boundary to itself (tangent bundle anomaly) and with diffeomorphisms acting on the vectors normal to the boundary (normal bundle anomaly). From the point of view of the D=2 CFT, these are gravitational and $`SU(2)_R`$ symmetry anomalies.
In the CFT the gravitational anomaly is obtained via descent from $`I_4=2\pi \frac{1}{24}(c\stackrel{~}{c})p_1`$, yielding
$$\delta S_{\mathrm{CFT}}=\frac{c\stackrel{~}{c}}{48}\frac{1}{2\pi }_{\mathrm{AdS}_3}\mathrm{Tr}(\mathrm{\Theta }d\omega ).$$
Equating this with (4.1) we find <sup>7</sup> There are two (cancelling) sign changes relative to the anomaly inflow in : the boundary at infinity has normal opposite to that of a defect in bulk; and also we are equating the two anomalies, as in AdS/CFT, rather than cancelling them, as in the anomaly inflow.
$$\stackrel{~}{c}c=\frac{1}{2}c_2q.$$
The computation of the normal bundle anomaly is similar. In this case the corresponding CFT anomaly is in the $`SU(2)_R`$ symmetry which, in our conventions, acts on the leftmovers so that the normal bundle anomaly contributes
$$c_{\mathrm{lin}}=\frac{1}{2}c_2q,$$
to $`c`$. The form of (4.1) and (4.1) are the same because these contributions arise from the same anomaly (4.1), decomposed into tangent and normal bundle part, and interpreted appropriately. These expressions capture the linear contributions to the central charges exactly. However, there are also $`๐ช(q^3)`$ contributions (see (3.1)) coming from the two-derivative part of the action, and so altogether we have
$$c=C_{IJK}q^Iq^Jq^K+\frac{1}{2}c_2q,\stackrel{~}{c}=C_{IJK}q^Iq^Jq^K+c_2q.$$
These are the results found in \[3,,4\].
The $`C_{IJK}q^Iq^Jq^K`$ contributions are actually quite subtle in the context of anomaly cancellation for M5-branes viewed as pointlike defects \[25,,26\]. The $`๐ช(q^3)`$ contribution to the normal bundle anomaly requires a subtle modification of the M-theory Chern-Simons term. By contrast, in the context of the smooth supergravity backgrounds considered here, this contribution is simple to understand because it comes from the leading two-derivative part of the action. We have phrased this in terms of computing the conformal anomaly, but we could have equally well computed the normal bundle anomaly directly. In our problem supersymmetry related these anomalies to one another, so a computation of either suffices.
4.3. Application: heterotic strings
An important special case of our computations is M-theory on $`K3\times T^2`$. Consider an $`M5`$-brane wrapped around the $`K3`$ and transverse to the $`T^2`$. In this case we have $`C_{IJK}q^Iq^Jq^K=0`$, and $`c_2q=24`$ because the Euler number of $`K3`$ is $`24`$. Therefore, (4.1) gives $`c=12`$ and $`\stackrel{~}{c}=24`$. These are the correct assignments for the heterotic string which, indeed, is a dual representation of an M5-brane on $`K3\times T^2`$. Thus we find the central charges of both sides of the heterotic strings; so we are sensitive to all excitations, rather than just the BPS states. In particular, from the Cardy formula (1.1) we get the entropy of nonsupersymmetric small black holes in agreement with the non-BPS entropy of the heterotic string. Although the formulae (4.1) have been known for some time, this agreement apparently has not been noticed before.
The recovery of both the central charges of the heterotic string sounds like an extremely powerful and surprising result when put, as above, in terms of the near horizon geometry, corrected by higher derivative terms in the action. However, from another point of view the agreement is almost trivial: a heterotic string propagating in a curved background suffers gravitational anomalies, because $`c\stackrel{~}{c}`$, and these must be cancelled by bulk terms, via the inflow mechanism. This works, of course; indeed, it would be one way to derive the anomalous coupling $`S_{\mathrm{anom}}`$, including the coefficient. Related to this, heterotic string theory in $`AdS_3\times N`$ has linear corrections that precisely reproduce the ones seen here . From either point of view, we should hardly be surprised when these couplings give back the heterotic string, when interpreted in terms of the near horizon geometry and its boundary at infinity. On the other hand, the fact that the agreement is essentially automatic does not make it any less valid, nor any less interesting.
4.4. Application: inclusion of angular momentum
Consider the BPS states of a heterotic string wrapped on an $`S^1`$ in $`T^5`$, with fixed winding number, rightmoving momentum, and angular momentum in a given 2-plane. The microscopic entropy is known to be
$$S=4\pi \sqrt{N_wN_pJ}.$$
Geometrically, the states correspond to rotating helical strings. The maximal angular momentum, $`J=N_wN_p`$ is attained when all the momentum is placed in oscillators of the lowest mode number, with polarizations in the angular momentum 2-plane. The profile of the helix is then a circle. If we decrease $`J`$ from its maximal value while holding $`N_{w,p}`$ fixed, then there are additional microstates in which the string wiggles away from its circular shape, either in the noncompact or internal dimensions. These additional states give rise to the entropy (4.1). As we will now argue, there is also a black object with the same charges and whose entropy agrees with (4.1).
Using heterotic/IIA duality, and lifting to M-theory, the configuration above describes a rotating helical M5-brane wrapped on $`K3\times T^2`$. The supergravity solution will have near horizon limit AdS$`{}_{3}{}^{}\times S^2\times K^3\times T^2`$. The rightmoving central charge is $`\stackrel{~}{c}=24N_w`$, since the M5-brane wraps $`K3`$ $`N_w`$ times. The level number $`h_R`$ appearing in the near horizon region differs from the total rightmoving momentum $`N_p`$ measured at infinity. Although we have not checked this explicitly in the present context, in other very similar cases (see, e.g. ) one finds that $`h_R`$ is obtained by subtracting from $`N_p`$ the momentum used up by the gyration:
$$h_R=N_pN_{\mathrm{gyro}}.$$
The mechanical gyration of the string carries momentum and angular momentum related by $`J_{\mathrm{gyro}}=\frac{\lambda }{2\pi }P_{\mathrm{gyro}}`$, where $`\lambda `$ is the wavelength of the gyration. Since our brane is wrapped $`N_w`$ times around a circle of radius $`R`$, the largest possible wavelength (which yields the highest entropy) is $`\lambda =2\pi RN_w`$, and so
$$h_R=N_p\frac{J}{N_w}.$$
The near horizon geometry will thus be a BTZ black hole with entropy given by the Cardy formula as
$$S=2\pi \sqrt{\frac{c}{6}h_R}=4\pi \sqrt{N_wN_pJ},$$
in agreement with (4.1). The black object could be thought of as a โsmallโ black ring.
With the replacements $`N_wN_5`$ and $`N_pN_1`$, (4.1) also gives the ground state entropy of the D1-D5 system on K3. Indeed there is a duality chain that relates the two systems. The M-theory configuration can be interpreted as IIA on $`K3\times S^1`$ with NS5-branes wrapped on the compact space and carrying momentum on the $`S^1`$. A T-duality on the $`S^1`$ followed by S-duality then yields the D1-D5 system. The ground state entropy of the D1-D5 system has recently been obtained in a different approach by Iizuka and Shigemori .
5. Discussion: corrections to all orders in $`1/Q`$ and beyond
The black hole entropy discussed in this paper has been presented in all cases in terms of the Cardy formula which is essentially semi-classical. It is interesting to think about how further corrections might be included. In particular, recent work has shown that is possible to reproduce the BPS entropy of the heterotic string to all orders in an expansion in inverse powers of the charges . Let us now show how our approach is naturally extended to include this agreement.
In evaluating the black hole partition function in section 2.3 we specified the black hole temperature $`\beta `$ and chemical potential $`\mu `$, which are conjugate to the mass and angular momentum of the black hole. We now note that we could also specify the values of any conserved charges or, alternatively, the boundary values of the corresponding gauge potentials.
In the case of M-theory on $`K3\times T^2`$ we have gauge fields $`A^I`$ that couple to charges $`Q_I`$ that correspond to wrapped M2-branes. We thus need the Euclidean action of black holes carrying these charges (as well as the M5-brane charge $`q^I`$). According to (3.1) this just gives a shift in $`h_R`$ which, from (2.1), changes the action to
$$I_{BH}(\overline{\tau },Q_I)=\frac{i\pi }{12}\frac{\stackrel{~}{c}}{\overline{\tau }}+2\pi i\overline{\tau }\frac{1}{2}C^{IJ}Q_IQ_J.$$
To focus on BPS states we set the left moving temperature to zero: $`\frac{1}{\tau }=0`$. Semi-classically, the potentials are related to the charges as
$$\varphi ^I=\frac{1}{\pi }\frac{I_{BH}}{Q_I},$$
so
$$I_{BH}(\varphi ^0,\varphi ^I)=\frac{\pi \stackrel{~}{c}}{6\varphi ^0}\frac{\pi }{2}\frac{C_{IJ}\varphi ^I\varphi ^J}{\varphi ^0},$$
where we renamed the right moving temperature
$$\varphi ^0=\frac{2}{i}\overline{\tau }.$$
The potentials $`\varphi ^0,\varphi ^I`$ defined in (5.1) and (5.1) were designed to agree with the conventions in the topological string literature \[11,,10\] which amounts to the equality
$$e^{\pi (Q_0\varphi ^0+q_I\varphi ^I)}=e^{2\pi i\overline{\tau }(Q_0+Q_IA_t^I)}.$$
Now, the expression (5.1) for the action is precisely the same as (the negative of) the free energy $`_{\mathrm{pert}}`$ appearing in (2.6) of , and at this point we can simply follow their analysis. In particular, the degeneracy of states $`\mathrm{\Omega }(Q_0,Q_I)`$ with the specified charge is given by the relation between the canonical and microcanonical ensembles
$$\mathrm{\Omega }(Q_0,Q_I)=๐\varphi e^{I_{BH}(\varphi ^0,\varphi ^I)+\pi (Q_0\varphi ^0+Q_I\varphi ^I)}.$$
Carrying out the integral yields a Bessel function which correctly accounts for the number of heterotic string states to all orders in inverse powers of charges. We refer the reader to for the details (and also to for an alternative approach). The point we wish to emphasize here is that the power law corrections to the black hole entropy are semi-classical in nature, and so they can be captured by our approach.
Ultimately, several other corrections must be included in order to account completely for the microscopic degeneracies including exponentially suppressed terms. For example, there are contributions from world-sheet and brane instantons and also, more dramatically, from semi-classical geometries distinct from the one contributing to the leading term. These corrections remain to be understood, both in the 4D topological string approach, and in the approach considered here.
Acknowledgments: We would like to thank Masaki Shigemori for discussions and an advance copy of . We thank the organizers of the Amsterdam workshop on Strings and Branes for hospitality as this work was completed. Also thanks to P. van Hove for correspondence. The work of PK is supported in part by the NSF grant PHY-00-99590. The work of FL is supported in part by the DoE.
Appendix A. Asymptotically flat M5-brane solution
For convenience, we give here the asymptotically flat solution representing M5-branes wrapped on 4-cycles of $`CY_3`$. We follow the conventions of . The $`CY_3`$ has harmonic $`(1,1)`$ forms $`J_I`$ and Kahler moduli $`X^I`$. The metric and 3-form are
$$\begin{array}{cc}\hfill ds^2& =ds_5^2+ds_{CY_3}^2\hfill \\ \hfill ๐& =A^IJ_I\hfill \end{array}$$
with
$$\begin{array}{cc}\hfill ds_5^2& =(\frac{1}{6}C_{IJK}H^IH^JH^K)^{1/3}(dt^2+dx_4^2)+(\frac{1}{6}C_{IJK}H^IH^JH^K)^{2/3}(dr^2+r^2d\mathrm{\Omega }_2^2)\hfill \\ \hfill A^I& =\frac{1}{2}q^I(1+\mathrm{cos}\theta )d\varphi \hfill \\ \hfill X^I& =\frac{H^I}{(\frac{1}{6}C_{IJK}H^IH^JH^K)^{1/3}}\hfill \\ \hfill H^I& =\overline{X}^I+\frac{q^I}{2r}.\hfill \end{array}$$
To examine the near horizon geometry we write
$$r=\frac{\frac{1}{6}C_{IJK}q^Iq^Jq^K}{2z^2}.$$
For $`z\mathrm{}`$ we then find the following AdS$`{}_{3}{}^{}\times S^2\times CY_3`$ geometry
$$\begin{array}{cc}\hfill ds_5^2& =\mathrm{}_{Ads}^2\frac{dt^2+dx_4^2+dz^2}{z^2}+\mathrm{}_{S^2}^2d\mathrm{\Omega }_2^2\hfill \\ \hfill X^I& =\frac{q^I}{(\frac{1}{6}C_{IJK}q^Iq^Jq^K)^{1/3}}\hfill \end{array}$$
with
$$\mathrm{}_{Ads}=2\mathrm{}_{S^2}=\left(\frac{1}{6}C_{IJK}q^Iq^Jq^K\right)^{1/3}.$$
The Brown-Henneaux computation of the central charge applied to this case gives
$$c=C_{IJK}q^Iq^Jq^K.$$
(A.1) and (A.1) are in perfect agreement with (3.1).
References
relax R. M. Wald, โBlack hole entropy is the Noether charge,โ Phys. Rev. D 48, 3427 (1993) \[arXiv:gr-qc/9307038\]. R. Wald, Phys. Rev. D 48 R3427 (1993); V. Iyer and R. M. Wald, โSome properties of Noether charge and a proposal for dynamical black hole entropy,โ Phys. Rev. D 50, 846 (1994) \[arXiv:gr-qc/9403028\]. โA Comparison of Noether charge and Euclidean methods for computing the entropy of stationary black holes,โ Phys. Rev. D 52, 4430 (1995) \[arXiv:gr-qc/9503052\]. relax K. Behrndt, G. Lopes Cardoso, B. de Wit, D. Lust, T. Mohaupt and W. A. Sabra, โHigher-order black-hole solutions in N = 2 supergravity and Calabi-Yau string backgrounds,โ Phys. Lett. B 429, 289 (1998) \[arXiv:hep-th/9801081\]; G. L. Cardoso, B. de Wit, J. Kappeli and T. Mohaupt, โExamples of stationary BPS solutions in N = 2 supergravity theories with $`R^2`$-interactions,โ Fortsch. Phys. 49, 557 (2001) \[arXiv:hep-th/0012232\]; โStationary BPS solutions in N = 2 supergravity with $`R^2`$ interactionsโ, JHEP 0012, 019 (2000) \[arXiv:hep-th/0009234\]; โSupersymmetric black hole solutions with $`R^2`$ interactionsโ, \[arXiv:hep-th/0003157\]; G. Lopes Cardoso, B. de Wit and T. Mohaupt, โArea law corrections from state counting and supergravityโ, Class. Quant. Grav. 17, 1007 (2000) \[arXiv:hep-th/9910179\]; โMacroscopic entropy formulae and non-holomorphic corrections for supersymmetric black holesโ, Nucl. Phys. B 567, 87 (2000) \[arXiv:hep-th/9906094\]; โDeviations from the area law for supersymmetric black holesโ, Fortsch. Phys. 48, 49 (2000) \[arXiv:hep-th/9904005\]; โCorrections to macroscopic supersymmetric black-hole entropyโ, Phys. Lett. B 451, 309 (1999) \[arXiv:hep-th/9812082\]. relax J. M. Maldacena, A. Strominger and E. Witten, โBlack hole entropy in M-theoryโ, JHEP 9712, 002 (1997); \[arXiv:hep-th/9711053\]. relax J. A. Harvey, R. Minasian and G. W. Moore, โNon-abelian tensor-multiplet anomalies,โ JHEP 9809, 004 (1998) \[arXiv:hep-th/9808060\]. relax A. Dabholkar, โExact counting of black hole microstatesโ, \[arXiv:hep-th/0409148\], A. Dabholkar, R. Kallosh and A. Maloney, โA stringy cloak for a classical singularityโ, JHEP 0412, 059 (2004), \[arXiv:hep-th/0410076\]. relax S. D. Mathur,โThe fuzzball proposal for black holes: An elementary review,โ arXiv:hep-th/0502050. relax A. Strominger, โBlack hole entropy from near-horizon microstatesโ, JHEP 9802, 009 (1998); \[arXiv:hep-th/9712251\]; V. Balasubramanian and F. Larsen, โNear horizon geometry and black holes in four dimensionsโ, Nucl. Phys. B 528, 229 (1998); \[arXiv:hep-th/9802198\]. relax J. D. Brown and M. Henneaux, โCentral Charges In The Canonical Realization Of Asymptotic Symmetries: An Example From Three-Dimensional Gravity,โ Commun. Math. Phys. 104, 207 (1986). relax A. Sen, โHow does a fundamental string stretch its horizon?,โ JHEP 0505, 059 (2005) \[arXiv:hep-th/0411255\]; โBlack holes, elementary strings and holomorphic anomaly,โ arXiv:hep-th/0502126. โStretching the horizon of a higher dimensional small black hole,โ arXiv:hep-th/0505122. relax A. Dabholkar, F. Denef, G. W. Moore and B. Pioline, โExact and asymptotic degeneracies of small black holesโ, \[arXiv:hep-th/0502157\]. relax H. Ooguri, A. Strominger and C. Vafa, โBlack hole attractors and the topological stringโ, Phys. Rev. D 70, 106007 (2004), \[arXiv:hep-th/0405146\]; E. Verlinde, โAttractors and the holomorphic anomalyโ, \[arXiv:hep-th/0412139\]; relax M. Henningson and K. Skenderis, โThe holographic Weyl anomaly,โ JHEP 9807, 023 (1998) \[arXiv:hep-th/9806087\]. relax V. Balasubramanian and P. Kraus, โA stress tensor for anti-de Sitter gravity,โ Commun. Math. Phys. 208, 413 (1999) \[arXiv:hep-th/9902121\]. relax J. M. Maldacena and A. Strominger, โAdS(3) black holes and a stringy exclusion principle,โ JHEP 9812, 005 (1998) \[arXiv:hep-th/9804085\]; R. Dijkgraaf, J. M. Maldacena, G. W. Moore and E. Verlinde, โA black hole farey tail,โ arXiv:hep-th/0005003. relax H. Saida and J. Soda, โStatistical entropy of BTZ black hole in higher curvature gravity,โ Phys. Lett. B 471, 358 (2000) \[arXiv:gr-qc/9909061\]. relax P. Kraus and F. Larsen, โAttractors and black rings,โ arXiv:hep-th/0503219. relax I. Bena, โSplitting hairs of the three charge black hole,โ Phys. Rev. D 70, 105018 (2004) \[arXiv:hep-th/0404073\]. relax F. Larsen and E. J. Martinec, โU(1) charges and moduli in the D1-D5 system,โ JHEP 9906, 019 (1999) \[arXiv:hep-th/9905064\]; โCurrents and moduli in the (4,0) theory,โ JHEP 9911, 002 (1999) \[arXiv:hep-th/9909088\]. relax R. Emparan and D. Mateos, โOscillator level for black holes and black rings,โ arXiv:hep-th/0506110. relax M. B. Green and J. H. Schwarz, โSupersymmetrical Dual String Theory. 2. Vertices And Trees,โ Nucl. Phys. B 198, 252 (1982); D. J. Gross and E. Witten, โSuperstring Modifications Of Einsteinโs Equations,โ Nucl. Phys. B 277, 1 (1986); W. Lerche, B. E. W. Nilsson and A. N. Schellekens, โHeterotic String Loop Calculation Of The Anomaly Cancelling Term,โ Nucl. Phys. B 289, 609 (1987); M. J. Duff, J. T. Liu and R. Minasian, โEleven-dimensional origin of string / string duality: A one-loop test,โ Nucl. Phys. B 452, 261 (1995) \[arXiv:hep-th/9506126\]; M. B. Green, M. Gutperle and P. Vanhove, โOne loop in eleven dimensions,โ Phys. Lett. B 409, 177 (1997) \[arXiv:hep-th/9706175\]; J. G. Russo and A. A. Tseytlin, โOne-loop four-graviton amplitude in eleven-dimensional supergravity,โ Nucl. Phys. B 508, 245 (1997) \[arXiv:hep-th/9707134\]; P. S. Howe and D. Tsimpis, โOn higher-order corrections in M theory,โ JHEP 0309, 038 (2003) \[arXiv:hep-th/0305129\]. relax A. A. Tseytlin, โR\**4 terms in 11 dimensions and conformal anomaly of (2,0) theory,โ Nucl. Phys. B 584, 233 (2000) \[arXiv:hep-th/0005072\]. relax P. S. Howe and D. Tsimpis, JHEP 0309, 038 (2003) \[arXiv:hep-th/0305129\]. relax I. Antoniadis, S. Ferrara, R. Minasian and K. S. Narain, โR\**4 couplings in M- and type II theories on Calabi-Yau spaces,โ Nucl. Phys. B 507, 571 (1997) \[arXiv:hep-th/9707013\]. relax E. Witten, โAnti-de Sitter space and holography,โ Adv. Theor. Math. Phys. 2, 253 (1998) \[arXiv:hep-th/9802150\]. relax E. Witten, โFive-brane effective action in M-theory,โ J. Geom. Phys. 22, 103 (1997) \[arXiv:hep-th/9610234\]. relax D. Freed, J. A. Harvey, R. Minasian and G. W. Moore, โGravitational anomaly cancellation for M-theory fivebranes,โ Adv. Theor. Math. Phys. 2, 601 (1998) \[arXiv:hep-th/9803205\]. relax D. Kutasov, F. Larsen and R. G. Leigh, โString theory in magnetic monopole backgrounds,โ Nucl. Phys. B 550, 183 (1999) \[arXiv:hep-th/9812027\]. relax J. G. Russo and L. Susskind, โAsymptotic level density in heterotic string theory and rotating black holes,โ Nucl. Phys. B 437, 611 (1995) \[arXiv:hep-th/9405117\]. relax D. Marolf and B. C. Palmer, โGyrating strings: A new instability of black strings?,โ Phys. Rev. D 70, 084045 (2004) \[arXiv:hep-th/0404139\]. relax N. Iizuka and M. Shigemori, โA Note on D1-D5-J System and 5D Small Black Ring,โ arXiv:hep-th/0506215.
|
warning/0506/astro-ph0506483.html
|
ar5iv
|
text
|
# A continuous Flaring- to Normal-branch transition in Sco~X-1
## 1 Introduction
Neutron star Low Mass X-ray Binaries are usually classified in two classes, the Z and the atoll sources; these names are based on the shape that individual sources describe in the color-color diagrams as a result of changes taking place in the sources (CDs, Hasinger & van der Klis (1989), van der Klis (2005)). The atoll sources trace a pattern consisting of a single curved branch (banana branch), together with a smaller โislandโ, while the pattern traced by Z sources is characterized by three branches called horizontal branch, normal branch and flaring branch respectively. Three types of Low-frequency Quasi-Periodic Oscillations (LFQPOs) have been associated with the position along the Z: the HBOs (horizontal branch oscillations), the NBOs (normal branch oscillations) and the FBOs (flaring branch oscillations).
NBOs with frequencies between 4.5 and 7 Hz have been reported from all Z sources. Their fractional rms amplitude is typically between 1 and 3 % in the 1-10 keV band, and are strongest near the middle of the NB. FBOs were seen only in two sources (Sco~X-1, Priedhorsky et al. (1986); GX~17+2, Homan et al. (2002)), while a broad excess that may be due to an FBO peak moving rapidly in frequency as the source moves rapidly up and down the FB has been reported in other sources (see e.g. GX~5-1, Jonker et al. (2002)). FBOs occur on a small part ($``$10 % of the total extent) of the Flaring Branch (FB) nearest the NB, and have frequencies in the range $``$6-25 Hz, increasing from the NB-FB vertex through the FB. With increasing frequency, the fractional amplitude of the FBOs remains approximately constant, while its width increases until the peak becomes too broad to distinguish from the broad band noise in the power density spectrum (the so-called high frequency noise, HFN).
NBOs and FBOs are thought to be physically related, since in Sco~X-1 the NBO frequency joins to the FBO frequency as the source moves from the NB to the FB. However, the increase from $``$6 Hz to $``$10 Hz occurs in a very short segment of the Z track located just at the vertex of the NB-FB junction, and is at present still unresolved (Dieters & van der Klis (2000)). The transition is even less clear in GX~17+2: Homan et al. (2002)) found that near the NB-FB vertex the frequency of the FBO was a factor of $``$2 higher than that of the NBO, which could mean that the NBO and FBO are harmonically related. However, by inspecting all single dynamical power spectra of observations with values of the curvilinear coordinate $`S_z`$ (tracking the position along the Z pattern, see van der Klis (1995)) around 2 (corresponding to the NB-FB vertex), they found in some cases QPOs with intermediate frequencies ($``$10 Hz), suggesting that the frequency does not jump directly from $``$7 to $``$14 Hz. However no clear transitions were found.
At present, it is currently believed that NBOs and FBOs are different manifestations of the same phenomenon, the properties of which rapidly change at around the position of the lower vertex in the Z track. However, the transition between them has not yet been unambiguously resolved. In this paper we present the analysis of a RXTE observation of Sco~X-1 in which for the first time a clear continuous fast transition from a Normal Branch Oscillation to a Flaring Branch Oscillation is seen and which is associated with the NB-FB vertex in the Z track.
## 2 Data analysis
We analyzed a RXTE/PCA observation of Sco~X-1, from the RXTE public archive, made on MJD 50230 (1996-05-27). The observation consists of three continuous data intervals of $``$1600, $``$550 and $``$2000 s duration respectively. All five PCA units were on during the whole observation. The PCA data were obtained in several simultaneous different modes (see Tab. 1). During the first interval, the pointing direction was offset by $``$0.3 deg, while in the other two intervals the offset was reduced to $``$0.005 deg. For this reason we could not build the expected Z-track across the whole observation. We concentrated our analysis on the first of the three data intervals, during which a variable low frequency ($``$6-15 Hz) QPO was detected.
We used Standard2 data to produce a color-color diagram (CD). For each 16 s data segment (the intrinsic resolution of the Standard2 mode), we defined two colors as ratios of count rates in two different energy bands. The energy bands used for the colors (soft and hard color) are given in Table 2. The CD is shown in Figure 1: Sco~X-1 was in its Flaring Branch (FB) at the beginning of the observation, and moved into the Normal Branch (NB) after $``$200 seconds.
Since the Z track is not complete, we could not use the usual rank number $`S_z`$ used in literature, which is defined using both vertices (Hasinger et al. 1990). In order to check the presence of a QPO along the Z track, we divided the CD in 14 intervals and calculated a power spectrum (with a Nyquist frequency of 128 Hz) for each of them by averaging power spectra calculated every 16 s data interval. The two power spectra corresponding to the extremes of the CD (#1 and #14), together with three power spectra approximately corresponding to the vertex (#9, #10 and #11) are shown in Figure 2.
We then fitted each power spectrum with a combination of a simple power law and a broad Lorentzian shape, approximating the broad band noise, and a narrow Lorentzian shape, approximating the QPO peak. For the intervals in the FB an additional narrow ($`\mathrm{\Delta }\nu /\nu `$1) Lorentzian shape was needed, in order to better approximate the asymmetric profile of the FBO peak. PDS fitting was carried out with the standard Xspec fitting package, by using a one-to-one energy-frequency conversion and a unit response. Because of the low intensity of the broad band noise, the parameters of the power law and those of the broad Lorentzian were not well constrained, and their behavior could not be studied. However, since the intensity of the QPO peak was high in all intervals, the uncertainties in the underlying continuum did not impact much on the estimate of the peak parameters.
In the upper panel of Figure 3 we plot the QPO centroid frequency as a function of the interval number. The evolution of the QPO peak is evident: the centroid frequency was $``$6 Hz when the source was in the upper part of the visible NB, slowly increased up to $``$7.5 Hz when the source approached the FB-NB vertex, then quickly increased up to $``$15 Hz in the FB. In the bottom panel of Figure 3, the total fractional rms and the QPO fractional rms are shown as a function of the interval number. Both slowly increase through the NB towards the vertex, and reach their maximum in the FB. The total rms slightly decreases after the vertex, while the QPO rms is consistent with remaining constant.
The transition appears to be very fast, as a result of the fast passage of the source through the NB-FB vertex. In order to track more finely the QPO frequency as a function of time, we produced a dynamical power spectrum for the whole observation by calculating a power spectrum every 4 seconds over PCA channels 0-35 (corresponding approximately to the 2-13 keV energy range) and with a time resolution of 1/128 s (corresponding to a Nyquist frequency of 64 Hz). The dynamical power spectrum and the correspondent light curve of the first 500 seconds of the observation, when the frequency transition occurred, are shown in Figure 4.
The transition is clearly visible in the dynamical power spectrum: the QPO peak at the beginning of the observation (when the source is in its FB) is at frequencies $``$14 Hz, has a maximum at $``$15 Hz after less than 100 seconds, and rapidly decreases (corresponding to the FB-NB vertex) after few hundred seconds to $``$6 Hz. The peak remains then visible in the dynamical power spectrum at $``$6 Hz in the remaining part of the observation, during which the source is in the NB.
## 3 Discussion
We presented the first continuous monitoring of a rapid transition between Flaring Branch and Normal Branch Oscillations. Even though the transition could not be continuously tracked through the Z pattern in the CD, the high statistics of the data makes it visible in the dynamical power spectrum, where it is resolved to take place in $``$100 seconds.
The presence of the NBO/FBO in the Z sources has often been related to the fact that these sources accrete at near-Eddington mass accretion rates. It is thought that at these high mass accretion rates a significant fraction of the accretion flow is in the form of a thick, perhaps nearly spherical flow. Also, the effects of radiation pressure are thought to play an important role in this regime. Fortner, Lamb & Miller (1989) suggested that the NBOs are the result of a radiation force/opacity feed-back mechanism within a spherical flow region. For reasonable physical parameters, the frequency of the NBOs is predicted to be $``$6 Hz at the onset of the process, and then to increase to $``$10 Hz as the luminosity approaches $`L_{Edd}`$. These oscillations can furthermore excite other modes with similar frequencies, which are expected to increase as the luminosity approaches and then exceeds $`L_{Edd}`$. However, it is not clear if the similarity of these two oscillation types can be as high as required to explain the smoothness of the transition reported in Sco~X-1 (see Figure 4). Another model for the NBOs was proposed by Alpar et al. (1992), in which the NBO frequency is basically that of sound waves in a thickened accretion disk. However, this model does not explain how the NBO changes into the FBO. A strong challenge for these models is the recent discovery of $``$6 Hz QPOs, with properties similar to those of NBOs, in atoll sources (see e.g. Wijnands, van der Klis & Rijkhorst 1999, Wijnands & van der Klis 1999, Belloni, Parolin & Casella 2004), in which the mass accretion rate is always lower than the Eddington limit.
Casella, Belloni & Stella (2005) recently reviewed and further extended the similarities between the LFQPOs observed in Z sources and the three main types of LFQPOs observed in Black-Hole Candidates (BHCs, see the cited work and references therein), and proposed a one-to-one association where the C-, B- and A-Type observed in BHCs correspond respectively to the HBO, NBO and FBO observed in Z sources. Whatever the physical mechanisms that determines these oscillations are, the presence of such mechanisms in both types of compact objects would clearly favor a disk origin for these oscillations, ruling out all models that involve any interaction with the surface or the magnetosphere of the neutron star (see e.g. Stella, Vietri & Morsink 1999; see also Kluzniak 2004).
In this context, the smooth continuous transition between the NBOs and the FBOs appears to be very different from the fast, unresolved transitions observed between type-B and type-A QPOs in BHCs (Nespoli et al. 2003; Casella et al. 2004), i.e. the equivalent of NBOs and FBOs in this analogy. At present it is in fact not clear whether type-B and type-A QPOs originate from the same physical phenomenon, the properties of which can change in a short time scale (less than a few tens of seconds in the known cases), or if they are two distinct phenomena, one of which (type-A) is too weak to be detected when the other (type-B) is present (see e.g. Nespoli et al. 2003). Available data at present do not permit us to address this issue, and more observations (and/or better statistics) are needed.
###### Acknowledgements.
This work was partially supported by the Italian Ministry of Education, University and Research under CO-FIN grants 2002027145 and 2003027534.
|
warning/0506/nlin0506018.html
|
ar5iv
|
text
|
# Quantum dynamical echo in two-level systems
## I Introduction
Since the overlap of states is invariant under unitary quantum dynamics, a new approach is necessary for measuring sensitivity of the evolution with respect to perturbation. A seminal idea for such investigation is to consider two evolutions of a single initial state corresponding to two slightly different Hamiltonians, as proposed by Peres peres84 . Let us consider a quantum system, which is initially in an arbitrary state $`|\alpha `$, evolves under the Hamiltonians $`H`$ and $`\stackrel{~}{H}=H+ฯตV`$ with $`ฯต`$ being perturbation parameter. If $`U(t)`$ and $`\stackrel{~}{U}(t)`$ are the corresponding evolution operators, the overlap intensity of the two evolved states or quantum fidelity
$$f(t)=|\alpha |\stackrel{~}{U}^{}(t)U(t)|\alpha |^2$$
(1)
is a measure of quantum sensitivity. This quantity can also be interpreted as the overlap between an evolved state, which is obtained by forward evolving in time under $`H`$ followed by backward evolution for the same time $`t`$ under $`\stackrel{~}{H}`$, and the initial state. Latter interpretation suggests that quantum fidelity (also called as Loschmidt echo) measures the accuracy to which a quantum state can be recovered by inverting the perturbed evolution.
Only in recent years, Peresโ proposal has evoked enormous interest with different interpretations. In particular, exponential decay of $`f(t)`$ in time has been identified as quantum manifestation of classically chaotic systems jala01 . Since fidelity is intimately related to decoherence cucc03 , the decaying nature could be a concern for practical implementation of algorithms to perform quantum computation. Loschmidt echo is realized as spin (polarization) echo in nuclear magnetic resonance experiments zhang92 , and attenuation of echo amplitude is found to be useful tool in characterizing many-body quantum systems usa98 . Fidelity is also a relevant quantity in spectroscopic studies on quantum stability of optically trapped two-level atoms ander03 .
While for most of the dynamical systems the quantum fidelity is either a decaying function or exhibits attenuated oscillations in time, for the harmonic oscillator it is periodic in first order perturbative approximation sankar03 . We may then enquire about other physical situations wherein the fidelity can be periodic. If $`f(t)=f(t+jT)`$ with integer $`j`$, $`f(jT)=1`$ as $`f(0)=1`$. That is to say that, after reversing the perturbed dynamics at $`t=jT`$ the initial state $`|\alpha `$ is completely recovered (up to the phase factor). We refer to this as periodic quantum echo with period $`T`$. Before addressing the possibility and utilization of such echo in many-body complex systems, it is instructive to resolve the dynamics of two-level quantum system (qubit) in this context. It goes without saying that any understanding of qubit is of fundamental importance in the subject of quantum computation niel00 as well. In the present note, we compute quantum fidelity to explore the echo scenario for two-level systems.
It should be noted that $`f(t)`$ depends on the form of perturbation and the choice of initial state. If the initial state is an eigenstate of one of the Hamiltonians (time independent), quantum fidelity is nothing but the survival probability under the evolution governed by the other Hamiltonian. Further, for a two-level system the survival probability is a simple periodic function in time. Hence we naively expect the fidelity to be a non-decaying function, for an arbitrary choice of initial state. In the next section, we show for time independent case that in general the quantum fidelity is quasiperiodic. It will be periodic provided the ratio of eigen-energy spacings of the two Hamiltonians are rational, implying that the two-level systems exhibit periodic quantum echoes. In section III, we dwell on such echo scenario for a specific time dependent problem. For a time periodic perturbation, it is shown that the well known Rabi oscillation of two-level system is directly reflected on the temporal behaviour of quantum fidelity. While this is seen as a simple mechanism to observe periodic quantum echoes in two-level systems, we point out that for certain initial superpositional states the Rabi oscillation vanishes and the fidelity is invariant in time.
## II Time independent case
Let us first consider two time independent Hamiltonians $`H`$ and $`\stackrel{~}{H}`$, whose eigenvalue equations are
$$H|\varphi _n=E_n|\varphi _n;\stackrel{~}{H}|\stackrel{~}{\varphi }_m=\stackrel{~}{E}_m|\stackrel{~}{\varphi }_m,$$
(2)
where $`m,n=1,2`$. Note that the eigen sets $`\{|\varphi _n\}`$ and $`\{|\stackrel{~}{\varphi }_m\}`$ individually form an orthonormal basis. Expanding the initial state $`|\alpha `$ in the eigen basis, one can write
$$|\alpha =u_1|\varphi _1+u_2|\varphi _2=v_1|\stackrel{~}{\varphi }_1+v_2|\stackrel{~}{\varphi }_2,$$
(3)
where $`u_1,u_2`$ and $`v_1,v_2`$ are complex numbers satisfying the normalization conditions, $`|u_1|^2+|u_2|^2=|v_1|^2+|v_2|^2=1`$. Then the fidelity, defined by eq. (1), can be expressed as
$$f(t)=\left|\underset{m,n}{}e^{i(\stackrel{~}{E}_mE_n)t/\mathrm{}}v_m^{}u_n\stackrel{~}{\varphi }_m|\varphi _n\right|^2.$$
(4)
Introducing the โfrequencyโ variables
$$\omega =\frac{E_2E_1}{\mathrm{}},\stackrel{~}{\omega }=\frac{\stackrel{~}{E}_2\stackrel{~}{E}_1}{\mathrm{}}$$
(5)
after some algebra the fidelity can be recasted as
$`f(t)`$ $`=`$ $`|v_1|^2\left[|u_1|^2|\stackrel{~}{\varphi }_1|\varphi _1|^2+|u_2|^2|\stackrel{~}{\varphi }_1|\varphi _2|^2\right]+|v_2|^2\left[|u_1|^2|\stackrel{~}{\varphi }_2|\varphi _1|^2+|u_2|^2|\stackrel{~}{\varphi }_2|\varphi _2|^2\right]`$ (6)
$`+\mathrm{\hspace{0.33em}2}\{(|v_1|^2|v_2|^2)[a_1\mathrm{cos}\omega tb_1\mathrm{sin}\omega t]+(|u_1|^2|u_2|^2)[a_2\mathrm{cos}\stackrel{~}{\omega }tb_2\mathrm{sin}\stackrel{~}{\omega }t]`$
$`+[a_3\mathrm{cos}(\stackrel{~}{\omega }\omega )t+b_3\mathrm{sin}(\stackrel{~}{\omega }\omega )t]+[a_4\mathrm{cos}(\stackrel{~}{\omega }+\omega )t+b_4\mathrm{sin}(\stackrel{~}{\omega }+\omega )t]\},`$
where
$$\begin{array}{ccc}a_1+ib_1\hfill & =\hfill & u_1u_2^{}\stackrel{~}{\varphi }_1|\varphi _1\varphi _2|\stackrel{~}{\varphi }_1,\hfill \\ a_2+ib_2\hfill & =\hfill & v_1v_2^{}\stackrel{~}{\varphi }_2|\varphi _1\varphi _1|\stackrel{~}{\varphi }_1,\hfill \\ a_3+ib_3\hfill & =\hfill & u_1u_2^{}v_1^{}v_2\stackrel{~}{\varphi }_1|\varphi _1\varphi _2|\stackrel{~}{\varphi }_2,\hfill \\ a_4+ib_4\hfill & =\hfill & u_1^{}u_2v_1^{}v_2\stackrel{~}{\varphi }_1|\varphi _2\varphi _1|\stackrel{~}{\varphi }_2,\hfill \end{array}$$
with $`a_i`$โs and $`b_i`$โs being real constants. In this form, the fidelity is a sum of four periodic functions and hence, in general $`f(t)`$ is quasiperiodic. We may then look at the expression (6) for the following choices of initial states.
(i) $`|\alpha =|\varphi _1`$; $`u_1=1,u_2=0`$:
$`f(t)`$ $`=`$ $`|\stackrel{~}{\varphi }_1|\varphi _1|^4+|\stackrel{~}{\varphi }_2|\varphi _1|^4`$ (7)
$`+\mathrm{\hspace{0.33em}2}|\stackrel{~}{\varphi }_1|\varphi _1|^2|\stackrel{~}{\varphi }_2|\varphi _1|^2\mathrm{cos}\stackrel{~}{\omega }t.`$
(ii) $`|\alpha =|\varphi _2`$; $`u_1=0,u_2=1`$:
$`f(t)`$ $`=`$ $`|\stackrel{~}{\varphi }_1|\varphi _2|^4+|\stackrel{~}{\varphi }_2|\varphi _2|^4`$ (8)
$`+\mathrm{\hspace{0.33em}2}|\stackrel{~}{\varphi }_1|\varphi _2|^2|\stackrel{~}{\varphi }_2|\varphi _2|^2\mathrm{cos}\stackrel{~}{\omega }t.`$
Since the $`H`$-evolution introduces just a phase for the above two choices of initial state, fidelity is nothing but the survival probability in $`\stackrel{~}{H}`$-evolution. It is a simple periodic function in time with period $`T=2\pi /|\stackrel{~}{\omega }|`$.
(iii) $`|\alpha =|\stackrel{~}{\varphi }_1`$; $`v_1=1,v_2=0`$:
$`f(t)`$ $`=`$ $`|\varphi _1|\stackrel{~}{\varphi }_1|^4+|\varphi _2|\stackrel{~}{\varphi }_1|^4`$ (9)
$`+\mathrm{\hspace{0.33em}2}|\varphi _1|\stackrel{~}{\varphi }_1|^2|\varphi _2|\stackrel{~}{\varphi }_1|^2\mathrm{cos}\omega t.`$
(iv) $`|\alpha =|\stackrel{~}{\varphi }_2`$; $`v_1=0,v_2=1`$:
$`f(t)`$ $`=`$ $`|\varphi _1|\stackrel{~}{\varphi }_2|^4+|\varphi _2|\stackrel{~}{\varphi }_2|^4`$ (10)
$`+\mathrm{\hspace{0.33em}2}|\varphi _1|\stackrel{~}{\varphi }_2|^2|\varphi _2|\stackrel{~}{\varphi }_2|^2\mathrm{cos}\omega t.`$
For the last two cases, $`\stackrel{~}{H}`$-evolution introduces just a phase. Hence, $`f(t)`$ is the survival probability in $`H`$-evolution and it oscillates with period $`T=2\pi /|\omega |`$.
On the other hand, for an arbitrary choice of initial state the fidelity is the sum of four periodic functions each of them oscillating with frequencies $`\omega ,\stackrel{~}{\omega },(\stackrel{~}{\omega }\omega )`$ and $`(\stackrel{~}{\omega }+\omega )`$. Then $`f(t)`$ can be periodic if and only if all the four frequencies are commensurate with each other. Since the last two frequencies are just the sum and difference of the first two, the required condition for $`f(t)`$ to be periodic is
$$\frac{\stackrel{~}{\omega }}{\omega }=\frac{\stackrel{~}{E}_2\stackrel{~}{E}_1}{E_2E_1}=\frac{p}{q},$$
(11)
where $`p`$ and $`q`$ are co-primes. Representing the Hamiltonian $`H`$ in a specific two-level basis states $`|1`$ and $`|2`$, energy eigenvalues are solutions of the quadratic equation
$$\left|\begin{array}{cc}H_{11}E& H_{12}\\ H_{12}^{}& H_{22}E\end{array}\right|=0,$$
(12)
where $`H_{ij}`$ are matrix elements of $`H`$ in the chosen basis. Denoting the two solutions as $`E_1`$ and $`E_2`$, we have
$$E_2E_1=\sqrt{(H_{11}H_{22})^2+4|H_{12}|^2}.$$
(13)
With similar expression for the Hamiltonian $`\stackrel{~}{H}`$, the condition for periodic fidelity becomes
$`\left({\displaystyle \frac{p}{q}}\right)^2\left[(H_{11}H_{22})^2+4|H_{12}|^2\right]`$
$`=(\stackrel{~}{H}_{11}\stackrel{~}{H}_{22})^2+4|\stackrel{~}{H}_{12}|^2.`$ (14)
Thus we arrive at a condition on matrix elements of the two Hamiltonians such that the quantum fidelity of an arbitrary state is periodic and the period $`T`$ can be calculated as follows. If $`\stackrel{~}{\omega }=\omega `$,
$$T=2\pi \mathrm{}/|E_2E_1|.$$
(15)
For $`\stackrel{~}{\omega }\omega `$, we have a relation between the frequencies deduced from eq. (11) as
$`p(p^2q^2)\omega `$ $`=`$ $`q(p^2q^2)\stackrel{~}{\omega }=pq(p+q)(\stackrel{~}{\omega }\omega )`$ (16)
$`=pq(pq)(\stackrel{~}{\omega }+\omega ),`$
implying that
$$T=2\pi \mathrm{}/|p(p^2q^2)(E_2E_1)|.$$
(17)
With this, we summarize our general treatment on time independent problem. For a given forward evolution of an arbitrary initial state $`|\alpha `$ governed by the Hamiltonian $`H`$, we shall choose another Hamiltonian $`\stackrel{~}{H}`$ such that its matrix elements in the chosen basis satisfies the condition (14). Then by reversing the evolution with $`\stackrel{~}{H}`$ at $`t=jT`$ ($`j`$ is integer), the initial state is completely recovered. We have then the scenario of periodic quantum echo.
### II.1 Perturbative approximation
If we consider $`\stackrel{~}{H}`$ as a perturbed Hamiltonian, that is, $`\stackrel{~}{H}=H+ฯตV`$ with $`|ฯต|1`$, we shall use the first order time-independent expansions for the eigenvalues
$$\stackrel{~}{E}_1E_1+ฯตV_{11};\stackrel{~}{E}_2E_2+ฯตV_{22},$$
(18)
and for the eigenfunctions
$$|\stackrel{~}{\varphi }_1|\varphi _1\frac{ฯตV_{21}}{(E_2E_1)}|\varphi _2;|\stackrel{~}{\varphi }_2|\varphi _2\frac{ฯตV_{12}}{(E_1E_2)}|\varphi _1,$$
(19)
where $`V_{mn}=\varphi _m|V|\varphi _n`$. Note that the above expansions are valid for $`|ฯตV_{12}||E_1E_2|`$. Then the expression (6) can be approximated up to the order in $`ฯต`$ as
$`f(t)`$ $``$ $`|u_1|^4+|u_2|^4+2|u_1|^2|u_2|^2\mathrm{cos}\omega _ฯตt2ฯต{\displaystyle \frac{|u_1|^2|u_2|^2}{E_2E_1}}\{A+[A\mathrm{cos}\omega t+B\mathrm{sin}\omega t]`$ (20)
$`[A\mathrm{cos}\omega _ฯตtB\mathrm{sin}\omega _ฯตt][A\mathrm{cos}(\omega +\omega _ฯต)t+B\mathrm{sin}(\omega +\omega _ฯต)t]\},`$
where
$$\omega _ฯต=\frac{ฯต(V_{22}V_{11})}{\mathrm{}};A+iB=\alpha |\varphi _1\varphi _2|\alpha V_{12}.$$
(21)
Thus the first order approximation to the fidelity is the sum of three periodic functions each of them oscillating with frequencies $`\omega `$, $`\omega _ฯต`$ and $`(\omega +\omega _ฯต)`$. Then the periodicity condition is
$$\frac{\omega _ฯต}{\omega }=\frac{ฯต(V_{22}V_{11})}{E_2E_1}=\frac{r}{s},$$
(22)
with $`r,s`$ being co-primes. If $`\omega _ฯต=\omega `$, the period of oscillation in this approximation
$$T^{}=2\pi \mathrm{}/|E_2E_1|.$$
(23)
For $`\omega _ฯต\omega `$, we have
$$r(r+s)\omega =s(r+s)\omega _ฯต=rs(\omega _ฯต+\omega )$$
(24)
and hence
$$T^{}=2\pi \mathrm{}/|r(r+s)(E_2E_1)|.$$
(25)
That is, if the difference between first order energy corrections is a rational multiple of the unperturbed energy level spacing, the quantum fidelity is approximately periodic with period $`T^{}`$. In other words, by reversing the perturbed dynamics at $`t=jT^{}`$ the initial state $`|\alpha `$ is nearly recovered.
If $`V_{11}=V_{22}`$ ($`\omega _ฯต=0`$), eq. (20) reduces to $`f(t)1`$. A trivial example of this case is that $`V=\text{constant}`$, corresponds to the shifting of levels which is dynamically insignificant. If the initial state is such that $`|u_1|=|u_2|^2`$, then the first order approximation reduces to
$$f(t)\frac{1}{2}\left[1+\mathrm{cos}\omega _ฯตt\right]$$
(26)
with $`T^{}=2\pi \mathrm{}/|ฯต(V_{22}V_{11})|`$.
## III Time dependent case
Having identified the scenario of periodic echo in time independent two-level systems, here we dwell on such a possibility for time dependent problems. Considering the eigenvalue equation of $`H`$, as given in eq. (2), the initial state is rewritten as
$$|\alpha =c_1(0)|\varphi _1+c_2(0)|\varphi _2,$$
(27)
with the normalization, $`|c_1(0)|^2+|c_2(0)|^2=1`$. Taking $`\stackrel{~}{H}=H+ฯตV(t)`$, the required evolutions are
$$\begin{array}{ccc}U(t)|\alpha \hfill & =\hfill & _nc_n(0)e^{iE_nt/\mathrm{}}|\varphi _n,\hfill \\ \stackrel{~}{U}(t)|\alpha \hfill & =\hfill & _mc_m(t)e^{iE_mt/\mathrm{}}|\varphi _m,\hfill \end{array}$$
(28)
with $`c_m(t)`$ satisfying the Schrรถdinger equation
$$\frac{i\mathrm{}}{ฯต}\left[\begin{array}{c}\dot{c}_1(t)\\ \dot{c}_2(t)\end{array}\right]=\left[\begin{array}{cc}V_{11}(t)\hfill & V_{12}(t)e^{i\omega t}\hfill \\ V_{12}^{}(t)e^{i\omega t}\hfill & V_{22}(t)\hfill \end{array}\right]\left[\begin{array}{c}c_1(t)\\ c_2(t)\end{array}\right],$$
(29)
where $`V_{mn}(t)=\varphi _m|V(t)|\varphi _n`$ and $`\omega =(E_2E_1)/\mathrm{}`$. We shall note that the solution of the above system of differential equations is exactly known only for a few specific forms of perturbation (see for example, ref. solved ). Using eq. (28) in (1), the quantum fidelity can be now expressed as
$`f(t)`$ $`=`$ $`|c_1(t)|^2|c_1(0)|^2+|c_2(t)|^2|c_2(0)|^2`$ (30)
$`+\mathrm{\hspace{0.33em}2}\text{Re}\left\{c_1^{}(t)c_2(t)c_1(0)c_2^{}(0)\right\}.`$
That is, if the transition probabilities $`|c_1(t)|^2`$ and $`|c_2(t)|^2`$ are periodic in time the system could possibly exhibit periodic echo. A simplest such situation is an oscillating perturbation of the form $`V_{11}(t)=V_{22}(t)=\mu `$ and $`V_{12}(t)=e^{i\nu t}`$, where $`\nu `$ is the perturbation frequency.
Defining $`\delta =\nu \omega `$, for the above choice of $`V(t)`$ the Schrรถdinger equation (29) may be rewritten as
$$\ddot{c}_1i\left[\delta \frac{2ฯต\mu }{\mathrm{}}\right]\dot{c}_1+\frac{1}{\mathrm{}^2}\left[ฯต^2(1\mu ^2)+ฯต\mu \mathrm{}\delta \right]c_1=0,$$
(31)
and a similar equation for $`c_2(t)`$ with $`\delta `$ being replaced by $`\delta `$. Defining the variables
$$\mathrm{\Omega }_r=\frac{2ฯต}{\mathrm{}};\mathrm{\Omega }=\sqrt{\delta ^2+\mathrm{\Omega }_r^2},$$
(32)
the general solution to eq. (31) is
$`c_1(t)`$ $`=`$ $`\mathrm{exp}\left[it({\displaystyle \frac{\delta }{2}}{\displaystyle \frac{ฯต\mu }{\mathrm{}}})\right]\{c_1(0)\mathrm{cos}\left({\displaystyle \frac{\mathrm{\Omega }t}{2}}\right)`$ (33)
$`{\displaystyle \frac{i}{\mathrm{\Omega }}}[\delta c_1(0)+\mathrm{\Omega }_rc_2(0)]\mathrm{sin}\left({\displaystyle \frac{\mathrm{\Omega }t}{2}}\right)\}.`$
To deduce the above solution, we note that with the choice
$$c_1(t)=e^{i\eta t}x(t),\eta =\frac{\delta }{2}\frac{ฯต\mu }{\mathrm{}},$$
(34)
eq. (31) is reduced to the harmonic oscillator equation, $`\ddot{x}+(\mathrm{\Omega }/2)^2x=0`$. From the general solution of the latter, after using (34) and (29), one obtains the solution (33). We also note that in the above example, resonance condition corresponds to $`\nu =\omega `$, which implies $`\delta =0`$ or $`\mathrm{\Omega }=\mathrm{\Omega }_r`$. Proceeding further with either the substitution of solution (33) in eq. (29) or solving $`c_2(t)`$ in a similar way, we obtain
$`c_2(t)`$ $`=`$ $`\mathrm{exp}[it({\displaystyle \frac{\delta }{2}}+{\displaystyle \frac{ฯต\mu }{\mathrm{}}})]\{c_2(0)\mathrm{cos}\left({\displaystyle \frac{\mathrm{\Omega }t}{2}}\right)`$ (35)
$`+{\displaystyle \frac{i}{\mathrm{\Omega }}}[\delta c_2(0)\mathrm{\Omega }_rc_1(0)]\mathrm{sin}\left({\displaystyle \frac{\mathrm{\Omega }t}{2}}\right)\}.`$
In what follows, it is useful to define the quantities
$$D=|c_2(0)|^2|c_1(0)|^2;a+ib=c_1(0)c_2^{}(0).$$
(36)
Using the above expressions for $`c_1(t)`$ and $`c_2(t)`$, we have the following:
$`|c_1(t)|^2`$ $`=`$ $`{\displaystyle \frac{1}{2\mathrm{\Omega }^2}}\left[(\delta ^2+\mathrm{\Omega }^2)|c_1(0)|^2+\mathrm{\Omega }_r^2|c_2(0)|^2\right]+{\displaystyle \frac{a\delta \mathrm{\Omega }_r}{\mathrm{\Omega }^2}}{\displaystyle \frac{1}{2\mathrm{\Omega }^2}}\left[D\mathrm{\Omega }_r^2+2a\delta \mathrm{\Omega }_r\right]\mathrm{cos}(\mathrm{\Omega }t){\displaystyle \frac{b\mathrm{\Omega }_r}{\mathrm{\Omega }}}\mathrm{sin}(\mathrm{\Omega }t),`$
$`|c_2(t)|^2`$ $`=`$ $`{\displaystyle \frac{1}{2\mathrm{\Omega }^2}}\left[(\delta ^2+\mathrm{\Omega }^2)|c_2(0)|^2+\mathrm{\Omega }_r^2|c_1(0)|^2\right]{\displaystyle \frac{a\delta \mathrm{\Omega }_r}{\mathrm{\Omega }^2}}+{\displaystyle \frac{1}{2\mathrm{\Omega }^2}}\left[D\mathrm{\Omega }_r^2+2a\delta \mathrm{\Omega }_r\right]\mathrm{cos}(\mathrm{\Omega }t)+{\displaystyle \frac{b\mathrm{\Omega }_r}{\mathrm{\Omega }}}\mathrm{sin}(\mathrm{\Omega }t),`$ (37)
and
$`2\text{Re}\left\{c_1^{}(t)c_2(t)c_1(0)c_2^{}(0)\right\}`$ $`=`$ $`{\displaystyle \frac{1}{\mathrm{\Omega }^2}}\{\mathrm{\Omega }_r(2a\mathrm{\Omega }_r\delta D)[a\mathrm{cos}(\delta t)+b\mathrm{sin}(\delta t)]+{\displaystyle \frac{1}{2}}[A_{}\mathrm{sin}(\delta +\mathrm{\Omega })t+B_{}\mathrm{cos}(\delta +\mathrm{\Omega })t]`$ (38)
$`+{\displaystyle \frac{1}{2}}[A_+\mathrm{sin}(\delta \mathrm{\Omega })t+B_+\mathrm{cos}(\delta \mathrm{\Omega })t]\},`$
where
$$A_\pm =bD\mathrm{\Omega }_r(\delta \pm \mathrm{\Omega })2ab\mathrm{\Omega }_r^2;B_\pm =aD\mathrm{\Omega }_r(\delta \pm \mathrm{\Omega })+2\left[a^2\delta ^2+b^2\mathrm{\Omega }^2\pm \delta \mathrm{\Omega }(a^2+b^2)\right].$$
It is now easy to verify the normalization condition, that is $`|c_1(t)|^2+|c_2(t)|^2=1`$, which is due to the unitary quantum evolution. If the initial state is such that $`c_1(0)=1`$ and $`c_2(0)=0`$, then $`a=b=0`$ and $`D=1`$. For this choice of initial state, at resonance ($`\delta =0`$ or $`\mathrm{\Omega }=\mathrm{\Omega }_r`$) we have
$$|c_1(t)|^2=\frac{1}{2}\left[1+\mathrm{cos}(\mathrm{\Omega }_rt)\right].$$
(39)
Therefore, at resonance if the system is initially in one of the unperturbed levels, the probability to remain in the same level oscillates periodically with the frequency $`\mathrm{\Omega }_r`$. This is the well known Rabi oscillation and $`\mathrm{\Omega }_r`$ is the so called Rabi frequency.
From the eqs. (37) and (38) we observe that the fidelity is the sum of four periodic functions with frequencies $`\delta ,\mathrm{\Omega },(\delta \mathrm{\Omega })`$ and $`(\delta +\mathrm{\Omega })`$. At resonance, eq. (37) becomes
$`|c_1(t)|^2`$ $`=`$ $`{\displaystyle \frac{1}{2}}\left[1D\mathrm{cos}(\mathrm{\Omega }_rt)2b\mathrm{sin}(\mathrm{\Omega }_rt)\right],`$
$`|c_2(t)|^2`$ $`=`$ $`{\displaystyle \frac{1}{2}}\left[1+D\mathrm{cos}(\mathrm{\Omega }_rt)+2b\mathrm{sin}(\mathrm{\Omega }_rt)\right],`$ (40)
and (38) is reduced to
$`2\text{Re}\left\{c_1^{}(t)c_2(t)c_1(0)c_2^{}(0)\right\}`$
$`=2a^2+2b^2\mathrm{cos}(\mathrm{\Omega }_rt)bD\mathrm{sin}(\mathrm{\Omega }_rt).`$ (41)
With this, the quantum fidelity takes the simple form
$$f(t)=\frac{1}{2}+2a^2+\left\{\frac{D^2}{2}+2b^2\right\}\mathrm{cos}(\mathrm{\Omega }_rt).$$
(42)
This shows that at resonance the quantum fidelity oscillates sinusoidally between $`1`$ and $`4a^2`$ with Rabi frequency. In other words, periodic behaviour of fidelity with period $`T=2\pi /|\mathrm{\Omega }_r|=\pi \mathrm{}/|ฯต|`$ is indeed a direct consequence of the Rabi oscillation seen in transition probabilities. Note that here $`ฯต`$ is arbitrary and the above calculations are exact. If $`ฯต`$ is small the period $`T`$ is large. For integer $`j`$, it is now easy to check that
$$f(jT)=\frac{1}{2}+\frac{1}{2}\left\{|c_1(0)|^2+|c_2(0)|^2\right\}^2=1.$$
(43)
That is, upon inverting the perturbed evolution at $`t=jT`$, the initial state is completely recovered. We may also consider this result as a simple mechanism to generate periodic quantum echo for a two-level system.
It should be emphasized that periodic perturbation does not necessarily imply that fidelity is periodic for arbitrary initial state. As it is evident from the eq. (42), $`f(t)`$ depends also on the choice of initial state. If the initial state is an eigen state of the unperturbed system i.e., $`|\alpha =|\varphi _1`$ or $`|\varphi _2`$, $`a=b=0`$ and $`D=1`$. In this case, eq. (42) becomes
$$f(t)=\frac{1}{2}\left[1+\mathrm{cos}(\mathrm{\Omega }_rt)\right]$$
(44)
and the fidelity oscillates between 1 and 0. On the other hand, if $`|\alpha =(|\varphi _1\pm |\varphi _2)/\sqrt{2}`$ then $`f(t)=1`$ and the fidelity is invariant in time. In other words, unperturbed and perturbed evolutions are one and the same. For all other choices of initial states, quantum dynamics is between the above two extremes.
## IV Conclusions
We have investigated the possibility of periodic dynamical echo for isolated two-level systems by computing the quantum fidelity. Considering two time independent Hamiltonians, the fidelity is shown to be quasiperiodic in general. It can be made periodic provided the ratio of corresponding eigen-energy spacings are rational. In terms of perturbative approximation, fidelity is nearly periodic if the difference between first order energy corrections is a rational multiple of unperturbed energy level spacing. The periodicity in fidelity implies that the two-level systems can exhibit periodic quantum echo. Further, we have considered a specific time-dependent problem, that is, a two-level system with oscillating time dependent perturbation. If the perturbation frequency is in resonance with atomic frequency, the fidelity is shown to display Rabi oscillation with period of oscillation inversely proportional to the strength of the perturbation. This is identified as a simple mechanism for generating periodic quantum echo.
It is yet to be seen if such periodic echoes of qubit can be exploited for any useful information processing. In this context, it will be interesting to extend similar studies for other exactly solvable problems and also for multilevel systems.
Acknowledgement
The work reported here forms part of the Council of Scientific and Industrial Research and the Department of Science & Technology, Government of India supported research projects.
|
warning/0506/astro-ph0506764.html
|
ar5iv
|
text
|
# CORRECTIONS FOR GRAVITATIONAL LENSING OF SUPERNOVAE: BETTER THAN AVERAGE?
## 1 INTRODUCTION
Observations of high-redshift Type Ia supernovae (SNIa) have in the last decade or so lead to a dramatic paradigm shift in cosmology (Perlmutter et al., 1998; Riess et al., 1998; Schmidt et al., 1998; Perlmutter et al., 1999; Knop et al., 2003; Tonry et al., 2003; Riess et al., 2004). Measurements of the luminosity distance to supernovae over a wide range of redshifts were used to break the degeneracy between cosmic fluids, as suggested by Goobar & Perlmutter (1995). The data clearly favors a universe dominated by repulsive dark energy, and which is presently undergoing accelerated expansion. The next step in observational cosmology is to test the nature of this dark energy, whether constant, i.e. compatible with Einsteinโs cosmological constant, or due to completely new physics. Observations of SNIa are among the leading astrophysical tools to explore this question further, as they probe the expansion history of the Universe directly. Large dedicated surveys are in progress (e.g. CFHTLS, ESSENCE, SDSSII) and even more ambitious space based projects are being planned for the future, e.g. the JDEM proposals, DESTINY, JEDI and SNAP.
One thing in common for all these projects is the very large projected number of SNIa that eventually will populate the Hubble diagram used to derive cosmological parameters. Clearly, systematic uncertainties will (soon) become the limiting factor. While some of these uncertainties are due to our lack of knowledge of the SNIa physics and intrinsic properties, others stem from possible interactions of the supernova light (rest-frame UV and optical) near the source or along the line-of-sight (l-o-s), e.g. extinction by dust in the host galaxy or intergalactic medium. In this work, we focus on the gravitational interaction of photons along the l-o-s, i.e. gravitational lensing. As supernova surveys become deeper, the measured source fluxes become increasingly more sensitive to the inhomogeneities in the matter distribution of the Universe. In Amanullah, Mรถrtsell & Goobar (2003), the JDEM/SNAP mission was simulated using the SNOC Monte-Carlo package (Goobar et al., 2002) and it was found that a careful statistical treatment could be used to optimize the fitting of cosmological parameters from the Hubble diagram of SNIa taking into account the (asymmetric) redshift dependent lensing magnification distribution. Lensing on individual SNe have also been studied, in e.g Lewis & Ibata (2001); Mรถrtsell, Gunnarsson & Goobar (2001); Benรญtez et al. (2002); Gunnarsson (2004) by modeling the effect from the galaxies close to the l-o-s to the SN.
In this work, we investigate the accuracy to which lensing (de)magnification can be estimated on individual supernovae. For that purpose, we create mock galaxy catalogs with properties (e.g., galaxy magnitudes, redshifts and spectral types) based on luminosity functions derived from observations by Dahlรฉn et al. (2005). Using the brightness of galaxies as a tracer of the gravitational fields along the l-o-s, we use the multiple lens-plane package Q-LET (Gunnarsson, 2004) to investigate the accuracy to which the magnification can be estimated as a function of the survey parameters, assumptions on M/L-ratios and halo shapes. In an accompanying paper (Jรถnsson et al., 2005), we apply the technique described here to investigate the lensing magnification probability distribution for 33 supernovae in the GOODS survey (Riess et al., 2004; Strolger et al., 2004).
In this paper, we will assume that dark matter halos in individual galaxy halos and small groups are most important for the lensing magnification of supernovae. For cluster size lenses, additional information is needed to model the gravitational potential, e.g., lensing of background galaxies. Though such a generalization of the method is straightforward, in the following we have assumed the uncertainty in the lensing magnification factor for the small fraction of SNe with foreground clusters to be of the same order as SNe with massive foreground galaxies. Also, we assume that the large scale dark matter structures in the Universe is traced by the luminous matter, i.e. that filaments and walls are populated by galaxies. This approach is reasonable as long as the luminous and dark matter are not anti-correlated, see ยง5.
In ยง2, we discuss whether one needs to correct for lensing at all. Section ยง3 describes the underlying theory and ยง4 treats our method for estimating the accuracy of the lensing corrections. We summarize and discuss our results in ยง5. Throughout the paper we use natural units, where $`c=G=1`$. We assume that the underlying cosmological parameters are $`H_0=70\mathrm{km}\mathrm{s}^1\mathrm{Mpc}^1`$, the matter density $`\mathrm{\Omega }_\mathrm{M}=0.3`$ and the dark energy density $`\mathrm{\Omega }_\mathrm{\Lambda }=0.7`$. When no explicit redshift dependence is shown, quantities refer to present values ($`z=0`$). Quoted magnitudes are in the Vega system.
## 2 TO CORRECT OR NOT TO CORRECT
Since the mean magnification due to gravitational lensing of a large number of sources is expected to be unity relative to a homogeneous universe, the question arises whether one should correct for the effect of gravitational lensing at all.
Because flux, $`f`$, is conserved, it is the mean of the magnification factor, $`\mu `$, that is equal to one, i.e. $`\overline{\mu }=1`$, or defining $`\mu =1+\delta `$ where $`\delta `$ is the fractional difference in luminosity from the unlensed (homogeneous universe) case, $`\overline{\delta }=0`$. The magnitude is given by $`m=2.5\mathrm{log}f+\mathrm{const}`$, and we can write
$$m=m_0\frac{2.5}{\mathrm{ln}10}\mathrm{ln}\mu ,$$
(1)
where $`m_0`$ is the unlensed magnitude. Taylor expanding $`\mathrm{ln}\mu =\mathrm{ln}(1+\delta )`$, we get
$$m=m_0\frac{2.5}{\mathrm{ln}10}\left[\delta \frac{\delta ^2}{2}+๐ช(\delta ^3)\right],$$
(2)
with mean value $`\overline{m}=m_0+0.54\overline{\delta ^2}+๐ช(\overline{\delta ^3})`$. From this it is clear that the average lensed magnitude need not be equal to the unlensed magnitude. Note that in current surveys, this effect is very small compared to, e.g., the intrinsic scatter of SN luminosities which is why the distinction is still unimportant. Note also that the mean magnification factor is unity only for random source positions. For an actual sample of observed SNIa, magnification bias can push the mean magnification to higher values.
However, given that we have a sample of random source position SNIa and neglect the small corrections to $`\overline{m}`$ (or perform our cosmology fit using flux units), then $`\overline{m}`$ is an unbiased estimator for the population mean of the observed magnitudes<sup>1</sup><sup>1</sup>1As long as the variance of $`m`$ is finite.. Under these circumstances, neglecting the scatter due to lensing does not cause any bias in the fitted cosmological parameters and good statistics will help in beating down the error (e.g., Holz & Linder, 2004). There could still be good reasons to consider correcting for lensing effects. If we are able to reduce the scatter in the observed magnitudes and keep $`\overline{m}`$ as an unbiased estimator, then we are able to make more accurate cosmology fits. There are also cases where it is non-trivial to quantify the importance of the magnification bias, e.g., the case of SN 1997ff. In a similar context, the ability to correct individual lines-of-sight for gravitational lensing magnification would have a profound impact in our ability use gravitational wave โsirensโ for measuring cosmological parameters, as their use as standard candles is ultimately limited by the lensing uncertainty (Holz & Hughes, 2005).
## 3 MODELING AN INHOMOGENEOUS UNIVERSE
In this section we present a method to investigate the effects of gravitational lensing in an inhomogeneous universe.
### 3.1 Halo Profiles
Neglecting gravitational lensing is equivalent to assuming that matter is homogeneously distributed in the Universe. However, on small scales, the Universe is certainly inhomogeneous. To investigate the effects of gravitational lensing of distant sources, a realistic model of the matter distribution in the Universe is needed. In the following, we describe how we (re)distribute the matter in our model universe using observations of the luminous matter.
We assume that each galaxy is surrounded by a dark matter halo and that the mass of this halo can be estimated from the galaxy luminosity. However, inferring masses of dark matter halos from luminosities of galaxies is non-trivial. The effects of lensing by a halo depends not only on its mass, but also on its density profile. Both the density profile and mass of dark matter halos are issues under debate. We have chosen to work mainly with two different halo models, Singular Isothermal Spheres (SIS) and the model of Navarro, Frenk and White (NFW; Navarro, Frenk & White, 1997).
The density profile of a SIS, $`\rho _{\mathrm{SIS}}(r)=\sigma ^2/(2\pi r^2)`$, is characterized by its l-o-s velocity dispersion $`\sigma `$, which can be estimated from the galaxy luminosity via the FaberโJackson (FโJ) or TullyโFisher (TโF) relations, approximately valid for elliptical and spiral galaxies respectively. Since the mass of a SIS halo diverges, $`m_{\mathrm{SIS}}(r)=2\sigma ^2r`$, we use a truncation radius $`r_\mathrm{t}`$. A commonly used scale for halo profiles in general is $`r_{200}`$, defined as the radius inside which the mean mass density is 200 times the present critical density. For a SIS halo, $`r_{200}`$ and the corresponding mass within this radius, $`m_{200}`$, are given by
$$r_{200}^{\mathrm{SIS}}=\frac{\sqrt{2}\sigma }{10H_0},m_{200}^{\mathrm{SIS}}=\frac{\sqrt{2}\sigma ^3}{5H_0}.$$
(3)
The density profile of a NFW halo is
$$\rho _{\mathrm{NFW}}(r)=\frac{\rho _\mathrm{s}}{(r/r_\mathrm{s})(1+r/r_\mathrm{s})^2},$$
(4)
where $`r_\mathrm{s}`$ is the scale radius for which approximately $`\rho _{\mathrm{NFW}}r^2`$ and $`\rho _\mathrm{s}`$ is the density at $`r0.5r_\mathrm{s}`$.
The NFW halo is fully determined by $`m_{200}`$ since $`r_{200}=\left[m_{200}/(100H_0^2)\right]^{1/3}`$ and the scale radius $`r_\mathrm{s}`$ and $`\rho _\mathrm{s}`$ can be found numerically from $`m_{200}`$ (Navarro, Frenk & White, 1997).
In the following, we assume that the mass within $`r_{200}`$ is roughly the same for the SIS and NFW halo profiles, i.e. $`m_{200}^{\mathrm{SIS}}=m_{200}^{\mathrm{NFW}}`$. We also set the truncation radius $`r_\mathrm{t}=r_{200}`$ for both SIS and NFW halos. Varying $`r_\mathrm{t}`$ does not alter gravitational lensing effects significantly, see ยง4.5.
### 3.2 The Smoothness Parameter
Very faint and/or small scale structures cannot all be seen in a magnitude limited survey. Also, for the method used in this paper, any matter not directly associated with individual galaxies such as completely dark halos and cluster halos need to be accounted for. In order to assure that the mean mass density in our model universe is kept constant, we keep the โremainingโ mass, not accounted for when relating the dark matter to the luminous matter, as a homogeneous component.
The homogeneous part can be characterized by the *smoothness parameter* $`\eta (z)`$, quantifying the fraction of smoothly distributed matter in our model universe (or our lack of knowledge on the dark matter distribution in the real Universe). Since the fraction of galaxies observed at a given magnitude limit is a function of redshift, and also since the Universe evolves, the smoothness parameter is expected to vary with redshift.
The smoothness parameter in a given survey can be computed from the observed density of matter in clumps, i.e. in our case galaxies surrounded by dark matter halos, $`\rho _\mathrm{g}(z)`$. If the redshift dependence of $`\rho _\mathrm{g}(z)`$ can be factorized into a term $`(1+z)^3`$, scaling like the matter density, and an unknown factor $`f(z)`$ originating from the magnitude limit of the survey and evolution, we can write
$$\rho _\mathrm{g}(z)=\rho _\mathrm{g}(0)(1+z)^3f(z).$$
(5)
Then the smoothness parameter is simply given by
$$\eta (z)=1\frac{\mathrm{\Omega }_\mathrm{G}}{\mathrm{\Omega }_\mathrm{M}}f(z),$$
(6)
where the density in galaxies at $`z=0`$ has been scaled with the present critical density to $`\mathrm{\Omega }_\mathrm{G}`$. Once the galaxies have been associated with halos of definite masses, the comoving density of clumps as a function of redshift $`\mathrm{\Omega }_\mathrm{G}f(z)`$ can be estimated. We divide the distribution of galaxies into redshift bins and estimate $`\mathrm{\Omega }_\mathrm{G}f(z)`$ in each bin. The density of clumps in the $`i`$:th bin, centered on redshift $`z_i`$, is obtained through
$$\mathrm{\Omega }_\mathrm{G}f(z_i)=\frac{1}{\rho _\mathrm{c}}\frac{_jm_j}{V_i},$$
(7)
where $`m_j`$ is the mass of a clump and $`\rho _\mathrm{c}`$ is the critical density. The comoving volume of the $`i`$:th bin is given by
$$V_i=_{z_i\mathrm{\Delta }z/2}^{z_i+\mathrm{\Delta }z/2}\frac{D_\mathrm{A}^2(1+z)^2}{\left[\mathrm{\Omega }_\mathrm{M}(1+z)^3+\mathrm{\Omega }_\mathrm{\Lambda }\right]^{1/2}}\mathrm{\Delta }\mathrm{\Omega }๐z,$$
(8)
where $`\mathrm{\Delta }z`$ is the width of the bin, $`\mathrm{\Delta }\mathrm{\Omega }`$ is the solid angle under study and $`D_\mathrm{A}`$ is the angular diameter distance. Distances have been calculated using the angsiz routines described in Kayser et al. (1997), in which a smoothness parameter varying with redshift can be included. Note that the angular diameter distance $`D_\mathrm{A}`$ used to determine the volume element in Eq. (8) above is calculated using the filled-beam approximation ($`\eta =1`$), since the volume is governed by the global expansion rate, which in turn is governed by the properties on very large scales where the Universe is homogeneous.
### 3.3 Deriving Velocity Dispersions from Observed Luminosities
Galaxy halo masses can be estimated from the velocity dispersion of galaxies. We calculate the velocity dispersion of each galaxy using absolute magnitudes ($`M_B`$) combined with empirical FโJ and TโF relations. For ellipticals, we use the FโJ relation
$$\mathrm{log}_{10}\sigma =\mathrm{log}_{10}\sigma _{}\frac{0.4}{\gamma }(M_BM_B^{})$$
(9)
where $`M_B^{}`$ is the characteristic magnitude and $`\sigma _{}`$ is the normalization in velocity dispersion. We use $`\gamma =4.4`$, as derived by Mitchell et al. (2005) using Sloan Digital Sky Survey (SDSS) data. To derive $`\sigma _{}`$, we use equation (33) in Mitchell et al. (2005), where we use the relation $`M_r=M_B1.32`$ to convert SDSS $`r`$โband magnitudes in AB system to standard $`B`$โband Vega normalized magnitudes. We have here assumed a typical color $`M_BM_r=1.20`$ for ellipticals in the AB-system, and an AB to Vega relation B<sub>AB</sub>=B$`{}_{\mathrm{Vega}}{}^{}0.12`$. The normalization in velocity dispersion is given by
$$\mathrm{log}_{10}\sigma _{}=2.20.091(M_B^{}+19.47+0.85z),$$
(10)
where we use $`M_B^{}=21.04`$ derived for the earlyโtype population by Dahlรฉn et al. (2005). Equation (10) yields $`\sigma _{}=220`$ $`\mathrm{km}\mathrm{s}^1`$ at $`z=0`$. Combining equation (9) and (10) gives an expression for the velocity dispersion
$$\mathrm{log}_{10}\sigma =0.091(M_B4.74+0.85z^{}),$$
(11)
where we use $`z^{}=z`$ for redshifts $`z<1`$ and $`z^{}=1`$ for $`z>1`$. The redshift dependence of the relation accounts for the brightening of the stellar population with redshift. Since this evolution is poorly known at $`z>1`$, we assume a flat evolution at these redshifts. As a measurement of the error in the derived relation, we use the observed scatter in the SDSS measurements by Sheth et al. (2003)
$$\mathrm{rms}(\mathrm{log}_{10}\sigma )=0.079[1+0.17(M_B+19.705+0.85z^{})],$$
(12)
where we again have transformed SDSS $`r`$โband to standard $`B`$โband magnitudes.
For the spiral and later type population, we use the TโF relation derived by Pierce & Tully (1992), with correction for redshift calculated by Bรถhm et al. (2004)
$$\mathrm{log}_{10}V_{\mathrm{max}}=0.134(M_B\mathrm{\Delta }M_B+3.52),$$
(13)
where $`V_{\mathrm{max}}`$ is the maximum rotation velocity for the galaxy. The correction due to redshift dependence is
$$\mathrm{\Delta }M_B=1.22z^{}0.09.$$
(14)
The observed scatter in the relation derived by Pierce & Tully (1992) is $`\mathrm{rms}(M_B)=0.41`$, corresponding to
$$\mathrm{rms}(\mathrm{log}_{10}V_{\mathrm{max}})=0.06.$$
(15)
At $`M_B^{}`$, this is similar to the errors in the FโJ relation above. Finally, the velocity dispersion in spiral galaxies is related to the circular velocity via $`\sigma =V_{\mathrm{max}}/\sqrt{2}`$.
### 3.4 Gravitational Lensing with Multiple Lenses
A typical source l-o-s within some angular radius $`\theta _\mathrm{s}`$ will contain more than one lens. This requires the multiple lens-plane algorithm (see Schneider, Ehlers & Falco, 1992; Gunnarsson, 2004, for further details), which takes into account each lens along the l-o-s by projecting the lensโ mass distribution onto a plane and then traces the light-ray from the image plane (first lens-plane) back through all lens-planes up to the source plane where the magnification and intrinsic position can be found.
In the following we denote angular diameter distances between redshifts $`z_i`$ and $`z_j`$ by $`D_{ij}`$. We use $`o`$ for observer, $`s`$ for source and $`d`$ for lens (deflector). When $`z_i=0`$, that index is omitted.
Each halo is truncated in 3D at $`r=r_\mathrm{t}`$, then, upon projection onto a plane, the corresponding surface mass density will smoothly go to zero at the projected truncation radius. The projection can be done analytically for our lens models. For simplicity, we start by considering a single lens-plane. The equations can be simplified if we let $`\xi `$ be the impact parameter on a halo and define $`x=\xi /\xi _0`$ and $`x_\mathrm{t}=r_\mathrm{t}/\xi _0`$, where
$$\xi _0=\frac{4\pi \sigma ^2D_\mathrm{d}D_{\mathrm{ds}}}{D_\mathrm{s}}$$
(16)
for the SIS and
$$\xi _0=r_\mathrm{s}$$
(17)
for the NFW halo. Then, the projected density $`\kappa (x)`$ can be written as
$$\kappa _{\mathrm{SIS}}(x)=\frac{1}{\pi x}\mathrm{arctan}\left(\frac{\sqrt{x_\mathrm{t}^2x^2}}{x}\right)$$
(18)
and
$$\kappa _{\mathrm{NFW}}(x)=\frac{2\kappa _\mathrm{s}}{x^21}f(x),$$
(19)
where
$$f(x)=\{\begin{array}{ccc}\frac{\sqrt{x_\mathrm{t}^2x^2}}{1+x_\mathrm{t}}+\frac{1}{\sqrt{1x^2}}\left(\mathrm{arctanh}\left(\sqrt{\frac{x_\mathrm{t}^2x^2}{1x^2}}\right)\mathrm{arctanh}\left(\frac{\sqrt{\frac{x_\mathrm{t}^2x^2}{1x^2}}}{x_\mathrm{t}}\right)\right)& & x<1<x_\mathrm{t}\\ \frac{\sqrt{x_\mathrm{t}^2x^2}}{1+x_\mathrm{t}}+\frac{1}{\sqrt{x^21}}\left(\mathrm{arctan}\left(\frac{\sqrt{\frac{x_\mathrm{t}^2x^2}{x^21}}}{x_\mathrm{t}}\right)\mathrm{arctan}\left(\sqrt{\frac{x_\mathrm{t}^2x^2}{x^21}}\right)\right)& & 1<xx_\mathrm{t}\end{array}$$
(20)
and
$$\kappa _{\mathrm{NFW}}(1)=\frac{2}{3}\kappa _\mathrm{s}\left(\frac{x_\mathrm{t}^33x_\mathrm{t}+2}{\left(x_\mathrm{t}^21\right)^{\frac{3}{2}}}\right).$$
(21)
Here, $`\kappa _\mathrm{s}=\rho _\mathrm{s}r_\mathrm{s}/\mathrm{\Sigma }_{\mathrm{cr}}`$, where $`\mathrm{\Sigma }_{\mathrm{cr}}=D_\mathrm{s}/4\pi D_\mathrm{d}D_{\mathrm{ds}}`$ is a critical density related to strong lensing. Note that $`x_\mathrm{t}>1`$ must be assumed for the NFW and that $`\kappa =0`$ for $`x>x_\mathrm{t}`$ for both halo types.
The general expression for the deflection angle for circularly symmetric lenses is
$$\widehat{\alpha }(x)=s\alpha (x)=s\frac{2}{x}_0^xx^{}\kappa (x^{})๐x^{},$$
(22)
where $`s=\xi _0D_\mathrm{s}/D_\mathrm{d}D_{\mathrm{ds}}`$, resulting in
$$\alpha (x)=\frac{2}{\pi }\left(\mathrm{arctan}\left(\frac{\sqrt{x_\mathrm{t}^2x^2}}{x}\right)+\frac{x_\mathrm{t}\sqrt{x_\mathrm{t}^2x^2}}{x}\right)$$
(23)
for the SIS model. For the NFW halo, numerical evaluation is needed.
As the magnification factor, $`\mu ^{}`$, for all halo models is obtained with distances calculated with the $`z`$-dependent $`\eta `$ function, this is the universe relative to which $`\mu ^{}`$ is found (implying $`\mu ^{}1`$ for primary images). In the following, we will quote magnifications, $`\mu `$, relative to a universe with homogeneously distributed matter, the filled-beam value (fb) where $`\overline{\mu }=1`$. The magnifications are related by
$$\mu =\mu ^{}\left(\frac{D_\mathrm{s}^{\mathrm{fb}}}{D_\mathrm{s}^{\eta (\mathrm{z})}}\right)^2.$$
(24)
## 4 SIMULATED SURVEYS
In order to study gravitational lensing corrections, we perform Monte Carlo simulations where we calculate the magnification factor for random source positions in mock galaxy catalogs. By varying the assumptions of the galaxy mass distributions as well as the magnitude limit of the observations, we can estimate the accuracy to which it is possible to correct for the lensing magnification.
For all lensing calculations we have used the publicly available fortran 77 code Q-LET<sup>2</sup><sup>2</sup>2Available at http://www.physto.se/~cg/qlet/qlet.htm (Gunnarsson, 2004), although substantially modified. The code fully utilizes the multiple lens-plane algorithm and has been used previously by Gunnarsson (2004) and Riess et al. (2004) to study lensing effects on supernovae.
As our simulation base, we create for each Monte Carlo realization a mock galaxy catalog designed to reflect the distribution of galaxies expected in a circular cone around a random l-o-s. To characterize the galaxy population we use the $`B`$-band rest-frame Schechter luminosity function (LF) derived by Dahlรฉn et al. (2005) using GOODS CDF-S observations. The LF is used to generate the number of expected galaxies within the cone where we take into account Poissonian fluctuations but do not include effects of galaxy correlations or cosmic variance. The same LF is used to assign absolute magnitudes to each object within the range $`23<M_B<16`$. To account for evolutionary effects, we include a brightening of the $`B`$-band characteristic magnitude by $``$1 mag to redshift $`z=1`$ as discussed in ยง3.3.
A random spectral type is assigned according to the type-specific LF of early-types, late-types and starburst galaxies at $`z`$0.4 in Dahlรฉn et al. (2005). We thereafter assign early-type galaxies an elliptical morphology and late-type galaxies a spiral morphology. We assume that the fraction of galaxies with elliptical morphology is constant over the redshift range investigated.
The redshift of each object is assigned with a probability proportional to the volume element, $`dV(z)/dz`$, which is equivalent to assuming a constant comoving number density of galaxies with redshift, i.e. we do not include any evolution of the number densities due to e.g., mergers or large scale structures. The galaxy is finally given a random position within the l-o-s cone.
Besides redshift, absolute magnitude, spectral type and position, we also calculate the apparent magnitude in the observed $`I`$-band for each object. This allows us to draw subsamples from the catalog with imposed magnitude cutoffs as is the case for real observations. Furthermore, to resemble an observational situation where redshifts are determined photometrically, we also have the option to add a random error to the redshifts. These errors are calculated using simulations and depend on redshift, spectral type and detection S/N. For bright objects with high S/N, errors are typically $`\mathrm{\Delta }_z|z_{\mathrm{phot}}z_{\mathrm{true}}|/(1+z_{\mathrm{spec}})0.05`$, while errors for faint objects (mostly at high $`z`$) can be as large as $`\mathrm{\Delta }z0.3`$. The errors for early-type galaxies are typically a factor two smaller compared to late-type galaxies when comparing at the same apparent magnitudes. Besides this โGaussianโ contribution to the error distribution, a fraction of the galaxies may also be assigned โcatastrophic redshiftsโ with large errors. We discuss this further in ยง4.4.
Figure 1 shows the simulated accuracy of the photometric redshifts for a survey with limiting magnitude $`I<27`$ (S/N=10). The bottom panel shows the generated (input) redshifts vs. the photometric redshifts, while the top panel shows the difference between generated and photometric redshifts as a function of galaxy magnitude.
### 4.1 Understanding the Magnification Uncertainties
We have identified and addressed the following uncertainties when estimating the lensing magnification of a specific source given a galaxy catalog:
* Finite field size
* The intrinsic scatter in, and accuracy of, the FโJ and TโF relations
* Redshift and position uncertainties
* Choice of halo profile
* The magnitude limit
These sources of error are addressed individually in the following sections. Our reference model consists of NFW halos truncated at $`r_\mathrm{t}=r_{200}`$, velocity dispersion/circular velocity normalizations of 220 km s<sup>-1</sup> for ellipticals, 203 km s<sup>-1</sup> for spirals and source redshift $`z=1.5`$. In Figure 2, the Probability Distribution Function (PDF) for lensing magnifications for the reference model is shown. The lensing dispersion at $`z=1.5`$ is $`7\%`$. We analyze the results by comparing the distribution of magnifications in the reference model with the distribution obtained after performing a correction with the above-mentioned uncertainties. We denote the uncorrected value $`\mu _{\mathrm{ref}}`$ and the corrected one $`q_\mu `$, where
$$q_\mu =\frac{\mu _{\mathrm{ref}}}{\mu _{\mathrm{est}}},$$
(25)
where $`\mu _{\mathrm{est}}`$ is the estimated magnification factor including one or more uncertainties. Corrections will reduce the uncertainties from lensing whenever the width of the distribution of $`q_\mu `$ is smaller than the corresponding width in the $`\mu _{\mathrm{ref}}`$ distribution. If no uncertainties were present $`\mu _{\mathrm{est}}=\mu _{\mathrm{ref}}`$ and $`q_\mu =1`$ implying a perfect correction. As a measure of the width we give the standard deviation of the distribution. Since many of the distributions are non-Gaussian, we also report the 68 % and 95 % confidence levels.
### 4.2 Finite Field Size
The mean magnification relative to an homogeneous universe of a large number of sources lensed by randomly distributed matter is expected to be unity due to photon number conservation. When we model a lensing system, only galaxies within angular radius $`\theta _\mathrm{s}`$ of the position of the source on the sky are taken into account and thus $`\overline{\mu }<1`$. If $`\theta _\mathrm{s}`$ is increased, more lenses are added and the mean magnification increases. This dependence is illustrated in Figure 3, showing the mean magnification of 5000 point sources, as a function of $`\theta _\mathrm{s}`$. The mean magnification increases rapidly for small $`\theta _\mathrm{s}`$, but only slowly for $`\theta _\mathrm{s}`$ larger than an arc-minute, where the error is $`1\%`$. In our simulations, we use $`60^{\prime \prime }`$ as a cutoff to save computing time. In a real survey, a cutoff will have to be introduced for practical purposes since only a limited portion of the sky will be observed. In order to avoid a systematic bias due to the finite field size for a given survey, the computed magnifications should be corrected with a factor corresponding to the inverse of the mean magnification for the cutoff radius used. In the following, we have neglected this small ($`1\%`$) correction. Furthermore, going to larger $`\theta _\mathrm{s}`$ would not render $`\overline{\mu }`$ being exactly unity since some flux is lost whenever multiple imaging occurs. Q-LET gives the magnification and intrinsic source position of a given *image*, not observed position and magnification of a given source. Therefore, in the rare cases of multiple imaging, only one of the images will be taken into account resulting in some flux loss. Note also that since random l-o-s and random source positions are different (e.g., Schneider, Ehlers & Falco, 1992), we have to use a magnification dependent weighting procedure to see whether each simulated event should be kept or discarded in order to get a sample of random source positions (see e.g., Goobar et al., 2002).
### 4.3 The FโJ and TโF Relations
The FโJ and TโF relations give the velocity dispersions of the *luminous* matter and we make the assumption that the dark matter that constitutes the halo follows scales in the same way. Both the Faber-Jackson and the Tully-Fisher relations have an intrinsic scatter with an rms estimated in ยง3.3. To study the effect of this scatter, we add random offsets to the FโJ and TโF relations when calculating the halo mass. Since we do not want to bias the total mass in our simulations, we distribute the offsets using a Gaussian distribution in $`\sigma ^3`$ (since mass $`\sigma ^3`$). We derive the width of the Gaussian (one sigma value in $`\sigma ^3`$) from the rms in log$`{}_{10}{}^{}(\sigma )`$ and log$`{}_{10}{}^{}(V_{\mathrm{max}})`$ using Eqs. (11)-(12) (FโJ) and Eqs. (13)-(15) (TโF).
In panel a) in Figure 4, we compare the distribution of the corrected value $`q_\mu `$ due to the scatter in the FโJ and TโF relations with the distribution of magnifications in the reference model (dashed line). Note that if we knew all velocity dispersions and circular velocities exactly, $`q_\mu `$ would be represented by a $`\delta `$-function at $`q_\mu =1`$. We see that the intrinsic scatter in the velocity dispersion and circular velocity causes a dispersion of $`q_\mu `$ of approximately $`3\%`$, a factor of $`2.6`$ less than the original dispersion. Panel a) in Figure 5 shows $`\mu _{\mathrm{ref}}1`$ vs $`q_\mu 1`$ for each individual source. For $`82\%`$ of the sources, the corrected luminosity will be better than the uncorrected one.
Besides the intrinsic scatter in the FโJ and TโF relations, there is also a possible *systematic* uncertainty in $`\sigma _{}`$. To estimate this, we use SDSS data in Bernardi et al. (2003) where photometric and spectroscopic parameters for a sample of $`9000`$ early-type galaxies in the redshift range $`0.01<z<0.3`$ are given, including K-corrections and accurately measured velocity dispersions. We fit a straight line $`M_B=b\mathrm{log}\sigma +a`$ and estimate the error in $`\sigma _{}`$ for a given $`M_B^{}`$ by propagating the errors in the parameters $`a`$ and $`b`$. For $`M_B^{}=21.04`$, we obtain $`\sigma _{}=218\pm 7`$ km s<sup>-1</sup>. In panel b) and c) in Figure 4, we investigate the effect of a systematic shift of $`\pm 10`$ km s<sup>-1</sup> in $`\sigma _{}`$. The dispersion in $`q_\mu `$ due to such a shift is quite small or at the order of $`1\%`$. Panels b) and c) in Figure 5 shows $`\mu _{\mathrm{ref}}1`$ vs $`q_\mu 1`$ for each individual source. The corrected value will be better than the uncorrected for $`>95\%`$ of the sources.
A further source that may increase the scatter in the magnification is the possible misclassification of galaxy morphology. An elliptical galaxy wrongly classified as a spiral, leads to an underestimation of the underlying mass, and vice versa if a spiral is misclassified as an elliptical. To investigate the possible effect of this, we first use a set of simulated galaxies with known morphological types (i.e. an exponential radial profile for spirals and a de Vaucouleurs profile for ellipticals) and measure how many are correctly recovered in an observational setup resembling the GOODS. To classify galaxies, we use the GALFIT software (Peng et al., 2002), which measures the slope of the radial profile and therefore allows a discrimination between ellipticals and spirals. At a S/N=10 detection limit (m$``$25), we find that $``$25 % of the ellipticals and $``$8 % of the spirals are misclassified. The fraction of misclassified galaxies quickly drops to $``$1 % at a magnitude $``$2 mag brighter than the detection limit. We then use these results to estimate the effect of misclassification on the derived magnifications. We find that the increase in the scatter in the magnification is $``$0.5 % due to this effect. Therefore, misidentification of galaxy morphology should only affect the results marginally. However, for ground-based surveys with low resolution, the effect may be larger.
### 4.4 Redshift and Position Uncertainties
Uncertainties in the redshifts of the lenses will alter the results both through the uncertain distances between different lens-planes and by introducing an uncertainty in their absolute magnitudes used in the FโJ and TโF relations.
In an ideal observational situation, all redshifts are determined spectroscopically. In many real situations, however, only photometric redshift are available due to, e.g., the faintness of the galaxies and the large number of sources. To investigate the effects of photometric redshift uncertainties, we add a random offset to the redshift of each object. The size of the offset depends on redshift, apparent magnitude and spectral type of the object and is drawn from the simulated error distribution discussed above.
In panel d) in Figure 4, we show the distribution of the corrected value $`q_\mu `$ due to a random offset to the redshift of each lensing object. The induced error in the estimated magnification is less than $`1\%`$. The corresponding panel in Figure 5 shows $`\mu _{\mathrm{ref}}1`$ vs $`q_\mu 1`$ for each individual source. In this case, $`96\%`$ of the sources have corrected values which are better than when uncorrected.
Besides the Gaussian-like distribution of the photometric redshift errors investigated above, there is also a possibility that a fraction of the objects get โcatastrophic redshiftsโ with large errors, so called outliers. E.g., by comparing with $``$1400 spectroscopic redshifts in the CDF-S and HDF-N, we find that the GOODS photometric redshifts have about 3 % outliers with $`\mathrm{\Delta }_z>0.3`$. The redshift probability distribution, for a majority of these objects are characterized by a primary ($``$Gaussian) peak combined with a less pronounced secondary peak. Outliers are foremost objects assigned the redshift of the primary peak, but where the true redshift is that of the secondary peak. To estimate the effect of outliers, we use the galaxies in the GOODS and simulate the case where we distribute the photometric redshifts over the full probability distribution, including the secondary peak. This will allow $``$3 % outliers. We compare this with the case where we only include redshifts in the primary peak (i.e. objects with $`\mathrm{\Delta }_z<0.3`$). We find that the increase in lensing dispersion due to the population of outliers is less than 1 %. One reason for the small effect is that outliers are mainly faint and very blue objects, which therefore should have relatively small masses.
Any error in the exact positions of the lensing galaxies can also affect the resulting magnification. Apart from the observational error, such an effect can be due to a misalignment between the luminous and the dark matter in a given galaxy. We have investigated the effect of a Gaussian random shift with $`\sigma _{\mathrm{pos}}=0.5`$ arcseconds of all lensing galaxies along the l-o-s. Even such a large shift of all galaxy positions results in a distribution of corrected values $`q_\mu `$ with a dispersion of less than $`0.5\%`$.
### 4.5 Choice of Halo Profile
The choice of halo model is only important in those lens-planes where the light-ray passes through a halo. If passing outside, the halo will act as a point mass and when $`m_{\mathrm{tot}}^{\mathrm{halo}}=m_{200}^{\mathrm{SIS}}=m_{200}^{\mathrm{NFW}}`$ the two different halo models give exactly the same results.
We have performed simulations where all halos were of SIS instead of NFW type. The effect of different halo profiles are also present in the realistic and pessimistic case simulations below.
Panel e) in Figure 4 shows the PDF of $`q_\mu `$ when assuming SIS halos instead of NFW as in the reference model. The dispersion is less than $`1.5\%`$. In $`>90\%`$ of the cases, the corrected value is better than the uncorrected, see panel e) in Figure 5.
We have also investigated how important the assumption on the truncation radius is for the resulting magnification distribution by running tests with $`0.75\times r_{200}r_\mathrm{t}1.25\times r_{200}`$ for the SIS model. The uncertainty in the resulting magnification gives a distribution of corrected values $`q_\mu `$ with a dispersion of $`0.51\%`$. Since the NFW profile falls of as $`\rho _{\mathrm{NFW}}r^3`$ at large radii as compared to $`\rho _{\mathrm{SIS}}r^2`$ for SIS halos, this should be considered a very conservative limit on the effect of changing the truncation radius.
### 4.6 Magnitude Limits
Our reference model uses a constant comoving mass density of galaxies, implying a constant smoothness parameter, $`\eta `$. In a real scenario with an observational magnitude limit, an increasing fraction of galaxies drop out at higher redshift. The faint high redshift galaxies will not be seen and hence not included as lenses in the magnification calculation but instead added as homogeneously distributed matter. Therefore, when deriving the smoothness parameter from observations, $`\eta (z)`$ increases with redshift even if the โunderlyingโ smoothness parameter is constant. For each simulation with a finite magnitude limit, a new $`\eta `$-function is computed using the method described in ยง3.2.
For $`I=27`$ mag the distribution of $`q_\mu `$ is very narrow and for $`I=29`$ mag, it is in principle a $`\delta `$-function. Panel f) in Figure 4 shows the PDF of $`q_\mu `$ for $`I=23`$. The dispersion is $`2\%`$. In $`86\%`$ of the cases, the corrected value is better than the uncorrected, see panel f) in Figure 5.
In Figure 6, we compare $`q_\mu 1`$ as a function of source redshift for models with magnitude limits $`I=23`$ and $`I=25`$ with $`\mu _{\mathrm{ref}}1`$. Even for source redshifts as high as $`z2`$, a magnitude limit of $`I=23`$ does not significantly impair our results.
Photometric errors in the apparent magnitudes translates to an increased scatter in the absolute magnitudes and therefore also in the derived velocity dispersions and masses. At the faintest magnitude limits considered here, S/N=10, typical errors are $`0.1`$ mag. For ellipticals, this corresponds to an increased dispersion of $`\mathrm{\Delta }\mathrm{log}_{10}\sigma 0.01`$ (using Eq. 10). This is significantly less than the intrinsic scatter in the FโJ relation which is rms(log$`{}_{10}{}^{})\sigma 0.08`$ (at $`M_B^{}`$, Eq. 12). For the TโF relation the increased scatter due to photometric errors is $`\mathrm{\Delta }\mathrm{log}_{10}V_{\mathrm{max}}0.013`$ (using Eq. 13), again significantly less than the intrinsic scatter rms(log$`{}_{10}{}^{}V_{\mathrm{max}}^{})0.06`$. So even for the faintest galaxies considered, the errors in apparent magnitude should not affect results more than marginally.
### 4.7 Realistic and Pessimistic Scenarios
We have studied the uncertainty in the lensing correction in a realistic scenario where a reasonable error budget is assumed. In this case we have assumed 50 % NFW, 50 % SIS, no central value shift of the velocity dispersion normalization but a dispersion around this value and a magnitude limit of $`I=25`$. Lens redshifts were assumed to be distributed around their reference values. As the correct model we have used the reference NFW model as above.
A pessimistic scenario for space based surveys has also been studied where we have maximized the uncertainties in relating the luminous and the dark matter. However, one could easily imagine worse cases in a ground-based experiment. For our scenario the erroneous assumptions were: SIS halos, central value of velocity dispersion normalization shifted +10 km s<sup>-1</sup> and distributed around this value, a magnitude limit of $`I=25`$, lens redshifts assumed to be distributed around their reference model values, an offset in lens positions as described in ยง4.4 and finally a truncation radius of $`1.25\times r_{200}`$. We consider this being a pessimistic but not completely unrealistic scenario.
The left panels in Figure 7 show results for the realistic scenario, the right panels for the pessimistic scenario. For the realistic case, $`q_\mu `$ has a dispersion of $`3\%`$ and $`(q_\mu 1)<(\mu _{\mathrm{ref}}1)`$ for $`80\%`$ of the sources. In the pessimistic case, the corresponding numbers are $`3\%`$ and $`77\%`$, i.e. our ability to correct for lensing is more or less unimpaired when going from a realistic to a pessimistic scenario. The bottom row shows $`q_\mu 1`$ as a function of source redshift for the two scenarios. The confidence levels of the realistic case vs $`z`$ can be well fitted with straight lines and these are found in Table 1 expressed in magnitudes.
## 5 SUMMARY AND DISCUSSION
We have investigated the accuracy to which lensing magnification can be estimated on individual lines of sight using the observed properties of the foreground galaxies of each source.
The result depends on the uncertainties in translating observed galaxy luminosities to the (invisible) matter distribution in the lensing galaxies. We have shown that none of the studied uncertainties neither individually nor combined will render the corrected distribution of magnifications wider than the dispersion from lensing. Even for a pessimistic scenario, the dispersion due to lensing for a standard candle source at $`z=1.5`$ can be reduced by a factor $`2`$, comparable to the result for a realistic scenario. The reason our pessimistic case result is not significantly worse than for the realistic case, is that the uncertainties are dominated by the scatter in the FโJ and TโF relations for both scenarios<sup>3</sup><sup>3</sup>3We were able to increase the width in the $`q_\mu `$ distribution for the pessimistic case scenario with $`50\%`$ by modelling 20 % of the galaxies as point masses. However, we consider such a compact mass distribution at galaxy scales too contrived to be included in the simulations.. At lower redshifts ($`z0.5`$), the effects from lensing are small and correcting for lensing is not likely to improve the results.
Even though the fraction of SNe lines-of-sight passing through galaxy cluster lenses is expected to be relatively low, these are potentially important due to magnification bias and in surveys specifically aimed at using cluster potentials as gravitational telescopes (Gunnarsson & Goobar, 2003). In those cases, individual modeling of the cluster potentials using, e.g., weak and strong lensing of background galaxies is needed to correct the observed magnitudes for the lensing magnification. Alternatively, one can choose to discard SNe background to galaxy clusters with very uncertain matter distributions when estimating cosmological parameters.
Our method also takes into account gravitational lensing from large scale dark matter structures such as filaments and walls as long as the matter density is dominated by the individual galaxy size halos. If we very conservatively assume that large scale structures are completely uncorrelated with the luminous matter, we expect the lensing magnification contribution to be less than 2 % on scales larger than 5 arcminutes (for a source redshift of unity) (Cooray et al., 2005).
For a given galaxy catalog, the computed magnifications do not depend strongly on the cosmological parameters used. However, since the formation of matter structure is a function of cosmology, it should in principle be possible to determine the cosmology from the observed distribution of standard candle luminosities. In this case, the use of magnifications of e.g. SNIa would be similar to using shear of background galaxies as in weak lensing studies. Such a study would require large statistics of very well observed SNIa and will probably have to await future dedicated missions such as the proposed SNAP satellite <sup>4</sup><sup>4</sup>4http://snap.lbl.gov.
For current high-$`z`$ SNIa observations the concern is to correct for the magnification and investigate the possibility for magnification bias. Such a study for SNe in the GOODS fields is described in a accompanying paper (Jรถnsson et al., 2005) where magnification bias is shown to be negligible but lensing for individual SNe can be estimated quite robustly from foreground galaxy observations and thus be corrected for. In fact, as long as the luminous and dark matter are not anti-correlated, we would expect to be able to reduce the scatter in the Hubble diagram by assuming that dark matter follows light. Thus, we find that even though the exact relation between luminous and dark matter is uncertain, correcting for gravitational lensing using observed galaxy properties should be harmless at the worst and very useful at the best.
The authors would like to thank Mariangela Bernardi for help with the velocity dispersion data from the Sloan survey, Joakim Edsjรถ, Daniel Holz and Saul Perlmutter for helpful discussions during the course of the work. We are grateful to Swara Ravindranath for providing simulated galaxy catalogs used to derive the recovery fraction of galaxy morphologies. CG would like to thank the Swedish Research Council for financial support. AG is a Royal Swedish Academy Research Fellow supported by grants from the Swedish Research Council, the Knut and Alice Wallenberg Foundation and the Gรถran Gustafsson Foundation for Research in Natural Sciences and Medicine.
|
warning/0506/cond-mat0506071.html
|
ar5iv
|
text
|
# Anomalous quasiparticle transport in the superconducting state of CeCoIn5
## Abstract
We report on a study of thermal Hall conductivity $`\kappa _{xy}`$ in the superconducting state of CeCoIn<sub>5</sub>. The scaling relation and the density of states of the delocalized quasiparticles, both obtained from $`\kappa _{xy}`$, are consistent with $`d`$-wave superconducting symmetry. The onset of superconductivity is accompanied by a steep increase in the thermal Hall angle, pointing to a striking enhancement in the quasiparticle mean free path. This enhancement is drastically suppressed in a very weak magnetic field. These results highlight that CeCoIn<sub>5</sub> is unique among superconductors. A small Fermi energy, a large superconducting gap, a short coherence length, and a long mean free path all indicate that CeCoIn<sub>5</sub> is clearly in the superclean regime ($`\epsilon _F/\mathrm{\Delta }\mathrm{}/\xi `$), in which peculiar vortex state is expected.
Five years after the discovery of superconductivity in CeCoIn<sub>5</sub> Petrovic , this compound has become the focus of considerable attention. Indeed, CeCoIn<sub>5</sub> occupies a particular place among unconventional superconductors; it shares more features with high-$`T_c`$ cuprates than any other heavy-fermion (HF) superconductor. Most importantly, superconducting instability arises in the normal state that exhibits pronounced non-Fermi-liquid behavior due to the proximity of an antiferromagnetic (AFM) quantum critical point(QCP) Sidorov . Several measurements indicate that the superconducting gap has $`d`$-wave symmetry with line nodes perpendicular to the plane Movshovich ; Izawa ; Ormeno ; Rourke .
CeCoIn<sub>5</sub> exhibits several fascinating properties, which have never been observed in any other superconductor. In a strong magnetic field, the superconducting transition is of the first order, indicating a field-induced destruction of the superconducting state by Pauli paramagnetism Izawa ; Bianchi2 . Closely related to this, the emergence of a spatially inhomogeneous Fulde-Ferrel-Larkin-Ovchinnikov superconducting state has been reported in the vicinity of the upper critical field FFLO ; Kakuyanagi . Recent NMR spectra also have revealed an unusual electronic structure in the vortex core Kakuyanagi . Moreover the observation of a QCP in the vicinity of the upper critical field $`H_{c2}`$ for $`Hc`$ suggests that superconductivity prevails, preventing the development of the AFM order QCP .
Another issue of interest is the increase in the quasiparticle (QP) lifetime below $`T_c`$, indicated by thermal conductivity $`\kappa _{xx}`$ and microwave experiments Movshovich ; Izawa ; Ormeno . This feature of CeCoIn<sub>5</sub>, reminiscent of very clean high-$`T_c`$ cuprates, is not observed in other HF superconductors. Thermal Hall conductivity $`\kappa _{xy}`$, the non-diagonal element of the thermal conductivity tensor in a perpendicular magnetic field, is a powerful probe of this feature; it is purely electronic and the direct consequence of a transverse QP current, while $`\kappa _{xx}`$ includes both electronic and phononic contributions. Over the past few years, the study of the thermal Hall effect in high-$`T_c`$ cuprates has opened a new window on QP transport Zhang ; Zeini ; Simon ; Durst ; Ong .
In this Letter, we report on a study of longitudinal and transverse thermal conductivities of CeCoIn<sub>5</sub>. The results highlight a steep increase in the QP mean free path $`\mathrm{}`$ directly inferred from the temperature dependence of the thermal Hall angle $`\mathrm{\Theta }\mathrm{tan}^1\kappa _{xy}/\kappa _{xx}`$. The magnitude of $`\mathrm{}`$ estimated in this way can be compared with that extracted from the QP thermal diffusivity (i.e. the ratio of thermal conductivity to specific heat) and confirms the unusually small Fermi energy $`\epsilon _F`$ in CeCoIn<sub>5</sub>. On the other hand, even a small magnetic field leads to a dramatic decrease in $`\mathrm{}`$. This phenomenon, yet to be understood, is unique to CeCoIn<sub>5</sub>. We also found that $`T`$\- and $`H`$-dependence of $`\kappa _{xy}`$ supports the $`d`$-wave symmetry.
Single crystals of CeCoIn<sub>5</sub> ($`T_c`$ = 2.3 K) were grown by the self-flux method. Both $`\kappa _{xx}`$ and $`\kappa _{xy}`$ were measured by the steady-state method by applying the heat current along the direction with $`๐`$ $``$ $`๐`$ for $`๐ฏ`$ $``$ $`๐`$. The thermal gradients $`_xT`$ $`๐`$ and $`_yT`$ $`๐`$ were measured by RuO<sub>2</sub> thermometers. Above 0.4 K, no hysteresis was observed in sweeping $`H`$ measure . The sign of $`\kappa _{xy}`$ is negative, as for the electrical Hall conductivity $`\sigma _{xy}`$.
The inset of Fig. 1 shows the $`T`$-dependence of $`\kappa _{xx}/T`$. In zero field, upon entering the superconducting state, $`\kappa _{xx}/T`$ display a kink and exhibits a pronounced maximum at $``$0.8 K Movshovich ; Izawa . Figure 2 (a) depicts the $`H`$-dependence of $`\kappa _{xx}`$. Applying $`H`$, $`\kappa _{xx}`$ decreases up to $`H_{c2}`$ after showing an initial steep decrease. Figure 2(b) and the inset depict the $`H`$-dependence of $`|\kappa _{xy}|`$. A strong non-linear $`H`$-dependence is observed in $`|\kappa _{xy}|`$. Similar to $`\kappa _{xx}`$, the absolute slope of $`|\kappa _{xy}|`$ versus $`H`$ at high fields is reduced as the temperature is lowered. The transition to the normal state below $``$1 K for both $`\kappa _{xx}`$ and $`|\kappa _{xy}|`$ is marked by a pronounced jump, indicating a first-order transition Izawa ; Bianchi2 . (In CeCoIn<sub>5</sub> the upper critical field determined by the orbital effect $`H_{c2}^{orb}`$ is nearly 2.5 times larger than $`H_{c2}`$; $`H_{c2}^{orb}`$ 12 T.)
At low fields, as shown in the inset of Fig. 2(b), $`|\kappa _{xy}|`$ exhibits a steep increase with a linear dependence on $`H`$. At $`T`$0.64 K, $`|\kappa _{xy}|`$ exhibits a prominent peak at $``$0.06 T. It should be noted that a similar peak structure in $`\kappa _{xy}`$ has also been reported for ultra-clean YBCO single crystals Zhang . In Fig. 1, we plot the $`T`$-dependence of $`|\kappa _{xy}|/B`$ and the initial Hall slope $`|\kappa _{xy}^0|/Blim_{B0}|\kappa _{xy}|/B`$. The overall temperature dependence of $`|\kappa _{xy}^0|/B`$ is similar to $`\kappa _{xx}`$; as the temperature falls below $`T_c`$, it exhibits a pronounced maximum at $``$1 K. This behavior of $`|\kappa _{xy}^0|/B`$ again bears a striking resemblance to YBCO Zhang ; Zeini .
Before discussing the QP transport, let us examine the validity of the โtransverseโ Wiedemann-Franz (WF) law. Just above $`T_c`$, $`\kappa _{xy}`$ and $`\sigma _{xy}`$ yield a โtransverseโ Lorenz number very close to the expected WF value: $`L_{xy}=lim_{B0}\kappa _{xy}/\sigma _{xy}T1.05L_0`$ (with $`L_0=2.44\times 10^8`$ $`\mathrm{\Omega }`$W/K). This result confirms the purely electronic origin of $`\kappa _{xy}`$ and conforms with reports for copper and the normal state of YBCO Ong .
We next examine the scaling relation of $`\kappa _{xy}`$ with respect to $`T`$ and $`H`$ proposed in Ref. Simon . A scaling relation of the single variable $`x=t/\sqrt{h}`$ with $`t=T/T_c`$ and $`h=H/H_{c2}^{orb}`$ is derived as
$$\kappa _{xy}T^2F_{\kappa _{xy}}(x),$$
(1)
where $`F(x)`$ is a scaling function. As shown in Fig. 3, $`|\kappa _{xy}(T,H)|/T^2`$ collapses into a common function of $`x`$ at $`x`$ 0.07 at low temperatures within the error bar, suggesting a scaling relation, although not as prominent as in YBCO Zhang . The present scaling relation provides further support for $`d`$-wave symmetry in CeCoIn<sub>5</sub> Scaling .
At first glance, the field dependence of $`\kappa _{xx}`$ does not look like what is expected for a nodal superconductor. In contrast to fully gapped superconductors, heat transport in nodal superconductors is dominated by contributions from delocalized QP states rather than bound states associated with vortex cores Kubert ; Barash ; Vekhter ; vekhter2 ; Franz . The most remarkable effect on the thermal transport is the Doppler shift of the QPs in the presence of supercurrents around vortices. Usually, this effect leads to a $`\sqrt{H}`$ increase in the population of delocalized QPs and a subsequent increase in $`\kappa _{xx}(H)`$ that is nearly proportional to $`\sqrt{H}`$, as experimentally observed in several unconventional superconductors. The field dependence of $`\kappa _{xx}`$ observed for CeCoIn<sub>5</sub> does not show this behavior. We will argue below, that this is a result of an increase of the DOS compensated by a reduction of the mean free path, both induced by the magnetic field.
The QP mean free path is directly provided by the thermal Hall angle in the weak field limit $`\omega _c\tau 1`$,
$$\mathrm{tan}\mathrm{\Theta }\omega _c\tau \frac{eB\mathrm{}}{k_F\mathrm{}},$$
(2)
where $`\omega _c`$ is the cyclotron frequency, $`k_F`$ is the Fermi wave number. Figure 4 shows $`|\mathrm{tan}\mathrm{\Theta }|/B`$ at the zero field limit and at 5.2 T, slightly above $`H_{c2}`$, as a function of $`T`$, together with the electrical Hall angle $`(\mathrm{tan}\mathrm{\Theta }_e\sigma _{xy}/\sigma _{xx})`$ divided by $`B`$. The magnitude of $`|\mathrm{tan}\mathrm{\Theta }|/B`$ coincides well with that of $`|\mathrm{tan}\mathrm{\Theta }_e|/B`$ at the zero field limit, but at 5.2 T it is slightly larger. Below the coherence temperature, $`T^{}`$ 20 K shown by the arrow in Fig. 4, the resisitivity exhibits $`T`$-linear behavior. Below $`T^{}`$, the cotangent of the electrical Hall angle for $`B0`$ was reported to display a $`T^2`$ behavior, as shown by the dashed line, which represents lim$`{}_{B0}{}^{}|\mathrm{cot}\mathrm{\Theta }_e|/B=a+bT^2`$ with $`a`$ = 4.38 T<sup>-1</sup> and $`b`$ = 0.20 K$`{}_{}{}^{2}T_{}^{1}`$ Nakajima . Below $`T_c`$, $`|\mathrm{tan}\mathrm{\Theta }|/B`$ increases much faster than the extrapolated temperature dependence observed above $`T_c`$. This enhancement of almost one order of magnitude is a direct evidence of a drastic increase in the QP mean free path below $`T_c`$. The inset of Fig. 4 shows the value of $`\mathrm{}`$ below $`T_c`$ using $`k_F=1.85\times 10^9`$ cm<sup>-1</sup>. At $`T`$ = 0.46 K, $`\mathrm{}`$ has a value of 1.6 $`\mu `$m.
An alternative way of estimating the QP mean free path is to use the well-known link between $`\kappa _{xx}`$ and the specific heat $`C_e`$: $`\kappa _{xx}=\frac{1}{3}C_ev_F\mathrm{}`$. Now, at $`T`$ = 0.4K, with $`\kappa _{xx}=0.48W/K`$ and $`C_e`$ = 0.056J/K$``$ mol Movshovich , if we take $`v_F`$=2130 km/s (calculated using a Fermi energy, $`ฯต_F`$, of 15K Kim and a mass enhancement of $`m^{}=100m_e`$ Shishido ), the magnitude of $`\mathrm{}`$ of 1.1 $`\mu `$m is comparable to that yielded by $`\mathrm{tan}\mathrm{\Theta }/B`$. This quantitative consistency also confirms the very low value of $`ฯต_F`$ deduced from the temperature dependence of specific heat Kim .
Figure 5 displays the field dependence of the QP mean free path at $`T`$ = 0.46 K. As seen in the figure, the magnetic field dramatically suppresses the QP mean free path. Even at $`H`$ = 0.1 T ($`H/H_{c2}^{orb}`$ 1/100), $`\mathrm{}`$ is reduced by one order of magnitude. The inset of Fig. 5 shows the data on a log-log scale. For comparison, we plot the average distance between vortices $`a_v=\sqrt{\mathrm{\Phi }_0/B}`$ by a dashed line. At low fields, the QP mean free path is several times longer than the intervortex distance, but becomes comparable with $`a_v`$ at higher fields. This strong variation in the QP mean free path with magnetic field appears to be the origin of the unexpectedly flat field dependence of $`\kappa _{xx}`$ discussed above. In order to check whether the DOS of the delocalized QPs, $`N_{del}(E)`$, displays the expected $`\sqrt{H}`$ dependence, we can use the conjectures $`\kappa _{xx}N_{del}(E)\mathrm{}`$ and $`|\kappa _{xy}|/(B\kappa _{xx})\mathrm{}`$. Plotting $`\kappa _{xx}^^2B/|\kappa _{xy}|`$ as a function of $`H`$ reveals the field dependence of $`N_{del}(E)`$. As seen in the inset of Fig. 3, this ratio displays a field-dependence close to the $`\sqrt{H}`$ behavior expected for a $`d`$-wave superconductor.
The strong field dependence of $`\mathrm{}`$ is a feature that is not yet understood. There are two lines of thought to understand the QP transport. It has been argued that low energy QPs in a periodic vortex lattice are described by Bloch wavefunctions and are not scattered Kita . In contrast, in a strongly disordered vortex lattice, QP scattering is caused by Andreev scattering on the velocity field associated with the vortices. In this case, the QP mean free path is proportional to $`a_v`$. This argument was used to explain the โplateauโ in $`\kappa _{xx}(H)`$ observed in Bi2212, in which the vortex lattice is strongly distorted Krishana ; Aubin ; Franz . However, as indicated by small angle neutron scattering experiments Eskildsen , there is no reason to assume that the vortex lattice in clean CeCoIn<sub>5</sub> is strongly distorted.
The initial decrease of $`\mathrm{}`$ at low fields may be explained without invoking vortex scattering. At very low fields, where the condition $`\sqrt{H/H_{c2}^{orb}}<T/T_c`$ is satisfied, thermally excited QPs dominate over Doppler shifted QPs. It has been shown that in this regime the DOS enhanced by the Doppler shift leads to a suppression of the impurity scattering time Kubert ; Barash ; Vekhter ; vekhter2 . It is yet to be seen if this can explain the magnitude of the decrease observed at very low fields. It does not seem to be relevant to the $`H`$-dependence of $`\mathrm{}`$ at higher fields.
Although $`\sqrt{H}`$-dependent term in $`\mathrm{}`$ in a nearly periodic vortex lattice has been argued in Ref.vekhter2 , it is open question whether any deviation of the vortex from the perfect arrangement of the vortex lattice in principle produces significant effects. Several peculiarities of CeCoIn<sub>5</sub> may lead to unusual vortex-QP scattering. In fact, the very strong suppression of $`\mathrm{}`$ up to $`H_{c2}`$ has never been observed in any other superconductors, including UPd<sub>2</sub>Al<sub>3</sub> wata and YNi<sub>2</sub>B<sub>2</sub>C izawa2 with similar $`H_{c2}`$ values. One is the possible existence of antiferromagnetism in vortex cores. Several experiments indicate that the AFM phase is superseded by the superconducting transition QCP . This in turn suggests that the AFM correlation is strongly enhanced in the region around vortex cores Kakuyanagi . In this case the QPs may be significantly scattered by the AFM fluctuation in the core region. Further investigation is strongly required to clarify the origin of the peculiar QP transport in CeCoIn<sub>5</sub>.
Another feature to be considered is the energy scale of the QP spectrum in the vortex core set by the confinement energy $`\mathrm{}\omega _0\mathrm{\Delta }^2/\epsilon _F`$. For most superconductors, this energy level is negligibly small. For CeCoIn<sub>5</sub> however, $`\epsilon _F`$ 15 K and $`\mathrm{\Delta }`$ 5 K, so that $`\mathrm{}\omega _0`$ 1.5 K and the vortex spectrum becomes important at low temperatures. Moreover, when a vortex moves, energy dissipation is produced by the scattering of QPs within the vortex core. If the broadening of the QP states ($`\mathrm{}/\tau `$) turns out to be much smaller than the energy scale of the spectrum within the core, $`\omega _0\tau 1`$, the vortex system enters a new regime (the superclean regime) that is difficult to access in most superconductors. The superclean condition is equivalent to $`\mathrm{}/\xi \epsilon _F/\mathrm{\Delta }`$. In CeCoIn<sub>5</sub>, $`\mathrm{}`$$`\mu `$m and $`\xi `$ 5 nm yields $`\mathrm{}/\xi `$ 200 at low fields, which is much larger than $`\epsilon _F/\mathrm{\Delta }`$ 3. This is in sharp contrast to other superconductors in which $`\mathrm{}/\xi \epsilon _F/\mathrm{\Delta }`$. In the superclean regime, strong enhancement of the vortex viscosity, which leads to anomalous vortex dynamics including an extremely large vortex Hall angle, is expected Harris .
To conclude, we have measured the thermal Hall angle of CeCoIn<sub>5</sub> and found that it indicates a dramatic increase in the quasiparticle mean free path below $`T_c`$. In spite of the presence of a periodic vortex lattice, this enhancement is easily suppressed by a weak magnetic field. These results highlight that CeCoIn<sub>5</sub> is unique among superconductors. We found that $`\kappa _{xy}`$ displays the scaling relation expected for $`d`$-wave symmetry. Moreover the DOS of the delocalized quasiparticles obtained from the thermal Hall conductivity, are consistent with $`d`$-wave symmetry. Finally, the results indicate that CeCoIn<sub>5</sub> is in the superclean regime.
We thank S. Fujimoto, R. Ikeda, Y. Kato, T. Kita, H. Kontani, and I. Vekhter for their helpful discussions.
|
warning/0506/physics0506054.html
|
ar5iv
|
text
|
# The way back: from charge conservation to Maxwell equations
## 1 Introduction
Maxwell equations are frequently introduced - using gaussian units - in the following form :
$`๐`$ $`=`$ $`4\pi \rho `$ (1)
$`\times ๐{\displaystyle \frac{1}{c}}{\displaystyle \frac{๐}{t}}`$ $`=`$ $`{\displaystyle \frac{4\pi }{c}}๐ฃ`$ (2)
$`๐`$ $`=`$ $`0`$ (3)
$`\times ๐+{\displaystyle \frac{1}{c}}{\displaystyle \frac{๐}{t}}`$ $`=`$ $`\mathrm{๐}.`$ (4)
Equations (1) and (2) are called inhomogeneous - or Maxwell equations with sources -, while (3) and (4) are called homogeneous, or source-free equations. The four equations constitute a closed system because the couples $`(๐,๐)`$ and $`(๐,๐)`$ are related to each other through the so-called โconstitutive equationsโ. It is however not unusual to stress the fact that the equations with sources are, to some extent, conceptually different from the source-free equations. Indeed, Eqs.(3) and (4) can be understood as expressing a purely mathematical statement. To see this we start by considering a scalar field $`\phi (t,๐ฑ)`$ and a vector field $`๐(t,๐ฑ)`$, which are continuously differentiable but otherwise totally arbitrary. Then we construct the fields $`๐\times ๐`$ and $`๐\phi _t๐/c`$. Eq.(3) is then identically satisfied because the divergence of a curl vanishes. If we now take the curl of $`๐`$ and use the fact that the curl of a gradient vanishes, we see that Eq.(4) also holds true identically. We conclude that Eqs.(3) and (4) are satisfied by *arbitrary* fields, as long as these fields are constructed as above, starting from the given fields $`\phi `$ and $`๐`$. These two equations are therefore not characteristic of the electromagnetic field. They can be understood as a mathematical statement telling us that there are fields, $`\phi `$ and $`๐`$, out of which we can construct $`๐`$ and $`๐`$. The electromagnetic nature of these fields depends on the fact that they have to satisfy also equations (1) and (2), as long as $`๐=๐(๐,๐)`$ and $`๐=๐(๐,๐)`$. Equations (1) and (2) are the ones possessing a truly physical content. They are the ones which contain the sources that produce the field. It is the particular way in which these sources are related to the fields, what makes up the physical content of these equations.
Let us now turn to Maxwell equations as they are often written in tensorial form:
$`_\mu F^{\mu \nu }`$ $`=`$ $`{\displaystyle \frac{4\pi }{c}}j^\nu ,`$ (5)
$`_\mu F_{\nu \lambda }+_\nu F_{\lambda \mu }+_\lambda F_{\mu \nu }`$ $`=`$ $`0.`$ (6)
Here again, starting from an *arbitrary* four-vector $`A_\mu (x)`$ we may define an antisymmetric tensor $`F_{\mu \nu }_\mu A_\nu _\nu A_\mu `$. It is easy to see that this tensor identically satisfies the homogeneous equation (6), which is a Bianchi type identity. As before, if our $`A_\mu `$ has to describe an electromagnetic field, then it has to satisfy the inhomogeneous equation $`_\mu ^\mu A^\nu ^\nu _\mu A^\mu =4\pi j^\nu /c`$, which is another form of Eq.(5).
Summarizing, we can say that the homogeneous Maxwell equations can be considered as entailing a mathematical statement about the nature of the fields $`๐`$ and $`๐`$, or -correspondingly - about the tensor $`F_{\mu \nu }`$. The inhomogeneous Maxwell equations in turn are the ones possessing physical content. We must postulate that the electromagnetic field has to satisfy them. Now, all these things are very well known. What seems to be not so very well known is the fact that the inhomogeneous equations by themselves are also *not* characteristic of the electromagnetic field. Indeed, suppose we are given a scalar function $`\rho (t,๐ซ)`$ and a vector function $`๐ฃ(t,๐ซ)`$, both of which go to zero sufficiently rapidly as $`r\mathrm{}`$, and being such that they satisfy the equation
$$_t\rho +๐ฃ=0.$$
(7)
Then there exist vector fields $`๐(t,๐ซ)`$ and $`๐(t,๐ซ)`$ satisfying the inhomogeneous Maxwell equations
$`๐`$ $`=`$ $`\rho `$ (8)
$`\times ๐`$ $`=`$ $`๐ฃ+_t๐.`$ (9)
The existence of $`๐`$ and $`๐`$ can be proved by explicit construction. Such a construction rests on Helmholtz theorem , which is discussed below. For now, it suffices to say that - loosely speaking - โa vector field is determined by its divergence and its curlโ. Thus, according to Helmholtz theorem, Eq.(8) can be solved for $`๐`$ (though the solution is not unique) when $`\rho `$ is given. From equations (8) and (7) we see that $`\left(_t๐+๐ฃ\right)=0`$. Applying Helmholtz theorem again we can show that there is a field $`๐`$ whose curl is $`_t๐+๐ฃ`$. This is equation (9). Note that we have written the inhomogeneous Maxwell equations in MKS units, which are the convenient units for what follows.
We see then that the continuity equation (7) entails the inhomogeneous Maxwell equations. The continuity equation expresses the conservation of something. This something must not necessarily be electric charge. It could be mass as well, or any other quantity - like probability, for instance. We are thus led to conclude that the inhomogeneous Maxwell equations are also not characteristic of electromagnetism. They hold true whenever something is conserved. Putting things this way we bring to the fore the fundamental role played by the constitutive equations, $`๐=๐(๐,๐)`$ and $`๐=๐(๐,๐)`$, whatever their precise form might be. They constitute the link between the homogeneous and the inhomogeneous Maxwell equations. It is this link what turns the four equations into a closed system. Neither the inhomogeneous nor the homogeneous equations by themselves are characteristic of electromagnetism. They must be linked to one another in order to conform a closed system of equations that is characteristic of electromagnetic phenomena.
In the following section we discuss Helmholtz theorem. Although this theorem can be found in several textbooks and articles, for our purposes it is useful to present it in a form which brings to the fore its connection with Greenโs functions.
## 2 Helmholtz theorem
Here we discuss Helmholtz theorem by following an approach which is slightly different from the one presented in several textbooks. Helmholtz theorem states that a vector field $`๐ฏ`$ is completely determined by giving its divergence and its curl, together with its normal component, $`\widehat{๐ง}๐ฏ`$, at the boundary of the domain where such a vector field is to be determined. For physical applications it is natural to take as โboundaryโ an infinitely distant surface and $`๐ฏ`$ vanishing there. Helmholtz theorem then says that we can write $`๐ฏ`$ in terms of two potentials, $`U`$ and $`๐`$, in the form
$$๐ฏ(๐ฑ)=U(๐ฑ)+\times ๐(๐ฑ),$$
(10)
where $`U`$ and $`๐`$ can be expressed in terms of the divergence and the curl of $`๐ฏ(๐ซ)`$, respectively. Now, put in this form, Helmholtzโs theorem might appear as a result that is rather awkward to prove. Let us thus try to lay bare what motivates it. To this end, consider the following two vector identities, in which the Laplacian $`^2`$ appears:
$`\left(U\right)`$ $`=`$ $`^2U`$ (11)
$`\times \left(\times ๐\right)`$ $`=`$ $`\left(๐\right)^2๐.`$ (12)
Add to these relations the equation satisfied by a Green function $`G(๐ฑ,๐ฒ)`$, on which we impose the condition that it vanishes at infinity:
$`^2G(๐ฑ,๐ฒ)`$ $`=`$ $`\delta ^3(๐ฑ๐ฒ),`$ (13)
$`G(๐ฑ,๐ฒ)`$ $`=`$ $`{\displaystyle \frac{1}{4\pi \left|๐ฑ๐ฒ\right|}}.`$ (14)
By means of $`G(๐ฑ,๐ฒ)`$ we can introduce $`U`$ and $`๐`$ as โpotentialsโ associated with two given โdensitiesโ, $`\rho `$ and $`๐ฃ`$, through
$`U(๐ฑ)`$ $`=`$ $`{\displaystyle \frac{1}{4\pi }}{\displaystyle \frac{\rho (๐ฒ)}{\left|๐ฑ๐ฒ\right|}d^3y},`$ (15)
$`๐(๐ฑ)`$ $`=`$ $`{\displaystyle \frac{1}{4\pi }}{\displaystyle \frac{๐ฃ(๐ฒ)}{\left|๐ฑ๐ฒ\right|}d^3y}.`$ (16)
We assume that $`\rho `$ and $`๐ฃ`$ vanish at infinity. The potentials then satisfy
$`^2U(๐ฑ)`$ $`=`$ $`\rho (๐ฑ),`$ (17)
$`^2๐(๐ฑ)`$ $`=`$ $`๐ฃ(๐ฑ).`$ (18)
$`๐ฃ(๐ฑ)`$ $`=`$ $`0๐(๐ฑ)=0.`$ (19)
The validity of Eqs.(17,18) follows directly from the definitions given by Eqs.(15,16), together with Eqs.(13,14). In order to see that $`๐ฃ(๐ฑ)=0`$ implies that $`๐(๐ฑ)`$ is divergenless we need a little more ellaborated calculation:
$`๐(๐ฑ)`$ $`=`$ $`{\displaystyle \frac{1}{4\pi }}{\displaystyle _๐ฑ\left(\frac{๐ฃ(๐ฒ)}{\left|๐ฑ๐ฒ\right|}\right)d^3y}={\displaystyle \frac{1}{4\pi }}{\displaystyle ๐ฃ(๐ฒ)_๐ฑ\left(\frac{1}{\left|๐ฑ๐ฒ\right|}\right)d^3y}`$ (20)
$`=`$ $`{\displaystyle \frac{1}{4\pi }}{\displaystyle ๐ฃ(๐ฒ)_๐ฒ\left(\frac{1}{\left|๐ฑ๐ฒ\right|}\right)d^3y}`$ (21)
$`=`$ $`{\displaystyle \frac{1}{4\pi }}{\displaystyle \underset{i=1}{\overset{3}{}}j_i(๐ฒ)\frac{}{y^i}\left(\frac{1}{\left|๐ฑ๐ฒ\right|}\right)d^3y}`$ (22)
$`=`$ $`{\displaystyle \frac{1}{4\pi }}\left[{\displaystyle \frac{๐ฃ(๐ฒ)๐ง(๐ฒ)}{\left|๐ฑ๐ฒ\right|}๐S}{\displaystyle \underset{i=1}{\overset{3}{}}\frac{1}{\left|๐ฑ๐ฒ\right|}\frac{j_i(๐ฒ)}{y^i}d^3y}\right]=0.`$ (23)
In the last step - which resulted from an integration by parts - the volume integral was replaced by a surface integral using the divergence - or Stokes \- theorem. Such a surface integral vanishes when the volume of integration goes to infinity, because $`๐ฃ`$ has been assumed to vanish at infinity. The second term vanishes because of the requirement $`๐ฃ=0`$.
From Eqs.(11) and (12) together with (17,18,19), we see that
$`\left(U\right)`$ $`=`$ $`\rho (๐ฑ),`$ (24)
$`\times \left(\times ๐\right)`$ $`=`$ $`๐ฃ(๐ฑ).`$ (25)
This suggests us to define a field $`๐ฏ=U+\times ๐`$. Such a field satisfies
$$๐ฏ=\rho (๐ฑ),\text{ }\times ๐ฏ=๐ฃ(๐ฑ).$$
(26)
This way we arrive naturally at the following statement: if we are given the divergence $`\rho (๐ฑ)`$ and the curl $`๐ฃ(๐ฑ)`$ of a vector field $`๐ฏ(๐ฑ)`$ which vanishes at infinity, then we can write this field as $`๐ฏ=U+\times ๐`$, where $`U`$ and $`๐`$ are given in terms of $`\rho `$ and $`๐ฃ`$ by Eqs.(15) and (16). In order to see that $`\rho `$ and $`๐ฃ`$ uniquely determine $`๐ฏ`$, it suffices to show that when both the divergence and the curl of a field vanish, then the field itself vanishes identically. This follows from what we have done so far. Indeed, we have shown that the following equation holds true identically:
$$๐ฏ(๐ฑ)=_๐ฑ\left(G(๐ฑ,๐ฒ)๐ฏ(๐ฒ)d^3y\right)+_๐ฑ\times \left(G(๐ฑ,๐ฒ)\times ๐ฏ(๐ฒ)d^3y\right),$$
(27)
with the Greenโs function $`G(๐ฑ,๐ฒ)`$ satisfying Eqs.(13,14). Hence, if $`๐ฏ=0`$ and $`\times ๐ฏ=\mathrm{๐}`$, then $`๐ฏ=\mathrm{๐}`$. We conclude that given two fields, $`๐ฏ_1`$ and $`๐ฏ_2`$, having the same divergence and curl, they must in fact be the same field. This, because their difference $`๐ฏ=๐ฏ_1๐ฏ_2`$ vanishes identically, as a consequence of $`๐ฏ=0`$ and $`\times ๐ฏ=\mathrm{๐}`$.
Finally, let us first note that Eq.(27) holds for Greenโs functions other than the one defined in Eq.(14). Indeed, the only property we need to assume about the Green function $`G(๐ฑ,๐ฒ)`$ is that it be of the form $`G(๐ฑ๐ฒ)`$. This is true anyway, whenever $`G(๐ฑ,๐ฒ)`$ fulfills Eq.(13). As to the field $`๐ฏ`$, it has been assumed to vanish at infinity. In fact, it suffices to assume that it vanishes faster than $`1/r`$ for large $`r`$. Note also that if we prescribe only the divergence $`๐ฏ=\rho (๐ฑ)`$ of a field, then what we can deduce from this sole condition is that
$$๐ฏ(๐ฑ)=_๐ฑ\left(G(๐ฑ,๐ฒ)๐ฏ(๐ฒ)d^3y\right)+_๐ฑ\times ๐(๐ฑ),$$
(28)
with $`๐(๐ฑ)`$ arbitrary. If we instead prescribe only the curl $`\times ๐ฏ=๐ฃ(๐ฑ)`$ of a field, then we have
$$๐ฏ(๐ฑ)=_๐ฑ\times \left(G(๐ฑ,๐ฒ)\times ๐ฏ(๐ฒ)d^3y\right)+_๐ฑV(๐ฑ),$$
(29)
with $`V(๐ฑ)`$ arbitrary.
## 3 Maxwell equations and Helmoltz theorem
We have discussed Helmholtz theorem in the framework of $`R^3`$. That is, the vector fields we have considered are of the form $`๐ฏ(๐ฑ)`$. However, all the results we have obtained so far remain valid if we assume these fields to depend on a set of additional parameters. They can be assumed to have been there all the way, but without having been shown explicitly. Let us denote one of these parameters as $`t`$. For the moment, we do not assign to it any physical meaning. Of course, the notation anticipates that it will be identified in due course with the time variable.
Let us start by assuming that we are given the divergence $`\rho `$ of a field, which is a function not only of position but of the parameter $`t`$ as well, which we now make explicit, i.e., $`\rho =\rho (t,๐ซ)`$. Let our boundary condition be such that $`\rho `$ vanishes at spatial infinity. Helmholtz theorem states that there is a field, call it $`๐`$, satisfying
$$๐(t,๐ซ)=\rho (t,๐ซ).$$
(30)
As we have seen, the field $`๐(t,๐ซ)`$ is explicitly given by
$$๐(t,๐ซ)=_๐ซ\frac{\rho (t,๐ซ_1)}{4\pi \left|๐ซ๐ซ_1\right|}d^3r_1+_๐ซ\times ๐(t,๐ซ),$$
(31)
with $`๐`$ an arbitrary field that we are free to put equal to zero, if we want. We stress that $`t`$ plays, in all of this, only the role of a parameter that can be appended to the fields, *without having any dynamical meaning*. The field $`๐(t,๐ซ)`$ is required to satisfy only one condition we have put upon it, i.e., $`๐(t,๐ซ)=\rho (t,๐ซ).`$ The curl of $`๐`$ has been left unspecified, or else set equal to zero.
Consider now a field $`๐ฃ(t,๐ซ)`$ depending on the same parameter $`t`$ as $`\rho `$ does. Assume next that $`\rho (t,๐ซ)`$ and $`๐ฃ(t,๐ซ)`$ satisfy a continuity equation:
$$_t\rho +๐ฃ=0.$$
(32)
By using Eq.(30) the continuity equation can be written as
$$\left(_t๐+๐ฃ\right)=0.$$
(33)
The divergenless vector $`_t๐+๐ฃ`$ can thus be taken as being the curl of a field $`๐(t,๐ซ)`$. Indeed, according to what we have seen before, the equation $`\times ๐=๐ฃ+_t๐`$ can be solved as
$$๐(t,๐ซ)=_๐ซ\times \frac{๐ฃ(t,๐ซ_1)+_t๐(t,๐ซ_1)}{4\pi \left|๐ซ๐ซ_1\right|}d^3r_1+_๐ซV(t,๐ซ).$$
(34)
As long as we do not specify $`๐`$ the function $`V`$ remains undetermined. In any case, the Maxwell equations $`๐=\rho `$ and $`\times ๐_t๐=๐ฃ`$ hold true as a consequence of the continuity equation and Helmholtz theorem. However, these equations are not enough to determine the dynamics of the fields $`๐`$ and $`๐`$, even though we may ascribe to $`t`$ the meaning of time. This must be so because - to begin with \- the continuity equation alone does not entail enough information about the dynamics of $`\rho `$ and $`๐ฃ`$. But even in case we were provided with the complete dynamics of $`\rho `$ and $`๐ฃ`$, from a physical point of view it is clear that some assumptions must be made concerning the properties of the medium (e.g., โspace-timeโ) in order to fix the dynamics of the electromagnetic fields that will eventually propagate in such a medium.
Nonetheless, let us pursue a little bit further the mathematical approach suggested by Helmholtz theorem. The potentials $`U`$ and $`๐`$ in terms of which we defined the field $`๐ฏ(๐ฑ)`$ read here
$`\phi (t,๐ซ)`$ $`=`$ $`{\displaystyle \frac{\rho (t,๐ซ_1)}{4\pi \left|๐ซ๐ซ_1\right|}d^3r_1},`$ (35)
$`๐(t,๐ซ)`$ $`=`$ $`{\displaystyle \frac{๐ฃ(t,๐ซ_1)+_t๐(t,๐ซ_1)}{4\pi \left|๐ซ๐ซ_1\right|}d^3r_1},`$ (36)
respectively, and we have that $`๐(t,๐ซ)=_๐ซ\phi (t,๐ซ)+_๐ซ\times ๐(t,๐ซ)`$ and $`๐(t,๐ซ)=_๐ซ\times ๐(t,๐ซ)+_๐ซV(t,๐ซ)`$. We obtain then, from Eq.(35),
$$๐(t,๐ซ)=\frac{1}{4\pi }d^3r_1\frac{\rho (t,๐ซ_1)}{\left|๐ซ๐ซ_1\right|^2}\frac{\left(๐ซ๐ซ_1\right)}{\left|๐ซ๐ซ_1\right|}+_๐ซ\times ๐(t,๐ซ).$$
(37)
For the special case of a point-like charge moving along the curve $`๐ซ_0(t)`$ we put $`\rho (t,๐ซ)=q\delta (๐ซ๐ซ_0(t))`$ and the above expression reduces to
$$๐(t,๐ซ)=\frac{q}{4\pi \left|๐ซ๐ซ_0(t)\right|^2}\frac{๐ซ๐ซ_0(t)}{\left|๐ซ๐ซ_0(t)\right|}+_๐ซ\times ๐(t,๐ซ).$$
(38)
According to Eqs.(37) or (38) the field $`๐(t,๐ซ)`$ at time $`t`$ entails an instantaneous Coulomb field produced by a continuous charge distribution $`\rho `$, or else by a point-like charge $`q`$. Such a result would correspond to an instantaneous response of the field to any change suffered by the charge distribution. That would be in contradiction with the finite propagation-time needed by any signal. Whatever the field $`๐(t,๐ซ)`$ might be, it must contain a similar instantaneous contribution that cancels the former one, if we want the present approach to bear any physical interpretation. Such an issue has been discussed and cleared, in the case of the *complete* set of Maxwell equations, by showing that both the longitudinal and the transverse parts of the electric field contain instantaneous contributions, which turn out to cancel each other . Note also that by taking $`๐`$ equal to zero in Eq.(44) we have $`\times ๐=\mathrm{๐}`$ in our case, which is not what happens when $`๐`$ has to satisfy (together with $`๐`$) the complete system of Maxwell equations. In any event, as we have already stressed, it is necessary to add some additional information to the one derived from the continuity equation, in order to fix the dynamics of the fields. We do this in the following form. Instead of taking the potentials $`\phi `$ and $`๐`$ as given by Eqs.(35) and (36), we assume them as additional quantities, out of which we define the fields $`๐`$ and $`๐`$ through
$`๐(t,๐ซ)`$ $`=`$ $`_๐ซ\phi (t,๐ซ)_t๐(t,๐ซ),`$ (39)
$`๐(t,๐ซ)`$ $`=`$ $`_๐ซ\times ๐(t,๐ซ).`$ (40)
These fields obey then the homogeneous Maxwell equations identically:
$`๐`$ $`=`$ $`0`$ (41)
$`\times ๐+_t๐`$ $`=`$ $`\mathrm{๐}.`$ (42)
Side by side to these two Maxwell equations we write the inhomogeneous ones:
$`๐`$ $`=\rho `$ (43)
$`\times ๐`$ $`=๐ฃ+_t๐.`$ (44)
We stress once again that - up to this point - the homogeneous and the inhomogeneous equations are independent from one another. We may connect them through some *constitutive equations*, like, e.g.,
$`๐`$ $`=\epsilon ๐,`$ (45)
$`๐`$ $`=\mu ^1๐.`$ (46)
These equations are usually assumed to describe a linear medium of electrical permittivity $`\epsilon `$ and magnetic permeability $`\mu `$. A particular case of such a medium is vacuum, and the system of equations, Eqs.(41, 42, 43, 44), that arises out of a connection like the one given by Eqs.(45, 46) is what we know as the complete system of Maxwell equations.
Without connecting $`\left(๐\text{}๐\right)`$ with $`\left(๐\text{}๐\right)`$ through some constitutive equations, we have no closed system. The equations that we have written down for $`\left(๐\text{}๐\right)`$, that is Maxwell equations with sources, can also be written down for a fluid, for example. We can expect that any conclusion that can be derived in the realm of electrodynamics from the equations $`๐=\rho `$ and $`\times ๐=๐ฃ+_t๐`$ *without* coupling them to the source-free Maxwell equations, will have a corresponding result in the realm of fluid dynamics. This assertion can be illustrated by two examples: 1) A fluid having a point-like singularity in its density $`\rho `$ (one can achieve this approximately, by using an appropriate sink): one obtains in this case a velocity-field obeying a law that is mathematically identical to Coulombโs law . 2) A fluid where a so-called vortex tube appears (tornadoes and whirl-pools are associated phenomena), in which case - after approximating the vortex-tube by a line singularity - one obtains a velocity-field through an expression which is mathematically identical to the Biot-Savart law .
## 4 Tensorial formulation
The derivation of the inhomogeneous Maxwell equations as a consequence of charge conservation is nothing new . It follows as a direct application of a theorem of de Rahm for differential forms . According to this theorem, given a four-vector $`j^\alpha (x)`$ for which a continuity equation holds, i.e., $`_\alpha j^\alpha =0`$, there exists an antisymmetric tensor $`F^{\alpha \beta }=F^{\beta \alpha }`$ fulfilling $`_\alpha F^{\alpha \beta }=j^\beta `$. As we said before, this last equation is nothing but the tensorial form of the inhomogeneous Maxwell equations, Eqs.(43) and (44). Now, the tensor $`F^{\alpha \beta }`$ is not always derivable from a vector $`A^\alpha `$. In order to be derivable from $`A^\alpha `$ in the form $`F^{\alpha \beta }=^\alpha A^\beta ^\beta A^\alpha `$, it must satisfy the equation $`^\alpha F^{\beta \gamma }+^\beta F^{\gamma \alpha }+^\gamma F^{\alpha \beta }=0`$. This is the tensorial form of the homogeneous Maxwell equations, Eqs.(41,42). In other words, given $`j^\alpha `$ and $`A^\alpha `$, with $`j^\alpha `$ satisfying a continuity equation, we may introduce two antisymmetric tensors, $`F_{(1)}^{\alpha \beta }`$ and $`F_{(2)}^{\alpha \beta }`$. The first one can be determined so as to satisfy $`_\alpha F_{(1)}^{\alpha \beta }=j^\beta `$, according to de Rahmโs theorem. The second tensor, defined through $`F_{(2)}^{\alpha \beta }^\alpha A^\beta ^\beta A^\alpha `$, satisfies $`^\alpha F_{(2)}^{\beta \gamma }+^\beta F_{(2)}^{\gamma \alpha }+^\gamma F_{(2)}^{\alpha \beta }=0`$ identically. In order that these two equations do conform a closed system, i.e., the *total* system of Maxwell equations, we need to connect $`F_{(1)}^{\alpha \beta }`$ with $`F_{(2)}^{\alpha \beta }`$ through some constitutive relation. In the following we ellaborate on all this, but without employing the tools of differential forms on manifolds, which - in spite of their usefulness - cannot be said yet to be part of the lore of physics.
It is indeed not necessary to rest on de Rhamโs theorem and the theory of differential forms on manifolds, in order to derive the foregoing conclusions in tensorial form. One could start with the tensorial form of Helmholtz theorem and go-ahead with a similar reasoning as the one we have followed in the preceding sections. We shall however proceed by explicitly constructing a tensor fulfilling our requirements.
Let us thus start by assuming that we are given a vector field $`j^\alpha `$. We want to show that there is an antisymmetric tensor $`F^{\alpha \beta }`$ fulfilling
$$_\alpha F^{\alpha \beta }=j^\beta .$$
(47)
Note first that from Eq.(47) and the antisymmetry of $`F^{\alpha \beta }`$ it follows that $`j^\beta `$ must satisfy the continuity equation:
$$_\beta j^\beta =0.$$
(48)
We now demonstrate the existence of the tensor $`F^{\alpha \beta }`$ by explicit construction. To this end, we consider the Green function $`G(x,x^{})`$ satisfying
$$_\mu ^\mu G(x,x^{})=\delta \left(xx^{}\right).$$
(49)
Given $`G(x,x^{})`$ and $`j^\alpha `$ we introduce the potential $`A^\mu (x)`$ as
$$A^\mu (x)=G(x,x^{})j^\mu (x^{})d^4x^{},$$
(50)
and define
$`F^{\mu \nu }\left(x\right)`$ $``$ $`^\mu A^\nu (x)^\nu A^\mu (x)`$ (51)
$`=`$ $`{\displaystyle \left[^\mu G(x,x^{})j^\nu (x^{})^\nu G(x,x^{})j^\mu (x^{})\right]d^4x^{}}.`$ (52)
Let us now take the four-divergence of the above defined tensor $`F^{\mu \nu }\left(x\right)`$:
$$_\mu F^{\mu \nu }(x)=\left[_\mu ^\mu G(x,x^{})j^\nu (x^{})_\mu ^\nu G(x,x^{})j^\mu (x^{})\right]d^4x^{}.$$
(53)
Because $`G(x,x^{})`$ satisfies Eq.(49), the first integral in Eq.(53) is equal to $`j^\nu (x)`$. As for the second integral, in order to show that it is zero we do as follows. Because $`G(x,x^{})`$ satisfies Eq.(49), it must be a function of $`\left(xx^{}\right)`$, so that $`_\mu G(x,x^{})=_\mu ^{}G(x,x^{})`$, where $`_\mu ^{}/x^\mu `$. We use this property and integrate by parts the second term in (53); at the same time we replace the first term by $`j^\nu (x)`$:
$`_\mu F^{\mu \nu }(x)`$ $`=`$ $`j^\nu (x)+^\nu {\displaystyle \left[_\mu ^{}\left(G(x,x^{})j^\mu (x^{})\right)G(x,x^{})_\mu ^{}j^\mu (x^{})\right]d^4x^{}}`$ (54)
$`=`$ $`j^\nu (x)+^\nu {\displaystyle _\mu ^{}\left(G(x,x^{})j^\mu (x^{})\right)d^4x^{}}.`$ (55)
We may now employ the generalized Gauss theorem in order to show that the four-volume integral on the right-hand side of (55) vanishes. The four-volume has as its boundary a three-dimensional hypersurface $`S^{}`$ whose differential element we denote by $`dS_\mu ^{}`$. Thus, because $`j^\mu `$ vanishes at spatial infinity,
$$_\mu ^{}\left(G(x,x^{})j^\mu (x^{})\right)d^4x^{}=G(x,x^{})j^\mu (x^{})๐S_\mu ^{}=0,$$
(56)
when we let $`S^{}\mathrm{}`$, and with this result Eq.(55) reduces to (47).
Now, just as in the three-dimensional case, where the divergence of a field did not determine the field uniquely (see Eq.(28)), by subjecting $`F^{\mu \nu }`$ to the sole condition of fulfilling Eq.(47) we do not fix $`F^{\mu \nu }`$ completely. Indeed, the tensor $`K^{\mu \nu }`$, which is defined below in terms of an *arbitrary* four-vector $`B_\rho `$, fulfills also Eq.(47):
$`K^{\mu \nu }`$ $`=`$ $`F^{\mu \nu }{\displaystyle \frac{1}{2}}ฯต^{\mu \nu \rho \sigma }\left(_\rho B_\sigma _\sigma B_\rho \right)`$ (57)
$``$ $`F^{\mu \nu }{\displaystyle \frac{1}{2}}ฯต^{\mu \nu \rho \sigma }H_{\rho \sigma }F^{\mu \nu }\stackrel{~}{H}^{\mu \nu }.`$ (58)
Here, $`ฯต^{\mu \nu \rho \sigma }`$ is the totally antisymmetric Levi-Civita tensor (in fact, a tensor density). The four-divergences of $`K^{\mu \nu }`$ and $`F^{\mu \nu }`$ are the same because, due to the antisymmetry of $`ฯต^{\mu \nu \rho \sigma }`$ and the symmetry of partial derivatives like $`_\mu _\rho `$, we have
$$_\mu \stackrel{~}{H}^{\mu \nu }=\frac{1}{2}ฯต^{\mu \nu \rho \sigma }\left(_\mu _\rho B_\sigma _\mu _\sigma B_\rho \right)0.$$
(59)
Hence, we obtain Maxwell equation with sources:
$$_\alpha K^{\alpha \beta }=j^\beta ,$$
(60)
together with the identity $`_\mu F_{\nu \lambda }+_\nu F_{\lambda \mu }+_\lambda F_{\mu \nu }0`$, which follows from the definition of $`F^{\mu \nu }`$, as given in Eq.(51). Introducing the dual $`\stackrel{~}{F}^{\mu \nu }=ฯต^{\mu \nu \alpha \beta }F_{\alpha \beta }/2`$ of the tensor $`F_{\alpha \beta }`$ we can write the former identity in the form
$$_\mu \stackrel{~}{F}^{\mu \nu }=0.$$
(61)
Eqs.(60) and (61) are Maxwell equations in tensorial form. As we said before, they constitute a closed system as long as $`K^{\alpha \beta }`$ and $`F^{\mu \nu }`$ become related to each other by some constitutive equations. A general, linear algebraic, relationship between these tensors has the form
$$K^{\alpha \beta }=\frac{1}{2}\chi ^{\alpha \beta \rho \sigma }F_{\rho \sigma },$$
(62)
where $`\chi ^{\alpha \beta \rho \sigma }`$ is called the constitutive tensor. It has the following symmetry properties: $`\chi ^{\alpha \beta \rho \sigma }=\chi ^{\beta \alpha \rho \sigma }=\chi ^{\alpha \beta \sigma \rho }=\chi ^{\rho \sigma \alpha \beta }`$. In three-dimensional notation the components of $`K^{\alpha \beta }`$ are $`๐`$ and $`๐`$, whereas those of $`F^{\mu \nu }`$ are $`๐`$ and $`๐`$. For free-space, the only nonzero components of $`\chi ^{\alpha \beta \rho \sigma }`$ have the values $`\epsilon _0`$ and $`1/\mu _0`$, corresponding to the electrical permittivity $`\epsilon _0`$ and magnetic permeability $`\mu _0`$ of the vacuum.
The properties of the medium can be specified through an equation like (62), as well as by introducing some other quantities that describe the polarization and magnetization of the medium. In three-vector notation these quantities are the vectors $`๐`$ and $`๐`$, respectively. Their relation to $`(๐,๐)`$ and $`(๐,๐)`$ is given, in the simplest case, by
$`๐`$ $`=`$ $`\epsilon _0๐+๐,`$ (63)
$`๐`$ $`=`$ $`{\displaystyle \frac{1}{\mu _0}}๐๐.`$ (64)
In tensor notation, $`๐`$ and $`๐`$ are subsumed into an antisymmetric tensor:
$$M^{\alpha \beta }=\left(\begin{array}{cccc}0\hfill & P_1\hfill & P_2\hfill & P_3\hfill \\ P_1\hfill & 0\hfill & M_3\hfill & M_2\hfill \\ P_2\hfill & M_3\hfill & 0\hfill & M_1\hfill \\ P_3\hfill & M_2\hfill & M_1\hfill & 0\hfill \end{array}\right).$$
(65)
This choice corresponds to the assignment $`E_i=F_{0i}`$ for the electric field, and $`B_i=ฯต_{ijk}F_{jk}/2`$ for the magnetic field (Latin indices run from $`1`$ to $`3`$). By relating $`M^{\alpha \beta }`$ to $`K^{\alpha \beta }`$ through $`K^{\alpha \beta }=F^{\alpha \beta }M^{\alpha \beta }`$ we can rewrite the inhomogeneous Maxwell equation (60) as
$$_\alpha F^{\alpha \beta }=j^\beta +_\alpha M^{\alpha \beta }.$$
(66)
Written in this form, the inhomogeneous Maxwell equation makes the magnetization-polarization tensor $`M^{\alpha \beta }`$ appear as a source of the electromagnetic field $`F^{\alpha \beta }^\mu A^\nu (x)^\nu A^\mu (x)`$. The constitutive equations are given in this case as a connection between $`M^{\alpha \beta }`$ and $`F^{\alpha \beta }`$ . At any rate, one has to make some hypothesis concerning the electromagnetic properties of the medium - be it vacuum or any other kind of medium - in order to obtain the closed system of Maxwell equations. The simplest assumption is to attribute to the medium the property of reacting locally and instantaneously to the presence of a field. It could be, however, that such an assumption describes reality as a first approximation only.
Finally, we want to stress the central role played by the Green function $`G(x,x^{})`$. We assumed this function to satisfy Eq.(49), an equation entailing the velocity of light. One possible solution of (49) is given by the retarded Green function
$$G(x,x^{})=\frac{1}{4\pi \left|๐ซ๐ซ^{}\right|}\delta (tt^{}\left|๐ซ๐ซ^{}\right|/c).$$
(67)
This is the solution of Eq.(49) to which we ascribe physical meaning. By using it in Eq.(50) we are actually prescribing how the source $`j^\mu (x^{})`$ at a space-time point $`x^{}`$ gives rise to an electromagnetic field $`A^\mu (x)`$ at a distant point $`x`$; a field that virtually acts upon a second charge or current density that is located at such a distant point. There is therefore a fundamental piece of information concerning the electromagnetic properties of space-time that is already contained in the Green function, be it given through the special form of $`G(x,x^{})`$, as in Eq.(67), or through the equation it has to satisfy, e.g., Eq.(49).
## 5 Summary and Conclusions
We showed that starting from charge conservation one can arrive at equations which are mathematically identical to Maxwell equations with sources. These equations are therefore tightly linked to a general statement telling us that something is conserved. Consider anything - charge, matter, or whatever \- that is contained inside an arbitrary volume. Consider also that this thing is in a quantity that changes with time. If the change is exclusively due to a flow through the volumeโs boundary, a continuity equation holds true. Then, as a consequence of it, a pair of Maxwell-like equations must be fulfilled by some auxiliary fields, which take the role ascribed to $`๐`$ and $`๐`$ in Maxwell equations.
That Maxwell equations with sources follow from charge conservation is a mathematical fact that has been known since a couple of years , although it is usually not mentioned in standard textbooks of electromagnetism. Maxwell equations with sources involve the fields $`๐`$ and $`๐`$, whereas the source-free equations involve the fields $`๐`$ and $`๐`$. It is through some constitutive equations connecting $`\left(๐\text{}๐\right)`$ with $`\left(๐\text{}๐\right)`$ that we obtain a closed system, i.e., the complete system of Maxwell equations. The constitutive equations express, in some way or another, the underlying properties of the medium where the fields act or are produced. From this perspective, the Maxwell equations entail besides charge conservation some properties of the medium, yet to be unraveled. These properties are effectively described, in the simplest case, through the permittivity $`\epsilon `$ and the permeability $`\mu `$ of the medium. The first one refers to electrical, the second one to magnetic, properties of the medium, be it vacuum or any other one. It is just when the equations for $`\left(๐\text{}๐\right)`$ together with those for $`\left(๐\text{}๐\right)`$ do conform a closed system, that we can derive a wave equation for these fields. The velocity of wave propagation is then given by $`c=(\epsilon \mu )^{1/2}`$, the velocity of light. This must be in accordance with the assumptions we make when choosing a physically meaningful Green function. It is remarkable that the velocity of light can be decomposed in terms of a product of two independent parameters. However, the development of physics has led us to see $`c`$ as a fundamental constant of Nature, instead of $`\epsilon `$ and $`\mu `$. Nevertheless, currently discussed and open questions related to accelerated observers, Unruh radiation, self-force on a charge, magnetic monopoles and the like, might well require an approach where the role of $`c`$ recedes in favor of quantities like $`\epsilon `$ and $`\mu `$. Maxwell equations, when written in the - by now - most commonly used Gaussian units, do not include but the single constant $`c`$, hiding so $`\epsilon `$ and $`\mu `$ from our view. These last two constants might well be key pieces that remain buried under the beauty of a unified theory of electromagnetic phenomena, which is the version of electrodynamics that we know and use today. A version that should not be regarded as a closed chapter in the book of classical physics.
## 6 Acknowledgments
The author is very much indebted to Professor F. W. Hehl for his comments concerning a first version of the present article, as well as for drawing his attention to the rich literature on a series of topics related to the present work.
|
warning/0506/astro-ph0506389.html
|
ar5iv
|
text
|
# Kaluza-Klein Dark Matter and Galactic Antiprotons
## I Introduction
Non-baryonic dark matter has been shown to be the dominant matter component of our Universe by several independent measurements โ see Freedman and Turner (2003) for a review. The recently published WMAP results D. N. Spergel et al. (2003), combined with ACBAR, CBI and 2dFGRS, lead to precise estimates of the baryonic, matter and total densities : $`\mathrm{\Omega }_bh^2=0.0224\pm 0.0009`$, $`\mathrm{\Omega }_mh^2=0.135\pm 0.009`$ and $`\mathrm{\Omega }_{tot}=1.02\pm 0.02`$. Weakly Interacting Massive Particles (WIMPs) are the favourite candidates to account for the Cold (non-baryonic) Dark Matter (CDM) as the required relic density can be naturally generated. Theoretically well-founded, neutralinos are certainly the most extensively studied example. On the other hand, in spite of the very important efforts devoted to direct and indirect searches in this direction, supersymmetric particles have not yet been discovered and alternative candidates should be considered. Among them, Kaluza-Klein (KK) particles are promising. So far, they arise as stable viable WIMPs in two frameworks : In Universal Extra Dimensions (UED) Appelquist et al. (2001) and in some warped geometries ร la Randall-Sundrum Randall and Sundrum (1999).
In the case of UED, all standard model fields propagate in one or more flat compact extra dimensions โ unlike models with Large Extra Dimensions ร la ADD Arkani-Hamed et al. (1998). As a result, the combination of a translation by $`\pi R`$ with a flip of sign of all odd states in the KK Fourier decomposition of the bulk fields โ named KK-parity โ is conserved. This implies that the lightest first level KK particle (LKP) cannot decay into standard model modes and is stable. Such a Kaluza-Klein particle is likely to be associated with the first KK excitation of the photon, more precisely the first excitation of the hypercharge gauge boson Cheng et al. (2002a), and is refered to as $`B^{(1)}`$. Depending on the number of dimensions and on the mass difference between the LKP and the NLKP โ next to LKP โ the $`B^{(1)}`$ mass is expected to lie in the range 300 โ 1000 GeV if it is to account for dark matter Servant and Tait (2002a) โ for a recent analysis see Kakizaki et al. (2005). Although not very narrow, this range is much smaller than in the neutralino case and this approach has much less free parameters. Furthermore, this range is fully compatible with experimental constraints which lead โ in the $`D=5`$ case โ from precision electroweak measurements to compactification radii satisfying $`R^1\text{ }>300`$ GeV. Direct detection of the $`B^{(1)}`$ LKP has been studied in germanium, sodium iodine and xenon detectors Servant and Tait (2002b); Cheng et al. (2002b). Indirect detection through gamma-rays Cheng et al. (2002b); Bertone et al. (2003); Bergstrom et al. (2005a); Baltz and Hooper (2004); Bergstrom et al. (2005b), neutrinos and synchrotron flux Bertone et al. (2003), or through positrons Cheng et al. (2002b); Hooper and Kribs (2004) has also been considered. The neutrino spectrum from LKP annihilation in the Sun was investigated in Hooper and Kribs (2003). Constraints on UED models from radion cosmology have also been studied Kolb et al. (2003).
The second class of Kaluza-Klein WIMPs arises in higher dimensional warped Grand Unified Theories Agashe and Servant (2004, 2005). In these models, a stable KK fermion can arise as a consequence of imposing proton stability in a way very reminiscent to R-parity stabilizing the lightest supersymmetric particle in supersymmetric models. The symmetry is called $`Z_3`$ and the Lightest $`Z_3`$ Particle (LZP) is stable since it cannot decay into standard model particles. It is actually associated with a KK Dirac right-handed neutrino with a mass in the 1 GeV to 1 TeV range. This RH neutrino has gauge interactions in particular with additional KK $`Z^{}`$ gauge bosons. Nevertheless, its interactions with ordinary matter are feeble because they involve heavy gauge bosons with a mass $`\text{ }>3`$ TeV. This opens the possibility of a weakly heavy gauge bosons with a mass $`M_{\mathrm{KK}}\text{ }>3`$ TeV. This opens the possibility of a weakly interacting Dirac RH neutrino. Indirect detection of โwarped dark matterโ in neutrino telescopes, gamma ray telescopes and cosmic positron experiments was investigated in Hooper and Servant (2005). In principle, the LZP is not necessarily the lightest KK particle. There might be lighter KK modes but which are unstable because they are not charged under $`Z_3`$. In practise though, and in the models of Agashe and Servant (2004, 2005), the RH neutrino LZP turns out to be the lightest KK particle due to various phenomenological constraints. Thus, in the following, we will use the generic appellation โLKPโ for both UED and warped types of KK dark matter.
In the present paper, we study the cosmic antiprotons that should be emitted as a result of LKP annihilations in the halo of the Milky Way. Those cosmic rays are of particular interest as the $`\overline{p}/p`$ ratio is both small โ smaller than $`10^4`$ whatever the energy โ and well-known Donato et al. (2001). The antiproton flux has been mostly measured by stratospheric balloon borne detectors โ IMAX IMAX Collaboration โ J. W. Mitchell et al. (1996), MASS MASS Collaboration โ G. Basini et al. (1999), CAPRICE CAPRICE Collaboration โ M. Boezio et al. (1997, 2001) and BESS BESS Collaboration โ S. Orito et al. (2000); BESS Collaboration โ T. Maeno et al. (2001); BESS Collaboration โ Y. Asoaka et al. (2002) โ flying at the top of the atmosphere. The interactions of high energy particles impinging on the latter generate a background to be removed in order to measure a signal that is compatible โ given the uncertainties โ with a pure secondary production arising from the spallations of cosmic ray nuclei on the interstellar gas of the Milky Way disk. The antiproton flux will be measured with unprecedented accuracy by the forthcoming space experiment AMS AMS Collaboration (2005) that has already flown on the space shuttle AMS Collaboration โ M. Aguilar et al. (2002). Small deviations from a pure secondary energy spectrum โ expected in our case if LKP particles annihilate in the galactic halo โ are potentially detectable by AMS. In section II, the source term is computed by convolving the LKP number density and cross sections with the relevant fragmentation functions. Section III is devoted to the propagation scheme and the astrophysical parameters. Finally, the primary antiproton flux resulting from LKP annihilations is compared in section IV with the secondary background and some perspectives are drawn. We will show that antiprotons can at least constrain the lowest values of LKP masses.
## II Source term
The production rate $`q_{\overline{p}}^{\mathrm{LKP}}`$ of antiprotons is obtained from the convolution over the various annihilation channels $`f`$ of the appropriate annihilation cross section $`<\sigma v>_f`$ with the fragmentation function $`\left(dN_{\overline{p}}/dT_{\overline{p}}\right)_f`$. It can be written as :
$$q_{\overline{p}}^{\mathrm{LKP}}(r,T_{\overline{p}})=\frac{1}{2}\underset{\mathrm{channel}f}{}<\sigma v>_f\left(\frac{dN_{\overline{p}}}{dT_{\overline{p}}}\right)_f\left\{n_{\mathrm{LKP}}(r)\frac{\rho _{\mathrm{LKP}}(r)}{M_{\mathrm{LKP}}}\right\}^2.$$
(1)
The LKP particles in the initial state are identical hence the overall factor of $`1/2`$. The distance between the production point and the galactic center is denoted by $`r`$ while $`n_{\mathrm{LKP}}(r)`$ and $`\rho _{\mathrm{LKP}}(r)`$ respectively stand for the LKP number and mass densities. Antiprotons are produced with a kinetic energy $`T_{\overline{p}}`$ that ranges from 0 up to the LKP mass $`M_{\mathrm{LKP}}`$. The previous relation features the four key ingredients that participate into the Kaluza-Klein antiproton souce term $`q_{\overline{p}}^{\mathrm{LKP}}`$.
To commence, for UED dark matter, the LKP annihilation cross section into fermions is given in the non-relativistic expansion limit by Servant and Tait (2002a) :
$$\sigma |\stackrel{}{v_1}\stackrel{}{v_2}|\{B^{(1)}B^{(1)}f\overline{f}\}=\frac{8\pi }{9}N_c(Y_L^4+Y_R{}_{}{}^{4})\frac{\alpha _1^2}{M_{\mathrm{LKP}}^2}(1v^2),$$
(2)
where $`N_c`$, $`Y_L`$ and $`Y_R`$ are respectively the number of colors and the left and right hypercharges of the resulting fermion $`f`$ whereas $`2\stackrel{}{v}=\stackrel{}{v_1}\stackrel{}{v_2}`$. Velocities within the Milky Way are typically non-relativistic so that the factor $`1v^2`$ can safely be approximated by 1. Notice that in contrast with neutralino dark matter, the annihilation into fermions is not helicity-suppressed. For warped dark matter, there is no simple analytical formula โ in particular, couplings depend in a non-trivial way on the LKP mass โ but we can summarize the situation as follows. For LKPs lighter than approximately 100 GeV, LKP annihilations proceed dominantly via s-channel $`Z`$-exchange. For larger masses, annihilation into top quarks via the t-channel exchange of the GUT KK gauge boson $`X_s`$ or into $`t\overline{t}`$, $`b\overline{b}`$, $`W^+W^{}`$ and $`Zh`$ via the s-channel KK $`Z^{}`$ exchange dominates. We refer the reader to the appendices of Agashe and Servant (2005) for details. In the rest of the paper, we have taken a Kaluza-Klein gauge boson mass of $`M_{\mathrm{KK}}=3`$ TeV and varied the LZP mass from 30 to 70 GeV. Median values for the annihilation cross section have been assumed here. They correspond to the couplings $`g_{10}=(g^{}+g_s)/2=0.785`$ and $`c_{\nu _L^{}}=0.4`$.
The second ingredient is the computation of the number of antiprotons with kinetic energy between $`T_{\overline{p}}`$ and $`T_{\overline{p}}+dT_{\overline{p}}`$ formed within a jet induced by a $`q\overline{q}`$ pair of energy $`M_{\mathrm{LKP}}`$. This was evaluated with the high-energy physics frequently-used Monte-Carlo event generator pythia Tjostrand (1994), based on the so-called string fragmentation model.
The square of the LKP number density $`n_{\mathrm{LKP}}`$ enters into the annihilation rate (1) and scales as $`M_{\mathrm{LKP}}^2`$. In UED models, the LKP requires a mass in the range between 700 and 900 GeV in order to generate the observed thermal relic density, unless other KK modes participate in the freeze-out process Servant and Tait (2002a). If this is the case, somewhat smaller masses are possible. A recent analysis, taking into account the effects of second level KK modes, indicates that the upper edge of this mass range is favored Kakizaki et al. (2005). In any case, we should keep in mind that the precise prediction of the LKP relic density depends on the particular KK mass spectrum which is used and is somewhat model-dependent. We will therefore consider masses in the lower range $`300`$ GeV which is the most favorable case as far as the antiproton signal is concerned. As for the warped LKP, it can thermally generate the observed quantity of dark matter in two mass ranges : near the $`Z`$-resonance with $`M_{\mathrm{LZP}}`$ 20 to 80 GeV and for considerably heavier masses โ $`M_{\mathrm{LZP}}\text{ }>`$ several hundred GeV โ Agashe and Servant (2004, 2005). Again, we will restrict ourselves to the more easily accessible lowest masses.
The distribution of dark matter inside galaxies is still an open and very debated issue. From one side, results from cosmological N-body simulations in $`\mathrm{\Lambda }`$-CDM models Navarro et al. (1996); Fukushige and Makino (1997); B. Moore et al. (1999) indicate a universal and coreless dark matter density profile. At small radii, the latter diverges with the distance $`r`$ from the galactic center as $`r^\gamma `$ with $`\gamma `$ 1 to 1.5. This implies a strongly peaked dark matter density at galactic centers. Very recent results obtained from simulations of halo formation Navarro et al. (2004); Diemand et al. (2004) strongly disfavour a singularity as steep as 1.5 and seem to point toward slopes logarithmically dependent on the distance from the galactic center and no steeper than $``$ 1.2. It should indeed be noticed that these cusps are predicted in regions which are usually smaller than the typical resolution size of the simulations. On the other side, several analysis of rotational curves observed for galaxies of different morphological types Borriello and Salucci (2001); de Blok and Bosma (2002); R. A. Swaters et al. (2003); Weldrake et al. (2003); Gentile et al. (2004); Donato et al. (2004a) put serious doubts on the existence of dark matter cusps in the central regions of the considered objects. Instead of a central singularity, these studies rather suggest a cored dark matter distribution, flattened toward the central regions. In the present analysis, we will consider the generic dark matter distribution
$$\rho _{\mathrm{CDM}}(r)=\rho _{\mathrm{CDM}}\left\{\frac{r_{}}{r}\right\}^\gamma \left\{\frac{1+\left(r_{}/a\right)^\alpha }{1+\left(r/a\right)^\alpha }\right\}^{\left(\beta \gamma \right)/\alpha },$$
(3)
where $`r_{}=8`$ kpc is the distance of the Solar System from the galactic center. The local โ Solar System โ CDM density has been set equal to $`\rho _{\mathrm{CDM}}=0.3`$ GeV cm<sup>-3</sup>. In the case of the pseudo-isothermal profile, the typical length scale $`a`$ is the radius of the central core. The profile indices $`\alpha `$, $`\beta `$ and $`\gamma `$ for the dark matter distributions which we have considered here are indicated in Tab. 1.
As already underlined in Maurin and Taillet (2003); Donato et al. (2004b) โ and as it will be clear also from the results presented in the following of the present paper โ the diffusion of primary CDM generated antiprotons is only very mildly dependent on the chosen dark matter distribution function.
## III Galactic propagation: control of uncertainties
Propagation in the Galaxy, while studied for a long time is not a simple matter. A realistic description should take into account the coupling between gas, magnetic field and cosmic rays (CRs). This is far from being reached โ at least at the Galactic scale. Our lack of knowledge about the structure of magnetic turbulences and their spatial distribution โ probably related to the regions of star formation โ hampers any clear and unambiguous description of the transport of CRs. So far, one major approximation assumed in all โ but a very few number โ of papers is that diffusion in the Galaxy does not depend on the galactic position. Even with this simplification, transport of CRs is not straightforward. It involves the now classical following ingredients : diffusion โ random walk on magnetic inhomogeneities โ and convection โ directed outward the galactic disk โ which compete for the spatial transport, especially at low energy. Regarding the energetic balance, energy losses โ Coulomb, ionization and adiabatic โ replenish the low energy tail whereas momentum diffusion โ reacceleration โ produces, on average, a gain in energy in the GeV/nucleon region. Finally, spallations may destroy CRs, preferentially at low energies.
Whatever the model retained for propagating antiprotons, it is very important to understand the origin of uncertainties in the propagated spectra. At a given energy $`E`$, spatial transport is sensitive to the following parameters โ see Fig. 2 in Maurin et al. (2003) : the diffusion coefficient normalization and slope in $`K(E)=\beta K_0^\delta `$ where the rigidity $`=p/Z`$, the halo height $`L`$, the wind velocity $`V_c`$ perpendicular to the disk โ choosen to be constant in our model โ and the Alfvรฉnic speed $`V_a`$ of the scatterers โ the uncertainty due to this latter parameter is less significant compared to the previous processes. Only special combinations of these parameters can account for the measured B/C ratio. The abundance of boron relative to carbon โ two typical elements which are respectively from secondary and primary origin โ is a very good tracer of the history of CRs propagation. In a previous study, a degeneracy between these combinations was found Maurin et al. (2001), leading to a wide uncertainty in the underlying parameters of the model, although they gave the same B/C ratio. In a second study, the same combinations were used to compute the secondary antiproton signal in the same model Donato et al. (2001). The induced uncertainty for this secondary flux in the region of interest โ a few hundreds of MeV โ was found to be small โ about 10%. This was expected, as all these species follow the same propagation paths, being emitted and detected in the disk.
The situation is quite different for primary exotic species, as most diffusion paths start in the diffusive halo Taillet and Maurin (2003); Maurin and Taillet (2003). The previous degeneracy is broken. The induced uncertainties on primary antiprotons are studied in details in Donato et al. (2004b) and are found to be as large as a factor $``$ 100 for supersymmetric particles. We briefly recall here, on a physical basis, the dependance of the uncertainty on each parameter, as it also applies to the present study. First, the halo height $`L`$ determines i) the total number of sources inside the diffusive region and ii) the effective radial range of diffusion, i.e. the distance that a CR can travel from a source before escaping from the Galaxy. Cosmic rays coming from farther than $`L`$ have an exponentially low probability to be detected on Earth. Notice that this second point explains why the evaluated fluxes are not very sensitive to the shape of the dark matter halo in the galactic center region โ see below in section IV and in Donato et al. (2004b). Second, the galactic wind wipes the particles away from the disk. It is well known that the effect of $`V_c`$ is similar to that of $`L`$ when sources are located in the disk โ see Jones (1978) and $`L^{}K(E)/L`$ โ but this is not true for sources in the halo. The two effects i) and ii) actually turn out to be of greater magnitude for $`L^{}`$ than for $`L`$.
It should be kept in mind that the parameters $`L`$, $`V_c`$ and $`K_0`$ are correlated. In the subset of parameters giving the observed B/C ratio, low values of $`K_0`$ generally correspond to low $`L`$ and large $`V_c`$ and thus low $`L^{}`$, so that the signal is expected to decrease with decreasing $`K_0`$. Notice that the effect of $`K_0`$ is not only through the correlation to $`L`$ and $`V_c`$ : the reader is referred to Donato et al. (2004b) โ see in particular its section III โ for an explicit analysis of all the effects.
A conservative estimate โ based on the full range of B/C allowed propagation parameters โ leads to variations of about two orders of magnitude of the primary antiproton flux. Notice that this does not include the nuclear and particle physics uncertainties. This range could be narrowed by using constraints coming from other species of cosmic rays. Actually, using radioactive Donato et al. (2002) or heavy Combet et al. (2005) species only yield a minor improvement. They enable to shrink the parameter space but leave unchanged the values leading to the extremal fluxes. The final extreme and median parameters which we have considered in this analysis are borrowed from Donato et al. (2004b). They are displayed in Tab. 2.
## IV Flux and conclusions
The case of UED models is featured in Fig. 1 to 3 where the LKP mass has been set equal to 300 GeV and to 1 TeV. The interstellar antiproton yields are plotted as a function of interstellar kinetic energy $`T_{\overline{p}}^{\mathrm{IS}}`$ for three different halo profiles. The canonical isothermal, NFW and Moore models respectively correspond to the solid blue, dashed red and dot-dashed magenta curves. The solid black line is the conventional secondary component. We have somewhat improved the previous estimate Donato et al. (2001) by taking adiabatic losses into account. The maximum, median and minimum diffusion configurations respectively correspond to Fig. 1, 2 and 3. A few remarks are in order.
To commence, because the square of the LKP mass enters into the denominator of the $`B^{(1)}`$ annihilation cross section โ see relation (2) โ the antiproton source term $`q_{\overline{p}}^{\mathrm{LKP}}`$ varies globally like $`M_{\mathrm{LKP}}^4`$. As a consequence, when the LKP mass is increased from 300 GeV to 1 TeV, the antiproton fluxes drop by a factor of $`(10/3)^4120`$. This downward shift of the curves by two orders of magnitude is clearly present in the figures.
Then, as already discussed in section III, the particular choice for the galactic cosmic ray diffusion parameters strongly affects the primary yields whereas the secondary component varies very little. From the maximal to the minimal configurations โ see Tab. 2 โ primary antiproton fluxes decrease by two orders of magnitude. That sensitivity combined with fairly similar shapes for the primary and secondary energy spectra do not strengthen the case of the antiproton signal as a clear signature for UED dark matter. Even in the most favorable case of a 300 GeV $`B^{(1)}`$ boson and for maximal galactic diffusion โ see Fig. 1 โ the secondary background overcomes the primary LKP signal up to an antiproton kinetic energy of 100 GeV. Notice however that above that energy and in the case of a Moore halo profile, the signal may eventually become comparable to the background, leading to an excess of antiprotons at high energy that has already been noticed by Bringmann (2005). Unfortunately, that distinctive spectral feature vanishes as soon as other configurations for the galactic cosmic ray propagation are selected. In Fig. 3, the antiproton yield is $``$ 30 times smaller than the secondary flux for an antiproton kinetic energy of 100 GeV.
Finally, the cusp at the Milky Way center does not affect much the primary antiproton signal. Varying the DM halo profile from a mild canonical isothermal distribution to the extreme case of a Moore divergence results in an increase of the primary yields by at most a factor $``$ 2 to 3 in the case of maximal galactic diffusion. That increase is much less significant for the median diffusion case and has disappeared in Fig. 3. As is clear in Tab. 2, the minimal diffusion configuration corresponds to a thickness of the confinement layers of only 1 kpc associated with a strong galactic convection wind that wipes away any particle originating from the galactic central cusp.
Notice that our galactic diffusion code relies on the expansion of the radial dependence of the cosmic ray abundances as a series of the Bessel functions $`J_0(\alpha _ir/R_{\mathrm{gal}})`$ where $`\alpha _i`$ is the i-th zero of the function $`J_0`$ and where $`R_{\mathrm{gal}}`$ is the radius of the propagation region. Because taking properly into account a central divergence like $`r^{2\gamma }`$ with $`\gamma =1`$ โ NFW โ or 1.3 โ Moore โ would necessitate an infinite number of such functions in the above expansion and would lead to numerical instability, we have renormalized the DM distribution in the vicinity of the Milky Way center without modifying the absolute number of its annihilations. More precisely, the actual DM density within a sphere of radius $`r_c`$ is given by
$$\frac{\rho (r)}{\rho _c}=\left\{\frac{r_c}{r}\right\}^\gamma ,$$
(4)
where $`\rho _c\rho (r_c)`$. The central cusp boosts the LKP annihilations by a factor of
$$\eta =\frac{3}{3\mathrm{\hspace{0.17em}2}\gamma }$$
(5)
with respect to the case of a uniform distribution with constant density $`\rho _c`$. We have replaced the divergent distribution (4) by the milder profile
$$\left\{\frac{\rho (rr_c)}{\rho _c}\right\}^2=\mathrm{\hspace{0.33em}1}+\left\{\frac{2\pi ^2}{3}\left(\eta 1\right)\mathrm{sin}_c^2\left(\frac{\pi r}{r_c}\right)\right\},$$
(6)
where $`\mathrm{sin}_c(x)\mathrm{sin}(x)/x`$. That renormalized density leads to the same number of LKP annihilations as the actual cusp. We have set $`r_c=500`$ pc and our primary flux calculations converge with only $`N_{\mathrm{Bes}}=300`$ terms in the Bessel expansion. A smaller value of $`r_c`$ would require a larger $`N_{\mathrm{Bes}}`$ and is not actually necessary insofar as the antiproton Green function that connects the solar system to the galactic central region varies smoothly over the latter Taillet and Maurin (2003); Maurin and Taillet (2003). In the minimal case for cosmic ray propagation โ presented in Fig. 3 โ it even vanishes.
Because of the uncertainties in the cosmic ray galactic propagation, we conclude that the antiproton signal is not the best tool to observe UED Kaluza-Klein species in the halo of the Milky Way. Direct detection is not very promising either since observable rates at current instruments are typically less than one event per year Servant and Tait (2002b). On the contrary, since a pair of $`B^{(1)}`$ bosons may annihilate directly into light fermions, the positron signal should exhibit a characteristic spectral spike spreading toward low energies as a result of positron energy losses during propagation Cheng et al. (2002b). Notice however that the positron annihilation signal arising in the case of an isothermal halo needs to be amplified by a factor of $``$ 60 Pochon (2005) before being detectable by AMS-02 AMS Collaboration (2005). An enhancement by a factor of $``$ 200 with respect to a pure NFW cusp is also necessary to reproduce โ albeit below 0.8 TeV Bergstrom et al. (2005a) โ the flat gamma ray spectrum which the HESS collaboration has detected at the center of the Milky Way HESS Collaboration โ F. Aharonian et al. (2004). The positron and gamma ray spectra are harder for Kaluza-Klein species than for neutralinos. Those signals have therefore been advocated as promising signatures. A word of caution is in order at that stage. The positron signature requires to be significantly enhanced in order to be detectable. If we now assume a boost factor of $``$ a hundred as suggested by recent numerical simulations Diemand et al. (2005) that point toward the presence of numerous mini-clumps in the DM galactic halo, primary antiprotons should be copiously produced since even in the most pessimistic diffusion scheme of Fig. 3, the signal exceeds the background above $`T_{\overline{p}}^{\mathrm{IS}}40`$ GeV in the case of a 300 GeV $`B^{(1)}`$ boson.
For the warped models of Agashe and Servant (2004, 2005), the LZP may be much lighter than the UED dark matter candidate. The primary antiproton flux โ at the top of the atmosphere โ is plotted in Fig. 4 as a function of antiproton kinetic energy $`T_{\overline{p}}^{\mathrm{TOA}}`$ for five different values of the LZP mass. The most optimistic galactic diffusion scheme as well as a canonical isothermal DM halo have been assumed for the primary signal. Observations from various experiments performed during solar minimum BESS Collaboration โ S. Orito et al. (2000); BESS Collaboration โ T. Maeno et al. (2001); AMS Collaboration โ M. Aguilar et al. (2002); CAPRICE Collaboration โ M. Boezio et al. (2001) are well explained by the narrow band within which the background of secondary antiprotons lies irrespective of the galactic propagation conditions. The curves corresponding to $`M_{\mathrm{LZP}}=40`$ โ short dashed magenta โ and 50 GeV โ long dashed red โ exceed the background and should have already led to a detection would our assumptions on galactic diffusion and halo profile be correct. For $`M_{\mathrm{LZP}}=M_{\mathrm{Z}^0}/2`$, the LZP annihilation is actually driven by the $`Z`$-resonance and is significantly enhanced.
As in the previous discussion of the UED models, the LZP antiproton signal sensitively depends on galactic cosmic ray propagation. In Fig. 5, the primary yields of a 40 and 50 GeV LZP decrease by two orders of magnitude between the most optimistic and the most pessimistic diffusion cases of Tab. 2. In the latter configuration, the antiproton signal is now well below the background. The halo profile is also a source of uncertainty as is clear in Fig. 6 where a 60 GeV LZP is exhibited. The maximal galactic diffusion that has been assumed in that case makes it possible for antiprotons from the central cusp to reach the solar circle and to lift the degeneracy among the various DM distributions. Should the minimal diffusion scheme be preferred, the three colored curves would be one and the same.
As featured in Fig. 4 to 6, the LZP antiproton signal is in the vicinity of the secondary background and therefore in the ballpark for detection. That is why we have explored the effect of a clumpy DM distribution by taking into account an overall boost factor in our estimates of the primary yields which we have compared to observations. We have actually performed a $`\chi ^2`$ test to assess the compatibility between our theoretical predictions for both secondary and primary components and the experimental data. All the available measurements have been used BESS Collaboration โ S. Orito et al. (2000); BESS Collaboration โ T. Maeno et al. (2001); BESS Collaboration โ Y. Asoaka et al. (2002); AMS Collaboration โ M. Aguilar et al. (2002); CAPRICE Collaboration โ M. Boezio et al. (2001, 1997); IMAX Collaboration โ J. W. Mitchell et al. (1996); MASS Collaboration โ G. Basini et al. (1999) except the Buffington point which is known for being one order of magnitude above all the others. Those experiments are either balloon borne โ IMAX, MASS, CAPRICE and BESS โ or space borne as AMS. In addition to statistical effects, they suffer from uncertainties associated with instrumental misreconstructions โ e.g. from electrons โ and from the atmosphere component contamination which has to be removed โ unless only antiprotons above the geomagnetic cutoff are taken into account. Within the error bars, the IS fluxes inferred from those experiments are now in reasonable agreement. A dramatic improvement is expected in the forthcoming years with AMS-02 which will be implemented on the International Space Station for 3 years starting in 2008 : both statistic and systematic errors are expected to be reduced by several orders of magnitude for antiprotons above 0.5 GeV. To take into account the differences in solar activity between those observations, the modulation has been applied in the force field scheme with three different field values $`\varphi `$ : 500 MV, 700 MV and 1000 MV, depending on the periods. The errors used are the experimental statistical uncertainties. Fig. 7 features the $`\chi ^2`$ per degree of freedom as a function of the boost factor for $`M_{\mathrm{LZP}}=30`$ Gev and a Kaluza-Klein scale $`M_{\mathrm{KK}}`$ of 3 TeV. The red, green and black curves respectively correspond to the maximum, median and minimum cosmic ray diffusion configurations. It should be emphasized that the value of that $`\chi ^2`$ must be taken with care as it indicates that the uncertainties have been underestimated, making any quantitative statistical conclusion impossible to reach. To give a crude estimate of the rejection power of this study, we have decided that models leading to a $`\chi ^2/\mathrm{d}.\mathrm{o}.\mathrm{f}.`$ larger than twice its minimum value โ for the considered parameters โ are excluded. In the case of a real $`\chi ^2`$ distribution, it would correspond roughly to the 99.9 % confidence level โ we use $`50`$ degrees of freedom. In Tab. 3, the LZP mass has been varied from 30 to 80 GeV whereas the Kaluza-Klein scale $`M_{\mathrm{KK}}`$ has been set equal to 3 and 6 TeV. The DM annihilation boost factor above which the primary antiproton signal is too strong to be compatible with the observations is displayed for each configuration. If a boost factor of $``$ one hundred is assumed, all the configurations with $`M_{\mathrm{KK}}`$ = 3 TeV are excluded whereas the LZP antiproton signal is potentially detectable for larger Kaluza-Klein scales.
Notice finally that direct detection experiments already exclude almost entirely a Kaluza-Klein scale $`M_{\mathrm{KK}}`$ of 3 TeV whereas larger values are allowed Agashe and Servant (2004). The LZP may also directly annihilate into light fermions and can produce the same kind of distortion in the positron spectrum as UED dark matter species. The HEAT excess HEAT Collaboration โ S. W. Barwick et al. (1997); S. Coutu et al. (1999) is actually well reproduced by a 40 or 50 GeV LZP if the boost factor is respectively set equal to $``$ 40 and 30 Hooper and Servant (2005). Antiproton calculations suffer from large uncertainties as regards the galactic cosmic ray propagation. The difficulty to reach a conclusion as regards the detectability of a primary antiproton signal has been illustrated in this article. We would like to stress that the same kind of ambiguities should also affect secondary and primary positrons with a magnitude that is yet to be determined.
Acknowledgements : P.S. would like to thank the french programme national de cosmologie PNC and the groupement de recherche on phรฉnomรจnes cosmiques de haute รฉnergie PCHE for their financial support.
|
warning/0506/astro-ph0506338.html
|
ar5iv
|
text
|
# Mira Variables in the OGLE Bulge fields Table 1 is available in electronic form at the CDS via anonymous ftp to cdsarc.u-strasbg.fr (130.79.128.5) or via http://cdsweb.u-strasbg.fr/cgi-bin/qcat?J/A+A/. Figure 1 and the Appendices are available in the on-line edition of A&A.
## 1 Introduction
In the course of the micro lensing surveys in the 1990โs the monitoring of the Small and Large Magellanic Clouds has revealed an amazing number and variety of variable stars. A big impact was felt and is being felt in any area of variable star research, like Cepheids and RR Lyrae stars. Also in the area of variability in red variables (RVs) and AGB stars there has been remarkable progress. Wood et al. (1999) and Wood (2000) were the first to identify and label different sequences โABCโ thought to represent the classical Mira sequence (โCโ) and overtone pulsators (โA,Bโ), and sequence โDโ which is not yet satisfactorily explained (Olivier & Wood 2003, Wood et al. 2004. Stars on these sequence are referred to as having Long Secondary PeriodsโLSPs). This view has subsequently been confirmed and expanded upon by Noda et al. (2002), Lebzelter et al. (2002), Cioni et al. (2003), Ita et al. (2004) and Kiss & Bedding (2003, 2004), Fraser et al. (2005). These works differ in the source of the variability data (macho, ogle, eros, moa), area (SMC or LMC), associated infrared data (Siding Spring 2.3m, denis, 2mass, sirius), and selection on pulsation amplitude or infrared colours. In a recent paper, Groenewegen (2004; hereafter G04) analysed the ogle data in the SMC and LMC, and correlated these sources with the denis and 2mass surveys. The paper discussed the variability properties of three samples: about 2300 spectroscopically confirmed AGB stars, around 400 previously known LPV variables and about 570 candidate dust-obscured AGB stars.
The present paper is an extension of the analysis in G04 to the ogle data in the direction of the Galactic Bulge (GB). Also for this area of the sky, several papers exist that use the results of the micro lensing surveys and have extended previous classical works on Bulge variable stars, like those of Lloyd Evans (1976), Glass & Feast (1982), Whitelock et al. (1991), Glass et al. (1995; hereafter GWCF), Alard et al. (1996) and Glass et al. (2001).
Alard et al. (2001; herafter ABC01) correlated ISOGAL sources within the NGC 6522 and Sgr I Baade windows with the MACHO database and present a list of 332 stars with complete 4-band $`V,R`$ and , magnitudes. Schultheis & Glass (2001) extended Alard et al. by also considering the denis and 2mass data in those fields in general, and for the variables in particular. Glass & Schultheis (2002; hereafter GS02) investigated a sample of 174 M-giants in the NGC 6522 Baade window and correlated them with denis ISOGAL and MACHO. Many stars of spectral type M5 and all M6 and later show variation, whereas subtypes M1-M4 do not (see also Glass et al. 1999).
Glass & Schultheis (2003; hereafter GS03) investigated the variable stars in the NGC 6522 Baadeโs window using MACHO data, and also used denis IR data. Of the 1661 selected stars 1085 were found to be variable. They present $`K`$-band $`PL`$-relations for sequences โABCDโ.
Wray et al. (2004) investigated small amplitude red giants variables in a sub-set of 33 ogle fields. They identified two groups that seem to correspond to groups โA-โ and โB-โ in Ita et al. (2004; also see G04).
In our paper we describe results on Mira variables selected from ogle Bulge fields. The paper is structured as follows. In Section 2 the ogle, 2mass and denis surveys are described. In Section 3 the model for the lightcurve analysis is briefly presented. In the remaining of the paper different results are described. The Period-Luminosity diagram is discussed in Section 5. A description of the Mira population in respect to the overall bulge population is given in Section 8. In Section 9 we show that the Miras are distributed in a bar-like structure and give the orientation. In the final section we give the distance to the Galactic Centre, based on the period-luminosity relation.
## 2 The data sets
The ogle-ii micro lensing experiment observed fourty-nine fields in the direction of the GB. Each field has a size 14.2โฒ$`\times `$57โฒ and was observed in $`BVI`$, with an absolute photometric accuracy of 0.01-0.02 mag (Udalski et al. 2002). Table 4 lists the galactic coordinates of the field centers and the total number of sources detected in these fields.
Wozniak et al. (2002) present a catalog of about 222 000 variable objects based on the ogle observations covering 1997-1999, applying the difference image analysis (DIA) technique on the $`I`$-band data. The data files containing the $`I`$-band data of the candidate variable stars was downloaded from the ogle homepage (http://sirius.astrouw.edu.pl/ogle/). According to Wozniak et al., the level of contamination by spurious detections is about 10%, but we presume this level is much less at the brighter magnitudes of the LPVs considered here. Table 4 lists the number of detected variable stars per field (Wozniak et al. 2002).
The denis survey is a survey of the southern hemisphere in $`IJK_\mathrm{s}`$ (Epchtein et al. 1999). The second data release available through ViZier was used (The DENIS consortium, 2003). The 221801 ogle objects were correlated on position using a 3โณ search radius and 59894 matches were found.
The 2mass survey is an all-sky survey in the $`JHK_\mathrm{s}`$ near-infrared bands. On March 25, 2003 the 2mass team released the all-sky point source catalog (Cutri et al. 2003). The easiest way to check if a star is included in the 2mass database is by uplinking a source table with coordinates to the 2mass homepage. Such a table was prepared for the 221801 ogle objects and correlated on position using a 3โณ radius. Data on 182361 objects were returned.
## 3 Lightcurve analysis
The model to analyse the lightcurves is described in detail in Appendices A-C in G04.
Briefly, a first code (see for details Appendix A in G04) sequentially reads in the $`I`$-band data for the 222 000 objects, determines periods through Fourier analysis, fits sine and cosine functions to the light curve through linear least-squares fitting and makes the final correlation with the pre-prepared denis and 2mass source lists. All the relevant output quantities are written to file.
This file is read in by the second code (see for details Appendix B in G04). A further selection may be applied (typically on period, amplitude and mean $`I`$-magnitude), multiple entries are filtered out (i.e. objects that appear in different ogle fields), and a correlation is made with pre-prepared lists of known non-LPVs and known LPVs or AGB stars. The output of the second code is a list with LPV candidates.
The third step (for details see Appendix C in G04) consists of a visual inspection of the fits to the light curves of the candidate LPVs and a literature study through a correlation with the simbad database. Non-LPVs are removed, and sometimes the fitting is redone. The final list of LPV candidates is compiled.
Details on the small changes in the codes w.r.t. the implementation in G04 are given in Appendix A of the present paper.
## 4 Comparison of the datasets
### 4.1 Astrometry
As in G04, the correlation between the ogle objects and known LPVs and AGB stars, and known non-LPVs, is actually done in 2 steps. In the first step the correlation is made (for a 3โณ search radius), and the differences and spread in $`\mathrm{\Delta }`$RA $`\mathrm{cos}(\delta )`$ and $`\mathrm{\Delta }\delta `$ are determined. These mean offsets are then applied in most cases to make the final cross-correlation, and this usually increases the number of matches. The results are listed in Table 3.
### 4.2 Photometry
As in G04, a comparison was made between the (mean) ogle $`I`$ and the (single-epoch) denis $`I`$, and between the (single-epoch) denis $`JK`$ and the (single-epoch) 2mass $`JK`$ magnitudes. This was done by selecting those objects with an amplitude in the $`I`$-band of $`<0.05`$ mag.
Figure 2 shows the final results when offsets $`I`$(denis-ogle) = $`0.03`$, $`J`$(denis-2mass)= $`0.02`$, and $`K`$(denis-2mass)= $`0.03`$ are applied. The latter values are consistent with those derived in OOS03 based on the 2mass second incremental data release who found $`J`$(denis-2mass)= $`0.02\pm 0.09`$, and $`K`$(denis-2mass)= $`0.00\pm 0.07`$.
## 5 Period-Luminosity relations
The full machinery outlined in Section 3 was performed. As in G04, all derived periods are given in Table 1 and are shown in Figure 1. The present paper discusses only objects which have at least one period with an $`I`$-band amplitude larger than 0.45 magnitudes<sup>1</sup><sup>1</sup>1The amplitude, $`A`$, is used in the mathematical sense in the present paper, $`y=A\mathrm{sin}x`$. The peak-to-peak amplitude is 0.90 mag., i.e. classical Mira variables (e.g. Hughes 1989). After visual inspection of the lightcurves a sample of 2691 such objects remain. The number of objects per field is listed in the last column of Tab. 4.
Table 1 lists the stars in the sample, the fitted periods with errors and amplitudes, and the denis and 2mass photometry of the associated sources. Table 2 lists alternative names and references from the literature. Figure 1 presents lightcurves and their fits.
In the discussion that follows, magnitudes are de-reddened using the $`A_\mathrm{V}`$ values that correspond to the respective ogle field taken from Sumi (2004; and $`A_\mathrm{V}`$ = 6.0 for the field 44 that they do not discuss), and selective reddenings of $`A_\mathrm{I}/A_\mathrm{V}=0.49`$, $`A_\mathrm{J}/A_\mathrm{V}=0.27`$, $`A_\mathrm{H}/A_\mathrm{V}=0.20`$, $`A_\mathrm{K}/A_\mathrm{V}=0.12`$ (Draine 2003) and implicitly assuming that all objects suffer this reddening value (i.e. ignoring differential reddening within a field, and ignoring that foreground and background objects would suffer a different reddening). Sumiโs method is based on the absolute magnitude of the Red Clump giants and is absolute calibrated using the $`(VK)`$ colours of 20 RR Lyrae stars in Baadeโs window. Popowski et al. (2003) present an extinction map (over 9000 resolution elements of 4x4 arc minute size) towards the GB based on MACHO $`V,R`$ photometry, under the assumption that colour-magnitude diagrams would look similar in the absence of extinction. For the centre of the ogle fields it was checked if there was a tile in the Popowski et al. set within 0.05 degrees distance. For those, the value of the visual extinction has been listed, next to the value in Sumi in Table 4. The rms difference $`A_\mathrm{V}`$ for the 21 field with values from both references is 0.18. Finally, Schultheis et al. (1999b) presented a reddening map for the inner GB comparing denis $`J,K`$ photometry to isochrones. Table 4 lists for two ogle fields the values they find: SC44 which was not considered by Sumi (2004), and SC5 for which Sumi derive a larger $`A_\mathrm{V}`$ than Schultheis et al.: 5.73 versus 4.13.
In the further discussion we only use periods that fulfil the following conditions are used in the calculations (with $`\mathrm{\Delta }P`$ the error in the period): $`\mathrm{\Delta }P/P<0.01`$ for $`P<500^d`$; $`\mathrm{\Delta }P<5^d`$ for $`500^d<P<800^d`$ and $`\mathrm{\Delta }P<1.5^d`$ for $`P>800^d`$. The latter constraint is necessary because the long periods become comparable to the length of the dataset.
Figure 3 shows the $`K`$-band $`PL`$-relation for all periods which have an $`I`$-band amplitude larger than 0.45 magnitude and $`(JK)_0<2.0`$ among the 2691 stars. The cut in $`(JK)`$ colour is needed to prevent that the $`K`$-magnitude is affected by circumstellar extinction, as shown in G04. Like G04, the $`K`$-magnitude is on the 2mass system, and is the average of the denis and 2mass photometry. In particular, if both denis and 2mass $`K`$-band data is available, the denis data point is corrected as explained above (i.e. 0.03 mag added), and averaged with the 2mass data point. This should take out some of the scatter in the $`PL`$-diagram, as the effect of the variability in the $`K`$-band is reduced. If only denis is available, the corrected value is used. In the left-hand panel the boundaries of the boxes โA-, A+, B-, B+, C, Dโ have been taken from G04, but shifted by $`4.0`$ to account for the approximate difference in distance modulus (DM), as, e.g., follows from the recent determination of 7.9 $`\pm `$ 0.4 kpc (Eisenhauer et al. 2003) for the distance to the GC, and 18.50 for the DM to the LMC (e.g. recent reviews by Walker 2003, Feast 2004a).
There is a reasonably well defined sequence in Box โCโ, but when compared to the similar figure for the SMC and LMC in G04 (his figure 3) some differences can be remarked as well. In particular, for the present Bulge sample there are a few objects located in Box โB+โ, and in particular many in Box โDโ. In the SMC and LMC, for this cut in amplitude, there are none in Box โB+โ and few in โDโ. Several issues may play a role. Applying a certain cut in amplitude may sample slightly different variables in SMC, LMC and Bulge. Figure 3 in G04 clearly shows how lowering the cut in amplitude results in a populating Box โB+โ and then โA+โ, and increases the number of objects in โDโ. Another effect is the possible contribution of objects in the foreground and background of the Bulge, the depth of the Bulge, and fourthly, the orientation of the Bar, as the ogle fields span 20 degrees in longitude (this last effect will be discussed later). Finally, the difference in DM may be different from the adopted value of 4.0.
To verify if the objects in Box โDโ actually show LSP, they were all visually inspected, and in fact few have, in agreement with the finding for LMC and SMC (for amplitudes $`>`$0.45 mag). This would call for a enlargement of Box โCโ to properly sample the $`PL`$-relation of the large amplitude (Mira) variables. To define this enlarged box the $`PL`$-relation was inspected for each field independently. The right-hand panel in Figure 3 shows the finally adopted boundaries of Box โCโ, which implies that Box โDโ has contracted. Stars inside this redefined Box will be used to define the $`PL`$-relation. The $`K`$-band $`PL`$-relation is determined to be:
$$m_\mathrm{K}=(3.37\pm 0.09)\mathrm{log}P+(15.44\pm 0.21)$$
(1)
with an rms of 0.42 and based on 1292 stars, and is shown in Fig. 3. The value of the slope is consistent with the median value when the $`PL`$-relation is determined for all fields individually.
For reference, fitting all stars in Figure 3, for a fixed slope of $`3.37`$ results in a ZP of 15.47 $`\pm `$ 0.55.
## 6 Historical versus current periods
Table 5 compares the period derived in the present paper (the one with the largest amplitude) with values derived in the literature. There are three cases where a previously determined period may be a harmonic of the present period but overall there is good agreement between periods. In the 12 cases where there is a period available from LE76 (with the photographic plates taken between 1969 and 1971, hence 28 years of time difference with ogle) there is no clear case for a star that changed period. By comparison, G04 found that about 8% of LMC variables changed their period by more than 10% over a about 17 year timespan. To find 0 out of 12 in the present sample is consistent with this.
## 7 Colour-colour diagrams
The 2mass and denis Colour-colour and colour-magnitude diagrams are shown in Fig. 4, together with that of spectroscopically confirmed M-stars in the LMC (see also Figure 12 in G04). There appear to be more redder stars in the Bulge sample, but this is likely due to a under representation in the LMC sample as this was restricted to spectroscopically known M-stars (i.e. in general optically selected). The sample of candidate infrared-selected AGB stars in the LMC \[Figure 12 in G04\] does cover the $`(IK)`$ and $`(JK)`$ colour range observed in the Bulge).
The other main difference is that the Bulge stars are redder by $`2`$ mag in $`(IJ)_0`$ compared to both LMC and SMC stars, as was also shown by Lebzelter et al. (2002) in a comparison of LMC and Bulge variable stars. As the diagrams involving $`J,H,K`$ colours appear similar, it seems that this difference in $`(IJ)`$ must be due to a difference in $`I`$. The $`I`$-band measurements of M stars is strongly affected by the TiO and VO molecular absorption features (Lanรงon & Wood 2000). It is expected that for larger metallicities these lines will be stronger (Schultheis et al. 1999a) which will lead to redder $`(IJ)`$ colours.
The bullets connected by a line in the Bulge denis $`(IJ)(JK)`$ colour-colour diagram are the average colours of M1, M2, .., M6, M6.5, M7, M8 giants in the NGC 6522 Baadeโs window (Blanco 1986, GS02). There is a spread of typically 0.3-0.5 mag in $`(IJ)`$ and 0.2-0.3 mag in $`(JK)`$ around these means, and there is only 1 M8 giant in their sample. The colours of the Miras follow those of normal giants well until M6.5, when the Miras become redder in $`(IJ)`$.
There are also stars redder in $`(JK)`$ than the single M8 star in the sample of GS02, indicating either the presence of later spectral types or the on-set of circumstellar reddening.
The conclusion of GS02 that โMany M5 and all stars M6 and later show variation, whereas subtypes (M1-M4) do notโ is confirmed here, as there are essentially no objects located in the region of the denis $`(IJ)(JK)`$ colour-colour diagram occupied by spectral types of M4 and earlier.
## 8 Mira bulge population as function of latitude
Figure 5 shows the period distribution of Miras in Box โCโ. A distinction is made between all Miras and those with $`(JK)_0<2.0`$ (dashed histograms). The latter selection minimises any influence of circumstellar extinction. For comparison, the period distribution of LMC and SMC Miras is also shown<sup>2</sup><sup>2</sup>2Derived following the implementation of the code and definition of the โboxesโ as in G04, and applying the same selection criteria as in the present paper, i.e. $`I`$-amplitude larger than 0.45 mag. . The Kolmogorov-Smirnov (KS) test is performed to indicate that the probability that the period distributions of Bulge-LMC, Bulge-SMC, LMC-SMC are the same for all stars (those with $`(JK)_0<2`$, respectively) is, respectively 0.36 (10<sup>-8</sup>), 0.05 (0.31) and 0.05 (0.05).
Any difference, in particular between Bulge and LMC period distribution, is difficult to quantify further as this depends in a complicated way on the Star Formation History and evolutionary tracks ($`T_{\mathrm{eff}}`$ \- Luminosity - Mass - metallicity). Regarding the period distribution of Bulge Miras as such, previous studies are limited to selected small fields (e.g. TLE, GWCF, Glass et al. 2001). Whitelock et al. (1991) present the period distribution of about 140 IRAS sources but no direct comparison is possible because of the difference in the selection criteria of the samples.
Figure 6 shows the period distribution of selected fields with very similar longitudes that cover a range in latitudes (the stars with $`(JK)_0<2.0`$ are shown as dashed histogram again). To add a field even closer to the GC than surveyed by ogle the data in Glass et al. (2001, 2002) is considered on a field centered on $`l=0.05,b=0.05`$. They present the results of a $`K`$-band survey of 24 $`\times `$ 24 arcmin<sup>2</sup> for LPVs down to $`K12.0`$. From the list of 409 stars, 14 were removed because of double entries, quality index of zero, uncertain or no listed period or amplitude. The coordinates were uploaded to the IPAC webserver and 2mass data within 2.5โณ was retrieved for these 395 stars to get information on the $`(JK)`$ colour. As an additional check, and to eliminate multiple stars within the search circle, it was verified that the single-epoch 2mass $`K`$ magnitude is consistent with the mean $`K`$-magnitude and amplitude listed in Glass et al. For 345 stars 2mass data is available. The magnitudes are corrected for interstellar reddening using the extinction map of the inner GB at 2โฒ resolution by Schultheis et al. (1999b). The extinction value of the nearest available grid point in this map is taken. The extinction values range between 18.5 and 30.4 with a mean of 24.7. The top panel in Figure 6 list 333 stars with $`K>7`$ (to eliminate 3 very likely foreground objects), and $`K`$-band amplitude larger than 0.35 (to correspond roughly to the cut in $`I`$-band amplitude of 0.45 mag), and 88 (the histogram with slanted hatching), or 236 (dashed histogram) stars which also have $`(JK)_0<2.0`$. The latter sample is the one that results when the reddening values from Schultheis et al. are multiplied by 1.35. They mention themselves that the reddening may be underestimated in the direction of the GC because of $`J`$-band non-detections. For the one field in common, their value is a factor 1.3-1.4 smaller than derived by Sumi (2004). In addition, for their default reddening (the histogram with slanted hatching in Figure 6) there would be many stars even at periods shorter than about 250 days which still would have $`(JK)_0>2.0`$ which is not observed in the other fields. This could off course be real, but it is generally believed (e.g. Launhardt et al. 2002) that the population of low- and intermediate mass stars in the Nuclear Bulge (the inner about 30 pc from the GC) and GB are similar, but that in the former there is an overabundance of $`10^710^8`$ year old stars. In this picture one would expect the period distributions to be similar at shorter periods, essentially independent of latitude. Therefore the period distribution of stars with $`(JK)_0<2.0`$ for the increased reddening is adopted.
The KS test is performed on consecutive fields in latitude for the distributions based on the stars with $`(JK)_0<2.0`$. It is found that the probability that the distributions are the same is $`10^{10}`$ for the $`b=0.05/`$ $`1.21`$ fields, 0.50 for the $`b=1.21/`$ $`1.39`$ fields, 0.80 for the $`b=1.39/`$ $`1.81`$, and $`>0.99`$ for the fields at more negative latitudes. The conclusion is that the period distributions of the fields at and below $`1.2`$ degree are statistically indistinguisable, but that the field at $`0.05`$ latitude has a significantly different period distribution (the probability that this distribution is the same as the distribution of the combined 6 ogle fields is $`10^{22}`$). This conclusion is independent of the assumed reddening of the inner Bulge field, which influences how many stars will have $`(JK)_0<2.0`$. For the default reddening of Schultheis et al., the probability that the distributions are the same for the $`b=0.05/`$ $`1.21`$ fields is still only 0.0033. The difference in the period distributions is especially clear at longer periods. Of the 236 stars in the inner field with $`(JK)_0<2.0`$ 61 have $`P>500`$ days, while in the other fields this is 3 out of 367.
The difference in period distribution might be due to an under representation of short period stars in the inner field. However, Figure 4 in Glass et al. illustrates that the expected $`K`$-magnitudes at short periods are not fainter than the completeness limit of their survey. In fact, Glass et al. mention that they expect that the number of short-period Miras ($`P<250`$ days) is at least 75% complete. As a test, one-third of stars with $`P<250`$ days were randomly duplicated and added to the sample, and the KS test repeated to find again a large difference between the period distribution of the field at $`0.05`$ degrees and the other fields.
The difference is emphasised in Figure 7 where the scaled period distribution of stars which have $`(JK)_0<2.0`$ in the 5 fields between $`b=1.39\mathrm{ยฐ}`$ and $`b=5.8\mathrm{ยฐ}`$ has been subtracted from the inner field. The scaling was done in such a way that at shorter periods the two distributions would cancel at a level of 1$`\sigma `$ (based on Poisson errors). Even if the scaling is done in a slightly different way the result is always very similar, in the sense that there is a significant ($`>4\sigma `$) overabundance of LPVs in the inner field bewteen about 350 and 600 days.
The conclusion is that there is a significant population of LPVs with period $`\stackrel{>}{}500`$ days present in the inner field, which remains barely present at latitude $`1.2\mathrm{ยฐ}`$, and is absent for $`b\stackrel{<}{}1.4\mathrm{ยฐ}`$. This was indirectly noted by Glass et al. who noticed that the average period of the stars in this field at $`b=0.05\mathrm{ยฐ}`$ is 427$`d`$ (and that of the known OH/IR stars 524$`d`$), while the average period in the Sgr i window ($`b=2.6\mathrm{ยฐ}`$) is 333$`d`$, with no known OH/IR stars (GWCF).
To quantify the nature of the Mira Bulge population, synthetic AGB evolutionary models have been calculated, which are described in detail in Appendix C.
In brief, the synthetic AGB code of Wagenhuber & Groenewegen (1998) was finetuned to reproduce the models of Vassiliadis & Wood (1993) and then extended to more initial masses and including mass loss on the RGB. For several initial masses the fundamental mode period distribution was calculated for stars inside the observed instability strip and when the mass loss was below a critical value to simulate the fact that they should be optically visible. Vassiliadis & Wood (1993) provide calculations for 4 different metal abundances: $`Z=0.016,0.008,0.004`$ and 0.001. We used the models for $`Z=0.016`$, representing a solar mix, which are most appropriate for our Bulge sample (e.g. Rich 1998). We also show that our results are essentially unchanged if $`Z`$ = 0.01 or 0.02 are adopted. From the comparison of the observed period distribution for fields more than $`1.2\mathrm{ยฐ}`$ away from the galactic centre with the theoretical ones, we deduce that the periods can be explained with a population of stars with Main Sequence masses in the range of 1.5 to 2.0 M. A possible extension to smaller masses is possible, but not necessary to explain the periods below 200 days. To explain the excess periods in the range of 350-600 days observed closer to the centre we need initial masses in the range 2.5 - 3 M. The presence of more massive stars in the inner field at $`b`$ = $`0.05\mathrm{ยฐ}`$ cannot be excluded, as it turns out that for more massive stars the optically visible Mira phase is essentially absent. Sevenster (1999) analysed OH/IR stars (which are LPVs with longer periods and higher mass loss rates than the Miras) in the inner Galaxy and came to the conclusion that OH/IR stars in the bulge have a minimum intial mass of about 1.3 M, based on an analysis of infrared colours, compatible with our results. We briefly mention here the result from Olivier et al. (2001) who studied a sample of LPVs in the solar neighbourhood with periods in the 300 to 800 days range. They conclude that majority of these stars had initial masses in the range of 1 - 2 M, with an average value of 1.3 M, lower than what we find for the 300 to 600 days range sample. This difference may be explained by the fact that our conclusions are only valid for a sample with no or only low mass loss rates ($`\stackrel{<}{}5\times 10^6\text{M}\text{}/yr`$ ), contrary to their sample which was selected to contain stars with significant mass loss ($`10^5\text{M}\text{}/yr`$). As can be seen in the Vassiliadis & Wood (1993) models, the period increases considerably when the stellar mass is reduced by the mass loss process.
We do not see a variation in the period distributions for the higher latitude fields (beyond $`1\mathrm{ยฐ}`$ latitude) and can consider this as a homogeneous โbulgeโ population, which according to the Vassiliadis & Wood (1993) model has ages in the range of 1 to about 3 Gyr. The excess population closer to the Galactic Centre is younger than 1 Gyr. According to Launhardt et al. (2002), the Nuclear Bulge (approximately the central degree) contains besides the bulge population seen at higher latitudes also an additional population due to recent star formation closer to the galactic centre. Blommaert et al. (1998) find that the extrapolation of the number density of bulge OH/IR stars towards the galactic centre would explain half of the galactic centre OH/IR population, but that an additional population, intrinsic to the galactic centre, exists which agrees with what we see in the distribution here.
The formation history of the Bulge is still a matter of debate. In several works like in Kuijken & Rich (2002) and recently in Zoccali et al. (2004), the bulge is considered to be old ($`>10`$ Gyr) and formed on a relatively short timescale ($`<1`$ Gyr) (e.g. Ferreras et al. 2003). On basis of the modelling of colour-magnitude diagrams, Zoccali et al. claim that no trace is found of any younger stellar population than 10 Gyr. The bulge Miras do not fit in this picture as, according to our analysis, they are considerably younger. The field studied by Zoccali et al. is centered at ($`l,b`$) = (0.277, $`6.167`$) so at a slighter higher latitude than our extreme fields ($`b5.8\mathrm{ยฐ}`$). Although we are limited by the small number of Miras detected at the highest latitude fields, we do not see a change in period distribution for those fields. Zoccali et al. (2004) acknowledge the presence of Miras, but consider them as part of the old population. It is true that Miras are also detected in globular clusters and thus can be associated with old ages, as is the case for a 1 M star in the Vassiliadis & Wood (1993) model but these stars produce periods shorter than 200 days (Figure C.1), insufficient to explain the period distribution seen in the bulge. The periods of Miras in Globular Clusters range from 150 to 300 days (Frogel & Whitelock 1998) and so only overlap with the shorter periods of the bulge Miras.
Our results agree more with the analysis of the infrared ISOGAL survey discussed in van Loon et al. (2003). They conclude that the bulk of the bulge population is old (more than 7 Gyr) but that a fraction of the stars is of intermediate age (1 to several Gyr). The Miras in our study can thus be considered as the intermediate age population seen in their analysis. van Loon et al. also see evidence for an even younger population ($`<200`$ Myr), but according to our findings, this would be restricted to the area close to the Galactic Centre.
Our discussion on the ages of the Mira stars is based on the assumption that they have evolved from single stars. An alternative scenario as suggested by Renzini & Greggio (1990), would be that the brighter (longer period) Miras could evolve from close binaries where the components coalesced to form one single star. This could lead to an underestimation of the age as the Mira essentially is the product of lower mass and thus older stars. This scenario may seem in better agreement with the idea that the bulge consists of an old stellar population. It however suffers from the same problem as the intermediate age population, in the sense that also no clear evidence for Blue Stragglers (which would be the Main Sequence counterpart of the Miras) is found (Kuijken & Rich, 2002).
If indeed the bulk of the bulge population is old and formed quickly and if the Miras are of intermediate age, then our Miras must be representatives of a population which was added at a later stage and it is unclear how it relates to the overall bulge. An interesting scenario suggested in Kormendy & Kennicutt (2004) is the one in which a secondary bulge or also called pseudo-bulge forms within an old bulge. Such a process would be connected to the presence of a โbarโ which would add โdiskyโ material into the old classical bulge. The Miras are indeed situated in a bar-structure as is discussed in the following section.
## 9 The orientation of the Bar
Table 6 lists the zero points (ZPs) of the the $`K`$-band $`PL`$-relation (for a fixed slope of $`3.37`$) for the ogle fields individually. To increase the statistics, some neighbouring fields have been added together, as indicated in the first column of the table. The galactic coordinates listed are the mean values of all individual objects, rather than the mean of the field centres. Figure 9 plots these ZPs (with error bars) as a function of Galactic longitude. There is a clear correlation; the formal weighted fit has a slope of $`0.023\pm 0.005`$ (magnitude/degree). Restricting the fields to those with longitudes $`5<l<+5`$ (reducing the contamination by disk stars, see Appendix B) the fit becomes:
$$m_\mathrm{K}=(0.0192\pm 0.0087)l+(15.484\pm 0.019)$$
(2)
with an rms of 0.10 and based on 32 fields.
The interpretation of this correlation is that the Bulge Miras are located in the Galactic Bar that has a certain orientation towards the observer. A similar correlation was found by Wray (2004) who concluded that an appropriately chosen ZP in $`I`$ for the small amplitude ogle variables in their sample (which they identify as to correspond to in Box โA-โ) correlated with Galactic longitude. No estimate for the orientation of the Bar was given however.
In Appendix B Monte Carlo simulations are carried out in order to quantify two issues: can these observations be used to constrain the orientation of the Galactic Bar, and, second, given the specific location of the ogle fields, if there is any bias in the derived zero point compared to a fiducial ZP when all Miras would be located exactly in the Galactic Centre (GC). As described in Appendix B, for a spatial distribution of Bulge and Disk stars following Binney et al. (1997), viewing angles $`\varphi `$ of 43 and 79 degrees (see the orientation in Figure 8) result in slopes (magnitude versus $`l`$, Eq. 2) in agreement with observations. However, the model with $`\varphi =43\mathrm{ยฐ}`$ gives a much better fit to the number of stars per field. The bias in the ZPs is essentially independent of viewing angles, and for the best fitting model the observed ZP derived for all stars (Eq. 1) is too bright by 0.018 mag ($`\pm `$ 0.013), while the ZP in Eq. 2 is too bright by 0.002 ($`\pm `$ 0.021) mag.
The preferred value of $`\varphi =43\mathrm{ยฐ}`$ is in agreement with the values of about $`45\mathrm{ยฐ}`$ by Whitelock (1992), based on 104 IRAS detected Mira variables, and the preferred value of $`46\mathrm{ยฐ}`$ by Sevenster et al. (1999), based on an analysis of OH/IR stars in the inner Galaxy.
Other values in the literature are usually much larger, between 60 and 80 degrees: Dwek et al. (1995) and Binney et al. (1997), based on COBE-DIRBE data, Stanek et al. (1997), based on bulge red clump stars, Robin et al. (2003) and Picaud & Robin, based on colour-magnitude fitting. Sevenster et al. (1999), however, argues that these values are commonly found when no velocity data is available, the longitude range is too narrow or when low latitudes are excluded. It is also possible that these studies are tracing other populations, which may be differently distributed than the Miras. Whitelock et al. and Sevenster et al. do use populations closely related to the Mira stars and find an angle of the bar close to the one we derive.
## 10 The distance to the Galactic Centre
A slope of the $`K`$-band $`PL`$-relation of $`3.37\pm 0.09`$ is derived. There appear not to exist many previous determination of this quantity. Recently, GS03 derived a $`PL`$-relation in NGC 6522 based on 34 MACHO variables with $`r`$-amplitude $`>`$1.5 and denis $`K`$ photometry: $`m_\mathrm{K}=4.6\mathrm{log}P+18.1`$. No errors or rms were given, asโby their own accountโthis fit was made by eye. Much better agreement is found with GWCF. Based on multi-epoch data of 55 stars they found $`m_\mathrm{K}=(3.47\pm 0.35)\mathrm{log}P+(15.64\pm 0.86)`$ (rms=0.35) in the Sgr i field.
Zero points for the $`K`$-band $`PL`$-relation have been derived in two ways. First, a direct fit to all stars resulting in (15.44 $`\pm `$ 0.21), and secondly, determining ZPs per (sub)-field, and fitting this as a function of $`l`$, resulting in (15.484 $`\pm `$ 0.019).
Applying the small bias corrections discussed at the end of Sect. 9 and averaging over the two estimates, the adopted $`K`$-band $`PL`$-relation for Miras at the GC is:
$$m_\mathrm{K}=3.37\mathrm{log}P+(15.47\pm 0.03)$$
(3)
The derived $`PL`$-relation can be compared to the one derived for 83 O-rich LPVs in the LMC derived in G04: $`m_\mathrm{K}=(3.52\pm 0.16)\mathrm{log}P+(19.56\pm 0.38)`$, with an rms of 0.26. Since the slopes are not exactly the same, the magnitudes are compared at the approximate mean period of $`\mathrm{log}P=2.45`$. The difference in magnitude is 3.72. Adopting the LMC based slope of $`3.52`$ for the GB Miras, and re-fitting the ZP, the bias corrected ZP would become 15.85, resulting in a GB-LMC DM difference of 3.71, essentially the same value. If the distance to the GC is assumed to be 7.94 kpc (Eisenhauer et al. 2003; in a recent preprint this was even lowered to 7.62 $`\pm `$ 0.32 kpc, Eisenhauer et al. (2005)), then the LMC would be at a DM = 18.21, or if the DM to the LMC is assumed to be 18.50 (Walker 2003, Feast 2004a), then the GC would be at 9.0 kpc. A similar result was found by GWCF who derived a distance to the GC of 8.9 $`\pm `$ 0.7 kpc, assuming 18.55 for the LMC DM and $`\varphi =45\mathrm{ยฐ}`$. The analysis so far has assumed no metallicity dependence of the Mira $`PL`$-relation. Wood (1990) present linear non-adiabatic pulsation calculations that suggest a dependence of the form $`\mathrm{log}P0.46\mathrm{log}Z+1.59\mathrm{log}L`$, but he notes that in the $`K`$-band the dependence is expected to be weaker and following the example he presents one infers a dependence of $`0.25\mathrm{log}Z`$ in the $`K`$-band. In G04 $`K`$-band $`PL`$-relations were derived for carbon-miras in the SMC and LMC. At a characteristic period of $`\mathrm{log}P=2.45`$ one infers a relative difference in DM of 0.38, which is smaller than the commonly quoted value of near 0.50 (0.48-0.53 $`\pm `$ 0.11, FO cepheids \[Bono et al. 2002\], 0.46-0.51 $`\pm `$ 0.15, FU cepheids \[Groenewegen 2000\], 0.44 $`\pm `$ 0.05, TRGB \[Cioni et al. 2000\]). This may hint at a metallicity dependence of the Mira $`K`$-band $`PL`$-relation. To test this hypothesis, a correction to the $`K`$-magnitude of $`+\beta \mathrm{log}Z`$ will be assumed<sup>3</sup><sup>3</sup>3that is, $`M_\mathrm{K}=M_\mathrm{K}(\mathrm{ref})+\beta \mathrm{log}(Z/Z_{\mathrm{ref}})`$, where $`M_\mathrm{K}(\mathrm{ref})`$ is the known magnitude is a galaxy with metallicity $`Z_{\mathrm{ref}}`$, and $`M_\mathrm{K}`$ the magnitude it would have in a galaxy with metallicity $`Z`$ (for both O- and C-rich LPVs), and the Bulge, LMC, and SMC will be assumed to have solar, solar/2 and solar/4 metallicity, respectively. For a value $`\beta =0.25`$ the relative SMC-LMC DM based on the C-Miras is increased from 0.38 to 0.46, while the relative DM LMC-GC is increased from 3.72 to 3.80. If the relative SMC-LMC DM is fixed at 0.50, then $`\beta =0.40`$ is required, and the relative DM LMC-GC becomes 3.84 for that value. For a LMC DM of 18.50, the distance to the GC then becomes 8.6 kpc. The error in this value is somewhat difficult to estimate as the $`PL`$-relations derived in G04 and here are fromโat bestโthe average of two $`K`$ values. Work by Feast et al. (1989) indicates that in the case of multi-epoch data (and for the small depth effect in the LMC) the intrinsic dispersion in the $`PL`$-relation is about 0.13 mag. Therefore we assign an error of 0.18 to the difference in DM. This implies an error of 0.7 kpc. Based on this large sample of Mira variables in the direction of the GB the conclusion is that the distance to the GC is between 8.6 and 9.0 ($`\pm 0.7`$) kpc, depending on the metallicity dependence of the $`K`$-band $`PL`$-relation. Feast (2004b) discusses the zeropoint of the Mira $`K`$-band $`PL`$-relation, and adopting the slope observed in the LMC ($`3.47`$) derives a zeropoint of 1.00 $`\pm `$ 0.08, averaging over independently derived ZPs from trigonometric parallaxes, OH VLBI expansion parallaxes and Galactic Globular Clusters. Adopting a slope of $`3.47`$ and refitting the ZP of the Bulge sample, the bias corrected value becomes 15.73 $`\pm `$ 0.03, and without metallicity correction (consistent with the assumption above about the metallicities in Bulge, LMC, SMC) leads to a distance to the GC of 8.8 $`\pm `$ 0.4 kpc. This independent distance estimate is in between the values derived using no or a strong metallicity dependent zero point.
###### Acknowledgements.
This research has made use of the SIMBAD database, operated at CDS, Strasbourg, France. This publication makes use of data products from the Two Micron All Sky Survey, which is a joint project of the University of Massachusetts and the Infrared Processing and Analysis Center/California Institute of Technology, funded by the National Aeronautics and Space Administration and the National Science Foundation.
## Appendix A The light curve analysis model
Some small changes w.r.t. the implementation of the lightcurve analysise model in G04 are described.
The only change in the first part of the code is the level at which a period is accepted as significant. This level was set at significance = $`1.0\times 10^{16}`$, compared to $`5.5\times 10^{11}`$ in G04. This was possible asโcontrary to G04โonly objects with large amplitudes were searched for. The resulting increase in the number of spurious periods was then caught in the process of visual inspection.
As in G04 a list of known objects (both known non-LPVs, and known AGB giants and LPVs) was compiled to ease automatic association. The list comprises:
(1) 14833 IRAS sources within 10 deg. radius of the centre of the 49 ogle fields at RA = 268.87, Dec = $`31.03`$<sup>4</sup><sup>4</sup>4Retrieved from the infrared science archive at IPAC, http://irsa.ipac.caltech.edu/,
(2) 51141 ISOGAL sources from Omont et al. (2003, hereafter OGA03; those within the extreme values of the ogle field boundaries 261.612 $`<=`$ RA $`<=`$ 276.133, $`40.726<=`$ Dec $`<=21.328`$),
(3) 268 pulsating variable stars, 1650 Eclipsing Binaries, and 943 Miscellaneous variable stars from ogle-i (Udalski et al. 1994, 1995a, 1995b, 1996, 1997),
(4) 332 objects from Alard et al. (2001, hereafter ABC01) who correlated ISOGAL sources within the NGC 6522 and Sgr I Baade windows with the MACHO database,
(5) 2353 objects from Ojha et al. (2003, hereafter OOS03) who studied sources in 9 ISOGAL fields,
(6) 174 M-giants later than spectral type M0 in the NGC 6522 Baadeโs window from Glass & Schultheis (2002, hereafter GS02),
(7) 421 objects from Glass et al. (2001, and erratum), who monitored in $`K`$ over four years an 24x24 arcmin area near the Galactic Centre,
(8) 122 objects from Alard et al. (1996, identified as Ter-\[number\]) who identified LPV using red photographic plates,
(9) 494 objects from Lloyd-Evans (1976, hereafter Lloyd-Evans (1976) who identified Mira variables in the three Baade windows, identified as TLE-\[field\]-\[number\] in Table 2), Blanco et al. (1984; herafter BMB, who identified M giants in Baadeโs window), and Blanco (1984; herafter B84, who identified RR Lyrae variables) with coordinates listed in the simbad database to 1โณ or better,
(10) 33 Nova related objects from Cieslinski (2003). The total number of sources used in the automatic correlation is 72764.
## Appendix B Simulations of the Galactic Bulge and foreground disk stars
In this Appendix the calculations are described to model a population in the direction of the Galactic Bulge.
The basic model is essentially the one proposed by Binney et al. (1997) to model the dust-corrected near-infrared COBE/DIRBE surface brightness map of the inner galaxy. The number density of bulge stars is assumed to be:
$$f_\mathrm{b}=f_0\mathrm{exp}(a^2/a_\mathrm{m}^2)/(1+a/a_0)^\beta $$
(4)
with $`f_0=624`$, $`a_\mathrm{m}=1.9`$ kpc, $`a_0=0.10`$ kpc, $`\beta =1.8`$. Binney et al. assumed a tri-axial bulge with axial ratios 1 : 0.6 : 0.4. For numerical convenience a prolate ellipsoid is assumed here: $`a=\sqrt{x^2+(y/\eta )^2+(z/\eta )^2}`$ with the value of $`\eta =0.5`$ taken from Binney et al.
The number of Bulge objects up to a radius $`r`$ from the centre, that defines the probability density function in the simulation, is approximated as:
$$N_\mathrm{b}(r)=_0^r4\pi a^2f_\mathrm{b}(a)๐a$$
(5)
up to a maximum radius that is taken to be the co-rotation radius, with a default value of $`R_{\mathrm{cr}}=2.4`$ kpc, following Dwek et al. (1995).
The number density of disk stars is assumed to be:
$$f_\mathrm{d}=(\mathrm{exp}(z/z_0)+\alpha \mathrm{exp}(z/z_1))\times $$
$$R_\mathrm{d}(\mathrm{exp}(r/R_\mathrm{d})f_\mathrm{h}\mathrm{exp}(r/R_\mathrm{h}))$$
(6)
with $`z_0=0.210`$ kpc, $`z_1=0.042`$ kpc, $`\alpha =0.27`$, $`R_\mathrm{d}=2.5`$ kpc (Binney et al.) and $`R_\mathrm{h}=1.3`$ kpc (Picaud & Robin 2004). This functional form follows Binney et al., but also allows for a โholeโ in inner disk (the scaling parameter is 0 $`f_\mathrm{h}1`$, and identical to zero in Binney et al.). The total number of disk stars, and the probability density functions, are defined as:
$$N_\mathrm{d}N_{\mathrm{d},\mathrm{z}}(z)\times N_{\mathrm{d},\mathrm{r}}(r)$$
given by,
$$N_\mathrm{d}=[2_0^z\mathrm{exp}(z/z_0)+\alpha \mathrm{exp}(z/z_1)dz]\times $$
$$\left[_0^r2\pi rR_\mathrm{d}(\mathrm{exp}(r/R_\mathrm{d})f_\mathrm{h}\mathrm{exp}(r/R_\mathrm{h}))๐r\right]$$
(7)
up to maximum values $`z_{\mathrm{max}}=(R_{\mathrm{cr}}\eta )`$, and $`R_{\mathrm{max}}=8.0`$ kpc, respectively.
A disk or bulge star is generated according to the ratio $`N_\mathrm{b}/(N_\mathrm{b}+N_{\mathrm{d},\mathrm{z}}\times N_{\mathrm{d},\mathrm{r}})`$. In case of a disk star, its height above the plane, $`z`$, distance to the GC, $`r`$, and a random angle between 0 and $`2\pi `$ in the Galactic plane, are drawn according to the probability functions $`N_{\mathrm{d},\mathrm{z}}(z)`$ and $`N_{\mathrm{d},\mathrm{r}}(r)`$. Its coordinates $`x,y,z`$ are then known.
In case of a bulge star, the distance, $`a`$, to the GC is drawn according to the probability function $`N_\mathrm{b}`$, and then a star is randomly placed on the surface of the appropriate ellipsoid, to find $`x,y,z`$. These values are then rotated by an angle $`\varphi `$ in the Galactic plane (see Fig. 8).
The Galactic coordinates are then derived assuming a distance from the Sun to the GC of $`R_0=8.5`$ kpc, and height above the plane of $`z_0=+24`$ pc (Maรญz-Apellรกniz 2001)
In a second step, for every star, the known distance to the Sun is used to calculate its appararent magnitude, assuming an arbitrary $`M_\mathrm{K}`$ of $`7.5`$ mag with a Gaussian dispersion of $`\sigma _\mathrm{K}=0.15`$ mag. This is about the dispersion observed in the $`PL`$-relation in LMC Miras when multi-epoch photometry is available to accurately determine mean-light magnitudes (Feast et al. 1989).
In a third step, for every simulated star it is verified if it is located within one of the 40 lines-of-sight considered, listed in Table 6. The field sizes of 14.2โฒ$`\times `$57โฒ are approximated by a circle of radius 0.27 degrees. If so, it is assumed the star would have been โdetectedโ (Given the relative brightness of the LPVs it is assumed that completeness is not an issue).
The number of stars drawn is such that a total of about 1200 objects is โdetectedโ, similar to the actual number. At the end of the simulation, the average magnitude and dispersion per line-of-sight is determined, and a weighted least-square fit is made of the mean magnitude versus longitude, for all fields, and for those with $`l<5\mathrm{ยฐ}`$, as for the observations. In addition, the mean magnitude and dispersion for all โdetectedโ stars is determined.
Such a simulation is repeated 1000 times. Then, distribution functions and from that median and 1-sigma values of the following parameters are determined: the number of stars (total, disk, bulge), mean magnitude and sigma (for every line-of-sight), mean magnitude and sigma for all stars, and slope and error in the slope both when fitted over all longitudes, and for those fields with $`l<5\mathrm{ยฐ}`$.
For the standard model of Binney et al. described above (i.e. $`f_\mathrm{h}=0`$) it turns out that for two values of $`\varphi `$ a slope (because of the contamination by disk stars in the outer fields, the slope fitted over $`l<5\mathrm{ยฐ}`$ is considered for now on) in agreement with observations is found: $`\varphi =43`$ and $`79`$ degrees (with values between 25 and 85 degrees resulting in predicted slopes within 1-$`\sigma `$ of the observed one.)
Figures 10, 11 and 12 show the results for these two cases. Figure 10 shows the distribution on the Galactic Plane for a random sub-sample of all stars simulated, and illustrates a fundamental difference between the two cases. For large viewing angles the outer fields $`l\stackrel{>}{}10\mathrm{ยฐ}`$ will be dominated by disk stars. Figure 11 shows for the same random sub-sample the observed magnitude as a function of $`l`$. In Figure 12 the simulated mean magnitude and error for each field are compared to the observations in the top panel, while in the bottom panel the observed and predicted number of objects are compared. It is from this plot that one may conclude that the model with $`\varphi =43\mathrm{ยฐ}`$ is to be preferred over the one with $`\varphi =79\mathrm{ยฐ}`$ as the latter model predicts too few stars, especially in the outer fields. Comparing only the observed and predicted number of stars (in a $`\chi ^2`$ sense) a best fit is found for $`\varphi =35\mathrm{ยฐ}`$ (with values between 0 and 60 degrees resulting in a reduced $`\chi ^2`$ within 1 unit of the minimum $`\chi ^2`$). Combining the constraints from the slope and the number of stars a viewing angle of $`\varphi =43\pm 17`$ degrees is the preferred value.
One may consider the ratio of Bulge to Disk stars as uncertain, and therefore, a model was considered with $`\varphi =79\mathrm{ยฐ}`$ and $`f_0=350`$. The latter value was set so that the model predicted the observed number of stars in the $`l=11\mathrm{ยฐ}`$ field. Such a model would still underestimate the number of stars in the outer fields at positive $`l`$, and would give a slope no longer in agreement with observations.
Finally, a model including a hole in the inner disk is considered (i.e. $`f_\mathrm{h}=1`$). To have the same ratio of bulge to disk stars, $`f_0`$ was set to 425. The results are very similar and the best fitting angle is now 40 degrees.
For reference, the predicted number of disk and bulge stars for the viewing angles of 43 and 79 degrees, and for the model with $`\varphi =40\mathrm{ยฐ}`$ and the central hole in the disk, are listed in Tab. 7. As they are quite different, these predictions may be usefull when additional data (proper motion<sup>5</sup><sup>5</sup>5Sumi et al. (2004) present proper motions for 5 million objects in the ogle fields, centered on the expected $`I_0`$ magnitude and $`(VI)_0`$ colour of red clump giants but also including some red giants. Of the 2691 LPVs in the present study, 1612 are listed in Sumi et al., radial velocities) become available to constrain the ratio of disk to bulge objects as a function of galactic coordinates.
## Appendix C Comparing stellar evolution codes
From P. Woodโs webpage (http://www.mso.anu.edu.au/$``$wood/) the models described in Vassiliadis & Wood (1993, VW) were downloaded. These files list for the individual calculated models on the AGB the relevant stellar quantities (remaining stellar mass, luminosity and effective temperature amongst other quantities) and the evolutionary time. They are available for $`Z`$ = 0.016 (1.0, 1.5, 2.0, 2.5, 3.5, 5.0 M), $`Z`$ = 0.008 (0.945, 1.0, 1.5, 2.0, 2.5, 3.5, 5.0 M), $`Z`$ = 0.004 (0.89, 1.0, 1.5, 2.0, 2.5, 3.5, 5.0 M) and $`Z`$ = 0.001 (1.0, 1.5 M).
For our comparison with simulation we used the solar metallicity models, which are expected to be the most representative for our Bulge sample. However, different studies indicate that the Bulge may have quite a broad metallicity distribution, peaking at about $`0.25`$ dex with dispersion of 0.3 dex (see e.g. McWilliam & Rich 1994, Ramirez et al. 2000). The AGB lifetime, LPV lifetime and LPV period distribution was determined. The AGB lifetime is defined as the time between the first model in the file (the start of the AGB) upto the point where the remaining envelope mass becomes less than 0.04 M, or $`T_{\mathrm{eff}}>4500K`$, that is taken as the start of the post-AGB evolution.
For each timestep the fundamental period is calculated following VW. The star is assumed to be in the Mira instability strip when the bolometric magnitude is within 0.20 magnitude (the assumed width of the instability strip at a given period) of the $`PL`$-relation (Feast et al. 1989):
$$M_{\mathrm{bol}}=3.00\mathrm{log}P+2.85$$
(8)
assuming a LMC distance of 18.50, and when the mass loss rate is below a critical value, as this $`PL`$-relation is derived for optically visible objects and the LPV samples studied in G04 and in the present paper have been culled by only considering objects with $`(JK)_0<2.0`$. In such a way the lifetime and the period distribution of optically visible LPVs can be determined.
The critical mass loss rate is determined by taking for each mass in the grid of solar metallicities typical values of luminosity and effective temperature inside the instability strip and then using a radiative transfer program (Groenewegen 1993) with the appropriate model atmosphere for M-stars (Fluks et al. 1994), and typical dust properties (silicate dust, condensation temperature of 1500 K, dust-to-gas ratio 0.005, expansion velocity 10 km s<sup>-1</sup>) to determine the critical mass loss rate at which the star would become redder in $`(JK)`$ than 2.0. The critical mass loss rates found are between 4 and 20 $`\times 10^6`$ M yr<sup>-1</sup> depending on the initial mass of the model. In fact, the critical mass loss rate is observed to scale with $`\sqrt{L}`$, as expected as the dust optical depth predominantly determines the infrared colours, hence the mass loss rate is proportional to the stellar radius, all other things being equal. For the 1.0 and 1.5M initial mass models the mass loss rate inside the instability strip always remains below the critical mass loss rate. The adopted critical mass loss rate is 1.0 $`\times 10^5\sqrt{L/13000}`$ M yr<sup>-1</sup> . The results for the solar metallicty models are summarised in Table 8
We will now consider the synthetic AGB models of Wagenhuber & Groenewegen (1998; WG). The reason being that the VW models exist only for a limited number of initial masses and secondly because mass loss on the RGB was only included for initial masses below 1M, while it is well known that the effect of mass loss on the RGB is substantial also at and above 1 M. The VW mass loss rate recipe was implemented, and to mimic the VW tracks as closely as possible the mixing length parameter $`\alpha `$ (basically setting the effective temperature scale) in the WG models was tuned to give similar AGB and LPV lifetimes. The results are summarised in Table 8. It turns out that with $`\alpha =2.0`$ the lifetimes match very well especially at low initial mass.
Mass loss on the RGB is described by a Reimers law with a scaling factor $`\eta _{\mathrm{RGB}}=0.35`$. This gives the required mass loss (0.13, 0.16, 0.17 M for a star of 1.0M initial mass at $`Z=`$ 0.004, 0.008, 0.019, respectively; M. Salaris, private communication) to give the observed mean colour of Horizontal Branch stars in Galactic Globular Clusters. Table 9 summarises for a set of WG models with $`\eta _{\mathrm{RGB}}=0.35`$ and $`\alpha =2.0`$ the AGB and LPV lifetime, and Figure 13 displays the period distribution of optically visible stars inside the instability strip (normalised to one each time) for a few initial masses.
Table 9 also includes for a few initial masses the results if slightly different metallicities of $`Z`$= 0.01 and 0.02 are adopted, and Figure 14 shows the corresponding Period distribution. These results indicate that the effect of metallicity on the pulsation properties for the typical metallicities in the Bulge is small.
|
warning/0506/quant-ph0506218.html
|
ar5iv
|
text
|
# Influence of intrinsic decoherence on nonclassical properties of the output of a Bose-Einstein condensate
## I. INTRODUCTION
Recently, the development of dilute gas Bose-Einstein condensation (BEC) has opened up the study of atom laser \[4-7\], which is the matter wave analogs of optical lasers. The ideal atom laser beam is a single frequency de Broglie wave with well-defined intensity and phase. The first realization of a pulsed atom laser was achieved with sodium atoms at MIT by coupling a BEC from a magnetically trapped state to an untrapped state using a rf pulse . Then, it was repeated with long rf pulses and Raman transitions .
In recent years, much attention has been focused on the problem of nonlinear atomic optics. The four matter-wave mixing was realized in the remarkable experiment by making use of the optical technique of Bragg diffraction to the condensate. The possibility of optical control on the quantum statistics of the output matter wave was pointed out in the framework of nonlinear atomic optical optics . Several dynamical analysis concerning the nonclassical properties of the output of the trapped condensed atoms have been presented in Refs..
On the other hand, there has been increased interest in the problem about decoherence of BEC and atom laser \[16-19\]. One fact causing decoherence comes from the nonlinear interaction between atoms. For a single-mode condensate the significant effect of atomic collisions is to cause fluctuations in the energy and hence fluctuations in the frequency, thus causing increased phase uncertainty, eventually lead to decoherence . In Ref., the influence of the decoherence on quantum coherent atomic tunneling between two condensates is studied. It is shown that the decoherence leads to the decay of the population difference and the suppression of the coherent atomic tunneling. In this paper, we would like to investigate the influence of decoherence on the nonclassical properties of the output of a Bose-Einstein condensate by adopting Milburnโs model of intrinsic decoherence . The influence of decoherence on the nonclassical properties, such as sub-poisson distribution and quadrature squeezing of the atom laser beam is investigated. It is shown that, under very special conditions, the atom laser beam may exhibit stationary quadrature squeezing in such a decoherence model.
The paper is organized as follows: In Sec.II we briefly outline the simple model describing the output coupling of the trapped dilute condensed atoms, in which the nonlinear interaction between the atoms and the quantized motion of atom center of mass in the inhomogeneous magnetic field has been ignored. Then, the quantum dynamical behavior of this system in the intrinsic decoherence model is discussed by making use of Bogoliubov approximation. In Sec.III, the influence of decoherence on the nonclassical properties, such as sub-poisson distribution, quadrature squeezing effect is investigated. In Sec.IV, there are some discussions.
## II. THE MODEL
In this section, we consider the output coupling of the trapped dilute condensed atoms. For simplicity, the atoms are assumed to have two states, $`|T`$ and $`|F`$, with the initial condensation occurring in the trapped state $`|T`$. State $`|F`$, which is typically unconfined by the magnetic trap, is coupled to $`|T`$ by a one-mode squeezed optical field tuned near resonance with the $`|T|F`$ transition. The Hamiltonian of this system is given as ($`\mathrm{}=1`$)
$$H=\omega _0b^{}b+\omega _aa^{}a+\mathrm{\Omega }(ab^{}c+a^{}bc^{}),$$
$`(1)`$
where $`b`$($`b^{}`$) and $`c`$($`c^{}`$) are the annihilation (creation) operators of bosonic atoms for the untrapped state $`|F`$ and the trapped state $`|T`$ with transition frequency $`\omega _0`$, respectively. $`a`$($`a^{}`$) are the annihilation (creation) operators of the optical field with frequency $`\omega _a`$. Here $`\mathrm{\Omega }=\sqrt{\omega _a/2\epsilon _0V}`$ is the coupling constant, and $`V`$ is the effective mode volume and $`\epsilon _0`$ the vacuum permittivity. In the system (1), the atom-atom coupling and the nonlinear interaction between the atoms and the quantization motion of atomic center of mass in the trapped state by an inhomogeneous magnetic field has been ignored. It was shown that the above system leads to an oscillation behavior of the quantum statistics between the optical field and the output atomic laser beam .
In what follows, we outline the basic content of the Milburn model of decoherence. Based on an assumption that on sufficiently short time steps the quantum system does not evolves continuously under unitary evolution but rather in a stochastic sequence of identical unitary transformations, Milburn has derived the equation for the time evolution density operator $`\rho (t)`$ of the quantum system ,
$$\frac{d\rho (t)}{dt}=\gamma [\mathrm{exp}(\frac{i}{\gamma }H)\rho (t)\mathrm{exp}(\frac{i}{\gamma }H)\rho (t)],$$
$`(2)`$
where $`\gamma `$ is the mean frequency of the unitary time step. In the limit $`\gamma \mathrm{}`$, Eq.(2) reduces to the ordinary von Neuman equation for the density operator. It is easy to obtain the formal solution of Eq.(2) as follows,
$$\rho (t)=\underset{k=0}{\overset{\mathrm{}}{}}A_k(t)\rho (0)A_k^{}(t),$$
$`(3)`$
where the Kraus operator $`A_k(t)`$ is given by
$$A_k(t)=\frac{(\gamma t)^{k/2}}{\sqrt{k!}}e^{\gamma t/2}\mathrm{exp}(i\frac{kH}{\gamma }).$$
$`(4)`$
Obviously, $`_{k=0}^{\mathrm{}}A_k^{}(t)A_k(t)=I`$.
We assume that the initial state of system (1) is described by $`\rho (0)=|\mathrm{\Psi }(0)\mathrm{\Psi }(0)|`$. Here, $`|\mathrm{\Psi }(0)=|\alpha _T|\mathrm{\Phi }(0)_s`$ with $`|\alpha _T`$ the Glauber coherent state of the operator $`c`$ characterizing the condensed atoms in the trapped state $`|T`$; $`|\mathrm{\Phi }(0)_s=|0_F|\xi `$, $`|0_F`$ represents that there is initially no occupying atoms in the untrapped state $`|F`$, and the optical field is in the squeezed vacuum state $`|\xi =S(\xi )|0`$. $`S(\xi )=\mathrm{exp}(\xi a^2\xi ^{}a^2)`$ is the squeezed operator.
Substituting the $`\rho (0)`$ into Eq.(3), we obtain
$$\rho (t)=e^{\gamma t}\mathrm{exp}(\alpha c^{}\alpha ^{}c)\underset{k=0}{\overset{\mathrm{}}{}}\frac{(\gamma t)^k}{k!}M^k|0_{TT}0||\mathrm{\Phi }(0)_{ss}\mathrm{\Phi }(0)|M^k\mathrm{exp}(\alpha c^{}+\alpha ^{}c)$$
$`(5)`$
where
$$M^k=\mathrm{exp}[i\frac{k}{\gamma }(H_0+H_1)],$$
$`(6)`$
$$H_0=\omega _0b^{}b+\omega _aa^{}a+\mathrm{\Omega }(\alpha ab^{}+\alpha ^{}a^{}b),$$
$`(7)`$
$$H_1=\mathrm{\Omega }(acb^{}+a^{}c^{}b).$$
$`(8)`$
If $`|\alpha |1`$, we can adopt the Bogoliubov approximation , i.e., neglect $`H_1`$ in Eq.(5). Then in the Bogoliubov approximation, the density operator $`\rho (t)`$ can be reexpressed as follows,
$$\rho (t)=|\alpha _{TT}\alpha |\rho _s(t),$$
$`(9)`$
where
$$\rho _s(t)=e^{\gamma t}\underset{k=0}{\overset{\mathrm{}}{}}\frac{(\gamma t)^k}{k!}\mathrm{exp}(i\frac{kH_0}{\gamma })|0_{FF}0||\xi \xi |\mathrm{exp}(i\frac{kH_0}{\gamma }),$$
$`(10)`$
is the reduced density operator describing the subsystem of the untrapped atoms and optical field.
Now, we confine ourselves in the resonance case, i.e., $`\omega _0=\omega _a=\omega `$. We define the operators $`a(k)`$ ($`a^{}(k)`$) and $`b(k)`$ ($`b^{}(k)`$) as
$$a(k)=\mathrm{exp}(\frac{ikH_0}{\gamma })a\mathrm{exp}(\frac{ikH_0}{\gamma }),$$
$$a^{}(k)=\mathrm{exp}(\frac{ikH_0}{\gamma })a^{}\mathrm{exp}(\frac{ikH_0}{\gamma }),$$
$$b(k)=\mathrm{exp}(\frac{ikH_0}{\gamma })b\mathrm{exp}(\frac{ikH_0}{\gamma }),$$
$$b^{}(k)=\mathrm{exp}(\frac{ikH_0}{\gamma })b^{}\mathrm{exp}(\frac{ikH_0}{\gamma }).$$
$`(11)`$
It is easy to obtain that
$$a(k)=\mathrm{cos}(\frac{k\mathrm{\Omega }^{}}{\gamma })e^{ik\omega /\gamma }ai\mathrm{sin}(\frac{k\mathrm{\Omega }^{}}{\gamma })e^{i(\theta k\omega /\gamma )}b,$$
$$a^{}(k)=\mathrm{cos}(\frac{k\mathrm{\Omega }^{}}{\gamma })e^{ik\omega /\gamma }a^{}+i\mathrm{sin}(\frac{k\mathrm{\Omega }^{}}{\gamma })e^{i(\theta k\omega /\gamma )}b^{},$$
$$b(k)=\mathrm{cos}(\frac{k\mathrm{\Omega }^{}}{\gamma })e^{ik\omega /\gamma }bi\mathrm{sin}(\frac{k\mathrm{\Omega }^{}}{\gamma })e^{i(\theta +k\omega /\gamma )}a,$$
$$b^{}(k)=\mathrm{cos}(\frac{k\mathrm{\Omega }^{}}{\gamma })e^{ik\omega /\gamma }b^{}+i\mathrm{sin}(\frac{k\mathrm{\Omega }^{}}{\gamma })e^{i(\theta +k\omega /\gamma )}a^{},$$
$`(12)`$
where $`\mathrm{\Omega }^{}=|\alpha |\mathrm{\Omega }`$ and $`e^{i\theta }=\alpha /|\alpha |`$. By making use of Eq.(12), we can express average values of arbitrary operator functionals $`G(a,a^{};b,b^{})`$ as following
$$\text{Tr}[G(a,a^{};b,b^{})\rho _s(t)]=e^{\gamma t}\underset{k=0}{\overset{\mathrm{}}{}}\frac{(\gamma t)^k}{k!}\text{Tr}[G(a(k),a^{}(k);b(k),b^{}(k))|0_{FF}0||\xi \xi |].$$
$`(13)`$
## III. THE INFLUENCE OF INTRINSIC DECOHERENCE ON THE NONCLASSICAL PROPERTIES OF THE ATOM LASER BEAM
In this section, we investigate the nonclassical properties of the atom laser in the Milburnโs model of decoherence. The average numbers and the fluctuation of the output atoms as well as the out-state photon can be obtained by making use of the Eq.(13)
$$N_a(t)=\text{Tr}(a^{}a\rho _s(t))$$
$$=[\frac{1}{2}+\frac{1}{4}\mathrm{exp}(\gamma te^{2i\mathrm{\Omega }^{}/\gamma }\gamma t)+\frac{1}{4}\mathrm{exp}(\gamma te^{2i\mathrm{\Omega }^{}/\gamma }\gamma t)]\mathrm{sinh}^2r,$$
$`(14)`$
$$N_b(t)=\text{Tr}(b^{}b\rho _s(t))$$
$$=[\frac{1}{2}\frac{1}{4}\mathrm{exp}(\gamma te^{2i\mathrm{\Omega }^{}/\gamma }\gamma t)\frac{1}{4}\mathrm{exp}(\gamma te^{2i\mathrm{\Omega }^{}/\gamma }\gamma t)]\mathrm{sinh}^2r,$$
$`(15)`$
$$\mathrm{}N_a^2(t)=\text{Tr}(a^{}aa^{}a\rho _s(t))[\text{Tr}(a^{}a\rho _s(t))]^2$$
$$=\{\frac{3}{4}+e^{\gamma t}[\frac{1}{2}\mathrm{exp}(\gamma te^{2i\mathrm{\Omega }^{}/\gamma })+\frac{1}{2}\mathrm{exp}(\gamma te^{2i\mathrm{\Omega }^{}/\gamma })+\frac{1}{8}\mathrm{exp}(\gamma te^{4i\mathrm{\Omega }^{}/\gamma })+\frac{1}{8}\mathrm{exp}(\gamma te^{4i\mathrm{\Omega }^{}/\gamma })]\}\mathrm{sinh}^2r\mathrm{cosh}^2r$$
$$+\{\frac{1}{8}+\frac{1}{16}e^{\gamma t}[\mathrm{exp}(\gamma te^{4i\mathrm{\Omega }^{}/\gamma })+\mathrm{exp}(\gamma te^{4i\mathrm{\Omega }^{}/\gamma })]$$
$$\frac{1}{16}e^{2\gamma t}[\mathrm{exp}(2\gamma te^{2i\mathrm{\Omega }^{}/\gamma })+\mathrm{exp}(2\gamma te^{2i\mathrm{\Omega }^{}/\gamma })+2\mathrm{exp}(2\gamma t\mathrm{cos}(2\mathrm{\Omega }^{}/\gamma ))]\}\mathrm{sinh}^4r$$
$$+\{\frac{1}{8}\frac{1}{16}e^{\gamma t}[\mathrm{exp}(\gamma te^{4i\mathrm{\Omega }^{}/\gamma })+\mathrm{exp}(\gamma te^{4i\mathrm{\Omega }^{}/\gamma })]\}\mathrm{sinh}^2r,$$
$`(16)`$
$$\mathrm{}N_b^2(t)=\text{Tr}(b^{}bb^{}b\rho _s(t))[\text{Tr}(b^{}b\rho _s(t))]^2$$
$$=\{\frac{3}{4}e^{\gamma t}[\frac{1}{2}\mathrm{exp}(\gamma te^{2i\mathrm{\Omega }^{}/\gamma })+\frac{1}{2}\mathrm{exp}(\gamma te^{2i\mathrm{\Omega }^{}/\gamma })\frac{1}{8}\mathrm{exp}(\gamma te^{4i\mathrm{\Omega }^{}/\gamma })\frac{1}{8}\mathrm{exp}(\gamma te^{4i\mathrm{\Omega }^{}/\gamma })]\}\mathrm{sinh}^2r\mathrm{cosh}^2r$$
$$+\{\frac{1}{8}+\frac{1}{16}e^{\gamma t}[\mathrm{exp}(\gamma te^{4i\mathrm{\Omega }^{}/\gamma })+\mathrm{exp}(\gamma te^{4i\mathrm{\Omega }^{}/\gamma })]$$
$$\frac{1}{16}e^{2\gamma t}[\mathrm{exp}(2\gamma te^{2i\mathrm{\Omega }^{}/\gamma })+\mathrm{exp}(2\gamma te^{2i\mathrm{\Omega }^{}/\gamma })+2\mathrm{exp}(2\gamma t\mathrm{cos}(2\mathrm{\Omega }^{}/\gamma ))]\}\mathrm{sinh}^4r$$
$$+\{\frac{1}{8}\frac{1}{16}e^{\gamma t}[\mathrm{exp}(\gamma te^{4i\mathrm{\Omega }^{}/\gamma })+\mathrm{exp}(\gamma te^{4i\mathrm{\Omega }^{}/\gamma })]\}\mathrm{sinh}^2r,$$
$`(17)`$
where $`r=2|\xi |`$.
In order to discuss the quantum statistical properties of the atom laser and the out-state optical field, we can calculate the Mandel Q parameters defined as
$$Q_i(t)=\frac{\mathrm{}N_i^2(t)N_i(t)}{N_i(t)},(i=a,b)$$
$`(18)`$
$`Q_i(t)<0`$, $`Q_i(t)=0`$ or $`Q_i(t)>0`$ mean the quantum state of the optical field or the atom laser field satisfy sub-Poisson, Poisson or super-Poisson distribution, respectively. In Fig.1 and Fig.2, the Mandel Q parameters $`Q_a`$ and $`Q_b`$ are plotted as a function of the time $`t`$ for two different values of the parameter $`\gamma `$, respectively. We can observe that both the optical field and the atom laser beam satisfy the super-poisson distribution for any $`t>0`$ in the cases with finite values of the parameter $`\gamma `$.
We now turn to discuss the influence of intrinsic decoherence on quadrature squeezing of atom laser and the optical field. We introduce four quadrature operators defined by
$$X_1^{(a)}=\frac{1}{2}(a+a^{}),X_2^{(a)}=\frac{1}{2i}(aa^{}),X_1^{(b)}=\frac{1}{2}(b+b^{}),X_2^{(b)}=\frac{1}{2i}(bb^{})$$
$`(19)`$
These operators satisfy the commutation relation
$$[X_1^{(a)},X_2^{(a)}]=\frac{i}{2},[X_1^{(b)},X_2^{(b)}]=\frac{i}{2},$$
$`(20)`$
which implies the Heisenberg uncertainly relations
$$(\mathrm{}X_1^{(a)})^2(\mathrm{}X_2^{(a)})^2\frac{1}{16},(\mathrm{}X_1^{(b)})^2(\mathrm{}X_2^{(b)})^2\frac{1}{16}.$$
$`(21)`$
Squeezing is said to exist whenever $`(\mathrm{}X_i^{(j)})^2<1/4`$, ($`i=1,2`$), ($`j=a,b`$). In order to characterize the influence of intrinsic decoherence on the quadrature squeezing of the atom laser and optical field, we calculate the following squeezing coefficients
$$S_i^{(j)}=\frac{(\mathrm{}X_i^{(j)})^20.25}{0.25},$$
$`(22)`$
where $`1S_i^{(j)}<0`$ for quadrature squeezing. We express the squeezing coefficients as follows
$$S_1^{(a)}=2N_a(t)+\text{Re}\{e^{\gamma t}[\mathrm{exp}(\gamma te^{2i\omega /\gamma })+\frac{1}{2}\mathrm{exp}(\gamma te^{2i(\mathrm{\Omega }^{}\omega )/\gamma })+\frac{1}{2}\mathrm{exp}(\gamma te^{2i(\mathrm{\Omega }^{}+\omega )/\gamma })]\mathrm{sinh}r\mathrm{cosh}re^{i\varphi }\},$$
$`(23)`$
$$S_2^{(a)}=2N_a(t)\text{Re}\{e^{\gamma t}[\mathrm{exp}(\gamma te^{2i\omega /\gamma })+\frac{1}{2}\mathrm{exp}(\gamma te^{2i(\mathrm{\Omega }^{}\omega )/\gamma })+\frac{1}{2}\mathrm{exp}(\gamma te^{2i(\mathrm{\Omega }^{}+\omega )/\gamma })]\mathrm{sinh}r\mathrm{cosh}re^{i\varphi }\},$$
$`(24)`$
$$S_1^{(b)}=2N_b(t)+\text{Re}\{e^{\gamma t2i\theta }[\mathrm{exp}(\gamma te^{2i\omega /\gamma })$$
$$+\frac{1}{2}\mathrm{exp}(\gamma te^{2i(\mathrm{\Omega }^{}\omega )/\gamma })+\frac{1}{2}\mathrm{exp}(\gamma te^{2i(\mathrm{\Omega }^{}+\omega )/\gamma })]\mathrm{sinh}r\mathrm{cosh}re^{i\varphi }\},$$
$`(25)`$
$$S_2^{(b)}=2N_b(t)\text{Re}\{e^{\gamma t2i\theta }[\mathrm{exp}(\gamma te^{2i\omega /\gamma })$$
$$+\frac{1}{2}\mathrm{exp}(\gamma te^{2i(\mathrm{\Omega }^{}\omega )/\gamma })+\frac{1}{2}\mathrm{exp}(\gamma te^{2i(\mathrm{\Omega }^{}+\omega )/\gamma })]\mathrm{sinh}r\mathrm{cosh}re^{i\varphi }\},$$
$`(26)`$
where $`e^{i\varphi }=\xi /|\xi |`$. It is obvious that $`S_1^{(b)}+S_2^{(b)}=4N_b(t)`$, and $`N_b(t)`$ is always non-negative. So, we need only investigate $`S_1^{(b)}`$ or $`S_2^{(b)}`$ to explore the squeezing property of the atom laser. In what follows, it is assumed $`\varphi =\theta =0`$. We start our analysis of the squeezing properties of both the atom laser and the optical field in the limit case with $`\gamma \mathrm{}`$ and finite values of $`\omega `$ and $`\mathrm{\Omega }^{}`$, which means not any decoherence is presented. Then, the Eqs.(23-26) reduces to the results in Ref., in which the squeezing coefficients of both the atom laser and the optical field exhibit the Rabi-like oscillation. If $`\gamma `$ is a finite value but remains large, i.e. $`\omega /\gamma 1`$ and $`\mathrm{\Omega }^{}/\gamma 1`$, the expressions of the squeezing coefficients $`S_i^{(a)}`$ and $`S_i^{(b)}`$ ($`i=1,2`$) can be approximated as
$$S_1^{(a)}[1+\mathrm{cos}2\mathrm{\Omega }^{}t\mathrm{exp}(2\mathrm{\Omega }^2t/\gamma )]\mathrm{sinh}^2r+\{\mathrm{cos}2\omega t\mathrm{exp}(2\omega ^2t/\gamma )$$
$$+\frac{1}{2}\mathrm{cos}[2(\mathrm{\Omega }^{}\omega )t]\mathrm{exp}[2(\mathrm{\Omega }^{}\omega )^2t/\gamma ]+\frac{1}{2}\mathrm{cos}[2(\mathrm{\Omega }^{}+\omega )t]\mathrm{exp}[2(\mathrm{\Omega }^{}+\omega )^2t/\gamma ]\}\mathrm{sinh}r\mathrm{cosh}r,$$
$`(27)`$
$$S_2^{(a)}[1+\mathrm{cos}2\mathrm{\Omega }^{}t\mathrm{exp}(2\mathrm{\Omega }^2t/\gamma )]\mathrm{sinh}^2r\{\mathrm{cos}2\omega t\mathrm{exp}(2\omega ^2t/\gamma )$$
$$+\frac{1}{2}\mathrm{cos}[2(\mathrm{\Omega }^{}\omega )t]\mathrm{exp}[2(\mathrm{\Omega }^{}\omega )^2t/\gamma ]+\frac{1}{2}\mathrm{cos}[2(\mathrm{\Omega }^{}+\omega )t]\mathrm{exp}[2(\mathrm{\Omega }^{}+\omega )^2t/\gamma ]\}\mathrm{sinh}r\mathrm{cosh}r,$$
$`(28)`$
$$S_1^{(b)}[1\mathrm{cos}2\mathrm{\Omega }^{}t\mathrm{exp}(2\mathrm{\Omega }^2t/\gamma )]\mathrm{sinh}^2r+\{\mathrm{cos}2\omega t\mathrm{exp}(2\omega ^2t/\gamma )$$
$$+\frac{1}{2}\mathrm{cos}[2(\mathrm{\Omega }^{}\omega )t]\mathrm{exp}[2(\mathrm{\Omega }^{}\omega )^2t/\gamma ]+\frac{1}{2}\mathrm{cos}[2(\mathrm{\Omega }^{}+\omega )t]\mathrm{exp}[2(\mathrm{\Omega }^{}+\omega )^2t/\gamma ]\}\mathrm{sinh}r\mathrm{cosh}r,$$
$`(29)`$
$$S_2^{(b)}[1\mathrm{cos}2\mathrm{\Omega }^{}t\mathrm{exp}(2\mathrm{\Omega }^2t/\gamma )]\mathrm{sinh}^2r\{\mathrm{cos}2\omega t\mathrm{exp}(2\omega ^2t/\gamma )$$
$$+\frac{1}{2}\mathrm{cos}[2(\mathrm{\Omega }^{}\omega )t]\mathrm{exp}[2(\mathrm{\Omega }^{}\omega )^2t/\gamma ]+\frac{1}{2}\mathrm{cos}[2(\mathrm{\Omega }^{}+\omega )t]\mathrm{exp}[2(\mathrm{\Omega }^{}+\omega )^2t/\gamma ]\}\mathrm{sinh}r\mathrm{cosh}r,$$
$`(30)`$
From Eqs.(27-30), we find that all of the squeezing coefficients $`S_i^{(a)}`$ and $`S_i^{(b)}`$ ($`i=1,2`$) tend to a fixed positive value $`\mathrm{sinh}^2r`$ as the time $`t\mathrm{}`$, except that they tend to $`\mathrm{sinh}^2r\pm \frac{1}{2}\mathrm{sinh}r\mathrm{cosh}r`$ in the special case with $`\mathrm{\Omega }^{}=\omega `$. With the further decrease of $`\gamma `$, the oscillatory behaviors of the squeezing properties of both the atom laser and the optical field become frozen. In Fig.3, the squeezing coefficient $`S_2^{(b)}`$ is plotted as a function of time $`t`$ for three different values of parameter $`\gamma `$. With the decreases of parameter $`\gamma `$, we can observe rapid deterioration of the Rabi-like oscillation of squeezing coefficient. In Fig.3(b) and Fig3(c), if the time $`t`$ become very large, the squeezing coefficient $`S_2^{(b)}`$ tends to be a fixed positive value $`0.093`$, which means the decoherence eventually completely destroy the squeezing property of the atom laser. However, under very special conditions that $`\mathrm{\Omega }^{}=\omega \gamma `$, $`0<\mathrm{tanh}r<\frac{1}{2}`$ and $`\gamma `$ is large but remain finite, $`S_2^{(b)}`$ will tend to be a negative value as the time $`t`$ approaches to infinite, which means the stationary state of the atom laser gets squeezed. This can be clearly seen from Fig.4.
Recently, the sensitivity of quantum systems that are chaotic in a classical limit to small perturbations has been investigated in Ref., and the relation between the sensitive and decoherence has been discussed. In what follows, we briefly investigate influence of small perturbation on the quadrature squeezing of atom laser. From Eq.(26), we can observe that the dynamical behavior of $`S_2^{(b)}`$ is dependent on the values of the exponential facts $`\gamma [e^{2i(\mathrm{\Omega }^{}\omega )/\gamma }1]`$, $`\gamma [e^{\pm 2i\mathrm{\Omega }^{}/\gamma }1]`$, $`\gamma [e^{2i\omega /\gamma }1]`$, and $`\gamma [e^{2i(\mathrm{\Omega }^{}+\omega )/\gamma }1]`$. In the case with $`\mathrm{\Omega }^{}\omega \gamma `$, the term $`\gamma [e^{2i(\mathrm{\Omega }^{}\omega )/\gamma }1](2i\delta 2\delta ^2/\gamma )`$ plays a dominantly role in the long time dynamical behavior, where $`\delta =\mathrm{\Omega }^{}\omega `$. In Fig.5, we plotted $`S_2^{(b)}`$ as the function of time $`t`$ for different small values of $`\delta `$. Since $`\mathrm{\Omega }^{}`$ is dependent on the amplitude $`\alpha `$ of the initial coherent state of condensed atoms in the trapped state $`|T`$, we expect that the squeezing dynamical behavior of the atom laser is sensitive to the initial state of trapped condensed atoms, just as shown in Fig.5.
## IV. DISCUSSION
In this paper, we investigate nonclassical properties of the output of a Bose-Einstein condensate in Milburnโs model of intrinsic decoherence by making use of Bogoliubov approximation. It is shown that the squeezing properties of the atom laser field can be destroyed by the decoherence under most conditions. However, under some very special conditions, the squeezing properties of atom laser is robust against the decoherence. This phenomenon is highly dependent on the particular modelling of decoherence.
In what follows, we briefly discuss the relevance of our theoretical results to the realistic experimental conditions. By making use of the preliminary atom laser experiments , as a guide to realistic parameter values, we choose $`\omega =300`$kHz, $`\mathrm{\Omega }=60`$kHz, the initial trapped atom number $`N=|\alpha |^2=10^6`$, the mean frequency of the unitary time step $`\gamma =10^5`$MHz, and $`r=1`$ as a example to illustrate our theoretical results. In such a condition, the Rabi-like oscillation behaviors of the numbers of both the output atom and output photon become frozen after $`100\mu s`$. However, the atom laser persists in exhibiting oscillation of squeezing coefficient and its squeezing property is finally destroyed by decoherence till about $`0.15s`$.
Milburnโs model of intrinsic decoherence is currently a very active field of research. Nevertheless, so far little attention has been paid to its study in the context of ultracold quantum gases. However, due to their high sensitivity to decoherence, such systems could provide an interesting testing ground for the Milburn model. In this paper, we provide a first step in this direction by studying the influence of intrinsic decoherence on the output properties of a simple atom laser model. A thorough discussion of all other sources of decoherence that could be present in an experiment and comparison with the results presented in this work should be interesting and necessary in the future study. It is also very interesting to investigate the entanglement between output atoms and output photons and discuss the influence of intrinsic decoherence on entanglement \[25-28\]. Moreover, it is worth to discuss the possible quantum chaotic dynamical behavior of atom laser and the quantum-classical corresponding by fully taking account of the atom-atom interaction.
## ACNOWLEDGMENTS
This project was supported by the National Natural Science Foundation of China (Project NO. 10174066).
Figure Caption
The Mandel Q parameter $`Q_a`$ of optical field is plotted as a function of time $`t`$ with $`\mathrm{\Omega }^{}=1`$ and $`r=2`$ for two different values of the parameter $`\gamma `$, (Solid line) $`\gamma =\mathrm{}`$, (Dot line) $`\gamma =10^2`$.
The Mandel Q parameter $`Q_b`$ of atom laser field is plotted as a function of time $`t`$ with $`\mathrm{\Omega }^{}=1`$ and $`r=2`$ for two different values of the parameter $`\gamma `$, (Solid line) $`\gamma =\mathrm{}`$, (Dot line) $`\gamma =10^2`$.
The quadrature squeezing coefficient $`S_2^{(b)}`$ of the atom laser field is plotted as a function of time $`t`$ with $`\omega =0.1`$, $`\varphi =0`$, $`\theta =0`$, $`\mathrm{\Omega }^{}=\pi `$ and $`r=0.3`$ for three different values of the parameter $`\gamma `$, (a) $`\gamma =\mathrm{}`$, (b) $`\gamma =10^3`$, (c) $`\gamma =10^2`$.
The quadrature squeezing coefficient $`S_2^{(b)}`$ of the atom laser field is plotted as a function of time $`t`$ with $`\omega =10`$, $`\varphi =0`$, $`\theta =0`$, $`\mathrm{\Omega }^{}=10`$, $`r=0.3`$ and $`\gamma =10^2`$.
The quadrature squeezing coefficient $`S_2^{(b)}`$ of the atom laser field is plotted as a function of time $`t`$ with $`\omega =10`$, $`\varphi =0`$, $`\theta =0`$, $`r=0.4`$ and $`\gamma =10^2`$ for four different values of $`\mathrm{\Omega }^{}`$: (Solid line) $`\mathrm{\Omega }^{}=10`$, (Dot line) $`\mathrm{\Omega }^{}=10+10^7`$, (Dash Dot line) $`\mathrm{\Omega }^{}=10+2\times 10^7`$, (Dash Dot Dot line) $`\mathrm{\Omega }^{}=10+3\times 10^7`$.
|
warning/0506/astro-ph0506540.html
|
ar5iv
|
text
|
# Generation of Gaussian Density Fields
## 1 INTRODUCTION
Gaussian density field play a major role in cosmology. There is now strong evidence that the large-scale structure of the universe originated from the growth, by gravitational instability, of primordial density fluctuations. Observations of the temperature fluctuations of the Cosmic Microwave Background (CMB) indicate that these primordial fluctuation were Gaussian.
The formation of evolution of large-scale structure in the universe is a complex problem that requires a numerical approach. Typically, one creates a realization of the density fluctuations at early time, and uses a numerical algorithm to evolve this fluctuation all the way to the present (or to some redshift of interest). Since it is impossible to simulate the entire universe (which might very well be infinite), we normally assume that the universe is periodic at large scales. We can then divide the universe into identical cubes of volume $`V_{\mathrm{box}}=L_{\mathrm{box}}^3`$, and we only need to simulate one cube. This approximation is valid as long as the box size $`L_{\mathrm{box}}`$ is much larger than any existing large-scale structure in the universe. We can rephrase this by saying that the box must contain a โfairโ sample of the universe.
The density field can be represented in two different ways. In the Eulerian Representation, the box is divided into $`N\times N\times N`$ cells, and the density contrast $`\delta `$ is calculated at the center of each cell. In the Lagrangian Representation, $`N_p\times N_p\times N_p`$ equal-mass particles are laid down on a cubic grid inside the box, and are then displaced in order to represent the density fluctuation. The choice of representation depends on the particular algorithm that will be used to evolve the system from these initial conditions.
### 1.1 The Power Spectrum
A Gaussian density field can be represented as a superposition of plane waves of wavevectors $`๐ค`$ and complex amplitudes $`\delta _๐ค^{\mathrm{cont}}`$, where the superscript โcontโ stands for continuous, indicating that all values of $`๐ค`$ are allowed. The amplitudes are related to the power spectrum $`P(k)`$ by
$$|\delta _๐ค^{\mathrm{cont}}|^2P(k),$$
(1)
where $`k=|๐ค|`$. However, there is a lot of confusion in the literature about the constant of proportionality between $`|\delta _๐ค^{\mathrm{cont}}|^2`$ and $`P(k)`$, and even the units of $`\delta _๐ค^{\mathrm{cont}}`$ can vary from one author to another. We will clarify this issue in ยง2.
Note: Equation (1) is a convenient simplification. As we will discuss later, in a truly Gaussian random field, the amplitudes $`\delta _๐ค`$ are determined only in a statistical sense. Their real and imaginary parts are separately given by a Gaussian distribution whose variance is proportional to $`P(k)`$. However, using equation (1) greatly simplifies the derivation presented in ยง2, without affecting the results.
## 2 THE AMPLITUDES OF THE DENSITY MODES
We assume that the universe is periodic over a comoving cubic volume $`V_{\mathrm{box}}=L_{\mathrm{box}}^3`$. The density contrast $`\delta `$ can be decomposed into a sum of plane waves.
$$\delta (๐ซ)=\frac{1}{N^3}\underset{๐ค}{}\delta _๐ค^{\mathrm{disc}}e^{i๐ค๐ซ},$$
(2)
where $`๐ซ`$ is the comoving position. The wavevectors $`๐ค`$ are given by
$$๐ค=(l,m,n)k_0,l,m,n=\mathrm{},\mathrm{},1,0,1,\mathrm{},\mathrm{}.$$
(3)
where the fundamental wavenumber is
$$k_0=\frac{2\pi }{L_{\mathrm{box}}}.$$
(4)
The superscript โ$`\mathrm{disc}`$โ indicates that the amplitudes $`\delta _๐ค^{\mathrm{disc}}`$ form a discrete spectrum, that is, they are defined for particular, discrete values of $`๐ค`$.<sup>1</sup><sup>1</sup>1Other values of $`๐ค`$ would not satisfy the periodic boundary conditions The factor $`1/N^3`$ is not necessary at this point, and could be absorbed into the definition of $`\delta _๐ค^{\mathrm{disc}}`$. We introduce it to make the notation consistent with ยง3. To make $`\delta (๐ซ)`$ real, the coefficients $`\delta _๐ค^{\mathrm{disc}}`$ must satisfy the reality condition:
$$\delta _๐ค^{\mathrm{disc}}=(\delta _๐ค^{\mathrm{disc}})^{}.$$
(5)
Our first challenge is to relate the discrete sum in equation (2) to the continuous sum of modes present in the real universe, and to express the amplitudes $`\delta _๐ค^{\mathrm{disc}}`$ in terms of the power spectrum $`P(k)`$. To do so, we will consider the rms fluctuation of the density at a certain scale $`R`$, and match the expressions obtained in the discrete and continuous limits.
### 2.1 The Discrete Limit
Consider a sphere of radius $`R`$ centered at $`๐ซ_0`$. The mass inside that sphere is given by
$$M(๐ซ_0)=_{\mathrm{sph}(๐ซ_0)}\overline{\rho }\left[1+\delta (๐ซ)\right]d^3r=\overline{\rho }V_{\mathrm{sph}}+\frac{\overline{\rho }}{N^3}_{\mathrm{sph}(๐ซ_0)}d^3r\underset{๐ค}{}\delta _๐ค^{\mathrm{disc}}e^{i๐ค๐ซ},$$
(6)
where $`\overline{\rho }`$ is the average density, $`V_{\mathrm{sph}}=4\pi R^3/3`$ is the volume of the sphere, and the integral is computed over that volume. The relative mass excess in the sphere is given by
$$\frac{\mathrm{\Delta }M}{M}(๐ซ_0)=\frac{M(๐ซ_0)\overline{\rho }V_{\mathrm{sph}}}{\overline{\rho }V_{\mathrm{sph}}}=\frac{1}{N^3V_{\mathrm{sph}}}_{\mathrm{sph}(๐ซ_0)}d^3r\underset{๐ค}{}\delta _๐ค^{\mathrm{disc}}e^{i๐ค๐ซ}.$$
(7)
We introduce the following change of variables,
$$๐ซ=๐ซ_0+๐ฒ.$$
(8)
In $`๐ฒ`$-space, the sphere is now located at the origin, and equation (7) becomes
$$\frac{\mathrm{\Delta }M}{M}(๐ซ_0)=\frac{1}{N^3V_{\mathrm{sph}}}_{\mathrm{sph}(0)}d^3y\underset{๐ค}{}\delta _๐ค^{\mathrm{disc}}e^{i๐ค๐ซ_0}e^{i๐ค๐ฒ}.$$
(9)
We now square this expression, and get
$$\left(\frac{\mathrm{\Delta }M}{M}\right)^2(๐ซ_0)=\frac{9}{16\pi ^2R^6N^6}\left[_{\mathrm{sph}(0)}d^3y\underset{๐ค}{}\delta _๐ค^{\mathrm{disc}}e^{i๐ค๐ซ_0}e^{i๐ค๐ฒ}\right]\left[_{\mathrm{sph}(0)}d^3z\underset{๐ค^{}}{}\delta _๐ค^{}^{\mathrm{disc}}e^{i๐ค^{}๐ซ_0}e^{i๐ค^{}๐ณ}\right].$$
(10)
The variance of the density contrast at scale $`R`$ is obtained by averaging the above expression over all possible locations $`๐ซ_0`$ of the sphere inside the computational box,
$`\sigma _R^2`$ $``$ $`\left({\displaystyle \frac{\mathrm{\Delta }M}{M}}\right)^2_{V_{\mathrm{box}}}={\displaystyle \frac{1}{V_{\mathrm{box}}}}{\displaystyle _{V_{\mathrm{box}}}}d^3r_0\left({\displaystyle \frac{\mathrm{\Delta }M}{M}}\right)^2(๐ซ_0)`$ (11)
$`=`$ $`{\displaystyle \frac{1}{V_{\mathrm{box}}}}{\displaystyle \frac{9}{16\pi ^2R^6N^6}}{\displaystyle _{V_{\mathrm{box}}}}d^3r_0{\displaystyle _{\mathrm{sph}(0)}}d^3y{\displaystyle _{\mathrm{sph}(0)}}d^3z{\displaystyle \underset{๐ค}{}}{\displaystyle \underset{๐ค^{}}{}}\delta _๐ค^{\mathrm{disc}}\delta _๐ค^{}^{\mathrm{disc}}e^{i๐ค๐ฒ}e^{i๐ค^{}๐ณ}e^{i(๐ค+๐ค^{})๐ซ_0}.`$
The integral over $`V_{\mathrm{box}}`$ reduces to
$$_{V_{\mathrm{box}}}d^3r_0e^{i(๐ค+๐ค^{})๐ซ_0}=V_{\mathrm{box}}\delta _{๐ค,๐ค^{}}.$$
(12)
We substitute this expression in equation (11), and use the Kronecker $`\delta `$ to eliminate the summation over $`๐ค^{}`$. Equation (11) reduces to
$$\sigma _R^2=\frac{9}{16\pi ^2R^6N^6}\underset{๐ค}{}|\delta _๐ค^{\mathrm{disc}}|^2\left[_{\mathrm{sph}(0)}d^3ye^{i๐ค๐ฒ}\right]^2,$$
(13)
where we used equation (5) to get $`\delta _๐ค^{\mathrm{disc}}\delta _๐ค^{\mathrm{disc}}=(\delta _๐ค^{\mathrm{disc}})^{}\delta _๐ค^{\mathrm{disc}}=|\delta _๐ค^{\mathrm{disc}}|^2`$. The remaining integral can be evaluated easily (see Appendix A). Equation (13) reduces to
$$\sigma _R^2=\frac{1}{N^6}\underset{๐ค}{}|\delta _๐ค^{\mathrm{disc}}|^2W^2(kR),$$
(14)
where
$$W(u)\frac{3}{u^3}(\mathrm{sin}uu\mathrm{cos}u).$$
(15)
### 2.2 The Continuous Limit
The real universe is of course not periodic, in which case all values of $`๐ค`$ are allowed. To convert the expressions derived in ยง2.1 from the discrete limit to the continuous one, consider any function $`f_๐ค`$ that is summed over all allowed values of $`๐ค`$. In the discrete limit, we have
$$\underset{๐ค}{}f_๐ค^{\mathrm{disc}}=\underset{\mathrm{all}\mathrm{V}.\mathrm{E}.}{}f_๐ค^{\mathrm{disc}}=\frac{1}{k_0^3}\underset{\mathrm{all}\mathrm{V}.\mathrm{E}.}{}f_๐ค^{\mathrm{disc}}k_0^3=\frac{V_{\mathrm{box}}}{(2\pi )^3}\underset{\mathrm{all}\mathrm{V}.\mathrm{E}.}{}f_๐ค^{\mathrm{disc}}_{\mathrm{V}.\mathrm{E}.}d^3k,$$
(16)
where โV.E.โ represent a volume element in $`๐ค`$-space, which is a cube of volume $`k_0^3`$ centered around an allowed value of $`๐ค`$ (with $`k_0=2\pi /L_{\mathrm{box}}`$). Assuming that the function $`f_๐ค^{\mathrm{disc}}`$ does not vary significantly over one volume element, we can pull it inside the integral,
$$\underset{๐ค}{}f_๐ค^{\mathrm{disc}}\frac{V_{\mathrm{box}}}{(2\pi )^3}\underset{\mathrm{all}\mathrm{V}.\mathrm{E}.}{}_{\mathrm{V}.\mathrm{E}.}f_๐ค^{\mathrm{disc}}d^3k.$$
(17)
Of course, the effect of integrating over the volume element, and then summing over all volume elements, is to effectively integrate over all $`k`$-space, so equation (17) reduces to
$$\underset{๐ค}{}f_๐ค^{\mathrm{disc}}\frac{V_{\mathrm{box}}}{(2\pi )^3}f_๐ค^{\mathrm{disc}}d^3k.$$
(18)
We can rewrite this expression as
$$\underset{๐ค}{}f_๐ค^{\mathrm{disc}}f_๐ค^{\mathrm{cont}}d^3k,$$
(19)
where the continuous and discrete functions are related by
$$f_๐ค^{\mathrm{cont}}=\frac{V_{\mathrm{box}}}{(2\pi )^3}f_๐ค^{\mathrm{disc}}.$$
(20)
Using these formulae, we can rewrite equation (2) as
$$\delta (๐ซ)=\frac{1}{N^3}d^3k\delta _๐ค^{\mathrm{cont}}e^{i๐ค๐ซ},$$
(21)
where
$$\delta _๐ค^{\mathrm{cont}}=\frac{V_{\mathrm{box}}}{(2\pi )^3}\delta _๐ค^{\mathrm{disc}}.$$
(22)
Let us now convert equation (14) into an integral, as we did for equation (16). We get
$$\sigma _R^2=\frac{V_{\mathrm{box}}}{(2\pi )^3N^6}d^3k|\delta _๐ค^{\mathrm{disc}}|^2W^2(kR).$$
(23)
We substitute equation (22) into equation (23), and get
$$\sigma _R^2=\frac{(2\pi )^3}{V_{\mathrm{box}}N^6}d^3k|\delta _๐ค^{\mathrm{cont}}|^2W^2(kR).$$
(24)
We then need to related $`\sigma _R^2`$ to the power spectrum $`P(k)`$. The relation is given by Bunn & White (1997) as
$$\sigma _R^2=\frac{1}{2\pi ^2}_0^{\mathrm{}}๐kk^2P(k)W^2(kR).$$
(25)
This relation is obtained by performing an integration over angles, using the fact that $`P(๐ค)`$ is a function of $`k=|๐ค|`$ only. We can โundoโ this integration, simply by dividing equation (25) by $`4\pi `$. We get
$$\sigma _R^2=\frac{1}{(2\pi )^3}d^3kP(k)W^2(kR).$$
(26)
Comparing equations (23), (24), and (26), we get
$$P(k)=\frac{V_{\mathrm{box}}}{N^6}|\delta _๐ค^{\mathrm{disc}}|^2=\frac{(2\pi )^6}{V_{\mathrm{box}}N^6}|\delta _๐ค^{\mathrm{cont}}|^2.$$
(27)
This gives us the relation between $`P(k)`$, $`\delta _๐ค^{\mathrm{disc}}`$, and $`\delta _๐ค^{\mathrm{cont}}`$. Notice the $`P(k)`$ is neither the square of $`\delta _๐ค^{\mathrm{disc}}`$ nor the square of $`\delta _๐ค^{\mathrm{cont}}`$. Both $`P(k)`$ and $`\delta _๐ค^{\mathrm{cont}}`$ have dimensions of a volume while $`\delta _๐ค^{\mathrm{disc}}`$ is dimensionless.
Equations (25) and (26) define the normalization of the power spectrum. When using any power spectrum obtained from the literature, it is essential to check that these relations are satisfied. Variations by factors of $`2\pi `$ between different papers are quite common.
## 3 DIRECT CALCULATION OF THE DENSITY HARMONICS
Using the formalism described in ยง2, we now want to compute the density harmonics $`\widehat{\delta }_๐ค`$. We first lay down inside the computational volume $`V_{\mathrm{box}}`$ a regular cubic grid of size $`N\times N\times N`$, with grid spacing $`\mathrm{\Delta }=L_{\mathrm{box}}/N`$. The coordinates $`๐ซ`$ of the grid points are given by
$$๐ซ=(\alpha ,\beta ,\gamma )\mathrm{\Delta },\alpha ,\beta ,\gamma =0,1,\mathrm{},N1.$$
(28)
The presence of a grid results in a discretization of space, which in turns modifies the structure of the $`k`$-space. In equation (3), the values of $`๐ค`$ form an infinite cubic grid in $`k`$-space, since the indicies $`l`$, $`m`$, $`n`$ can take any integer value from $`\mathrm{}`$ to $`+\mathrm{}`$. However, the discretization of space limits the number of modes. Consider a mode with wavenumber
$$๐ค^{}=๐ค+(uN,vN,wN)k_0,$$
(29)
where $`u`$, $`v`$, and $`w`$ are integers. The exponential in equation (2) becomes
$`e^{i๐ค^{}๐ซ}`$ $`=`$ $`e^{i๐ค๐ซ}e^{i[uN,vN,wN][\alpha ,\beta ,\gamma ]k_0\mathrm{\Delta }}=e^{i๐ค๐ซ}e^{i[uN,vN,wN][\alpha ,\beta ,\gamma ](2\pi /N)}`$ (30)
$`=`$ $`e^{i๐ค๐ซ}e^{2\pi i[u,v,w][\alpha ,\beta ,\gamma ]}=e^{i๐ค๐ซ}.`$
Hence, the modes $`๐ค^{}`$ and $`๐ค`$ are inseparable. They represent a plane wave with the same effective wavenumber. Consequently, we will consider a finite $`k`$-space, in which the values $`l`$, $`m`$, $`n`$ do not run from $`\mathrm{}`$ to $`+\mathrm{}`$, but instead are limited to the range 0 to $`N1`$. Hence, in $`k`$-space, the density harmonics $`\widehat{\delta }(๐ค)`$ are also defined on a regular cubic grid of size $`N\times N\times N`$ . The wavevectors are given by
$$๐ค=(l,m,n)k_0,l,m,n=0,1,\mathrm{},N1.$$
(31)
In doing so, we are simply neglecting high-frequency modes. This is justified since the discreteness of the grid in $`r`$-space prevents us from resolving any structure that these modes represent. Hence, by using a grid in $`r`$-space, we are effectively performing a filtering of the density fluctuation at the scale of $`\mathrm{\Delta }`$, and such filtering eliminates high-frequency modes.
The Fourier transform and inverse Fourier transform are given respectively by
$`\widehat{\delta }(๐ค)`$ $`=`$ $`{\displaystyle \underset{๐ซ}{}}\delta (๐ซ)e^{i๐ค๐ซ},`$ (32)
$`\delta (๐ซ)`$ $`=`$ $`{\displaystyle \frac{1}{N^3}}{\displaystyle \underset{๐ค}{}}\widehat{\delta }(๐ค)e^{i๐ค๐ซ}.`$ (33)
Notice that equation (33) is the same as equation (2), with a slight change of notation: $`\delta _๐ค^{\mathrm{disc}}\widehat{\delta }(๐ค)`$. We introduce the notation
$`\widehat{\delta }(๐ค)`$ $`=`$ $`\widehat{\delta }_{lmn},`$ (34)
$`\delta (๐ซ)`$ $`=`$ $`\delta _{\alpha \beta \gamma }.`$ (35)
Equation (32) becomes
$`\widehat{\delta }_{lmn}`$ $`=`$ $`{\displaystyle \underset{\alpha ,\beta ,\gamma =0}{\overset{N1}{}}}\delta _{\alpha \beta \gamma }e^{i(2\pi /\mathrm{\Delta }N)\mathrm{\Delta }[l,m,n][\alpha ,\beta ,\gamma ]}={\displaystyle \underset{\alpha ,\beta ,\gamma =0}{\overset{N1}{}}}\delta _{\alpha \beta \gamma }e^{2\pi il\alpha /N}e^{2\pi im\beta /N}e^{2\pi in\gamma /N}`$ (36)
$`=`$ $`{\displaystyle \underset{\alpha ,\beta ,\gamma =0}{\overset{N1}{}}}\left(\mathrm{cos}{\displaystyle \frac{2\pi l\alpha }{N}}+i\mathrm{sin}{\displaystyle \frac{2\pi l\alpha }{N}}\right)\left(\mathrm{cos}{\displaystyle \frac{2\pi m\beta }{N}}+i\mathrm{sin}{\displaystyle \frac{2\pi m\beta }{N}}\right)\left(\mathrm{cos}{\displaystyle \frac{2\pi n\gamma }{N}}+i\mathrm{sin}{\displaystyle \frac{2\pi n\gamma }{N}}\right).`$
After expansion, this expression becomes
$$\widehat{\delta }_{lmn}=(\widehat{\delta }_{eee}+\widehat{\delta }_{eoo}+\widehat{\delta }_{oeo}+\widehat{\delta }_{ooe})+i(\widehat{\delta }_{eeo}+\widehat{\delta }_{eoe}+\widehat{\delta }_{oee}+\widehat{\delta }_{ooo}),$$
(37)
where we define
$`\widehat{\delta }_{eee}`$ $``$ $`{\displaystyle \underset{\alpha ,\beta ,\gamma =0}{\overset{N1}{}}}\delta _{\alpha \beta \gamma }\mathrm{cos}{\displaystyle \frac{2\pi l\alpha }{N}}\mathrm{cos}{\displaystyle \frac{2\pi m\beta }{N}}\mathrm{cos}{\displaystyle \frac{2\pi n\gamma }{N}},`$ (38)
$`\widehat{\delta }_{eeo}`$ $``$ $`{\displaystyle \underset{\alpha ,\beta ,\gamma =0}{\overset{N1}{}}}\delta _{\alpha \beta \gamma }\mathrm{cos}{\displaystyle \frac{2\pi l\alpha }{N}}\mathrm{cos}{\displaystyle \frac{2\pi m\beta }{N}}\mathrm{sin}{\displaystyle \frac{2\pi n\gamma }{N}},`$ (39)
$`\widehat{\delta }_{eoe}`$ $``$ $`{\displaystyle \underset{\alpha ,\beta ,\gamma =0}{\overset{N1}{}}}\delta _{\alpha \beta \gamma }\mathrm{cos}{\displaystyle \frac{2\pi l\alpha }{N}}\mathrm{sin}{\displaystyle \frac{2\pi m\beta }{N}}\mathrm{cos}{\displaystyle \frac{2\pi n\gamma }{N}},`$ (40)
$`\widehat{\delta }_{eoo}`$ $``$ $`{\displaystyle \underset{\alpha ,\beta ,\gamma =0}{\overset{N1}{}}}\delta _{\alpha \beta \gamma }\mathrm{cos}{\displaystyle \frac{2\pi l\alpha }{N}}\mathrm{sin}{\displaystyle \frac{2\pi m\beta }{N}}\mathrm{sin}{\displaystyle \frac{2\pi n\gamma }{N}},`$ (41)
$`\widehat{\delta }_{oee}`$ $``$ $`{\displaystyle \underset{\alpha ,\beta ,\gamma =0}{\overset{N1}{}}}\delta _{\alpha \beta \gamma }\mathrm{sin}{\displaystyle \frac{2\pi l\alpha }{N}}\mathrm{cos}{\displaystyle \frac{2\pi m\beta }{N}}\mathrm{cos}{\displaystyle \frac{2\pi n\gamma }{N}},`$ (42)
$`\widehat{\delta }_{oeo}`$ $``$ $`{\displaystyle \underset{\alpha ,\beta ,\gamma =0}{\overset{N1}{}}}\delta _{\alpha \beta \gamma }\mathrm{sin}{\displaystyle \frac{2\pi l\alpha }{N}}\mathrm{cos}{\displaystyle \frac{2\pi m\beta }{N}}\mathrm{sin}{\displaystyle \frac{2\pi n\gamma }{N}},`$ (43)
$`\widehat{\delta }_{ooe}`$ $``$ $`{\displaystyle \underset{\alpha ,\beta ,\gamma =0}{\overset{N1}{}}}\delta _{\alpha \beta \gamma }\mathrm{sin}{\displaystyle \frac{2\pi l\alpha }{N}}\mathrm{sin}{\displaystyle \frac{2\pi m\beta }{N}}\mathrm{cos}{\displaystyle \frac{2\pi n\gamma }{N}},`$ (44)
$`\widehat{\delta }_{ooo}`$ $``$ $`{\displaystyle \underset{\alpha ,\beta ,\gamma =0}{\overset{N1}{}}}\delta _{\alpha \beta \gamma }\mathrm{sin}{\displaystyle \frac{2\pi l\alpha }{N}}\mathrm{sin}{\displaystyle \frac{2\pi m\beta }{N}}\mathrm{sin}{\displaystyle \frac{2\pi n\gamma }{N}}.`$ (45)
Consider now the mode $`\widehat{\delta }_{Nl,m,n}`$. We replace $`l`$ by $`Nl`$ in equation (36), and get
$$\widehat{\delta }_{Nl,mn}=\underset{\alpha ,\beta ,\gamma =0}{\overset{N1}{}}\delta _{\alpha \beta \gamma }e^{2\pi i(Nl)\alpha /N}e^{2\pi im\beta /N}e^{2\pi in\gamma /N}=\underset{\alpha ,\beta ,\gamma =0}{\overset{N1}{}}\delta _{\alpha \beta \gamma }e^{2\pi i\alpha }e^{2\pi il\alpha /N}e^{2\pi im\beta /N}e^{2\pi in\gamma /N}.$$
(46)
Since $`\alpha `$ is an integer, the first exponential is always unity, and equation (46) reduces to
$$\widehat{\delta }_{Nl,mn}=\underset{\alpha ,\beta ,\gamma =0}{\overset{N1}{}}\left(\mathrm{cos}\frac{2\pi l\alpha }{N}i\mathrm{sin}\frac{2\pi l\alpha }{N}\right)\left(\mathrm{cos}\frac{2\pi m\beta }{N}+i\mathrm{sin}\frac{2\pi m\beta }{N}\right)\left(\mathrm{cos}\frac{2\pi n\gamma }{N}+i\mathrm{sin}\frac{2\pi n\gamma }{N}\right).$$
(47)
Comparing equations (36) and (47), we see that the effect of replacing $`l`$ by $`Nl`$ amounts to a change of sign of the first sine function. In equations (38)โ(45), that sine appears only in the $`\widehat{\delta }`$โs for which the first subscript is โ$`o`$.โ Hence, these $`\widehat{\delta }`$โs will change sign, and equation (37) will become
$$\widehat{\delta }_{Nl,mn}=(\widehat{\delta }_{eee}+\widehat{\delta }_{eoo}\widehat{\delta }_{oeo}\widehat{\delta }_{ooe})+i(\widehat{\delta }_{eeo}+\widehat{\delta }_{eoe}\widehat{\delta }_{oee}\widehat{\delta }_{ooo}),$$
(48)
We can directly generalize to the other indicies, or combinations of them. Replacing $`m`$ by $`Nm`$ changes the sign of the $`\widehat{\delta }`$โs for which the second subscript is โ$`o`$,โ and replacing $`n`$ by $`Nn`$ changes the sign of the $`\widehat{\delta }`$โs for which the third subscript is โ$`o`$.โ Hence,
$`\widehat{\delta }_{l,Nm,n}`$ $`=`$ $`(\widehat{\delta }_{eee}\widehat{\delta }_{eoo}+\widehat{\delta }_{oeo}\widehat{\delta }_{ooe})+i(\widehat{\delta }_{eeo}\widehat{\delta }_{eoe}+\widehat{\delta }_{oee}\widehat{\delta }_{ooo}),`$ (49)
$`\widehat{\delta }_{lm,Nn}`$ $`=`$ $`(\widehat{\delta }_{eee}\widehat{\delta }_{eoo}\widehat{\delta }_{oeo}+\widehat{\delta }_{ooe})+i(\widehat{\delta }_{eeo}+\widehat{\delta }_{eoe}+\widehat{\delta }_{oee}\widehat{\delta }_{ooo}),`$ (50)
$`\widehat{\delta }_{Nl,Nm,n}`$ $`=`$ $`(\widehat{\delta }_{eee}\widehat{\delta }_{eoo}\widehat{\delta }_{oeo}+\widehat{\delta }_{ooe})+i(\widehat{\delta }_{eeo}\widehat{\delta }_{eoe}\widehat{\delta }_{oee}+\widehat{\delta }_{ooo}),`$ (51)
$`\widehat{\delta }_{Nl,m,Nn}`$ $`=`$ $`(\widehat{\delta }_{eee}\widehat{\delta }_{eoo}+\widehat{\delta }_{oeo}\widehat{\delta }_{ooe})+i(\widehat{\delta }_{eeo}+\widehat{\delta }_{eoe}\widehat{\delta }_{oee}+\widehat{\delta }_{ooo}),`$ (52)
$`\widehat{\delta }_{l,Nm,Nn}`$ $`=`$ $`(\widehat{\delta }_{eee}+\widehat{\delta }_{eoo}\widehat{\delta }_{oeo}\widehat{\delta }_{ooe})+i(\widehat{\delta }_{eeo}\widehat{\delta }_{eoe}+\widehat{\delta }_{oee}+\widehat{\delta }_{ooo}),`$ (53)
$`\widehat{\delta }_{Nl,Nm,Nn}`$ $`=`$ $`(\widehat{\delta }_{eee}+\widehat{\delta }_{eoo}+\widehat{\delta }_{oeo}+\widehat{\delta }_{ooe})+i(\widehat{\delta }_{eeo}\widehat{\delta }_{eoe}\widehat{\delta }_{oee}\widehat{\delta }_{ooo}).`$ (54)
Hence, 8 different, but related harmonics can be represented by various combinations of 8 numbers: $`\widehat{\delta }_{eee}`$, $`\widehat{\delta }_{eeo}`$, $`\widehat{\delta }_{eoe}`$, $`\widehat{\delta }_{oee}`$, $`\widehat{\delta }_{eoo}`$, $`\widehat{\delta }_{oeo}`$, $`\widehat{\delta }_{ooe}`$, and $`\widehat{\delta }_{ooo}`$. This implies that the Fourier transform of a real field defined on a grid $`N\times N\times N`$ can be represented by $`N^3`$ real numbers stored on a similar grid, even though the Fourier transform $`\widehat{\delta }(๐ค`$) is complex. Notice also that the 8 โrelatedโ modes are located, in $`k`$-space, at the verticies of a rectangular box centered on the center of the $`k`$-space grid ($`l,m,n=N/2`$). This is illustrated in Figure 1.
From equations (37) and (48)โ(54), we see that related modes form 4 pairs of complex conjugates:
$`\widehat{\delta }_{lmn}`$ $`=`$ $`\widehat{\delta }_{Nl,Nm,Nn}^{},`$ (55)
$`\widehat{\delta }_{lm,Nn}`$ $`=`$ $`\widehat{\delta }_{Nl,Nm,n}^{},`$ (56)
$`\widehat{\delta }_{l,Nm,n}`$ $`=`$ $`\widehat{\delta }_{Nl,m,Nn}^{},`$ (57)
$`\widehat{\delta }_{l,Nm,Nn}`$ $`=`$ $`\widehat{\delta }_{Nl,mn}^{}.`$ (58)
### 3.1 General Case
Consider first the modes for which the indicies $`l`$, $`m`$, and $`n`$ are neither 0 nor $`N/2`$. In equations (55)โ(58), the amplitudes $`\widehat{\delta }`$ are provided by the power spectrum. From equation (27), we get
$$|\widehat{\delta }_๐ค^{\mathrm{disc}}|=N^3\left[\frac{P(k)}{V_{\mathrm{box}}}\right]^{1/2}.$$
(59)
Equation (59) provides the correct normalization of the power spectrum. However, there are two problems with this equation. First, it provides no mean of determining the phases of the complex numbers $`\widehat{\delta }_๐ค^{\mathrm{disc}}`$, and second, choosing random phases would result in a field that is not Gaussian. In a truly Gaussian field, equation (59) is only valid in a statistical sense. The correct approach is to compute the real and imaginary parts of $`\widehat{\delta }_๐ค^{\mathrm{disc}}`$ independently, by drawing them from a Gaussian distribution,
$`\mathrm{Re}\widehat{\delta }_๐ค^{\mathrm{disc}}`$ $`=`$ $`G_1(0,1)N^3\left[{\displaystyle \frac{P(k)}{2V_{\mathrm{box}}}}\right]^{1/2},`$ (60)
$`\mathrm{Im}\widehat{\delta }_๐ค^{\mathrm{disc}}`$ $`=`$ $`G_2(0,1)N^3\left[{\displaystyle \frac{P(k)}{2V_{\mathrm{box}}}}\right]^{1/2},`$ (61)
where $`G_1(0,1)`$ and $`G_2(0,1)`$ are random numbers drawn from a Gaussian distribution with mean 0 and standard deviation 1. This ensures that the resulting field is Gaussian.
Equations (60) and (61) provide us with the left-hand-sides of equations (55)โ(58). By equating these expressions with equations (37), (49), (50), and (53), and considering the real and imaginary parts separately, we get 8 equations,
$`\widehat{\delta }_{eee}+\widehat{\delta }_{eoo}+\widehat{\delta }_{oeo}+\widehat{\delta }_{ooe}`$ $`=`$ $`R_1,`$ (62)
$`\widehat{\delta }_{eeo}+\widehat{\delta }_{eoe}+\widehat{\delta }_{oee}+\widehat{\delta }_{ooo}`$ $`=`$ $`I_1,`$ (63)
$`\widehat{\delta }_{eee}\widehat{\delta }_{eoo}\widehat{\delta }_{oeo}+\widehat{\delta }_{ooe}`$ $`=`$ $`R_2,`$ (64)
$`\widehat{\delta }_{eeo}+\widehat{\delta }_{eoe}+\widehat{\delta }_{oee}\widehat{\delta }_{ooo}`$ $`=`$ $`I_2,`$ (65)
$`\widehat{\delta }_{eee}\widehat{\delta }_{eoo}+\widehat{\delta }_{oeo}\widehat{\delta }_{ooe}`$ $`=`$ $`R_3,`$ (66)
$`\widehat{\delta }_{eeo}\widehat{\delta }_{eoe}+\widehat{\delta }_{oee}\widehat{\delta }_{ooo}`$ $`=`$ $`I_3,`$ (67)
$`\widehat{\delta }_{eee}+\widehat{\delta }_{eoo}\widehat{\delta }_{oeo}\widehat{\delta }_{ooe}`$ $`=`$ $`R_4,`$ (68)
$`\widehat{\delta }_{eeo}\widehat{\delta }_{eoe}+\widehat{\delta }_{oee}+\widehat{\delta }_{ooo}`$ $`=`$ $`I_4,`$ (69)
where
$`R_1`$ $``$ $`\mathrm{Re}\widehat{\delta }_{lmn},`$ (70)
$`I_1`$ $``$ $`\mathrm{Im}\widehat{\delta }_{lmn},`$ (71)
$`R_2`$ $``$ $`\mathrm{Re}\widehat{\delta }_{lm,Nn},`$ (72)
$`I_2`$ $``$ $`\mathrm{Im}\widehat{\delta }_{lm,Nn},`$ (73)
$`R_3`$ $``$ $`\mathrm{Re}\widehat{\delta }_{l,Nm,n},`$ (74)
$`I_3`$ $``$ $`\mathrm{Im}\widehat{\delta }_{l,Nm,n},`$ (75)
$`R_4`$ $``$ $`\mathrm{Re}\widehat{\delta }_{l,Nm,Nn},`$ (76)
$`I_4`$ $``$ $`\mathrm{Im}\widehat{\delta }_{l,Nm,Nn}.`$ (77)
Altogether, we get two separate systems of 4 equations and 4 unknowns. Written in matrix form, these systems are:
$$๐\left[\begin{array}{c}\widehat{\delta }_{eee}\\ \widehat{\delta }_{eoo}\\ \widehat{\delta }_{oeo}\\ \widehat{\delta }_{ooe}\end{array}\right]=\left[\begin{array}{c}R_1\\ R_2\\ R_3\\ R_4\end{array}\right],๐\left[\begin{array}{c}\widehat{\delta }_{eeo}\\ \widehat{\delta }_{eoe}\\ \widehat{\delta }_{oee}\\ \widehat{\delta }_{ooo}\end{array}\right]=\left[\begin{array}{c}I_1\\ I_2\\ I_3\\ I_4\end{array}\right],$$
(78)
where the matrix M is given by
$$๐=\left[\begin{array}{cccc}1& 1& 1& 1\\ 1& 1& 1& 1\\ 1& 1& 1& 1\\ 1& 1& 1& 1\end{array}\right].$$
(79)
The inverse of the matrix $`๐`$ is simply $`๐^1=๐/4`$. Hence, the $`\widehat{\delta }`$โs are given by
$$\left[\begin{array}{c}\widehat{\delta }_{eee}\\ \widehat{\delta }_{eoo}\\ \widehat{\delta }_{oeo}\\ \widehat{\delta }_{ooe}\end{array}\right]=\frac{๐}{4}\left[\begin{array}{c}R_1\\ R_2\\ R_3\\ R_4\end{array}\right],\left[\begin{array}{c}\widehat{\delta }_{eeo}\\ \widehat{\delta }_{eoe}\\ \widehat{\delta }_{oee}\\ \widehat{\delta }_{ooo}\end{array}\right]=\frac{๐}{4}\left[\begin{array}{c}I_1\\ I_2\\ I_3\\ I_4\end{array}\right],$$
(80)
### 3.2 On a Face
Consider now the case where one of the indicies, say $`l`$, is equal to either 0 or $`N/2`$. These cases correspond to values of $`๐ค`$ located on a face of the first octant in $`k`$-space. The problem becomes simpler. With $`l=0`$, equation (36) reduces to
$`\widehat{\delta }_{0mn}`$ $`=`$ $`{\displaystyle \underset{\alpha ,\beta ,\gamma =0}{\overset{N1}{}}}\delta _{\alpha \beta \gamma }e^{2\pi im\beta /N}e^{2\pi in\gamma /N}`$ (81)
$`=`$ $`{\displaystyle \underset{\alpha ,\beta ,\gamma =0}{\overset{N1}{}}}\delta _{\alpha \beta \gamma }\left(\mathrm{cos}{\displaystyle \frac{2\pi m\beta }{N}}+i\mathrm{sin}{\displaystyle \frac{2\pi m\beta }{N}}\right)\left(\mathrm{cos}{\displaystyle \frac{2\pi n\gamma }{N}}+i\mathrm{sin}{\displaystyle \frac{2\pi n\gamma }{N}}\right).`$
After expansion, this expression becomes
$$\widehat{\delta }_{0mn}=(\widehat{\delta }_{ee}+\widehat{\delta }_{oo})+i(\widehat{\delta }_{eo}+\widehat{\delta }_{oe}),$$
(82)
where
$`\widehat{\delta }_{ee}`$ $`=`$ $`{\displaystyle \underset{\alpha ,\beta ,\gamma =0}{\overset{N1}{}}}\delta _{\alpha \beta \gamma }\mathrm{cos}{\displaystyle \frac{2\pi m\beta }{N}}\mathrm{cos}{\displaystyle \frac{2\pi n\gamma }{N}},`$ (83)
$`\widehat{\delta }_{eo}`$ $`=`$ $`{\displaystyle \underset{\alpha ,\beta ,\gamma =0}{\overset{N1}{}}}\delta _{\alpha \beta \gamma }\mathrm{cos}{\displaystyle \frac{2\pi m\beta }{N}}\mathrm{sin}{\displaystyle \frac{2\pi n\gamma }{N}},`$ (84)
$`\widehat{\delta }_{oe}`$ $`=`$ $`{\displaystyle \underset{\alpha ,\beta ,\gamma =0}{\overset{N1}{}}}\delta _{\alpha \beta \gamma }\mathrm{sin}{\displaystyle \frac{2\pi m\beta }{N}}\mathrm{cos}{\displaystyle \frac{2\pi n\gamma }{N}},`$ (85)
$`\widehat{\delta }_{oo}`$ $`=`$ $`{\displaystyle \underset{\alpha ,\beta ,\gamma =0}{\overset{N1}{}}}\delta _{\alpha \beta \gamma }\mathrm{sin}{\displaystyle \frac{2\pi m\beta }{N}}\mathrm{sin}{\displaystyle \frac{2\pi n\gamma }{N}}.`$ (86)
Consider now the mode $`\widehat{\delta }_{0,Nm,n}`$. We replace $`m`$ by $`Nm`$ in equation (81), and get
$$\widehat{\delta }_{0,Nm,n}=\underset{\alpha ,\beta ,\gamma =0}{\overset{N1}{}}\delta _{\alpha \beta \gamma }e^{2\pi i(Nm)\beta /N}e^{2\pi in\gamma /N}=\underset{\alpha ,\beta ,\gamma =0}{\overset{N1}{}}\delta _{\alpha \beta \gamma }e^{2\pi i\beta }e^{2\pi im\beta /N}e^{2\pi in\gamma /N}.$$
(87)
Since $`\beta `$ in an integer, the first exponential is always unity, and equation (87) reduces to
$$\widehat{\delta }_{0,Nm,n}=\underset{\alpha ,\beta ,\gamma =0}{\overset{N1}{}}\delta _{\alpha \beta \gamma }\left(\mathrm{cos}\frac{2\pi m\beta }{N}i\mathrm{sin}\frac{2\pi m\beta }{N}\right)\left(\mathrm{cos}\frac{2\pi n\gamma }{N}+i\mathrm{sin}\frac{2\pi n\gamma }{N}\right).$$
(88)
Comparing equations (81) and (88), the only difference is a change of sign of the first sine function. In equations (83)โ(86), that sine appears only in the $`\widehat{\delta }`$โs for which the first subscript is โ$`o`$.โ Hence, these $`\widehat{\delta }`$โs will change sign, and equation (82) will becomes
$$\widehat{\delta }_{0,Nm,n}=(\widehat{\delta }_{ee}\widehat{\delta }_{oo})+i(\widehat{\delta }_{eo}\widehat{\delta }_{oe}),$$
(89)
Similarly, we can easily show that replacing $`n`$ by $`Nn`$ results in a change of sign of the $`\widehat{\delta }`$โs for which the second subscript is โ$`o`$.โ Hence, we get
$`\widehat{\delta }_{0m,Nn}`$ $`=`$ $`(\widehat{\delta }_{ee}\widehat{\delta }_{oo})+i(\widehat{\delta }_{eo}+\widehat{\delta }_{oe}),`$ (90)
$`\widehat{\delta }_{0,Nm,Nn}`$ $`=`$ $`(\widehat{\delta }_{ee}+\widehat{\delta }_{oo})+i(\widehat{\delta }_{eo}\widehat{\delta }_{oe}).`$ (91)
These 4 modes are located at the verticies of a rectangle in $`k`$-space, as shown in Figure 2. They form two pairs of complex conjugates,
$`\widehat{\delta }_{0mn}`$ $`=`$ $`\widehat{\delta }_{0,Nm,Nn}^{},`$ (92)
$`\widehat{\delta }_{0m,Nn}`$ $`=`$ $`\widehat{\delta }_{0,Nm,n}^{}.`$ (93)
By equating these expressions with equations (82) and (90), and considering the real and imaginary parts separately, we get 4 equations,
$`\widehat{\delta }_{ee}+\widehat{\delta }_{oo}`$ $`=`$ $`R_1,`$ (94)
$`\widehat{\delta }_{eo}+\widehat{\delta }_{oe}`$ $`=`$ $`I_1,`$ (95)
$`\widehat{\delta }_{ee}\widehat{\delta }_{oo}`$ $`=`$ $`R_2,`$ (96)
$`\widehat{\delta }_{eo}+\widehat{\delta }_{oe}`$ $`=`$ $`I_2,`$ (97)
where
$`R_1`$ $``$ $`\mathrm{Re}\widehat{\delta }_{0mn},`$ (98)
$`I_1`$ $``$ $`\mathrm{Im}\widehat{\delta }_{0mn},`$ (99)
$`R_2`$ $``$ $`\mathrm{Re}\widehat{\delta }_{0m,Nn},`$ (100)
$`I_2`$ $``$ $`\mathrm{Im}\widehat{\delta }_{0m,Nn}.`$ (101)
Again, the numbers $`R_1`$, $`I_1`$, $`R_2`$, and $`I_2`$ are determined from equations (60) and (61). The solutions are
$`\widehat{\delta }_{ee}`$ $`=`$ $`{\displaystyle \frac{R_1+R_2}{2}},`$ (102)
$`\widehat{\delta }_{oo}`$ $`=`$ $`{\displaystyle \frac{R_1R_2}{2}},`$ (103)
$`\widehat{\delta }_{eo}`$ $`=`$ $`{\displaystyle \frac{I_1I_2}{2}},`$ (104)
$`\widehat{\delta }_{oe}`$ $`=`$ $`{\displaystyle \frac{I_1+I_2}{2}}.`$ (105)
Consider now the case $`l=N/2`$. Equation (36) reduces to
$`\widehat{\delta }_{N/2,mn}`$ $`=`$ $`{\displaystyle \underset{\alpha ,\beta ,\gamma =0}{\overset{N1}{}}}\delta _{\alpha \beta \gamma }e^{\pi il\alpha }e^{2\pi im\beta /N}e^{2\pi in\gamma /N}`$ (106)
$`=`$ $`{\displaystyle \underset{\alpha ,\beta ,\gamma =0}{\overset{N1}{}}}\delta _{\alpha \beta \gamma }(1)^\alpha \left(\mathrm{cos}{\displaystyle \frac{2\pi m\beta }{N}}+i\mathrm{sin}{\displaystyle \frac{2\pi m\beta }{N}}\right)\left(\mathrm{cos}{\displaystyle \frac{2\pi n\gamma }{N}}+i\mathrm{sin}{\displaystyle \frac{2\pi n\gamma }{N}}\right).`$
After expansion, this expression becomes
$$\widehat{\delta }_{0mn}=(\widehat{\delta }_{ee}+\widehat{\delta }_{oo})+i(\widehat{\delta }_{eo}+\widehat{\delta }_{oe}),$$
(107)
where
$`\widehat{\delta }_{ee}`$ $`=`$ $`{\displaystyle \underset{\alpha ,\beta ,\gamma =0}{\overset{N1}{}}}\delta _{\alpha \beta \gamma }(1)^\alpha \mathrm{cos}{\displaystyle \frac{2\pi m\beta }{N}}\mathrm{cos}{\displaystyle \frac{2\pi n\gamma }{N}},`$ (108)
$`\widehat{\delta }_{eo}`$ $`=`$ $`{\displaystyle \underset{\alpha ,\beta ,\gamma =0}{\overset{N1}{}}}\delta _{\alpha \beta \gamma }(1)^\alpha \mathrm{cos}{\displaystyle \frac{2\pi m\beta }{N}}\mathrm{sin}{\displaystyle \frac{2\pi n\gamma }{N}},`$ (109)
$`\widehat{\delta }_{oe}`$ $`=`$ $`{\displaystyle \underset{\alpha ,\beta ,\gamma =0}{\overset{N1}{}}}\delta _{\alpha \beta \gamma }(1)^\alpha \mathrm{sin}{\displaystyle \frac{2\pi m\beta }{N}}\mathrm{cos}{\displaystyle \frac{2\pi n\gamma }{N}},`$ (110)
$`\widehat{\delta }_{oo}`$ $`=`$ $`{\displaystyle \underset{\alpha ,\beta ,\gamma =0}{\overset{N1}{}}}\delta _{\alpha \beta \gamma }(1)^\alpha \mathrm{sin}{\displaystyle \frac{2\pi m\beta }{N}}\mathrm{sin}{\displaystyle \frac{2\pi n\gamma }{N}}.`$ (111)
We find the same expressions as for the case $`l=0`$, the only differences being the extra factor of $`(1)^\alpha `$ in the definitions of the $`\widehat{\delta }`$โs. Hence, the solutions (102)โ(105) are still valid in this case. These modes are shown in Figure 3.
We have only considered the cases when the first index, $`l`$, is either 0 or $`N/2`$, but these results can be generalized to the other indicies $`m`$ and $`n`$ as well, since the entire problem has cubic symmetry.
### 3.3 On an Edge
Consider now the case when two of the indicies, say $`l`$ and $`m`$, are equal to either 0 or $`N/2`$. These cases corresponds to values of $`๐ค`$ located on an edge of the first octant in $`k`$-space. With $`l=m=0`$, equation (36) reduces to
$$\widehat{\delta }_{00n}=\underset{\alpha ,\beta ,\gamma =0}{\overset{N1}{}}\delta _{\alpha \beta \gamma }e^{2\pi in\gamma /N}=\underset{\alpha ,\beta ,\gamma =0}{\overset{N1}{}}\delta _{\alpha \beta \gamma }\left(\mathrm{cos}\frac{2\pi in\gamma }{N}+i\mathrm{sin}\frac{2\pi in\gamma }{N}\right).$$
(112)
This expression becomes
$$\widehat{\delta }_{00n}=\widehat{\delta }_e+i\widehat{\delta }_o,$$
(113)
where
$`\widehat{\delta }_e`$ $`=`$ $`{\displaystyle \underset{\alpha ,\beta ,\gamma =0}{\overset{N1}{}}}\delta _{\alpha \beta \gamma }\mathrm{cos}{\displaystyle \frac{2\pi n\gamma }{N}},`$ (114)
$`\widehat{\delta }_o`$ $`=`$ $`{\displaystyle \underset{\alpha ,\beta ,\gamma =0}{\overset{N1}{}}}\delta _{\alpha \beta \gamma }\mathrm{sin}{\displaystyle \frac{2\pi n\gamma }{N}}.`$ (115)
Consider now the mode $`\widehat{\delta }_{00,Nn}`$. We replace $`n`$ by $`Nn`$ in equation (112), and get
$$\widehat{\delta }_{00,Nn}=\underset{\alpha ,\beta ,\gamma =0}{\overset{N1}{}}\delta _{\alpha \beta \gamma }e^{2\pi i(Nn)\gamma /N}=\underset{\alpha ,\beta ,\gamma =0}{\overset{N1}{}}\delta _{\alpha \beta \gamma }e^{2\pi i\gamma }e^{2\pi in\gamma /N}.$$
(116)
Since $`\gamma `$ in an integer, the first exponential is always unity, and equation (116) reduces to
$$\widehat{\delta }_{0,0,Nn}=\underset{\alpha ,\beta ,\gamma =0}{\overset{N1}{}}\delta _{\alpha \beta \gamma }\left(\mathrm{cos}\frac{2\pi n\gamma }{N}i\mathrm{sin}\frac{2\pi n\gamma }{N}\right).$$
(117)
Comparing equations (112) and (117), the only difference is a change of sign of the sine function. In equations (114) and (115), that sine appears only in the expression for $`\widehat{\delta }_o`$. Hence, equation (113) becomes
$$\widehat{\delta }_{00,Nn}=\widehat{\delta }_ei\widehat{\delta }_o,$$
(118)
These two modes are complex conjugates,
$`\widehat{\delta }_{00n}`$ $`=`$ $`\widehat{\delta }_{00,Nn}^{}.`$ (119)
They are shown in Figure 4. By equating equation (113) and (119), and considering the real and imaginary parts separately, we get
$`\widehat{\delta }_e`$ $`=`$ $`R_1,`$ (120)
$`\widehat{\delta }_o`$ $`=`$ $`I_1,`$ (121)
where
$`R_1`$ $``$ $`\mathrm{Re}\widehat{\delta }_{00n},`$ (122)
$`I_1`$ $``$ $`\mathrm{Im}\widehat{\delta }_{00n}.`$ (123)
The numbers $`R_1`$ and $`I_1`$ are determined from equations (60) and (61).
Consider now the case $`l=N/2`$, $`m=0`$, the case $`l=0`$, $`m=N/2`$, and the case $`l=m=N/2`$. We can easily show that the solutions (120) and (121) are still valid, using the same approach as in ยง3.2. Again, the only difference will be extra factors of $`(1)^\alpha `$ in the definitions of $`\widehat{\delta }_e`$ and $`\widehat{\delta }_o`$. These modes are shown in Figure 5. Using the cubic symmetry, we can then show that the solutions (120) and (121) applies to all cases for which two of the three indicies $`l`$, $`m`$, $`n`$ are equal to 0 or $`N/2`$ (12 combinations).
### 3.4 In a Corner
Finally, we consider the cases when all indicies are equal to 0 or $`N/2`$. These cases corresponds to values of $`๐ค`$ located in a corner of the first octant in $`k`$-space. With $`l=m=n=0`$, equation (36) reduces to
$$\widehat{\delta }_{000}=\underset{\alpha ,\beta ,\gamma =0}{\overset{N1}{}}\delta _{\alpha \beta \gamma }\delta _u,$$
(124)
where the subscript $`u`$ stands for โuniform,โ since the mode 000 corresponds to a null wavenumber, or an infinite wavelength. Notice that since $`\delta _{\alpha \beta \gamma }`$ is real, $`\delta _u`$ is real as well. Hence, for this mode, there is no imaginary part, and $`\delta _u`$ is determined form equation (60). This mode is shown in Figure 6.
Consider now the case $`l=N/2`$, $`m=0`$, $`n=0`$. Again, we can easily show that the solution (124) is still valid, using the same approach as in ยงยง3.2 and 3.3. The only difference will be an extra factor of $`(1)^\alpha `$ in the expression for $`\delta _u`$. Using the cubic symmetry, we can then show that the solution (124) applies to all cases for which the three indicies $`l`$, $`m`$, $`n`$ are equal to 0 or $`N/2`$ (8 combinations).
Notice that there is a fundamental difference between the mode $`\widehat{\delta }_{000}`$ and the other 7 modes, $`\widehat{\delta }_{00,N/2}`$, $`\widehat{\delta }_{0,N/2,0}`$, $`\mathrm{}`$, $`\widehat{\delta }_{N/2,N/2,N/2}`$, shown in Figures 6 and 7, respectively. The mode $`\widehat{\delta }_{000}`$ represents a perturbation of infinite wavelength, that is, a constant. Clearly that constant must be zero, otherwise the mean value of $`\delta (๐ซ)`$ integrated over the entire volume would be nonzero, and this would violate the assumption that the volume contains a fair sample of the universe. Therefore, for the mode $`\widehat{\delta }_{000}`$, and that mode only, we do not compute the amplitude $`|\widehat{\delta }_{000}|`$ from the power spectrum, but instead set that amplitude equal to zero.
### 3.5 Putting it All Together
We can now count the number of independent quantities necessary to represent all the density harmonics. Consider first the modes $`\widehat{\delta }_{lmn}`$ for which the indicies are neither 0 nor $`N/2`$. Excluding these values, each index can take $`N2`$ different values, which gives us $`(N2)^3`$ modes. As we showed in ยง3.1, these modes come in groups of 8, and within each group the complex values of the harmonics can be expressed as combinations of 8 real numbers $`\widehat{\delta }_{eee}`$, $`\widehat{\delta }_{eeo}`$, $`\mathrm{}`$, $`\widehat{\delta }_{ooo}`$. Hence, it takes a total of $`(N2)^3`$ real numbers to represent these modes.
Consider now the modes for which one of the indicies is equal to 0 or $`N/2`$. There are 6 possibilities, and for each one, the remaining two indicies can take $`N2`$ values each (all values except 0 and $`N/2`$). This gives us $`6(N2)^2`$ modes. As we showed in ยง3.2, these modes come in groups of four, and within each group the complex values of the harmonics can be expressed as combinations of four real numbers $`\widehat{\delta }_{ee}`$, $`\widehat{\delta }_{eo}`$, $`\widehat{\delta }_{oe}`$, and $`\widehat{\delta }_{oo}`$. Hence, it takes a total of $`6(N2)^2`$ real numbers to represent these modes.
Next, consider the modes for which two of the indicies are equal to 0 or $`N/2`$. There are 12 possibilities, and for each one, the remaining index can take $`N2`$ values (all values except 0 and $`N/2`$). This gives us $`12(N2)`$ modes. As we showed in ยง3.3, the amplitude these modes form complex conjugates pairs, and each pair can be represented by two real numbers $`\widehat{\delta }_e`$, and $`\widehat{\delta }_o`$. Hence, it takes a total of $`12(N2)`$ real numbers to represent these modes.
Finally, consider the modes for which all three indicies are equal to 0 or $`N/2`$. There are 8 such modes. As we showed in ยง3.4, these modes are real, hence it takes 8 real numbers to represent them. The total number of variables necessary to represent all the density harmonics is therefore
$$(N2)^3+6(N2)^2+12(n2)+8=N^3.$$
(125)
It takes $`N^3`$ numbers to represent the Fourier transform of a cube $`N\times N\times N`$ of real numbers, in spite of the fact that the Fourier transform is complex. Indeed, it is common for Fast Fourier Transform (FFT) subroutines to take a tridimensional array of number and to overwrite that array with the Fourier transform. This is the case for the FFT subroutines of Numerical Recipes (Press et al, 1992). When these subroutines compute the Fourier transform of a cube $`N\times N\times N`$, the results are written in the same cube, and the actual numbers stored in the cube are precisely the $`\widehat{\delta }_{xxx}`$โs, $`\widehat{\delta }_{xx}`$โs, $`\widehat{\delta }_x`$โs, and $`\delta _u`$โs derived in this section. Hence it is possible to generate the density harmonics directly in $`k`$-space, using the above formulae, store these numbers at the appropriate locations inside an $`N\times N\times N`$ cubic array, and then invoke the Numerical Recipes inverse FFT subroutines to generate the density field.
The Numerical Recipes convention for storing the density harmonics is the following: Consider a 3D array A, with indicies running from 0 to $`N1`$. We loop over all modes located in the first octant in $`k`$-space: $`0l,m,nN/2`$.
1. For modes with $`0<l,m,n<N/2`$ (the general case), the numbers $`\delta _{eee}`$, $`\delta _{eeo}`$, $`\mathrm{}`$, $`\delta _{ooo}`$ are stored in a $`2\times 2\times 2`$ cube located at A(2l,2m,2n), A(2l,2m,2n+1), $`\mathrm{}`$, A(2l+1,2m+1,2n+1).
2. For modes with $`l=0`$, the numbers $`\delta _{ee}`$, $`\delta _{eo}`$, $`\delta _{oe}`$, $`\delta _{oo}`$ are stored at A(0,2m,2n), A(0,2m,2n+1), A(0,2m+1,2n), and A(0,2m+1,2n+1), respectively. For modes with $`l=N/2`$, they are stored at A(1,2m,2n), A(1,2m,2n+1), A(1,2m+1,2n), and A(1,2m+1,2n+1), respectively. This is easily generalized to the other faces ($`m=0`$, $`m=N/2`$, $`n=0`$, $`n=N/2`$).
3. For modes with $`l=m=0`$, the numbers $`\delta _e`$ and $`\delta _o`$ are stored at A(0,0,2n) and A(0,0,2n+1), respectively. For modes with $`l=0`$, $`m=N/2`$, they are stored are stored at A(0,1,2n) and A(0,1,2n+1), respectively. This is easily generalized to the other edges.
4. The number $`\delta _u`$ is stored in a $`2\times 2\times 2`$ cube located in the corner of the array, at A(0,0,0), A(0,0,1), $`\mathrm{}`$, A(1,1,1). Then, the value of A(0,0,0) is set to 0, since this represents the mode $`(0,0,0)`$.
## 4 CALCULATION OF THE DENSITY FIELD
### 4.1 Eulerian Representation
With the expressions derived in ยง2, we have all the ingredients necessary to compute the density field on a grid. The steps are the following:
1. Choose a particular power spectrum $`P(k)`$, a box size $`L_{\mathrm{box}}`$, and a grid size $`N`$. This determines the fundamental wavenumber $`k_0=2\pi /L_{\mathrm{box}}`$. The allowed modes are given by equation (31).
2. For all modes with $`l,m,n0,N/2`$, group these modes in groups of 8, and for each group, calculate the quantities $`\delta _{eee}`$, $`\delta _{eeo}`$, $`\mathrm{}`$, $`\delta _{ooo}`$ using equations (70)โ(77) and (80). The quantities $`R_1`$, $`I_1`$, $`\mathrm{}`$, $`I_4`$ are determined from equations (60) and (61).
3. For all modes located on a face (one of the indicies $`l,m,n`$ equal to 0 or $`N/2`$), group these modes in groups of four, and for each group, calculate the quantities $`\delta _{ee}`$, $`\delta _{eo}`$, $`\delta _{oe}`$, $`\delta _{oo}`$ using equations (98)โ(101) and (102)โ(105).
4. For all modes located on an edge (two of the indicies $`l,m,n`$ equal to either 0 or $`N/2`$), group these modes in groups of two, and for each group, calculate the quantities $`\delta _e`$ and $`\delta _o`$, using equations (120)โ(101).
5. For the modes located in a corner (all indicies $`l,m,n`$ equal to either 0 or $`N/2`$), calculate the quantity $`\delta _u`$ using equation (124). For the mode $`(l,m,n)=(0,0,0)`$, replace the value of $`\delta _u`$ by 0.
6. Store the quantities $`\delta _{eee}`$, $`\mathrm{}`$, $`\delta _{ooo}`$, $`\delta _{ee}`$, $`\delta _{eo}`$, $`\delta _{oe}`$, $`\delta _{oo}`$, $`\delta _e`$, $`\delta _o`$, $`\delta _u`$ in a 3D, $`N\times N\times N`$ array, at the proper locations. These depends on the convention used by the FFT subroutines.
7. Compute the inverse FFT of the 3D array. The result will be the density field $`\delta (๐ซ)`$.
### 4.2 Lagrangian Representation
In the Lagrangian representation, the density field is represented by a distribution of equal-mass particles. We start by laying down $`N_p\times N_p\times N_p`$ particles on a cubic grid with grid spacing $`d=L_{\mathrm{box}}/N_p`$ inside the computational volume $`V_{\mathrm{box}}`$. The mass of the particles are given by $`M=\overline{\rho }V_{\mathrm{box}}/N_p^3`$, such that the mean density inside the box is equal to the mean background density $`\overline{\rho }`$. The particles are then displaced in order to represent the initial density field. This approach is valid only in the linear regime, defined by $`|\delta (๐ซ)|1`$. In this regime, the particle displacements $`\mathrm{\Delta }๐ซ`$ are significantly smallar than the initial separation $`d`$ between the particles, and can be computed using the Zelโdovich approximation (Zelโdovich, 1970). The displacements are given by
$$\mathrm{\Delta }๐ซ_j=\frac{i}{N^3}\underset{๐ค}{}\frac{\widehat{\delta }_๐ค๐ค}{k^2}e^{i๐ค๐ซ_j},$$
(126)
where $`๐ซ_j`$ is the position of particle $`j`$ before it is displaced, and $`๐ซ_j+\mathrm{\Delta }๐ซ_j`$ is the position after. Notice that the reality condition (5) ensures that $`\mathrm{\Delta }๐ซ_j`$ is real.
This method is straightforward, but can rapidly become unpractical. Equation (126) involves a sum over $`N^3`$ modes, and that sum must be performed for each of the $`N_p^3`$ particles. Since in typical cosmological simulations $`N_p`$ is chosen to be $`N/2`$, the number of operations scales like $`N^6`$. If it takes 5 minutes to set up initial conditions with $`64^3`$ particles, it will take 5.3 hours for $`128^3`$ particles, 2 weeks for $`256^3`$ particles, and 2.5 years for $`512^3`$ particles! An alternative method, which scales like $`N^3`$, was proposed by Efstathiou et al. (1985). Essentially, this approach uses the fact that the displacement of each particle is proportional to its peculiar acceleration, that can be calculated with a N-body simulation algorithm such as PM (Particle-Mesh) or $`\mathrm{P}^3\mathrm{M}`$ (Particle-Particle/Particle-Mesh). With this method, the number of operations scales roughly as $`N^3`$. I refer the reader to Hockney & Eastwood (1981) and Efstathiou et al. (1985) for details. It is worth noting that, unlike equation (126), the approach of Efstathiou et al. (1985) is approximative, and the high-frequency modes, near the Nyquist frequency $`k=(N_p/2)k_0`$, are often poorly represented by the particle distribution.
## 5 FILTERING
Our next task in to filter the density field at some scale $`s`$. The choice of scale must obey two conditions: $`s\mathrm{\Delta }`$ and $`sL_{\mathrm{box}}`$. The first condition is required by the discreteness of the grid and the second by the assumption of periodic boundary conditions. The filtered density field $`\delta _s(๐ซ)`$ at scale $`s`$ is given by
$$\delta _s(๐ซ)=_{V_{\mathrm{box}}}\delta (๐ซ^{})K_s(๐ซ๐ซ^{})d^3r^{},$$
(127)
where $`K_s`$ is the filter function. We will use a Gaussian filter given by
$$K_s(๐ฑ)=\frac{e^{x^2/2s^2}}{(2\pi )^{3/2}s^3}.$$
(128)
This filter function satisfied the normalization condition,
$$_{V_{\mathrm{box}}}K_s(๐ซ)d^3r=1,$$
(129)
as long as $`sL_{\mathrm{box}}`$.
It is well known that filtering in real space is equivalent to a multiplication in $`k`$-space. However, it is useful to redo the derivation, to ensure that we have all the correct factors of $`2\pi `$, $`N^3`$, and so on. First, we express the filter as an inverse Fourier transform,
$$K_s(๐ซ๐ซ^{})=\frac{1}{N^3}\underset{๐ค^{}}{}\widehat{K}_s(๐ค^{})e^{i๐ค^{}(๐ซ๐ซ^{})}.$$
(130)
We substitute equations (33) and (130) in equation (127), and get
$`\delta _s(๐ซ)`$ $`=`$ $`{\displaystyle \frac{1}{N^6}}{\displaystyle _{V_{\mathrm{box}}}}{\displaystyle \underset{๐ค}{}}\widehat{\delta }(๐ค)e^{i๐ค๐ซ}{\displaystyle \underset{๐ค^{}}{}}\widehat{K}_s(๐ค^{})e^{i๐ค^{}(๐ซ๐ซ^{})}d^3r^{}`$ (131)
$`=`$ $`{\displaystyle \frac{1}{N^6}}{\displaystyle \underset{๐ค}{}}{\displaystyle \underset{๐ค^{}}{}}\widehat{\delta }(๐ค)\widehat{K}_s(๐ค^{})e^{i๐ค๐ซ}{\displaystyle _{V_{\mathrm{box}}}}e^{i(๐ค^{}๐ค)๐ซ^{}}d^3r^{}.`$
The integral is equal to $`V_{\mathrm{box}}\delta _{๐ค,๐ค^{}}`$ (see eq. ), and we use the Kronecker $`\delta `$ to eliminate the sum over $`๐ค^{}`$. We get
$$\delta _s(๐ซ)=\frac{V_{\mathrm{box}}}{N^6}\underset{๐ค}{}\widehat{\delta }(๐ค)\widehat{K}_s(๐ค)e^{i๐ค๐ซ}.$$
(133)
We now need an expression for $`\widehat{K}_s(๐ค)`$. This function is the Fourier transform of the filter,
$$\widehat{K}_s(๐ค)=\underset{๐ฑ}{}K_s(๐ฑ)e^{i๐ค๐ฑ}=\frac{1}{(2\pi )^{3/2}s^3}\underset{๐ฑ}{}e^{x^2/2s^2}e^{i๐ค๐ฑ}.$$
(134)
We rewrite this expression as
$$\widehat{K}_s(๐ค)=\frac{1}{(2\pi )^{3/2}s^3}\left(\frac{N^3}{V_{\mathrm{box}}}\right)\underset{๐ฑ}{}e^{x^2/2s^2}e^{i๐ค๐ฑ}\left(\frac{V_{\mathrm{box}}}{N^3}\right).$$
(135)
The factor $`V_{\mathrm{box}}/N^3=\mathrm{\Delta }^3`$ represents the volume element around each point $`๐ฑ`$ in the $`N\times N\times N`$ grid. Since we assume $`s\mathrm{\Delta }`$, we can approximate the sum as an integral over the volume of the box (or, equivalently, regard the sum as a numerical approximation for the integral). Hence,
$$\widehat{K}_s(๐ค)=\frac{1}{(2\pi )^{3/2}s^3}\left(\frac{N^3}{V_{\mathrm{box}}}\right)_{V_{\mathrm{box}}}e^{x^2/2s^2}e^{i๐ค๐ฑ}d^3x.$$
(136)
Since we assume periodic boundary conditions, we are free to locate the origin anywhere inside the box. For instance, we can locate it in the center of the box. Since $`sL_{\mathrm{box}}`$, the integrant becomes negligible at the edge of the box. We can then extend the integration domain to all space,
$$\widehat{K}_s(๐ค)=\frac{1}{(2\pi )^{3/2}s^3}\left(\frac{N^3}{V_{\mathrm{box}}}\right)_{\text{all space}}e^{x^2/2s^2}e^{i๐ค๐ฑ}d^3x,$$
(137)
where this expression no longer assumes boundary conditions. The integral in equation (137) can be found in any textbook of Fourier transforms,
$$_{\text{all space}}e^{x^2/2s^2}e^{i๐ค๐ฑ}d^3x=(2\pi )^{3/2}s^3e^{(ks)^2/2}.$$
(138)
Hence,
$$\widehat{K}_s(๐ค)=\frac{N^3}{V_{\mathrm{box}}}e^{(ks)^2/2}.$$
(139)
We substitute this expression in equation (133), and get
$$\delta _s(๐ซ)=\frac{1}{N^3}\underset{๐ค}{}\widehat{\delta }(๐ค)e^{(ks)^2/2}e^{i๐ค๐ซ}.$$
(140)
Hence, to obtain a filtered density field, we generate the density harmonics using the method described in ยง3, and then multiply them by the factor $`e^{k^2s^2/2}`$ before taking the inverse Fourier transform.
## 6 SUMMARY
This paper presents in great detail the techniques used for generating Gaussian density fields. These techniques are well-known among experts in cosmological numerical simulations, but the specific details of the implementation are often difficult to find in the literature. Also, the notation tends to vary significantly from one author to another. The consequences is that any new researcher moving into this field has to either spend a great deal of effort rederiving all the technical details, or else rely on existing codes and use them as black boxes. The goal of this document is to improve the situation by presenting in a comprehensive form the basic theory behind the generation of Gaussian random fields.
I am very thankful to Yehuda Hoffman, Patrick McDonald, Matthew Pieri, Cรฉdric Grenon, and Matthew Craig for reading this manuscript and making valuable comments. This work was supported by the Canada Research Chair program and NSERC.
## Appendix A FOURIER TRANSFORM OF THE TOP HAT
Consider the following integral,
$$I=_{\mathrm{sph}(0)}d^3ye^{i๐ค๐ฒ},$$
(A1)
where the domain of integration is a sphere of radius $`R`$ centered at the origin. We consider a spherical coordinate system centered at the origin, with the $`z`$-axis pointing in the direction of $`๐ค`$. Equation (A1) becomes
$$I=_0^{2\pi }๐\varphi _0^\pi ๐\theta _0^R๐y(y^2\mathrm{sin}\theta )e^{iky\mathrm{cos}\theta },$$
(A2)
where $`k|๐ค|`$, $`y|๐ฒ|`$, and $`\theta `$ is the angle between $`๐ค`$ and $`๐ฒ`$. The integrations over $`\varphi `$ and $`\theta `$ are trivial. We get
$$I=2\pi _0^R๐yy^2\left[\frac{e^{iky\mathrm{cos}\theta }}{iky}\right]_0^\pi =2\pi _0^R๐yy^2\left[\frac{e^{iky}e^{iky}}{iky}\right]=4\pi _0^R๐y\frac{y\mathrm{sin}ky}{k}.$$
(A3)
The integral over $`y`$ is now trivial. We get
$$I=\frac{4\pi }{k^3}(\mathrm{sin}kRkR\mathrm{cos}kR)=\frac{4\pi R^3}{u^3}(\mathrm{sin}uu\mathrm{cos}u),$$
(A4)
where $`u=kR`$.
|
warning/0506/cs0506038.html
|
ar5iv
|
text
|
# A Game Theoretic Economics Framework to understanding the Information Security outsourcing Market
## 1 Introduction
Security outsourcing market where firms contract with outside information security vendors to meet their organizational demands has been growing at a double digit rate for the past $`3`$ years, and experts predict that this growth rate will continue through $`2008`$. Compared with the booming of the business, theory of security outsourcing is less developed. In view of this both buyers and MSSPs need to strategically understand the nature of this market.
Information security outsourcing is different from traditional outsourcing because information security is different from durable goods and other services outsourced such as payroll and accounting. As more and more firms automate processes, servers and the networks work like the brains and vessels of a firm. If any core system go down, the cost may be large due to lost data and lost revenue. What makes it worse is that security breaches are irreversible. While defects in manufacturing can be returned or wrong paychecks can be reissued, monetary loss due to down time is gone forever, and lost customer confidence may be hard to gain back. Therefore, while most industries put cost saving as the primary reason they outsource business processes other than security, firms that outsource information security state service quality is their primary motivation. This is supported by a survey by Jeffrey Kaplan published in Business Communication Review ($`2003`$). It is reported that 40.6% of the firms outsource network operations based on concerns for service quality.
Information asymmetry is another reason that firms have concerns outsourcing their security. Since buyers cannot observe and monitor MSSPsโ action, MSSPs, as profit maximizing companies, have an incentive to lower their effort level to reduce cost.
The model we present is a model where buyers and MSSPs engage in a repeated game with infinite horizon where MSSPsโ effort level in not observable to buyers. We show that under this information asymmetry, moral hazard problem will occur. Performance based contracts are recommended to avoid such moral hazard problem.
For comparison, we also provide results under perfect information, where buyers can have all information they need and shirking is not an option for MSSPs. Under the scenario of perfect information, the optimal solution(in terms how the contact is written) is a price-only contract. This solution is called first best because no deadweight loss is incurred under perfect information assumption.
Besides the optimal contract form, we are particularly interested in the effect of transaction cost on market equilibrium price. Transaction cost includes all cost spent on searching for, arguing and executing contracts with MSSPs. We argue in section(3.2) transaction cost can be very high in outsourcing non traditional services such as security because standard rules and procedures have not been established yet. We show that when transaction cost increases, price of security outsourcing will be lowered.
There is a large body of literature on IT outsourcing, including information security outsourcing as a sub-category. Ang and Straub (1998) did an empirical study on the U.S. banking industry and showed IT outsourcing is strongly influenced by the production cost advantage offered by IT service vendors. Transaction cost also influences outsourcing decisions with a much smaller effect. Though their result is based on data of US banking system, this result is probably true in a lot of areas outside the banking system. Based on their result, we will assume decrease in production cost out-weight increase in transaction cost throughout this paper. Lacity and Willcocksโ(1998) use US and UK organizations survey data and provide empirical evidence that the following practices are recommended to achieve cost saving expected: selective outsourcing, senior executives and IT manager make decisions together, invite both internal and external bids, short-term contract, detailed fee-for-service contract. This paper will provide theoretical support for the last practice. Mieghem (1999) builds a game theoretic model on production outsourcing where investment decision has to be made before market demand is revealed. After market demand is revealed, the firmโs production is limited to its investment level, and will use outside production(outsource) to meet excess demand. His paper studies three kinds of contracts 1), price-only contract, 2),incomplete contract and 3), state-dependent contract. He shows that only state-dependent contract is optimal in the sense that it eliminates all decentralizing cost<sup>1</sup><sup>1</sup>1centralized economy system assumes there is a social planner who make decision by pooling all available resources from different firms. Decentralized economy system is one where firms make their own decision using individual resources. It can be shown that outcome of centralized economy weakly dominates outcome of decentralized economy. Difference between the two is decentralization cost. His paper is related to security outsourcing because in security outsourcing, an implicity assumption of centralized economy is that all participants will work diligently. Therefore, with moral hazard problem, decentralization cost is caused by the possibility that MSSPs may shirk. This paper will investigate why state-contingent contract is preferred to non state-contingent contract from a information economics point of view. We argue that state-contingent contract is the optimal contract form when there is moral hazard problem.
The rest of this paper is organized as follows: In Section 2 and 3, we contrast information security outsourcing with other types of outsourcing. Next we set up an outsourcing model with perfect and imperfect information to discuss what optimal contract look like and what is the effect of transaction cost on prices in Section 4. In Section 5, related work on this topic is summarized. We end with a summary and conclusions in Section 6.
## 2 Outsourcing Theory
Outsourcing is defined as โall the subcontracting relationships between firms and the hiring of workers in non-traditional jobsโ (Heshmati 2003). Business Process Outsourcing (BPO), which includes outsourcing of human resources, finance and accounting, procurement, shared services, billing, customer care and so on, is estimated to grow at a 9.5% compound annual rate through 2007 reaching $173 billion by Gartner. IT Outsourcing (ITO) is expected to grow at a compound rate of 7.2% through 2008 reaching $253.1 billion in 2008. Furthermore, Information security outsourcing is predicted to grow from $4.1 billion in 2001 to $9.0 billion in 2006, a compound growth rate of double digits.
Behind this booming of outsourcing, the basic force is โcost efficiencyโ. As markets become more competitive, outsourcing is an essential way firms may reduce costs. By using information security outsourcing, firm only need to pay a fraction of their in-housing cost for outsourced security. Outsourcing can reduce cost either because suppliers has lower input costs and/or larger scale of production as in the case of offshore manufacturing outsourcing; or because the suppliers have expertise or more advanced technology as in payroll and IT outsourcing. However, at the same time of reducing production cost, buyers incur transaction costs searching for, signing, and executing contracts with suppliers. In the case of total outsourcing, when firms keep no in-house production, firms also lose sunk costs<sup>2</sup><sup>2</sup>2Firmโs investment specific to the outsourced process, which can be machines and plants that can only be used to produce the outsourced product or can be money spent on training technicians.
If cost reduction is the only concern for firms, firms will outsource when reduction in production cost exceeds increase in transaction cost. In standardized outsourcing procedures such as payroll and manufacture goods, transaction cost has been reduced as Coase predicted โThis(transaction) cost may be reduced but it will not be eliminated by emergence of specialist$`\mathrm{}`$โ. It is argued that transaction cost is some percentage of the contract value since the larger the project, the greater effort firms will spent on searching for a proper MSSP and the more coordination is needed between firm and MSSP after signing the contract.
The second outsourcing incentive is firms will be able to concentrate on their core competence by outsourcing support/routine functions. For example, although a lot computer companies are based in the U.S., most keyboards are produced in Asia. By outsourcing labor intensive processes to areas that are abundant in labor, firms achieve cost reduction and become more focused on core competence.
Yet another key reason for outsourcing is to obtain higher quality. Outside companies accumulate more experience by specializing in certain processes. They can afford larger investment on R&D to get updated technology and skills and better trained expertise. A large client base also contributes to the quality of goods and services of outside producers and service providers. They gain experience and knowledge by serving varied clients. Consulting, for example, the service providers have professional knowledge that a non-consulting firm can never afford to build by itself.
Argument against production outsourcing concerns unemployment issue as in off-shore outsourcing: while argument against security outsourcing focus on transaction cost control and service quality monitoring. We will analyze these two concerns on information security outsourcing in detail in the following section.
## 3 Security Outsourcing: What is Special?
In spite of all the advantages outsourcing may bring, some people think security should not be outsourced, or firms should be really careful when doing so.
### 3.1 Quality Measurement Difficulty
Security management is an art rather than science where we know how to achieve a best solution; here we do not even know what the best solutions are, nor do MSSPs. A security system can be a very complicated project. People may think that they are safe with firewalls and IDSs. Even so, firms have to decide which firewalls and IDSs to buy, how to allocate limited budget on combination of these devices to reach maximum level of security and how to manage these devices and tune them so that they secure your system enough and do not give too many alerts on harmless behaviors. The bright side is MSSPs are gaining experience on these issues quickly by their devotion and specialization in this area.
However, people argue that it is hard to evaluate products and services of MSSPs both ex ante and ex post. As security outsourcing market becoming prominent over the last few years; a large number of MSSPs emerged from diversified backgrounds. The largest ones include firms formed solely to solve internet security problems such as Counterpane, firms from research and computer production such as IBM, anti virus companies such as Symantec, firms from internet providers such as AT&T and so on. This diversification in background reflects on their diversified product and services making it really hard for the firms to compare and choose from them. (See appendix I for major MSSPs and their products.)
Also, evaluating MSSPsโ products by performance of their products is tricky because the outcome is highly random and can even be misleading. A better secured system may be down because of intensive attacks; systems that ignore patching notices from time to time may go well for a long time. On the other hand, it is not true that the more money spent on security, the fewer bleaches a system will have. Sophisticated hackers are more attracted to systems that are hard to break into.
However, a โbetterโ secured system should be less vulnerable in statistical sense in the long run. This paper will use *expected* performance to evaluate a security system. We assume buyers have access to historical data of MSSPโs service performance, and can generate a distribution of benefit from using security outsourcing.
### 3.2 Effective Cost Reduction?
Based on a survey on IT managers, directors and other decision makers from both firms that outsourced security and those who did not, cost reduction remains their focus.
There is evidence that security outsourcing will reduce production cost. Device management for example, which tunes and monitors firewalls, IDSs and runs vulnerability testing, a security personnel cost$8,000 to $16,000 per month. And to get 24\*7 support, this figure may need to be more than tripled. For the same functions, MSSPs charge between $600 and $4,000. For network monitoring, Counterpane, one of the most successful MSSPs, claims that it only charges a fraction of the money for net management a firm need to spend to do the security in house: โFrom an annualized basis, its going to cost you $1 million to $1.2 million just to look at the sam information we monitor, and our average contract ranges from $40,000 to $150,000 a year โ between 4% and 10% of what it would cost to do yourself $`\mathrm{}`$.
However, although security vendorsโ may provide huge reduction in production cost, transaction cost may be quite high. Since standard measure for security services has not been established and each MSSP uses their featured(different) technology, most of the time it is very hard to do comparison across different MSSPs. This quality measurement difficulty may increase transaction cost potentially.
Also, writing up the contract and decide who is responsible for what kind of losses due to security breaches can be painful. Firms would feel more comfortable if security vendors can take responsibility if losses occur. But it is not always the security vendorโs fault because no matter how well security devices are designed and tuned, there is always probability that the system is broken into. More tricky things can be if security vendors take responsibility for the losses, firms may not play due diligence as they should. Therefore, although this paper is devoted to discussion of MSSPsโ moral hazard behavior, the optimal contract needs to guard against firmsโ moral hazard behavior as well, which may increase transaction cost significantly. Therefore although we will assume that transaction cost is lower than reduction in production cost, effect of transaction cost needs to be further explored.
## 4 The Model
Based on above observation of how security outsourcing is special, We set up the model in the following way.
There are two sides on the security outsourcing market: potential security service buyers (โbuyersโ for short), and security vendors(MSSPs). Vendors and buyers all seek to maximize their individual profit.
Basic assumptions are:
* A1: Vendors are more cost efficient than firms; transaction cost is lower than production cost advantage.
* A2: Services provided by different security vendors are imperfect substitutes<sup>3</sup><sup>3</sup>3imperfect substitutes are goods that are not identical but have similar functions, e.g. lap-top and desk-tops. .
* A3: Buyers do not have moral hazard problem.
In the following three subsections, we show that:
1. With imperfect information, we have moral hazard problem on MSSP side. Optimal contract depends non-trivially on MSSPs performance.
2. With perfect information, optimal contract is a price-only contract.
3. With either perfect information or imperfect information, price is decreasing on transaction cost.
### 4.1 Optimal contract with imperfect information <br>โ Performance based contract
Due to imperfect information, actions of the players are not directly observable. Both MSSPs and security buyers can disobey their promises secretly. In this paper, we focus on how to avoid moral hazard behavior of MSSPs, and assume buyers will always follow the contract as it is. The optimal contract will be such that following the contract is the best choice for both players. We temporarily assume transaction cost is zero in this section.
Our analysis is based on principal-agent problem with infinite horizon following Spear and Srivastava(), where agentโs action is not observable to principal. principal is assumed to be risk neutral<sup>4</sup><sup>4</sup>4A risk neutral player only cares about average payoff.and agent risk averse<sup>5</sup><sup>5</sup>5A risk averse player gets lower utility if variance of his payoff increase. Here, MSSP is agent to principal buyer. We are allowed to assume security buyer is risk neutral because security buyers have access to insurance market and can buy insurance to mitigate risks that MSSPs cannot eliminate. However, the risk neutral assumption is not essential to the result. We can discuss risk averse buyers but it only make the mathematics more complicated without accomplishing anything. So we just keep the simple assumption that buyers are risk neutral.
Denote buyerโs period t benefit(before payment to MSSP) from security outsourcing as $`y_t`$. Because of the random nature of cyber attacks, $`y_t`$ is a random variable. Denote MSSPโs effort level in period t as $`a_t`$, $`a_t[\underset{ยฏ}{a},\overline{a}]`$. Then distribution of security service performance $`y_t`$ is conditional on MSSPโs effort $`a_t`$. Denote the distribution as $`f(y,a_t)`$. $`P_t`$ denotes buyerโs compensation(price) to MSSP in period t. History up to period t is denoted as: $`h_t=\{y_t,y_{t1},\mathrm{},y_0\}`$.
A price contract is composed of MSSPโs effort level and price buyer pays to MSSP: $`\{a_t(h_{t1}),P_t(h_t)\}`$. Notice that MSSPโs period t effort level $`a_t`$ depends only on history up to period t-1, since MSSP has to choose his effort level at beginning of period t before period t benefit $`y_t`$ is realized. Payment to MSSP in period t however depends on the whole performance history.
Let $`u(P_t)\varphi (a_t)`$ be net payoff to MSSP under contract $`\{a_t(h_{t1}),P_t(h_t)\}`$, where $`u(P_t)`$ is MSSPโs utility from payment $`P_t`$ and $`\varphi (a_t)`$ measures cost of working at effort level $`a_t`$. We assume $`u^{}>0`$, $`u^{\prime \prime }<0`$<sup>6</sup><sup>6</sup>6$`u^{\prime \prime }<0`$ comes from risk averse assumption. and $`\varphi ^{}>0`$. History $`h_t`$ evolve recursively by the following probability rule:
$`\pi (h_t|h_{t1})=f(y_t,a_t(h_{t1}))\pi (h_{t1})`$ (1)
Assume buyers and MSSP discount future payoff at same rate $`\rho ,\rho [0,1]`$, then buyer and MSSPโs period t expected payoff are $`(y_tP_t)f(y_t|a_t)๐y_t`$ and $`u(P_t)\varphi (a_t)`$:
Discount all future payoff to period 0, we have buyer and MSSPโs period 0 discounted payoff as:
$`B_t(P_t,a_t)`$ $`=`$ $`{\displaystyle \underset{j=0}{\overset{\mathrm{}}{}}}{\displaystyle \underset{h^{t+j}}{}}\rho ^j[{\displaystyle (y_tP_t)f(y_t,a_t)๐y_t}]\pi (h_{t+j},a_{t+j}|h_t)`$ (2)
$`M_t(P_t,a_t)`$ $`=`$ $`{\displaystyle \underset{j=0}{\overset{\mathrm{}}{}}}{\displaystyle \underset{h^{t+j}}{}}\rho ^j[u(P_t)\varphi (a_t)]\pi (h_{t+j},a_{t+j}|h_t)`$ (3)
Therefore, the maximization problem for security buyer is to choose a sequence of contracts $`\{P_t(y),a_t\}_{t=0}^{\mathrm{}}`$ to maximize discounted expected utility subject to the constraint that MSSP cannot benefit from deviating from the contract:
$`\underset{\{P_t(y)\}_{t=0}^{\mathrm{}},\{a_t\}_{t=0}^{\mathrm{}}}{\mathrm{max}}`$ $`B_t(P_t(y),a_t)`$
st $`M_t(P_t(y),a_t)M_t(P_t(y),\stackrel{~}{a}_t)\stackrel{~}{a}_t[\underset{ยฏ}{a},\overline{a}]`$ (4)
where, constraint in above maximization problem is called the incentive compatibility(IC) constraint. It show that the effort level $`a_t`$ is optimal for MSSP compared to any other possible effort level $`\stackrel{~}{a}_t`$.
Since the above problem has infinitely unknown variables, it is impossible to solve it directly. Instead, we rewrite it in the recursive form.
In the recursive form, principal maximize current periodโs payoff assuming he will behave optimally from next period on. Let $`v`$ denote payoff buyer promised to MSSP this period and $`w(y)`$ denote the promised payoff to MSSP next period. $`K(v)`$ be maximized payoff to buyer when MSSP gets v as promised expected payoff. Hence, $`K(w(y))`$ is buyerโs best possible payoff next period. Then the maximization problem in recursive form is:
$`K(v)`$ $`=`$ $`\underset{P(y),w(y),a}{\mathrm{max}}{\displaystyle [yP(y)+\rho K(w(y))]f(y,a)๐y}`$
st $`{\displaystyle [u(P(y))+\rho w(y)]f(y,a)๐y}\varphi (a)v\text{(PK)}`$ (5)
$`a\mathrm{arg}\mathrm{max}{\displaystyle [u(P(y))+\rho w(y)]f(y,a)๐y}\varphi (a)\text{(IC)}`$
The optimal contract should contain $`\{P(y),w(y),a\}`$. (PK) is short for โpromise keepingโ. It requires that if buyer promised MSSP payoff v, the contract should guarantee expected payoff to MSSP is at least v(equal to v in equilibrium). (IC) constraint is same as in (4).
The (IC) constraint implies the solution $`a`$ should satisfy both the following first order condition and second order condition:
$`(FOC)`$ $`{\displaystyle [u(P(y))+\rho w(y)]f_a(y,a)๐y}\varphi ^{}(a)`$ (6)
$`(SOC)`$ $`{\displaystyle [u(P(y))+\rho w(y)]f_{aa}(y,a)๐y}\varphi ^{\prime \prime }(a)0w(y)`$ (7)
Assumption:
* Convexity of distribution function condition(COFC):
$`F_{aa}0`$ (8)
where $`F(x,a)=_{\mathrm{}}^xf(y,a)๐y`$
Rogerson(1985) shows that when COFC is satisfied, (SOC) is guaranteed. We can use (FOC) to substitute (IC) constraint and get rid of the (SOC).
Let $`\lambda `$ be Lagrangian multiplier on (PK) constraint and $`\mu `$ be the multiplier on (IC)-(FOC) constraint. We have the Lagrangian equation:
$`L`$ $`=`$ $`{\displaystyle [yP(y)+\rho K(w(y))]f(y,a)๐y}`$ (9)
$`+\lambda ({\displaystyle [u(P(y))+\rho w(y)]f(y,a)๐y}\varphi (a)v)`$
$`+\mu ({\displaystyle [u(P(y))+\rho w(y)]f_a(y,a)๐y}\varphi ^{}(a))`$
Take first order conditions w.r.t $`P(y),w(y)`$ and $`a`$, we get the following first order conditions and the envelope condition:
$`\{P(y)\}`$ $`1+\lambda u^{}(P(y))+\mu u^{}(P(y)){\displaystyle \frac{f_a(y,a)}{f(y,a)}}=0`$ (10)
$`\{w(y)\}`$ $`\rho K^{}(w(y))+\rho \lambda +\mu \rho {\displaystyle \frac{f_a(y,a)}{f(y,a)}}=0`$ (11)
$`\{a\}`$ $`{\displaystyle }[yP(y)+\rho P(w(y)]f_a(y,a)dy`$ (12)
$`+\mu [{\displaystyle [u(P(x))+\rho w(y)]f_{aa}(y,a)๐y}\varphi ^{\prime \prime }(a)]=0`$
$`\{ENV\}`$ $`K^{}(v)=\lambda `$ (13)
First order conditions (10) and (11) implies:
$`{\displaystyle \frac{1}{u^{}(P(y))}}=K^{}(w(y))=\lambda +\mu {\displaystyle \frac{f_a(y,a)}{f(y,a)}}`$ (14)
Definition: MLRP(monotone likelihood ratio property)
* Likelihood ratio $`\frac{f_a(y,a)}{f(y,a)}`$ is monotone in $`y`$ or $`\frac{d}{dy}[\frac{f_a(y,a)}{f(y,a)}]0`$. This also implies: $`a>\stackrel{~}{a},y>\stackrel{~}{y},\frac{f(y,a)}{f(\stackrel{~}{y},a)}\frac{f(y,\stackrel{~}{a})}{f(\stackrel{~}{y},\stackrel{~}{a})}`$.
Intuitively, this means at a higher effort level $`a`$, it is more probable to get a higher benefit $`y`$ than at a lower effort level $`\stackrel{~}{a}`$.
Rogerson(1085) shows that when the density function $`f(y,a)`$ has monotone likelihood ratio property, $`\mu `$ the multiplier on (IC) constraint is positive.
When MLRP holds, $`\mu >0`$, equation (14) implies the following results:
$`y\frac{1}{u^{}(P(y))}P(y)`$.
Reason: $`u^{\prime \prime }(P(y))0`$
This result suggests contacts should be performance-based, i.e. payment to MSSP should be higher when benefit from security outsourcing increases and vice versa. And this supports empirical result of Lacity and Willcock(1998).
$`yK^{}(w(y))w(y)`$
Reason: $`K(w(y))`$ is best possible payoff of buyer next period when MSSPโs expected payoff is w(y). Since MSSPโs payoff comes from compensation $`P(y)`$ from buyer, the higher MSSPโs payoff $`w(y)`$ is, the lower buyerโs payoff$`K(w(y))`$ will be.
This result suggest buyer should reward MSSP with higher expected payoff for next period if buyer gets high benefit this period.
$`v\lambda P(y),w(y)`$
Reason: $`v\lambda `$ from the envelope condition (ENV). $`\lambda P(y),w(y)`$ follows from equation (14).
This result shows that if buyer promise MSSP a higher current expected payoff, buyer should increase both current period compensation and next period promised expected payoff.
To sum up, from Result 1 - 3, we suggest that optimal contract under moral hazard should depend on performance in a non-trivial way. And effect of performance is persistently on future compensations. The effect is carried over by promised value $`v`$ and $`w(y)`$ as shown in Result 2 and 3.
### 4.2 Optimal contract with perfect information <br>โ price only contract
With perfect information, buyer can monitor MSSPโs behavior very well. Then MSSP is not able to shirk and moral hazard problem does not exist. In this scenario, Maximization problem of buyer(5) reduces to:
$`K(v)`$ $`=`$ $`\underset{P(y),w(y),a}{\mathrm{max}}{\displaystyle [yP(y)+\rho K(w(y))]f(y,a)๐y}`$
st $`{\displaystyle [u(P(y))+\rho w(y)]f(y,a)๐y}\varphi (a)v\text{(PK)}`$ (15)
Corresponding first order conditions are:
$`\{P(y)\}`$ $`1+\lambda u^{}(P(y))=0`$ (16)
$`\{w(y)\}`$ $`\rho K^{}(w(y))+\rho \lambda =0`$ (17)
$`\{a\}`$ $`{\displaystyle }[yP(y)+\rho P(w(y)]f_a(y,a)dy=0`$ (18)
$`\{ENV\}`$ $`K^{}(v)=\lambda `$ (19)
Equation 16 and (17) imply:
$`{\displaystyle \frac{1}{u^{}(P(y))}}=K^{}(w(y))=\lambda `$ (20)
This suggests that without moral hazard problem, optimal compensation and next period promised value does not depend on this periodโs outcome $`y`$. Constant compensation and promised value would be optimal.
### 4.3 Effect of transaction cost
#### 4.3.1 Effect from game between buyer and MSSP
In this section, we will study how transaction cost affects equilibrium market price. No matter whether buyer has perfect information about MSSPโs effort level or not, existence of transaction cost reduces buyers compensation to MSSP.
As in section(4.1), we use $`P(y)`$ to denote buyerโs compensation to MSSP. Since buyers will also need to pay transaction cost on top of service price, the actual out of pocket price buyers of MSSP face is $`(1+\alpha )P(y)`$, where $`\alpha P(y)`$ is the transaction cost<sup>7</sup><sup>7</sup>7transaction cost is modelled as a percentage of contract value because as the project gets larger, buyer and vendor need to spend more time and money on the negotiation and coordination part \[Coll04\]. A Survey done by Barthelemy(2001) shows that transaction cost is up to 6% for contracts lower than $10million value.
With transaction cost, we modify the maximization problem of buyer as:
$`K(v)`$ $`=`$ $`\underset{P(y),w(y),a}{\mathrm{max}}{\displaystyle [y(1+\alpha )P(y)+\rho K(w(y))]f(y,a)๐y}`$
st $`{\displaystyle [u(P(y))+\rho w(y)]f(y,a)๐y}\varphi (a)v\text{(PK)}`$ (21)
$`a\mathrm{arg}\mathrm{max}{\displaystyle [u(P(y))+\rho w(y)]f(y,a)๐y}\varphi (a)\text{(IC)}`$
Corresponding first order conditions are:
$`\{P(y)\}`$ $`(1+\alpha )+\lambda u^{}(P(y))+\mu u^{}(P(y)){\displaystyle \frac{f_a(y,a)}{f(y,a)}}=0`$ (22)
$`\{w(y)\}`$ $`\rho K^{}(w(y))+\rho \lambda +\mu \rho {\displaystyle \frac{f_a(y,a)}{f(y,a)}}=0`$ (23)
$`\{a\}`$ $`{\displaystyle }[yP(y)+\rho P(w(y)]f_a(y,a)dy`$ (24)
$`+\mu [{\displaystyle [u(P(x))+\rho w(y)]f_{aa}(y,a)๐y}\varphi ^{\prime \prime }(a)]=0`$
$`\{ENV\}`$ $`K^{}(v)=\lambda `$ (25)
From first order conditions (22) we have
$`{\displaystyle \frac{1+\alpha }{u^{}(P(y))}}=\lambda +\mu {\displaystyle \frac{f_a(y,a)}{f(y,a)}}`$ (26)
Similarly, under perfect information, we have:
$`{\displaystyle \frac{1+\alpha }{u^{}(P(y))}}=\lambda `$ (27)
Compare with equation (14) and equation (20), it can be implied that all other things same, compensation $`P(y)`$ is smaller with transaction cost.
#### 4.3.2 Effect from game among MSSPs
Another effect of transaction cost on market price comes from competition among MSSPs. This effect also suggests when transaction cost increase, nominal market price will decrease.
* A3: Vendors engage in a price competition against each other.
We will derive the Nash Equilibrium<sup>8</sup><sup>8</sup>8A strategy vector x with payoff vector $`\pi `$ is called a Nash Equilibrium if $`\pi _i(x_i,x_i)\pi _i(\stackrel{~}{x_i},x_i),\stackrel{~}{x_i}X_i,i`$. $`X_i`$ is set of all possible actions player $`i`$ can take. This condition means that Nash Equilibrium is such that no player can benefit from unilateral deviations. price under the assumption A1-A3. For this section, to see effect of MSSPsโ competitions, we ignore effect of buyers, and assume perfect information(as shown in section(4.2), optimal contract specifies a non-performance-dependent price, $`P(y)`$ is replaced with $`P`$). We will show that MSSPs will lower price to bear part of the transaction cost due to competition with other MSSPs. Division of the transaction cost between buyers and vendors depends on demand elasticity for security products.
A price competition is where every MSSP uses price as a strategic variable, and is free to choose a price that maximizes their profit given price of other vendors. Explicitly, profit maximization problem for vendor i is:
$$\underset{P^i}{\mathrm{max}}\{P^iN^i((1+\alpha )P)C^i(N^i((1+\alpha )P)\}$$
$`P`$ denotes the price vector $`\{P^i,i=1,\mathrm{},V\}=\{P^i,P^i\}`$, where $`P^i`$ is market price MSSP$`i`$ charges. $`P^i`$ is the price vector of prices of all other MSSPs except MSSP$`i`$ charges. $`N^i`$ is demand for MSSP$`i`$โs service, which depends on market prices. It also depends on service quality MSSPs provide implicitly. $`C^i`$ is MSSP$`i`$โs total cost of servicing $`N^i`$ customers. Then the above maximization problem shows how MSSP$`i`$ maximize its net profit(revenue minus cost) by choosing $`P^i`$ when other vendors charge price $`P^i`$.
$`C^i`$ includes both fixed cost($`FC`$) which does not change with number of customers and variable cost($`VC`$) which does. Explicitly,
$$C^i(N^i())=FC+VC(N^i()),$$
(28)
$`C()`$ increases with number of customers.
Optimal price MSSP$`i`$ should charge solves the following first order condition of the maximization problem w.r.t $`P^i`$:
$$N^i()+P^i\frac{N^i()}{P^i}(1+\alpha )=C^{}(N^i())\frac{N^i()}{P^i}(1+\alpha )$$
(29)
Divide both sides of equation (29) with $`\frac{N^i()}{P^i}(1+\alpha )`$ and rearrange terms, we get:
$$P^i(1\frac{1}{\eta ^i(1+\alpha )})=C^{}(N^i())\text{i}=1,\mathrm{},V$$
(30)
where $`\eta ^i=(N^i()/N^i)/(P^i/P^i)`$, which represents percentage change in demand due to percentage change in price, the price elasticity of vendor iโs demand. It measures how sensitive market demand changes with price. Because $`d()/(P)<0`$(demand and price move in opposite directions), a negative sign is added so that $`\eta >0`$.
solving $`P^i`$ from optimizing condition (30), $`P^i`$ is a function of $`P^i`$, $`\alpha `$ and $`\eta `$:
$$P^i=r(P^i,\alpha ,\eta )$$
(31)
Equation(31) can be viewed as response function of MSSP $`i`$ on prices of other security MSSPs $`P^i`$. Therefore, for all MSSPs on the market, $`i=1,\mathrm{},V`$, we can form a equation system:
$`P^1=r(P^1,\alpha ,\eta ),`$
$`P^2=r(P^2,\alpha ,\eta ),`$
$`\mathrm{}`$
$`P^V=r(P^V,\alpha ,\eta )`$ (32)
The Nash Equilibrium of this price competition is a price vector (*strategies*) that solves the above equation system and a corresponding vector of profit(*payoffs*). Under regularity conditions, this equilibrium price vector exists and is unique.
To give an idea how this Nash Equilibrium price look like, we present a graphic solution for the simplified case when $`V=2`$. Then optimization conditions (32) reduce to the following:
$`P^1=r(P^2,\alpha ,\eta )`$
$`P^2=r(P^1,\alpha ,\eta )`$ (33)
To make things easier, we make two more assumptions:
* A4. Marginal cost $`C_i^{}()`$ is constant, i.e. it costs MSSP $`i`$ same amount of money to serve one additional buyer.
* A5. $`\frac{\eta ^i}{(P^i/P^i)}>0`$, meaning, as MSSP $`i`$โs service becomes more expensive relative to services of other MSSPs, demand for MSSP $`i`$โs service become more elastic. In other word, a same percentage increase in $`P^i`$ will induce greater percentage reduction in $`N^i`$ for higher $`P^i/P^i`$ then lower.
Two response curves $`P^i=r(P^i,\alpha ,\eta ),i=1,2`$ are plotted in figure 1 where the horizontal axe represent MSSP 1โs price and the vertical axe represent MSSP 2โs price. Under A4 and A5, Feenstra showed that both reaction curves have positive slopes. Then slope of MSSP 1โs response curve is larger than slope of that of MSSP 2โs as shown in Fig.1(a).
Because response curve is the locus of MSSPโs best responses given the other MSSPโs action, the intersection point E is the equilibrium point where both MSSPs are are choosing optimally and simultaneously. By definition, they are the Nash Equilibrium prices. Observe that this Nash Equilibrium is a stable equilibrium in the sense that no matter what price the MSSPs start off with, they will eventually arrive at point E, as shown by the arrows in Fig.1(a).
Denote price vendor $`i`$ would charge by $`P_0^i`$ when there is no transaction cost($`\alpha =0`$), from equation system (33),
$`P_0^i=r(P^i,\alpha =0,\eta ),i=1,2`$ (34)
Totally differentiate optimization condition (30),
$`dP^i(1{\displaystyle \frac{1}{\eta ^i(1+\alpha )}})+P^i{\displaystyle \frac{d\eta ^i}{\eta ^{i2}(1+\alpha )}}+P^i{\displaystyle \frac{d\alpha }{\eta ^i(1+\alpha )^2}}=C^{\prime \prime }(N())`$ (35)
By A4
$`C^{\prime \prime }(N())=0`$ (36)
Equation (35) implies:
$`dP^i(1{\displaystyle \frac{1}{\eta ^i(1+\alpha )}}+{\displaystyle \frac{\frac{d\eta ^i/\eta ^i}{dP^i/P^i}}{\eta ^i(1+\alpha )}})=P^i{\displaystyle \frac{d\alpha }{\eta ^i(1+\alpha )}}`$ (37)
Assume:
* A6. $`\frac{d\eta ^i/\eta ^i}{dP^i/P^i}>1\eta ^i(1+\alpha )`$
Under assumption (4.3.2),
$`1{\displaystyle \frac{1}{\eta ^i(1+\alpha )}}+{\displaystyle \frac{\frac{d\eta ^i/\eta ^i}{dP^i/P^i}}{\eta ^i(1+\alpha )}}>0`$ (38)
Equation(37) implies
$`d\alpha >0dP^i<0`$ (39)
This shows that when transaction cost increases, MSSPs reduce their prices correspondingly.
Graphically, the reaction curve $`P^1=r(P^2,\alpha ,\eta )`$ shifts to the left and $`P^2=r(P^1,\alpha ,\eta )`$ shifts down. therefore, compare with the reaction curves when there is no transaction cost. As shown in Figure-1(b), reaction curves with transaction cost intersect at lower price level for both MSSPs. Remember that the intersection of reaction curves is the Nash Equilibrium of the game.
As shown above, under assumptions 1-6, existence of transaction cost reduces prices charged by MSSPs. The extend of reduction depends on how sensitive market demand is to prices.
## 5 Related Work
### 5.1 Empirical Work
Empirical works on this issue were mostly done with surveys. Ang and Straub (1998) performed a well designed survey on banks of different sizes with items measuring degree of IT outsourcing, production cost advantage, transaction cost, financial slack (archive data also used here) outsourcing degree and firm size. And they found that production cost advantage is the main driving force of IT outsourcing, transaction cost dampens outsourcing intention, but has a much smaller effect. They also reported evidence that degree of IT outsourcing decreases with firm size. They argued that this is because large firms are more likely to generate economies of scale in their IT department, therefore are more likely to produce IT services in-house. Lacity and Willcocks (1998) measures success or failure of a IT outsourcing based on seven factors, and found that outsourcing scope, length of contract term, contract type are among the most important factors that decides how successful an IT outsourcing is. Poppo and Zenger (1998) includes technological uncertainty, measurement difficult and quality satisfaction in their model, and showed that when it is harder to measure performances, firm become less satisfied with costs. Ang and Cummings (1997) found empirical evidence that in hyper-competitive environments, not only firms act strategically, but security vendors also.
### 5.2 Analytical Work
Analytical papers on the other hand have a strong game theoretic flavor. Mieghem (1999) built a multivariate, multidimensional competitive model, and investigated effect of subcontracting complexity on coordination.
Ang and Cummings argued that organizations respond strategically under hyper-competitive environments. Whang employed a game theoretical approach to explain asymmetric information and incentive compatible issue in software development.
## 6 Conclusion
Security outsourcing market benefits both vendors and buyers if it works properly. In the first place, security outsourcing offers cost reduction for buyers. We showed that for security outsourcing, optimal form of contract should be performance-based. Also, we showed that with transaction cost, price paid to MSSPs are lower than otherwise. MSSPs take part of the transaction cost to stimulate demand.
|
warning/0506/astro-ph0506211.html
|
ar5iv
|
text
|
# Internal dynamics of the ๐งโผ0.8 cluster RX J0152.7-1357 Based in part on observations carried out at the European Southern Observatory using the ESO Very Large Telescope on Cerro Paranal (ESO programs 166.A-0701, 69.A-0683, and 72.A-0759) and the ESO New Technology Telescope on Cerro La Silla (ESO program 61.A-0676).
## 1 Introduction
Clusters of galaxies are visible tracers of the network of matter in the Universe, marking the high-density regions where filaments of dark matter join together. In the hierarchical scenario of largeโscale structure, clusters form via merging of smaller clumps and accretion of material from large scale filaments (e.g., Borgani & Guzzo bor01 (2001); Evrard & Gioia evr02 (2002)). From the observational side, signatures of past merging processes are found in cluster substructure and evidences for ongoing cluster mergers are rapidly accumulating (e.g., Bรถhringer & Schuecker boh02 (2002); Buote buo02 (2002); Girardi & Biviano gir02 (2002); Evrard evr04 (2004)).
Over the last few years significant progress has been made to extend the above studies from local to distant clusters. Pioneering analyses suggest that no evidence of dynamical evolution is shown by the cluster population out to $`z0.3`$โ0.4 (Adami et al. ada00 (2000); Girardi & Mezzetti gir01 (2001); but cf. Plionis pli02 (2002)). On the other hand, $`z>0.5`$ clusters have more Xโray substructures than lowerโ$`z`$ clusters (Jeltema et al. jel05 (2005)) and most clusters identified at $`z>0.8`$ show an elongated, clumpy, or possibly filamentary structure (e.g., Donahue et al. don98 (1998); Gioia et al. gio99 (1999); Rosati ros04 (2004)) thus suggesting that present observations are approaching the epoch of cluster formation. Our results on RX J0152.7$``$1357 at $`z0.8`$ add further insights on this issue.
The galaxy cluster RX J0152.7$``$1357 was discovered in the ROSAT Deep Cluster Survey (RDCS, Rosati et al. ros98 (1998)) in the ROSAT PSPC field rp60000rn00 observed in January 1992. It was independently discovered in the Wide Angle ROSAT Pointed Survey (WARPS, Ebeling et al. ebe00 (2000)) and reported in the Bright SHARC survey (Romer et al. rom00 (2000)). It appeared also in the list of Xโray extended sources obtained from Einstein IPC data by Oppenheimer et al. (opp97 (1997)).
The BeppoSax observations were used to derive a cluster Xโray bolometric luminosity $`L_{\mathrm{X},\mathrm{bol}}=(22\pm 5)\times 10^{44}`$ erg s<sup>-1</sup> ($`h=0.5`$ and $`q_0=0.5`$), and a gas temperature $`kT=6.46_{1.19}^{+1.74}`$ keV (Della Ceca et al. del00 (2000)). RX J0152.7$``$1357 is characterized by a complex morphology with at least two cores, both in the optical and Xโray data as recovered by Keck imaging and BeppoโSAX data (Della Ceca et al. del00 (2000)). Observations with Chandra also show a complex structure in the intra-cluster medium with the presence in the central cluster region of two peaks in the Xโray emission $`95{}_{}{}^{}^{}`$ apart (North-East: R.A.=$`1^\mathrm{h}52^\mathrm{m}44\text{s}\text{.}\mathrm{\hspace{0.17em}18}`$, Dec.=$`13{}_{}{}^{}57\mathrm{}15{}_{}{}^{}^{}\text{.}\mathrm{\hspace{0.17em}84}`$; South-West: R.A.=$`1^\mathrm{h}52^\mathrm{m}39\text{s}\text{.}\mathrm{\hspace{0.17em}89}`$, Dec.=$`13{}_{}{}^{}58\mathrm{}27{}_{}{}^{}^{}\text{.}\mathrm{\hspace{0.17em}48}`$ \[J2000.0\]), and a possible third peak to the East (R.A.=$`1^\mathrm{h}52^\mathrm{m}52\text{s}\text{.}\mathrm{\hspace{0.17em}42}`$, Dec.=$`13\mathrm{ยฐ}58\mathrm{}5{}_{}{}^{}^{}\text{.}\mathrm{\hspace{0.17em}52}`$ \[J2000.0\]), see Maughan et al. (mau03 (2003)). The existence of an Eastern peak was confirmed by spectroscopic VLT data and an independent analysis of the Chandra data by Demarco et al. (dem05 (2005)), who detect it at the $`>3\sigma `$ c.l. in X-rays (see their Fig. 1). Chandra observations gave a gas temperature for the North-East and South-West central Xโray clumps of $`kT=5.5_{0.8}^{+0.9}`$ keV and $`kT=5.2_{0.9}^{+1.1}`$ keV, respectively (Maughan et al. mau03 (2003)). A complex structure with several clumps is also shown by the gravitational lensing analysis of Jee et al. (jee05 (2005)): in particular, the mass clump A corresponds to the Eastern Xโray peak.
A number of evidences suggest that RX J0152.7$``$1357 may be undergoing a merger: the displacement between peaks of gas distribution and of galaxy/dark matter distribution (Maughan et al. mau03 (2003); Jee et al. jee05 (2005)); the possible presence of a shock front (Maughan et al. mau03 (2003)); the presence of galaxies showing a very recent star formation episode (Jรธrgensen et al. jor05 (2005)); the segregation of starโforming and non starโforming galaxies probably induced by the intraโcluster medium interaction (Homeier et al. hom05 (2005)).
Demarco et al. (dem05 (2005)) have performed an extensive spectroscopic survey of RX J0152.7$``$1357 based on observations carried out with FORS1 and FORS2 on the ESO Very Large Telescope, obtaining more than 200 redshifts in the cluster field. Their analysis shows that RX J0152.7$``$1357 is characterized by a large velocity dispersion, $`1600`$ km s<sup>-1</sup>, and indicates a very complex structure. In particular, the galaxy populations inhabiting the regions around the three main Xโray peaks are characterized by different kinematical behaviour, in agreement with a cluster merging scenario.
On the basis of Demarco et al. data we further investigate the internal dynamics of RX J0152.7$``$1357. The spatial and kinematical analysis of member galaxies is a powerful way to detect and measure the amount of substructure, to identify and analyze possible preโmerging clumps or merger remnants (Girardi & Biviano gir02 (2002) and refs. therein). This optical information is complementary to Xโray information since galaxies and intraโcluster gas react on different time scales during a merger (see, e.g., numerical simulations by Roettiger et al. roe97 (1997); Ricker & Sarazin ric01 (2001); Schindler sch02 (2002)).
The paper is organized as follows. We describe member selection and present our results for global properties of RX J0152.7$``$1357 in Sect. 2. We present our analysis of internal dynamics in Sect. 3. We discuss our results suggesting a tentative picture of the dynamical status of RX J0152.7$``$1357 in Sect. 4. We summarize our results in Sect. 5.
Unless otherwise stated, we give errors at the 68% confidence level (hereafter c.l.)
Throughout the paper, we assume a flat cosmology with $`\mathrm{\Omega }_m=0.3`$, $`\mathrm{\Omega }_\mathrm{\Lambda }=0.7`$ and $`H_0=70`$ km s$`^1`$Mpc<sup>-1</sup>. For this cosmological model 1 arcmin corresponds to $`458`$ kpc at the cluster redshift.
## 2 Member selection and global properties
Our data sample consists of the spectroscopic survey of RX J0152.7$``$1357 presented by Demarco et al. (dem05 (2005)), i.e. 187 galaxies with available redshift (see their Tables 4 and 5). We assume a typical redshift error of $`8\times 10^4`$ according to the authors prescriptions.
The identification of cluster members proceeds in two steps, following a procedure already used for nearby and mediumโredshift clusters (Fadda et al. fad96 (1996); Girardi et al. gir96 (1996); Girardi & Mezzetti gir01 (2001)).
First, we perform the clusterโmember selection in velocity space by using only redshift information. We apply the adaptive kernel method (Pisani pis93 (1993)) to find the significant ($`>99\%`$ c.l.) peaks in the velocity distribution. This procedure detects RX J0152.7$``$1357 as a well isolated peak at $`z=0.836`$ assigning 103 galaxies considered as candidate cluster members (see Fig. 1). Out of nonโmember galaxies, 61 and 23 are foreground and background galaxies, respectively. In particular, a second significant peak of 31 galaxies is shown at $`z=0.638`$ suggesting the presence of a foreground system.
All the galaxies assigned to the RX J0152.7$``$1357 peak are analyzed in the second step, which uses the combination of position and velocity information. We apply the procedure of the โshifting gapperโ by Fadda et al. (fad96 (1996)). This procedure rejects galaxies that are too far in velocity from the main body of galaxies and within a fixed bin that shifts along the distance from the cluster center. The procedure is iterated until the number of cluster members converges to a stable value. We use a gap of $`1000`$ km s$`^1`$โ in the cluster rest-frame โ and a bin of 0.6 Mpc, or large enough to include 15 galaxies. As for the center we consider the position of the biweight center, i.e. we perform the biweight mean-estimator (ROSTAT package; Beers et al. bee90 (1990)) for ascension and declination separately: this center is positioned between the North-East and South-West Xโray peaks (see ยง 1). The choice of using either one of the two Xโray peaks as cluster center does not affect the final results.
The shiftingโgapper procedure rejects eight galaxies to give 95 fiducial members. The list of selected members corresponds to that in Table 4 of Demarco et al. (dem05 (2005)), but excluding galaxies $`\mathrm{\#}`$306,509,557,650,895,1146,1239. Fig. 2 shows the plot of rest-frame velocity $`V_{\mathrm{rf}}=(czcz)/(1+z)`$ vs. clustercentric distance $`R`$ of galaxies in the main redshift peak. Finally, we recompute the biweight center on the 95 cluster members obtaining: R.A.=$`1^\mathrm{h}52^\mathrm{m}41\text{s}\text{.}\mathrm{\hspace{0.17em}669}`$, Dec.=$`13{}_{}{}^{}\mathrm{\hspace{0.17em}57}\mathrm{}\mathrm{\hspace{0.17em}58}{}_{}{}^{}^{}\text{.}\mathrm{\hspace{0.17em}32}`$ (J2000.0). Unless otherwise stated, we adopt this as cluster center.
By applying the biweight estimator to cluster members (Beers et al. bee90 (1990)), we compute a mean cluster redshift of $`z=0.8357\pm 0.0005`$. We estimate the lineโofโsight (LOS) velocity dispersion, $`\sigma _\mathrm{V}`$, by using the biweight estimator and applying the cosmological correction and the standard correction for velocity errors (Danese et al. dan80 (1980)). We obtain $`\sigma _\mathrm{V}=1322_{68}^{+74}`$ km s<sup>-1</sup>, where errors are estimated through a bootstrap technique.
To evaluate the robustness of the $`\sigma _\mathrm{V}`$ estimate we analyze the integral velocity dispersion profile (Fig. 3). The value of $`\sigma _\mathrm{V}(<R)`$ sharply varies in the internal cluster region. A similar behaviour is shown by the mean velocity $`V(<R)`$ suggesting that a mix of clumps at different redshifts is the likely cause for the high value of the velocity dispersion rather than individual contaminating fieldโgalaxies. A robust value of $`\sigma _\mathrm{V}`$ is reached in the external cluster regions where the profile flattens, as found for most nearby clusters (e.g., Fadda et al. fad96 (1996)).
The question of the presence of substructure is deferred to the following sections. Here we assume that the system is in dynamical equilibrium to compute virial global quantities.
Following the prescriptions of Girardi & Mezzetti (gir01 (2001)), we assume for the radius of the quasiโvirialized region R$`{}_{\mathrm{vir}}{}^{}=0.17\times \sigma _\mathrm{V}/H(z)=2.0`$ Mpc (see their eq. 1 after introducing the scaling with $`H(z)`$, see also eq. 8 of Carlberg et al. car97 (1997) for $`R_{200}`$). Thus the cluster is sampled out to a significant region, i.e. $`R_{\mathrm{out}}=0.82\times R_{\mathrm{vir}}`$.
We compute the virial mass (Limber & Mathews lim60 (1960); see also, e.g., Girardi et al. gir98 (1998)) using the data for the $`N_g`$ observed galaxies:
$$M=3\pi /2\sigma _\mathrm{V}^2R_{\mathrm{PV}}/GC,$$
(1)
where $`C`$ is the surface term correction (The & White the86 (1986)), and $`R_{\mathrm{PV}}`$, equal to two times the (projected) harmonic radius, is:
$$R_{\mathrm{PV}}=N_g(N_g1)/(\mathrm{\Sigma }_{ij}R_{ij}^1),$$
(2)
where $`R_{ij}`$ is the projected distance between two galaxies.
The estimate of $`\sigma _\mathrm{V}`$ is generally robust when computed within a large cluster region (see Fig. 3 for RX J0152.7$``$1357 and Fadda et al. fad96 (1996) for other examples). The value of $`R_{\mathrm{PV}}`$ depends on the size of the sampled region and possibly on the quality of the spatial sampling (e.g., whether the cluster is uniformly sampled or not). Here we obtain $`R_{PV}=(1.45\pm 0.05)`$ Mpc, where the error is obtained via a jacknife procedure. The value of $`C`$ strongly depends on the radial component of the velocity dispersion at the radius of the sampled region and could be obtained by analyzing the velocityโdispersion profile, although this procedure would require several hundreds of galaxies. We apply the correction obtained in the literature by combining data of many clusters sampled out to about $`R_{\mathrm{vir}}`$ ($`C/M_\mathrm{V}20\%`$, Carlberg et al. car97 (1997); Girardi et al. gir98 (1998)). We obtain $`M(<R_{\mathrm{out}}=1.65\mathrm{Mpc})=(2.2\pm 0.3)\times 10^{15}`$$`M_{}`$. Calling into question the quality of the spatial sampling, one could use an alternative estimate of $`R_{\mathrm{PV}}`$ on the basis of the knowledge of the galaxy distribution. We assume a Kingโlike distribution, with parameters typical of nearby/mediumโredshift clusters: a core radius $`R_C=1/20\times R_{\mathrm{vir}}`$ and a slopeโparameter $`\beta _{\mathrm{fit}}=0.8`$, i.e. the volume galaxy density at large radii goes as $`r^{3\beta _{fit}}=r^{2.4}`$ (see G98 and Girardi & Mezzetti gir01 (2001)). We obtain $`R_{\mathrm{PV}}=1.25`$ Mpc, with a $`25\%`$ error, thus in agreement with the above direct estimate. The mass is then $`M(<R_{\mathrm{out}})=(1.9\pm 0.5)\times 10^{15}`$$`M_{}`$, in good agreement with our first estimate.
We can use the second of the above approaches to obtain the mass within the whole assumed virialized region, which is larger than that sampled by observations, $`M(<R_{\mathrm{vir}}=2.0\mathrm{Mpc})=(2.2\pm 0.6)\times 10^{15}`$$`M_{}`$.
## 3 Dynamical analysis
### 3.1 Velocity distribution
We analyze the velocity distribution to look for possible deviations from Gaussianity that could provide important signatures of complex dynamics. For the following tests the null hypothesis is that the velocity distribution is a single Gaussian. We base our analysis on shape estimators, i.e. the kurtosis and the skewness. As for the kurtosis, we find $`K=2.04\pm 0.49`$, that indicates a $`2\sigma `$ departure from a Gaussian distribution (reference value $`K=3`$). In addition, we compute the scaled tail index ($`STI`$), which also measures the shape of a distribution, but is based on order statistics of the dataset instead of its moments (see, e.g., Beers et al. bee91 (1991)). This estimator, $`STI=0.860`$, indicates that the tails are underpopulated if the parent population is really a single Gaussian with a c.l. between $`90\%`$ and $`95\%`$, (see Table 2 of Bird & Beers bir93 (1993)). Finally, also the Wโtest (Shapiro & Wilk sha65 (1965)) rejects the null hypothesis of a Gaussian parent distribution at the $`98\%`$ c.l..
Then we investigate the presence of gaps in the distribution, which can be the signature of subclustering. A weighted gap in the space of the ordered velocities is defined as the difference between two contiguous velocities, weighted by the location of these velocities with respect to the middle of the data. We obtain values for these gaps relative to their average size, precisely the midmean of the weighted-gap distribution. We look for normalized gaps larger than 2.25 since in random draws of a Gaussian distribution they arise at most in about $`3\%`$ of the cases, independent of the sample size (Wainer and Schacht wai78 (1978); see also Beers et al. bee91 (1991)). Three significant gaps (2.312, 2.366, 2.395) in the ordered velocity dataset are detected (see Fig. 4). From low to high velocities the dataset is divided in parts containing 39, 29, 3, and 24 galaxies: thus the gaps individuate substantially three main subsets.
In order to detect the presence of groups within our velocity dataset we use the Kayeโs mixture model (KMM) test (Ashman et al. ash94 (1994)). The KMM algorithm fits a user-specified number of Gaussian distributions to a dataset and assesses the improvement of that fit over a single Gaussian. In addition, it provides the maximum-likelihood estimate of the unknown n-mode Gaussians and an assignment of objects into groups. KMM is most appropriate in situations where theoretical and/or empirical arguments indicate that a Gaussian model is reasonable. This is valid in the case of cluster velocity distributions, where gravitational interactions drive the system toward a Gaussian distribution. However, one of the major uncertainties of this method is the optimal choice of the number of groups for the partition. Moreover, only in mixture models with equal covariance matrices for all components the algorithm converges, while this is not always true for the heteroscedastic case (see Ashman et al. 1994, for further details).
Our search for significant gaps suggests the presence of two Gaussians (separated by the two very close second and third gaps at $`252\times 10^3`$ km s<sup>-1</sup>) or possibly three Gaussians (corresponding to the three main subsets). In the homoscedastic case the KMM algorithm fits a twoโgroup partition by rejecting the single Gaussian at the $`97.4\%`$ c.l. (as obtained from the likelihood ratio test). The threeโgroup partition is fitted at the $`97.9\%`$ c.l.. In the heteroscedastic case we use the results of the gap analysis to determine the first guess and we fit two velocity groups around the guess meanโvelocities of $`249\times 10^3`$ and $`254\times 10^3`$ km s<sup>-1</sup>. The algorithm fits a twoโgroup partition at the $`99.4\%`$ c.l. Similarly, we fit three velocity groups around the guess meanโvelocities of $`247\times 10^3`$, $`250\times 10^3`$, and $`254\times 10^3`$ km s$`^1`$to obtain a threeโgroup partition at the $`97.2\%`$ c.l. The high probability value obtained in the heteroscedastic bimodal case suggests the presence of a main cluster of 76 galaxies (KMM1), with the presence of a highโvelocity clump of 19 galaxies (KMM2). In turn, the main cluster can be subdivided in two clumps of 19 and 57 galaxies according to the heteroscedastic trimodal case (KMM1a and KMM1b, respectively). Table 1 lists the kinematical properties of these clumps: the three corresponding Gaussians are displayed in Fig. 4. Fig. 5 shows that the galaxies of the KMM1a group mainly populate the SouthโWest central region of the cluster. This spatial segregation suggests to investigate the velocity field in more detail.
### 3.2 Velocity field
The cluster velocity field may be influenced by the presence of internal substructures, possible cluster rotation, and the presence of other structures on larger scales, such as nearby clusters, surrounding superclusters, and filaments. Each asymmetry effect could produce a velocity gradient in the cluster velocity field.
To investigate the velocity field of RX J0152.7$``$1357 we divide galaxies in a lowโ and a highโvelocity samples by using the median value of galaxy velocities $`\overline{V}=250626`$ km s$`^1`$and check the difference between the spatial distributions of the two samples. Highโ and lowโvelocity galaxies appear segregated in the EโW direction (see Fig. 6). The corresponding spatial distributions are different at the $`99.2\%`$ c.l. according to the twoโdimensional KolmogorovโSmirnov test (hereafter 2DKSโtest; see Fasano & Franceschini fas87 (1987), as implemented by Press et al. pre92 (1992)).
To estimate the direction of the velocity gradient we perform a multiple linear regression fit to the observed velocities with respect to the galaxy positions in the plane of the sky (see also den Hartog & Katgert den96 (1996); Girardi et al. gir96 (1996)). We find a position angle on the celestial sphere of $`PA=97{}_{}{}^{}\pm 16^{}`$ (measured counterโclockโwise from North), i.e. higherโvelocity galaxies lie in the EastโSouth-East region of the cluster, in agreement with the visual impression of galaxy distribution in Fig. 6. To assess the significance of this velocity gradient we perform 1000 Monte Carlo simulations by randomly shuffling the galaxy velocities and for each simulation we determine the coefficient of multiple determination ($`RC^2`$, see e.g., NAG Fortran Workstation Handbook nag86 (1986)). We define the significance of the velocity gradient as the fraction of times in which the $`RC^2`$ of the simulated data is smaller than the observed $`RC^2`$. We find that the velocity gradient is significant at the $`98.3\%`$ c.l..
We also analyze the central cluster region using 22 galaxies within a radius of 0.4 Mpc. This choice allows us to include the position of both Xโray peaks and exclude the East region populated by higherโvelocity galaxies only. We find a very significant ($`99.5\%`$) position angle of $`PA=59{}_{}{}^{}{}_{}{}^{+28^{}}_{25^{}}`$, i.e. higherโvelocity galaxies lie in the direction of the North-East Xโray clump.
### 3.3 3D substructure and detection of subclumps
The existence of correlations between positions and velocities of cluster galaxies is a footprint of real substructures. Here we combine velocity and position information to compute the $`\mathrm{\Delta }`$โstatistics devised by Dressler & Schectman (dre88 (1988)). This test is sensitive to spatially compact subsystems that have either an average velocity that differs from the cluster mean, or a velocity dispersion that differs from the global one, or both. We find $`\mathrm{\Delta }=154`$ for the value of the parameter which gives the cumulative deviation of the local kinematical parameters (velocity and velocity dispersion) from the global cluster parameters. To compute the significance of substructure we run 1000 Monte Carlo simulations, randomly shuffling the galaxy velocities, and obtain a value of $`>99.9\%`$.
This technique also provides information on the positions of substructures. Fig. 7 shows the distribution on the sky of all galaxies, each marked by a circle: the larger the circle, the larger the deviation $`\delta _i`$ of the local parameters from the global cluster parameters, i.e. the higher the evidence for substructure. A clump of galaxies with low velocity is the likely cause of large values of $`\delta _i`$ in the region which lies closely at SouthโWest of the cluster center, i.e. in correspondence of the South-West Xโray peak. The other possible substructure, populated by highโvelocity galaxies, lies in the Eastern region.
To assign galaxies to the 3Dโsubclumps, we resort to the technique developed by Biviano et al. (biv02 (2002)), who used the individual $`\delta _i`$โvalues of the Dressler & Schectman method. The critical point is to determine the value of $`\delta _i`$ that optimally separates between internal and external substructures. To this aim we consider the $`\delta _i`$โvalues of all 1000 Monte Carlo simulations already used to determine the significance of the substructure (see above). The resulting distribution of $`\delta _i`$ is compared to the observed one finding a difference of $`99.8\%`$ c.l. according to the KSโtest. The โsimulatedโ distribution is normalized to produce the observed number of galaxies and compared to the observed distribution in Fig. 8: the latter shows a tail at large values. This tail is populated by galaxies that presumably are in substructures.
For the selection of galaxies within substructures we choose the value of $`\delta =3.35`$, since only after the rejection of the values $`\delta _i>\delta `$, the observed and simulated distributions are no longer distinguishable according to the KSโtest. With this choice, 14 galaxies of the cluster are assigned to substructures: six to the central SouthโWest clump (DS-SW\*) and eight to the East clump (DS-E\*), see Fig. 9. The velocity dispersions computed for these structures, $`\sigma _\mathrm{V}300`$ km s$`^1`$and $`650`$ km s$`^1`$for DS-S\* and DS-E\* clumps, respectively, are likely to be considered as lower limits since our analysis does not guarantee the detection of all substructure members.
We consider also a more relaxed criteria, by selecting galaxies with $`\delta _i>3`$ as suggested by the histogram of Fig. 8: Table 1 shows that the results for the SouthโWest clump (DS-SW vs. DS-SW\*) are very robust, while only one additional galaxy in the Eastern clump (DS-E vs. DS-E\*) leads to an increase of 200 km s$`^1`$in the velocity dispersion. We also consider the remaining 76 galaxies of the main structure (DS-M). DS-M does not contain significant structure according to the DresslerโSchectman test. However, since we cannot exclude a residual contamination from substructure members, its value of velocity dispersion $`\sigma _\mathrm{V}1300`$ km s$`^1`$is likely to be an upper limit.
The Dressler-Schectman results superseed those of the KMM test. Again, there is the presence of a lowโvelocity clump and now its SouthโWest position is better defined by the detection of the DS-SW clump. The presence of a highโvelocity clump is confirmed and located at the East by the detection of the DS-E clump. Moreover, the location of DS clumps well coincide with Xโray peaks of extended emissions.
### 3.4 Analysis of Xโray โ centered clumps
The good spatial agreement between detected galaxy clumps and peaks of Xโray emission prompts us to analyze the profiles of mean velocity and velocity dispersion of galaxy systems corresponding to the South-West, East, and North-East Xโray peaks, i.e. using the position of the Xโray peaks as system-centers (see Figs. 10, Fig. 11, and Fig. 12, respectively). This allows an independent analysis of the individual galaxy clumps. An increase of the velocityโdispersion profile in their central regions might be due to dynamical friction and galaxy merging (e.g., Menci & Fusco-Femiano men96 (1996); Girardi et al. gir98 (1998); Biviano & Katgert biv04 (2004)), or simply induced by the presence of interlopers or of a secondary clump (e.g., Girardi et al. gir96 (1996)). The latter hypothesis can be investigated by looking at the behaviour of the mean velocity profile. Figs. 10, 11, and 12 show velocityโdispersion and meanโvelocity profiles, and regions likely not contaminated by other clumps and thus reliable for kinematical analysis. Detailed results of this analysis are included in Table 1 where the clumps are named as SW, E, and NE.
The analysis of the SouthโWest central region has indicated the presence of a lowโvelocity clump with a low velocityโdispersion (of 300โ400 km s$`^1`$according to DS-SW and KMM1a results). Fig. 10 shows how the velocityโdispersion increases with the distance from the South-West Xโray peak. The meanโvelocity shows a sharp change very close to the Xโray peak, at $`0.2`$ Mpc. This suggests a strong contamination of galaxies from other structures. We consider two possible uncontaminated regions: one within 0.2 Mpc, where we find $`\sigma _\mathrm{V}500`$ km s<sup>-1</sup>, and one within 0.18 Mpc, where we find $`\sigma _\mathrm{V}300`$ km s<sup>-1</sup>. Such a sharp change of $`\sigma _\mathrm{V}`$ is induced just by the rejection of two galaxies, one of which has an anomalously high velocity. The value of $`\sigma _\mathrm{V}`$ for the SW-clump is further analyzed in Sect. 3.5 and discussed in Sect. 4.1.
The DresslerโSchectman analysis of the East region has indicated the presence of a highโvelocity clump with a velocity dispersion of about 600โ800 km s<sup>-1</sup>. By choosing the Xโray peak as center (Fig. 11), the mean velocity changes at $`0.40.5`$ Mpc from the Xโray peak. Inside this region, we obtain $`\sigma _\mathrm{V}700`$ km s$`^1`$for the E-clump
Fig. 12 refers to the region around the North-East Xโray peak. The main mass clump is located in this same position, according to the gravitational lensing analysis (Jee et al. jee05 (2005)). We have shown that this region is mostly populated by galaxies having velocities intermediate between those of the above clumps, and likely forms a high velocity dispersion structure (i.e., KMM1b clump in Sect. 3.1, and DS-M system in Sect. 3.3). Fig. 12 shows an increase of the integral velocityโdispersion profile at about 0.4 Mpc from the Xโray peak, and a corresponding sharp change in the mean velocity. Moreover, galaxies of both DS-SW and DS-E substructures lie beyond 0.4 Mpc from the NorthโEast Xโray peak. Thus, the value $`\sigma _\mathrm{V}900`$ km s$`^1`$, computed within 0.4 Mpc , should be reliable.
The three Xโray clumps differ from each other in mean velocities at a c.l. $`>99\%`$, according to the meansโtest (e.g., Press et al. pre92 (1992)).
Assuming that each of the three galaxy clumps is a system in dynamical equilibrium, for each clump we compute the virial radius and the mass contained inside with the same procedure adopted in Sect. 2 (see Table 2). The large uncertainties associated to the mass values are due to poor number statistics.
### 3.5 Spectralโtype segregation
We check for possible spectralโtype segregation of galaxies, both in position and in velocity space, by using the classification of Demarco et al. (dem05 (2005), see their Table 4), i.e. passive galaxies (k), galaxies with significant Balmer lines โ likely post-starbursts (k+a/a+k) and galaxies with relevant emission lines (e/k+a+\[OII\]). The sample of cluster members contains 56, 7, and 32 passive, postโstarburst, and emissionโline galaxies, respectively.
Figure 13 shows the spatial distribution of galaxies of different types. As already noted by Demarco et al., emissionโline galaxies avoid the regions of the subclumps (see also Homeier et al. hom05 (2005)). The same behaviour is shown by postโstarburst galaxies. When comparing spatial distributions of passive (k) and โactiveโ (k+a/a+k/e/k+a+\[OII\]) galaxies we find a very strong difference: $`>99.99\%`$, according to the 2DKS-test.
As for the velocity distributions, no difference is found between passive and โactiveโ galaxies, according to the KSโtests. Moreover, mean velocities and velocity dispersions of the two populations (see Table 1) do not significantly differ according to the meansโ and Fโtest (e.g., Press et al. pre92 (1992)). This suggests that our sample of member galaxies is not significantly contaminated by interlopers. In fact, possible field galaxies would preferably contaminate the sample of โactiveโ galaxies causing a difference in the kinematical properties with respect to the sample of passive galaxies, e.g., enhancing the velocity dispersion or changing the mean velocity.
Finally, we perform again the analysis of mean velocity and velocityโdispersion profiles of Sect. 3.4, by considering passive galaxies only. We draw different conclusions only for the SW-clump. Fig. 14 shows that the mean velocity now changes only at $`0.3`$ Mpc from the South-West Xโray peak and the velocity dispersion does not increase any longer in the central region. Within $`0.3`$ Mpc, we compute for the SW-clump a value of $`321_{59}^{+132}`$ km s<sup>-1</sup>, in good agreement with the lower estimate of $`\sigma _\mathrm{V}`$ obtained in Sect. 3.4 (see Table 1).
## 4 Discussion
Out of 187 galaxies with available redshift we assign 95 members to RX J0152.7$``$1357. This galaxy selection is more restrictive than made by Demarco et al. (dem05 (2005): 102 members giving a velocity dispersion of $`1600`$ km s<sup>-1</sup>) due to our analysis of position and velocity combined together. In particular, we reject a small group of galaxies at $`z=0.864`$โ0.867. In spite of this more restrictive member selection, the value we obtain for the LOS velocity dispersion is still rather high, $`\sigma _\mathrm{V}=1322_{68}^{+74}`$ km s<sup>-1</sup>, and lies in the highโtail of the $`\sigma _\mathrm{V}`$โdistribution of nearby/medium redshift clusters (see, e.g., Fadda et al. fad96 (1996); Mazure et al. maz96 (1996); Girardi & Mezzetti gir01 (2001)). The position on the $`L_{\mathrm{X},\mathrm{bol}}`$$`\sigma _\mathrm{V}`$ plane is consistent with the relation provided by Borgani et al. (1999) for moderately distant clusters and by Wu et al. (1999) for local clusters. As for the $`\sigma _\mathrm{V}`$$`T_X`$ relation, assuming the densityโenergy equipartition between gas and galaxies, i.e. $`\beta _{\mathrm{spec}}=1`$ (e.g., Girardi et al. gir96 (1996), gir98 (1998); Xue & Wu xue00 (2000)), where $`\beta _{\mathrm{spec}}=\sigma _\mathrm{V}^2/(kT/\mu m_p)`$ with $`\mu =0.58`$ the mean molecular weight and $`m_p`$ the proton mass, our value of $`\sigma _\mathrm{V}`$ corresponds to $`kT=10.6_{1.1}^{+1.2}`$ keV. This value is more than 2$`\sigma `$ higher than the Xโray temperature determined from BeppoSAX observations (Della Ceca et al. del00 (2000)) and more than 3$`\sigma `$ higher than those of the North-East and South-West Xโray systems as determined from Chandra data (Maughan et al. mau03 (2003); Huo et al. huo04 (2004)). This suggests a strong departure from dynamical equilibrium and, in fact, we find evidence for nonโGaussianity of the velocity distribution, presence of a velocity gradient and significant substructure.
We find no kinematical difference between passive and โactiveโ galaxy populations. This suggests that our sample of member galaxies is not significantly contaminated by interlopers. In fact, possible field galaxies would preferably contaminate the sample of โactiveโ galaxies causing a difference in the kinematical properties with respect to the sample of passive galaxies, e.g., enhancing the velocity dispersion or changing the mean velocity.
Instead, our analysis shows that the high value of $`\sigma _\mathrm{V}`$ is due to the complex structure of this system, i.e. to the presence of three galaxy clumps of different meanโvelocity. Using optical data only we detect the lowโvelocity SW-clump in the central regions and the highโvelocity E-clump, which lie close to the South-West and East peaks detected by the Xโray analysis. The North-East Xโray peak is then associated to the main galaxy structure. In particular, the high relative velocity between the NE- and SW-clumps, $`V_\mathrm{r}=(V_{\mathrm{NE}}V_{\mathrm{SW}})/(1+z)=1531`$ km s<sup>-1</sup>, explains the high value of $`\sigma _\mathrm{V}`$ measured in the central cluster region and the presence of a velocity gradient there (see Figs. 3 and 6), while the global velocity gradient is induced by the presence of the highโvelocity E-clump in external cluster regions. The presence of the three galaxy clumps was already suggested by Demarco et al. (dem05 (2005)) from the inspection of the velocity distribution in relation to the spatial location of galaxies. Moreover, the NE-, SW-, and E- clumps correspond to three clumps in the mass distribution as obtained from the weak lensing analysis (Jee et al. jee05 (2005): C, F, and A subclumps, respectively).
As for the mass of the whole cluster, from the global analysis of Sect. 2 we obtain $`M(<2.0\mathrm{Mpc})=(2.2\pm 0.6)\times 10^{15}`$$`M_{}`$. Since the system is not virialized, but likely bound (see the discussion below), this estimate might overestimate the mass even by a factor two. Adding the mass estimates of each clump within its virial radius (see Table 2, Sect. 3.4), we obtain $`M=1.2_{0.3}^{+0.5}\times 10^{15}`$$`M_{}`$: this estimate should be considered as a lower value within 2.0 Mpc , since it does not consider other small clumps or isolated infalling, bound galaxies, as well the likely possibility that the three clumps extend outside the virial radius. Thus, we conclude that the cluster mass within 2 Mpc lies in the range of $`(1.22.2)\times 10^{15}`$$`M_{}`$, which is typical to that of very massive clusters (e.g., Girardi et al. gir98 (1998); Girardi & Mezzetti gir01 (2001)).
Our mass estimate is consistent with that of Maughan et al. (mau03 (2003)) of $`(1.1\pm 0.2)\times 10^{15}`$$`M_{}`$, based on Chandra Xโray analysis and considering only the two central clumps within 1.4 Mpc. To compare with results from weak lensing analyses we also compute the projected mass, by considering the global cluster geometry as formed by the three clumps at the cluster redshift. Each clump is described by the Kingโlike mass distribution (see Sect. 2) or, alternatively, by a NFW profile where the massโdependent concentration parameter is taken from Navarro et al. (nav97 (1997)) and rescaled by the factor 1+z (Bullock et al. bul01 (2001); Dolag et al. dol04 (2004)). The mass distribution of each clump is truncated at one virial radius or, alternatively, at two virial radii. In the following we indicate the range of our results. We find the projected mass within 1 Mpc from the center of the main clump (NE-clump) to be $`(915)`$$`\times 10^{14}M_{}`$, higher than that, 5$`\times 10^{14}M_{}`$, of Jee et al. (jee05 (2005)), but in agreement with the value $`>1\times 10^{15}`$$`M_{}`$of Huo et al. (huo04 (2004), see their Figure 10). Indeed, both Huo et al. and Jee et al. compare their weak lensing results with an isothermal sphere with $`\sigma _\mathrm{V}=`$900โ1000 km s<sup>-1</sup>, in agreement with the value of $`\sigma _\mathrm{V}`$ that we measure for the main galaxy clump. However, the weakโlensing mass lies above or below the isothermal sphere mass for Jee et al. and Huo et al., respectively.
### 4.1 Individual clumps
Our estimate of $`\sigma _\mathrm{V}`$ for the NE-clump well agrees with that of Demarco et al. (dem05 (2005)) and corresponds to $`kT=4.8_{0.4}^{+1.8}`$ keV, in agreement with the observed gas temperature of $`6`$ KeV (Maughan et al. mau03 (2003); Tozzi et al. toz03 (2003); Huo et al. huo04 (2004)). Similarly, our mass estimate, $`M(<R_{\mathrm{vir}}=1.35\mathrm{Mpc})=7.0_{2.1}^{+2.9}`$$`\times 10^{14}M_{}`$, well agrees with the Xโray mass by Maughan et al. \[mau03 (2003), $`M(<1.4\mathrm{Mpc})=7.0_{1.5}^{+1.7}`$$`\times 10^{14}M_{}`$\]. To compare our results with other studies we rescale $`M(<R_{\mathrm{vir}})`$ at their radii by using the Kingโlike profile or, alternatively, the NFW profile (see above). In the following we give the two values obtained from the rescaling, reliable with a $`30\%`$ lowerโerror and a $`40\%`$ upperโerror, as derived from the estimate of $`M(<R_{\mathrm{vir}})`$. Our estimates well agree with those of other studies: $`M(<0.43\mathrm{Mpc})=`$(1.9โ2.6)$`\times 10^{14}M_{}`$, cf. with $`(2.5\pm 0.9)`$$`\times 10^{14}M_{}`$by Demarco et al. (dem05 (2005)), based on galaxy dynamics; $`M(r<65{}_{}{}^{}^{}=0.496\mathrm{Mpc})=`$(2.3โ2.9)$`\times 10^{14}M_{}`$, cf. with $`(3\pm 1)`$$`\times 10^{14}M_{}`$by Joy et al. (joy01 (2001)), based on the SunyaevโZeldovich effect; $`M(r<0.753\mathrm{Mpc})`$=(3.8โ4.3)$`\times 10^{14}M_{}`$, cf. with $`(2.66\pm 0.77)`$$`\times 10^{14}M_{}`$by Ettori et al. (ett04 (2004)), based on Chandra Xโray data; $`M(r<1\mathrm{Mpc})=`$(5.1โ5.4)$`\times 10^{14}M_{}`$, cf. with $`5`$$`\times 10^{14}M_{}`$by Huo et al. (huo04 (2004)), based on Chandra Xโray data.
As for the SW-clump, the results in the literature are not yet clear. In fact, the Xโray temperature suggests that the North-East and the South-West clumps are similar in mass (Maughan et al. mau03 (2003); Huo et al. huo04 (2004)), while both the optical analysis by Demarco et al. (dem05 (2005)) and the weak lensing analysis by Jee et al. (jee05 (2005)) find that the South-West clump is about half massive than the North-East clump. Our analysis of the $`\sigma _\mathrm{V}`$โprofile gives two alternative values for $`\sigma _\mathrm{V}`$: the larger value is consistent with that found by Demarco et al. (dem05 (2005), cf. $`\sigma _\mathrm{V}=503_{96}^{+439}`$ km s$`^1`$vs. their $`737\pm 126`$ km s<sup>-1</sup>) and with the observed gas temperatures of 5-6 keV (Maughan et al. mau03 (2003); Huo et al. huo04 (2004)), while the lower estimate, $`\sigma _\mathrm{V}=301_{107}^{+122}`$, is significantly different. This uncertainty is due to the fact that the $`\sigma _\mathrm{V}`$ profile increases in central regions (see Fig.10) and thus the $`\sigma _\mathrm{V}`$ estimate strongly depends on the considered region. Demarco et al. considered a region (based on Xโray data) larger than our region (based on kinematical data). Our analysis of passive galaxies also gives a small value, $`\sigma _\mathrm{V}300`$ km s$`^1`$, thus suggesting two alternative hypothesis: 1) high values of $`\sigma _\mathrm{V}`$ are due to galaxyโcontamination by other clumps, so that the SW-clump should be considered as a very small group, 2) we are dealing with a very relaxed core hosted in a highโ$`\sigma _\mathrm{V}`$, massive cluster. The second hypothesis is consistent with the observations of nearby clusters where $`\sigma _\mathrm{V}`$ of the subsample of bright central elliptical galaxies is lower than $`\sigma _\mathrm{V}`$ of the whole cluster (Biviano & Katgert biv04 (2004)), a phenomenon possibly due to dynamical friction and galaxy merging (e.g., Menci & Fusco-Femiano men96 (1996)). Only a deeper galaxy sample would allow us to better trace and separate the NorthโEast and the South-West systems and thus discriminate between the two hypotheses. However, the SW-clump appears to be so dense of galaxies that we are inclined to believe in the detection of a clusterโcore. In this case, we note that: a) our mass estimate would be an underestimate of the global mass of the Southern cluster; b) our results would be reconciled with high values of gas temperature and Xโray luminosity (Maughan et al. mau03 (2003); Tozzi et al. toz03 (2003); Huo et al. huo04 (2004)).
As for the Eastern clump, the level of Xโray emission in the Chandra image is much lower than those of the North-East or the South-West clumps (see Fig. 1 of Demarco et al. dem05 (2005)). On the contrary, its gravitationalโlensing mass is comparable to that of the South-West clump (see A and F clumps by Jee et al. jee05 (2005)), and our estimate of velocity dispersion is typical of a massive cluster, $`\sigma _\mathrm{V}700`$ km s<sup>-1</sup>. This discrepancy with Xโray luminosity suggests that this galaxy system is far from being virialized, maybe elongated along the LOS (thus giving a high $`\sigma _\mathrm{V}`$ and a high projected lensing mass), with the gas component not very dense. In particular, the Eastern Xโray peak might be associated to a small group embedded in a largeโscale structure filament connecting to the cluster from the EastโSouth-East region, which is populated by higher velocity โmaybe more distant โ galaxies (see Fig. 6). In the case of a bound, but non virialized structure, we might have overestimated the mass even by a factor two.
We finally compare the projected mass of the three clumps within a radius of $`20{}_{}{}^{}^{}`$ with the results from weak lensing by Jee et al. (jee05 (2005)). The resulting values for projected masses of the NE-, SW-, and E-clumps lie in the ranges $`(1.62.3)`$$`\times 10^{14}M_{}`$, $`(0.50.7)`$$`\times 10^{14}M_{}`$, and $`(1.01.5)`$$`\times 10^{14}M_{}`$, all values somewhat higher than those reported in Table2 of Jee et al. for clumps C, F, and A, respectively.
### 4.2 Analytic calculations of the dynamical status
Here, we investigate the relative dynamics of the NE- and SW-clumps in the central cluster region using different analytic approaches which are based on an energy integral formalism in the framework of locally flat spacetime and Newtonian gravity (e.g., Beers et al. bee82 (1982)). The values of the relevant observable quantities for the twoโclump system are: the relative LOS velocity, $`V_\mathrm{r}=1531`$ km s<sup>-1</sup>, the projected linear distance between the two clumps, $`D=0.66`$ Mpc, the mass of the system obtained adding the masses of the two clumps each within its virial radius, $`M_{\mathrm{sys}}=8.2_{2.2}^{+3.6}`$$`\times 10^{14}M_{}`$.
First, we consider the Newtonian criterion for gravitational binding stated in terms of the observables as $`V_\mathrm{r}^2D2GM_{\mathrm{sys}}sin^2\alpha cos\alpha `$, where $`\alpha `$ is the projection angle between the plane of the sky and the line connecting the centers of two clumps. The faint curve in Fig. 15 separates the bound and unbound regions according to the Newtonian criterion (above and below the curve, respectively). Considering the value of $`M_{\mathrm{sys}}`$, the NE+SW system is bound between $`30^{}`$ and $`77^{}`$: the corresponding probability, computed considering the solid angles (i.e., $`_{30}^{77}cos\alpha ๐\alpha `$), is $`47\%`$. We also consider the implemented criterion $`V_\mathrm{r}^2D2GMsin^2\alpha _\mathrm{V}cos\alpha _D`$, which introduces different angles for projection of distance and velocity, not assuming strictly radial motion between the clumps (Hughes et al. hug95 (1995)). We obtain a binding probability of $`44\%`$.
Then, we apply the analytical twoโbody model introduced by Beers et al. (bee82 (1982)) and Thompson (tho82 (1982); see also Lubin et al. lub98 (1998) for a recent application). This model assumes radial orbits for the clumps, with no shear or net rotation of the system. Furthermore, the clumps are assumed to start their evolution at time $`t_0=0`$ with separation $`d_0=0`$, and are moving apart or coming together for the first time in their history, i.e. we are assuming that we are seeing the cluster prior to merging. The bimodal model solution gives the total system mass $`M_{\mathrm{sys}}`$ as a function of $`\alpha `$ (e.g., Gregory & Thompson gre84 (1984)). Fig. 15 compares the bimodalโmodel solutions with the observed mass of the system, which is the most uncertain observational parameter. The present bound outgoing solutions (i.e. expanding), BO, are clearly inconsistent with the observed mass. The possible solutions span these cases: the bound and present incoming solution (i.e. collapsing), $`BI_a`$ and $`BI_b`$, and the unbound-outgoing solution, $`UO`$. For the incoming case there are two solutions because of the ambiguity in the projection angle $`\alpha `$. We compute the probabilities associated to each solution assuming that the region of $`M_{\mathrm{sys}}`$ values between 1$`\sigma `$ bands is equally probable for individual solutions: $`P_{\mathrm{BIa}}65\%`$, $`P_{\mathrm{BIb}}35\%`$, $`P_{\mathrm{UO}}<1\times 10^4\%`$.
There are several limitations to characterize the dynamics of the central region of RX J0152.7$``$1357 using these models. For instance, possible underestimates of the masses, e.g., if the clumps extend outside the virial radius or if the SW-clump is only the core of the South-West system (see above), lead to binding probabilities larger than those computed above. Moreover, the models do not take into accounts the mass distribution in the clumps when the separation of the clumps is comparable with their size (i.e. at small $`\alpha `$) and do not consider the possible effect of the E-clump. Finally, the twoโbody model breaks down in a regime where the NE- and SW-clumps are already strongly interacting, as suggested by several evidences: the displacement between peaks of gas distribution and of galaxy/dark matter distribution (Maughan et al. mau03 (2003); Huo et al. huo04 (2004)); Jee et al. jee05 (2005)); the possible presence of a shock front (Maughan et al. mau03 (2003)); the presence of galaxies showing a very recent star formation episode (Jรธrgensen et al. jor05 (2005)); the segregation of starโforming and non starโforming galaxies probably induced by the interaction with the intraโcluster medium (Homeier et al. hom05 (2005)).
Looking at galaxies only we cannot discriminate between a preโ or postโmerging phase since the galaxy component is very robust against mergers, e.g., two clusters can pass through one another without destroying the individual optical components (e.g., White & Fabian whi95 (1995); Roettiger et al. roe97 (1997)). Note, for instance, that the properties of the SW-clump resemble those of clusterโcores destined to survive tidal disruption during the merger: size comparable to the cluster core and mass $`<0.05\times `$ cluster mass (see GonzรกlezโCasado et al. gon94 (1994)). These cores will be detectable in the host cluster as a substructure for a long time. Since the gas component shows two well distinct entities in the central cluster regions, we presume that the merging is not too advanced, i.e. well before the coalescence.
Under the assumption that the two central clumps are already very close, we apply the above dynamical models to the system made of the \[(NE+SW)+E\] clumps, too. The values of the relevant observable quantities are: $`V_\mathrm{r}=1401`$ km s<sup>-1</sup>, $`D=1.09`$ Mpc, and $`M_{\mathrm{sys}}=11.7_{2.6}^{+4.7}`$ $`\times 10^{14}M_{}`$. We obtain that the binding probabilities are $`48\%`$, and $`45\%`$, according to the Newtonian criterion and its implementation, respectively; while the twoโbody model gives a probability $`>99.9\%`$ for the bound incoming solution.
## 5 Summary & conclusions
We present the results of the dynamical analysis of the cluster of galaxies RX J0152.7$``$1357, one of the most massive structures known at $`z>0.8`$. The Xโray emission is known to have two clumps in the central regions, and a third clump $`1`$ Mpc to the East. Our analysis is based on velocities and positions of member galaxies taken from the extensive spectroscopic survey performed by Demarco et al. (dem05 (2005)), i.e. 187 galaxies having redshift in the cluster region.
We find that RX J0152.7$``$1357 appears as a well isolated peak in the redshift space at $`z=0.836`$, and select 95 cluster members. We compute a value for the LOS velocity dispersion of galaxies, $`\sigma _\mathrm{V}=1322_{68}^{+74}`$ km s<sup>-1</sup>, much larger than expected for a relaxed cluster with an observed Xโray temperature of $`56`$ keV.
We find evidence that this cluster is far from dynamical equilibrium, as shown by:
* the non Gaussianity of the velocity distribution according to different tests, at the 90โ98$`\%`$ c.l., as well as the presence of significant velocity gaps;
* the correlation between velocities and positions of galaxies at the $`>99\%`$ c.l., and the presence of a velocity gradient;
* the presence of significant substructures at the $`>99.9\%`$ c.l..
To detect and analyze possible subsystems we used different methods.
* By applying the KMM method we find that a twoโclumps, and likely a threeโclumps partition of the velocity distribution is significantly better than a single Gaussian to describe the velocity distribution; in particular, the galaxies of KMM1a group are mainly located in the SouthโWest central region.
* By combining positions and velocities in the Dressler & Schectman statistics we detect two substructures, well corresponding in location to the SouthโWest and East Xโray peaks, in addition to the main cluster component identified with the NorthโEast Xโray peak.
* Taking advantage of Xโray peak determination, we analyze the three galaxy clumps centered in these peaks through the profiles of mean velocity and velocity dispersion. This analysis allows us to estimate the clump region that is likely not contaminated by galaxies of other clumps and to evaluate the kinematical properties.
In summary, our analysis shows that the high value of $`\sigma _\mathrm{V}`$ is due to the complex structure of RX J0152.7$``$1357, i.e. to the presence of three galaxy clumps of different meanโvelocity. Using optical data we detect a lowโvelocity clump (with $`\sigma _\mathrm{V}=`$300โ500 km s<sup>-1</sup>) in the central SouthโWest region and a highโvelocity clump (with $`\sigma _\mathrm{V}`$700 km s<sup>-1</sup>) in the Eastern region, nicely matching the position of the SouthโWest and East peaks detected in the Xโray emission. The central NorthโEast Xโray peak is associated to the main galaxy structure having intermediate velocity and $`\sigma _\mathrm{V}900`$ km s<sup>-1</sup>. The three clumps differ from each other in mean velocities at a c.l. $`>99\%`$ (relative LOS velocities are $`>1000`$ km s<sup>-1</sup>).
The mass of the whole system within 2 Mpc is estimated to be (1.2โ2.2)$`\times 10^{15}`$$`M_{}`$, where the upper and lower limits come from the virial analysis of the cluster as a whole and from the sum of virial masses of the three individual clumps, respectively.
Analytic calculations, based on the two-body model, indicate that the system is most likely bound, destined to merge. In particular, we suggest that the SouthโWest clump is not a small group, but rather the dense core of a massive cluster, able to survive tidal disruption during the merger.
In conclusion, RX J0152.7$``$1357 reveals a very complex structure, with several clumps likely destined to merge in a very massive cluster. Our results lend further support to the picture that massive clusters at $`z>0.8`$ are dynamically complex and, therefore, likely to be young. This indicates that we are approaching the epoch at which such massive structures take shapes from the evolution of the cosmic web. On-going extensive spectroscopic surveys of such systems at $`z1`$ and beyond, combined with detailed analyses of their gaseus and dark matter components (now possible with weak lensing analysis of HST-ACS data; Jee et al. jee05 (2005); Lombardi et al. lom05 (2005)), will shed new light on cluster formation processes.
###### Acknowledgements.
We thank Andrea Biviano, Massimo Ramella, and Paolo Tozzi for useful discussions. We are grateful to the anonymous referee for helpful comments. This work has been partially supported by the Italian Space Agency (ASI), by the Istituto Nazionale di Astrofisica (INAF) through grant D4/03/IS, and by the Istituto Nazionale di Fisica Nucleare (INFN) through grant PD-51.
|
warning/0506/hep-th0506212.html
|
ar5iv
|
text
|
# Unifying phantom inflation with late-time acceleration: scalar phantom-non-phantom transition model and generalized holographic dark energy
## I Introduction
The recent astrophysical data indicate that effective equation of state parameter $`w_{\mathrm{eff}}`$ of dark energy lies in the interval: $`1.48<w_{\mathrm{eff}}<0.72`$ HM . In other words, it is quite possible that current universe lives (or enters) at effective phantom era (for review of observational data indicating to phantom late universe seeobse and last ref. from HM and for recent discussion of various approaches to late-time phantom cosmology, see phantom ; tsujikawa and references therein). However, it is not clear how to relate the late-time phantom cosmology with early-time inflation. For instance, the transition from decelerating phase to dark energy universe is not yet well understood (possibly because it is not clear what is dark energy itself). Nevertheless, there are attempts to unify the early time inflation where phantoms are essential with accelerated phantom universe unification . The unified inflation/acceleration universe occurs for some models of modified gravitymodified as well as for complicated, non-standard equation of state (EOS) for the universe salvatore (for recent discussion of such (phantomic) EOS, see EOS ; inh ). The attempts to use phantoms in early universe may be found also in inflation .
In the present work we suggest the scenario where within the same theory, quite naturally there occurs both phenomena: early-time phantom inflation and late-time phantom acceleration. The circles of phantom-dominated and non-phantom dominated epoch in such universe suggest that probably the universe is (partially) oscillating. In the next section we consider gravity-scalar theory with scalar-dependent coupling in front of kinetic term and scalar potential. We should note that the scalar factor in the kinetic term does not play any role each in the phontom or non-phantom phase and can be absorbed into the redefinition of the scalar field. Right on the transition point, however, the factor cannot be absorbed into the redefinition and play the role to connect two phases smoothly. In the number of explicit examples it is demonstrated how transitions between phantom and non-phantom phases occur and that two phases are smoothly connected with each other. As a result, there occurs the universe which contains at least two phantom phases corresponding to early time inflation and late time acceleration. The bridge between phantom phases correspond to the standard non-phantom cosmology (radiation/matter dominated, expanding or shrinking one). In oscillating universe there may emerge multiple phantom/non-phantom transitions (eras).
Section three is devoted to the study of generalized holographic dark energy where infrared cutoff depends on the combination of Hubble rate, the particle and future event horizons, life-time of the universe and even cosmological constant. Here, the analogy with AdS/CFT correspondence may be pointed out: there also IR or UV cutoffs represent some combination (depending on the order of the expansion in large N). It is shown that in such model quite naturally the crossing of phantom divide occurs. When including dark matter, the natural solution of coincidence problem follows. Finally, it is demonstrated that the unification of phantom inflation with phantom dark energy universe is also possible. Some summary and outlook are given in the last section.
## II Phantom inflation and late-time acceleration in scalar theory
In the present section we will discuss usual gravity with scalar field. The possibility to have unified phantom inflation with phantom late-time acceleration is shown. This occurs via phantom-non-phantom transition. Let us start from the following action:
$$S=d^4x\sqrt{g}\left\{\frac{1}{2\kappa ^2}R\frac{1}{2}\omega (\varphi )_\mu \varphi ^\mu \varphi V(\varphi )\right\}.$$
(1)
Here $`\omega (\varphi )`$ and $`V(\varphi )`$ are functions of the scalar field $`\varphi `$. Such scalar theory reminds about self-coupled dilaton naftulin or about sigma-model. The spatially-flat FRW metric is
$$ds^2=dt^2+a(t)^2\underset{i=1}{\overset{3}{}}\left(dx^i\right)^2.$$
(2)
The scalar field $`\varphi `$ only depends on the time coordinate $`t`$. Then the FRW equations are given by
$$\frac{3}{\kappa ^2}H^2=\rho ,\frac{2}{\kappa ^2}\dot{H}=p+\rho .$$
(3)
Here the energy density $`\rho `$ and the pressure $`p`$ are
$$\rho =\frac{1}{2}\omega (\varphi )\dot{\varphi }^2+V(\varphi ),p=\frac{1}{2}\omega (\varphi )\dot{\varphi }^2V(\varphi ).$$
(4)
By combining (3) and (4), one obtains
$$\omega (\varphi )\dot{\varphi }^2=\frac{2}{\kappa ^2}\dot{H},V(\varphi )=\frac{1}{\kappa ^2}\left(3H^2+\dot{H}\right).$$
(5)
The interesting case is that $`\omega (\varphi )`$ and $`V(\varphi )`$ are defined in terms of single function $`f(\varphi )`$ as
$$\omega (\varphi )=\frac{2}{\kappa ^2}f^{}(\varphi ),V(\varphi )=\frac{1}{\kappa ^2}\left(3f(\varphi )^2+f^{}(\varphi )\right).$$
(6)
Hence, the following solution may be presented
$$\varphi =t,H=f(t).$$
(7)
One can check the solution (7) satisfies the scalar field equation:
$$0=\omega (\varphi )\ddot{\varphi }+\frac{1}{2}\omega ^{}(\varphi )\dot{\varphi }^2+3H\omega (\varphi )\dot{\varphi }+V^{}(\varphi ).$$
(8)
Then any cosmology defined by $`H=f(t)`$ in (7) can be realized by (6).
As clear from the first equation (5), when $`\dot{H}`$ is positive, which corresponds to the phantom phase, $`\omega `$ should be negative, that is, the kinetic term of the scalar field has non-canonical sign. On the other hand, when $`\dot{H}`$ is negative, corresponding to the non-phantom phase, $`\omega `$ should be positive and the sign of the kinetic term of the scalar field is canonical. If we restrict in one of phantom or non-phantom phase, the function $`\omega (\varphi )`$ can be absorbed into the field redefinition given by
$$\phi =^\varphi ๐\varphi \sqrt{\omega (\varphi )},$$
(9)
in non-phantom phase or
$$\phi =^\varphi ๐\varphi \sqrt{\omega (\varphi )},$$
(10)
in phantom phase. Usually, at least locally, Eq.(9) or Eq.(10) can be solved with respect to $`\varphi `$ as $`\varphi =\varphi (\phi )`$. Then the action (1) can be rewritten as
$$S=d^4x\sqrt{g}\left\{\frac{1}{2\kappa ^2}R\frac{1}{2}_\mu \phi ^\mu \phi \stackrel{~}{V}(\phi )\right\}.$$
(11)
Here
$$\stackrel{~}{V}(\phi )V\left(\varphi (\phi )\right).$$
(12)
In the sign $``$ of (11), the minus sign corresponds to the non-phantom phase and the plus one to the phantom phase. Then both of $`\omega (\varphi )`$ and $`V(\varphi )`$ in the action (1) do not correspond to physical degrees of freedom but only one combination given by $`\stackrel{~}{V}(\phi )`$ has real freedom in each of the phantom or non-phantom phase and defines the real dynamics of the system. The redefinition (9) or (10), however, has a discontinuity between two phases. When explicitly keeping $`\omega (\varphi )`$, the two phases are smoothly connected with each other (kind of phase transitions). Hence, the function $`\omega (\varphi )`$ gives only redundant degree of freedom and does not correspond to the extra degree of freedom of the system ( in the phantom or non-phantom phase). It plays the important role just in the point of the transition between the phantom phase and non-phantom phase. By using the redundancy of $`\omega (\varphi )`$, in any physcally equivalent model, one may choose, just for example, $`\omega (\varphi )`$ as $`\omega (\varphi )=\omega _0\left(\varphi \varphi _0\right)`$ with constants $`\omega _0`$ and $`\varphi _0`$. If we further choose $`\omega _0`$ to be positive, the region given by $`\varphi >\varphi _0`$ corresponds to the non-phantom phase, the region $`\varphi <\varphi _0`$ to the phantom phase, and the point $`\varphi =\varphi _0`$ to the point of the transition between two phases.
Since the second FRW equation is given by
$$p=\frac{1}{\kappa ^2}\left(2\dot{H}+3H^2\right),$$
(13)
by combining the first FRW equation, the effective equation of state parameter $`w_{\mathrm{eff}}`$ looks as
$$w_{\mathrm{eff}}=\frac{p}{\rho }=1\frac{2\dot{H}}{3H^2}.$$
(14)
After this discussion, one may consider some toy models to realize above phantom-non-phantom transition to unify the phantom inflation with late-time phantom acceleration.
As a first example, we consider the following model
$$f(\varphi )=\frac{\alpha }{3}(T_0+\varphi )^3\beta (T_0+\varphi )+\gamma ,\gamma \frac{\alpha }{3}T_0^3+\beta T_0,$$
(15)
with the constants $`\alpha `$, $`\beta `$, and $`T_0`$, which give
$`\omega (\varphi )`$ $`=`$ $`{\displaystyle \frac{2}{\kappa ^2}}\left\{\alpha (T_0+\varphi )^2\beta \right\},`$
$`V(\varphi )`$ $`=`$ $`{\displaystyle \frac{3}{\kappa ^2}}\{{\displaystyle \frac{\alpha ^2}{3}}(T_0+\varphi )^62\alpha \beta (T_0+\varphi )^4+\alpha \gamma (T_0+\varphi )^3`$ (16)
$`+(\alpha +3\beta ^2)(T_0+\varphi )^22\beta \gamma (T_0+\varphi )+3\gamma ^2\}.`$
Then the solution can be given by
$$H=\frac{\alpha }{3}(T_0+t)^3\beta (T_0+t)+\gamma ,\varphi =t,$$
(17)
or
$$a=a_0\mathrm{e}^{\frac{\alpha }{12}\left(T_0+t\right)^4\frac{\beta }{2}\left(T_0+t\right)^2+\gamma \left(T_0+t\right)}.$$
(18)
As $`H`$ vanishes at $`t=0`$, $`a`$ has a minimum there. Then the universe is shrinking when $`t<0`$ and expanding when $`t>0`$. Since
$$\dot{H}=\alpha (T_0+t)^2\beta ,$$
(19)
$`\dot{H}`$ vanishes at
$$t=t_\pm T_0\pm \sqrt{\frac{\beta }{\alpha }}>0.$$
(20)
Hence, $`w_{\mathrm{eff}}`$ (14) is greater than $`1`$ when $`t_{}<t<t_+`$ (non-phantom phase) and less than $`1`$ when $`0<t<t_{}`$ or $`t>t_+`$ (phantom phase). There occurs the phantom inflation when $`0<t<t_{}`$ and late-time acceleration when $`t>t_+`$. We should also note there does not occur the Big Rip singularity in the solution (18) and $`w_{\mathrm{eff}}`$ goes to $`1`$ in the limit of $`t\mathrm{}`$. Thus, the model (II) may provide a unification of the inflation generated by phantom and the late time phantom acceleration of the universe.
As a second example, we consider the model given by
$$f(\varphi )=h_0+h_1\mathrm{sin}(\nu \varphi ),$$
(21)
with constants $`h_0`$, $`h_1`$, and $`\nu `$, which give
$`\omega (\varphi )`$ $`=`$ $`{\displaystyle \frac{2h_1\nu }{\kappa ^2}}\mathrm{cos}(\nu \varphi ),`$
$`V(\varphi )`$ $`=`$ $`{\displaystyle \frac{3}{\kappa ^2}}\left(3h_0^2+6h_0h_1\mathrm{sin}(\nu \varphi )+h_1\nu \mathrm{cos}(\nu \varphi )+h_1^2\mathrm{sin}^2(\nu \varphi )\right).`$ (22)
The Hubble rate $`H`$ is given by
$$H=h_0+h_1\mathrm{sin}(\nu t),$$
(23)
which is oscillating. When $`h_0>h_1>0`$, $`H`$ is always positive and the universe is expanding. Since
$$\dot{H}=h_1\nu \mathrm{cos}(\nu t),$$
(24)
when $`h_1\nu >0`$, $`w_{\mathrm{eff}}`$ (14) is greater than $`1`$ (non-phantom phase) when
$$\left(2n\frac{1}{2}\right)\pi <\nu t<\left(2n+\frac{1}{2}\right)\pi ,$$
(25)
and less than $`1`$ (phantom phase) when
$$\left(2n+\frac{1}{2}\right)\pi <\nu t<\left(2n+\frac{3}{2}\right)\pi .$$
(26)
In (25) and (26), $`n`$ is an integer. Hence, in the model (II), there occur multiply oscillations between phantom and non-phantom phases. It could be that our universe currently corresponds to late-time acceleration phase in such oscillatory regime.
The third example is given by scalar function :
$$f(\varphi )=h_0\left(\frac{1}{\varphi }+\frac{1}{t_s\varphi }\right),$$
(27)
with constants $`h_0`$ and $`t_s`$, which give
$`\omega (\varphi )`$ $`=`$ $`{\displaystyle \frac{2h_0t_s\left(2\varphi t_s\right)}{\kappa ^2\varphi ^2\left(t_s\varphi \right)^2}},`$
$`V(\varphi )`$ $`=`$ $`{\displaystyle \frac{h_0t_s\left\{2\varphi +\left(3h_01\right)t_s\right\}}{\kappa ^2\varphi ^2\left(t_s\varphi \right)^2}}.`$ (28)
Then the Hubble rate and the scale factor $`a`$ are given by,
$$H=h_0\left(\frac{1}{t}+\frac{1}{t_st}\right),a=a_0\left(\frac{t}{t_st}\right)^{h_0},$$
(29)
This was obtained from the two scalars model in tsujikawa . As $`a=0`$ at $`t=0`$, the universe starts at $`t=0`$. Note that there is a Big Rip type singularity at $`t=t_s`$. Since
$$\dot{H}=\frac{h_0t_s\left(2tt_s\right)}{t^2\left(t_st\right)^2},$$
(30)
when $`0<t<t_s/2`$, the universe is in non-phantom phase but when $`t_s/2<t<t_s`$, it is in phantom phase. In fact $`w_{\mathrm{eff}}`$ (14) is greater than $`1`$ when $`0<t<t_s/2`$ and less than $`1`$ when $`t_s/2<t<t_s`$, Hence, again the unified phantom inflation/acceleration universe may emerge.
As a fourth example, we now consider
$$f(\varphi )=h_0^2\left(\frac{1}{t_0^2\varphi ^2}+\frac{1}{\varphi ^2+t_1^2}\right).$$
(31)
Here $`h_0`$, $`t_0`$, and $`t_1`$ are positive constants.It is assumed $`t_0>t_1`$. Then $`\omega (\varphi )`$ and $`V(\varphi )`$ are
$`\omega (\varphi )`$ $`=`$ $`{\displaystyle \frac{8h_0^2\left(t_0^2+t_1^2\right)\varphi \left(\varphi ^2\frac{t_0^2t_1^2}{2}\right)}{\kappa ^2\left(t_0^2\varphi ^2\right)^2\left(\varphi ^2t_1^2\right)^2}},`$
$`V(\varphi )`$ $`=`$ $`{\displaystyle \frac{h_0^2\left(t_0^2+t_1^2\right)}{\kappa ^2\left(t_0^2\varphi ^2\right)^2\left(\varphi ^2t_1^2\right)^2}}\left\{3h_0^2\left(t_0^2+t_1^2\right)+4\varphi \left(\varphi ^2{\displaystyle \frac{t_0^2t_1^2}{2}}\right)\right\}.`$ (32)
The Hubble rate $`H`$ and the scale factor $`a(t)`$ follow as
$`H`$ $`=`$ $`h_0^2\left({\displaystyle \frac{1}{t_0^2t^2}}+{\displaystyle \frac{1}{t^2+t_1^2}}\right),`$
$`a`$ $`=`$ $`a_0\left({\displaystyle \frac{t+t_0}{t_0t}}\right)^{\frac{h_0^2}{2t_0}}\mathrm{e}^{\frac{h_0^2}{t_1}\mathrm{Arctan}\frac{t_1}{t}}.`$ (33)
Since $`a=0`$ at $`t=t_0`$, one may regard $`t=t_0`$ corresponds to the creation of the universe. Since
$$\dot{H}=\frac{4h_0^2\left(t_0^2+t_1^2\right)t\left(t^2\frac{t_0^2t_1^2}{2}\right)}{\left(t_0^2t^2\right)^2\left(t^2t_1^2\right)^2},$$
(34)
we find $`H`$ has two minimum at $`t=t_\pm \pm \sqrt{\frac{t_0^2t_1^2}{2}}`$ and at $`t=0`$, $`H`$ has a local maximum. Hence, phantom phase occurs when $`t_{}<t<0`$ and $`t>t_+`$ and non-phantom phase when $`t_0>t>t_{}`$ and $`0<t<t_+`$. We also note that there is a Big Rip type singularity at $`t=t_0`$.
As is discussed in CNO , the solution (7) is stable in the phantom phase but unstable in the non-phantom phase. The instablity becomes very large when crossing $`w=1`$. In order to avoid this problem, one may consider two scalar model. In case of one scalar model, the large instability occurs since the coefficient of the kinetic term $`\omega (\varphi )`$ in (1) vanishes at the crossing $`w=1`$ point. In the two scalar model, we can choose the corresponding coefficients do not vanish anywhere. Then we may expect that such a divergence of the instability would not occur, which we now check explicitly.
We now consider two scalar model like
$$S=d^4x\sqrt{g}\left\{\frac{1}{2\kappa ^2}R\frac{1}{2}\omega (\varphi )_\mu \varphi ^\mu \varphi \frac{1}{2}\eta (\chi )_\mu \chi ^\mu \chi V(\varphi ,\chi )\right\}.$$
(35)
Here $`\eta (\chi )`$ is a function of the scalar field $`\chi `$. The FRW equations give
$$\omega (\varphi )\dot{\varphi }^2+\eta (\chi )\dot{\chi }^2=\frac{2}{\kappa ^2}\dot{H},V(\varphi ,\chi )=\frac{1}{\kappa ^2}\left(3H^2+\dot{H}\right).$$
(36)
Then if
$$\omega (t)+\eta (t)=\frac{2}{\kappa ^2}f^{}(t),V(t,t)=\frac{1}{\kappa ^2}\left(3f(t)^2+f^{}(t)\right),$$
(37)
the explicit solution follows
$$\varphi =\chi =t,H=f(t).$$
(38)
One may choose that $`\omega `$ should be always positive and $`\eta `$ be always negative, for example
$`\omega (\varphi )`$ $`=`$ $`{\displaystyle \frac{2}{\kappa ^2}}\left\{f^{}(\varphi )\sqrt{\alpha ^2+f^{}(\varphi )^2}\right\}>0,`$
$`\eta (\chi )`$ $`=`$ $`{\displaystyle \frac{2}{\kappa ^2}}\sqrt{\alpha ^2+f^{}(\chi )^2}<0.`$ (39)
Here $`\alpha `$ is a constant. Define a new function $`\stackrel{~}{f}(\varphi ,\chi )`$ by
$$\stackrel{~}{f}(\varphi ,\chi )\frac{\kappa ^2}{2}\left(๐\varphi \omega (\varphi )+๐\chi \eta (\chi )\right),$$
(40)
which gives
$$\stackrel{~}{f}(t,t)=f(t).$$
(41)
If $`V(\varphi ,\chi )`$ is given by using $`\stackrel{~}{f}(\varphi ,\chi )`$ as
$$V(\varphi ,\chi )=\frac{1}{\kappa ^2}\left(3\stackrel{~}{f}(\varphi ,\chi )^2+\frac{\stackrel{~}{f}(\varphi ,\chi )}{\varphi }+\frac{\stackrel{~}{f}(\varphi ,\chi )}{\chi }\right),$$
(42)
the FRW and the scalar field equations are also satisfied:
$`0`$ $`=`$ $`\omega (\varphi )\ddot{\varphi }+{\displaystyle \frac{1}{2}}\omega ^{}(\varphi )\dot{\varphi }^2+3H\omega (\varphi )\dot{\varphi }+{\displaystyle \frac{\stackrel{~}{V}(\varphi ,\chi )}{\varphi }},`$
$`0`$ $`=`$ $`\eta (\chi )\ddot{\chi }+{\displaystyle \frac{1}{2}}\eta ^{}(\chi )\dot{\chi }^2+3H\eta (\chi )\dot{\chi }+{\displaystyle \frac{\stackrel{~}{V}(\varphi ,\chi )}{\chi }}.`$ (43)
In case of one scalar model, the instability becomes infinite at the crossing $`w=1`$ point (from higher than phantom value), since the coefficient of the kinetic term $`\omega (\varphi )`$ in (1) vanishes at the point. In the two scalar model in (35), the coefficients $`\omega (\varphi )`$ and $`\eta (\varphi )`$ do not vanish anywhere, as in (II). Then we may expect that such a divergence of the instability does not occur.
By introducing the new quantities, $`X_\varphi `$, $`X_\chi `$, and $`Y`$ as
$$X_\varphi \dot{\varphi },X_\chi \dot{\chi },Y\frac{\stackrel{~}{f}(\varphi ,\chi )}{H},$$
(44)
the FRW equations and the scalar field equations (II) are:
$`{\displaystyle \frac{dX_\varphi }{dN}}`$ $`=`$ $`{\displaystyle \frac{\omega ^{}(\varphi )}{2H\omega (\varphi )}}\left(X_\varphi ^21\right)3(X_\varphi Y),`$
$`{\displaystyle \frac{dX_\chi }{dN}}`$ $`=`$ $`{\displaystyle \frac{\eta ^{}(\chi )}{2H\eta (\chi )}}\left(X_\chi ^21\right)3(X_\chi Y),`$
$`{\displaystyle \frac{dY}{dN}}`$ $`=`$ $`{\displaystyle \frac{1}{2\kappa ^2H^2}}\left\{X_\varphi \left(X_\varphi Y1\right)+X_\chi \left(X_\chi Y1\right)\right\}.`$ (45)
Here $`d/dNH^1d/dt`$. In the solution (38), $`X_\varphi =X_\chi =Y=1`$. The following perturbation may be considered
$$X_\varphi =1+\delta X_\varphi ,X_\chi =1+\delta X_\chi ,Y=1+\delta Y.$$
(46)
Hence
$$\frac{d}{dN}\left(\begin{array}{c}\delta X_\varphi \\ \delta X_\chi \\ \delta Y\end{array}\right)=M\left(\begin{array}{c}\delta X_\varphi \\ \delta X_\chi \\ \delta Y\end{array}\right),M\left(\begin{array}{ccc}\frac{\omega ^{}(\varphi )}{H\omega (\varphi )}3& 0& 3\\ 0& \frac{\eta ^{}(\chi )}{H\eta (\chi )}3& 3\\ \frac{1}{2\kappa ^2H^2}& \frac{1}{2\kappa ^2H^2}& \frac{1}{\kappa ^2H^2}\end{array}\right).$$
(47)
The eigenvalues of the matrix $`M`$ are given by solving the following eigenvalue equation
$`0`$ $`=`$ $`\left(\lambda +{\displaystyle \frac{\omega ^{}(\varphi )}{H\omega (\varphi )}}+3\right)\left(\lambda +{\displaystyle \frac{\eta ^{}(\chi )}{H\eta (\chi )}}+3\right)\left(\lambda {\displaystyle \frac{1}{\kappa ^2H^2}}\right)`$ (48)
$`+{\displaystyle \frac{3}{2\kappa ^2H^2}}\left(\lambda +{\displaystyle \frac{\omega ^{}(\varphi )}{H\omega (\varphi )}}+3\right)+{\displaystyle \frac{3}{2\kappa ^2H^2}}\left(\lambda +{\displaystyle \frac{\eta ^{}(\chi )}{H\eta (\chi )}}+3\right).`$
The eigenvalues (48) for the two scalar model are clearly finite. Hence, the instability could be finite. In fact, right on the transition point where $`\dot{H}=f^{}(t)=0`$ and therefore $`f^{}(\varphi )=f^{}(\chi )=0`$, for the choice in (II), we find
$$\omega (\varphi )=\eta (\chi )=\frac{2\alpha }{\kappa ^2},\omega ^{}(\varphi )=\frac{2\ddot{H}}{\kappa ^2},\eta ^{}(\chi )=0.$$
(49)
Then the eignvalue equation (48) reduces to
$$0=\lambda ^3+\left(AB+6\right)\lambda ^2+\left(AB3A3B+9\right)\lambda \frac{3}{2}AB+9B,A\frac{\ddot{H}}{\alpha },B\frac{1}{\kappa ^2H^2}.$$
(50)
Here we have chosen $`\alpha >0`$. Then the eignevalues are surely finite, which tells that even if the solution (38) could not be stable, the solution has non-vanishing measure and therefore the transition from non-phantom phase to phantom one can surely occur. We should also note that the solution (38) can be in fact stable. For example, we consider the case $`A,B0`$. Then Eq.(50) further reduces to
$$0=\lambda \left(\lambda +3\right)^2.$$
(51)
Then the eignvalues are given by $`0`$ and $`3`$. Since there is no positive eigenvalue, the solution (38) is stable in the case.
As an example, we consider $`f(t)=h_0+h_1\mathrm{sin}(\nu t)`$ in (21). Here it is assumed $`h_0`$, $`h_1`$, and $`\nu `$ are positive. By choosing $`\alpha =h_1\nu `$ in (II), one finds
$`\omega (\varphi )={\displaystyle \frac{2h_1\nu }{\kappa ^2}}\left\{\mathrm{cos}(\nu \varphi )\sqrt{1+\mathrm{cos}^2(\nu \varphi )}\right\},\eta (\chi )={\displaystyle \frac{2h_1}{\nu }}\kappa ^2\sqrt{1+\mathrm{cos}^2(\nu \chi )},`$
$`\stackrel{~}{f}(\varphi ,\chi )=h_0+h_1\mathrm{sin}(\nu \varphi ){\displaystyle \frac{\sqrt{2}}{\nu }}\left\{E(1/\sqrt{2},\nu \varphi )E(1/\sqrt{2},\nu \chi )\right\}.`$ (52)
Here $`E(k,x)`$ is the second kind elliptic integral defined by
$$E(k,x)=_0^x๐x\sqrt{1k^2\mathrm{sin}^2x}.$$
(53)
Then even in two scalar model, the cosmology is given by (23). Similarly any cosmology (including unified inflation/acceleration) can be realized by using the two scalar model, as in the examples with single scalar field.
Thus, we presented several toy models showing the natural possibility to unify early-time inflation with late-time acceleration via phantom-non-phantom transitions in scalar theory. Much work remains to be done in order to decide if such theoretic possibility is realistic one. In next section, we demonstrate that similar effect is possible also in generalized holographic dark energy.
## III Generalized holographic dark energy and unification of phantom inflation with phantom acceleration.
Let us start from the holographic dark energy model Li (see also refs.FEH where further support for holographic DE was given). Denote the infrared cutoff by $`L_\mathrm{\Lambda }`$, which has a dimension of length. If the holographic dark energy $`\rho _\mathrm{\Lambda }`$ is given by,
$$\rho _\mathrm{\Lambda }=\frac{3c^2}{\kappa ^2L_\mathrm{\Lambda }^2},$$
(54)
with a numerical constant $`c`$, the first FRW equation
$$\frac{3}{\kappa ^2}H^2=\rho _\mathrm{\Lambda },$$
(55)
can be written as
$$H=\frac{c}{L_\mathrm{\Lambda }}.$$
(56)
Here it is assumed that $`c`$ is positive to assure the expansion of the universe. In (55), we do not include the contribution from the matter. The next question is the choice of infrared cut-off. For instance, identifying it with Hubble parameter does not lead to accelerating universe. Hence, one is forced to consider other choices.
The particle horizon $`L_p`$ and future horizon $`L_f`$ are defined by
$$L_pa_0^t\frac{dt}{a},L_fa_t^{\mathrm{}}\frac{dt}{a}.$$
(57)
For the FRW metric with the flat spacial part:
$$ds^2=dt^2+a(t)^2\underset{i=1,2,3}{}\left(dx^i\right)^2.$$
(58)
Identifying $`L_\mathrm{\Lambda }`$ with $`L_p`$ or $`L_f`$, one obtains the following equation:
$$\frac{d}{dt}\left(\frac{c}{aH}\right)=\pm \frac{1}{a}.$$
(59)
Here, the plus (resp. minus) sign corresponds to the particle (resp. future) horizon. The solution of (59) is given by
$$a=a_0t^{h_0},$$
(60)
with
$$h_0=\frac{1}{1\pm \frac{1}{c}}.$$
(61)
Then, in the case $`L_\mathrm{\Lambda }=L_f`$, the universe is accelerating ($`h_0>1`$ or $`w=1+2/3h_0<1/3`$). When $`c>1`$ in the case $`L_\mathrm{\Lambda }=L_p`$, $`h_0`$ becomes negative and the universe is shrinking. If the theory is invariant under the change of the direction of time, one may change $`t`$ with $`t`$. Furthermore by properly shifting the origin of time, we obtain, instead of (60),
$$a=a_0\left(t_st\right)^{h_0}.$$
(62)
This tells us that there will be a Big Rip singularity at $`t=t_s`$ (for classification of future, finite-time singularities and list of related references, see tsujikawa ). Since the direction of time is changed, the particle horizon becomes a future like one:
$$L_p\stackrel{~}{L}_fa_t^{t_s}\frac{dt}{a}=a_0^{\mathrm{}}\frac{da}{Ha^2}.$$
(63)
By using (14) for (60) and (62), we find
$$w_{\mathrm{eff}}=1+\frac{2}{3h_0}.$$
(64)
Note that if $`L_\mathrm{\Lambda }`$ is chosen as a future horizon in (57), the deSitter space
$$a=a_0\mathrm{e}^{\frac{t}{l}}\left(H=\frac{1}{l}\right)$$
(65)
can be a solution. Since $`L_f`$ is now given by $`L_f=l`$, the holographic dark energy (54) is given by $`\rho _\mathrm{\Lambda }=\frac{c^2}{\kappa ^2l^2}`$. When $`c=1`$, the first FRW equation $`\frac{3}{\kappa ^2}H^2=\rho _\mathrm{\Lambda }`$ is identically satisfied. If $`c1`$, the deSitter space is not a solution. If $`L_\mathrm{\Lambda }`$ is chosen to be the particle horizon, the deSitter solution does not exist, either, since the particle horizon $`L_p`$ (57) is not a constant: $`L_p=(1\mathrm{e}^{\frac{t}{l}})/l`$. Hence, the essentials of holographic dark energy are discussed.
In general, $`L_\mathrm{\Lambda }`$ could be a combination (a function) of both, $`L_p`$, $`L_f`$eli . Furthermore, if the span of life of the universe is finite, the span $`t_s`$ can be an infrared cutoff. If the span of life of the universe is finite, the definition of the future horizon $`L_f`$ (57) is not well-posed, since $`t`$ cannot go to infinity. Then, one may redefine the future horizon as in (63)
$$L_f\stackrel{~}{L}_fa_t^{t_s}\frac{dt}{a}=a_0^{\mathrm{}}\frac{da}{Ha^2}.$$
(66)
Since there can be many choices for the infrared cutoff, in analogy with AdS/CFT one may assume $`L_\mathrm{\Lambda }`$ is the function of $`L_p`$, $`\stackrel{~}{L}_f`$, and $`t_s`$, as long as these quantities are finite:
$$L_\mathrm{\Lambda }=L_\mathrm{\Lambda }(L_p,\stackrel{~}{L}_f,t_s).$$
(67)
As an example, we consider the generalized holographic dark energy from above class eli
$$\frac{L_\mathrm{\Lambda }}{c}=\frac{\left(\frac{t_sB(1+h_0,1h_0)}{L_p+\stackrel{~}{L}_f}\right)^{1/h_0}}{h_0\left\{1+\left(\frac{t_sB(1+h_0,1h_0)}{L_p+\stackrel{~}{L}_f}\right)^{1/h_0}\right\}^2},h_0>0.$$
(68)
Here $`B(p,q)`$ is a beta-function defined by
$$B(p,q)_0^{\mathrm{}}\frac{dtt^{p1}}{\left(1+t\right)^{p+q}}.$$
(69)
Eq.(68) leads to the solution:
$$H=h_0\left(\frac{1}{t}+\frac{1}{t_st}\right),\text{or}a=a_0\left(\frac{t}{t_st}\right)^{h_0}.$$
(70)
In fact, one finds
$$L_p+\stackrel{~}{L}_f=a_0^{t_s}\frac{dt}{a}=t_s\left(\frac{t}{t_st}\right)^{h_0}B(1+h_0,1h_0),$$
(71)
and therefore
$$\frac{c}{L_\mathrm{\Lambda }}=h_0\left(\frac{1}{t}+\frac{1}{t_st}\right)=H,$$
(72)
which satisfies (56). For the solution (70), $`w_{\mathrm{eff}}`$ defined in (14) is time-dependent and looks as
$$w_{\mathrm{eff}}=1+\frac{2\left(t_s2t\right)}{3h_0t_s}.$$
(73)
Then $`w_{\mathrm{eff}}=1`$ at $`t=t_s/2`$ and we find $`w1+2/(3h_0)>1`$ when $`t0`$ and $`w_{\mathrm{eff}}12/(3h_0)<1`$ when $`tt_s`$. Hence, there occurs the crossing of $`w_{\mathrm{eff}}=1`$ in our generalized holographic dark energy model.
One can also include the matter whose parameter of the equation of state is $`w`$: $`\rho _m=wp_m`$ into the consideration. In the following, we define $`h_0`$ by $`h_0(2/3)/(1+w)`$. However, it is assumed that there is an interaction between the holographic mattersamendola . The energy-conservation law is taken as follows
$$\dot{\rho }_m+3H\left(\rho _m+p_m\right)=3H\frac{4\rho _0}{3h_0}\frac{\left\{1+\left(\frac{t_sB(1+h_0,1h_0)}{L_p+\stackrel{~}{L}_f}\right)^{1/h_0}\right\}^3}{\left(\frac{t_sB(1+h_0,1h_0)}{L_p+\stackrel{~}{L}_f}\right)^{2/h_0}}.$$
(74)
It is also assumed that
$$\frac{L_\mathrm{\Lambda }}{c}=\left(1\frac{\kappa ^2\rho _0}{3h_0^2}\right)^{1/2}\frac{\left(\frac{t_sB(1+h_0,1h_0)}{L_p+\stackrel{~}{L}_f}\right)^{1/h_0}}{h_0\left\{1+\left(\frac{t_sB(1+h_0,1h_0)}{L_p+\stackrel{~}{L}_f}\right)^{1/h_0}\right\}^2},$$
(75)
The first FRW equation is then modified as $`(H^2/\kappa ^2)=\rho _\mathrm{\Lambda }+\rho _m`$ and we find (70) again and
$$\rho _m=\rho _0\left(\frac{1}{t}+\frac{1}{t_st}\right)^2.$$
(76)
Then the ratio of the energy density $`\rho _m`$ of matter with respect to that of the holographic dark energy corresponding to (54) is a constant:
$$\frac{\rho _\mathrm{\Lambda }}{\rho _m}=\frac{3h_0^2}{\kappa ^2\rho _0}\left(1\frac{\kappa ^2\rho _0}{3h_0^2}\right).$$
(77)
Hence, one sees that coincidence problem may be solved in generalized holographic dark energy. Note that similar scenario was proposed in PZ , where the ratio of the matter energy density and the holographic energy can be a constant by introducing the interaction, as in (74), between the matter and the holographic energy. For the naive model PZ , the effective $`w_{\mathrm{eff}}`$ (14) is constant. As shown here, even if $`w_{\mathrm{eff}}`$ depends on time, the ratio can be constant. In more general case the ratio between the matter energy density and the dark energy density is not constant WGA .
As more general case than (67), we may consider the case that $`L_\mathrm{\Lambda }`$ depends on the Hubble rate $`H`$ and the length scale $`l`$ coming from the cosmological constant $`\mathrm{\Lambda }=12/l^2`$, if the cosmological constant does not vanish:
$$L_\mathrm{\Lambda }=L_\mathrm{\Lambda }(L_p,\stackrel{~}{L}_f,t_s,H,l)\text{or}L_\mathrm{\Lambda }(L_p,L_f,t_s,H,l).$$
(78)
As an example, the generalized holographic dark energy theory from such class may be considered
$$\frac{c}{L_\mathrm{\Lambda }}=\frac{1}{\alpha L_f}\left\{\alpha +1+2\left(\alpha ^2\alpha 1\right)\frac{L_f}{\alpha l}+2\left(\alpha ^32\alpha ^2+\alpha +1\right)\left(\frac{L_f}{\alpha l}\right)^2\right\}.$$
(79)
Here $`\alpha `$ is a positive dimensionless parameter. Since
$`\left(\alpha ^2\alpha 1\right)^22\left(\alpha ^32\alpha ^2+\alpha +1\right)=\left(\alpha ^2{\displaystyle \frac{1}{2}}\right)^22\alpha {\displaystyle \frac{3}{4}}<0`$
$`\alpha ^32\alpha ^2+\alpha +1=\alpha (\alpha 1)^2+1>0,`$ (80)
$`L_\mathrm{\Lambda }`$ is always positive as long as $`\alpha `$ is positive. Then a cosmological solution is given by
$$a=\frac{t^{\alpha +1}\mathrm{e}^{\frac{t}{l}}}{L_0\alpha \left(1+\frac{t}{\alpha l}\right)},L_f=\frac{t}{\alpha \left(1+\frac{t}{\alpha l}\right)},$$
(81)
which leads to
$$H=\frac{1+\alpha \left(1+\frac{t}{\alpha l}\right)^2}{t\left(1+\frac{t}{\alpha l}\right)}.$$
(82)
As clear from (81), if $`\alpha <0`$, there occurs the Big Rip like singularity at $`t=\alpha l`$, where $`a`$ diverges. When $`\alpha <0`$, $`L_\mathrm{\Lambda }`$ can be negative, in such a case as $`H=\frac{c}{L_\mathrm{\Lambda }}`$ if we do not include the matter, the universe is shrinking.
In (81), $`L_0`$ is a constant of the integration. When $`t`$ is small, $`H`$ behaves as the inverse power of $`t`$:
$$H\frac{\alpha +1}{t},$$
(83)
which shows that the universe is filled with the fluid with $`w=(\alpha 1)/(3(\alpha +1))`$. On the other hand, when $`t`$ is large, $`H`$ goes to a constant:
$$H\frac{1}{l},$$
(84)
which tells that the universe becomes deSitter space asymptotically. Instead of (79), one may consider the model as
$$\frac{c}{L_\mathrm{\Lambda }}=\frac{H}{\alpha +1}+\frac{1}{L_f}\left\{1+\frac{2\left(\alpha ^2\alpha 1\right)}{\alpha +1}\frac{L_f}{\alpha l}+\left(1\frac{\alpha ^2(\alpha +2)}{\alpha +1}\right)\left(\frac{L_f}{\alpha l}\right)^2\right\}.$$
(85)
Then the solution (81) follows again.
In general, the FRW equation has the following form:
$$\frac{3}{\kappa ^2}H^2=\frac{3c^2}{\kappa ^2L_\mathrm{\Lambda }^2}+\rho _m.$$
(86)
If one defines the matter energy $`E_m`$, the Casimir energy $`E_c`$, and the Hubble entropy $`S_H`$ by
$$E_m\rho _mL_\mathrm{\Lambda }^3,E_c\frac{3c^2L_\mathrm{\Lambda }}{\kappa ^2},S_H\frac{18\pi cL_\mathrm{\Lambda }^3}{\kappa ^2}H,$$
(87)
we can rewrite the FRW equation in a Cardy-Verlinde (holographic) form verlinde :
$$S_H^2=\left(2\pi L_\mathrm{\Lambda }\right)^2E_c\left(E_c+E_m\right).$$
(88)
Different from the deSitter case KeLi , the Hubble entropy $`S_H`$ is not a constant and depends on time. In case $`\rho _m=0`$, by using (56) one has
$$S_H=\frac{18\pi c^2L_\mathrm{\Lambda }^2}{\kappa ^2}=\frac{18\pi c^4}{\kappa ^2H^2}.$$
(89)
Thus for the model (68), using (70), we find
$$S_H=\frac{18\pi c^4t^2\left(t_st\right)^2}{\kappa ^2h_0^2t_s^2},$$
(90)
which vanishes at $`t=0`$ and $`t=t_s`$. The maximum of $`S_H`$ (90) is obtained when $`t=t_s/2`$. For the generalized models (74) and (75), where matter is included, $`S_H`$ (90) is modified by a constant factor:
$$S_H=\left(1\frac{\kappa ^2\rho _0}{3h_0^2}\right)^3\frac{18\pi c^4t^2\left(t_st\right)^2}{\kappa ^2h_0^2t_s^2}.$$
(91)
In case of (79) or (86), it follows
$$S_H=\frac{18\pi c^4t^2\left(1+\frac{t}{\alpha l}\right)^2}{\kappa ^2\left\{1+\alpha \left(1+\frac{t}{\alpha l}\right)^2\right\}^2},$$
(92)
which vanishes at $`t=0`$ and goes to a constant
$$S_H\frac{18\pi c^4l^2}{\kappa ^2},$$
(93)
when $`t\mathrm{}`$. In (92), $`\alpha `$ can be negative in general but $`S_H`$ is positive. Although the Hubble entropy $`S_H`$ is always positive in (89), $`S_H`$ (91) can be negative if $`\frac{\kappa ^2\rho _0}{3h_0^2}>1`$. In such a case, if $`S_H`$ gives a upper bound for the entropy of the universe, the entropy should be negative. Such a negative entropy has been observed for phantom era in BNOV . Nevertheless, when phantom era is transient in the late-time universe it may occur that the universe entropy remains to be positive inh .
As a little bit complicated example, we may consider
$$\frac{c}{L_\mathrm{\Lambda }}=\frac{\alpha }{3}T^3\beta T+\gamma .$$
(94)
Here $`\alpha `$, $`\beta `$, and $`\gamma `$ are positive constants satisfying
$$\gamma >\frac{2\beta }{3}\sqrt{\frac{\beta }{\alpha }},$$
(95)
and $`T`$ is defined by
$`T`$ $``$ $`{\displaystyle \frac{H\gamma +\frac{4\alpha }{3\beta ^2}\left(H+3\gamma \right)\mathrm{ln}\frac{L}{F}}{\frac{\alpha }{3\beta ^2}\left(H+3\gamma \right)^2\beta \frac{4\alpha }{3\beta }\mathrm{ln}\frac{L}{F}}},`$
$`L`$ $``$ $`{\displaystyle \frac{a(t)}{a(0)}}F,`$
$`F`$ $``$ $`{\displaystyle _{\mathrm{}}^{\mathrm{}}}\mathrm{e}^{\frac{\alpha }{12}t^4+\frac{\beta }{2}t^2\gamma t}๐t.`$ (96)
The condition (III) tells that, as a function of $`T`$, $`H`$ vanishes only once. By defining $`T_0<0`$ via $`H(T=T_0)=0`$, one gets
$$\gamma =\frac{\alpha }{3}T_0^3+\beta T_0.$$
(97)
Then the solution can be given by (17) or (18), again. Thus, the generalized dark energy model (94) may provide a unification of the inflation generated by phantom and the late time phantom acceleration of the universe.
In the model (94), the Hubble entropy $`S_H`$ (87) is given by
$$S_H=\frac{18\pi c^4}{\kappa ^2\left\{\frac{\alpha }{3}\left(t+T_0\right)^3\beta \left(t+T_0\right)+\gamma \right\}^2},$$
(98)
which is always positive. From (97), one finds $`S_H`$ diverges at $`t=0`$. Since $`S_H`$ is proportional to $`H^2`$, $`S_H`$ decreases when $`0<t<t_{}`$, increases when $`t_{}<t<t_+`$, and decreases again when $`t>t_+`$. In the limit $`t\mathrm{}`$, $`S_H`$ vanishes. Therefore, in the phantom phase, $`S_H`$ is decreasing and in the non-phantom phase, it is increasing. This finishes the discussion of holographic entropy bounds for generalized holographic dark energy model.
## IV Discussion
In summary, it is suggested the scenario where phantom cosmology may be key element at early-time as well as at late-time universe. Specific model under consideration suggests the phantom-non-phantom transitions during the evolution of the universe. This may be easily realized due to presence of scalar coupling in front of kinetic term: this coupling function may change its sign on cosmological scales. As a result, even multiply phantom-non-phantom transitions are possible. The intermediate region between very early and very late universe may correspond to standard (radiation/matter dominated) cosmology.
The generalized holographic dark energy where infrared cutoff is identified with combination of the natural FRW parameters: Hubble rate, particle and future horizons, span of life of the universe (when its life-time is finite) and even with cosmological constant is suggested. The possibility to have the crossing of phantom divide there, as well as solution of coincidence problem (when matter presents) is demonstrated. The holographic entropy bound is obtained and the regime where it may be negative is discussed. It is interesting that holographic entropy is decreasing in phantom phase in accord with earlier observation of ref.BNOV . Finally, the possibility to unify phantom inflation with late-time acceleration is demonstrated also in generalized holographic dark energy. It is clear that in similar way one can suggest unified cosmological scenario for tachyon phantoms and for time-dependent, phantomic equations of state.
Much work remains to be done in order to understand if such combined scenario is realistic one: the standard cosmological problems (especially at early universe) should be first discussed. In order to start such investigation one awaits the final answer to fundamental question: Do we currently live in phantom universe? Hopefully, the answer comes soon.
## Acknowledgments
The research by SDO has been partly supported by RFBR grant 06-01-00609 (Russia), by LRSS project n4489.2006.02 (Russia) and by project FIS2005-01181 (MEC, Spain).
|
warning/0506/math0506543.html
|
ar5iv
|
text
|
# Untitled Document
SPhT/03-148
Geodesic Distance in Planar Graphs: An Integrable Approach
P. Di Francesco<sup>1</sup> philippe@spht.saclay.cea.fr
Service de Physique Thรฉorique, CEA/DSM/SPhT
Unitรฉ de recherche associรฉe au CNRS
CEA/Saclay
91191 Gif sur Yvette Cedex, France
We discuss the enumeration of planar graphs using bijections with suitably decorated trees, which allow for keeping track of the geodesic distances between faces of the graph. The corresponding generating functions obey non-linear recursion relations on the geodesic distance. These are solved by use of stationary multi-soliton tau-functions of suitable reductions of the KP hierarchy. We obtain a unified formulation of the (multi-) critical continuum limit describing large graphs with marked points at large geodesic distances, and obtain integrable differential equations for the corresponding scaling functions. This provides a continuum formulation of two-dimensional quantum gravity, in terms of the geodesic distance.
10/03
1. Introduction
The work presented in this note was initially motivated by the need to better understand on a combinatorial level the various results obtained via matrix models on the enumeration of graphs with fixed topology (see also and and references therein). The early work on this subject dates back to a combinatorist, W. Tutte , who managed to enumerate many of the planar versions of these using recursion relations in the spirit of what we call today โloop equationsโ for matrix models. Such an approach, though combinatorial, failed to really explain the simplicity of the algebraic equations determining the generating functions for planar graphs. Another motivation comes from the physics of two-dimensional quantum gravity. At the discrete level, coupling matter to gravity simply amounts to define a statistical model (typically with local Boltzmann weights) on a fluctuating base space, in the form of random discretized surfaces. The continuum version of this involves field-theoretical descriptions of random surfaces with critical matter , via the coupling of conformal field theories (matter) to the Liouville field theory (metrics of the underlying space).
The interpretation of the planar graph results remained elusive until the groundbreaking work of G. Schaeffer , who finally gave a beautifully simple combinatorial explanation for these algebraic equations, in terms of decorated trees. The idea was simply to establish bijections between classes of planar graphs and suitably decorated trees, then easily enumerated via algebraic relations obeyed by their generating functions. This technique proved quite general, and was extended to many classes of planar graphs, including graphs of arbitrary valence , special classes of bipartite graphs called constellations and other classes of bipartite graphs , including the particular cases of hard objects on planar graphs and of the Ising model on planar graphs .
The great advantage of this bijective enumeration is that it allows for keeping track of some details of the graphs in the language of trees. An important example of this concerns the geodesic distance between say the vertices of random quadrangulations, namely planar graphs with only tetravalent faces. In , it was shown that the geodesic distance of all vertices from an origin vertex of the graph may translate into integer vertex labels in some corresponding trees, themselves realizing a discrete version of the Brownian snake, further studied and extended in in the language of spatial branching processes. In , it was shown that the generating functions for planar graphs with two external legs obeyed non-linear recursion relations on the maximal geodesic distance between the legs, and it is the purpose of this note to clarify and extend the results of this paper.
Note that no continuum (field-theoretical) treatment of two-dimensional quantum gravity in terms of geodesic distance is available yet, only some partial results were obtained using a transfer matrix formalism leading to some conjectural continuum expression for a scaling function of the geodesic distance in the case of pure gravity without matter. The present work provides an alternative solution and gives access to a host of scaling functions for various (multi-) critical models of matter systems coupled to two-dimensional quantum gravity.
The paper is organized as follows. In Sects. 2,3 and 4, we recall some known facts on the bijective enumeration of planar graphs with respectively even valences, arbitrary (even and odd) valences, and bicolored vertices. This relies on an iterative cutting procedure which, starting from a planar graph, produces a decorated rooted tree. In turn the tree may be closed back in a unique way into a planar graph, and we use this bijection to recover the algebraic relations satisfied by the various counting functions involved. Sect. 5 is devoted to the introduction of the geodesic distance in the various enumeration problems at hand: we present a unified picture involving formal operators $`Q`$ generating the descendent trees around a vertex, and allowing for keeping track of the geodesic distance from the root to the external face of the planar graph to which the tree closes back. The main results are recursion relations on the geodesic distance satisfied by the generating functions for planar graphs with legs. In Sect. 6, we solve exactly a number of these recursion relations, by expanding the generating functions at large maximal geodesic distance $`n`$, and resumming the resulting series. These display a remarkable โintegrableโ structure in that they generically involve tau-functions of the KP hierarchy. Sect. 7 is devoted to the critical continuum limit of the problem, in which we consider large planar graphs and large geodesic distances. The generating functions calculated in Sect. 6 are shown to yield universal scaling functions, characteristic of the various (multi-) critical points of random surfaces with matter and with marked points at a fixed geodesic distance. We propose a generalization of our results based on differential equations obeyed by the scaling functions, and illustrate it in the case of the Ising model on random surfaces. Finally, we gather a few concluding remarks in Sect. 8, where in particular we discuss the integrable structure of our recursion relations, and of their continuum counterparts.
2. Planar graphs and trees I: the case of even valences
In this section, we recall some results on the enumeration of planar graphs of even valence with two extra โlegsโ. This is done via a bijection between planar graphs and decorated trees, easily enumerated.
2.1. Tetravalent graphs
Fig.1: The bijection between planar tetravalent graphs with two legs (a) and rooted ternary blossom-trees (d) is obtained via the following cutting procedure. We first visit all edges of the graph in counterclockwise direction (b) and cut them iff the resulting graph is still connected: we have indicated the succession of cut edges by their number in the order of visit, from 1 to 6 here. The cut edges are then replaced by pairs of white and black leaves (c). Moreover the incoming leg is replaced by a white leaf and the outcoming one by a root, finally yielding a rooted ternary blossom-tree (d).
We wish to compute the generating function $`RR(g)`$ for planar tetravalent graphs with two distinguished (say in- and out-coming) univalent vertices, and with a weight $`g`$ per tetravalent vertex. These will be referred to as โtwo-leg diagramsโ in the following, the legs denoting simply the edges connecting the two univalent vertices to the graph. For definiteness, we will always represent these planar graphs with the incoming leg adjacent to the external face. Note that the outcoming one need not be adjacent to the same face, as in the example of Fig.1 (a).
The computation of $`R`$ relies on the following bijection, illustrated in Fig.1, between two-leg diagrams and so-called rooted blossom-trees . Starting from a two-leg diagram (Fig.1 (a)), let us first visit all edges adjacent to the external face, starting from the incoming leg and in counterclockwise order. Successively, each visited edge is cut iff the cut diagram remains connected (Fig.1 (b)). The cut ends of the edge are then decorated respectively with a black and a white leaf. Once all edges adjacent to the external face have been visited, we repeat the procedure with the newly cut diagram. The process ends when all faces of the original diagram have been merged with the external one. The resulting graph is nothing but a planar tree (it has only one face), with black and white leaves (Fig.1 (c)). Note that by construction there is exactly one black leaf connected to each internal vertex of the tree. Finally, we replace the incoming vertex by a white leaf and the outcoming one by a root, so that there is exactly one more white leaf than black ones (Fig.1 (d)). We define rooted blossom-trees as planar rooted ternary trees with black and white leaves, and such that there is exactly one black leaf at each internal vertex. Our cutting procedure has produced a rooted blossom-tree out of any two-leg diagram. The process however is readily seen to be invertible as there is a unique way of re-connecting the black-white leaf pairs into edges, by connecting each black leaf to the first available white leaf in counterclockwise order around the tree (the dashed lines of Fig.1 (c)). This establishes the desired bijection between the two objects.
Couting rooted blossom-trees is now an easy task, performed for instance by inspection of all possible environments of the vertex connected to the root. This leads straightforwardly to the relation
$$\begin{array}{cc}& \text{}\hfill \\ & R=1+3gR^2\hfill \end{array}$$
where the first term $`1`$ counts the possibility that the root be directly connected to a leaf, and the second term accounts for the three possible positions for the black leaf around this vertex, itself receiving the weight $`g`$, while the two other descendents of the vertex are themselves rooted blossom-trees. From its very definition as counting function, $`R`$ admits a power series expansion in $`g`$, with $`R=1+O(g)`$. This fixes it uniquely to be
$$R=\frac{1\sqrt{112g}}{6g}$$
The series for $`R`$ has a finite convergence radius $`g_c=1/12`$. When $`g`$ approaches $`g_c`$ (critical limit), the contribution of large graphs becomes dominant, and we learn that the number of graphs with $`N`$ vertices behaves as $`g_c^N/N^{3/2}`$ for large $`N`$.
Note that $`R(g)=C(3g)`$ where $`C`$ denotes the generating function for Catalan numbers $`c_N=\left(\genfrac{}{}{0pt}{}{2N}{N}\right)/(N+1)`$, which count among other things the rooted planar binary trees with $`N`$ inner vertices, with the convention that $`c_0=1`$. The number of rooted blossom trees with $`N`$ inner vertices is obtained by considering rooted planar binary trees with all leaves white and by decorating each vertex with a black leaf: it reads therefore $`3^Nc_N`$ as there are three choices for the position of the black leaf at each inner vertex.
2.2. General case of graphs with even valences
The bijection of previous section may be extended to include two-leg-diagrams of graphs with arbitrary even valences $`v=4,6,8\mathrm{}`$ We repeat the exact same cutting procedure and end up with some generalized rooted blossom-trees, such that each inner vertex say of valence $`v=2k`$ has exactly $`k1`$ black leaves attached to it. The corresponding generating function $`RR(g_4,g_6,g_8,\mathrm{})`$ with say weights $`g_{2k}`$ per $`2k`$-valent vertex, $`k=2,3,4\mathrm{}`$ obeys the following relation
$$R=1+\underset{k2}{}g_{2k}\left(\genfrac{}{}{0pt}{}{2k1}{k}\right)R^k$$
obtained again by inspecting all possible environments of the vertex attached to the root. The combinatorial factor $`\left(\genfrac{}{}{0pt}{}{2k1}{k}\right)`$ accounts for the number of choices for the positions of the $`k1`$ black leaves around the vertex, while the remaining $`k`$ descendents are themselves rooted blossom-trees.
Again, $`R`$ is the unique solution to (2.1) that admits a power series expansion in the $`g_{2i}`$โs, with $`R=1+O(g_{2i})`$ for all $`i2`$.
3. Planar graphs and trees II: arbitrary valences
This section extends the results of the previous one to graphs with both even and odd valences. The first consequence of allowing for odd valences is the existence of one-leg diagrams with only one (outcoming) external leg. These will be represented in the plane like two-leg diagrams, but the leg need not be adjacent to the external face.
3.1. Trivalent graphs
Let $`SS(g)`$ and $`RR(g)`$ denote the generating functions for respectively one- and two-leg diagrams of trivalent planar graphs, with a weight $`g`$ per trivalent vertex. Applying the cutting procedure of previous sections to one and two-leg diagrams, we end up with two types of rooted blossom trees which we call S-trees and R-trees respectively. Note that the unique (outcoming) leg of the one-leg diagrams is replaced by a root in the corresponding blossom-tree. To characterize S- and R-trees, let us introduce the charge $`q`$ of a tree as its number of white leaves minus that of black ones. For instance, the blossom-trees of Sects. 2.1 and 2.2 above have all charge $`q=1`$, the same holds for the present R-trees, while the S-trees are neutral, with charge $`q=0`$. Now S-trees and R-trees are characterized among rooted binary blossom-trees as having only descendent subtrees (not reduced to black leaves) of charge 0 or 1, while their total charge is $`0`$ and $`1`$ respectively (this is easily done by following the effect of the cutting procedure on the original graph, see for details). The cutting procedure establishes a bijection between one- and two-leg diagrams and S- and R-trees respectively.
The latter are easily counted, again by inspection of the local environment of the vertex attached to the root. We get the coupled relations:
$$\begin{array}{cc}& \text{}\hfill \\ & S=2gR+gS^2\hfill \\ & \text{}\hfill \\ & R=1+2gRS\hfill \end{array}$$
respectively displaying contributions from a vertex with total charge 0 (with one black leaf with $`q=1`$ and one descendent R-tree with $`q=1`$, and two possible positions for the black leaf, or with two descendent S-trees), and from a vertex with total charge $`1`$ (with one descendent S-tree with $`q=0`$ and one descendent R-tree with $`q=1`$, and two possible relative positions for these).
The generating functions $`R,S`$ are uniquely determined by the relations (3.1) and the fact that they admit power series expansions $`R=1+O(g)`$, $`S=O(g)`$.
3.2. General case
The trivalent case is easily extended to the case of arbitrary (even or odd) valences weighted by $`g_3,g_4,g_5,\mathrm{}`$ per tri-, tetra-, penta-,โฆ valent vertex, in which the very same cutting procedure now leads to generalized rooted blossom S- and R-trees, now blossom trees of arbitrary valences $`v=3,4,5,\mathrm{}`$ again further characterized by the fact that all their descendent subtrees not reduced to a black leaf have charge $`0`$ or $`1`$, and by their total charge 0 and 1 respectively. This allows to count them straightforwardly, with the coupled relations:
$$\begin{array}{cc}\hfill S& =\underset{k3}{}g_k\underset{j=0}{\overset{[\frac{k1}{2}]}{}}\left(\genfrac{}{}{0pt}{}{k1}{j}\right)\left(\genfrac{}{}{0pt}{}{k1j}{j}\right)R^jS^{k12j}\hfill \\ \hfill R& =1+\underset{k3}{}g_k\underset{j=0}{\overset{[\frac{k2}{2}]}{}}\left(\genfrac{}{}{0pt}{}{k1}{j}\right)\left(\genfrac{}{}{0pt}{}{k1j}{j+1}\right)R^{j+1}S^{k22j}\hfill \end{array}$$
where the combinatorial factors account for the possible ways of positioning $`j`$ black leaves, $`j`$ or $`j+1`$ descendent R-subtrees and the remaining S-subtrees on the vertex attached to the root. Again, eqs.(3.1) determine completely $`R`$ and $`S`$ with $`R=1+O(g_i)`$ and $`S=O(g_i)`$ for all $`i3`$.
For illustration, in the case of tri/tetravalent graphs, where only $`g_3,g_4`$ are non-zero, we have the equations
$$\begin{array}{cc}\hfill S& =g_3(2R+S^2)+g_4(6RS+S^3)\hfill \\ \hfill R& =1+2g_3RS+3g_4R(R+S^2)\hfill \end{array}$$
4. Planar graphs and trees III: bipartite graphs
We now turn to the slightly more involved case of bipartite (i.e. vertex-bicolored, say black and white) graphs. In the language of matrix models, these correspond to the case of two coupled matrices.
4.1. $`p`$-valent case
Let us consider two-leg diagrams of vertex-bicolored $`p`$-valent planar graphs, and their generating functions with a weight $`g`$ (resp. $`\stackrel{~}{g}`$) per $`p`$-valent black (resp. white) vertex. We must also indicate the color of the vertices to which the in and out-coming legs are attached, and this leads to a priori distinct generating functions. For simplicity, we restrict ourselves to only diagrams with incoming (resp. outcoming) leg attached to a white (resp. black) vertex with generating function $`RR^{}(g,\stackrel{~}{g})`$, and also include the single graph whose incoming and outcoming legs are directly attached to one-another, without vertex, contributing $`1`$ to $`R`$.
Fig.2: A sample bipartite $`p`$-valent two-leg diagram with $`p=4`$ (a) is iteratively cut into a rooted bipartite blossom-tree (c) by the usual procedure, with the restriction that only edges originating from a black vertex may be cut. The edges to be cut are indicated in (b) with their order of visit, counterclockwise around the graph.
The above cutting procedure is now slightly modified, with the additional constraint that an edge may be cut only if it moreover originates from a black vertex (see Fig.2 for an illustration in the case $`p=4`$). This leads to a new kind of blossom-trees (Fig.2 (c)), with bicolored vertices, and such that black or white leaves may only be connected to vertices of the same color, while each black vertex has exactly $`p2`$ black leaves attached to it. The tree still has the property of having one more white leaf than black ones (i.e. a total charge of $`q=1`$), as the incoming leg is replaced by a white leaf (which is compatible with the above rule, as the incoming leg is attached to a $`p`$-valent white vertex). Moreover, the root (former outcoming vertex) is attached to a black vertex. The generating function $`R`$ obeys the relations:
$$\begin{array}{cc}\hfill R& =1+(p1)gX\hfill \\ \hfill X& =\stackrel{~}{g}R^{p1}\hfill \end{array}$$
The first line is obtained by inspection of all possible environments of the root: (i) it may simply have one white leaf attached to it or (ii) it may have a black vertex attached to it, itself with $`p2`$ black leaves and a descendent rooted blossom-tree of total charge $`+(p1)`$ with a black root, and with generating function $`X`$. The second line expresses these latter trees according to the environment of the white vertex attached to the root, having $`p1`$ descendent rooted blossom-trees of charge $`+1`$, all generated by $`R`$. We finally get
$$R=1+(p1)g\stackrel{~}{g}R^{p1}$$
Note that the generating function $`C_p(x)`$ for rooted $`p`$-valent planar trees with a weight $`x`$ per vertex satisfies $`C_p(x)=1+xC_p(x)^{p1}`$ and $`C_p(x)=1+O(x)`$. The corresponding number of trees with $`N`$ vertices reads $`C_N^{(p)}=\frac{1}{1+(p1)N}\left(\genfrac{}{}{0pt}{}{(p1)N}{N}\right)`$. These numbers are also known as the Fuss-Catalan numbers, and reduce to the ordinary Catalan numbers for $`p=3`$. Finally the number of rooted blossom-trees with $`N`$ vertices is simply $`R|_{g^N\stackrel{~}{g}^N}=(p1)^NC_N^{(p)}`$. For large $`N`$, it behaves as $`g_c^N/N^{3/2}`$, where $`g_c=(p2)^{p2}/(p1)^p`$.
4.2. $`p`$-constellations
Fig.3: The cutting procedure is applied to a two-leg diagram of a 3-constellation (a), with incoming (resp. outcoming) leg attached to a white (resp. black) vertex. The edges are visited in counterclockwise order around the graph (b), and cut iff (i) this leaves the resulting graph connected (ii) the cut edge originates from a black vertex. We have indicated the chronological order of the cut edges, from 1 to 6. We finally replace each cut edge by a pair of black/white leaves, and the incoming (resp. outcoming) leg by a white leaf (resp. root).
The $`p`$-constellations, introduced in , are vertex-bicolored (black and white) graphs such that say black vertices have all valence $`p`$, while white ones may have valences arbitrary multiples of $`p`$. We again consider two-leg diagrams of such planar graphs with say incoming leg connected to a white vertex and outcoming leg connected to a black vertex (see Fig.3 (a) for an illustration with $`p=3`$), or both legs being directly connected without vertex, and count them with a weight $`g`$ per ($`p`$-valent) black vertex and weights $`\stackrel{~}{g}_m`$ for $`mp`$-valent white vertices, $`m=1,2,3\mathrm{}`$ Let $`R=R^{}(g;\{\stackrel{~}{g}_i\})`$ denote the corresponding generating function. The cutting procedure, illustrated in Fig.3, remains the same as in the previous section, and leaves us with rooted blossom-trees with bicolored vertices of total charge $`+1`$, such that leaves may only be connected to vertices of their own color, and that descendent subtrees may be of two types: if their first vertex is black, they have total charge $`1`$; if its is white, they have total charge $`p1`$, the descendents of the root vertex of the latter being themselves only blossom trees of charge $`1`$ or bunches of $`p1`$ black leaves attached to a black vertex.
By inspection of all possible local environments of the vertex attached to the root, we may derive the following relations for $`R`$:
$$\begin{array}{cc}\hfill R& =1+(p1)gX\hfill \\ \hfill X& =\underset{m1}{}\stackrel{~}{g}_m\left(\genfrac{}{}{0pt}{}{mp1}{m1}\right)Y^{m1}R^{(p1)m}\hfill \\ \hfill Y& =g\hfill \end{array}$$
where $`R`$, $`X`$, $`Y`$ generate rooted blossom-trees of respective charges $`1`$, $`p1`$, $`1p`$, starting respectively with a black, white, black vertex. $`Y=g`$ is due to the fact that the only tree contributing to $`Y`$ has a black vertex and $`p1`$ black leaves attached to it. In the case $`p=3`$, this reads
$$\begin{array}{cc}& \text{}\hfill \\ & R=1+2gX\hfill \\ & \text{}\hfill \\ & X=\stackrel{~}{g}_1R^2+5\stackrel{~}{g}_2YR^4+\mathrm{}\hfill \\ & \text{}\hfill \\ & Y=g\hfill \end{array}$$
Eliminating $`Y`$ and $`X`$ from (4.1), we arrive at the single algebraic equation
$$R=1+(p1)\underset{m1}{}\stackrel{~}{g}_m\left(\genfrac{}{}{0pt}{}{mp1}{m1}\right)g^{m1}R^{(p1)m}$$
$`R`$ is the unique solution to this equation such that $`R=1+O(g,\stackrel{~}{g}_i)`$.
Note that in the case $`p=2`$ of 2-constellations, we recover the even-valent graph result (2.1), upon taking $`g_{2k}=\stackrel{~}{g}_k`$ for $`k2`$, while $`\stackrel{~}{g}_1=0`$ and $`g=1`$. Indeed, in 2-constellations, the (2-valent) black vertices may be viewed as decorations of the edges of an arbitrary graph with only white even-valent vertices. Setting $`g=1`$ precisely allows to forget about these decorations, while $`\stackrel{~}{g}_1=0`$ simply eliminates 2-valent white vertices.
4.3. Planar bipartite graphs and the Ising model
Constellations are easily tractable objects, essentially due to the triviality of one type of vertex (black here), whose valence remains fixed. More generally, we would like to consider in all generality vertex-bicolored graphs of arbitrary even valences for vertices of both colors. For the sake of simplicity, we will restrict ourselves to two-leg diagrams in which both legs are attached to white vertices. Let us denote by $`RR^{}(\{g_{2i}\};\{\stackrel{~}{g}_{2i}\})`$ the generating function for two-leg diagrams of planar bipartite graphs with weights $`g_{2i}`$ (resp. $`\stackrel{~}{g}_{2i}`$) per $`2i`$-valent black (resp. white) vertex, $`i=1,2,\mathrm{}`$, and such that both legs are attached to white vertices.
Applying the now usual cutting procedure to two-leg diagrams, we end up with blossom-trees of total charge $`+1`$ with bicolored vertices of arbitrary even valences. Their characterization however is quite delicate, as it involves describing all their possible descendent subtrees. These come in two forms according to the color of their vertex attached to the root: if the latter is white, the possible descendent subtrees are either reduced to a black leaf (of charge $`1`$) or rooted vertex-bicolored blossom trees of total charges $`1`$, $`3`$, $`5`$, โฆ ; if it is black, the possible descendent subtrees are rooted vertex-bicolored blossom trees of charges $`1`$, $`1`$, $`3`$, $`5`$, โฆ Let us introduce the generating functions $`R_i`$ for vertex-bicolored blossom-trees of total charge $`2i1`$, $`i=1,2,3,\mathrm{}`$ and whose root is attached to a white vertex, together with the generating functions $`X_i`$ for vertex-bicolored blossom-trees of total charge $`12i`$, $`i=1,2,3,\mathrm{}`$ and whose root is attached to a black vertex, while $`VX_0`$ generates vertex-bicolored blossom-trees of total charge $`1`$ and whose root is attached to a black vertex, or the tree made of a single leaf attached to the root, without any vertex (contributing $`1`$ to $`V`$). We also introduce the generating function $`R_0=1`$ for the tree made of a single black leaf attached to the root, without any vertex. We now simply have to enumerate all possible environments of the vertex attached to the root of each of these trees, according to the type of its attached descendent subtrees. This gives the following system
$$\begin{array}{cc}\hfill V& =1+\underset{k1}{}g_{2k}\underset{\genfrac{}{}{0pt}{}{j_1,j_2,\mathrm{},j_{2k1}0}{\mathrm{\Sigma }j_l=k}}{}R_{j_1}R_{j_2}\mathrm{}R_{j_{2k1}}\hfill \\ \hfill X_m& =\underset{k1}{}g_{2k}\underset{\genfrac{}{}{0pt}{}{j_1,j_2,\mathrm{},j_{2k1}0}{\mathrm{\Sigma }j_l=km}}{}R_{j_1}R_{j_2}\mathrm{}R_{j_{2k1}},m=1,2,3,\mathrm{}\hfill \\ \hfill R_m& =\underset{k1}{}\stackrel{~}{g}_{2k}\underset{\genfrac{}{}{0pt}{}{j_1,j_2,\mathrm{},j_{2k1}0}{\mathrm{\Sigma }j_l=km}}{}X_{j_1}X_{j_2}\mathrm{}X_{j_{2k1}},m=1,2,3,\mathrm{}\hfill \end{array}$$
The desired generating function $`R=R_1`$ is the unique solution to this system where all $`R`$โs, $`X`$โs and $`V`$ admit power series expansions of the $`g`$โs and $`\stackrel{~}{g}`$โs.
The case of the Ising model on planar tetravalent graphs may be viewed as a particular case of the above, in which only $`g_2,g_4,\stackrel{~}{g}_2,\stackrel{~}{g}_4`$ are non-zero. To see this, recall that the Ising model on tetravalent planar graphs is defined by say coloring the vertices of an arbitrary tetravalent planar graph in black or white (colors stand here for the spin up or down), and counting the configurations with different โnearest neighbor interactionโ weights for edges connected to vertices of the same color (weight $`e^K`$) or of different colors (weight $`1`$), while black (resp. white) vertices are counted with a weight $`ge^H`$ (resp. $`ge^H`$). Here $`K`$ and $`H`$ are respectively the spin coupling and the external magnetic field of the Ising model. To make the contact with our model, we just have to resum all possible configurations obtained by adding arbitrary numbers of 2-valent black and white vertices on the edges of Ising configurations, in such a way that bicoloration is restored. This entails adding any chain of black, white, black, โฆ, white 2-valent vertices between any white and black tetravalent vertices connected by an edge, or any chain of alternating white, black, โฆ, white 2-valent vertices between any two black tetravalent vertices connected by an edge or else any chain of black, white, โฆ, black 2-valent vertices between any two white tetravalent vertices connected by an edge. Doing the resummations within the configurations of our bicolored graphs produces an effective edge interaction weight $`w_{ab}`$ according to the colors $`a,b`$ of the adjacent vertices:
$$\begin{array}{cc}\hfill w_{}& =\frac{\stackrel{~}{g}_2}{1g_2\stackrel{~}{g}_2}=\text{}\hfill \\ \hfill w_{}& =\frac{g_2}{1g_2\stackrel{~}{g}_2}=\text{}\hfill \\ \hfill w_{}& =w_{}=\frac{1}{1g_2\stackrel{~}{g}_2}=\text{}\hfill \end{array}$$
hence to identify our model with the Ising one, we must take $`g_2=\stackrel{~}{g}_2=e^K`$, while $`g_4=(1e^{2K})^2ge^H`$ and $`\stackrel{~}{g}_4=(1e^{2K})^2ge^H`$, and the external legs must receive the extra weights $`1/\sqrt{1e^{2K}}`$ each.
Restricting to the symmetric case $`g_2=\stackrel{~}{g}_2c`$ and $`g_4=\stackrel{~}{g}_4g`$, the above equations simply read
$$\begin{array}{cc}& \text{}\hfill \\ & V=1+cR_1+3gR_1^2+3gR_2\hfill \\ & \text{}\hfill \\ & X_1=c+3gR_1\hfill \\ & \text{}\hfill \\ & X_2=g\hfill \\ & \text{}\hfill \\ & R_1=cV+3gV^2X_1\hfill \\ & \text{}\hfill \\ & R_2=gV^3\hfill \end{array}$$
Eliminating $`R_2`$ and $`X_1`$, we get
$$V=\frac{R}{c+3gR}\mathrm{with}R(c+3gR)^2(1(c+3gR)^2)=(c+3gR)^3+3g^2R^3$$
Here, $`RR_1`$ is the generating function for two-leg diagrams of planar tetravalent graphs with Ising (black or white) spins decorating their vertices, with interaction weights $`w_{}=w_{}=c`$, $`w_{}=w_{}=1`$ and a weight $`g/(1c^2)^2`$ per tetravalent vertex, and such that the two legs are attached to univalent black vertices, themselves weighted by $`1/\sqrt{1c^2}`$.
5. Geodesic distances
In the previous sections, we have enumerated various one- and two-leg diagrams by establishing bijections with suitable classes of blossom-trees. Note that in the plane representation we have chosen, the face $`F_1`$ adjacent to the unique leg for one-leg diagrams is not necessarily the external face $`F_0`$. Accordingly in the case of two-leg diagrams, the face $`F_1`$ adjacent to the out-coming leg need not be the external face $`F_0`$, itself adjacent to the in-coming leg. In this section, we show how to keep track of the geodesic distance between $`F_0`$ and $`F_1`$, namely the smallest number of edges to be crossed in a path from $`F_0`$ to $`F_1`$. By a slight abuse of language, this distance will also be referred to as the distance between the legs.
5.1. Keeping track of the geodesic distances
The main feature of the previous sections is a sort of unified formulation of planar graphs in the language of blossom-trees. Note that going back from blossom-trees to graphs is a straightforward step, as there is a unique way of reconnecting the black and white leaves into edges: this is done by simply connecting each black leaf to the first available white leaf in counterclockwise direction. This process leaves us in the case of trees of charge $`1`$ with exactly one unmatched white leaf, which is taken as outcoming leg, while the root is the incoming one. For trees of charge 0, all pairs are exhausted and only the root remains as uniqe leg. In both cases, we note that the geodesic distance between the faces $`F_0`$ (external) and $`F_1`$ (adjacent to the former root) is simply given by the number of black-white edge pairs that separate the root from the external face after recombination. Keeping track of this geodesic distance simply amounts to keeping track of the black leaves โin excessโ that require encompassing the root to be connected to their white alter ego.
Fig.4: The contour walk of a rooted blossom-tree. Visiting the tree (a) in clockwise direction starting from the root, one keeps a record of the type of leaves encountered in the form of a walk on the integer line (b), starting at the origin, and with steps up (for a black leaf) and down (for a white one). The maximum reached by the walk is nothing but the geodesic distance separating the root from the external face in the recombined graph. This distance is 3 in the present case, as one readily checks by closing the tree back into a planar graph (with tetra/hexavalent vertices here). Concentrating on the environment of the vertex attached to the root, we see that each descendent subtree corresponds to a portion of the walk (c), with a certain relative maximum. Expressing the global maximum of the contour walk in terms of the relative maxima of its portions allows for writing a recursion relation for the generating function for rooted blossom-trees whose root is at a maximum distance $`n`$ from the external face of the recombined graph.
This is done in all generality by attaching to each blossom-tree a โcontour walkโ, namely a walk on the relative integer line with steps $`\pm 1`$, obtained as follows (see Fig.4 for an example). One starts from the root of the blossom-tree and visits in clockwise direction all leaves around the tree. Starting from the coordinate $`0`$, we make a step $`+1`$ (resp. $`1`$) for each encountered black (resp. white) leaf. In a blossom-tree of charge $`k`$, such a walk will end up at coordinate $`k`$. Now the number of excess black leaves responsible for the geodesic distance between $`F_1`$ and $`F_0`$ is simply the maximum coordinate reached by the contour walk.
In view of this result, and of the form of all relations determining the blossom-trees of the previous sections, it is natural to introduce by analogy with the generating functions say $`X`$ for some particular type of blossom-trees the generating function $`X_n`$ for the same blossom-trees with a geodesic distance of at most $`n`$ between $`F_1`$ and $`F_0`$. To obtain from $`X_n`$ the generating function for blossom-trees with geodesic distance equal to $`n`$ betwen $`F_0`$ and $`F_1`$, we simply have to take the difference $`X_nX_{n1}`$. Keeping track of $`n`$ then boils down to expressing the maximum of the contour walk of a tree in terms of those of the individual contour walks of the blossom-trees descending from the vertex attached to the root, following the same inspection procedure as before. This is done case by case in the following sections.
5.2. Even valences
Let us start with the tetravalent case. We find that eq. (2.1) must be transformed into
$$R_n=1+gR_n(R_{n1}+R_n+R_{n+1})$$
where $`R_n`$ denotes the generating function for tetravalent planar graphs with two legs at distance at most $`n`$, and weight $`g`$ per vertex. The three terms on the r.h.s. correspond to respectively the black leaf on the right, in the middle or on the left of the two other descendents of the vertex attached to the root (see the picture of eq.(2.1)). It is clear that the presence of the black leaf acts as a shift by $`1`$ on the local distance $`n`$ while $`R_n`$ accompanies a shift by $`+1`$, when going clockwise around the vertex.
To have a compact notation for the general result, let us introduce a formal orthonormal basis $`|n`$, $`n\text{ZZ}`$, with $`m|n=\delta _{m,n}`$, and an operator $`\sigma `$ acting as a shift $`\sigma |n=|n+1`$, and its formal inverse $`\sigma ^1`$ such that $`\sigma ^1|n=|n1`$. We introduce the operator
$$Q=\sigma +\sigma ^1\widehat{r}$$
where $`\widehat{r}`$ simply acts diagonally as $`\widehat{r}|n=R_n|n`$. Note that the shift $`\sigma `$ may be represented by a black leaf, while its inverse always accompanies an $`R`$. Then eq. (5.1) takes the form
$$1=n1|(QgQ^3)|n$$
More generally, in the case of arbitrary even valences, we have to write
$$1=n1|(Q\underset{k2}{}g_{2k}Q^{2k1})|n$$
For instance in the case of tetra- and hexa-valent graphs, eq. (5.1) reads explicitly
$$\begin{array}{cc}\hfill R_n& =1+g_4R_n(R_{n+1}+R_n+R_{n1})+g_6R_n(R_{n+1}R_{n+2}+R_{n+1}R_{n1}\hfill \\ & +R_{n1}R_{n2}+R_{n+1}^2+R_{n1}^2+R_n(2R_{n+1}+R_n+2R_{n1}))\hfill \end{array}$$
A remark is in order. The equations (5.1)-(5.1) are valid only for $`n0`$, provided we use $`R_k=0`$, $`k=1,2,\mathrm{}`$ wherever they occur in the r.h.s. With these boundary conditions, and the general fact that $`R_n`$ possess a power series expansion in $`g`$, with $`R_n=1+O(g)`$ for $`n0`$, all $`R_n`$โs are then uniquely determined by (5.1) order by order in $`g`$.
5.3. Arbitrary valences
Let $`S_n`$ (resp. $`R_n`$) denote the generating function for one- (resp. two-) leg diagrams of planar graphs with arbitrary valences with weights $`g_i`$ per $`i`$-valent vertex, and such that $`F_0`$ (the external face) and $`F_1`$ (the face adjacent to the unique (resp. outcoming) leg) are distant by at most $`n`$. For trivalent graphs, we find that eqs. (3.1) must be transformed into
$$\begin{array}{cc}\hfill S_n& =g(R_n+R_{n1})+gS_n^2\hfill \\ \hfill R_n& =1+gR_n(S_{n+1}+S_n)\hfill \end{array}$$
where in addition to the situation of previous section we simply note that $`S`$โs donโt affect the distance counting (no shift). This suggests to introduce in the general case of arbitrary valences the operator
$$Q=\sigma +\sigma ^1\widehat{s}\sigma +\sigma ^1\widehat{r}$$
where $`\widehat{s}`$ acts diagonally on the basis $`|n`$ as $`\widehat{s}|n=S_n|n`$, and in terms of which we simply have to write
$$\begin{array}{cc}\hfill 0& =n|(Q\underset{i3}{}g_iQ^{i1})|n\hfill \\ \hfill 1& =n1|(Q\underset{i3}{}g_iQ^{i1})|n\hfill \end{array}$$
For tri- and tetra-valent graphs, this reads
$$\begin{array}{cc}\hfill S_n& =g_3(R_n+R_{n1}+S_n^2)+g_4(R_n(S_{n+1}+2S_n)+R_{n1}(S_{n1}+2S_n)+S_n^3)\hfill \\ \hfill R_n& =1+g_3R_n(S_n+S_{n+1})+g_4R_n(S_n^2+S_nS_{n+1}+S_{n+1}^2+R_{n+1}+R_n+R_{n1})\hfill \end{array}$$
Note that when $`g_3=0`$, we find the solution $`S_n=0`$ identically, and eq. (5.1) reduces to the tetravalent case (5.1). More generally, imposing that all odd $`g`$โs vanish leads to the solution $`S_n=0`$ (as there are no one-leg diagrams with only even valences), and we recover the even-valent case of previous section.
5.4. Constellations
Let $`R_n`$ denote the generating function for two-leg diagrams of $`p`$-constellations with a weight $`g`$ per white ($`p`$-valent) vertex, and weights $`\stackrel{~}{g}_i`$ per black $`pi`$-valent vertex, $`i=1,2,\mathrm{}`$, whose incoming (resp. outcoming) leg is attached to a black (resp. white) vertex (or both are connected without vertex and the corresponding unique graph contributes $`1`$ to $`R_n`$ for all $`n`$) and such that the geodesic distance between the two legs is at most $`n`$. Here the notion of geodesic distance is defined according to the rules used in the cutting procedure of Sect.5.3, namely the geodesic distance between $`F_0`$ and $`F_1`$ is the minimal number of edges to be crossed in a path going from $`F_0`$ to $`F_1`$, and such that at each edge-crossing the white vertex is always on the right. Following the same reasoning as in the previous sections, we are now led to the introduction of two operators $`Q_1`$ and $`Q_2`$ which generate, upon taking powers, the successive decorations of the vertex attached to the root, respectively in the case of a black and white vertex. We have
$$\begin{array}{cc}\hfill Q_1& =\sigma +\sigma ^1\widehat{x}\sigma ^{2p}\hfill \\ \hfill Q_2& =\sigma ^1\widehat{r}+\sigma ^{p2}\widehat{y}\sigma \hfill \end{array}$$
where the shift $`\sigma `$ represents a single black leaf, while $`\widehat{x}`$ represents rooted blossom-trees of charge $`p1`$ whose first vertex is white. Again, $`\widehat{x},\widehat{y},\widehat{r}`$ act diagonally on the basis $`|n`$ with eigenvalues $`X_n,Y_n,R_n`$ respectively. The equations (4.1) now become
$$\begin{array}{cc}\hfill 1& =n1|(Q_2gQ_1^{p1})|n\hfill \\ \hfill 0& =n+p1|(Q_2gQ_1^{p1})|n\hfill \\ \hfill 0& =np+1|(Q_1\underset{i1}{}\stackrel{~}{g}_iQ_2^{pi1})|n\hfill \end{array}$$
In the particular case of 3-constellations, with say only $`g,\stackrel{~}{g}_1,\stackrel{~}{g}_2`$ non-zero, these read for instance
$$\begin{array}{cc}\hfill R_n& =1+g(X_n+X_{n1})\hfill \\ \hfill Y_n& =g\hfill \\ \hfill X_n& =\stackrel{~}{g}_1R_nR_{n+1}+\stackrel{~}{g}_2gR_nR_{n+1}(R_{n+3}R_{n+2}+R_{n+2}R_{n+1}+R_{n+1}R_n+R_nR_{n1})\hfill \end{array}$$
while in the case of only (black and white) $`p`$-valent vertices of Sect. 4.1, eqs.(5.1) read
$$\begin{array}{cc}\hfill R_n& =1+g(X_n+X_{n1}+\mathrm{}+X_{np+2})\hfill \\ \hfill X_n& =\stackrel{~}{g}_1R_nR_{n+1}\mathrm{}R_{n+p2}\hfill \end{array}$$
for the generating function $`R_n`$ for two-leg-diagrams of bipartite $`p`$-valent graphs with incoming leg attached to a white vertex and outcoming leg attached to a black one, and such that the geodesic distance from the in- to the out-coming leg is at most $`n`$.
5.5. Bipartite even-valent graphs and the Ising model
In the general case of bipartite even-valent graphs, we are led to the introduction of two operators $`Q_1,Q_2`$ with the following structure
$$\begin{array}{cc}\hfill Q_1& =\sigma +\underset{k1}{}\sigma ^{12k}\widehat{r}^{(k)}\hfill \\ \hfill Q_2& =\sigma ^1\widehat{v}+\underset{k1}{}\sigma ^1\widehat{x}^{(k)}\sigma ^{2k}\hfill \end{array}$$
where the $`\widehat{r}^{(k)},\widehat{x}^{(k)},\widehat{v}`$ all act diagonally with eigenvalues $`R_n^{(k)},X_n^{(k)},V_n`$. The latter are nothing but the generating functions for sets of rooted blossom-trees restricted by $`n`$, respectively starting with a white, black, black vertex, and with charges $`2k1,12k,1`$ respectively. We are actually interested in computing $`R_nR_n^{(1)}`$, the generating function for two-leg diagrams of bipartite planar graphs with weights $`g_{2i},\stackrel{~}{g}_{2i}`$ per $`2i`$-valent black, white vertex, such that moreover the two legs are distant by at most $`n`$. Again, the distance from $`F_0`$ to $`F_1`$ is defined as the minimal number of edges to be crossed in a path from $`F_0`$ to $`F_1`$, such that at each edge crossing the white vertex is always on the right. This definition allows to keep track of this distance on the trees themselves, as the number of โexcessโ black leaves which upon recombination with white ones encompass the root of the tree. The equations determining $`R_n`$ are simply
$$\begin{array}{cc}\hfill 1& =n1|(Q_2\underset{i1}{}g_{2i}Q_1^{2i1})|n\hfill \\ \hfill 0& =n+2m1|(Q_2\underset{i1}{}g_{2i}Q_1^{2i1})|n,m=1,2,\mathrm{}\hfill \\ \hfill 0& =n2m+1|(Q_1\underset{i1}{}\stackrel{~}{g}_{2i}Q_2^{2i1})|n,m=1,2,\mathrm{}\hfill \end{array}$$
In the abovementioned case of the Ising model with only $`g_2=\stackrel{~}{g}_2=c`$ and $`g_4=\stackrel{~}{g}_4=g`$ non-zero, we must take $`Q_1=\sigma +\sigma ^1\widehat{r}^{(1)}+\sigma ^3\widehat{r}^{(2)}`$ and $`Q_2=\sigma ^1\widehat{v}+\sigma ^1\widehat{x}^{(1)}\sigma ^2+\sigma ^1\widehat{x}^{(2)}\sigma ^4`$, and eq.(5.1) reduces to
$$\begin{array}{cc}\hfill V_n& =1+cR_n+gR_n(R_{n+1}+R_n+R_{n1})+g(R_n^{(2)}+R_{n+1}^{(2)}+R_{n+2}^{(2)})\hfill \\ \hfill X_n^{(1)}& =c+g(R_n+R_{n1}+R_{n2})\hfill \\ \hfill X_n^{(2)}& =g\hfill \\ \hfill R_n& =cV_n+gV_n(V_{n+1}X_{n+2}^{(1)}+V_nX_{n+1}^{(1)}+V_{n1}X_n^{(1)})\hfill \\ \hfill R_n^{(2)}& =gV_nV_{n1}V_{n2}\hfill \end{array}$$
We first remark that $`R_n=V_nX_{n+1}^{(1)}`$ by comparing the second and fourth lines of eq.(5.1). This is a particular case of a general duality between $`Q_1`$ and $`Q_2`$ in the symmetric case when $`g_i=\stackrel{~}{g}_i`$ for all $`i`$, where we may write $`Q_1=Q_2^{}`$, where $`\sigma ^{}=\sigma ^1v`$, $`(AB)^{}=B^{}A^{}`$ for all operators $`A`$, $`B`$, and $`f^{}=f`$ for all diagonal operators. Here this implies $`\sigma ^1x^{(1)}\sigma ^2=r^{(1)}v^1\sigma `$ and $`\sigma ^1x^{(2)}\sigma ^4=r^{(2)}(v^1\sigma )^3`$, i.e. $`R_n=V_nX_{n+1}^{(1)}`$ and $`R_n^{(2)}=X_{n+1}^{(2)}V_nV_{n1}V_{n2}`$. Finally eliminating $`X_n^{(1)}`$ and $`R_n^{(2)}`$ from eq.(5.1), we are left with
$$\begin{array}{cc}\hfill V_n(1g^2(V_{n+1}V_{n+2}& +V_{n+1}V_{n1}+V_{n1}V_{n2}))=1+R_n(c+g(R_{n+1}+R_n+R_{n1})\hfill \\ \hfill R_n& =V_n(c+g(R_{n+1}+R_n+R_{n1})\hfill \end{array}$$
An important remark is in order about the generating function $`R_n`$. Although $`n`$ has the meaning of a maximal geodesic distance between the two legs in the bipartite graph picture, it loses somewhat of its meaning in the correspondence with Ising model configurations. Indeed, within configurations of the Ising model on tetravalent planar graphs, $`n`$ is not the obvious geodesic distance between the faces adjacent to the in- and out-coming legs, as its definition involves first transforming the graph into a bipartite one, and it then corresponds to a distance where edge-crossing is permitted only if the white vertex is on the right. This restriction is almost irrelevant, as there are in general sufficiently many successions of black, white, blackโฆ bivalent vertices to allow for crossing edges in both directions. One case however is troublesome: when an Ising edge connects a black vertex to a white one, in the absence of intermediate bivalent vertices (not necessary here as the bicoloration is already ensured), the edge may only be crossed in one direction, leaving the white vertex on the right. This introduces a bias in the notion of distance, having to do with the matter configurations on the graph. An analogous situation was encountered in in the case of hard dimers on tetravalent planar graphs, namely on configurations of tetravalent planar graphs where edges may (or may not) be occupied by dimers which repel one-another in such a way that no two adjacent edges can be simultaneously occupied. In this case indeed, the notion of geodesic distance is biased by the dimers, in that the occupied edges cannot be crossed in paths from $`F_0`$ to $`F_1`$. In both Ising and hard-dimer cases, the matter interfers with the space, by modifying the rules governing distances. Consequently, we think it is interesting to investigate the dependence of the Ising two-leg diagrams on this special distance, and it may eventually be that its difference with the true geodesic distance becomes irrelevant in large graphs.
6. Exact solutions
6.1. Finding exact solutions: a general scheme
All the equations listed in Sect. 5, despite their diversity, are all basically of the same form: (possibly coupled) algebraic recursion relations expressing $`R_n`$ (and the other generating functions involved) in terms of a finite number of previous terms $`R_{n1},R_{n2},\mathrm{},R_{nk}`$. In principle the boundary data needed to entirely determine $`R_n`$ should consist of $`k`$ consecutive initial values of $`R_j`$. It turns out however that we may drastically simplify, namely divide by $`2`$ this required number of initial data by simply requiring that $`lim_n\mathrm{}R_n`$ exists, and that it moreover coincides with the generating function $`R`$. Indeed, this is nothing but restating the definitions of $`R_n`$ and $`R`$, as the latter was first obtained regardless of the geodesic distance between legs, while the limit $`n\mathrm{}`$ of the former amounts to removing the geodesic distance constraint in the counting of graphs. In the same fashion, all other generating functions involved tend to their obvious limiting values when $`n\mathrm{}`$. This allows to linearize the various recursion relations at large $`n`$, by setting say $`R_n=R\rho _n`$, and similarly for the other generating functions involved. At first order in $`\rho _n`$ and its other counterparts, we obtain (possibly coupled) linear recursion relations. We immediately deduce that $`\rho _nx^n`$ (or a linear combination involving $`x`$โs of the same modulus) for some solution $`x`$ (with modulus less than 1) to the characteristic equation of the linearized recursion relations.
To completely solve our equations, we start by determining the exact form of the linearized solution, in general a linear combination $`\rho _n_{j=1}^ka_j(x_j)^n`$ where $`x_j`$, $`j=1,2,\mathrm{},k`$ denote all the (generically distinct) solutions of the linearized characteristic equation with modulus less than 1. In a second step, we obtain order by order in the $`(x_j)^n`$ the higher order contributions to the true solution, expanded at large $`n`$. These take in general the form of recursion relations for the coefficients of the multiple expansion in powers of the $`(x_j)^n`$. Solving these recursion relations, and resumming the resulting series allows us to finally obtain compact expressions for the exact solutions to the non-linear recursion relations at hand. The result still depends on the initial parameters $`a_j`$, $`j=1,2,\mathrm{},k`$, which then are fixed by requiring that the terms involving the $`k`$ first $`R_k,R_{k+1},\mathrm{},R_1`$ drop out of the recursion relations.
In the following sections, we simply present the solutions, as we have found them. A case by case proof by substitution is left as an exercise to the reader. In many situations, the proof boils down to a certain identity between Chebyshev polynomials of the first kind, as will be apparent soon.
6.2. Tetravalent case
For pedagogical purposes, we detail in this simple case the general scheme presented in the previous section. Substituting $`R_n=R\rho _n`$ into eq.(5.1) we get at first order in $`\rho _n`$:
$$\rho _n(13gR)=gR(\rho _{n1}+\rho _n+\rho _{n+1})+O(\rho _n^2)$$
The linearized characteristic equation therefore reads
$$1gR\left(x+\frac{1}{x}+4\right)=0$$
For $`g<g_c=1/12`$, there is generically a unique solution $`xx(g)`$ with modulus less than 1 to this equation, and we find that at first order $`\rho _n=ax^n+O(x^{2n})`$. We may now infer the general form $`\rho _n=_{j1}a_jx^{nj}`$, $`a_1=a`$ for the complete solution, where the coefficients $`a_j`$ are to be determined order by order in $`x^n`$. We find explicitly
$$a_{k+1}=\underset{j=1}{\overset{k}{}}\left(\frac{x^j+\frac{1}{x^j}+1}{x^{k+1}+\frac{1}{x^{k+1}}x\frac{1}{x}}\right)a_ja_{k+1j}$$
solved recursively as
$$a_k=a\frac{1x^k}{1x}\left(\frac{ax}{(1x)(1x^2)}\right)^{k1}$$
Picking $`a=x(1x)(1x^2)\lambda `$, $`R_n`$ is easily resummed into
$$R_n=R\frac{u_nu_{n+3}}{u_{n+1}u_{n+2}},u_n=1\lambda x^{n+1}$$
This is the general solution to eq.(5.1), that converges for large $`n`$. To see why, it is simplest to substitute the form (6.1) into the initial equation (5.1) which then boils down to the following quartic relation
$$u_nu_{n+1}u_{n+2}u_{n+3}=\frac{1}{R}u_{n+1}^2u_{n+2}^2+gR(u_{n1}u_{n+2}^2u_{n+3}+u_n^2u_{n+3}^2+u_nu_{n+1}^2u_{n+4})$$
Substituting $`u_n=1\lambda x^{n+1}`$ into this, we just have to check that the zeros of the l.h.s. as a degree 4 polynomial in $`\mathrm{\Lambda }=\lambda x^n`$ match those of the r.h.s. as moreover the equation reduces to eq. (2.1) for $`\mathrm{\Lambda }=0`$.
Finally, the โintegrationโ constant $`\lambda `$ is now fixed by further requiring that eq.(5.1) makes sense at $`n=0`$, in which case the term $`R_1`$ must drop off the r.h.s. of the recursion relation, in other words we have to impose $`R_1=0`$. This simply gives $`\lambda =1`$, and finally the exact solution to our combinatorial problem reads
$$R_n=R\frac{(1x^{n+1})(1x^{n+4})}{(1x^{n+2})(1x^{n+3})}=R\frac{U_nU_{n+3}}{U_{n+1}U_{n+2}},U_nU_n\left(\sqrt{x}+\frac{1}{\sqrt{x}}\right)=\frac{x^{\frac{n+1}{2}}x^{\frac{n+1}{2}}}{x^{\frac{1}{2}}x^{\frac{1}{2}}}$$
where we have identified the Chebyshev polynomials $`U_n`$ of the first kind.
As a simple application, the formula (6.1) gives access to the generating function for two-leg diagrams whose legs lie in the same face, already identified as that of rooted tetravalent planar maps or quadrangulations. We find that
$$R_0=R\frac{x+\frac{1}{x}}{x+\frac{1}{x}+1}=R\frac{14gR}{13gR}=RgR^3$$
with $`R`$ as in (2.1). This result was first obtained by Tutte in a completely different, though combinatorial, manner.
6.3. Even valences
Let us for definiteness consider the equation (5.1) with only $`g_4,g_6,\mathrm{},g_{2m+2}`$ non-zero. Linearizing again the equation at large $`n`$ and solving for the leading $`\rho _nx^n`$, we find that $`x`$ must obey the following characteristic equation:
$$\begin{array}{cc}\hfill 0=\chi _m(x)& 1\underset{k=0}{\overset{m}{}}g_{2k+2}R^k\underset{l=0}{\overset{k}{}}\left(\genfrac{}{}{0pt}{}{2k+1}{l}\right)\frac{1}{x^{kl}}\frac{1x^{2k+12l}}{1x}\hfill \\ & =1\underset{k=1}{\overset{m}{}}g_{2k+2}R^k\underset{l=0}{\overset{k}{}}\left(\genfrac{}{}{0pt}{}{2k+1}{l}\right)U_{2k2l}(w)\hfill \end{array}$$
expressed in terms of Chebyshev polynomials of $`w=\sqrt{x}+1/\sqrt{x}`$. Note that $`\chi _m(x)`$ is actually a degree $`m`$ polynomial in $`x+1/x`$, with generically $`m`$ distinct roots with modulus less than 1, denoted by $`x_1,x_2,\mathrm{},x_m`$. Repeating the straightforward, though tedious, exercise of previous section, we end up with the following exact solution
$$\begin{array}{cc}& R_n=R\frac{u_n^{(m)}u_{n+3}^{(m)}}{u_{n+1}^{(m)}u_{n+2}^{(m)}}\hfill \\ & u_n^{(m)}=\underset{l=0}{\overset{m}{}}(1)^l\underset{1m_1<\mathrm{}<m_lm}{}\underset{i=1}{\overset{l}{}}\lambda _{m_i}x_{m_i}^{n+m}\underset{1i<jl}{}c_{m_i,m_j}\hfill \\ & c_{a,b}\frac{(x_ax_b)^2}{(1x_ax_b)^2}\hfill \end{array}$$
Remarkably, the structure of $`u_n^{(m)}`$ matches exactly that of an $`N`$-soliton tau-function of the KP hierarchy which reads
$$\begin{array}{cc}& \tau =\underset{r=0}{\overset{N}{}}\underset{i_1<\mathrm{}<i_r}{}\underset{\mu =1}{\overset{r}{}}e^{\eta _{i_\mu }}\left(\underset{\mu <\nu }{}c_{i_\mu ,i_\nu }\right)\hfill \\ & c_{i,j}=\frac{(p_ip_j)(q_iq_j)}{(p_iq_j)(q_ip_j)}\hfill \end{array}$$
where the $`\eta `$โs contain the timesโ dependence of the KP hierarchy. Our solution (6.1) simply amounts to identifying $`N=m`$, $`e^{\eta _i}=x_i^{n+m}\lambda _i`$, $`p_i=x_i`$ and $`q_i=1/x_i`$, $`i=1,2,\mathrm{},m`$. This surprising relation suggests the existence of an underlying integrable structure for our recursion relations, but remains mysterious.
Again, to further fix the โintegration constantsโ $`\lambda _1,\lambda _2,\mathrm{},\lambda _m`$, we simply have to express that $`u_1=u_2=\mathrm{}=u_m=0`$, with the result
$$\lambda _i=\underset{ji}{}\frac{1x_ix_j}{x_ix_j}i=1,2,\mathrm{},m$$
and finally the complete solution to (5.1) reads
$$R_n=R\frac{U_n(w_1,\mathrm{},w_m)U_{n+3}(w_1,\mathrm{},w_m)}{U_{n+1}(w_1,\mathrm{},w_m)U_{n+2}(w_1,\mathrm{},w_m)}$$
where
$$U_n(w_1,\mathrm{},w_m)det\left[U_{n+2j2}(w_i)\right]_{1i,jm}$$
in terms of Chebyshev polynomials of the first kind expressed at the values $`w_i=\sqrt{x_i}+1/\sqrt{x_i}`$, $`i=1,2,\mathrm{},m`$. The precise proof of these statements can be found in .
For illustration, in the tetra/hexavalent case $`m=2`$, we have the characteristic equation
$$0=\chi _2(x)1g_4R\left(x+\frac{1}{x}+4\right)g_6R^2\left(x^2+6x+16+\frac{6}{x}+\frac{1}{x^2}\right)$$
and for instance the result (6.1)(6.1) reads for $`n=0`$:
$$R_0=R\frac{1+(x_1+\frac{1}{x_1})(x_2+\frac{1}{x_2})}{1+(1+x_1+\frac{1}{x_1})(1+x_2+\frac{1}{x_2})}=R\frac{14g_4R15g_6R^2}{13g_4R10g_6R^2}$$
where we have reexpressed the symmetric functions of $`x_1+1/x_1`$ and $`x_2+1/x_2`$ in terms of the coefficients of $`\chi _2(x)`$. The result is nothing but the generating function for rooted tetra/hexavalent planar graphs.
6.4. Trivalent case
Repeating the exercise of Sect.6.2 for the trivalent case of eqs. (5.1), we have found the following solution
$$R_n=R\frac{u_nu_{n+2}}{u_{n+1}^2},u_n=1\lambda x^{n+1}$$
while
$$S_n=S\frac{t_n}{u_nu_{n+1}},t_n=gR^2(1x)(1x^2)\lambda x^n$$
In eqs.(6.1)(6.1), the parameter $`x`$ is the unique solution to the linearized characteristic equation
$$1g^2R^3\left(x+\frac{1}{x}+2\right)=0$$
such that $`|x|<1`$.
Requiring that $`R_1=0`$, fixes $`\lambda =1`$, and we obtain the complete solution
$$\begin{array}{cc}\hfill R_n& =R\frac{(1x^{n+1})(1x^{n+3})}{(1x^{n+2})^2}\hfill \\ \hfill S_n& =SgR^2\frac{(1x)(1x^2)x^n}{(1x^{n+1})(1x^{n+2})}\hfill \end{array}$$
from which we read off the following compact expressions for $`R_0`$ and $`S_0`$, respectively generating two- and one-leg diagrams of trivalent planar graphs, with respectively the two legs in the same face and the leg in the external face:
$$\begin{array}{cc}\hfill R_0& =R\frac{x+\frac{1}{x}+1}{x+\frac{1}{x}+2}=Rg^2R^4\hfill \\ \hfill S_0& =SgR^2\hfill \end{array}$$
In particular, $`R_0`$ is the generating function for rooted trivalent planar graphs.
6.5. Arbitrary valences
The case of arbitrary valences still awaits a good solution, in the same spirit as the case of even valences. It is however possible to find โintegrable-likeโ solutions to the corresponding recursion relations (containing only one integration constant), which for the time being are still too restrictive to describe the general case. Indeed, we need a sufficient number of integration constants to allow for satisfying all the necessary initial conditions of our combinatorial problem. This number is exactly the degree of the characteristic equation of the linearized recursions, when expressed as a polynomial in $`x+1/x`$.
For arbitrary values of $`g_3,g_4,g_5,\mathrm{}`$ we have the following โone-$`x`$โ solutions:
$$\begin{array}{cc}\hfill R_n& =R\frac{(1\lambda x^{n+1})(1\lambda x^{n+3})}{(1\lambda x^{n+2})^2}\hfill \\ \hfill S_n& =S\sqrt{Rx}\frac{(1x)^2\lambda x^n}{(1\lambda x^{n+1})(1\lambda x^{n+2})}\hfill \end{array}$$
where $`x`$ is any solution with modulus less than 1 to the corresponding linearized characteristic equation. The latter reads for instance in the case of tri/tetravalent graphs where only $`g_3`$ and $`g_4`$ are non-zero:
$$\left(g_4R(x+\frac{1}{x}+4)+S(2g_3+3g_4S)1\right)^2R(g_3+3g_4S)^2(x+\frac{1}{x}+2)=0$$
and amounts to
$$\sqrt{R}(g_3+3g_4S)(\sqrt{x}+\frac{1}{\sqrt{x}})=1g_4R(x+\frac{1}{x}+4)S(2g_3+3g_4S)$$
as $`R`$ has a power series expansion of the form $`R=1+O(g_3,g_4)`$. In this particular case, it would be desirable to obtain the full solution involving the two roots $`x_1`$ and $`x_2`$ of (6.1) and two integration constants, to be able to solve simultaneously the two boundary conditions $`R_1=0`$ and $`R_1S_1=0`$ (to be understood as $`lim_{n1}R_nS_n=0`$) obtained from (5.1) at $`n=0`$, and clearly not satisfied by (6.1).
6.6. Bipartite $`p`$-valent case
Repeating the usual exercise with eq.(5.1) in which we first eliminate $`X_n`$, we have found the following solution, including only one integration constant:
$$R_n=R\frac{u_nu_{n+p+1}}{u_{n+1}u_{n+p}},u_n=1\lambda x^{n+1}$$
valid provided $`x`$ is chosen among the roots of modulus less than 1 of the linearized characteristic equation:
$$1g\stackrel{~}{g}_1R^{p2}\frac{1}{x^{p2}}(1+x+x^2+\mathrm{}+x^{p2})^2=0$$
As noticed in the case of planar graphs of arbitrary valences, the degree of this polynomial of $`x+1/x`$ is however $`p2`$, hence only for $`p=3`$ is the solution (6.1) completely general. This is the case of bipartite trivalent planar graphs, for which eq.(5.1) reduces to $`R_n=1+g\stackrel{~}{g}_1R_n(R_{n+1}+R_{n1})`$. In this case, the solution is further fixed by requiring $`R_1=0`$, hence $`\lambda =1`$, and it reads
$$R_n=R\frac{(1x^{n+1})(1x^{n+5})}{(1x^{n+2})(1x^{n+4})}$$
with $`1g\stackrel{~}{g}_1R(x+1/x+2)=0`$, $`|x|<1`$. $`R_n`$ is the generating function for bipartite trivalent graphs with two legs attached to vertices of opposite colors, and geodesic distance between those less or equal to $`n`$. In particular,
$$R_0=R\frac{x^2+\frac{1}{x^2}+x+\frac{1}{x}+1}{(x+\frac{1}{x})(x+\frac{1}{x}+2)}=R\frac{13g\stackrel{~}{g}_1R+g^2\stackrel{~}{g}_1^2R^2}{12g\stackrel{~}{g}_1R}$$
is the generating function for rooted bipartite trivalent planar graphs (with weights $`g`$/$`\stackrel{~}{g}_1`$ per black/white vertex)
For $`p4`$, we must work out the generalizations of (6.1), which now read as follows. Introducing
$$\begin{array}{cc}\hfill p(x)& =x(1+x+\mathrm{}+x^{p2}),q(x)=p(1/x)\hfill \\ \hfill p_i& =p(x_i),q_i=q(x_i)\hfill \\ \hfill c_{i,j}& =\frac{(p_ip_j)(q_iq_j)}{(p_iq_j)(q_ip_j)}\hfill \end{array}$$
where $`x_i`$, $`i=1,2,\mathrm{},p2`$ denote the generically distinct roots of eq.(6.1) with modulus less than 1, the general solution now takes the form
$$R_n=R\frac{u_n^{(m)}u_{n+p+1}^{(m)}}{u_{n+1}^{(m)}u_{n+p}^{(m)}}$$
with $`m=p2`$ and as before
$$u_n^{(m)}=\underset{l=0}{\overset{m}{}}(1)^l\underset{1i_1<i_2<\mathrm{}<i_lm}{}\underset{t=1}{\overset{l}{}}\lambda _{i_l}x_{i_l}^n\underset{1r<sl}{}c_{i_r,i_s}$$
Note the absolutely remarkable fact that we obtain again an expression in terms of the tau-function for the KP hierarchy, but with different kinematics, in the form of an implicit relation between the $`p`$โs and $`q`$โs of eq.(6.1).
Imposing moreover the vanishing of the first terms $`u_1=u_2=\mathrm{}=u_{p+1}`$, we find that
$$\lambda _i=x_i^{(p1)m1}\underset{ji}{}\frac{q_ip_j}{p_ip_j}$$
This fixes completely the solution to our combinatorial problem, and in particular gives a compact expression for the generating function $`R_0`$ for rooted bipartite $`p`$-valent planar graphs. For $`p=4`$ for instance, it reads
$$R_0=R\frac{15g\stackrel{~}{g}_1R^2+3g^2\stackrel{~}{g}_1^2R^4}{13g\stackrel{~}{g}_1R^2}$$
while $`R`$ satisfies $`R=1+3g\stackrel{~}{g}_1R^3`$.
6.7. Constellations
The solution for general $`p`$-constellations is similar to that for pure $`p`$-valent bipartite graphs. Indeed, we find that the general solution has the exact same form (6.1) as in the previous section, with $`u_n^{(m)}`$ given by (6.1) and $`p_i,q_i,c_{i,j}`$ defined as in (6.1). The only difference is that now $`m`$ may take a larger value say $`(p1)k1`$ in the case when only $`g,\stackrel{~}{g}_1,\stackrel{~}{g}_2,\mathrm{},\stackrel{~}{g}_k`$ are non-zero (namely of constellations with white $`p`$ valent vertices, and black vertices with valences $`p`$, $`2p`$, โฆ,$`kp`$). The $`x`$โs entering these formulas are now the generically distinct roots with modulus less than 1 of the linearized characteristic equation, itself a polynomial of degree $`(p1)k1`$ of $`x+1/x`$, reading
$$1=\left(1+\frac{1}{x}+\mathrm{}+\frac{1}{x^{p2}}\right)\underset{i1}{}g^i\stackrel{~}{g}_iR^{(p1)i2}\underset{m=0}{\overset{(p1)i1}{}}\underset{j=0}{\overset{i1}{}}\left(\genfrac{}{}{0pt}{}{j+m}{m}\right)\left(\genfrac{}{}{0pt}{}{pi2jm}{ij1}\right)x^{mj(p1)}$$
For illustration, in the case of $`3`$-constellations with say only $`g`$ and $`\stackrel{~}{g}_2`$ non-zero ($`R_n`$ obeys eq.(5.1) with $`\stackrel{~}{g}_1=0`$), we get the linearized characteristic equation:
$$1g^2\stackrel{~}{g}_2R^3(x+\frac{1}{x}+2)(x^2+\frac{1}{x^2}+x+\frac{1}{x}+6)=0$$
of degree $`3`$ in $`x+1/x`$. Picking the three roots with $`|x_i|<1`$, we finally get the generating function for rooted 3-constellations with white trivalent and black hexavalent vertices
$$R_0=R\frac{117g^2\stackrel{~}{g}_2R^3+25g^4\stackrel{~}{g}_2^2R^6}{110g^2\stackrel{~}{g}_2R^3}$$
while $`R`$ satisfies the equation (4.1) with only $`g`$ and $`\stackrel{~}{g}_2`$ non-zero, namely $`R=1+10g\stackrel{~}{g}_2R^4`$.
6.8. Ising model
We have not been able to find a nice structure for the general solution of the even-valent bipartite case in general. In the particular case of the Ising model with zero magnetic field (eq.(5.1)), we have been able to derive all possible solutions involving only one integration constant.
The usual linearization of eq.(5.1) yields the following characteristic equation
$$\begin{array}{cc}\hfill \left(x+\frac{1}{x}+\frac{c}{gV(13gV)}\frac{1gV}{gV}\right)& (x+\frac{1}{x}\frac{c}{gV(13gV)}\frac{1gV}{gV})\times \hfill \\ & \times (x+\frac{1}{x}+\frac{1+gV}{gV})=0\hfill \end{array}$$
of degree $`3`$ in $`x+1/x`$. As already discussed before, we would need in principle to find the full solution to (5.1) that converges at large $`n`$, including all three $`x_1,x_2,x_3`$, respectively the roots of the three factors in eq.(6.1) with modulus less than 1, and therefore including also three integration constants. We now display the solutions with one $`x`$ for each of the three factors in (6.1).
For both $`x=x_1`$ and $`x=x_2`$, we have found that
$$V_n=V\frac{u_nu_{n+3}}{u_{n+1}u_{n+2}},u_n=1\lambda x^n$$
while for $`x=x_1`$:
$$R_n=RV\frac{\lambda (1x)(1x^2)x^n}{u_{n+1}u_{n+2}}$$
and for $`x=x_2`$:
$$R_n=R+V\frac{\lambda (1x)(1x^2)x^n}{u_{n+1}u_{n+2}}$$
For $`x=x_3`$ however, the solution is quite different:
$$\begin{array}{cc}\hfill V_n& =V\frac{u_nu_{n+3}}{u_{n+1}u_{n+2}}\hfill \\ \hfill u_n& =12\lambda x^nz\lambda ^2x^{2n}\hfill \\ \hfill z& =\frac{\left((c2)(x+\frac{1}{x})+c8\right)\left((c+2)(x+\frac{1}{x})+c+8\right)}{\left(x^2+\frac{1}{x^2}(c4)(x+\frac{1}{x})(c2)\right)\left(x^2+\frac{1}{x^2}+(c+4)(x+\frac{1}{x})+(c+2)\right)}\hfill \end{array}$$
The growing complexity of the solutions lets us expect a quite involved general three $`x`$-solution, yet to be found.
7. Continuum limit
The exact solutions of Sect. 6 allow us to investigate the properties of the corresponding classes of planar graphs in terms of the geodesic distance, in the limit when the latter becomes large. This limit must clearly be taken simultaneously with the so-called critical limit of large graphs, reached in turn by letting the various weights per vertex approach some critical locus, corresponding to approaching the finite radii of convergence of the various combinatorial series involved.
The large $`n`$ limit of $`R_n=R+a_1x_1^n+\mathrm{}`$ where $`x_1`$ is the largest root with modulus less than 1 of the characteristic equation, leads to the natural definition of a correlation length $`\xi =\mathrm{Log}|x_1|`$, governing the exponential decay of $`R_nR\mathrm{exp}(n/\xi )`$ as a function of the geodesic distance $`n`$. A good continuum limit may therefore be reached by letting $`x_1`$ (and possibly other $`x`$โs) tend to the value $`1`$, while simultaneously keeping $`n/\xi `$, the continuum geodesic distance, fixed. This in turn implies certain relations between the vertex weights are reached, via the characteristic equation relating them to the $`x`$โs. These relations express nothing but the abovementioned critical limit, which must therefore be taken simultaneously with the continuum one.
7.1. Tetravalent case
We illustrate the above with the case of tetravalent graphs, with $`R_n`$ given by (6.1). The critical limit $`gg_c=1/12`$ is reached by taking say
$$g=g_c(1ฯต^4),R=\frac{R_c}{1+ฯต^2}$$
with $`R_c=2`$, and the solution of eq.(6.1) with modulus less than 1 reads
$$x(g)=\frac{1+2ฯต^2ฯต\sqrt{3(2+ฯต^2)}}{1ฯต^2}$$
hence $`x=e^{ฯต\sqrt{6}}+O(ฯต^2)`$ as $`ฯต0`$, and $`\xi 1/ฯต`$. Finally setting
$$n=\frac{r}{ฯต}$$
we may simply express the continuum limit of the quantity $`R_n`$, or more interestingly that of $`RR_n`$, generating two-leg diagrams of tetravalent planar graphs with distance at least $`n`$ between the two legs. We find that
$$(r)=\underset{ฯต0}{lim}\frac{RR_n}{ฯต^2R}=\frac{3}{\mathrm{sinh}^2\left(\sqrt{\frac{3}{2}}r\right)}$$
This is the continuum two-point correlation function for random surfaces with two marked points at (rescaled) geodesic distance larger or equal to $`r`$. It coincides with the scaling function derived in , by use of transfer matrix formalism. This in turn yields the continuum two-point correlation function for random surfaces with two marked points at (rescaled) geodesic distance $`r`$:
$$๐ข(r)=^{}(r)=3\sqrt{6}\frac{\mathrm{cosh}\left(\sqrt{\frac{3}{2}}r\right)}{\mathrm{sinh}^3\left(\sqrt{\frac{3}{2}}r\right)}$$
This result may in turn be interpreted in terms of graphs with large but finite number $`N`$ of vertices. Indeed, the above scaling relations (7.1) and (7.1) imply the following relation between the correlation length and the deviation from the critical point $`\xi (g_cg)^\nu `$, with the exponent $`\nu =1/4`$, and therefore the fractal dimension $`d_F=1/\nu =4`$ for the present model of random planar surfaces. More concretely, this tells us in particular that the number of faces in a large graph lying at geodesic distance $`n`$ from the external one behaves as $`n^{d_F}=n^4`$. A simple measure of this number is indeed the ratio $`R_n|_{g^N}/R_0|_{g^N}`$ of the corresponding coefficients of $`g^N`$ in the two power series expansions, giving the proportion of graphs with the two legs distant by at most $`n`$ to that with the two legs in the same face. Using our exact solution and performing a saddle-point expansion, we find
$$\underset{N\mathrm{}}{lim}\frac{R_n|_{g^N}}{R_0|_{g^N}}\frac{3}{56}n^4$$
giving an explicit illustration of the fractal dimension 4.
This suggests to set $`n=rN^{1/4}`$ and to write $`R_n|_{g^N}`$, again by use of a saddle-point expansion, as
$$R_n|_{g^N}\frac{4}{\pi }\frac{(12)^N}{N^{3/2}}_0^{\mathrm{}}๐uu^2\left(1+\mathrm{Re}(r\sqrt{iu})\right)$$
and similarly for $`R|_{g^N}`$ with $``$ replaced by $`0`$. The ratio $`R_n|_{g^N}/R|_{g^N}`$ gives the probability $`P(r)`$ for a random surface with two marked points that their geodesic distance be less or equal to $`r`$:
$$P(r)=\frac{2}{\sqrt{\pi }}_0^{\mathrm{}}๐uu^2e^{u^2}\left(1+\mathrm{Re}(r\sqrt{iu})\right)$$
7.2. Arbitrary even valences, multicriticality
Repeating the analysis of previous section for the case of graphs with even valences say up to $`2m+2`$, various critical points may be reached in the space of weights $`g_j`$. Concretely, picking the particular weights that ensure the following form for (2.1):
$$\frac{g_cg}{g_c}=\left(\frac{V_cV}{V_c}\right)^{m+1}$$
where $`V=gR`$, $`g_4=g`$, $`g_{2j}=g^jz_j`$ fixed by the form (7.1) and $`V_c=m/6`$, $`g_c=m/(6(m+1))`$, the corresponding multicritical limit is obtained by setting
$$g=g_c(1ฯต^{2(m+1)}),R=R_c\frac{1ฯต^2}{1ฯต^{2(m+1)}}$$
with $`R_c=V_c/g_c=m+1`$. Remarkably, the characteristic equation (6.1) then turns into
$$\chi _m(x)=\left(\frac{V_cV}{V_c}\right)^mP_m\left(\frac{1ฯต^2}{ฯต^2}\left(x+\frac{1}{x}2\right)\right)=0$$
for some fixed degree $`m`$ polynomial
$$P_m(u)=\underset{l=0}{\overset{m}{}}(u)^l\frac{l!}{(2l+1)!}\frac{m!}{(ml)!}$$
This allows to get the leading behavior of the various $`x`$โs as
$$x_i=e^{a_iฯต}+O(ฯต^2)$$
where $`a_i^2`$ are the roots of $`P_m`$, and $`a_i`$ are taken with positive real part. Consequently all the $`x`$โs tend to 1 simultaneously, and the correlation length of the problem reads $`\xi 1/ฯต`$, hence we now have the relation $`\xi (g_cg)^\nu `$ with $`\nu =2(m+1)`$, i.e. a fractal dimension $`d_F=2(m+1)`$.
The multicritical continuum limit is therefore still obtained by setting (7.1), and the exact solution (6.1) yields the following two-point correlation of random surfaces with multicritical weights (known to simulate non-unitary matter conformal field theories with central charges $`c(2,2m+1)=13(2m1)^2/(2m+1)`$ coupled to two-dimensional quantum gravity), with two marked points at geodesic distance less or equal to $`r`$:
$$(r)=2\frac{d^2}{dr^2}\mathrm{Log}๐ฒ(\mathrm{sinh}\left(a_1\frac{r}{2}\right),\mathrm{sinh}\left(a_2\frac{r}{2}\right),\mathrm{},\mathrm{sinh}\left(a_m\frac{r}{2}\right))$$
where $`๐ฒ(f_1,f_2,\mathrm{},f_m)`$ stands for the Wronskian determinant $`det\left[f_i^{(j1)}\right]_{1i,jm}`$.
It is known from matrix model solutions that the general case of arbitrary valences leads to the same multicritical points. In particular we expect the scaling functions (7.1) to be the same at these points. The same remark applies to constellations as well. New multicritical points corresponding to conformal theories with central charges $`c(p,q)=16(pq)^2/(pq)`$ for $`p,q`$ two coprime integers can be reached within the framework of two-matrix models, corresponding to the general bipartite graphs. In the latter case, we expect some new scaling functions, characteristic of these other universality classes. An example will be given in next section, when discussing the Ising model.
7.3. Critical/continuum limit in general
The (multi-) critical continuum limits are reached by letting a number of the roots $`x`$ of the characteristic equation tend to 1 simultaneously.
An alternative route for deriving the critical continuum limit of say the tetravalent case would have been to postulate the form $`R_n=R(1ฯต^2(nฯต))`$ and plug this ansatz into the recursion relation (5.1). With $`g=g_c(1ฯต^4)`$, expanding the equation up to order $`4`$ in $`ฯต`$, we arrive at the following differential equation for $``$:
$$^{\prime \prime }(r)3^2(r)6(r)=0$$
The function $``$ of eq.(7.1) is the unique solution to (7.1) such that $`(0)=\mathrm{}`$ and $`(\mathrm{})=0`$.
More generally, we may derive differential equations for the multicritical cases of Sect.7.2 as well. These take the form
$$_{m+1}[1+]=_{m+1}[1]$$
where $`_m[u]`$ is the $`m`$th KdV residue $`(d^2u)^{m1/2}|_{d^1}`$ where $`dd/dr`$ . For instance, in the case of multicritical tetra/hexavalent graphs ($`m=2`$) eq.(7.1) reads
$$^{(4)}(r)10(r)^{\prime \prime }(r)10^{\prime \prime }(r)5(^{}(r))^2+10((r))^3+30((r))^2+30(r)=0$$
Even more generally, recall that the KdV residues naturally arise (e.g. in the context of matrix model solutions) when solving the differential operator equation $`[P,Q]=1`$, where $`Q=d^2u`$ and $`P`$ some degree $`2m+1`$ differential operator. The equation indeed boils down to $`2d/dr(_{m+1}[u])=1`$. In the present case, we rather have to write the equation $`[P,Q]=0`$, which turns into $`_{m+1}[u]=`$const.
At the discrete level, comparing say (5.1) with the equations determining the matrix model solution for tetravalent graphs of arbitrary genus, we simply would have to replace $`1`$ by $`n/N`$ in the r.h.s. of (5.1). We may think of the differential operator $`Q=d^2u`$ as the continuum limit of the operator $`Q`$ (5.1) defined in Sect.5.2, now acting on functions of the variable $`r=nฯต`$. In the other cases described in Sect.6, we always have such an operator $`Q`$ at hand and again the recursion relations resemble strongly those obtained in the solutions of the multi-matrix models describing two-dimensional quantum gravity coupled with matter, up to the same substitution $`1n/N`$. This suggests that the relevant (coupled) differential equations governing the (multi-) critical continuum limit should read $`[P,Q]=0`$, with $`Q`$ the continuum limit of the operator say $`Q_1`$ used in our approach, taking the form of a differential operator of degree $`q`$ say, and $`P`$ a differential operator of degree $`p`$ coprime with $`q`$. Our claim is that the generalized two-point functions $`(๐)`$ for random surfaces in the presence of critical matter (corresponding to conformal field theories with central charges $`c(p,q)<1`$), with two marked points at geodesic distance less than $`r`$, should be governed by $`[P,Q]=0`$, where $`Q=d^qqud_.^{q2}..`$, and $`u=1+`$. This is illustrated in the case of the Ising model in the next section.
7.4. Ising model
The multicritical limit of the Ising model is obtained as follows. Starting from the equations (4.1) rewritten as $`W(R)=0`$, $`W`$ a polynomial of degree $`5`$, the tri-critical points are determined by setting $`W^{}(R)=W^{\prime \prime }(R)=0`$, and we find that $`c_c=\pm 4`$ while $`g_c=10/9`$, and $`R_c=3/5`$, $`V_c=3/10`$. The tricritical limit is approached by setting
$$c=4,R=R_c(1ฯต^2)g=g_c(1\frac{16}{5}ฯต^6)$$
We note that in the characteristic equation (6.1) only the first and last factor tend to $`x+1/x2`$ as $`ฯต0`$, while the middle one tends to $`x+1/x+10`$. This means that only two of the three $`x`$โs tend to 1 in this limit. More precisely, we have $`x_3=e^{\sqrt{6}ฯต}+O(ฯต^2)`$ and $`x_1=e^{2\sqrt{3}ฯต}+O(ฯต^2)`$. This displays the fractal dimension of graphs with critical Ising configurations, namely $`d_F=6`$, obtained by expressing the correlation length $`\xi (g_cg)^\nu `$, where $`\nu =1/6`$. Recall however that the distance $`n`$ or its rescaled version $`r`$ are slightly different from the true geodesic distance in the Ising tetravalent graphs, and the fractal dimension measured here might be different from that associated to the true geodesic distance.
Further substituting
$$\begin{array}{cc}\hfill R_n& =R(1ฯต^2(nฯต)),r=nฯต\hfill \\ \hfill V_n& =\frac{R_n}{c+g(R_{n+1}+R_n+R_{n1})}\hfill \end{array}$$
into the recursion relations (5.1) together with (7.1), we find by expanding up to order 6 in $`ฯต`$ that the two-point function $``$ obeys the differential equation
$$^{(4)}(r)18(r)^{\prime \prime }(r)18^{\prime \prime }(r)9(^{}(r))^2+24((r))^3+72((r))^2+72(r)=0$$
This equation is precisely what one would get by writing $`[P,Q]=0`$ for differential operators $`Q=d^33ud3u^{}/2`$ and $`P`$ of order $`4`$, $`u=1+`$.
Looking for convergent solutions in the form $`(r)=ae^{kr}`$, we find that $`k_1=\sqrt{6}`$ or $`k_2=2\sqrt{3}`$, corresponding to $`e^{k_1r}=x_3^n`$ and $`e^{k_2r}=x_1^n`$ respectively. Proceeding like in the discrete case, we may now solve the differential equation order by order in $`e^{k_ir}`$, with a double power series expansion $`(r)=_{m,p0}a_{m,p}e^{r(mk_1+pk_2)}`$, in terms of the two integration constants $`\lambda =a_{1,0}`$, $`\mu =a_{0,1}`$, while $`a_{0,0}=0`$. The differential equation (7.1) indeed simply amounts to a recursion relation on the coefficients $`a_{m,p}`$. Resumming the series for $`(r)`$ finally yields:
$$\begin{array}{cc}\hfill (r)& =\frac{d^2}{dr^2}\mathrm{Log}(1\frac{\lambda }{6}e^{r\sqrt{6}}\frac{\mu }{12}e^{2r\sqrt{3}}\frac{\lambda ^2}{288}e^{2r\sqrt{6}}\hfill \\ & \frac{1712\sqrt{2}}{72}\lambda \mu e^{r(\sqrt{6}+2\sqrt{3})}+\frac{577408\sqrt{2}}{3456}\lambda ^2\mu e^{2r(\sqrt{6}+\sqrt{3})})\hfill \end{array}$$
The two integration constants are further fixed by the boundary conditions. The latter are obtained by requiring that the recursion relations (5.1) also make sense at $`n=0,1`$, namely $`R_1=V_1=0`$, while $`lim_{n0}V_{n1}V_{n2}=0`$. The first conditions give $`(0)=\mathrm{}`$, while the latter implies $`R_1R_2=0`$. As in the case of planar graphs with only tetra/hexavalent vertices, this implies a higher order vanishing of the argument of the logarithm in (7.1), which plays the role of continuum limit of $`u_n`$, while the condition implies that both $`u_1`$ and $`u_2`$ vanish. This is easily solved into $`\lambda =12(17+12\sqrt{2})`$ and $`\mu =12(4+3\sqrt{2})`$, so that finally
$$(r)=\frac{d^2}{dr^2}\mathrm{Log}\left(\mathrm{sinh}\left(r(\sqrt{6}+\sqrt{3})\right)+(17+12\sqrt{2})\mathrm{sinh}\left(r(\sqrt{6}\sqrt{3})\right)2(4+3\sqrt{2})\mathrm{sinh}(r\sqrt{3})\right)$$
and we also get the correlation $`๐ข(r)=^{}(r)`$ for the Ising model on random surfaces, with two marked points at (special) geodesic distance $`r`$.
8. Conclusion
In this note, we have addressed the problem of enumeration of various types of planar graphs with external legs, while keeping track of a suitable geodesic distance between these legs. The basic tool we used are blossom-trees, namely trees carrying the minimal information needed to close them back into planar graphs. This information is essentially contained in the two types of leaves (black and white), and we have devised a compact algebraic way of keeping track of the geodesic distance between legs, by introducing operators $`Q`$ describing the structure of the rooted trees around their first vertex. This operator acts formally on a basis $`|n`$ indexed by relative integers, and may as well be viewed as acting on sequences $`\{p_n\}_{n\text{ZZ}}`$, via the shift operator $`\sigma `$ and its relative integer powers, and a number of (combinatorial) diagonal operators. This definition is clearly borrowed from that of the $`Q`$ operator of matrix models, that generate the multiplication by an eigenvalue $`\lambda `$ on the basis of (right) monic (bi-)orthogonal polynomials $`p_n(\lambda )`$, $`n=0,1,2\mathrm{}`$ In this framework, the main recursion relations for the coefficients of $`Q`$ are obtained by considering the operator $`P`$, acting on the $`p_n(\lambda )`$ by differentiation w.r.t. $`\lambda `$. The canonical relation $`[P,Q]=1`$ determines in fact all the functions of the problem, and turns into differential equations in the critical scaling limit. Here, by analogy with this case, we have been led to set $`[P,Q]=0`$, for some operator $`P`$ still awaiting a good combinatorial meaning. Nevertheless, this equation also turns into differential equations for the physical quantities of our problem. The mysterious part of this correspondence is that the natural variable in the matrix model approach is the rescaled cosmological constant, which generates the topological expansion of the free energy, namely the expansion in powers of $`r`$ of the function $`u(r)`$ has coefficients corresponding to planar graphs of fixed genus. This must be contrasted with the present findings, where $`r`$ has the meaning of geodesic distance. This seems to indicate that a more general structure should exist, that includes both the topological and geodesic directions, probably some suitably defined matrix model of sorts.
In view of these strong analogies with matrix model solutions, we may want to characterize the class of possibly decorated planar graphs for which a tree formulation exists as that for which there exists a matrix model formulation admitting a solution via orthogonal polynomials. This would exclude for instance the case of the three-state Potts model, a generalization of the Ising model whose configurations are graphs with vertices of three possible colors, and edge weights $`w_{a,b}=e^{K\delta _{a,b}}`$ according to the colors $`a,b`$ of the adjacent vertices. More generally, loop models on graphs also correspond to matrix models without orthogonal polynomial solutions: their configurations are simply mutually- and self-avoiding loops drawn on the edges of planar graphs, with a weight $`n`$ per loop (the so-called O(n) model, extensively solved in ). It would be extremely interesting to investigate any of these models using tree techniques.
A final striking outcome of our work is the emergence of soliton-like tau functions entering the explicit exact formulas for a number of generating functions for two-leg diagrams with legs distant by at most $`n`$. A reason why this should happen in the first place may be perhaps traced back to the integrability of the recursion relations we have obtained. A first indication of this integrability is the existence of โintegrals of motionโ for these equations. For instance, the recursion relation of the tetravalent case, eq.(5.1), has the following integral of motion:
$$\begin{array}{cc}\hfill f(R_n,R_{n+1})& =\mathrm{const}.\hfill \\ \hfill f(x,y)& =xy(1gxgy)xy\hfill \end{array}$$
as is immediately checked by forming
$$f(R_n,R_{n+1})f(R_n,R_{n1})=(R_{n+1}R_{n1})(R_n1gR_n(R_{n+1}+R_n+R_{n1}))$$
More generally, one may construct such integrals of motion for all the models studied in this note. Again, it may be that this integrability property relates to the existence, for the same planar graphs but with arbitrary topology rather than planar with fixed geodesic distances, of matrix model formulations that are solvable via orthogonal polynomial techniques.
Acknowledgments This note summarizes work essentially done in collaboration with J. Bouttier and E. Guitter. We also thank M. Bousquet-Mรฉlou and G. Schaeffer for fruitful discussions on constellations and bipartite graphs.
References
relax E. Brรฉzin, C. Itzykson, G. Parisi and J.-B. Zuber, Planar Diagrams, Comm. Math. Phys. 59 (1978) 35-51. relax P. Di Francesco, P. Ginsparg and J. ZinnโJustin, 2D Gravity and Random Matrices, Physics Reports 254 (1995) 1-131. relax B. Eynard, Random Matrices, Saclay Lecture Notes (2000), available at http://www-spht.cea.fr/lectures\_notes.shtml relax W. Tutte, A Census of planar triangulations Canad. Jour. of Math. 14 (1962) 21-38; A Census of Hamiltonian polygons Canad. Jour. of Math. 14 (1962) 402-417; A Census of slicings Canad. Jour. of Math. 14 (1962) 708-722; A Census of Planar Maps, Canad. Jour. of Math. 15 (1963) 249-271. relax V.G. Knizhnik, A.M. Polyakov and A.B. Zamolodchikov, Fractal Structure of 2D Quantum Gravity, Mod. Phys. Lett. A3 (1988) 819-826; F. David, Conformal Field Theories Coupled to 2D Gravity in the Conformal Gauge, Mod. Phys. Lett. A3 (1988) 1651-1656; J. Distler and H. Kawai, Conformal Field Theory and 2D Quantum Gravity, Nucl. Phys. B321 (1989) 509-527. relax G. Schaeffer, Bijective census and random generation of Eulerian planar maps, Electronic Journal of Combinatorics, vol. 4 (1997) R20; see also G. Schaeffer, Conjugaison dโarbres et cartes combinatoires alรฉatoires PhD Thesis, Universitรฉ Bordeaux I (1998). relax J. Bouttier, P. Di Francesco and E. Guitter, Census of planar maps: from the one-matrix model solution to a combinatorial proof, Nucl. Phys. B645\[PM\] (2002) 477-499. relax M. Bousquet-Mรฉlou and G. Schaeffer, Enumeration of planar constellations, Adv. in Applied Math., 24 (2000) 337-368. relax J. Bouttier, P. Di Francesco and E. Guitter, Counting colored Random Triangulations, Nucl.Phys. B641 (2002) 519-532. relax D. Poulalhon and G. Schaeffer, A note on bipartite Eulerian planar maps, preprint (2002), available at http://www.loria.fr/$``$schaeffe/ relax J. Bouttier, P. Di Francesco and E. Guitter, Combinatorics of hard particles on planar maps, Nucl. Phys. B655 (2003) 313-341. relax M. Bousquet-Mรฉlou and G. Schaeffer, The degree distribution in bipartite planar maps: application to the Ising model, preprint math.CO/0211070. relax P. Chassaing and G. Schaeffer, Random Planar Lattices and Integrated SuperBrownian Excursion, preprint (2002), to appear in Probability Theory and Related Fields, math.CO/0205226. relax J. Bouttier, P. Di Francesco and E. Guitter, Random trees between two walls: Exact partition function, J. Bouttier, P. Di Francesco and E. Guitter, Saclay preprint t03/086 and cond-mat/0306602 (2003), to appear in J. Phys. A: Math. Gen. (2003). relax J. Bouttier, P. Di Francesco and E. Guitter, Geodesic distance in planar graphs, Nucl. Phys. B 663\[FS\] (2003) 535-567. relax H. Kawai, N. Kawamoto, T. Mogami and Y. Watabiki, Transfer Matrix Formalism for Two-Dimensional Quantum Gravity and Fractal Structures of Space-time, Phys. Lett. B 306 (1993) 19-26. relax J. Ambjรธrn and Y. Watabiki, Scaling in quantum gravity, Nucl.Phys. B445 (1995) 129-144. relax J. Ambjรธrn, J. Jurkiewicz and Y. Watabiki, On the fractal structure of two-dimensional quantum gravity, Nucl.Phys. B454 (1995) 313-342. relax M. Jimbo and T. Miwa, Solitons and infinite dimensional Lie algebras, Publ. RIMS, Kyoto Univ. 19 No. 3 (1983) 943-1001, eq.(2.12). relax I. Gelfand and L. Dikii, Fractional powers of operators and Hamiltonian systems, Funct. Anal. Appl. 10:4 (1976) 13. relax B. Eynard and C. Kristjansen, Exact Solution of the O(n) Model on a Random Lattice, Nucl.Phys. B455 (1995) 577-618, and More on the exact solution of the O(n) model on a random lattice and an investigation of the case $`|n|>2`$, Nucl.Phys. B466 (1996) 463-487.
|
warning/0506/astro-ph0506423.html
|
ar5iv
|
text
|
# The Rise of Dwarfs and the Fall of Giants: Galaxy Formation Feedback Signatures in the Halo Satellite Luminosity Function
## 1. Introduction
The complex physical processes associated with reionization of the intergalactic medium (IGM) at a redshift above 6 is expected to leave characteristic scales and features in the star formation history (e.g., Cen 2003), in the supernovae distribution (e.g., Mesinger et al. 2005), and, potentially, in the luminosity distributions of galaxies. In the case of galaxy statistics, for example, the feedback related to supernovae heating in small mass halos and photoionization during reionization (Benson et al., 2002a). have been used to explain the flattened faint-end slope of the galaxy luminosity function (LF).
Since one averages galaxy statistics over large volumes and, thereby, averages over any inhomogeneities and differences in time scales and dispersions coming from galaxy formation and evolution processes, it is not possible to address detailed physics related to reionization with the galaxy LF alone. On the other hand, the luminosity distribution of satellites in dark matter halos, as a function of the halo mass, may be an ideal probe of reionization physics. In this respect, the lack of an abundant population of low luminosity galaxies in the local group, relative to expectations from cold dark matter cosmological models, has been explained in terms of photoionization (Bullock et al., 2000; Benson et al., 2003). The subsequent squelching of galaxy formation in dark matter halos below a certain mass scale, corresponding to the temperature to which IGM is heated, explains the environmental dependence of the faint-end slope of the cluster LF (Tully et al., 2002).
In addition to effects related to photoionization, the cluster satellite LF should also show signatures of feedback associated with starformation. The satellite luminosity distribution can be described through the halo occupation number (Cooray & Sheth, 2002), minus the central galaxy, conditioned in terms of satellite luminosity. Here, we construct an empirical model for the conditional luminosity function (CLF; Yang, Mo, & van den Bosch 2003; Cooray & Milosavljeviฤ 2005; Cooray 2005) of satellites in dark matter halos, based on the subhalo mass function and compare to observed measurements; at the bright-end, we make use of CLFs measured by Yang et al. (2005) using the 2dFGRS (Cole et al., 2001) galaxy group catalog, and extend this comparison to the faint-end using cluster LFs measured by Trentham & Hodgkin (2002) and Hilker et al. (2003). We argue that cluster LFs show two scales, one associated with photoionization at halo mass scales around $`5\times 10^{10}`$ M, resulting in the disappearance of dwarfs satellites in galaxy groups relative to clusters, and another scale related to an overall suppression of galaxy formation in subhalos below $`10^{11}`$ M, independent of the total system mass.
This Letter is organized as follows: In ยง 2, we describe the construction of the satellite CLF of dark matter halos. In ยง 3, we compare our LF with the observed LFs of satellites in groups and clusters and discuss interesting physics that could explain the observed features. We make use of the current concordance cosmological model (Spergel et al., 2003).
## 2. Satellite Luminosity Function
The CLF, denoted by $`\mathrm{\Phi }(L|M)`$, is the average number of galaxies with luminosities between $`L`$ and $`L+dL`$ that reside in halos of mass $`M`$ (Yang, Mo, & van den Bosch, 2003; Cooray, 2005). Following Cooray & Milosavljeviฤ (2005), we separate the CLF into terms associated with central and satellite galaxies, $`\mathrm{\Phi }(L|M)=\mathrm{\Phi }_\mathrm{c}(L|M)+\mathrm{\Phi }_\mathrm{s}(L|M)`$. In previous studies (Cooray, 2005), central galaxy CLF was described with a log-normal distribution in luminosity with a mean $`L_\mathrm{c}(M)`$ and dispersion $`\mathrm{\Sigma }_c`$, while the satellite CLF is assumed to be a power law. Here, we focus on the satellite CLF and, instead of an a priori assumption on a power-law CLF, we model it using the subhalo mass function.
In this approach, each satellite in a subhalo mass $`M_s`$ has a log-normal luminosity distribution of
$`\varphi _\mathrm{s}(L|M_s)`$ $`=`$ $`{\displaystyle \frac{\varphi (M_s)}{\sqrt{2\pi }\mathrm{ln}(10)\mathrm{\Sigma }L}}\mathrm{exp}\left\{{\displaystyle \frac{\mathrm{log}_{10}[L/L_\mathrm{s}(M_s)]^2}{2\mathrm{\Sigma }^2}}\right\},`$ (1)
where the normalization $`\varphi (M_s)`$ is fixed such that $`\mathrm{\Phi }(L|M_s)L๐L=L_s(M_s)`$. Given the luminosity distribution of each satellite, the CLF of satellites, in a parent halo mass $`M`$, is
$$\mathrm{\Phi }_\mathrm{s}(L|M)=_0^{\mathrm{}}\varphi _\mathrm{s}(L|M_s)\frac{dn_s(M_s|M)}{dM_s}๐M_s,$$
(2)
where $`dn_s(M_s|M)/dM_s`$ is the subhalo mass function of dark matter halos given the parent halo mass $`M`$. Here, we use the analytical form
$$\frac{dn_s(M_s|M)}{dM_s}=\frac{\gamma }{\beta ^2M\mathrm{\Gamma }(2\alpha )}\left(\frac{M_s}{\beta M}\right)^\alpha \mathrm{exp}\left(\frac{M_s}{\beta M}\right),$$
(3)
where $`\alpha =1.91`$, $`\beta =0.39`$, and $`\gamma =0.18`$ (Vale & Ostriker, 2004). Our conclusions do not change significantly if we use the description of van den Bosch et al. (2005) where $`\alpha `$ and $`\gamma `$ are functions of mass, $`M`$, with $`\alpha `$ varying roughly over 10% as $`M`$ is varied from group to cluster mass scales. Since CLF measurements are averaged over a sample of dark matter halos in a narrow range in mass (Yang et al., 2005), we also calculate the mass-averaged satellite CLF by averaging over the dark matter halo mass function, $`dn/dM`$ (Sheth & Tormen 1999), over the mass range of interest.
In our model, an important ingredient is the $`L_s(M_s)`$ relation which describes the luminosity of a subhalo given the subhalo mass. Here, we follow the procedure related to modeling the field galaxy LF (Cooray & Milosavljeviฤ, 2005), and employ a fitting function to describe the relation between the luminosity of a halo and the dark matter mass of that halo. In Vale & Ostriker (2004), this relation was obtained through a model description of the 2dFGRS LF (Norberg et al., 2001) using the global subhalo mass function, $`n_{\mathrm{sh}}(M_s)=๐n/๐M๐n_s(M_s|M)/๐M_s๐M`$. The relation is
$$L(M)=L_0\frac{(M/M_1)^a}{[b+(M/M_1)^{cd}]^{1/d}}.$$
(4)
The relevant parameters for the $`b_J`$-band, as appropriate for 2dFGRS data, are $`L_0=5.7\times 10^9L_{\mathrm{}}`$, $`M_1=10^{11}M_{\mathrm{}}`$, $`a=4.0`$, $`b=0.57`$, $`c=3.72`$, and $`d=0.23`$ (Vale & Ostriker, 2004). Though this relation was used in Cooray (2005) to describe the field galaxy LF, whose statistics are dominated by central galaxies, the same relation should also remain valid for subhalos as well. The remaining parameter in our model for the satellite LF is $`\mathrm{\Sigma }`$ and we set this to be 0.17 based on the value needed to explain the exponential drop-off in the field galaxy LF (Cooray, 2005).
Note that at low subhalo masses, $`dn_s/dM_sM_s^\alpha `$ where $`\alpha =1.91`$, independent of the halo mass, or varies from 2.0 to 1.9 when parent halo mass varies from $`M10^{11}\mathrm{M}_{}`$ to $`10^{15}\mathrm{M}_{}`$ (van den Bosch et al., 2005). The satellite CLF is $`\mathrm{\Phi }_\mathrm{s}(L|M)\varphi _\mathrm{s}(L|M_s)(dn_s/dM_s)๐M_s`$. If $`L_\mathrm{s}M_s^\eta `$, we can write $`\mathrm{\Phi }_\mathrm{s}(L|M)L^{1\alpha /\eta +1/\eta }\delta (L^{}L)๐L^{}`$, where we have ignored the scatter in the $`L_\mathrm{s}`$$`M_s`$ relation by setting $`\mathrm{\Sigma }0`$. The faint-end of the satellite LF then scales as $`\mathrm{\Phi }_\mathrm{s}(L|M)L^{1\alpha /\eta +1/\eta }`$.
## 3. Results and Discussion
In Figure 1, we show $`b_J`$-band CLFs based on the 2dFGRS group catalog (Yang et al., 2005) and an extension to the faint-end based on nearby cluster LFs in the B-band. The satellite CLFs show several interesting trends. As one moves to a higher mass for the central halo, from (a) to (d) in Figure 1, one finds the faint-end of the satellite CLF to be filled with dwarf galaxies. Since the 2dFGRS is incomplete below absolute magnitudes of $`M_{b_J}`$ of -17, we make use of B-band LF measurements by Trentham & Hodgkin (2002), whose faint-end statistics are generally dominated by the dwarf galaxy population. At the high mass end of halos, corresponding to clusters like Virgo, with an assumed halo mass of $`10^{14}`$ M, and Coma, with mass between $`(7`$ and $`10)\times 10^{14}`$ M, satellites begin to trace the expected slope predicted by the subhalo mass function. On the other hand, in low mass groups, such as Ursa Major, with an assumed mass of $`(5`$ to $`8)\times 10^{12}`$ M, the faint-end dwarf population does not trace the subhalo mass function. The statistics of the faint-end dwarf galaxy population have been discussed in Tully et al. (2002) in the context of photoionization effect resulting from reionization. These authors argue that massive clusters such as Virgo started to form prior to reionization, while low mass groups such as Ursa Major, where the abundance of dwarf galaxies is relatively smaller, formed subsequent to reionization. As shown in Figure 1(c) and (d), the faint-end dwarf LF is associated with subhalo masses below $`5\times 10^{10}`$ M. The photoionized, and heated, gas is not expected to cool in halos of below this mass scale after reionization. Such an argument is consistent with Figure 1.
Based on satellite LFs alone, we find that reionization may be more complex than simply assuming that the universe reionized completely at a single redshift. For example, in Figure 1(a) and (b), we also show the LF of dwarf galaxies in Fornax. We plot these data in both panels due to a large uncertainty, or variation, in the quoted total mass of Fornax in the literature. Regardless of the exact mass of Fornax, the presence of more dwarfs than Ursa Major suggests that Fornax formed prior to the latter system and probably during reionization such that gas managed to cool in a fraction of small dark matter halos while the rest was affected. The large scatter in dwarf population of similar mass groups may be evidence for inhomogeneous reionization or non-uniform feedback processes.
What is new and intriguing is this: when compared to the expectation based on the sub-halo mass function and the luminosity-mass relation of Vale & Ostriker (2004), one sees a relative decrease in the number of satellite galaxies at luminosities below $`3\times 10^9`$ L, corresponding to mass scales below few times $`10^{11}`$ M, in all halo systems. Since at these mass scales, $`3<\eta <4`$, we expect $`\mathrm{\Phi }_\mathrm{s}(L|M)`$ to scale as $`L^{1.2}`$ to $`L^{1.3}`$; we do not expect the faint-end slope to be flatter than $`L^{1.2}`$, unless the subhalo mass function slope is changed; for $`\mathrm{\Phi }_\mathrm{s}(L|M)`$ to be luminosity independent, $`\alpha 1`$. Thus, to suppress galaxies at luminosities below $`3\times 10^9`$ L, we include an efficiency function to the subhalo mass function to characterize the subhalo mass distribution where satellite galaxies present. Here, we take an analytical description of the form $`f(M_s)=0.5(1+\mathrm{erf}[(\mathrm{log}M_s\mathrm{log}M_c)]/\sigma )`$, such that $`f(M_s)0`$ when $`M_sM_c`$ and $`f(M_s)0`$ when $`M_sM_c`$. To explain the flattening of the satellite CLF, we set $`M_c=10^{11}`$ M and $`\sigma =1.5`$; with such a broad dispersion, $`f(M_s)dn(M_s|M)/dM_s`$ flattens (see, Figure 2) instead of becoming zero even when $`M10^9`$ M; the flattening of the satellite CLF does not imply that all subhalos below some critical mass scale is affected, but one sees a broad distribution of subhalo masses where, statistically, satellite galaxies are not present. Instead of modifying the subhalo mass function, we can also vary the $`L_s(M_s)`$ relation and set a steep slope for $`\eta `$. While in this case all subhalos contain galaxies, the average luminosity would be lower for subhalos in groups relative to same mass subhalos in clusters. In such a scenario, it is also hard to understand the sudden appearance of dwarfs in clusters, given the dip in the LF of giants regardless of system mass.
We now offer a physical explanation for this unique feature. Supernovae-powered winds from abundant dwarf galaxies at $`L<2\times 10^8`$ M are expected to be strong (Dekel & Silk 1986; Mori et al. 2002). We adopt the view that these winds transport metals and energy into the IGM. Following Cen & Bryan (2001), the temperature of the IGM is
$`T_{SN}(z)=1.3\times 10^4\mathrm{K}\left({\displaystyle \frac{E_{SN}}{1.2\times 10^{51}\mathrm{erg}}}\right)`$ (5)
$`\times `$ $`\left({\displaystyle \frac{M_C}{0.2\mathrm{M}_{}}}\right)\left({\displaystyle \frac{\eta }{0.3}}\right)\left({\displaystyle \frac{[C/H]}{1\times 10^3}}\right)\left({\displaystyle \frac{4}{1+z}}\right)^2`$
where $`M_C`$ is the mass of carbon ejected by one supernova; $`E_{SN}`$ is the total energy output of one supernova; $`\eta `$ is the fraction of that energy that is eventually deposited in the IGM in the form of thermal energy (Mori et al. 2002), and $`[C/H]`$ is the ratio of carbon number density to hydrogen number density of the gas in solar units. We have assumed that energy is deposited at some high redshift, perhaps $`z69`$, and the temperature of the IGM subsequently decays adiabatically. We adopt all the fiducial values for $`E_{SN}`$, $`M_C`$, $`\eta `$ and $`[C/H]`$, which gives us an added temperature to the IGM of $`1.3\times 10^4`$K at $`z3`$; this explains the Doppler width issue of the Ly$`\alpha `$ forest (Cen & Bryan 2001).
Figure 3 shows the evolution of the nonlinear mass in the standard cold dark matter model (solid curve), the Jeans mass for two extreme conditions (two short dashed curves), and the temperature of the IGM (long dashed curve). We see that, depending on detailed gasdynamics, the majority of halos formed at $`z3.44.4`$ and with masses $`10^{10}6\times 10^{10}h^1\mathrm{M}_{}`$, are significantly deprived of gas, potentially explaining the dip in the satellite LFs seen in Figure 1.
Our results suggest that most subhalos of 10<sup>10</sup> M to $`10^{11}`$ M in massive clusters should not have galaxies in them (so-called โdark halosโ). Lensing-based studies indicate the agreement between dark matter subhalo mass function and the mass function of subhalos associated with galaxies in clusters when $`M_s>3\times 10^{11}`$ M (Natarajan & Springel, 2004). While the observed difference below this mass scale is considered to be due to an observational limitation, it is useful to extend lensing studies to measure the subhalo mass function below $`10^{11}`$ M. The dip in the LF could be detected as a dip in the mass function of subhalos with galaxies (Figure 2). Furthermore, a large sample of satellite LFs, down to dwarf galaxy magnitudes and with better determined parent halo masses, have the potential to address more detailed physics of reionization and galaxy formation processes through a better characterization of features in the CLF.
AC thanks Frank van den Bosch for helpful correspondence and X. Yang for LF measurements. This work was completed while AC was at the Aspen Center for Physics. This work is supported in part by grants AST-0206299, AST-0407176 and NAG5-13381 to RC.
|
warning/0506/astro-ph0506534.html
|
ar5iv
|
text
|
# On cosmic acceleration without dark energy
## I Introduction
Recent observations of the expansion history of the Universe indicate that the Universe is presently undergoing a phase of accelerated expansion original ; acceleratedreview . The accelerated expansion is usually interpreted as evidence either for a โdark energyโ (DE) component to the mass-energy density of the Universe, or for a modification of gravity at large distances. In this paper we explore another possibility, namely that the accelerated expansion is due to the presence of inhomogeneities in the Universe.
In the homogeneous, isotropic, Friedmann-Robertson-Walker (FRW) cosmology, the acceleration (or deceleration) of the expansion may be expressed in terms of a dimensionless parameter $`q`$, proportional to the second time derivative of the cosmic scale factor $`a`$. It is uniquely determined in terms of the relative densities and the equations of state of the various fluids by (overdots denote time derivatives),
$$q\frac{\ddot{a}a}{\dot{a}^2}=\frac{1}{2}\mathrm{\Omega }_{\mathrm{TOTAL}}+\frac{3}{2}\underset{i}{}w_i\mathrm{\Omega }_i,$$
(1)
where $`\mathrm{\Omega }_{\mathrm{TOTAL}}`$ is the total density parameter and the factors $`\mathrm{\Omega }_i`$ are the relative contributions of the various components of the energy density with equation of state $`w_i=P_i/\rho _i`$ ($`P_i`$ and $`\rho _i`$ being the pressure and energy density of $`i`$-th fluid). The expansion accelerates if $`q<0`$. Observations seem to require DE with present values $`w_{DE}1`$ and $`\mathrm{\Omega }_{DE}0.7`$ rp . The negative value of $`w_{DE}`$, indicating a violation of the energy condition $`w>1/3`$ hawking , is usually interpreted as the effect of a mysterious dark energy fluid of unknown nature or a cosmological constant of surprisingly small magnitude.
The existence of a negative-pressure fluid or a cosmological constant would have profound implications for physics as well as cosmology. While the observational evidence for the acceleration of the Universe is now compelling, it is important to keep in mind that the evidence for dark energy is indirect; it is inferred from the observed time evolution of the expansion rate of the Universe. What is known is that the expansion history of the Universe is not described by the expansion history of an Einsteinโde Sitter Universe (a spatially flat, matter-dominated FRW model). While such a departure may be caused by dark energy, there are other possibilities. One possibility is that general relativity is not a good description of gravity on large distance scales. Another possibility is that the Universe is matter-dominated and described by general relativity, and the departure of the expansion rate from the Einsteinโde Sitter model is the result of back reactions of cosmological perturbations. This explanation is the most conservative, since it assumes neither a cosmological constant, a negative-pressure fluid, nor a modification of general relativity.
In this paper we explore the possibility that backreactions of cosmological perturbations is the source of the accelerated expansion KMNR ; oldKMNR ; japan ; oldrasanen ; rasanen ; alessio ; sw ; bmr ; rtolman . The idea is as follows. We know there exist cosmological perturbations; after all, the Universe is inhomogeneous. To describe the time evolution of a patch of the Universe as large as our local Hubble radius one has to construct the effective dynamics from which observable average properties can be inferred. Of course, this implies a scale-dependent description of inhomogeneities. Suppose further that our Universe is filled with pressureless matter and no DE. If inhomogeneities evolve with time, a local observer would infer that our Universe is not expanding as a homogeneous and isotropic FRW matter-dominated Universe with Hubble rate $`H(t)t^{2/3}`$, where $`t`$ is cosmic time. On the contrary, the Universe would appear to have an expansion rate with a time evolution that depends on the nature and time evolution of these perturbations. Potentially, this could lead to an accelerated expansion.
Our idea is actually intimately connected with the general problem of how the nonlinear dynamics of cosmological perturbations on small scales may affect the large-scale โbackgroundโ geometry. Let us start by discussing this issue in some generality.
The standard approach to cosmology is based both on observational facts, such as the near-perfect isotropy of the Cosmic Microwave Background (CMB) radiation, and on an a priori philosophical assumption, the so-called Cosmological or Copernican Principle. According to the Cosmological Principle all comoving cosmic observers at a given cosmic time should see identical properties around them (isotropy around all cosmic observers implies homogeneity, hence the FRW line element). The Cosmological Principle allows one to circumvent our inability to obtain information about the Universe outside our past light-cone by assuming that a symmetry principle is valid everywhere. By using the Cosmological Principle, we assume that we are able to determine conditions many Hubble radii away from us by using observational data within our past light-cone, whose region of influence is, by definition, limited to one Hubble radius ellis84 .
An alternative procedure, dubbed Observational Cosmology, has also been proposed. It aims at constructing a cosmological model solely in terms of observational facts, thereby avoiding any a priori assumptions of global symmetry. It dates back to the works by Kristian and Sachs in 1966 ks and Ellis in 1984 ellis84 . A remarkable feature of this approach is that, by using Einsteinโs equations, the dataset observable within our past light-cone is precisely sufficient to determine the space-time and its matter content within the same light-cone (see Ref. ellis84 and references therein).
A crucial ingredient of the Observational Cosmology approach, which is shared by any realistic cosmological model-fitting procedure, is smoothing. Observations tell us that the Universe is far from homogeneous and isotropic on small scales. Observationally, we know that homogeneity, e.g., in the galaxy distribution, is only achieved over some large smoothing scale (see e.g., Ref. lahav ). When we refer to homogeneity and isotropy of the Universe we tacitly assume that spatial smoothing over some suitably large filtering scale has been applied so that fine-grained details can be ignored (see in this respect the discussion in Refs. carf ; carf2 ). In other words, by the mere assumption that the same background model can be used to describe the properties of nearby and very distant objects in the Universe, the smoothing process is implicit in the way we fit a FRW model to observations. Cosmological parameters like the Hubble expansion rate or the energy density of the various cosmic components are to be considered as volume averaged quantities: only these can be compared with observations ellisnew .
There is, however, a technical difficulty inherent in any smoothing procedure. While matter smoothing is somewhat straightforward (e.g., in the fluid description), smoothing of the space-time metric is more complex and immediately leads to an important and unexpected feature, pointed out by Ellis ellis84 . Let us assume that Einsteinโs equations hold on some suitably small scale where the Universe is highly inhomogeneous and anisotropic. Next, suppose we smooth over some larger scale. Einsteinโs equations are nonlinear: smoothing and evolution (i.e., going to the field equations) will not commute. Hence, the Einstein tensor computed from the smoothed metric would generally differ from that computed from the smoothed stress-energy tensor. The difference is a tensor appearing on the right-hand side (RHS) of Einsteinโs equations that leads to an extra term in the effective Friedmann equations describing the dynamics of the smoothing domain.
How can this fact be related to the acceleration of the Universe? The answer is that this extra source term need not satisfy the usual energy conditions (according to which our Universe can only decelerate) even if the original matter stress-energy tensor does. The fact that the effective stress-energy tensor emerging after smoothing could lead to a violation of the energy conditions was originally recognized by Ellis ellis84 .<sup>1</sup><sup>1</sup>1A closely related discussion can also be found in Ref. turb . As we will discuss in Sec. II, explicit calculations of the effective Friedmann equations buchert1 ; buchert2 ; japan confirm that acceleration is indeed possible even if our Universe is filled solely with matter.
A closely related question is what is the appropriate scale for which the smoothing procedure can fit the standard picture of a homogeneous and isotropic Universe on large scales? Our choice will be that of smoothing over a volume of size comparable with present-day Hubble volume. The precise size of the averaging volume does not matter, provided it is large enough that the fair sample hypothesis applies, i.e., that volume averages yield an accurate approximation of statistical ensemble averages. We will nonetheless refer to scales within (outside) the averaging domain as โsub-Hubbleโ (โsuper-Hubbleโ) perturbations. In doing this we are however promoting our super-Hubble, or โzeroโ-mode, to the role of FRW-like background.
The next question is what are the scales that determine the dynamics of our local background. To answer this question we have to recall what happens in the standard FRW models. The evolution of the global scale factor $`a(t)`$, the zero-mode of FRW models, is fully determined by the matter content of the Universe through the value of $`\mathrm{\Omega }_{\mathrm{TOTAL}}`$ and $`\mathrm{\Omega }_i`$, and via the equation of state $`w_i`$ of its components. That is where microphysics enters the game. In other words, in the standard FRW picture the evolution of the Universe as a whole is determined by the dynamics of matter on sub-Hubble scales. Similarly, in the Observational Cosmology approach the dynamics of our local background must be determined by the observed behavior of matter inhomogeneities within our past-light-cone. That is where the backreaction of sub-Hubble inhomogeneities enters the game. This picture will become clear in Sec. II, where we will introduce two scalars, the so-called kinematical backreaction $`Q_D`$ and the mean spatial curvature $`R_D`$, that enter the expression for the energy density and pressure in the effective Friedmann equations governing the mean evolution of our local domain $`D`$. The crucial point is that in the fully general relativistic framework these two scalars are linked together by an integrability condition (which has no analogue in the Newtonian context), whose solution provides the effective equation of state of the backreaction. In order to solve this equation and establish the typical size of these terms one needs a non-perturbative and non-Newtonian approach to the evolution of cosmological irregularities, as pointed out in Refs. ehlers ; bks .
The fact that the average dynamics naturally leads to new terms in the source implies that it is legitimate to use an effective Friedmann description, provided one takes into account that the effective sources of these equations contain the back-reaction terms.
What will result from our analysis is that the evolution of sub-Hubble perturbations leads to an instability of the perturbative expansion due to the presence of large contributions which depend on a combination of Newtonian and post-Newtonian terms. This instability indicates that the effective scale factor describing the dynamics of our local Hubble patch is fed by the evolution of inhomogeneities within the Hubble radius. This cross-talk between the small-scale dynamics and the effective average dynamics described by super-Hubble, or โzeroโ-mode, playing the role of FRW-like background might be the crucial ingredient of backreaction that can lead to cosmic acceleration without dark energy. In Ref. KMNR a deviation from the pure matter-dominated evolution was obtained by a combination of sub- and super-Hubble modes generated by inflation, the latter being improperly used to amplify the backreaction. In this paper, we will show that the deviation from a matter-dominated background is entirely due to the nonlinear evolution of sub-Hubble modes which may cause a large backreaction (technically due to the disappearance of the filter modeling the volume average), while the super-Hubble modes play no dynamical role.
At this point it may be useful to contrast the differences between our approach dealing with inhomogeneities with the traditional approach. In the traditional approach one averages over inhomogeneities and forms a time-dependent average energy density $`\rho (\stackrel{}{x},t)`$ (although the standard procedure is to calculate averages with the unperturbed spatial metric!). One then uses for the dynamics of the โzero-modeโ \[$`a(t)`$\] the dynamics of a homogeneous universe with energy density $`\rho (t)=\rho (\stackrel{}{x},t)`$. One then regards inhomogeneities as a purely โlocalโ effect, for instance, leading to peculiar velocities. In this approach inhomogeneities can not result in acceleration.
In our approach, we take note of the fact that the expansion rate of an inhomogeneous universe of average density $`\rho (\stackrel{}{x},t)`$ (using the inhomogeneous spatial metric to calculate the spatial average!) is not the same as the expansion rate of a universe with the same average density. In order to account for this we encode the expansion dynamics into a new zero mode (or scale factor) $`a_D(t)`$ (which will be properly defined in the next section). It is the dynamics of this renormalized scale factor that is best used to calculate observables and will determine whether the Universe accelerates. In our approach the effect of short-wavelength inhomogeneities is not just a local effect, but renormalizes the long-wavelength dynamics.
Our paper is organized as follows. In Sec. II we summarize the effective Friedmann description of an inhomogeneous Universe after smoothing. In Sec. III we discuss how acceleration in our Hubble patch can result from the backreaction of perturbations. Conclusions are drawn in Sec. IV. The Appendix presents the main results of a fourth-order gradient-expansion technique.
## II Effective Friedmann equations in an inhomogeneous Universe
The goal of this section is to compute the time dependence of the local expansion rate of the Universe. For a generic fluid we may take the four-velocity to be $`u^\mu =(1,\stackrel{}{0})`$, which amounts to saying that a local observer is comoving with the energy flow of the fluid. For the case of irrotational dust considered in this paper we have the freedom to work in the synchronous and comoving gauge with line element
$$ds^2=dt^2+h_{ij}(๐ฑ,t)dx^idx^j,$$
(2)
where $`t`$ is cosmic time.
A fundamental quantity in our analysis is the velocity gradient tensor, which is purely spatial and symmetric because of irrotationality. It is defined as
$$\mathrm{\Theta }_j^i=u_{;j}^i=\frac{1}{2}h^{ik}\dot{h}_{kj}.$$
(3)
Here dots denote derivatives with respect to cosmic time. The tensor $`\mathrm{\Theta }_j^i`$, represents the extrinsic curvature of the spatial hypersurfaces orthogonal to the fluid flow. It may be written as
$$\mathrm{\Theta }_j^i=\mathrm{\Theta }\delta _j^i+\sigma _j^i.$$
(4)
Here $`\mathrm{\Theta }`$ is called the volume-expansion scalar, reducing to $`3H`$ ($`H`$ is the usual Hubble rate) in the homogeneous and isotropic FRW case. The traceless tensor $`\sigma _j^i`$ is called the shear.
The evolution equations for the expansion and the shear come from the space-space components of Einsteinโs equations (see e.g., Ref. mater ). They read, respectively, ($`\rho `$ is the mass density, $`R`$ and $`R_j^i`$ are the spatial Ricci scalar and tensor respectively of comoving space-like hypersurfaces)
$`\dot{\mathrm{\Theta }}+\mathrm{\Theta }^2+R`$ $`=`$ $`12\pi G\rho ,`$ (5)
$`\dot{\sigma }_j^i+\mathrm{\Theta }\sigma _j^i+R_j^i{\displaystyle \frac{1}{3}}R\delta _j^i`$ $`=`$ $`0.`$ (6)
The $`00`$ component of Einsteinโs equations is the energy constraint
$$\frac{2}{3}\mathrm{\Theta }^22\sigma ^2+R=16\pi G\rho ,$$
(7)
where $`\sigma ^2\frac{1}{2}\sigma _j^i\sigma _i^j`$. The $`0i`$ components yield the momentum constraint
$$\sigma _{j|i}^i\frac{2}{3}\mathrm{\Theta }_{,j}=0,$$
(8)
where the vertical bar denotes covariant differentiation in the three-space with metric $`h_{ij}`$. The mass density, in turn, can be obtained from the continuity equation
$$\dot{\rho }=\mathrm{\Theta }\rho ,$$
(9)
whose solution reads
$$\rho =\rho _0\left(h/h_0\right)^{1/2},$$
(10)
where $`h\mathrm{det}h_{ij}`$ and the initial conditions have been arbitrarily set at the present time $`t_0`$. Combining the expansion evolution equation with the energy constraint gives the Raychaudhuri equation,
$$\dot{\mathrm{\Theta }}+\frac{1}{3}\mathrm{\Theta }^2+2\sigma ^2+4\pi G\rho =0.$$
(11)
From the latter equation it is straightforward to verify that irrotational pressure fluid elements cannot locally undergo accelerated expansion. (This point was emphasized by Hirata and Seljak seljak .) Indeed, defining a local deceleration parameter and using the Raychaudhuri equation, one finds
$$q\left(3\dot{\mathrm{\Theta }}+\mathrm{\Theta }^2\right)/\mathrm{\Theta }^2=6(\sigma ^2+2\pi G\rho )/\mathrm{\Theta }^20.$$
(12)
While it is true that locally the expansion does not accelerate, it is incorrect to assume that acceleration can not occur when the fluid is coarse-grained over a finite domain. The reason is trivial: the time derivative of the average of $`\mathrm{\Theta }`$ and the average of the time derivative of $`\mathrm{\Theta }`$ are not the same because of the time dependence of the coarse-graining volume.
Let us denote the coarse-grained value of a quantity $``$ by its average over a spatial domain $`D`$:<sup>2</sup><sup>2</sup>2Notice that one is not allowed to define the mean cosmological parameters only through an average over directions flanagan ; seljak as cosmological observables, such as the Hubble rate, receive unacceptably large corrections from the same Newtonian terms which become harmless surface terms when averaging over a large volume oldKMNR . We acknowledge discussions with U. Seljak about this issue.
$$_D=\frac{_D\sqrt{h}d^3x}{_D\sqrt{h}d^3x}.$$
(13)
We will take the domain to be comparable with the size of the present Hubble volume<sup>3</sup><sup>3</sup>3The correct definition of our comoving Hubble radius is $`R_H(t_0)=e^{\mathrm{\Psi }_\mathrm{}0}^{t_0}๐te^{\mathrm{\Psi }_{\mathrm{}}(t)}`$..
A first important property follows directly from the smoothing procedure itself: for a generic function $``$ one has buchert1 ; buchert2
$$_D^{}\dot{}_D=\mathrm{\Theta }_D\mathrm{\Theta }_D_D,$$
(14)
where we have not considered terms originating from the peculiar motion of the boundary, since we will eventually consider only comoving domains in what follows. In particular, for the local expansion rate one finds
$$\mathrm{\Theta }_D^{}=\dot{\mathrm{\Theta }}_D+\mathrm{\Theta }^2_D\mathrm{\Theta }_D^2\dot{\mathrm{\Theta }}_D.$$
(15)
Although $`\dot{\mathrm{\Theta }}_D\frac{1}{3}\mathrm{\Theta }^2_D0`$, the coarse-grained deceleration parameter $`q_D3\mathrm{\Theta }_D^{}/\mathrm{\Theta }_D^21`$ is related to $`\mathrm{\Theta }_D^{}`$, which is not the same as $`\dot{\mathrm{\Theta }}_D`$. It is precisely this commutation rule that allows for the possibility of acceleration in our local patch in spite of the fact that fluid elements cannot individually undergo accelerated expansion. This simple argument circumvents the no-go theorem adopted in Refs. flanagan ; seljak (and later in giov ), according to which there can be no acceleration in our local Hubble patch if the Universe only contains irrotational dust.
Indeed, let us follow the work of Buchert buchert1 ; buchert2 and define a dimensionless scale factor
$$a_D(t)\left(\frac{V_D}{V_{D_0}}\right)^{1/3};V_D=_D\sqrt{h}d^3x,$$
(16)
where $`V_D`$ is the volume of our coarse-graining domain (the subscript โ$`0`$โ denotes the present time). As an averaging volume we may take a large comoving domain, so that $`D`$ is constant in time. Alternative choices are however possible (see, e.g., Ref. elsto for a discussion of different averaging procedures).<sup>4</sup><sup>4</sup>4One could alternatively average over a volume of size of the order of the instantaneous particle horizon. The effective dynamics in this case will be accompanied by a stochastic source originated by the statistical nature of the perturbations.
The coarse-grained Hubble rate $`H_D`$ will be
$$H_D=\frac{\dot{a}_D}{a_D}=\frac{1}{3}\mathrm{\Theta }_D.$$
(17)
(Notice that with such a coarse-graining, $`H_D`$ coincides with the quantity $`\overline{H}`$ defined in Ref. KMNR ). By properly smoothing Einsteinโs equations over the volume $`V_D`$, one obtains buchert1 ; buchert2
$`{\displaystyle \frac{\ddot{a}_D}{a_D}}`$ $`=`$ $`{\displaystyle \frac{4\pi G}{3}}\left(\rho _{\mathrm{eff}}+3P_{\mathrm{eff}}\right),`$ (18)
$`\left({\displaystyle \frac{\dot{a}_D}{a_D}}\right)^2`$ $`=`$ $`{\displaystyle \frac{8\pi G}{3}}\rho _{\mathrm{eff}},`$ (19)
where we have defined effective energy density and pressure terms
$`\rho _{\mathrm{eff}}`$ $`=`$ $`\rho _D{\displaystyle \frac{Q_D}{16\pi G}}{\displaystyle \frac{R_D}{16\pi G}}`$ (20)
$`P_{\mathrm{eff}}`$ $`=`$ $`{\displaystyle \frac{Q_D}{16\pi G}}+{\displaystyle \frac{R_D}{48\pi G}},`$ (21)
and we have introduced the kinematical backreaction
$$Q_D=\frac{2}{3}\left(\mathrm{\Theta }^2_D\mathrm{\Theta }_D^2\right)2\sigma ^2_D.$$
(22)
From the effective Friedmann equations, Eqs. (18) and (19), obtained by Buchert in Ref. buchert1 , one immediately obtains the continuity equation for our effective fluid
$$\dot{\rho }_{\mathrm{eff}}=3H_D\left(\rho _{\mathrm{eff}}+P_{\mathrm{eff}}\right).$$
(23)
Note that the smoothed continuity equation differs from the local continuity equation. On the other hand, owing to the fact that our coarse-graining volume is comoving with the mass flow, mass conservation is preserved by the smoothing procedure, implying
$$\rho _D^{}=3H_D\rho _D.$$
(24)
The two quantities $`Q_D`$ and $`R_D`$ are not independent. This can be seen by taking the derivative of Eq. (19) and using Eq. (24). The consistency of the system of Eqs. (18), (19), and (24) requires that $`Q_D`$ and $`R_D`$ satisfy the integrability condition buchert1
$$\left(a_D^6Q_D\right)^{}+a_D^4\left(a_D^2R_D\right)^{}=0.$$
(25)
One should stress that the latter equation, i.e., the link between kinematical backreaction $`Q_D`$ and mean curvature $`R_D`$, is a genuine General Relativistic (GR) effect, having no analogue in Newtonian theory, as the curvature $`R`$ of comoving hypersurfaces vanishes identically in the Newtonian limit ellis ; mater ; buchert1 , implying that there exist globally flat Eulerian coordinates $`X^i`$. Indeed, in the Newtonian case, it is immediate to verify that $`Q_D`$ is exactly (i.e., at any order in perturbation theory) given by the volume integral of a total-derivative term in Eulerian coordinates ehlers ,
$$Q_D^{\mathrm{Newtonian}}=\left[๐ฎ\left(๐ฎ\right)\left(๐ฎ\right)๐ฎ\right]_D,$$
(26)
where $`๐ฎ`$ is the peculiar velocity, so that by the Gauss theorem it can be transformed into a pure boundary term. It is precisely by this reason that any analysis of backreaction based on the Newtonian approximation, such as that recently performed in v1 of Ref. fry , is not relevant: it will invariably lead to a tiny effect, and to the absence of any acceleration. Indeed, if inhomogeneities only exist on scales much smaller than our Hubble radius and if peculiar velocities are small on the boundary of our Hubble patch then, within the Newtonian approximation, the standard FRW matter-dominated model can be applied without any substantial correction from the backreaction ehlers . Such a drawback of the Newtonian approximation was also noticed in Refs. oldrasanen ; oldKMNR . This exact result demonstrates that in order to deal with the backreaction, going beyond the Newtonian approximation is mandatory, as also stressed in Ref. japan . Studies of the average dynamics including the lowest post-Newtonian gradient terms in the weak field limit were considered in Refs. newton and v2 of fry . However, further and more sizeable terms are expected to contribute to the backreaction once the effective dynamics of the system (including the kinematical backreaction) is considered. We will come back to this issue in subsection IIIC.
The GR integrability condition makes it clear how acceleration in our local Hubble patch is possible. Indeed, it is immediate to realize that the general condition for acceleration in a domain with mean density $`\rho _D`$ is
$$Q_D>4\pi G\rho _D.$$
(27)
Moreover, a particular solution of the integrability condition for constant $`Q_D`$ and $`R_D`$ is
$$Q_D=\frac{1}{3}R_D=\text{const.},$$
(28)
which, for negative mean curvature mimics a cosmological constant, $`\mathrm{\Lambda }_{\mathrm{eff}}=Q_D`$. More in general, if $`Q_D`$ is positive, it may mimic a dynamical dark energy or quintessence.
So far the considerations have been rather general. Now we write the spatial metric in the general form
$$h_{ij}a^2(t)e^{2\mathrm{\Psi }(๐ฑ,t)}\left[\delta _{ij}+\chi _{ij}(๐ฑ,t)\right],$$
(29)
where $`a(t)t^{2/3}`$ is the usual FRW scale-factor for a flat, matter-dominated Universe and the traceless tensor $`\chi _{ij}`$ contains the remaining modes of the metric, namely one more scalar, as well as vector and tensor modes.<sup>5</sup><sup>5</sup>5Indices of $`\chi _{ij}`$ will be raised by the Kronecker symbol: $`\chi _j^i\delta ^{ik}\chi _{kj}`$ and $`\chi ^{ij}\delta ^{ik}\delta ^{jl}\chi _{kl}`$. Next, when we consider the expansion in some domain $`D`$, we can split the gravitational potential $`\mathrm{\Psi }`$ into two parts: $`\mathrm{\Psi }=\mathrm{\Psi }_{\mathrm{}}+\mathrm{\Psi }_s`$, where $`\mathrm{\Psi }_{\mathrm{}}`$ is the long-wavelength mode and $`\mathrm{\Psi }_s`$ is a collection of short-wavelength modes. Of course โlongโ and โshortโ describe wavelengths compared to the size of the domain $`D`$. We can easily take into account the effect of $`\mathrm{\Psi }_{\mathrm{}}`$ by noting that within $`D`$ the factor $`e^{2\mathrm{\Psi }_{\mathrm{}}}`$ is a space-independent conformal rescaling of the spatial metric. Let us then write
$$h_{ij}=a^2(t)e^{2\mathrm{\Psi }_{\mathrm{}}(t)}\stackrel{~}{h}_{ij}(๐ฑ,t),$$
(30)
with $`\stackrel{~}{h}_{ij}=e^{2\mathrm{\Psi }_s(๐ฑ,t)}\left[\delta _{ij}+\chi _{ij}(๐ฑ,t)\right]`$. The expansion scalar and shear then become
$`\mathrm{\Theta }`$ $`=`$ $`3{\displaystyle \frac{\dot{a}}{a}}+\stackrel{~}{\mathrm{\Theta }}3\dot{\mathrm{\Psi }}_{\mathrm{}},`$
$`\sigma _j^i`$ $`=`$ $`\stackrel{~}{\sigma }_j^i,`$ (31)
where $`\stackrel{~}{\mathrm{\Theta }}`$ and $`\stackrel{~}{\sigma }_j^i`$ are calculated with $`\stackrel{~}{h}_{ij}`$. Note that $`\stackrel{~}{\mathrm{\Theta }}`$ and $`\stackrel{~}{\sigma }_j^i`$ do not depend explicitly on $`\mathrm{\Psi }_{\mathrm{}}`$. It should be kept in mind that the local expansion rate is $`\mathrm{\Theta }`$, not $`\stackrel{~}{\mathrm{\Theta }}`$.
The Ricci tensor of comoving space-like hypersurfaces is given by
$$R_j^i=a^2e^{2\mathrm{\Psi }_{\mathrm{}}}\left[\stackrel{~}{R}_j^i+\stackrel{~}{}^i\stackrel{~}{}_j\mathrm{\Psi }_{\mathrm{}}+\stackrel{~}{}^2\mathrm{\Psi }_{\mathrm{}}\delta _j^i+\stackrel{~}{}^i\mathrm{\Psi }_{\mathrm{}}\stackrel{~}{}_j\mathrm{\Psi }_{\mathrm{}}\stackrel{~}{}^k\mathrm{\Psi }_{\mathrm{}}\stackrel{~}{}_k\mathrm{\Psi }_{\mathrm{}}\delta _j^i\right],$$
(32)
where $`\stackrel{~}{R}_j^i`$ is the Ricci scalar of the metric $`\stackrel{~}{h}_{ij}`$ and the symbol $`\stackrel{~}{}_i`$ denotes the covariant derivative in the 3-space with metric $`\stackrel{~}{h}_{ij}`$. For the Ricci scalar we find
$`R`$ $`=`$ $`a^2e^{2\mathrm{\Psi }_{\mathrm{}}}\left[\stackrel{~}{R}+4\stackrel{~}{}^2\mathrm{\Psi }_{\mathrm{}}2\stackrel{~}{}^k\mathrm{\Psi }_{\mathrm{}}\stackrel{~}{}_k\mathrm{\Psi }_{\mathrm{}}\right]`$ (33)
$`R_D`$ $`=`$ $`a^2e^{2\mathrm{\Psi }_{\mathrm{}}}\stackrel{~}{R}+4\stackrel{~}{}^2\mathrm{\Psi }_{\mathrm{}}2\stackrel{~}{}^k\mathrm{\Psi }_{\mathrm{}}\stackrel{~}{}_k\mathrm{\Psi }_{\mathrm{}}_D.`$ (34)
Notice that $`a_D`$ coincides with the scale factor $`\overline{a}`$ adopted in Ref. KMNR , provided we take
$$a_D(t)=a(t)e^{\mathrm{\Psi }_{\mathrm{}}(t)+\mathrm{\Psi }_\mathrm{}0}$$
(35)
with
$$\mathrm{\Psi }_{\mathrm{}}(๐ฑ_{\mathrm{obs}},t)\mathrm{ln}a\frac{1}{3}\mathrm{ln}\left(_D\sqrt{h}d^3x\right)+\mathrm{const}.,$$
(36)
where the residual dependence of $`\mathrm{\Psi }_{\mathrm{}}`$ on the spatial coordinate $`๐ฑ_{\mathrm{obs}}`$ labels the individual comoving volume patch, i.e., the specific cosmic observer we are considering. Using Eq. (10) we can rewrite Eq. (36) in the form
$$\mathrm{\Psi }_{\mathrm{}}(t)=\frac{1}{3}\mathrm{ln}(1+\delta _{\mathrm{FRW}})^1_{D_{\mathrm{in}}}+\mathrm{const}.,$$
(37)
where $`\delta _{\mathrm{FRW}}`$ is the density contrast with respect to the mean density of a flat, matter-dominated FRW (Einstein-de Sitter) model, defined through $`\rho =\left(1+\delta _{\mathrm{FRW}}\right)/\left(6\pi Gt^2\right)`$, and โinโ denotes the initial time, which for simplicity we have taken to coincide with the end of inflation. For any quantity $``$
$$_{D_{\mathrm{in}}}=\frac{_D\sqrt{h_{\mathrm{in}}}d^3x}{_D\sqrt{h_{\mathrm{in}}}d^3x}.$$
(38)
By inspecting Eq. (37) one immediately realizes that acceleration may be achieved in those Hubble patches where the mean rarefaction factor $`\left(1+\delta _{\mathrm{FRW}}\right)^1_{D_{\mathrm{in}}}`$ grows fast enough to compensate for the Einstein-de Sitter expansion rate $`H=2/3t`$. Note also that the integral defining $`\mathrm{\Psi }_{\mathrm{}}`$ is dominated by the dynamics of the most underdense fluid elements, not by the densest ones, so the complex dynamics of highly nonlinear mass concentrations never enters the calculation; by the same reasoning, any intrinsic limitation related to caustic formation would not affect the validity of our backreaction treatment.
The kinematical backreaction $`Q_D`$ is non-vanishing and gets contributions only from $`\stackrel{~}{\mathrm{\Theta }}`$ and from the shear $`\stackrel{~}{\sigma }_j^i`$:
$$Q_D=\frac{2}{3}\stackrel{~}{\mathrm{\Theta }}^2_D2\stackrel{~}{\sigma }^2_D,$$
(39)
where we used the fact that $`\stackrel{~}{\mathrm{\Theta }}_D=0`$ by construction.
In order to have a qualitative understanding of why acceleration can be the natural outcome of the backreaction, let us rewrite the mean expansion rate in terms of the peculiar expansion rate $`\theta `$, defined by $`\mathrm{\Theta }=3H+\theta `$:
$$H_D=\frac{2}{3t}+\frac{1}{3}\frac{\left(1+\delta _{\mathrm{FRW}}\right)^1\theta _{D_{\mathrm{in}}}}{\left(1+\delta _{\mathrm{FRW}}\right)^1_{D_{\mathrm{in}}}},$$
(40)
which shows that the mean expansion rate receives a correction with respect to the FRW background value by the peculiar expansion rate of mostly underdense regions (where $`\theta `$ is largest), which give the largest contribution to the average. However, as we already noticed, acceleration may be achieved when $`Q_D`$ is positive and large enough. This requires a large variance of the volume expansion rate within the averaging domain. At early times, when perturbations are linear, $`\mathrm{\Theta }`$ is narrowly peaked around its mean background value $`3H`$. When non-linearities set in, the variance increases because of the simultaneous presence of largely under- and over-dense regions (in fact, counting only under-dense regions would reduce the variance leading to an under-estimate of $`Q_D`$) comment .
## III The appearance of instabilities
In order to discuss the dynamics of our local Hubble patch, one may proceed along two complementary directions. Either one tries to encode the effect of perturbations into the local scale factor $`a_D`$ as done in Ref. KMNR , or one may try to construct an effective equation of state by computing the backreaction terms $`Q_D`$ and $`R_D`$. We will try to follow a combination of them to see under which circumstances acceleration in our Hubble patch can be achieved.
### III.1 The effect of super-Hubble modes
The mean curvature generally depends on both sub- and super-Hubble modes. Since $`Q_D`$ does not depend explicitly on perturbations with wavelengths larger than the Hubble radius, one immediately concludes that if one considers super-Hubble modes only, $`Q_D`$ vanishes and from the integrability condition $`R_D`$ scales like $`a_D^2`$. Therefore the effect of pure super-Hubble perturbations is limited to generating a true local curvature term which may be important but can not accelerate the expansion of the Universe.
As we already anticipated in the Introduction, the effective scale factor describing the dynamics of our local Hubble patch is fed by the evolution of inhomogeneities within the Hubble radius. This cross-talk between the small-scale dynamics and the effective average dynamics is a crucial ingredient, as also pointed out in Ref. KMNR . Before coming to this crucial point, let us pause for a moment and show that there is another more technical way to achieve the same conclusion about the role played by super-Hubble modes starting from the spatial-gradient expansion of Einstein equations.
The spatial-gradient expansion is a nonlinear approximation method suitable to describe the long-wavelength part of inhomogeneities in the Universe. This scheme is based on the assumption that observables like the local curvature can be expanded in powers of gradients of the perturbations. To account for the effect of super-Hubble modes at late times we may adopt the so-called renormalization group method applied to the gradient expansion of Einstein equations nambu . This will result in the renormalized long-wavelength solution to Einstein equations, valid also at late times until the long-wavelength perturbations enter the horizon.
Here we sketch a non-perturbative technique to solve Einsteinโs equations in an inhomogeneous Universe. A more detailed presentation of the method will be presented elsewhere pilli .
Our approach makes use of a gradient-expansion approximation (see Refs. LK ; Tom ; Sal ; Der ; Nam ; BS ; 4grad ). The idea is to describe the dynamics of irregularities in a Universe which contains inhomogeneities on scales larger than the Hubble radius. Working in the synchronous gauge one expands Einsteinโs equations starting from a space-dependent โseedโ metric. The lowest order solution corresponds to the so-called long-wavelength approximation, while adding higher-order gradients leads to a more accurate solution, which hopefully converges toward the exact one.
The gradient-expansion technique amounts to keeping a finite number of spatial derivatives. This approximation technique is non-perturbative in the sense that by solving for the metric coefficients $`\mathrm{\Psi }`$ and $`\chi _{ij}`$ up to $`2n`$ spatial gradients one obtains terms of any order in the conventional perturbative expansion containing up to $`2n`$ gradients. For the purpose of this paper, working to four derivatives in $`\mathrm{\Psi }`$ and $`\chi _{ij}`$ will suffice. Note that because the scalar $`\mathrm{\Psi }`$ appears in the argument of an exponential in the way we write the spatial metric, our gradient-expansion sensibly differs from that used in Refs. LK ; Tom ; Sal ; Der ; Nam ; BS ; 4grad . Even if $`\mathrm{\Psi }`$ is obtained up to a finite number of spatial gradients, $`h_{ij}`$ will necessarily contain gradient terms of any order.
Since the cosmological perturbations are generated during inflation, it is natural to set initial conditions for the gravitational perturbations $`\mathrm{\Psi }`$ and $`\chi _{ij}`$ at the end of inflation (effectively coinciding with $`t=t_{\mathrm{in}}=0`$). If so, the spatial metric in the super-horizon regime is given by $`h_{ij}=a^2e^{2\zeta }\delta _{ij}`$ where $`\zeta `$ is the curvature perturbation en . It is related to the so-called peculiar gravitational potential $`\phi (๐ฑ)`$ defined by $`\phi =3\zeta /5`$, in a matter-dominated Universe. The comoving curvature perturbation is constant on super-horizon scales (up to gradients) when only adiabatic modes are present and the decaying mode is disregarded. Therefore, the initial conditions at $`t=0`$ are $`\mathrm{\Psi }_{\mathrm{in}}\mathrm{\Psi }(t=0)=5\phi /3`$ and $`\chi _{ij}(t=0)=0`$. Notice that since cosmological perturbations generated during single-field models of inflation are very nearly Gaussian with a nearly flat spectrum en , we infer that $`\phi `$ should be regarded as a nearly scale-invariant, quasi-Gaussian random field.
From this analysis we see that $`\mathrm{\Psi }`$ contains at least a zero-derivative term: this will be our seed metric perturbation, which is the necessary ingredient for gravitational instability to develop. The traceless tensor $`\chi _{ij}`$ has at least two spatial gradients. The only exception might come from linear or higher-order tensor modes appearing at the end of inflation. Nonetheless, accounting for these contributions does not quantitatively affect our results.
Let us adopt the gradient-expansion in order to obtain $`\mathrm{\Psi }`$ and $`\chi _i^j`$ including up to four gradients of the initial seed potential $`\phi `$. Under this assumption, the inverse spatial metric can be written as
$$h^{ij}=a^2e^{2\mathrm{\Psi }}\left(\delta ^{ij}\chi ^{ij}+\chi _k^i\chi ^{kj}\right).$$
(41)
Writing $`\mathrm{\Theta }3H+\theta `$ where $`H=\dot{a}/a`$, one finds (up to higher-derivative terms)
$`\theta `$ $`=`$ $`3\dot{\mathrm{\Psi }}{\displaystyle \frac{1}{2}}\chi ^{kl}\dot{\chi }_{kl}`$ (42)
$`\sigma _j^i`$ $`=`$ $`{\displaystyle \frac{1}{2}}\dot{\chi }_j^i{\displaystyle \frac{1}{2}}\chi ^{ik}\dot{\chi }_{kj}+{\displaystyle \frac{1}{6}}\chi ^{lk}\dot{\chi }_{lk}\delta _j^i.`$ (43)
To solve for these quantities one also needs the 3D Ricci tensor and Ricci scalar of the constant-time hypersurfaces. We start by calculating the 3D Christoffel symbols, which read (up to higher-derivative terms)
$$\mathrm{\Gamma }_{jk}^i=\mathrm{\Psi }_{,k}\delta _j^i\mathrm{\Psi }_{,j}\delta _k^i+\mathrm{\Psi }^{,i}\delta _{jk}+\frac{1}{2}\left(\chi _{j,k}^i+\chi _{k,j}^i\chi _{jk}^{,i}\right)+\mathrm{\Psi }^{,i}\chi _{jk}\mathrm{\Psi }_{,l}\chi ^{il}\delta _{jk}.$$
(44)
The Ricci tensor and Ricci scalar read, respectively,
$`R_j^i`$ $``$ $`{\displaystyle \frac{_j^i}{a^2}}={\displaystyle \frac{e^{2\mathrm{\Psi }}}{a^2}}[\mathrm{\Psi }_{,j}^{,i}+^2\mathrm{\Psi }\delta _j^i+\mathrm{\Psi }^{,i}\mathrm{\Psi }_{,j}(\mathrm{\Psi })^2\delta _j^i\chi ^{ik}\mathrm{\Psi }_{,kj}\chi ^{ik}\mathrm{\Psi }^{,k}\mathrm{\Psi }_{,j}`$ (45)
$`+{\displaystyle \frac{1}{2}}\left(\chi _{,kj}^{ik}+\chi _{j,k}^{k,i}^2\chi _j^i\right)\mathrm{\Psi }^{,kl}\chi _{kl}\delta _j^i\mathrm{\Psi }_{,k}\chi _{,l}^{kl}\delta _j^i`$
$`+{\displaystyle \frac{1}{2}}\mathrm{\Psi }^{,k}(\chi _{k,j}^i\chi _{kj}^{,i}+\chi _{j,k}^i)+\mathrm{\Psi }_{,k}\mathrm{\Psi }_{,l}\chi ^{kl}\delta _j^i]`$
$`R`$ $``$ $`{\displaystyle \frac{}{a^2}}={\displaystyle \frac{e^{2\mathrm{\Psi }}}{a^2}}\left[4^2\mathrm{\Psi }2\left(\mathrm{\Psi }\right)^2+\chi _{,ij}^{ij}4\chi ^{ij}\mathrm{\Psi }_{,ij}4\chi _{,i}^{ij}\mathrm{\Psi }_{,j}+2\chi ^{ij}\mathrm{\Psi }_{,i}\mathrm{\Psi }_{,j}\right].`$ (46)
The evolution equations for the peculiar volume expansion scalar and for the shear immediately follow from Eqs. (5), (6) and (7)
$`\dot{\theta }+3H\theta +{\displaystyle \frac{1}{2}}\theta ^2+{\displaystyle \frac{3}{2}}\sigma ^2`$ $`=`$ $`{\displaystyle \frac{1}{4a^2}},`$
$`\dot{\sigma }_j^i+3H\sigma _j^i+\theta \sigma _j^i`$ $`=`$ $`{\displaystyle \frac{1}{a^2}}\left(_j^i{\displaystyle \frac{1}{3}}\delta _j^i\right).`$ (47)
Replacing our expressions for $`\theta `$, $`\chi _j^i`$, $`_j^i`$ and $``$ in terms of the metric coefficients and retaining only terms containing up to four spatial derivatives, one obtains differential equations for $`\mathrm{\Psi }`$ and $`\chi _{ij}`$, namely
$`\ddot{\mathrm{\Psi }}+3H\dot{\mathrm{\Psi }}`$ $`=`$ $`{\displaystyle \frac{3}{2}}\dot{\mathrm{\Psi }}^2{\displaystyle \frac{1}{2}}H\chi ^{kl}\dot{\chi }_{kl}{\displaystyle \frac{5}{48}}\dot{\chi }^{kl}\dot{\chi }_{kl}{\displaystyle \frac{1}{6}}\chi ^{kl}\ddot{\chi }_{kl}+{\displaystyle \frac{1}{12a^2}},`$ (48)
$`\ddot{\chi }_j^i+3H\dot{\chi }_j^i`$ $`=`$ $`3H\left(\chi ^{ik}\dot{\chi }_{kj}{\displaystyle \frac{1}{3}}\chi ^{kl}\dot{\chi }_{kl}\delta _j^i\right)+\left(\dot{\chi }^{ik}\dot{\chi }_{kj}{\displaystyle \frac{1}{3}}\dot{\chi }^{kl}\dot{\chi }_{kl}\delta _j^i\right)`$ (49)
$`+`$ $`\left(\chi ^{ik}\ddot{\chi }_{kj}{\displaystyle \frac{1}{3}}\chi ^{kl}\ddot{\chi }_{kl}\delta _j^i\right)+3\dot{\mathrm{\Psi }}\dot{\chi }_j^i{\displaystyle \frac{2}{a^2}}\left(_j^i{\displaystyle \frac{1}{3}}\delta _j^i\right).`$
These equations can be solved iteratively. With two gradients only, the conformal Ricci tensor $`_j^i`$ and scalar $``$ coincide with their initial values, i.e., with the curvature of the seed conformal metric $`e^{2\mathrm{\Psi }_{\mathrm{in}}}\delta _{ij}`$. Up to two gradients we obtain
$`\mathrm{\Psi }`$ $`=`$ $`{\displaystyle \frac{5}{3}}\phi +{\displaystyle \frac{1}{18}}\left({\displaystyle \frac{a}{a_0}}\right)\left({\displaystyle \frac{2}{H_0}}\right)^2e^{10\phi /3}\left[^2\phi {\displaystyle \frac{5}{6}}\left(\phi \right)^2\right],`$ (50)
$`\chi _j^i`$ $`=`$ $`{\displaystyle \frac{1}{3}}\left({\displaystyle \frac{a}{a_0}}\right)\left({\displaystyle \frac{2}{H_0}}\right)^2e^{10\phi /3}\left[D_j^i\phi +{\displaystyle \frac{5}{3}}\left(\phi ^{,i}\phi _{,j}{\displaystyle \frac{1}{3}}\left(\phi \right)^2\delta _j^i\right)\right],`$ (51)
where $`D_j^i^i_j\frac{1}{3}^2\delta _j^i`$.
From Eq. (50) we obtain the volume expansion scalar and the shear tensor:
$`\theta `$ $`=`$ $`{\displaystyle \frac{1}{3}}\left({\displaystyle \frac{a}{a_0}}\right)^{1/2}\left({\displaystyle \frac{2}{H_0}}\right)e^{10\phi /3}\left[^2\phi {\displaystyle \frac{5}{6}}\left(\phi \right)^2\right]`$ (52)
$`\sigma _j^i`$ $`=`$ $`{\displaystyle \frac{1}{3}}\left({\displaystyle \frac{a}{a_0}}\right)^{1/2}\left({\displaystyle \frac{2}{H_0}}\right)e^{10\phi /3}\left[D_j^i\phi +{\displaystyle \frac{5}{3}}\left(\phi ^{,i}\phi _{,j}{\displaystyle \frac{1}{3}}\left(\phi \right)^2\delta _j^i\right)\right].`$ (53)
Let us first explain how the renormalization group method works at the level of two gradients starting from the solution of Eq. (50) written in the form
$$\mathrm{\Psi }=\mathrm{\Psi }_{\mathrm{in}}+e^{2\mathrm{\Psi }_{\mathrm{in}}}ฯต\left(aa_{\mathrm{in}}\right),ฯต\frac{1}{18}\left(\frac{1}{a_0}\right)\left(\frac{2}{H_0}\right)^2\left[^2\phi \frac{5}{6}\left(\phi \right)^2\right],$$
(54)
where $`ฯต1`$. By taking the limit $`ฯต1`$ we can isolate the long-wavelength part of $`\mathrm{\Psi }`$, in other words, $`\mathrm{\Psi }_{\mathrm{}}`$.
The constant $`\mathrm{\Psi }_{\mathrm{in}}`$ represents the value of the gravitational potential at some initial instant of time when the scale factor is $`a_{\mathrm{in}}`$. We regularize the $`๐ช(ฯต)`$ secular term by introducing an arbitrary โscale factorโ $`\mu `$ and a renormalized constant $`\mathrm{\Psi }_{\mathrm{in}}=\mathrm{\Psi }_R(\mu )+ฯต\delta \mathrm{\Psi }(\mu ,a_{\mathrm{in}})`$. If we split the term $`(aa_{\mathrm{in}})`$ into $`(a\mu +\mu a_{\mathrm{in}})`$, then to first order in $`ฯต`$
$$\mathrm{\Psi }=\mathrm{\Psi }_R(\mu )+ฯต\delta \mathrm{\Psi }(\mu ,a_{\mathrm{in}})+ฯตe^{2\mathrm{\Psi }_R(\mu )}(a\mu +\mu a_{\mathrm{in}}).$$
(55)
The counterterm $`\delta \mathrm{\Psi }`$ is determined in such a way to absorb the $`(\mu a_{\mathrm{in}})`$-dependent term in the gravitational potential $`\mathrm{\Psi }`$:
$$\delta \mathrm{\Psi }(\mu ,a_{\mathrm{in}})+e^{2\mathrm{\Psi }_R(\mu )}(\mu a_{\mathrm{in}})=0.$$
(56)
This defines the renormalization-group transformation
$$\mathrm{\Psi }_R(\mu )=\mathrm{\Psi }_{\mathrm{in}}+ฯตe^{2\mathrm{\Psi }_R(\mu )}(\mu a_{\mathrm{in}}),$$
(57)
and the renormalization group equation
$$\frac{\mathrm{\Psi }_R(\mu )}{\mu }=ฯตe^{2\mathrm{\Psi }_R(\mu )}.$$
(58)
The solution of Eq. (58) is
$$\mathrm{\Psi }_R(\mu )=\frac{1}{2}\mathrm{ln}\left(c_22ฯต\mu \right),$$
(59)
where $`c_2=e^{10\phi /3}`$ is the constant of integration. Equating $`\mu `$ to the generic scale factor we find that the renormalized improved solution for the gravitational potential at the level of two gradients is given by
$$\mathrm{\Psi }=\mathrm{\Psi }_R\left(\mu =a\right)=\frac{5}{3}\phi \frac{1}{2}\mathrm{ln}\left[1\frac{1}{9}\left(\frac{a}{a_0}\right)\left(\frac{2}{H_0}\right)^2e^{10\phi /3}\left(^2\phi \frac{5}{6}\left(\phi \right)^2\right)\right].$$
(60)
If expanded up to two gradients, this solution coincides with Eq. (50). Since by construction one should take the long-wavelength part of the argument of the logarithm in the solution of Eq. (60), it is easy to see that the latter matches Eq. (36) expanded up to two gradients. Indeed, write Eq. (36) as
$$\mathrm{\Psi }_{\mathrm{}}(๐ฑ,t)\mathrm{\Psi }_{\mathrm{}}(๐ฑ,t_{\mathrm{in}})=\frac{1}{3}\mathrm{ln}\left(\frac{_De^{3\mathrm{\Psi }}d^3x}{_De^{3\mathrm{\Psi }_{\mathrm{in}}}d^3x}\right).$$
(61)
Inserting Eq. (50) into Eq. (61) and expanding up to two gradients using the fact that $`\mathrm{\Psi }_{\mathrm{}}(๐ฑ,t_{\mathrm{in}})=5\phi /3`$, we obtain
$`\mathrm{\Psi }_{\mathrm{}}(๐ฑ,t)\mathrm{\Psi }_{\mathrm{}}(๐ฑ,t_{\mathrm{in}})`$ $``$ $`{\displaystyle \frac{1}{3}}\mathrm{ln}\left({\displaystyle \frac{_De^{3\mathrm{\Psi }_{\mathrm{in}}}\left(13e^{10\phi /3}ฯตa\right)d^3x}{_De^{3\mathrm{\Psi }_{\mathrm{in}}}d^3x}}\right)`$ (62)
$``$ $`{\displaystyle \frac{1}{3}}\mathrm{ln}\left(13{\displaystyle \frac{_De^{3\mathrm{\Psi }_{\mathrm{in}}}e^{10\phi /3}ฯตad^3x}{_De^{3\mathrm{\Psi }_{\mathrm{in}}}d^3x}}\right)`$
$``$ $`{\displaystyle \frac{1}{2}}\mathrm{ln}\left(1{\displaystyle \frac{1}{9}}\left({\displaystyle \frac{a}{a_0}}\right)\left({\displaystyle \frac{2}{H_0}}\right)^2e^{10\phi /3}\left[^2\phi {\displaystyle \frac{5}{6}}\left(\phi \right)^2\right]_{D_{\mathrm{in}}}\right),`$
which coincides with Eq. (60). Notice, in particular, that Eq. (60) differs from the toy gravitational potential adopted by Hirata and Seljak seljak .
Let us compute the corresponding deceleration parameter
$$q=\frac{\dot{H}_D}{H_D^2}1=\frac{\dot{H}\ddot{\mathrm{\Psi }}_{\mathrm{}}}{\left(H\dot{\mathrm{\Psi }}_{\mathrm{}}\right)^2},$$
(63)
where we have used the fact that $`H_D=H\dot{\mathrm{\Psi }}_{\mathrm{}}`$. Inserting Eq. (60) into Eq. (63), we find that at late times
$$q\frac{3}{2}\frac{2}{3}1=0,$$
(64)
i.e., the deceleration parameter tends to zero. This result confirms our expectation that at the (resummed) lowest order in the gradient expansion, the Universe turns out to be curvature-dominated at late times, which is equivalent to a Universe with effective equation of state $`w=1/3`$. What about the resummation of the long-wavelength perturbations at higher order in gradient terms? The curvature term is a series of gradients, and can be written in the form
$$R=\underset{n1}{}e^{2n\mathrm{\Psi }_{\mathrm{in}}}c_na^{n2},$$
(65)
where $`c_n=๐ช\left(^{2n}\right)`$ is a coefficient containing $`2n`$ gradients. Repeating the resummation procedure outlined for the case of two gradients, one can easily show that at any given order $`n`$ the renormalized solution reads
$$\mathrm{\Psi }_R(a)\frac{1}{2n}\mathrm{ln}\left(12nc_na^n\right).$$
(66)
Since at very late times $`\mathrm{\Psi }_R\frac{1}{2}\mathrm{ln}a`$, the corresponding deceleration parameter at late times goes like in Eq. (64). We conclude that at any order in the gradients, at late times the effect of the resummation of the long-wavelength perturbations is simply to generate a curvature term. This conclusion may be obtained also by inspecting Eq. (65) after the constant of integration $`\mathrm{\Psi }_{\mathrm{in}}`$ has been promoted to the renormalized quantity $`\mathrm{\Psi }_R`$. Each term in the series gives a contribution to $`R`$ which scales as $`a^3t^2`$. If only long-wavelength perturbations were present, the true scale factor $`a_D`$ would scale like
$$a_D=ae^{\frac{1}{2}\mathrm{ln}a}=a^{3/2},$$
(67)
and $`R_D`$ would scale like $`a_D^2`$. Therefore, if only long-wavelength perturbations were present, at large times and at any order in the gradients the line-element would take the form of a curvature dominated Universe, with $`h_{ij}t^2C_{ij}(๐ฑ)`$, where $`C_{ij}(๐ฑ)`$ is a function of spatial coordinates only.
In summary, super-Hubble perturbations cannot be distinguished from the background for local observers. Thus a Universe which is pure matter and has only super-Hubble perturbations, looks like a FRW universe to the local observer. Even if we started with a flat Universe plus perturbations, it is clear that the local observer will interpret what she/he sees as a FRW model with curvature (it would need a fine tuning to have k=0 within the Hubble patch). Now, as there is only matter and curvature in that model, the curvature will eventually dominate at late times, as a (open) non-flat matter Universe is dominated by curvature at late times.
### III.2 The effect of sub-Hubble modes
Dealing with the backreaction of sub-Hubble perturbations, and therefore attacking the issue of the cross-talk between the sub-Hubble modes and the homogeneous mode, is more difficult than dealing with the super-Hubble modes because, as we shall see, the gradient expansion displays an instability of the perturbative series.
In the effective Friedmann description of the inhomogeneous Universe one wishes to compute the typical value of the local observables averaged over the comoving volume $`D`$. By that we mean the ensemble average of such a volume average. The cosmological perturbations are treated as variables that take random values over different realizations of volumes $`D`$. In other words, we calculate the typical value of a quantity for a region of given size as the statistical mean over many different similar regions. This typical value is generically accompanied by a variance. If the size of the comoving volume $`D`$ is much smaller than the global inflationary volume, then we can imagine placing this volume in random locations within a region whose size is much bigger than the size of $`D`$. By the ergodic property, this is equivalent to taking random samples of the ensemble for a fixed location of the box. In other words, one can replace the expected value of a given quantity averaged over a given comoving volume $`D`$ with the ensemble average of the volume average, denoted by $`\overline{\mathrm{}}_D`$. Since we are interested in the role of sub-Hubble perturbations which cause a tiny variation of the value of the gravitational potential from one Hubble patch to another, the variance of the local mean observables is small. Under these circumstances, we can safely replace the spatial average with the ensemble average. This automatically implies that the perturbations which contribute to the effective dynamics are no longer restricted to receive contributions peaked at modes comparable to the Hubble-size (technically, this means that the window filter function defining the size of the comoving volume $`D`$ plays no role) and therefore can be much bigger than of order $`10^5`$ (or powers of it)<sup>6</sup><sup>6</sup>6It is important to stress that although the evolution of the kinematical backreaction and the mean curvature are obtained for averaged fields restricted to the domain $`D`$, the solutions to the averaged equations are actually influenced by inhomogeneities outside the domain $`D`$ too, since the initial data are to be constructed non-locally and so take the fields on the whole Cauchy hypersurface into account. We thank T. Buchert for discussions on this issue..
Let us start by considering the lowest order in a gradient expansion, i.e., keeping only two spatial derivatives. The mean local curvature will be non-vanishing, but $`Q_D`$ will be zero at this order, as $`\dot{\mathrm{\Psi }}`$ contains at least two spatial derivatives. In such a case, the integrability relation Eq. (25) immediately shows that the only consistent solution is
$$R_Da_D^2,$$
(68)
i.e., the effect of sub-Hubble perturbations at this order is to generate a standard curvature-like term in the effective Friedmann equations, scaling as the inverse square of the scale factor. This simple result holds at any order in perturbation theory (provided that one keeps only two spatial derivatives) and represents a straightforward extension of what found in Ref. chung where it was shown that to second order in spatial gradients and in the gravitational potential, cosmological perturbations amount only to a renormalization of the local spatial curvature (this result valid up to two gradients was though improperly applied to the findings of Ref. KMNR , where more than two spatial derivatives were included, for instance through the physical redshift; this point was also noticed in Ref. rasanen ). The result of Eq. (68) is reminiscent of the so-called vacuole model (see, e.g., Ref. hammer ).<sup>7</sup><sup>7</sup>7We thank S. Carroll for correspondence on this issue. Consider indeed a spherical region of a perfectly uniform Universe. Suppose that the matter inside that spherical region is squeezed into a smaller uniform spherical distribution with higher density. By mass conservation there will be a region in between the overdense sphere and the external Universe that is completely empty. Einsteinโs equation are exactly solvable for this situation in terms of the Tolman-Bondi metric. The outside FRW Universe is totally unaffected by such a rearrangement. By Birkhoffโs theorem, the empty shell will be described by the Schwarzschild metric. Finally, the interior region will behave like a homogeneous and isotropic FRW Universe, but with different values of the cosmological parameters. These parameters will exactly obey the conventional Friedmann equations, and someone who lived inside there would have no way of telling that those parameters did not describe the entire Universe.
However, the situation changes if we consistently account for higher-order derivative terms both in $`R_D`$ and in $`Q_D`$. The lowest non-zero contribution to $`Q_D`$ contains four spatial gradients and goes like $`a^2H^2a^1`$; this corresponds to a similar term with four gradients in the mean spatial curvature.
Let us further elaborate on these findings. In all generality one can write
$`Q_D`$ $`=`$ $`{\displaystyle \underset{n=2}{\overset{\mathrm{}}{}}}q_na^{n3}`$
$`R_D`$ $`=`$ $`{\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}r_na^{n3},`$ (69)
where $`q_n`$ and $`r_n`$ are expansion coefficients containing $`2n`$ spatial gradients. (Note that $`q_1=0`$, as $`Q_D`$ starts from 4 gradients.) One may wonder about the actual range of validity of the gradient-expansion technique. At first sight it might appear to be valid only to describe inhomogeneities on super-Hubble scales, i.e., for comoving wave-numbers $`kaH`$. However, this is not really the case! As one can easily check, terms of order $`n`$ in the expansion, i.e., terms with $`2n`$ gradients, contain the peculiar gravitational potential $`\phi `$ to power $`m`$ with $`2nmn`$. The dominant contribution at each order $`n`$ (i.e., with $`2n`$ gradients) is Newtonian, i.e., coming from terms of the type $`(^2\phi )^n`$. However, these terms both in $`Q_D`$ and $`R_D`$ sum up to produce negligible surface terms when averaged over a large volume, so that the leading terms become the first post-Newtonian ones, i.e., those proportional to $`(^2\phi )^{n1}(\phi )^2`$. In other words the expansion is shielded from the effect of the Newtonian terms, which could in principle be almost arbitrarily large, by the volume averaging. The same protection mechanism, however, does not apply to the non-Newtonian terms in the expansion, simply because they cannot be recast as surface terms. This simple reasoning immediately leads to the conclusion that the actual limit of validity of our expansion at order $`n`$, is set by $`(k/aH)^{2n}\phi ^{n+1}1`$. Because of the nearly-Gaussian nature of our inflationary seed $`\phi `$, it is clear that the lowest-order term able to produce a big contribution to $`Q_D`$ and $`R_D`$ appears for $`n=3`$, i.e., a term with six gradients. The importance of the six-derivative post-Newtonian terms has indeed been stressed also by Notari alessio . It is a disconnected fourth-order moment of $`\phi `$ of the type
$$(^2\phi )^2/H_0^4(\phi )^2/H_0^2,$$
(70)
having assumed that the spatial average coincides with the ensemble average. At this level an instability of the perturbative expansion is produced by the combination of the small post-Newtonian term $`(\phi )^2/H_0^2`$ (of order $`10^5`$) with the Newtonian term $`(^2\phi )^2/H_0^4`$, which can be almost arbitrarily large oldKMNR , due to the logarithmic dependence on the ultraviolet cut-off (for a scale-invariant spectrum and cold dark matter transfer function).<sup>8</sup><sup>8</sup>8Unlike the standard perturbative approach, we need not require small matter density fluctuations $`^2\phi /H_0^21`$, in our approach. This is because in the evaluation of mean observables, powers of $`^2\phi /H_0^2`$ either give rise to tiny surface terms or get multiplied by the small post-Newtonian term $`(\phi )^2/H_0^2`$.
It is important to stress that the six derivative terms give a contribution to $`Q_D`$ which scales like $`a^3H^2=`$ constant; similarly the six-gradients contribution to the smoothed curvature scales like $`a^2/a^2=`$ constant. So these terms give rise to a sizeable effective cosmological constant-like term in our local Friedmann equations. In order to estimate correctly the six-gradients terms one needs the metric coefficients $`\mathrm{\Psi }`$ and $`\chi _j^i`$ up to four gradients (whose explicit expressions are given in the Appendix).
The existence of a large contribution at six gradients, however, suggests that higher-order gradient terms will similarly lead to large corrections to the FRW background expansion rate. This is indeed the case. In the large-$`n`$ limit there will be large contributions coming from perturbations in the quasi-linear regime ($`|\delta _{FRW}|1`$). These generic conclusions, however, also tell us that stopping the expansion at six gradients would be completely arbitrary and that, in any case, the perturbative approach cries for a more refined treatment than simply counting powers of the scale factor as done in Ref. alessio . The existence of large corrections to the background should be taken strictly as evidence for an instability of the FRW background caused by nonlinear structure formation in the Universe. The actual quantitative evaluation of their effect on the expansion rate of the Universe would however require a truly non-perturbative approach, which is clearly beyond the aim of this paper.
Connected to this fact is a technical obstacle in extending the validity of the gradient expansion to late times and/or to the nonlinear regime. This comes from the fact that the metric determinant may become negative, indicating an internal inconsistency of the approximation. In Ref. 4grad the problem is solved by using an โimprovedโ approximation scheme which expresses the metric as a โsquare;โ this choice guarantees non-negativity and leads to a GR extension of the classical Zelโdovich approximation of Newtonian theory. It is then shown that with suitable choice of the initial seed, such an improved approximation provides an excellent match to an exact inhomogeneous solution of Einsteinโs field equations, the so called Szekeres metric szek , which describes locally axisymmetric (pancake) collapse of irrotational dust. Alternatively, exploiting the non-perturbative continuity equation, $`(1+\delta _{\mathrm{FRW}})^1=_{t_{\mathrm{in}}}^t๐t\theta `$, one can easily convince her/himself that the determinant of the metric is always well-defined.
In order to take one step forward in the gradient-expansion approach, we will use the same renormalization group technique previously applied to deal with the backreaction of super-Hubble modes. Let us start by dealing with the case of two gradients. Can we apply the renormalization technique to the case of sub-Hubble perturbations up to two gradients? The answer is yes, since the spatial averages of objects like $`^2\phi /H_0^2`$ and $`\left(\phi \right)^2/H_0^2`$ can be replaced by the corresponding ensemble averages and therefore are tiny (of the order of $`10^5`$). The renormalized growing solutions at two gradients reads therefore<sup>9</sup><sup>9</sup>9The long-wavelength part of the factor $`e^{5\phi (๐ฑ)/3}`$, associated with each spatial gradient, can be re-absorbed by a redefinition of the spatial coordinates, as noticed in Ref. seljak , and does not play any role when evaluating ensemble averages as well as in the backreaction problem (we thank M. Porrati for correspondence on this issue); the small-wavelength part, on the other hand, can be expanded as $`(1+5\phi _s/3)1`$ because $`\phi _s10^5`$. Nonetheless, we prefer to show them explicitly because they provide the initial condition $`C_{\mathrm{in}}`$ for the renormalization approach.
$`\mathrm{\Psi }`$ $`=`$ $`{\displaystyle \frac{5}{3}}\phi {\displaystyle \frac{1}{2}}\mathrm{ln}\left[1{\displaystyle \frac{1}{9}}\left({\displaystyle \frac{a}{a_0}}\right)\left({\displaystyle \frac{2}{H_0}}\right)^2e^{10\phi /3}\left(^2\phi {\displaystyle \frac{5}{6}}\left(\phi \right)^2\right)\right],`$
$`\chi _j^i`$ $`=`$ $`{\displaystyle \frac{1}{2}}\mathrm{ln}\left\{1{\displaystyle \frac{2}{3}}\left({\displaystyle \frac{a}{a_0}}\right)\left({\displaystyle \frac{2}{H_0}}\right)^2e^{10\phi /3}\left[D_j^i\phi +{\displaystyle \frac{5}{3}}\left(\phi ^{,i}\phi _{,j}{\displaystyle \frac{1}{3}}\left(\phi \right)^2\delta _j^i\right)\right]\right\}.`$ (71)
The next step consists in solving for the cosmological perturbations at four gradients. The equations of motion for the gravitational potential $`\mathrm{\Psi }`$ and for $`\chi _j^i`$ at four gradients are given by Eqs. (48) and (49) where the sources in the right-hand-side are computed inserting the solutions of Eq. (III.2).
Upon defining the coefficient
$$=\frac{1}{9}e^{10\phi /3}\left(^2\phi \frac{5}{6}\left(\phi \right)^2\right),$$
(72)
the matrix
$$_j^i=\left(^i_j\phi \frac{5}{6}^i\phi _j\phi \right),$$
(73)
and the traceless matrix
$$_j^i=\frac{2}{3}e^{10\phi /3}\left[D_j^i\phi +\frac{5}{3}\left(\phi ^{,i}\phi _{,j}\frac{1}{3}\left(\phi \right)^2\delta _j^i\right)\right],$$
(74)
the growing solution at four gradients for the gravitational potential assumes the form
$`\mathrm{\Psi }`$ $``$ $`{\displaystyle \frac{1}{24}}\mathrm{Tr}\left[\mathrm{ln}\left(1\left({\displaystyle \frac{a}{a_0}}\right)\left({\displaystyle \frac{2}{H_0}}\right)^2\right)\mathrm{ln}\left(1\left({\displaystyle \frac{a}{a_0}}\right)\left({\displaystyle \frac{2}{H_0}}\right)^2\right)\right]`$ (75)
$`+`$ $`{\displaystyle \frac{5}{36}}\mathrm{Tr}\left[\mathrm{ln}\left(1\left({\displaystyle \frac{a}{a_0}}\right)\left({\displaystyle \frac{2}{H_0}}\right)^2\right)\mathrm{ln}\left(1\left({\displaystyle \frac{a}{a_0}}\right)\left({\displaystyle \frac{2}{H_0}}\right)^2\right)\right].`$
A similar solution can be obtained for $`\chi _j^i`$ at four gradients. At late times the solution for the gravitational potential grows like $`(\mathrm{ln}a)^2`$. A renormalization procedure can be applied to the solutions at four gradients because there are still โsmallโ perturbative terms at hand, for instance terms like $`\left(\phi \right)^4`$ whose spatial average is small. This amounts to saying that one has to take the solution of Eq. (75), expand the arguments of the logarithms and apply the renormalization procedure described previously at second order in the โperturbative parameterโ $``$. The renormalized solution for the gravitational potential will grow like $`(\mathrm{ln}a)`$ at large times.
The lesson to learn from this computation is that, if we proceed further and go to six gradients, the unrenormalized gravitational potential, as well as $`\chi _j^i`$, will grow like $`(\mathrm{ln}a)^3`$. This is surely a step forward compared to the simple counting of powers of the scale factor which predicts that, at six gradients, the gravitational potential should grow like $`a^3`$. However, at this stage the renormalization procedure fails because it involves terms like the one in Eq. (70), which may be easily of order unity. Even the resummed perturbative expansion shows an instability produced by the combination of post-Newtonian and Newtonian terms; solutions with $`2n`$ gradients are expected to behave like $`(\mathrm{ln}a)^n`$. If taken at face value, such a time-behavior of the gravitational potential would lead to acceleration of our local Hubble patch. To put this indication on firmer grounds (or to disprove it), however, one should go beyond the perturbative approach adopted in this paper. Our result may spur the efforts toward the search for a nonperturbative description of the dynamics of the system which would account for combinations of large Newtonian and small post-Newtonian terms.
As a concluding remark of this subsection, we address a common objection to the use of the synchronous and comoving gauge in addressing the backreaction problem, namely that the occurrence of shell-crossing singularities (caustics) in the evolution of collisionless fluids might prevent the analysis to be carried over into the fully non-linear regime. We would like to point out that the instability we find in the gradient expansion is unrelated to shell-crossing singularities. This can be immediately appreciated by noting that: i) shell-crossing instabilities imply the emergence of divergent gradients terms, while our instability shows up through an infinite number of finite gradient terms; ii) shell crossing is well known to lead to an infinite Newtonian term, while our effect involves a tiny Newtonian term. It should also be stressed that the occurrence of caustics does not represent a serious limitation of our approach; indeed, the very fact that caustics only carry a small amount of mass implies that they can be easily smeared over a finite region out in such a way that their presence does not affect the mean expansion rate of the Universe.
For the sake of completeness, in the next subsection we will address the problem at hand within the commonly used weak-field approximation in the Poisson gauge.
### III.3 The backreaction in the weak-field approximation
So far, in evaluating the effect of backreaction we have been making use of a perturbative approach in which non-linear dynamical quantities are explicitly expressed in terms of the inflationary seed perturbation $`\phi `$ and its spatial derivatives. This is the reason why higher and higher gradients of $`\phi `$ appear in our results. The same conclusion would hold also in different gauges as well as by using different perturbative schemes.
One might however wonder whether the back-reaction problem can be approached directly in terms of non-linearly evolved variables. Related to this issue is the gauge choice. Non-perturbative approaches are indeed possible both in the comoving gauge adopted so far (see Ref. mater ), and in the more commonly used Poisson gauge bert . Working out the effects of back-reaction in a non-comoving gauge is indeed fully legitimate, provided a well-defined space-time splitting is performed, e.g., by means of the ADM approach buchert2 . We will here only sketch how back-reaction effects can be evaluated in the Poisson gauge, leaving to a subsequent paper a more detailed and quantitative analysis of the problem.
The line-element of the Poisson gauge reads (see, e.g., Ref. carbone )
$$ds^2=a^2(\tau )\left\{\left(1+2\varphi _P\right)d\tau ^22V_id\tau dx^i+\left[\left(12\mathrm{\Psi }_P\right)\delta _{ij}+h_{ij}^{(T)}\right]dx^idx^j\right\}.$$
(76)
where $`\tau `$ is the conformal time and $`a(\tau )\tau ^2`$ is the FRW background scale-factor for our irrotational dust source. It is important to stress that this line-element is meant to include perturbative terms of any order around the FRW background. The quantities $`V_i`$ are pure vectors, i.e., they are divergenceless, $`^iV_i=0`$, while $`h_{ij}^{(T)}`$ represent traceless and transverse (i.e., pure tensor) modes, $`h_i^{(T)i}=^ih_{ij}^{(T)}=0`$ (spatial indices are raised by the Kronecker symbol). Vector and tensor metric modes are, respectively, of $`๐ช(1/c^3)`$ and $`๐ช(1/c^4)`$. To leading order in powers of $`1/c`$ the above line-element is known to take the well-known weak-field form
$$ds^2=a^2(\tau )\left[(1+2\varphi _P)d\tau ^2+\left(12\psi _P\right)\delta _{ij}dx^idx^j\right],$$
(77)
where the scalars $`\varphi _P`$ and $`\psi _P`$ are both $`๐ช(1/c^2)`$ and $`\varphi _P=\psi _P=\mathrm{\Phi }_N/c^2`$; the Newtonian gravitational potential $`\mathrm{\Phi }_N`$ is related to density fluctuations $`\delta \rho `$ by the cosmological Poisson equation $`^2\mathrm{\Phi }_N=4\pi Ga^2\delta \rho `$. It is easy to realize that this form is accurate enough to describe structure formation within the Hubble radius as long as the considered wavelengths are much larger than the Schwarzschild radius of collapsing bodies peebles .
The crucial point is that the kinematical back-reaction will contain the relevant term buchert2 $`N^2\mathrm{\Theta }^2_D`$, where $`\mathrm{\Theta }=u_{;\mu }^\mu `$ ($`u^\mu `$ being the fluid four-velocity). Here $`N`$ is the inhomogeneous lapse function needed to express the Poisson-gauge coordinate time $`t_P=๐\tau a(\tau )`$ as a function of the proper time $`t`$ of comoving observers. This issue was already pointed out in Ref. oldrasanen where an approximate explicit expression for $`N`$ was given (a second-order perturbative expression can be found in Ref. oldKMNR ); a term of the type $`\left(\mathrm{\Phi }_N\right)^2`$ appears explicitly in $`N`$. What is important for us here is that $`Q_D`$ will clearly display the same type of post-Newtonian (hence non-total derivative) terms which were found in the comoving gauge using our gradient expansion, namely terms of the type
$$\left(^2\mathrm{\Phi }_v\right)^2\left(\mathrm{\Phi }_N\right)^2,$$
(78)
where $`\mathrm{\Phi }_v`$ is the velocity potential, which coincides (up to a sign) with the gravitational potential $`\mathrm{\Phi }_N`$ on linear scales; more generality $`\mathrm{\Phi }_v`$ and $`\mathrm{\Phi }_N`$ are connected by a cosmological Bernoulli equation (see, e.g., Ref. mater ). Similar terms appear in the mean curvature when projecting onto the comoving observer frame. We stress again that the terms of the type (78) appear only when considering the correct effective description of the average dynamics which has to include the kinematical backreaction term. Notice that this does not amount to saying that post-Newtonian effects are relevant in the dynamical evolution of the gravitational and velocity potentials themselves. Indeed the expression (78) requires evaluation of the generally non-linear potentials $`\mathrm{\Phi }_v`$ and $`\mathrm{\Phi }_N`$ which may be readily obtained through the use of standard $`N`$-body simulations. Owing to the non-linear (hence non-Gaussian) nature of the potentials $`\mathrm{\Phi }_v`$ and $`\mathrm{\Phi }_N`$, the average (78) contains both a disconnected term, as in our previous treatment, and a non-zero connected four-point moment which is dominated by mildly non-linear scales, of order a few Mpc.
Contrary to what happens in the synchronous gauge when the result is expressed in terms of the initial seed $`\phi `$, in the weak-field approximation the number of gradients is expected to be finite and the complexity of the problem resides in the non-perturbative evaluation of the evolved potentials $`\mathrm{\Phi }_v`$ and $`\mathrm{\Phi }_N`$.
It is interesting to note that the combination in Eq. (78) provides a contribution to $`Q_D`$ which is of the order of $`H^2`$ and, using the linear dependence, nearly constant in time.
## IV Conclusions
The most astonishing recent observational result in cosmology is the indication that our Universe is presently undergoing a phase of accelerated expansion. One possible explanation of the observations is that the Universe is homogeneously filled with a fluid with negative pressure that counteracts the attractive gravitational force of matter fields. Another possible explanation is a modification of GR on large distance scales.
In this paper we have elaborated on the alternative idea that the backreaction of cosmological perturbations may cause the cosmic acceleration KMNR ; oldKMNR ; japan ; oldrasanen ; rasanen ; alessio ; sw ; bmr ; rtolman . Following Buchert buchert1 ; buchert2 , we have provided the effective Friedmann equations describing an inhomogeneous Universe after smoothing. The effective dynamics is governed by two scalars: the so-called kinematical backreaction $`Q_D`$ and the mean spatial curvature $`R_D`$. They enter in the expression for the effective energy density and pressure in the Friedmann equations governing the mean evolution of a local domain $`D`$. For positive $`Q_D`$, acceleration in our local Hubble patch may be attained despite the fact that fluid elements cannot individually undergo accelerated expansion. Indeed, the very fact that the smoothing process does not commute with the time evolution invalidates the no-go theorem, which states that there can be no acceleration in our local Hubble patch if the Universe only contains irrotational dust.
Through the renormalization group technique, we have then shown that super-Hubble modes can be resummed at any order in perturbation theory yielding a local curvature term $`a_D^2`$ at large times. We then turned our attention to the backreaction originating from modes within our Hubble radius, studying perturbatively their time-behavior. In this case our findings indicate that an instability occurs in the perturbative expansion, which may be not taken care of by the renormalization group procedure since terms of the form $`H^2\delta _{\mathrm{FRW}}^2(v/c)^2`$ (where $`v`$ is the peculiar velocity) start appearing both in $`Q_D`$ and in the mean spatial curvature. Such terms are not as small as order $`10^5H^2`$; on the contrary the averaging procedure allows the combination of post-Newtonian and Newtonian terms to acquire values of order $`H^2`$. Since the perturbation approach breaks down, we may not predict on firm grounds that backreaction is responsible for the present-day acceleration of the Universe. However, it is intriguing that such an instability shows up only recently in the evolution of the Universe and that this picture is further supported by a very general result; as shown explicitely by Buchert et al. bks , even a tiny back-reaction term can drive the cosmological parameters on the averaging domain far away from their global values of the standard FRW model, thus modifying the global expansion history of the Universe. Other aspects of the scenario discussed in this paper, such as the dynamics of perturbations on observable scales, will be the subject of a forthcoming publication.
*
## Appendix A Fourth-order gradient-expansion approximation to the solution of Einsteinโs field equations
For completeness, we give here the explicit expression for $`\mathrm{\Psi }`$ and $`\chi _j^i`$ up to four gradients (we refer the reader to Ref. pilli for the detailed derivation of these results). We have
$`\mathrm{\Psi }`$ $`=`$ $`{\displaystyle \frac{5}{3}}\phi +{\displaystyle \frac{1}{18}}\left({\displaystyle \frac{a}{a_0}}\right)\left({\displaystyle \frac{2}{H_0}}\right)^2e^{10\phi /3}\left[^2\phi {\displaystyle \frac{5}{6}}\left(\phi \right)^2\right]+{\displaystyle \frac{1}{504}}\left({\displaystyle \frac{a}{a_0}}\right)^2\left({\displaystyle \frac{2}{H_0}}\right)^4e^{20\phi /3}`$ (79)
$`\times \left[{\displaystyle \frac{23}{9}}\left(^2\phi \right)^2{\displaystyle \frac{10}{3}}\phi ^{,ij}\phi _{,ij}{\displaystyle \frac{100}{9}}\phi _{,i}\phi _{,j}\phi ^{,ij}+{\displaystyle \frac{35}{27}}^2\phi \left(\phi \right)^2{\displaystyle \frac{1675}{324}}\left(\phi \right)^2\left(\phi \right)^2\right]`$
$`\chi _j^i`$ $`=`$ $`{\displaystyle \frac{1}{3}}\left({\displaystyle \frac{a}{a_0}}\right)\left({\displaystyle \frac{2}{H_0}}\right)^2e^{10\phi /3}\left[D_j^i\phi +{\displaystyle \frac{5}{3}}\left(\phi ^{,i}\phi _{,j}{\displaystyle \frac{1}{3}}\left(\phi \right)^2\delta _j^i\right)\right]`$ (80)
$`+{\displaystyle \frac{1}{504}}\left({\displaystyle \frac{a}{a_0}}\right)^2\left({\displaystyle \frac{2}{H_0}}\right)^4e^{20\phi /3}\{38(\phi ^{,ki}\phi _{,kj}{\displaystyle \frac{1}{3}}\phi _{,kl}\phi ^{,kl}\delta _j^i)`$
$`{\displaystyle \frac{128}{3}}\left[(^2\phi )\phi _{,j}^{,i}{\displaystyle \frac{1}{3}}(^2\phi )^2\delta _j^i\right]+{\displaystyle \frac{890}{27}}(^2\phi )(\phi )^2\delta _j^i{\displaystyle \frac{250}{9}}(\phi )^2\phi _{,j}^{,i}`$
$`{\displaystyle \frac{640}{9}}(^2\phi )\phi ^{,i}\phi _{,j}{\displaystyle \frac{380}{9}}\phi _{,k}\phi _{,l}\phi ^{,kl}\delta _j^i+{\displaystyle \frac{190}{3}}\left(\phi ^{,ki}\phi _{,k}\phi _{,j}+\phi ^{,i}\phi _{,kj}\phi ^{,k}\right)`$
$`+{\displaystyle \frac{1600}{27}}(\phi )^2(\phi ^{,i}\phi _{,j}{\displaystyle \frac{1}{3}}(\phi )^2\delta _j^i)\}.`$
One can verify that these solutions satisfy the energy constraint and the momentum constraint up to the relevant number of gradients. It is also important to stress that these expressions reproduce the perturbative second-order metric (see, e.g., Ref. oldKMNR ) when only terms up to second order in $`\phi `$ are kept.
###### Acknowledgements.
It is a pleasure to thank M. Bruni, T. Buchert, S. Carroll, D. Chung, D. Eisenstein, G. Ellis, G. Gelmini, A. Guth, J. Maldacena, Y. Nambu, P. Naselsky, A. Notari, J. Peebles, A. Pillepich, L. Pilo, M. Porrati, S. Rรคsรคnen, V. Sahni, R. Scherrer, U. Seljak, N. Straumann, G. Veneziano and L. Verde, for discussions during the various stages of this work.
|
warning/0506/cs0506058.html
|
ar5iv
|
text
|
# An MSE Based Transfer Chart to Analyze Iterative Decoding Schemes
## I Introduction
An Extrinsic Information Transfer (EXIT) chart is an insightful and extremely useful tool to analyze iterative decoding schemes. In an EXIT chart, the mutual information transfer characteristics of the component decoders is plotted to study the convergence behavior graphically.
Consider a serial concatenation of convolutional codes shown in Fig. 1. For this case, EXIT charts have the following two properties. One, the EXIT curve of the inner code should lie above the EXIT curve (after reflecting about the line $`y=x`$) of the outer code for the iterations to converge to the correct codeword. Two, the area under the EXIT chart is related to the rate of the code. If the a priori information is assumed to be from an erasure channel, Ashikhmin et al showed that for any code of rate $`R`$, the area under the exit curve is $`1R`$. Based on these properties it is easy to see that an optimum code can be designed by matching the EXIT charts. Recently, this technique has been used to design codes that work well with iterative decoding/signal processing . An EXIT chart is usually plotted assuming that the a priori LLRs have a Gaussian distribution. But so far the area property has been proved only for the erasure case. Therefore designing codes by matching EXIT charts for Gaussian a priori LLRs does not have a theoretical justification, although it appears to work well in several cases.
In this paper, we define a new measure based on the mean squared error (MSE) instead of mutual information, and describe an MSE chart similar to an EXIT chart. For this new measure, when the a priori information is from an AWGN channel, we theoretically prove an area property that is similar in flavor to the area property of EXIT charts in erasure channels. We then use this result to prove that matching of the MSE transfer curves of the component decoders is optimal when both the a priori and extrinsic LLRs are Gaussian. This result is then extended to prove that EXIT chart matching is also optimal. The proof is based on the recent result of Guo, Shamai and Verdu that relates the information rate to MMSE and it shows the utility of Guo et alโs fundamental result.
We use the area properties derived for the MSE chart to show that for an AWGN channel, the EXIT chart of a capacity achieving code is flat. This has recently been proved by Peleg et al in . However, the proof in this paper is slightly different from theirs.
In , several different measures used to analyze iterative decoding were studied and it was concluded that some measures were robust to different channels. However, in order to compute these measures knowledge of the transmitted bits was required and, hence, could not be done at the receiver. We show that the measure proposed here is robust and can be computed without knowledge of the transmitted bits.
The paper is organized as follows. In section II we present the notation used in this paper. In section III we outline some existing measures and propose a new measure. In section IV we show the area property. We prove the optimality of matching for Gaussian LLRs in section V. In section VI we prove that the EXIT chart of capacity achieving codes is flat. We summarize our results in section VII.
## II Notation
We use $`\stackrel{}{X}`$ to represent a vector and $`X_1,\mathrm{},X_n`$ to denote its elements. We denote a set containing elements $`X_i,\mathrm{},X_j`$ by $`X_i^j`$. We use $`|\stackrel{}{X}|`$ to denote $`(X_i^2)^{0.5}`$. We use $`\varphi (\stackrel{}{X}|\stackrel{}{Y})`$ to denote the average minimum mean squared error in estimating the elements of $`\stackrel{}{X}`$ given $`\stackrel{}{Y}`$, that is $`\varphi (\stackrel{}{X}|\stackrel{}{Y})=\frac{1}{n}E_{\stackrel{}{X},\stackrel{}{Y}}\left[\right|\stackrel{}{X}E_{\stackrel{}{X}|\stackrel{}{Y}}[\stackrel{}{X}|\stackrel{}{Y}]|^2]`$. We drop the subscript in the expectation operator $`E[]`$ whenever it is unambiguous.
For the AWGN channel $`Y=\sqrt{\gamma }X+N`$, with $`X\{+1,1\}`$ with $`P(X=1)=p`$, and $`N`$ is a Gaussian random variable with zero mean and unit variance, we use $`I_2(\gamma ,p)`$ to denote the mutual information between $`X`$ and $`Y`$ and $`\varphi (\gamma ,p)`$ to denote the minimum mean squared error in estimating $`X`$ from $`Y`$. When $`p=0.5`$ we represent mutual information and MMSE by just $`I_2(\gamma )`$ and $`\varphi (\gamma )`$ respectively.
If $`Y`$ is the output of an AWGN channel with snr $`\gamma `$ and input $`X`$, to highlight that $`\varphi (X|Y)`$ is a function of $`\gamma `$ we write it as $`\varphi (X|Y,\gamma )`$. We do not encounter cases where the snr is unknown in this paper.
We will use $`\lambda `$ and $`\rho `$ to represent the edge perspective degree profile of the variable nodes and the check nodes in an LDPC code, where $`n`$ is the total number of edges, $`n\lambda _i`$ is the number of edges connected to degree $`i`$ bit nodes and $`n\rho _i`$ is the number of edges connected to degree $`i`$ check nodes.
## III Measures
Consider the serial concatenation scheme and the corresponding iterative decoder shown in Fig. 1. Let $`L(x_k)`$, $`L_{ap}(x_k)`$ and $`L_{ext}(x_k)`$ be the $`\mathrm{log}`$ likelihood ratio (LLR), a priori LLR, and extrinsic LLR on bit $`x_k`$. Further, let us assume that the two component decoders produce true a posteriori estimates $`L(x_k)`$ based on $`L_{ap}`$ and any other observation from the channel. It has been observed that the pdf of $`L(x_k)`$ can be assumed to be Gaussian with mean $`mx_k`$ and variance $`2m`$, denoted by $`๐ฉ(mx_k,2m)`$. Based on this assumption, we plot a curve for each of the decoder blocks. We assume an a priori LLR $`๐ฉ(mx_k,2m)`$ and generate extrinsic LLR for the inner decoder. We extract some parameter from these LLRs, $`F(L)`$, and plot $`F(L_{ap})`$ against $`F(L_{ext})`$. For the outer decoder again we do a similar computation but plot $`F(L_{ext})`$ against $`F(L_{ap})`$. This is illustrated in Fig. 2 for $`F(L)=I(X;L)`$, in which case such chart is called an EXIT chart. The path taken by the iterations is also shown in the curve. It is clear from the chart that the iterations will converge to the correct codeword if the curves do not cross each other.
We get different charts depending on the parameter that is extracted. Some of the measures that have been considered previously are
Mutual Information measure used in EXIT charts defined in is given by
$$F(L)=I(X;L)$$
(1)
Fidelity measure was defined in as
$$F(L)=\theta =E[x_k\mathrm{tanh}(L(x_k)/2)]$$
(2)
In a measure $`\eta `$ was defined as
$$F(L)=\eta =E[L^2(x_k)]$$
(3)
In , it was shown that measures M1 and M2 are robust and predict the performance of iterative decoding well. Measure M3 was proposed as a measure that could be computed without knowing $`x_k`$โs and, hence, could be used at the receiver. However, in , it is shown that this measure is not robust.
### III-A Proposed Measure
We propose a new measure $`\varphi `$
$$F(L)=1\varphi =E[\mathrm{tanh}^2(L(x_k)/2)]$$
(4)
Any APP decoder computes $`L_{ext}(x_k)`$ from some channel observations $`Y`$ and the a prior information on bits $`x_1^{k1}`$ and $`x_{k+1}^n`$. When the APP decoder is a true APP decoder
$$L_{ext}(x_k)=\mathrm{log}\left(\frac{P(X_k=1|Y,L_{ap}(x_1^{k1},x_{k+1}^n))}{P(X_k=1|Y,L_{ap}(x_1^{k1},x_{k+1}^n))}\right)$$
(5)
The MMSE estimate of $`x_k`$ given $`Y`$ and $`L_{ap}(x_1^{k1},x_{k+1}^n)`$ is given by
$`\widehat{x}_k`$ $`=`$ $`P(X_k=1|Y,L_{ap}(x_1^{k1},x_{k+1}^n))P(X_k=1|Y,L_{ap}(x_1^{k1},x_{k+1}^n))`$ (6)
$`=`$ $`{\displaystyle \frac{e^{L_{ext}(x_k)}}{1+e^{L_{ext}(x_k)}}}{\displaystyle \frac{1}{1+e^{L_{ext}(x_k)}}}=\mathrm{tanh}(L_{ext}(x_k)/2)`$ (7)
The MMSE is given by
$$E[(x_k\widehat{x}_k)^2]=1E[x_k.\widehat{x}_k]=1E\left[(\widehat{x}_k)^2\right]$$
(8)
Therefore we have
$$MMSE=\varphi =1\theta $$
(9)
From the definition (4) it can be seen that M4 can be computed without knowledge of $`x_k`$. Since M4 is equal to M2 when the component decoders are true APP decoders it is robust as well. Let us denote the transfer chart obtained using measure M4 as an MSE chart.
It is easy to see that when both the a priori and extrinsic information are from erasure channels, the MSE chart and the EXIT charts become identical. Therefore the area properties derived for the EXIT charts in the erasure case also apply to the MSE chart. In the next section we derive some area properties for the MSE chart in the Gaussian case.
## IV Area Property
In this section we derive some relationships between the rate and the MSE curve of the inner and outer code of the serial concatenation scheme shown in Fig. 1. The motivation for the relationships presented here is the following result by Guo et al that connects MMSE and mutual information.
For a Gaussian channel $`Y=\sqrt{snr}X+N;`$where $`N๐ฉ(0,1)`$, if $`\widehat{X}`$ is the MMSE estimate of $`X`$ given $`Y`$ then
$$\frac{d}{dsnr}I(X;Y)=\frac{\mathrm{log}_2e}{2}E[(X\widehat{X})^2]$$
(10)
Using this result when X is binary we get
$$\frac{d}{d\gamma }I_2(\gamma ,p)=\frac{1}{\mathrm{ln}4}\varphi (\gamma ,p)$$
(11)
###### Theorem 1
Consider a system where $`\stackrel{}{X}`$ is chosen from a code $`๐`$ and transmitted over a Gaussian channel with signal to noise ratio $`\gamma `$. Let $`\stackrel{}{Y}`$ denote the output of the Gaussian channel. Let $`\stackrel{}{Z}`$ represent side information available about $`\stackrel{}{X}`$. For this system we have
$`{\displaystyle _0^{\mathrm{}}}\varphi (\stackrel{}{X}|\stackrel{}{X}๐,\stackrel{}{Y},\stackrel{}{Z},\gamma )๐\gamma ={\displaystyle \frac{\mathrm{ln}4}{n}}H(\stackrel{}{X}|\stackrel{}{X}๐,\stackrel{}{Z})`$ (12)
where $`n`$ is the length of the codeword $`\stackrel{}{X}`$.
###### Proof:
This system is similar to the general additive noise channel model shown in Fig. 3. We have
$$I(\stackrel{}{X};\stackrel{}{Y},\stackrel{}{Z}|\stackrel{}{X}๐)=I(\stackrel{}{X};\stackrel{}{Z}|\stackrel{}{X}๐)+I(\stackrel{}{X};\stackrel{}{Y}|\stackrel{}{Z},\stackrel{}{X}๐)$$
(13)
Differentiating both sides with respect to $`\gamma `$ and noting that $`I(\stackrel{}{X};\stackrel{}{Z}|\stackrel{}{X}๐)`$ is independent of $`\gamma `$ we have
$$\frac{d}{d\gamma }I(\stackrel{}{X};\stackrel{}{Y},\stackrel{}{Z}|\stackrel{}{X}๐)=\frac{d}{d\gamma }I(\stackrel{}{X};\stackrel{}{Y}|\stackrel{}{X}๐,\stackrel{}{Z})$$
(14)
Given $`\stackrel{}{Z}`$ the channel between $`\stackrel{}{X}`$ and $`\stackrel{}{Y}`$ is Gaussian. By using the relationship derived by Guo et al we have
$`{\displaystyle \frac{d}{d\gamma }}I(\stackrel{}{X};\stackrel{}{Y},\stackrel{}{Z}|\stackrel{}{X}๐)`$ $`=`$ $`{\displaystyle \frac{E[|\stackrel{}{X}E[\stackrel{}{X}|\stackrel{}{X}๐,\stackrel{}{Y},\stackrel{}{Z},\gamma ]|^2]}{\mathrm{ln}4}}`$ (15)
$`=`$ $`{\displaystyle \frac{n}{\mathrm{ln}4}}\varphi (\stackrel{}{X}|\stackrel{}{X}๐,\stackrel{}{Y},\stackrel{}{Z},\gamma )`$
Now integrating both sides with respect to $`\gamma `$ we have
$`{\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{d}{d\gamma }}I(\stackrel{}{X};\stackrel{}{Y},\stackrel{}{Z}|\stackrel{}{X}๐)๐\gamma `$ (16)
$`=I(\stackrel{}{X};\stackrel{}{Y},\stackrel{}{Z}|\stackrel{}{X}๐)|_{\gamma =\mathrm{}}I(\stackrel{}{X};\stackrel{}{Y},\stackrel{}{Z}|\stackrel{}{X}๐)|_{\gamma =0}`$
$`=H(\stackrel{}{X}|\stackrel{}{X}๐)I(\stackrel{}{X};\stackrel{}{Z}|\stackrel{}{X}๐)`$
$`=H(\stackrel{}{X}|\stackrel{}{Z},\stackrel{}{X}๐)`$
$`={\displaystyle \frac{n}{\mathrm{ln}4}}{\displaystyle _0^{\mathrm{}}}\varphi (\stackrel{}{X}|\stackrel{}{X}๐,\stackrel{}{Y},\stackrel{}{Z},\gamma )๐\gamma `$
Note that in a typical concatenation scheme, $`\stackrel{}{X}`$ is the input to the inner encoder. However, here we use the term inner code to refer to a set of constraints satisfied by $`\stackrel{}{X}`$. This difference will me made clear in example 2.
###### Corollary 1
For any code $`๐`$ of rate $`R`$
$$_0^{\mathrm{}}\varphi (\stackrel{}{X}|\stackrel{}{X}๐,\stackrel{}{Y},\gamma )๐\gamma =R\mathrm{ln}4$$
(17)
where $`\stackrel{}{X}`$ represents a length $`n`$ codeword, $`\stackrel{}{Y}`$ represents the received signal when $`\stackrel{}{X}`$ is transmitted over an AWGN channel with signal to noise ratio $`\gamma `$ and $`\varphi (\stackrel{}{X}|\stackrel{}{X}๐,\stackrel{}{Y},\gamma )`$ is the MMSE is estimating $`\stackrel{}{X}`$ given that $`\stackrel{}{Y}`$ is the received signal when a codeword was transmitted.
###### Proof:
Follows from Theorem 1 when there is no side information as $`H(\stackrel{}{X}|\stackrel{}{X}๐)=R`$. โ
To plot the transfer characteristic of a component code, it is assumed that the a priori information is from a Gaussian channel. For a true APP decoder, $`\mathrm{tanh}(L(X_k))=\mathrm{tanh}(L_{ap}(X_k)+L_{ext}(X_k))=E[X_k|\stackrel{}{X}๐,\stackrel{}{Y},\gamma ]`$. For the outer code in a concatenation scheme, the $`\gamma `$ in (16), corresponds to the SNR of the a priori channel. Hence, if we plot the MMSE at the output, $`\varphi (\stackrel{}{X}|\stackrel{}{X}๐,\stackrel{}{Y},\gamma )=1\mathrm{tanh}^2(L(X_k))`$, as a function of the a priori snr then the area under the curve is equal to the rate of the code times $`\mathrm{ln}4`$.
###### Example 1
In Fig. 4 we plot the MMSE as a function of SNR for different rate 1/2 codes. It can be seen that the area under the MMSE curve for the different codes is nearly the same. Numerical computations show that the area is nearly $`\mathrm{ln}2`$.
In context of iterative decoding, corollary 1 provides a nice relationship between the area under the MMSE vs SNR curve and the rate of an outer code. Theorem 1 links the area under the MMSE vs SNR curve of an inner code to an information theoretic quantity but its relation to the maximum rate supported is not clear. In the following lemma, for a special case, when the outer code is chosen independent of the inner code, we derive a relationship between the maximum outer code rate supported and the area under the MMSE vs a priori snr curve of the inner decoder. Note however, that this special case is what is typically encountered in iterative decoding.
###### Example 2
Consider the design of a good LDPC code designed for an AWGN channel with signal to noise ratio $`\gamma `$. We can treat this as a concatenated code where $`\stackrel{}{X}`$ represent the edges and $`\stackrel{}{Z}`$ the channel observations. In this case, the inner code represents the restrictions imposed on $`\stackrel{}{X}`$ by the irregular repeat code (Fig. 5). The outer code is a single parity check (SPC) code. We are interested in finding a relationship between the rate of the SPC outer code and the area under the MMSE chart for the inner irregular repeat code. In this case it can be easily seen that the rate of the outer code $`1\frac{\rho _i}{i}`$ is bounded above by $`1\frac{\lambda _i}{i}(1I_2(\gamma ))=1\frac{1}{n}H(X|Z)`$. The following Lemma generalizes this result.
###### Lemma 1
If an outer code $`๐_{\mathrm{๐จ๐ฎ๐ญ}}`$ is chosen independent of the inner code $`๐_{\mathrm{๐ข๐ง}}`$, then, the maximum rate of the outer code that can be used while achieving a vanishing probability of error is given by $`R_{out}1\frac{1}{n}H(\stackrel{}{X}|\stackrel{}{X}๐_{\mathrm{๐ข๐ง}},\stackrel{}{Z})`$ where $`\stackrel{}{Z}`$ represents the channel observation and $`n`$ represents length of $`\stackrel{}{X}`$. We will refer to this upper bound as $`R_{outer}^{max}`$.
###### Proof:
Let $`m`$ be the length of the outer codewords and let $`\stackrel{}{X}^{}`$ represent a length $`m`$ vector. Consider a sequence $`S`$ of length $`Nmn`$. We say $`S๐_{\mathrm{๐ข๐ง}}`$ if $`S(ln+1,\mathrm{},ln+n)`$ is a sequence in $`๐_{\mathrm{๐ข๐ง}}`$ for all $`l`$. Similarly we say $`S๐_{\mathrm{๐จ๐ฎ๐ญ}}`$ if $`S(lm+1,\mathrm{},lm+m)`$ is a sequence in $`๐_{\mathrm{๐จ๐ฎ๐ญ}}`$ for all $`l`$. We say that $`๐_{\mathrm{๐ข๐ง}}`$ and $`๐_{\mathrm{๐จ๐ฎ๐ญ}}`$ are chosen independently if for a random sequence $`S`$ the events $`S๐_{\mathrm{๐ข๐ง}}`$ and $`S๐_{\mathrm{๐จ๐ฎ๐ญ}}`$ are independent, i.e., $`P(S๐_{\mathrm{๐ข๐ง}}\text{ and }S๐_{\mathrm{๐จ๐ฎ๐ญ}})=P(S๐_{\mathrm{๐ข๐ง}})P(S๐_{\mathrm{๐จ๐ฎ๐ญ}})`$.
The number of length $`Nmn`$ sequences that belong to $`๐_{\mathrm{๐ข๐ง}}`$ is $`2^{NmH(\stackrel{}{X}|\stackrel{}{X}๐_{\mathrm{๐ข๐ง}})}`$. Number of length $`Nmn`$ sequences that belong to $`๐_{\mathrm{๐จ๐ฎ๐ญ}}`$ is $`2^{NnH(\stackrel{}{X}^{}|\stackrel{}{X}^{}๐_{\mathrm{๐จ๐ฎ๐ญ}})}`$. We have
$$P(S๐_{\mathrm{๐ข๐ง}}\text{ and }S๐_{\mathrm{๐จ๐ฎ๐ญ}})=\frac{2^{NmH(\stackrel{}{X}|\stackrel{}{X}๐_{\mathrm{๐ข๐ง}})}}{2^{nmN}}\frac{2^{NnH(\stackrel{}{X}^{}|\stackrel{}{X}^{}๐_{\mathrm{๐จ๐ฎ๐ญ}})}}{2^{nmN}}$$
(18)
and the number of sequences that belong to both $`๐_{\mathrm{๐ข๐ง}}`$ and $`๐_{\mathrm{๐จ๐ฎ๐ญ}}`$ is
$$2^{nmN}\frac{2^{NmH(\stackrel{}{X}|\stackrel{}{X}๐_{\mathrm{๐ข๐ง}})}}{2^{nmN}}\frac{2^{NnH(\stackrel{}{X}^{}|\stackrel{}{X}^{}๐_{\mathrm{๐จ๐ฎ๐ญ}})}}{2^{nmN}}=2^{NmH(\stackrel{}{X}|\stackrel{}{X}๐_{\mathrm{๐ข๐ง}})+NnH(\stackrel{}{X}^{}|\stackrel{}{X}^{}๐_{\mathrm{๐จ๐ฎ๐ญ}})Nmn}$$
(19)
If with some choice of outer code, the decoder is always able to recover $`S`$ from the channel observations, then the total number of sequences $`S`$ should be less than $`2^{NmI(\stackrel{}{X};\stackrel{}{Z}|\stackrel{}{X}๐_{\mathrm{๐ข๐ง}})}`$. Therefore we have
$$NmH(\stackrel{}{X}|\stackrel{}{X}๐_{\mathrm{๐ข๐ง}})+NnH(\stackrel{}{X}^{}|\stackrel{}{X}^{}๐_{\mathrm{๐จ๐ฎ๐ญ}})NmnNmI(\stackrel{}{X};\stackrel{}{Z}|\stackrel{}{X}๐_{\mathrm{๐ข๐ง}})$$
(20)
which implies
$$\frac{1}{m}H(\stackrel{}{X}^{}|\stackrel{}{X}^{}๐_{\mathrm{๐จ๐ฎ๐ญ}})1\frac{1}{n}H(\stackrel{}{X}|\stackrel{}{X}๐_{\mathrm{๐ข๐ง}},Z)$$
(21)
In the general case (when the outer code is not independent of the inner code), there seems to be no such relationship. For example, consider the LDPC code in Example 2 but consider another outer code constructed from a good rate $`R`$ code ($`R<I_2(\gamma )`$) by repeating $`c_j`$, the $`j`$th coded bit, $`d_j`$ times, where $`d_j`$ is the degree of the $`j`$th bit node. In this case the rate of the outer code is $`\frac{\lambda _i}{i}R`$. Its relationship to $`H(X|Z)`$ is not straightforward.
The inner decoder has side information about the coded bits from the channel output apart from the a priori information. The transfer characteristics is obtained by increasing the snr of the a priori channel from $`0`$ to $`\mathrm{}`$. The outer code and inner code are usually separated by a random interleaver which makes the inner code and outer code independent. Therefore from Theorem 1 and Lemma 1 it follows that for an inner decoder, the area under the plot of MMSE at the output against snr of the a priori channel is equal to $`\mathrm{ln}4(1R)`$, where $`R`$ is the maximum rate of outer code supported by the inner code. This can be easily verified for the following examples.
###### Example 3
Consider an uncoded AWGN channel with signal to noise ratio $`SNR`$ as an inner code. Let $`X`$ be the transmitted bit and let $`Z`$ be the received signal. Let $`Y`$ be the output when $`X`$ is sent over another AWGN channel with snr $`\gamma `$. Clearly, MMSE in estimating $`X`$ from $`Y`$ and $`Z`$ is same as MMSE in estimating X from the output of an AWGN channel with an snr of $`\gamma +SNR`$. We have
$`{\displaystyle _0^{\mathrm{}}}\varphi (X|Y,Z,\gamma )๐\gamma `$ $`=`$ $`{\displaystyle _0^{\mathrm{}}}\varphi (\gamma +SNR,p)๐\gamma `$ (22)
$`=`$ $`{\displaystyle _{SNR}^{\mathrm{}}}\varphi (\gamma ,p)๐\gamma `$
$`=`$ $`\mathrm{ln}4(H(p)I_2(SNR,p))`$ (23)
(23) follows from (11). When $`p=0.5`$ we get $`\mathrm{ln}4(1I_2(SNR))`$. $`R_{outer}^{max}=I_2(SNR)`$ in this case.
###### Example 4
Consider an uncoded erasure channel with erasure probability $`ฯต`$ as an inner code. Let equiprobable bits $`X`$ be the transmitted bits and let $`Z`$ be the received signal. Let $`Y`$ be the output when $`X`$ is sent over an AWGN channel with snr $`\gamma `$. We have $`\varphi (X|Y,Z,\gamma )=(1ฯต)0+ฯต\varphi (X|Y,\gamma )`$. Therefore
$`{\displaystyle _0^{\mathrm{}}}\varphi (X|Y,Z,\gamma )๐\gamma `$ $`=`$ $`ฯต{\displaystyle _0^{\mathrm{}}}\varphi (\gamma )๐\gamma `$
$`=`$ $`ฯต\mathrm{ln}4\text{ From (}\text{11}\text{)}`$
Hence $`R_{outer}^{max}=1ฯต`$ which is exactly the capacity of this channel.
###### Example 5
Consider an inner code corresponding to an LDPC code over an AWGN channel. Let $`\stackrel{}{X}`$ represent the edges, $`\stackrel{}{Y}`$ the a priori messages and let $`\stackrel{}{Z}`$ represent the channel information at the bit nodes. The MMSE for an edge connected to a bit node of degree $`i`$ is $`\varphi (i\gamma +SNR)`$. Let $`\{\lambda _i\}`$ and $`\{\rho _i\}`$ represent the degree profile of the LDPC code in edge perspective. We have
$`{\displaystyle _0^{\mathrm{}}}\varphi (\stackrel{}{X}|\stackrel{}{X}๐,\stackrel{}{Y},\stackrel{}{Z},\gamma )๐\gamma `$ $`=`$ $`{\displaystyle \underset{i=1}{\overset{N_v}{}}}\lambda _i{\displaystyle _0^{\mathrm{}}}\varphi (i\gamma +SNR)๐\gamma `$
$`=`$ $`\mathrm{ln}4{\displaystyle \underset{i=1}{\overset{N_v}{}}}{\displaystyle \frac{\lambda _i}{i}}(1I_2(SNR))`$
For an LDPC code that works well at $`SNR`$, we have
$`1{\displaystyle \frac{\frac{\rho _i}{i}}{\frac{\lambda _i}{i}}}`$ $``$ $`I_2(SNR)`$
$`R_{outer}=1{\displaystyle \frac{\rho _i}{i}}`$ $``$ $`1{\displaystyle \frac{\lambda _i}{i}(1I_2(SNR))}`$
It is interesting to compare the area property derived here with that derived by Ashikhmin et al in . It was shown that the area under the EXIT curve, when both the a priori and extrinsic information can be modelled to be from erasure channels, is given by
$$\text{Area}=\left(\frac{1}{n}\underset{i=1}{\overset{n}{}}H(X_i|\stackrel{}{X}๐)\right)^2\left[1\frac{H(\stackrel{}{X}|\stackrel{}{Z},\stackrel{}{X}๐)}{_{i=1}^nH(X_i|\stackrel{}{X}๐)}\right]$$
(24)
(24) was obtained by modifying equations (22), (23) in to suit the notation used in this paper.
In the special case when $`H(X_i)=1`$ the area becomes $`1\frac{1}{n}H(\stackrel{}{X}|\stackrel{}{Z},\stackrel{}{X}๐)`$. For an outer code of rate $`R`$ the area is therefore is $`1R`$. For some specific inner codes $`C_{in}`$, it was shown that $`1\frac{1}{n}H(\stackrel{}{X}|\stackrel{}{Z},\stackrel{}{X}๐_{\mathrm{๐ข๐ง}})`$ is the maximum rate of the outer code that can be used in iterative decoding to achieve error free communication.
In this paper, we have proved that $`1\frac{1}{n}H(\stackrel{}{X}|\stackrel{}{Z},\stackrel{}{X}๐_{\mathrm{๐ข๐ง}})`$ is indeed the maximum rate of outer code that can be used for reliable communication when the outer code and inner code are independently chosen. This makes the area property derived for EXIT charts more concrete.
We note that in the case when $`H(X_i|\stackrel{}{X}๐)1`$, the simple relationship between the area under the EXIT chart and rate does not hold. However the relationship between area and rate of the outer code and the relationship between area and $`R_{outer}^{max}`$ for the inner code continue to hold for the MMSE vs SNR plot.
### IV-A Area Property for MSE chart
Let us assume that the bits about which information is exchanged in an iterative decoding scheme (usually the coded bits of the outer code) are equiprobable. Further, let the a priori and the extrinsic information can be modelled as though the bits were transmitted over an AWGN channel. Let us refer to the SNRs of these channels as $`snr_{ap}`$ and $`snr_{ext}`$. We first note that if a true APP decoder is employed, MSE is equal to the MMSE. We denote the MMSE corresponding to the a priori, the extrinsic, and the output LLR by $`MMSE_{ap}`$, $`MMSE_{ext}`$ and $`MMSE_{out}`$ respectively.
We will refer to a plot of $`MSE_{ext}`$ versus $`MSE_{ap}`$ as an MSE transfer curve. An MSE chart then has two MSE transfer curves, one for the inner decoder and one for the outer decoder. The area properties proved so far are for a plot of the $`MMSE_{out}`$ (not $`MMSE_{ext}`$) versus the $`snr_{ap}`$. With the Gaussian assumption, the $`MMSE_{out}`$ vs $`snr_{ap}`$ plot can be generated from the MSE transfer curve using the transformation $`(1MMSE_{ap},1MMSE_{ext})(\varphi (\varphi ^1(MMSE_{ap})+\varphi ^1(MMSE_{ext})),\varphi ^1(MMSE_{ap}))`$. The area properties derived thus apply for the MSE transfer curve under this transformation. For convenience, we use $`\gamma _{ap}`$ and $`\gamma _{ext}`$ to denote $`\varphi ^1(MMSE_{ap})`$ and $`\varphi ^1(MMSE_{ext})`$, respectively. We will use subscripts $`inner`$ and $`outer`$ to refer to quantities corresponding to the inner and outer decoders.
###### Lemma 2
If the a a priori and extrinsic information can be represented as information from Gaussian channels then for a rate $`R`$ code we have $`_0^{\mathrm{}}\varphi (\gamma _{ap}+\gamma _{ext})๐\gamma _{ext}=(1R)\mathrm{ln}4`$
###### Proof:
Consider the transfer curve as a continuous curve from $`(0,0)`$ to $`(1,1)`$ by connecting any discontinuity in $`\varphi (\gamma _{ext})`$ by vertical lines. With every point $`(x,y)=(1\varphi (\gamma _{ap}),1\varphi (\gamma _{ext}))`$ on the transfer curve associate a variable $`z=x^2+y^2`$. The reason for introducing this variable is to make it easy to handle the possibility discontinuity of the MSE transfer curve. It is easy to see that $`\gamma _{ap}`$ and $`\gamma _{ext}`$ are both continuous and increasing functions of $`z`$ such that $`\gamma _{ap}`$ and $`\gamma _{ext}`$ are $`0`$ at $`z=0`$ and $`\mathrm{}`$ at $`z=2`$.
$`\mathrm{ln}4`$ $`=`$ $`{\displaystyle _{z=0}^2}\varphi (\gamma _{ap}+\gamma _{ext})d(\gamma _{ap}+\gamma _{ext})`$
$`=`$ $`{\displaystyle _{z=0}^2}\varphi (\gamma _{ap}+\gamma _{ext})๐\gamma _{ap}+{\displaystyle _{z=0}^2}\varphi (\gamma _{ap}+\gamma _{ext})๐\gamma _{ext}`$
$`=`$ $`{\displaystyle _{\gamma _{ap}=0}^{\mathrm{}}}\varphi (\gamma _{ap}+\gamma _{ext})๐\gamma _{ap}+{\displaystyle _{\gamma _{ext}=0}^{\mathrm{}}}\varphi (\gamma _{ap}+\gamma _{ext})๐\gamma _{ext}`$
$`=`$ $`R\mathrm{ln}4+{\displaystyle _0^{\mathrm{}}}\varphi (\gamma _{ap}+\gamma _{ext})๐\gamma _{ext}`$
###### Lemma 3
For an inner code when the a a priori and extrinsic information can be represented as information from Gaussian channels then the maximum supported outer code rate ($`R_{outer}^{max}`$) is given by $`_0^{\mathrm{}}\varphi (\gamma _{ap}+\gamma _{ext})๐\gamma _{ext}/(\mathrm{ln}4)`$
The proof is similar to the proof in the previous lemma.
###### Example 6
Consider a repetition code of rate $`1/N`$. In this case when the a priori information is from an AWGN channel of snr $`\gamma _{ap}`$ then the extrinsic information can be modelled as information from a Gaussian channel of snr $`(N1)\gamma _{ap}`$. We have
$$_0^{\mathrm{}}\varphi (\gamma _{ap}+\gamma _{ext})๐\gamma _{ext}=(N1)_0^{\mathrm{}}\varphi (\gamma _{ap}+(N1)\gamma _{ap})๐\gamma _{ap}=\frac{N1}{N}_0^{\mathrm{}}\varphi (x)๐x=\left(1\frac{1}{N}\right)\mathrm{ln}4$$
(25)
This verifies Lemma 2.
## V Optimality of Matching
In this section we prove that the MSE curve of the outer code has to be matched to the MSE curve of the inner code when the extrinsic information resembles that from an AWGN channel.
###### Lemma 4
For two codes $`C_1`$ and $`C_2`$ such that $`1MMSE_{ext}^{C_1}(\gamma _{ap})1MMSE_{ext}^{C_2}(\gamma _{ap})\gamma _{ap}`$, $`R_1R_2`$ with equality only when the two curves overlap.
###### Proof:
We have
$`MMSE_{ext}^{C_1}(\gamma _{ap})`$ $``$ $`MMSE_{ext}^{C_2}(\gamma _{ap})`$
$`\gamma _{ext}^{C_1}(\gamma _{ap})`$ $``$ $`\gamma _{ext}^{C_2}(\gamma _{ap})`$
$`\varphi (\gamma _{ap}+\gamma _{ext}^{C_1}(\gamma _{ap}))`$ $``$ $`\varphi (\gamma _{ap}+\gamma _{ext}^{C_2}(\gamma _{ap}))`$
$`R_1`$ $``$ $`R_2`$
It is easy to see that equality occurs only when the two curves overlap. โ
From Corollary 1 and Lemma 3 it follows that a code that is matched exactly to the channel has a rate equal to the rate supported by the inner code. Therefore from Lemma 4 it is easy to see that any outer code whose flipped MSE curve lies below the inner code and is not matched to the inner code has a rate lesser than that supported by the inner code.
We note that under the Gaussian assumption the MSE curve and EXIT curve are related by a one to one function. Therefore since matching is optimal for MSE chart, it is optimal for EXIT charts.
As a consequence of the results derived so far, under the Gaussian assumptions, the following properties hold.
1. With Gaussian assumption on messages from outer decoder to inner decoder.
$`{\displaystyle \frac{1}{\mathrm{ln}4}}{\displaystyle \varphi _{inner}^1๐\gamma _{ap,inner}}=1R_{outer}^{max}`$ (26)
$`{\displaystyle \frac{1}{\mathrm{ln}4}}{\displaystyle \varphi _{outer}^2๐\gamma _{ext,outer}}=1R_{outer}`$ (27)
and for the iterative decoder to converge to the correct codeword $`\varphi _{inner}^1<\varphi _{outer}^2`$. Here $`\varphi ^1`$ and $`\varphi ^2`$ are used to denote the MMSE expressed as a function of the a priori and the extrinsic snr respectively.
2. With Gaussian assumption on messages from inner decoder to outer decoder, we have
$`{\displaystyle \frac{1}{\mathrm{ln}4}}{\displaystyle \varphi _{inner}^2๐\gamma _{ext,inner}}=R_{outer}^{max}`$ (28)
$`{\displaystyle \frac{1}{\mathrm{ln}4}}{\displaystyle \varphi _{outer}^1๐\gamma _{ap,outer}}=R_{outer}`$ (29)
and for the iterative decoder to converge to the correct codeword $`\varphi _{inner}^2>\varphi _{outer}^1`$.
Depending on the distribution of the exchanged messages, one of the above mentioned properties may be used to analyze and design component codes. For example, in an LDPC code the bit to check messages closely resembles information from an AWGN channel. In this case we plot MMSE against SNR extrinsic for the inner code and against SNR a priori for the outer code. In Fig. 6 we plot these curves for a rate 0.5 LDPC code that was designed using EXIT charts for an snr of 0.5dB. The degree profile designed LDPC code is $`\rho _3=1`$, $`\lambda _2=0.254`$, $`\lambda _4=0.419`$, and, $`\lambda _{18}=0.327`$. The threshold for a bit error rate of $`10^4`$ is .55dB. The threshold predicted using these curves is 0.51dB. In Fig. 7 we plot these curve for a (3,6) LDPC code. The threshold predicted is 1.05dB and the actual threshold is around 1.1dB.
## VI Area Property of Capacity Achieving Codes over AWGN channel
The optimality of matching proved in the previous section assumed that the extrinsic information resembles information from an AWGN channel. In this section we prove the optimality of matching for the AWGN channel without making any assumption on the extrinsic information. We show that the EXIT curve of any capacity achieving code is flat and is matched to the channel. It is also seen that the area under the EXIT curve of any rate $`R`$ capacity achieving code is equal to $`1R`$.
Consider a capacity achieving binary code $`๐`$ of rate $`R=I_2(SNR)`$ being transmitted over an AWGN channel with signal to noise ratio $`\gamma `$. Since the code decodes perfectly when $`\gamma >SNR`$ the MMSE in estimating the transmitted codeword $`X`$ from the received symbols $`Y`$ is 0. Therefore from Corollary 1 we get
$`R`$ $`=`$ $`{\displaystyle \frac{1}{\mathrm{ln}4}}{\displaystyle _0^{SNR}}\varphi (\stackrel{}{X}|\stackrel{}{X}๐,\stackrel{}{Y},\gamma )๐\gamma `$
$`=`$ $`{\displaystyle \frac{1}{n\mathrm{ln}4}}{\displaystyle _0^{SNR}}{\displaystyle \underset{i=1}{\overset{n}{}}}\varphi (X_i|\stackrel{}{X}๐,\stackrel{}{Y},\gamma )d\gamma `$
$``$ $`{\displaystyle \frac{1}{n\mathrm{ln}4}}{\displaystyle _0^{SNR}}{\displaystyle \underset{i=1}{\overset{n}{}}}\varphi (X_i|Y_i,\gamma )d\gamma `$
$`=`$ $`{\displaystyle \frac{1}{n\mathrm{ln}4}}{\displaystyle _0^{SNR}}{\displaystyle \underset{i=1}{\overset{n}{}}}\varphi (\gamma ,P(X_i=1))d\gamma `$
$`=`$ $`{\displaystyle \frac{1}{n}}{\displaystyle \underset{i=1}{\overset{n}{}}}I_2(SNR,P(X_i=1))`$ (31)
$``$ $`I_2(SNR)`$ (32)
The inequality in (VI) is because MMSE error in estimating $`A`$ from both $`B`$ and $`C`$ is always less that MMSE error in estimating $`A`$ from $`B`$. It is also easy to prove that $`\varphi (A|B,C)=\varphi (A|B)`$ only when $`E[A|B,C]=E[A|B]`$. (31) follows from (11).
Since $`R=I_2(SNR)`$ the inequalities in (VI) and (31) have to be equalities. Therefore we have $`P(X_i=1)=0.5i`$. We also have
$$_0^{SNR}\varphi (X_i|\stackrel{}{X}๐,\stackrel{}{Y},\gamma )๐\gamma =_0^{SNR}\varphi (X_i|Y_i,\gamma )๐\gamma $$
(33)
Since $`\varphi (X_i|Y_i,\gamma )\varphi (X_i|\stackrel{}{X}๐,\stackrel{}{Y},\gamma )`$, from (33) it follows that $`\varphi (X_i|Y_i,\gamma )=\varphi (X_i|\stackrel{}{X}๐,\stackrel{}{Y},\gamma )`$ for almost every $`\gamma [0,SNR]`$.
Now, using the fact that MMSE in both the cases is a decreasing function of $`\gamma `$ and the fact that $`\varphi (X_i|Y_i,\gamma )`$ is continuous, it can be shown that
$$\varphi (X_i|\stackrel{}{X}๐,\stackrel{}{Y},\gamma )=\varphi (X_i|Y_i,\gamma )\gamma <SNR,i$$
(34)
It is easy to see that the MMSE estimate $`E[X_i|\stackrel{}{X}๐,\stackrel{}{Y},\gamma ]=E[X_i|Y_i,\gamma ]`$. Therefore for $`\gamma <SNR`$ we have
$`\mathrm{tanh}\left({\displaystyle \frac{L_{ap}(X_i)+L_{ext}(X_i)}{2}}\right)`$ $`=`$ $`\mathrm{tanh}\left({\displaystyle \frac{L_{ap}(X_i)}{2}}\right)`$
$`L_{ext}(X_i)`$ $`=`$ $`0`$ (35)
Therefore $`I(X;L_{ext})=0`$ when $`\gamma <SNR`$.
When $`\gamma >SNR`$, $`L_{ap}+L_{ext}=+\mathrm{}`$ when $`X=1`$ is transmitted. Since $`L_{ap}<\mathrm{}`$ we have $`L_{ext}=+\mathrm{}`$ when $`X=1`$. Similarly we have $`L_{ext}=\mathrm{}`$ when $`X=1`$. Therefore $`I(X;L_{ext})=1`$ when $`\gamma >SNR`$.
We note that in a very similar approach has been used to arrive at the same result. The proof presented here though is simpler and avoids some of the steps in .
## VII Conclusion
We proposed a new measure based on MSE for analyzing the convergence behavior of iterative decoding schemes. This measure is robust and can be computed without the knowledge of the transmitted bits. Under Gaussian assumptions, we showed a mapping from the MSE chart such that for any code the area under the map is equal to the rate. We used this to prove that curve fitting is optimum in the MSE chart and then extended it to the EXIT chart case. For the AWGN channel, without making any assumptions on the distribution of extrinsic LLRs, we showed that capacity achieving codes have an EXIT chart that is flat and matched to the channel .
|
warning/0506/hep-ph0506212.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
Magnetic fields are ubiquitous in the universe. Most galaxies, cluster of galaxies and even the Coma supercluster and radio galaxies at redshift $`z2`$ have been found to be endowed with magnetic fields (for reviews see-).
The average field strength of the interstellar magnetic field in our Galaxy has been observed to be $`34\mu `$G. Spiral galaxies in general seem to have magnetic fields with strength of the order of $`10\mu `$G. The structure of these magnetic fields is determined by a large scale component with a coherence length of the order of the size of the visible disk and a small-scale component of tangled fields. There are a few spiral galaxies with exceptionally strong magnetic fields of the order of $`50\mu `$G, which also have a very high star formation rate .
Magnetic fields associated with elliptical galaxies have field strengths comparable to those observed in spiral galaxies. However, their structure seems to be quite distinct from that found for magnetic fields in spiral galaxies. The coherence length is much smaller than the corresponding galactic scales and the structure appears to be random.
In clusters of galaxies, magnetic fields of strength of the order of upto a few $`\mu `$G are found in the intracluster medium. The cluster center regions indicate strong magnetic fields with typical field strengths of the order of $`1030\mu `$G and in exceptional cases upto $`70\mu `$G . The coherence length of the magnetic fields is of the order of the scale of the cluster galaxies.
There is also evidence for the existence of magnetic fields in structures on supercluster scale. The Coma-Abell 1367 supercluster is observed to have a magnetic field of strength $`0.20.6\mu `$G . Finally, observations indicate the existence of magnetic fields in redshift $`z2`$ radio galaxies .
There is no direct observational evidence of magnetic fields that are not associated with any collapsing or virialized structure . However, it is possible to put upper bounds on the strength of such cosmological magnetic fields from anisotropy measurements of the cosmic microwave background and from the abundances of light elements predicted by standard big bang nucleosynthesis .
To explain the widespread existence of large scale magnetic fields in the universe it is commonly assumed that a tiny magnetic seed field at the epoch of galaxy formation is amplified by a dynamo mechanism to its present strength of a few microgauss in our Galaxy -. The dynamo amplifies an initial seed magnetic field exponentially. The amplification factor depends on the growth rate for the dominant mode of the dynamo and the amount of time during which the dynamo operates. In a flat universe with no cosmological constant the initial seed magnetic field needs to have at least a field strength of the order of $`10^{20}`$ G to explain the current $`\mu `$G galactic field today , -. However, as it was pointed out in , this bound depends on the cosmological model. In a flat universe with non-vanishing cosmological constant the lower limit on the required initial magnetic field strength can be lowered significantly. For reasonable cosmological parameter the required strength of the initial seed magnetic field is of the order of $`10^{30}`$G.
There are different proposals for the origin of the magnetic seed field -. A class of proposed models involves the creation of magnetic seed fields during an inflationary stage of the very early universe . In order to produce a magnetic seed field of significant strength the conformal invariance of Maxwellโs equations has to be broken, for example, by gravitational couplings of the photon .
The conformal invariance of Maxwellโs equations in four dimensions can also be broken if an embedding into a higher dimensional space-time with time-varying extra spatial dimensions is considered. In relation with the creation of seed magnetic fields this was first investigated in . It is assumed that the $`D`$ dimensional space-time can be written as a direct product of a three dimensional space and an $`n`$ dimensional space. Vacuum space-times of this type are provided by the Kasner solutions, which in general admit two classes of solutions: either expanding three dimensions and collapsing extra dimensions or vice versa. The higher dimensional background with dynamical extra dimensions is matched to a standard four dimensional radiation dominated universe with static extra dimensions. In it was found that magnetic fields of cosmologically interesting strength can be generated only in the case of contracting three dimensions and growing extra dimensions. The novel feature of the model under consideration here is that momenta along the extra dimensions are also taken into account. The final spectrum is obtained by integrating over these internal momenta . This leads to the generation of magnetic seed fields of cosmologically interesting strength in the case of expanding three dimensions and contracting extra dimensions. Imposing bounds from observations an upper bound on the strength of the magnetic field can be found.
Models with extra dimensions arise naturally in string/ M-theory which also led to the possibility of large extra dimensions . In higher dimensional gravity the four dimensional Planck scale $`M_4`$ is no longer fundamental, instead the higher dimensional Planck scale $`M_D`$ becomes the fundamental scale. With the assumption that the $`n`$ extra dimensions have a characteristic size $`R`$, using Gaussโ law, the $`D`$-dimensional and the four-dimensional Planck masses $`M_4`$ and $`M_D`$, respectively, are related by
$`M_4^2=R^nM_D^{n+2}.`$ (1.1)
Experiments show that Newtonian gravity is valid at least down to length scales of the order of 1 mm . This implies a lower bound on the ratio $`M_D/M_4`$.
## 2 Magnetic fields from extra dimensions
The background space-time is assumed to be homogeneous and anisotropic with a line element,
$`ds^2=a^2(\eta )\left[d\eta ^2\delta _{ij}dx^idx^j\right]b^2(\eta )\delta _{AB}dy^Ady^B,`$ (2.2)
where $`i,j=1,..,3`$ and $`A,B=4,..,3+n`$, $`n1`$. $`a(\eta )`$ and $`b(\eta )`$ are the scale factor of the external, 3-dimensional space and the internal, $`n`$-dimensional space, respectively.
It is assumed that for $`\eta <\eta _1`$ both scale-factors are functions of time. At $`\eta =\eta _1`$ this is matched to a radiation dominated four dimensional flat universe with static extra dimensions, $`b(\eta )=const.`$. The solutions are given by
$`a(\eta )`$ $`=`$ $`a_1\left({\displaystyle \frac{\eta }{\eta _1}}\right)^\sigma ,b(\eta )=b_1\left({\displaystyle \frac{\eta }{\eta _1}}\right)^\lambda ,\mathrm{for}\eta <\eta _1`$ (2.3)
$`a(\eta )`$ $`=`$ $`a_1\left({\displaystyle \frac{\eta +2\eta _1}{\eta _1}}\right),b(\eta )=b_1,\mathrm{for}\eta \eta _1`$ (2.4)
In the following we set $`a_1=1=b_1`$.
For $`\eta <\eta _1`$ the solution is given by the vacuum Kasner metric, which determines the exponents $`\sigma `$ and $`\lambda `$ as functions of the number of extra dimensions $`n`$. These are related to the Kasner exponents $`\alpha _E`$ and $`\alpha _I`$, satisfying the Kasner conditions $`3\alpha _E+n\alpha _I=1`$ and $`3\alpha _E^2+n\alpha _I^2=1`$, by
$`\sigma ={\displaystyle \frac{\alpha _E}{1\alpha _E}},\lambda ={\displaystyle \frac{\alpha _I}{1\alpha _E}}.`$ (2.5)
In the case of an expanding, external space and a contracting, internal space the exponents $`\sigma `$ and $`\lambda `$ are of the form ,
$`\sigma ={\displaystyle \frac{1}{2}}\left(\sqrt{{\displaystyle \frac{3n}{n+2}}}1\right),\lambda =\sqrt{{\displaystyle \frac{3}{n(n+2)}}}.`$ (2.6)
Maxwellโs equations in $`D`$ dimensions are given by $`_{\stackrel{~}{A}}F^{\stackrel{~}{A}\stackrel{~}{B}}=0`$ with $`F_{\stackrel{~}{A}\stackrel{~}{B}}=_{[\stackrel{~}{A}}A_{\stackrel{~}{B}]}`$, $`\stackrel{~}{A},\stackrel{~}{B}=0,..,n+3`$. Here the interest is the electromagnetic field in the (3+1)-dimensional space-time. Thus it is assumed that $`A_i=A_i(x^i,y^B,\eta )`$ and $`A_B=0`$. Using the radiation gauge $`A_0=0`$, $`_iA^i=0`$, Maxwellโs equations imply
$`{\displaystyle \frac{1}{b^n}}_0\left[b^n_0A_i\right]+{\displaystyle \underset{j=1}{\overset{3}{}}}_j_jA_i+\left({\displaystyle \frac{a}{b}}\right)^2{\displaystyle \underset{B=4}{\overset{3+n}{}}}_B_BA_i=0,`$ (2.7)
where $`_0\frac{}{\eta }`$, $`_i\frac{}{x^i}`$ and $`_B\frac{}{y^B}`$.
Furthermore, the canonical field $`\mathrm{\Psi }_i=b^{\frac{n}{2}}A_i`$ is introduced and the following expansion is used
$`\mathrm{\Psi }_i(\eta ,x^i,y^A)={\displaystyle \frac{d^3kd^nq}{(2\pi )^{\frac{3+n}{2}}}\underset{\alpha }{}e_i^\alpha (๐ฅ)\left[a_{l,\alpha }\mathrm{\Psi }_l(\eta )e^{i๐ฅ๐}+a_{l,\alpha }^{}\mathrm{\Psi }_l^{}(\eta )e^{i๐ฅ๐}\right]},`$ (2.8)
where $`l^\mu `$ is a $`(3+n)`$vector with components $`l^ik^i`$, $`l^Aq^A`$. Moreover, $`๐ฅ๐=๐ค๐ฑ+๐ช๐ฒ`$. $`\alpha `$ runs over the polarizations. In the background (2.3), $`\eta <\eta _1`$, this results in the mode equation
$`\mathrm{\Psi }_l^{\prime \prime }+\left[k^2+\left({\displaystyle \frac{\eta }{\eta _1}}\right)^{2\beta }q^2{\displaystyle \frac{N}{\eta ^2}}\right]\mathrm{\Psi }_l=0,`$ (2.9)
where $`{}_{}{}^{}\frac{}{\eta }`$ and $`N\frac{1}{4}\left(n\lambda 1\right)^2\frac{1}{4}`$. Furthermore, $`\beta \sigma \lambda `$. $`\beta <0`$ since only solutions with contracting extra dimensions will be discussed. $`1\beta <1/(1+\sqrt{3})`$, where the lower boundary corresponds to $`n=1`$ and the upper bound gives the value for large $`n`$.
Equation (2.9) can be solved in a closed form for one extra dimension $`n=1`$. In general, for $`n>1`$, to our knowledge, apart from the case $`n=6`$, there are no solutions in closed form. However, it is possible to find approximate solutions.
* For $`n=1`$ and $`\eta <\eta _1`$ the equation for $`\mathrm{\Psi }_l`$ (cf. equation (2.9)) reads
$`\mathrm{\Psi }_l^{\prime \prime }+\left[k^2+\left({\displaystyle \frac{\eta }{\eta _1}}\right)^2q^2+{\displaystyle \frac{1}{4\eta ^2}}\right]\mathrm{\Psi }_l=0,`$ (2.10)
which is solved by
$`\mathrm{\Psi }_l={\displaystyle \frac{\sqrt{\pi }}{2}}e^{\frac{\pi }{2}q\eta _1}{\displaystyle \frac{\left(k\eta \right)^{\frac{1}{2}}}{\sqrt{k}}}H_{iq\eta _1}^{(2)}(k\eta ),`$ (2.11)
satisfying the Wronskian condition $`\mathrm{\Psi }_l^{}\mathrm{\Psi }_l\mathrm{\Psi }_l^{}\mathrm{\Psi }_l^{}=i`$ and $`H_\nu ^{(2)}(z)`$ is the Hankel function of the second kind.
* For $`n>1`$ and $`\eta <\eta _1`$, in general approximate solutions can be found to the mode equation (2.9). In this case, there is a natural distinction into two cases .
1. For $`\left(\frac{\eta }{\eta _1}\right)^{2\beta }q^2<k^2`$, or $`\omega _q<\omega _k`$ in terms of the physical frequencies $`\omega _k=k/a(\eta )`$ and $`\omega _q=q/b(\eta )`$, equation (2.9) becomes approximately,
$`\mathrm{\Psi }_l^{\prime \prime }+\left[k^2{\displaystyle \frac{N}{\eta ^2}}\right]\mathrm{\Psi }_l=0,`$ (2.12)
which is solved by
$`\mathrm{\Psi }_l={\displaystyle \frac{\sqrt{\pi }}{2}}{\displaystyle \frac{\sqrt{k\eta }}{\sqrt{k}}}H_\mu ^{(2)}(k\eta ),`$ (2.13)
where $`H_\mu ^{(2)}`$ is the Hankel function of the second kind and $`\mu ^2\frac{1}{4}+N\mu =\frac{1}{2}(n\lambda 1)`$. The mode functions satisfy the Wronskian condition.
2. For $`\left(\frac{\eta }{\eta _1}\right)^{2\beta }q^2>k^2`$, or $`\omega _q>\omega _k`$, equation (2.9) can be approximated by,
$`\mathrm{\Psi }_l^{\prime \prime }+\left[\left({\displaystyle \frac{\eta }{\eta _1}}\right)^{2\beta }q^2{\displaystyle \frac{N}{\eta ^2}}\right]\mathrm{\Psi }_l=0,`$ (2.14)
which is solved by
$`\mathrm{\Psi }_l={\displaystyle \frac{\sqrt{\pi }}{2}}\left(\kappa \eta \right)^{\frac{1}{2}}H_{\mu \kappa }^{(2)}\left[\left(q\eta \right)\kappa \left({\displaystyle \frac{\eta }{\eta _1}}\right)^\beta \right],`$ (2.15)
where $`\kappa \frac{1}{\beta +1}`$ and $`\mu =\frac{1}{2}(n\lambda 1)`$.
The case $`q=0`$ is covered by the first case, $`\left(\frac{\eta }{\eta _1}\right)^{2\beta }q^2<k^2`$, thus the solutions are not written explicitly.
In the background (2.4), for $`\eta \eta _1`$, the mode equation is given by
$`\mathrm{\Psi }_l^{\prime \prime }+\left[k^2+\left({\displaystyle \frac{\eta +2\eta _1}{\eta _1}}\right)^2q^2\right]\mathrm{\Psi }_l=0.`$ (2.16)
Introducing $`z\left(\frac{2q}{\eta _1}\right)^{\frac{1}{2}}\left(\eta +2\eta _1\right)`$ and $`\alpha \frac{\eta _1k^2}{2q}`$ this can be transformed into the equation for parabolic cylinder functions ,
$`{\displaystyle \frac{d^2\mathrm{\Psi }_l}{dz^2}}+\left[{\displaystyle \frac{z^2}{4}}\alpha \right]\mathrm{\Psi }_l=0,`$ (2.17)
which is solved by
$`\mathrm{\Psi }_l={\displaystyle \frac{1}{\sqrt{2}}}\left({\displaystyle \frac{\eta _1}{2q}}\right)^{\frac{1}{4}}\left[c_{}E(\alpha ,z)+c_+E^{}(\alpha ,z)\right],`$ (2.18)
where the Wronskian condition on the mode functions was applied and the normalization for the Bogoliubov coefficients $`|c_+|^2|c_{}|^2=1`$ was used. Using the approximations ((19.24) of ) gives expressions for $`\mathrm{\Psi }_l`$ and $`\mathrm{\Psi }_l^{}`$ at $`\eta =\eta _1`$.
1. Namely, for $`\omega _q/\omega _k<1`$, it is found that
$`\mathrm{\Psi }_l(\eta _1)`$ $``$ $`{\displaystyle \frac{1}{\sqrt{2k}}}[c_{}e^{ik\eta _1+i\frac{\pi }{4}}+c_+e^{ik\eta _1i\frac{\pi }{4}}]`$
$`\mathrm{\Psi }_l^{}(\eta _1)`$ $``$ $`\sqrt{{\displaystyle \frac{k}{2}}}[c_{}e^{ik\eta _1i\frac{\pi }{4}}+c_+e^{ik\eta _1+i\frac{\pi }{4}}].`$ (2.19)
2. For $`\omega _q/\omega _k>1`$ it follows that
$`\mathrm{\Psi }_l(\eta _1)`$ $``$ $`{\displaystyle \frac{1}{\sqrt{2q}}}[c_{}e^{i\frac{q\eta _1}{2}+i\frac{\pi }{4}}+c_+e^{i\frac{q\eta _1}{2}i\frac{\pi }{4}}]`$
$`\mathrm{\Psi }_l^{}(\eta _1)`$ $``$ $`\sqrt{{\displaystyle \frac{q}{2}}}[c_{}e^{i\frac{q\eta _1}{2}i\frac{\pi }{4}}+c_+e^{i\frac{q\eta _1}{2}+i\frac{\pi }{4}}].`$ (2.20)
The total magnetic energy density is given by
$`\rho =2{\displaystyle \frac{R^n}{(2\pi )^{n+3}}}{\displaystyle \left[\left(\frac{k}{a}\right)^2+\left(\frac{q}{b}\right)^2\right]^{\frac{1}{2}}|c_{}|^2๐V},`$ (2.21)
where, assuming that the volume consists of two spheres, $`dV=\frac{1}{a^3b^n}\frac{2\pi ^{\frac{3}{2}}}{\mathrm{\Gamma }(\frac{3}{2})}k^2dk\frac{2\pi ^{\frac{n}{2}}}{\mathrm{\Gamma }(\frac{n}{2})}q^{n1}dq`$. At $`\eta =\eta _1`$ the comoving wavenumbers $`k`$ and $`q`$ are equal to the physical momenta, since $`a_1=1=b_1`$. The spectral energy density $`\rho (\omega _k)=d\rho /d\mathrm{log}\omega _k`$ is then given by
$`\rho (\omega _k)=16{\displaystyle \frac{R^n}{(2\pi )^{n+3}}}{\displaystyle \frac{\pi ^{1+\frac{n}{2}}}{\mathrm{\Gamma }(\frac{n}{2})}}\omega _k^{4+n}{\displaystyle ๐Y[1+Y^2]^{\frac{1}{2}}Y^{n1}|c_{}|^2},`$ (2.22)
where $`Y\frac{\omega _q}{\omega _k}`$, and $`\omega _k=\frac{k}{a}`$, $`\omega _q=\frac{q}{b}`$.
During most of its history the universe had a very high conductivity, implying that a primordial magnetic field evolves while its flux is conserved. This makes the dimensionless ratio $`r\rho _B/\rho _\gamma `$ approximately constant , where $`\rho _B`$ is the magnetic field energy density and $`\rho _\gamma `$ is the energy density of the background radiation. Thus $`r`$ is a good measure of the strength of a cosmological magnetic field. Furthermore, $`r=\mathrm{\Omega }_{em}/\mathrm{\Omega }_\gamma `$, where $`\mathrm{\Omega }=\rho /\rho _c`$ with $`\rho _c`$ the critical energy density, and $`\mathrm{\Omega }_\gamma =(H_1/H)^2(a_1/a)^4`$. Thus expressing the critical energy density in terms of the $`D`$-dimensional Planck mass $`M_D`$, $`\rho _c=\frac{3}{8\pi }R^nM_D^{n+2}H^2`$, leads to
$`r(\omega _k)={\displaystyle \frac{16}{3}}{\displaystyle \frac{8\pi }{(2\pi )^{n+3}}}{\displaystyle \frac{\pi ^{1+\frac{n}{2}}}{\mathrm{\Gamma }(\frac{n}{2})}}a^n\left({\displaystyle \frac{H_1}{M_D}}\right)^{n+2}\left({\displaystyle \frac{\omega _k}{\omega _1}}\right)^{4+n}{\displaystyle _0^{Y_{max}}}๐YY^{n1}\left[1+Y^2\right]^{\frac{1}{2}}|c_{}|^2,`$ (2.23)
where $`\omega _1\frac{k_1}{a}`$ and the maximal comoving wave number $`k_1H_1`$. Furthermore, an upper cut-off $`Y_{max}=\omega _{q_{max}}/\omega _k`$ has been introduced. This is justified by the sudden transition approximation, which is used here, since at the transition time, $`\eta =\eta _1`$, the metric is continuous but not its first derivative. This means that for modes with periods much larger than the duration of the transition phase, the transition phase can be treated as instantaneous. However, without an upper cut-off this type of approximation leads to an ultraviolet divergence .
For $`q>0`$ and $`n=1`$, that is one extra dimension, continuously matching at $`\eta =\eta _1`$ the solutions (2.11) and (2.18) on superhorizon scales $`k\eta _11`$, $`q\eta _11`$ leads to the following Bogoliubov coefficients for $`\omega _q/\omega _k<1`$ and $`\omega _q/\omega _k>1`$,
$`c_{}e^{ik\eta _1}`$ $``$ $`{\displaystyle \frac{1}{\sqrt{2\pi }}}{\displaystyle \frac{1}{\sqrt{k\eta _1}}}\left[1+{\displaystyle \frac{1}{2}}\mathrm{ln}k\eta _1ik\eta _1\mathrm{ln}k\eta _1\right]e^{i\frac{\pi }{4}}\mathrm{for}Y<1,`$ (2.24)
$`c_{}e^{i\frac{q\eta _1}{2}}`$ $``$ $`{\displaystyle \frac{1}{\sqrt{2\pi }}}{\displaystyle \frac{1}{\sqrt{q\eta _1}}}\left[1+{\displaystyle \frac{1}{2}}\mathrm{ln}k\eta _1iq\eta _1\mathrm{ln}k\eta _1\right]e^{i\frac{\pi }{4}}\mathrm{for}Y>1.`$ (2.25)
Neglecting subleading terms, it follows that the ratio $`r(\omega _k)`$ is given by
$`r(\omega _k){\displaystyle \frac{1}{3\pi ^3}}\left({\displaystyle \frac{H_1}{M_4}}\right)^3\left({\displaystyle \frac{M_5}{M_4}}\right)^3\left({\displaystyle \frac{\omega _k}{\omega _1}}\right)^3\left(\mathrm{ln}{\displaystyle \frac{\omega _k}{\omega _1}}\right)^2{\displaystyle \frac{\omega _{q_{max}}}{\omega _1}},`$ (2.26)
where $`\omega _{q_{max}}(\eta )=\frac{q_{max}}{b}`$ and it was assumed that $`\omega _{q_{max}}>\omega _k`$.
The case $`Y<1`$ includes the limit $`q=0`$. Therefore together with $`\rho _{em}(\omega _k)=2\frac{\omega _k^4}{\pi ^2}|c_{}(\omega _k)|^2`$ the following expression for the ratio of magnetic to background radiation energy density is obtained for $`q=0`$, $`n=1`$,
$`r(\omega _k){\displaystyle \frac{2}{3\pi ^2}}\left({\displaystyle \frac{H_1}{M_4}}\right)^2\left({\displaystyle \frac{\omega _k}{\omega _1}}\right)^3\left(\mathrm{ln}{\displaystyle \frac{\omega _k}{\omega _1}}\right)^2.`$ (2.27)
For more than one extra dimension, $`n>1`$, the solutions for $`\mathrm{\Psi }_l`$ and $`\mathrm{\Psi }_l^{}`$ for $`\eta <\eta _1`$ and $`\eta >\eta _1`$ are matched at $`\eta =\eta _1`$ for $`Y<1`$ and $`Y>1`$ for superhorizon modes, $`k\eta _11`$, $`q\eta _11`$. This leads to the following expressions for the Bogoliubov coefficient $`c_{}`$
$`c_{}e^{ik\eta _1}`$ $``$ $`{\displaystyle \frac{2^{\mu \frac{3}{2}}}{\sqrt{\pi }}}\mathrm{\Gamma }(\mu )\left(k\eta _1\right)^{\frac{1}{2}\mu }\left[\left(\mu {\displaystyle \frac{1}{2}}\right){\displaystyle \frac{1}{k\eta _1}}+i\right]e^{i\frac{\pi }{4}}\mathrm{for}Y<1,`$ (2.28)
$`c_{}e^{i\frac{q\eta _1}{2}}`$ $``$ $`{\displaystyle \frac{2^{\mu \kappa \frac{3}{2}}}{\sqrt{\pi }}}\mathrm{\Gamma }(\mu \kappa )\left(\kappa q\eta _1\right)^{\frac{1}{2}\mu \kappa }\left[\left(\mu {\displaystyle \frac{1}{2}}\right){\displaystyle \frac{1}{q\eta _1}}+i\right]e^{i\frac{\pi }{4}}\mathrm{for}Y>1.`$ (2.29)
Using the expressions for $`|c_{}|`$ for $`Y<1`$ and $`Y>1`$, as provided by equations (2.28) and (2.29) for more than one extra dimension, $`n>1`$, leads to the ratio of magnetic spectral energy density to background radiation density,
$`r(\omega _k)`$ $``$ $`๐ฉa^{1+2\mu \kappa n}\left({\displaystyle \frac{H_1}{M_D}}\right)^{n+2}\left({\displaystyle \frac{\omega _{q_{max}}}{\omega _1}}\right)^{n2\mu \kappa }\left({\displaystyle \frac{\omega _k}{\omega _1}}\right)^3`$ (2.30)
where
$`๐ฉ`$ $``$ $`{\displaystyle \frac{16}{3}}{\displaystyle \frac{8\pi }{(2\pi )^{n+3}}}{\displaystyle \frac{\pi ^{1+\frac{n}{2}}}{\mathrm{\Gamma }(\frac{n}{2})}}{\displaystyle \frac{2^{2\mu \kappa 3}}{\pi (n2\mu \kappa )}}\mathrm{\Gamma }^2(\mu \kappa )\kappa ^{12\mu \kappa }\left(\mu {\displaystyle \frac{1}{2}}\right)^2`$
where subleading terms have been omitted and $`\omega _{q_{max}}>\omega _k`$ was assumed. The resulting spectrum is growing in frequency.
The expression for $`q=0`$ can be derived using the expression for $`c_{}`$ for $`Y<1`$ (cf. equation (2.28)). Together with $`\rho _{em}=2\frac{\omega ^4}{\pi ^2}|c_{}|^2`$ this implies for $`q=0`$, $`n>1`$
$`r(\omega _k){\displaystyle \frac{2^{n\lambda 2}}{3\pi ^2}}\mathrm{\Gamma }^2\left({\displaystyle \frac{n\lambda 1}{2}}\right)\left(2n\lambda \right)^2\left({\displaystyle \frac{H_1}{M_4}}\right)^2\left({\displaystyle \frac{\omega _k}{\omega _1}}\right)^{4n\lambda }.`$ (2.31)
Furthermore, $`n\lambda =\sqrt{\frac{3n}{n+2}}`$. Since $`n\lambda <4`$ the resulting spectrum for $`r(\omega _k)`$ is increasing in frequency.
## 3 Constraining the model
The expressions for the ratio $`r(\omega _k)`$ determining the ratio of the energy density of the magnetic field in comparison with the energy density of the background radiation contain several parameters apart from the physical frequencies $`\omega _k`$ and $`\omega _{q_{max}}`$. The free parameters are the Hubble parameter at the time of transition $`H_1`$, the $`D`$-dimensional Planck mass $`M_D`$ and the number of extra dimensions $`n`$.
There are several constraints from observations. $`r(\omega _k)`$ has to be less than one for all frequencies in order not to overclose the universe. For $`r(\omega _k)`$ increasing with frequency this implies $`r(\omega _1)<1`$. This is the case for the spectra given by equations (2.30) and (2.31) applicable for backgrounds with more than one extra dimension. In the case of one extra dimension the expressions for $`r(\omega _k)`$ (cf. equations (2.26) and (2.27)) have a maximum at some frequency $`\omega _2`$. Thus the constraint $`r(\omega _2)<1`$ is imposed.
Newtonian gravity has been tested down to length scales of the order of 1 mm . This implies the constraint $`\frac{M_D}{M_4}(1.616\times 10^{32})^{\frac{n}{n+2}}`$. Furthermore, with $`T_1`$ the temperature at the beginning of the radiation epoch, big bang nucleosynthesis requires that $`T_1>10`$ MeV. This imposes a bound on $`H_1`$ by using $`\frac{H_1}{M_4}=1.66g_{}^{\frac{1}{2}}(T_1)\left(\frac{T_1}{M_4}\right)^2`$, where for $`T_1>300`$ GeV the number of effective degrees of freedom is given by $`g_{}(T_1)=106.75`$ , namely, $`\mathrm{log}\frac{H_1}{M_4}>40.94`$.
The ratio $`r`$ calculated at the galactic scale $`\omega _G^1`$ of order of 1 Mpc determines the strength of the primordial seed magnetic field at the time of galaxy formation. In the standard picture of a galactic magnetic dynamo operating since the time of galaxy formation, a seed magnetic field of at least $`B_s10^{20}`$G , corresponding to $`r(\omega _G)>10^{37}`$, is needed to explain the currently observed galactic magnetic field of a few $`\mu `$G . However, taking into account a non-vanishing cosmological constant, it was shown in that initial magnetic seed field strengths can be much below $`10^{20}`$G. Thus $`r(\omega _G)`$ can be as low as $`10^{57}`$ and correspondingly the magnetic seed field $`B_s10^{30}`$ G.
In the following, using the constraint $`r(\omega _1)<1`$ or $`r(\omega _2)<1`$, respectively, the constraint from the size of the extra dimension and from big bang nucleosynthesis an upper limit on the ratio $`r(\omega _G)`$ and thus the strength of the magnetic seed field strength at the time of galaxy formation is derived. The strength of the seed field in terms of $`r`$ is given by $`B_s3r^{\frac{1}{2}}\times 10^2`$ G .
In addition, the maximally amplified frequency calculated with respect to present day $`\omega _1(\eta _0)`$ is given by $`\omega _16\times 10^{11}\mathrm{Hz}\left(\frac{H_1}{M_4}\right)^{\frac{1}{2}}`$ and the frequency corresponding to galactic scale, $`\omega _G10^{14}`$Hz . Furthermore, $`r(\omega _G)`$ is assumed to be of the form $`r(\omega _G)=10^m`$ where the exponent $`m`$ will be constrained by observational bounds. In the standard picture of the galactic dynamo, $`m37`$. In the following an upper bound on $`m`$ will be found.
For one extra dimension, $`n=1`$, the spectra (2.26) and (2.27) have a maximum at $`\frac{\omega _2}{\omega _1}=e^{\frac{2}{3}}`$. Thus the constraint of the critical density is imposed by requiring $`r(\omega _2)<1`$.
In the case where the momenta lying in the extra dimension are not taken into account, that is $`q=0`$, $`r(\omega _G)=10^m`$ where $`\omega _G=10^{14}`$Hz implies,
$`m=\mathrm{log}{\displaystyle \frac{2}{3\pi ^2}}+{\displaystyle \frac{1}{2}}\mathrm{log}{\displaystyle \frac{H_1}{M_4}}+3\mathrm{log}{\displaystyle \frac{10^{14}}{6\times 10^{11}}}+\mathrm{log}[\mathrm{ln}{\displaystyle \frac{10^{14}}{6\times 10^{11}}}1.1513\mathrm{log}{\displaystyle \frac{H_1}{M_4}}]^2.`$ (3.32)
Big bang nucleosynthesis requires $`\mathrm{log}\frac{H_1}{M_4}>40.94`$ and the constraint $`r(\omega _2)<1`$ implies $`\mathrm{log}\frac{H_1}{M_4}<1.2`$. Evaluating $`m`$ at the upper limit $`\mathrm{log}\frac{H_1}{M_4}=1.2`$ gives $`r(\omega _G)<10^{74}`$ corresponding to a magnetic seed field strength of $`B_s<10^{39}`$ G. Thus magnetic fields created in this setting are too weak in order to seed the galactic magnetic dynamo.
For $`q>0`$ and $`n=1`$ the various constraints mentioned above applied to the expression for $`r(\omega _k)`$ (cf. equation (2.26)) lead to the constraint on $`m`$
$`m<\mathrm{log}{\displaystyle \frac{9e^2}{4}}+3\mathrm{log}{\displaystyle \frac{10^{14}}{6\times 10^{11}}}{\displaystyle \frac{3}{2}}\mathrm{log}{\displaystyle \frac{H_1}{M_4}}+\mathrm{log}[\mathrm{ln}{\displaystyle \frac{10^{14}}{6\times 10^{11}}}1.1513\mathrm{log}{\displaystyle \frac{H_1}{M_4}}]^2.`$ (3.33)
Evaluating $`m`$ at the lower bound $`\mathrm{log}\frac{H_1}{M_4}=40.94`$ results in the bound $`r(\omega _G)<10^{13}`$ corresponding to a magnetic seed field strength of $`B_s<10^8`$ G. Thus in this case the lower bound on the magnetic seed field imposed by the galactic dynamo can be satisfied easily.
Assuming that $`T_1M_D`$ results in an additional constraint on $`\mathrm{log}\frac{H_1}{M_4}`$ by using the bound on the size of the extra dimensions. Namely, for any $`n`$,
$`\mathrm{log}{\displaystyle \frac{H_1}{M_4}}>\mathrm{log}17.15+{\displaystyle \frac{2n}{n+2}}\mathrm{log}(1.616\times 10^{32}).`$ (3.34)
This gives a bound on $`\mathrm{log}\frac{H_1}{M_4}`$ stronger than the one from big bang nucleosynthesis only upto three extra dimensions $`n3`$. In particular in the case at hand, for $`n=1`$, it implies $`\mathrm{log}\frac{H_1}{M_4}>19.96`$. Evaluating $`m`$ at this value of $`\mathrm{log}\frac{H_1}{M_4}`$ leads to $`r(\omega _G)<10^{43}`$ and correspondingly the magnetic seed field strength $`B_s<10^{23}`$ G. Thus in the case where $`T_1M_5`$ the created magnetic seed field satisfies the weaker bound of $`B_s>10^{30}`$ G.
For more than one extra dimension $`n>1`$ and $`q>0`$ the constraint on $`r(\omega _k)`$ (cf. equation (2.30)) at $`\omega _1`$ together with $`r(\omega _G)=10^m`$ leads to
$`m<{\displaystyle \frac{3}{2}}\mathrm{log}{\displaystyle \frac{H_1}{M_4}}+3\mathrm{log}{\displaystyle \frac{10^{14}}{6\times 10^{11}}}.`$ (3.35)
Using the constraint from big bang nucleosynthesis $`\mathrm{log}\frac{H_1}{M_4}>40.94`$ results in $`m<15.9`$ and thus $`r(\omega _G)<10^{16}`$ and hence seed magnetic fields with strengths upto $`B_s<10^{10}`$ G can be created. Assuming that the temperature at the beginning of the radiation epoch, $`T_1`$, is given by $`M_D`$, that is $`T_1M_D`$, changes the bound on $`m`$ for two and three extra dimensions (cf. equation (3.34)). In this case, for $`n=2`$ extra dimensions, $`m<31.5`$ implying $`r(\omega _G)<10^{32}`$ and the magnetic field strength $`B_s<10^{18}`$ G. For $`n=3`$ extra dimensions, $`m<21.95`$ and hence $`r(\omega _G)<10^{22}`$ and the magnetic field strength $`B_s<10^{13}`$G.
This is to be compared with the case where the internal momenta are not taken into account . Applying the constraints to equation (2.31) implies
$`m<\left(1{\displaystyle \frac{n\lambda }{4}}\right)\mathrm{log}\left[{\displaystyle \frac{2^{n\lambda 2}}{3\pi ^2}}\mathrm{\Gamma }^2\left({\displaystyle \frac{n\lambda 1}{2}}\right)\left(2n\lambda \right)^2\right]+\left(4n\lambda \right)\mathrm{log}{\displaystyle \frac{10^{14}}{6\times 10^{11}}}.`$ (3.36)
In this case the bound on $`m`$ depends on the number of extra dimensions $`n`$. This is related to the fact that the spectral index in the expression for $`r(\omega _k)`$ (cf. equation (2.31)) is given by $`4n\lambda `$ and thus depends explicitly on the number of dimensions. In the case, where $`n>1`$ and $`q>0`$, the spectral index is 3, independent of the number of extra dimensions. In figure 1 the magnetic seed field strength $`B_s`$ is plotted as a function of the number of extra dimensions $`n`$ in the case $`n>1`$, $`q=0`$. As can be seen the resulting values for $`B_s`$ are very small, much below even the weaker constraint, $`B_s>10^{30}`$G .
In the cases $`n=1`$ and $`n>1`$ for $`q>0`$, $`\omega _{q_{max}}=q_{max}/b`$ appears as a parameter in the expressions for $`r(\omega _k)`$ (cf. equations (2.26) and (2.30)). Assuming that $`q_{max}k_1`$ leads to $`\omega _{q_{max}}/\omega _1a`$. Using this in $`r(\omega _2)<1`$, for $`n=1`$, and in $`r(\omega _1)<1`$, for $`n>1`$, leads in both cases to a constraint of the form
$`๐ฉ_{}a_0\left({\displaystyle \frac{H_1}{M_4}}\right)^{n+2}\left({\displaystyle \frac{M_D}{M_4}}\right)^{(n+2)}<1,`$ (3.37)
where $`๐ฉ_{}=\frac{4}{27e^2\pi ^3}`$ for $`n=1`$ and $`๐ฉ_{}=๐ฉ`$ for $`n>1`$. If there are no additional constraints then equation (3.37) implies a lower bound on $`M_D/M_4`$, which has to be compared with the lower bound provided by the size of the extra dimensions. However, if in addition $`T_1M_D`$ is imposed, then equation (3.37) together with $`\frac{H_1}{M_4}1.66g_{}^{\frac{1}{2}}\left(\frac{M_D}{M_4}\right)^2`$ implies an upper bound on $`M_D/M_4`$, namely,
$`\mathrm{log}{\displaystyle \frac{M_D}{M_4}}<{\displaystyle \frac{\mathrm{log}\left(3\times 10^{31}๐ฉ_{}\right)}{n+3}}{\displaystyle \frac{n+\frac{5}{2}}{n+3}}\mathrm{log}1.66g_{}^{\frac{1}{2}},`$ (3.38)
where $`a_03\times 10^{31}\left(\frac{H_1}{M_4}\right)^{\frac{1}{2}}`$ was used. This bound is always larger than the lower bound on $`\mathrm{log}\frac{M_D}{M_4}`$ provided by the size of the extra dimensions. Thus, the assumption $`T_1M_D`$ is consistent with the various constraints. Moreover, although this upper bound on $`M_D/M_4`$ leads to an upper bound on $`H_1/M_4`$, the maximal strength of the magnetic seed field is not changed, since for $`n1,q>0`$ this was evaluated at the lower boundary of $`\mathrm{log}\frac{H_1}{M_4}`$.
## 4 Conclusions
The origin of magnetic fields on galactic and extragalactic scales is still an open problem. Different types of mechanisms have been proposed. In particular, in the creation of magnetic fields due to dynamical extra dimensions was proposed. Along these lines, here, a model consisting of two phases has been investigated. A higher dimensional epoch with three expanding, external (spatial) dimensions and $`n`$ contracting, internal dimensions is matched to a standard radiation dominated phase with static extra dimensions. Taking the internal momenta into account the final expression for the ratio $`r`$ of magnetic field energy density to background radiation energy density is obtained by integrating over the internal modes. In doing so the sudden approximation requires the introduction of a maximal frequency in the internal momentum space. The resulting spectrum is constrained by imposing bounds from observations, such as, the constraint from critical energy density, the size of the extra dimensions and big bang nucleosynthesis.
For one extra dimension, $`n=1`$, it was found that in the case where the momenta along the extra dimension are not taken into account, $`q=0`$, only very weak magnetic seed fields are created, $`B_s<10^{39}`$ G. However, in the case $`q>0`$ magnetic seed fields as strong as $`10^8`$ G can be obtained in general. Imposing the additional constraint $`T_1M_5`$ leads to magnetic seed fields $`B_s<10^{23}`$ G which satisfy the lower bound in a $`\mathrm{\Lambda }`$ universe .
In models with more than one extra dimension, $`n>1`$, strong magnetic seed fields can be created if the internal momenta are taken into account. In particular, not assuming that the temperature at the beginning of the radiation epoch is of the order of the $`D`$-dimensional Planck scale allows for the creation of seed magnetic fields with strengths of upto $`10^{10}`$ G. For more than three extra dimensions, this also holds if $`T_1M_D`$ is assumed. With this assumption for two and three extra dimensions results in weaker magnetic seed fields, with maximal field strengths, $`B_s<10^{18}`$ G for two extra dimensions and $`B_s<10^{13}`$G for three extra dimensions.
Therefore, in this particular model with extra dimensions, taking into account the momenta along the extra dimensions allows for the creation of strong magnetic fields.
## 5 Acknowledgements
I would like to thank M. A. Vรกzquez-Mozo for useful discussions. This work has been supported by the programme โRamรณn y Cajalโ of the M.E.C. (Spain). Partial support by Spanish Science Ministry Grants FPA 2002-02037 and BFM 2003-02121 is acknowledged.
|
warning/0506/hep-th0506196.html
|
ar5iv
|
text
|
# Geometric transitions and integrable systems
## 1 Introduction
Large $`N`$ duality has been at the center of many recent developments in topological string theory. In particular B-model transitions have revealed a fascinating interplay of random matrix models, integrable systems and Calabi-Yau geometry.
In this paper we generalize the results of reviewed in section two to a new class of conifold transitions among noncompact Calabi-Yau threefolds. As explained in section three, the starting point of our construction is a configuration
$$\begin{array}{ccccc}\stackrel{~}{๐บ}& & \stackrel{~}{๐ด}& & \\ & & & & \\ ๐บ& & ๐ด& & ๐ณ.\end{array}$$
of moduli spaces which generalizes the essential geometric features of the local transitions studied in . Here $`๐ณ`$ is a component of the moduli space of projective or quasi-projective Calabi-Yau threefolds and $`๐ด`$ is a subspace of $`๐ณ`$ parameterizing Calabi-Yau manifolds with isolated conifold singularities which admit a (quasi-)projective small resolution. The deepest stratum $`๐บ`$, which is a key element of the construction, parameterizes Calabi-Yau spaces with a genus $`g`$ curve $`\mathrm{\Sigma }`$ of $`A_1`$ singularities. The spaces $`\stackrel{~}{๐บ},\stackrel{~}{๐ด}`$ are moduli spaces of the resolution of Calabi-Yau spaces in $`๐บ`$ and respectively $`๐ด`$. Such geometric structures have been considered before in the physics literature in relation to $`N=2`$ gauge theories and open string superpotentials. Here we will show that they play a key role in B-model geometric transitions.
Our main construction is carried out in section four. We consider noncompact Calabi-Yau spaces fibered by affine quadrics over a fixed genus $`g`$ curve $`\mathrm{\Sigma }`$. A special feature of this model is that the moduli spaces $`\stackrel{~}{๐ด}`$, $`๐ณ`$ are isomorphic to the total spaces of vector bundles over $`๐บ`$.
Large $`N`$ duality is an equivalence between B-type open-closed topological strings on a resolved threefold corresponding to a point in $`\stackrel{~}{๐ด}`$ and closed topological strings on a generic threefold in $`๐ณ`$. In the present paper we establish this result for genus zero topological amplitudes in the geometric framework described above. The proof involves two parts.
The genus zero dynamics for closed B-topological strings on Calabi-Yau spaces is usually encoded in the intermediate Jacobian fibration over the moduli space, which supports an integrable system structure . In our case we show that the relevant integrable system for a family of threefolds parameterized by a normal slice to $`๐บ`$ in $`๐ณ`$ is the $`A_1`$ Hitchin integrable system. In particular, the normal slice $`๐ณ_s`$ at a point $`s๐บ`$ is isomorphic to the space of quadratic differentials on $`\mathrm{\Sigma }`$, which is the base of the Hitchin systems. This follows from a structure result for the intermediate Jacobians of the Calabi-Yau threefolds in $`๐ณ`$ proved in section five.
The second part of the proof is more physical in nature and involves B-topological open string dynamics on a small resolution parameterized by a generic point in $`\stackrel{~}{๐ด}`$. In section six we construct the holomorphic Chern-Simons theory which captures open string target space dynamics using the formalism of D-brane categories. Then we argue that the holomorphic Chern-Simons functional integral reduces to a finite dimensional integral on a real cycle in the product $`\text{Sym}^N(\mathrm{\Sigma })\times \text{Sym}^N(\mathrm{\Sigma })`$. This can be regarded as a generalized matrix model in which the eigenvalues are parameterized by a compact Riemann surface. The final result of this section is that the large $`N`$ planar limit of this generalized matrix model is captured by the same $`A_1`$ Hitchin system that was found in section five. This concludes the physical proof of genus zero large $`N`$ duality.
Acknowledgments. We are very grateful to Bogdan Florea and Antonella Grassi for collaboration at an early stage of the project and many useful discussions. We would also like to thank Jacques Distler, Sheldon Katz, Balรกzs Szendrรถi and Cumrun Vafa for helpful discussions. D.-E. D. would also like to acknowledge the partial support of the Alfred P. Sloan foundation and the hospitality of KITP Santa Barbara and The Aspen Center for Physics where part of this work was performed. The research of R.D. was supported by a NWO Spinoza grant and the FOM program String Theory and Quantum Gravity. R.D. was partially supported by NSF grant DMS 0104354 and FRG grant 0139799 for โThe Geometry of Superstringsโ. T.P. was partially supported by NSF grants FRG 0139799 and DMS 0403884. The work of C.M.H. was supported in part by a Marie Curie Fellowship under contract MEIF-CT-2003-500687, the Israel-US Binational Science Foundation, the ISF Centers of Excellence Program and Minerva.
## 2 Review of Dijkgraaf-Vafa transitions
In this section we will review large $`N`$ duality for a class of geometric transitions among noncompact Calabi-Yau threefolds first studied in . Adopting the common terminology in the physics literature, for us a geometric transition will be an extremal transition connecting two different components of a moduli space of Calabi-Yau threefolds through a degeneration. The degenerations usually considered in this context are nodal Calabi-Yau threefolds with isolated ODP singularities. More complicated singularities, such as rational double points, can also appear as junctions of geometric transitions and support very interesting large $`N`$ physics . We will not look at these more complicated geometries here but they certainly deserve a thorough investigation from the point of view of Dijkgraaf-Vafa quantization.
In the situation considered in , we have a moduli space $`๐ณ`$ of noncompact Calabi-Yau hypersurfaces in $`X_l^4`$ defined by equations of the form
(1)
$$uv+y^2l(x)=0$$
where $`l(x)`$ is an arbitrary polynomial of degree $`2n`$. The moduli space $`๐ณ`$ is the complex vector space of dimension $`2n+1`$ parameterizing the coefficients of $`l(x)`$. The degeneration takes place along a subvariety $`๐ด๐ณ`$ characterized by the property that
(2)
$$m(x)=(W_m^{}(x))^2$$
for $`m๐ด`$, where $`W_m(x)`$ is an arbitrary polynomial of degree $`n+1`$. Therefore $`๐ด`$ is a $`(n+1)`$-dimensional subvariety in $`๐ณ`$. In the following we will call $`W_m(x)`$ the classical superpotential for reasons that will shortly become clear. It is easy to check that if the roots of $`W_m^{}(x)`$ are distinct, $`X_m`$ has $`n`$ isolated ODPs given by solutions of the equations
$$u=v=y=0,W_m^{}(x)=0$$
We will refer to such points in $`๐ด`$ as generic points. If $`W_m^{}(x)`$ has coincident roots, $`X_m`$ develops more complicated singularities. A special role in the theory will be played by the singular point $`s`$ of $`๐ด`$ for which the polynomial $`s(x)`$ (and hence $`W_m^{}(x)`$) is identically zero.
For a generic point $`m๐ด`$ we can easily construct a quasi-projective crepant resolution of $`X_m`$ by blowing-up $`^4`$ along the subvariety
$$u=0,yW_m^{}(x)=0$$
This resolution is not unique since we can obtain a different one for example by blowing up $`^4`$ along
$$v=0,y+W_m^{}(x)=0,$$
and we can also consider obvious variations. However all these resolutions are related by flops. Therefore we will have a moduli space $`\stackrel{~}{๐ด}`$ of smooth Calabi-Yau threefolds $`\stackrel{~}{X}_{\stackrel{~}{m}}`$ and a finite-to one (in fact in this case two to one) surjective map $`\rho :\stackrel{~}{๐ด}๐ด`$ so that $`\stackrel{~}{X}_{\stackrel{~}{m}}`$ is a quasi-projective crepant resolution of $`X_{\rho (\stackrel{~}{m})}`$. We will denote by $`C_1,\mathrm{},C_n`$ the exceptional curves on $`\stackrel{~}{X}_{\stackrel{~}{m}}`$.
Resolving the singular threefolds $`X_m`$ corresponding to the special points in $`๐ด`$ where $`W_m^{}(x)`$ acquires multiple roots is more involved. Here we will only discuss the extreme case of the threefold $`X_s`$ corresponding to the singular point $`s๐ด`$. Note that $`s`$ is a branch point for the cover $`\rho :\stackrel{~}{๐ด}๐ด`$. The inverse image $`\rho ^1(s)`$ consists of a single point $`\stackrel{~}{s}\stackrel{~}{๐ด}`$. Since $`W_s(x)0`$, the singular threefold $`X_s`$ is determined by the equation
$$uv+y^2=0.$$
Therefore $`X_s`$ is isomorphic to a direct product of the form $`\times Y`$ where $`Y`$ is the singular quadric surface described by the same equation in $`^3`$ with coordinates $`(u,v,y)`$. In particular $`X_s`$ has a line $`u=v=y=0`$ of $`A_1`$ singularities. The resolution $`\stackrel{~}{X}_{\stackrel{~}{s}}`$ is isomorphic to the direct product $`\times \stackrel{~}{Y}`$, where $`\stackrel{~}{Y}`$ is the minimal resolution of $`Y`$; $`\stackrel{~}{Y}`$ is isomorphic to the total space of the line bundle $`๐ช(2)`$ over $`^1`$, and the map $`\stackrel{~}{Y}Y`$ is the contraction of the zero section.
Note that a resolution $`\stackrel{~}{X}_{\stackrel{~}{m}}`$ corresponding to a generic point is related to $`\stackrel{~}{X}_{\stackrel{~}{s}}`$ by a complex structure deformation and has the structure of a fibration in quadrics over the complex line. Consider the projection map $`\stackrel{~}{\pi }:\stackrel{~}{X}_{\stackrel{~}{m}}`$ defined in terms of local coordinates by forgetting $`(y,u,v)`$. Using the equation (1), the fibers of this projection are easily seen to be affine quadrics in $`^3`$. The fibers over points in $``$ different from the zeroes of $`W_m(x)`$ are smooth affine quadrics which can be described as smoothings of the A<sub>1</sub> singularity. The fiber over a zero of $`W_m(x)`$ is isomorphic to the resolution $`\stackrel{~}{Y}`$ of the $`A_1`$ singularity. In particular these fibers contain the exceptional curves $`C_1,\mathrm{},C_n`$ which may be identified with the zero section of $`\stackrel{~}{Y}`$.
In the framework of topological B-strings, large $`N`$ duality relates topological open strings on a resolution $`\stackrel{~}{X}_{\stackrel{~}{m}}`$ to topological closed strings on a smoothing $`X_l`$. The topological open string B-model is constructed by wrapping $`N_i`$ topological B-branes on the $`i`$-th exceptional curve $`C_i`$, $`i=1,\mathrm{},n`$ in $`\stackrel{~}{X}_{\stackrel{~}{m}}`$. The precise statement for extremal transitions among general projective or quasi-projective Calabi-Yau varieties is not known<sup>1</sup><sup>1</sup>1To first order however, large N duality for general transitions can be made precise, and is formulated and proved in .. Here we will explain how large $`N`$ duality works for the special class of transitions introduced above. Our discussion follows closely .
First we should explain what we mean by a topological open B-model defined by wrapping branes on the exceptional curves. From a rigorous mathematical point of view, topological $`๐`$-branes should be described in terms of derived objects โ or, more concretely complexes of vector bundles โ on the threefold $`\stackrel{~}{X}_{\stackrel{~}{m}}`$. Although this formalism will be very useful for generalizations, in the present case it suffices to think informally of a D-brane with multiplicity $`N_i`$ wrapping an exceptional curve $`C_i`$ as a rank $`N_i`$ vector bundle $`E_i`$ on $`C_i`$. More specifically for the purpose of large $`N`$ duality we will consider trivial bundles of the form $`E_i=C_i\times ^{N_i}`$. In the following we fix a threefold $`\stackrel{~}{X}_{\stackrel{~}{m}}`$ with $`m=\rho (\stackrel{~}{m})`$ generic.
The dynamics of such a brane should be described in terms of a set of off-shell fields which in this case are $`C^{\mathrm{}}`$ bundle valued differential forms on $`C_i`$ of the form
$$A^{0,p}\left(\text{End}(E_i)\mathrm{\Lambda }^qN_{C_i/\stackrel{~}{X}_{\stackrel{~}{m}}}\right)$$
subject to the constraint $`p+q=1`$. In principle, one would like to write down a holomorphic Chern-Simons action functional for such fields using first principles and work out the rules for quantization. However, this is quite difficult to do in practice because the quantization of holomorphic Chern-Simons theories is typically untractable.
One of the main insights of is that this program can actually be carried out in the present geometric situation as follows. First construct the holomorphic Chern-Simons action for B-branes on the resolution of the singular threefold $`\stackrel{~}{X}_{\stackrel{~}{s}}`$. Then construct the holomorphic Chern-Simons action for branes on a generic threefold $`\stackrel{~}{X}_{\stackrel{~}{m}}`$ by adding a superpotential deformation. The end result is that the dynamics of B-branes at a generic point $`\stackrel{~}{m}\stackrel{~}{๐ด}`$ is captured by a holomorphic matrix model.
The first step is easy. The resolved threefold $`\stackrel{~}{X}_{\stackrel{~}{s}}`$ is isomorphic to the product $`\times \stackrel{~}{Y}`$, where $`\stackrel{~}{Y}`$ is isomorphic to the total space of $`๐ช(2)`$ over $`^1`$. Therefore $`\stackrel{~}{X}_{\stackrel{~}{s}}`$ contains a ruled surface $`S=\times ^1`$ where the rational fibers of the ruling are exceptional curves on $`\stackrel{~}{X}_{\stackrel{~}{s}}`$ obtained by resolving the line of $`A_1`$ singularities. In particular all fibers of $`S^1`$ are $`(0,2)`$ curves on $`\stackrel{~}{X}_{\stackrel{~}{s}}`$. We consider a system of $`N`$ topological B-branes wrapping a given fiber $`C`$ of the ruling, where
$$N=\underset{i=1}{\overset{n}{}}N_i$$
Informally this means that we pick up a trivial bundle $`E`$ of the form $`C\times ^N`$. We want to write down a holomorphic Chern-Simons action functional on the set of off-shell fields
$$A^{0,p}\left(\text{End}(E)\mathrm{\Lambda }^qN_{C/\stackrel{~}{X}_{\stackrel{~}{s}}}\right)=A^{0,p}\left(\mathrm{\Lambda }^qN_{C/\stackrel{~}{X}_{\stackrel{~}{s}}}\right)M_N()$$
with $`p+q=1`$. Since $`N_{C/\stackrel{~}{X}_{\stackrel{~}{s}}}๐ช_C\mathrm{\Omega }_C^1`$ we are left with three off-shell fields
$$\varphi ^{0,1}A^{0,1}M_N(),\varphi ^{0,0}A^{0,0}M_N(),\varphi ^{1,0}A^{0,0}\mathrm{\Omega }_C^1M_N().$$
In order to write down the holomorphic Chern-Simons action, we have to regard the holomorphic bundle $`E`$ as a $`C^{\mathrm{}}`$ bundle equipped with a $`(0,1)`$ connection which in this case can be taken to be the trivial Dolbeault operator $`\overline{}`$ on $`C`$. The field $`\varphi ^{0,1}`$ represents an off-shell deformation of the background connection which can be eliminated by performing a gauge transformation. Therefore it suffices to write down an action for the remaining fields $`\varphi ^{0,0},\varphi ^{1,0}`$. This is simply given by
(3)
$$S_{\stackrel{~}{s}}=_C\mathrm{Tr}\left(\varphi ^{1,0}\overline{}\varphi ^{0,0}\right).$$
This is the holomorphic Chern-Simons action for B-branes on $`\stackrel{~}{X}_{\stackrel{~}{s}}`$. The Chern-Simons actions for the branes on an arbitrary threefold $`\stackrel{~}{X}_{\stackrel{~}{m}}`$ is constructed in by adding a superpotential term to the functional (3).
From our perspective, this construction can be best summarized as follows. Our final goal is to construct the partition function for B-branes on $`\stackrel{~}{X}_{\stackrel{~}{m}}`$, at least as a formal perturbative expansion. The partition function of any gauge theory is obtained by formally integrating over all fields in the action and dividing by the volume of the gauge group. Usually, if the fluctuations of the theory are described by some complex fields $`\psi `$, the path integral can be formally written as an integral over the space of fields $`D\psi D\overline{\varphi }e^{S(\varphi ,\overline{\varphi })}`$. However, the holomorphic Chern-Simons action depends only on the holomorphic part of the fields; the antiholomorphic part is absent. Then the quantum theory should be formally defined by integrating the holomorphic measure $`D\varphi e^{S(\varphi )}`$ over a suitable middle dimensional real cycle $`\mathrm{\Gamma }`$ in the space of fields . Therefore the formal expression of the functional integral is
(4)
$$Z=\frac{1}{\text{vol}(๐ข)}_\mathrm{\Gamma }D\varphi e^{S(\varphi )}.$$
The cycle $`\mathrm{\Gamma }`$ should be regarded as part of the data specifying the quantum theory. The dependence of the physical quantities of the choice of $`\mathrm{\Gamma }`$ has been thoroughly investigated for holomorphic matrix models in . Since we are interested only in a semiclassical expansion around the critical points, the choice of the contour is irrelevant. We will only assume that one can find such a cycle so that the functional integral is (at least formally) well defined and the usual perturbative techniques are valid.
As always, the functional integral (4) reduces to an integral over (a middle dimensional real cycle in) the moduli space $``$ of critical points of the action modulo gauge transformations. The holomorphic measure of the moduli space integral should be determined in principle by integrating out the massive modes, provided that the original measure $`D\varphi `$ is at least formally well defined.
For the action written down in equation (3), the critical point equations are
$$\overline{}\varphi ^{0,0}=0,\overline{}\varphi ^{1,0}=0.$$
The set of solutions is parameterized by the complex vector space of $`N\times N`$ complex matrices. As explained above in order to formulate the quantum theory we have to specify a real middle dimensional cycle in the space of solutions $`M_N()`$. A convenient choice is to restrict the functional integral to the subspace of hermitian matrices. Therefore the partition function of B-branes on $`\stackrel{~}{X}_{\stackrel{~}{s}}`$ is given by the matrix integral
(5)
$$Z_{\stackrel{~}{s}}=\frac{1}{\text{vol}(U(N))}๐\varphi $$
where $`d\varphi `$ is the linear measure on the space of hermitian $`N\times N`$ matrices. The coefficient $`1/\text{vol}(U(N))`$ is due to a residual $`U(N)`$ gauge symmetry which acts on $`\varphi `$ by conjugation.
Now, recall that our goal is to find an effective description for the dynamics of off-shell fluctuations around a B-brane configuration on $`\stackrel{~}{X}_{\stackrel{~}{m}}`$ specified by the multiplicities $`N_1`$, $`i=1,\mathrm{},n`$. According to , the quantum fluctuations about such a background are governed by the perturbative expansion of a deformed matrix integral of the form
(6)
$$Z_{\stackrel{~}{m}}=\frac{1}{\text{vol}(U(N))}๐\varphi e^{\frac{1}{g_s}W_m(\varphi )}$$
where $`m=\rho (\stackrel{~}{m})`$, and $`W_m(x)`$ is the polynomial function introduced in equation (2). Although this deformation has been recently derived by topological open string computation in , a more geometric treatment is better suited for our purposes.
The main observation is that on a generic threefold $`\stackrel{~}{X}_{\stackrel{~}{m}}`$ there exists a family of transverse holomorphic deformations of the exceptional curves $`C_i`$, $`i=1,\mathrm{},n`$ parameterized by the complex line. Transverse holomorphic deformations of a holomorphic cycle are $`C^{\mathrm{}}`$-deformations of the cycle which depend holomorphically on the deformation parameters. To construct a transverse holomorphic family which includes the $`C_i`$โs, we will use the special geometry of the fibration $`\stackrel{~}{X}_{\stackrel{~}{m}}`$. We constructed $`\stackrel{~}{X}_{\stackrel{~}{m}}`$ as a quasi-projective small resolution of the singular threefold $`X_m`$ defined by $`uv+y^2=(W_m^{}(x))^2`$, given by a generic polynomial $`W_m`$ of degree $`n+1`$. Viewing the polynomial $`W_m^{}(x)`$ as a degree $`n`$ map from $``$ to $``$ we can identify $`X_m`$ with the fiber product
where $`Z`$ is the conifold hypersurface $`uv+y^2=w^2`$ in $`^4`$ with coordinates $`(w,y,u,v)`$. Hence we can obtain the quasi-projective small resolution $`\stackrel{~}{X}_{\stackrel{~}{m}}`$ of $`X_m`$ as the fiber product
with a resolved conifold. Explicitly $`\stackrel{~}{Z}`$ is the total space of the vector bundle $`๐ช(1)๐ช(1)`$ on $`^1`$ and the map $`\stackrel{~}{Z}`$ corresponds to the natural epimorphism of vector bundles $`๐ช(1)๐ช(1)๐ช`$. In this picture for $`\stackrel{~}{X}_{\stackrel{~}{m}}`$, the exceptional curves $`C_i`$, $`i=1,\mathrm{},n`$ are simply the preimages of the zero section $`C\stackrel{~}{Z}`$ of
$$\stackrel{~}{Z}=๐ช(1)๐ช(1)^1.$$
Now, the fibration $`\stackrel{~}{Z}`$ is well known to be the complement of the fiber at infinity of the twistor family for the Taub-NUT hyperkahler metric on the surface $`\stackrel{~}{Y}=\mathrm{tot}(๐ช_^1(2))`$. In particular, the family of twistor lines on $`\stackrel{~}{Z}`$ gives a $`C^{\mathrm{}}`$ trivialization $`\tau :\stackrel{~}{Y}\times \stackrel{~}{}\stackrel{~}{Z}`$ of the fibration $`\stackrel{~}{Z}`$, which is transverse holomorphic by construction. Note that the twistor-line trivialization identifies the surface $`\stackrel{~}{Y}\times \{0\}\stackrel{~}{Y}\times `$ holomorphically with the zero fiber of $`\stackrel{~}{Z}`$. and $`C\stackrel{~}{Y}\stackrel{~}{Z}_0`$ can be viewed either as the zero section of $`๐ช(2)`$ or as the zero section of $`๐ช(1)๐ช(1)`$. The $`\tau `$-image of the holomorphic family
gives a transverse holomorphic family of two spheres
which over $`0`$ specializes to the holomorphic $`(1,1)`$ curve $`C\stackrel{~}{Z}`$. Finally, we can pull back this family to the fiber product $`\stackrel{~}{X}_{\stackrel{~}{m}}`$ to obtain a transverse holomorphic family
of 2-spheres, which is parameterized by $``$ and includes the exceptional curves $`C_i`$.
Given such a family, there is a pure geometric construction for a holomorphic function $`๐ฒ`$ on the parameter space whose critical points coincide with the locations of the holomorphic fibers. Such a function is called a superpotential and is determined by the Abel-Jacobi map associated to the transverse holomorphic family. To construct this function for the present model, pick an arbitrary reference point $`p_0`$ in $``$ and for any point $`p\{p_0\}`$ pick an arbitrary path $`\gamma `$ joining $`p_0`$ and $`p`$. For each path $`\gamma `$ we can construct a canonical three-chain $`\mathrm{\Gamma }`$, $`i=1,\mathrm{},n`$ with boundary
$$\mathrm{\Gamma }=๐_p๐_{p_0}$$
which is swept by the cycles $`๐_q`$ in the family with $`q\gamma `$. Then we have
(7)
$$๐ฒ=_\gamma \mathrm{\Omega }_{\stackrel{~}{X}_{\stackrel{~}{m}}}$$
where $`\mathrm{\Omega }_{\stackrel{~}{X}_{\stackrel{~}{m}}}`$ is a global holomorphic three-form on $`\stackrel{~}{X}_{\stackrel{~}{m}}`$. Usually such a function would be multivalued because of the choices involved, but this complication does not appear in the present example because $`H_3(\stackrel{~}{X}_{\stackrel{~}{m}},)=0`$. Finally, one can check by an explicit computation that $`๐ฒ`$ agrees with $`W_m`$ up to an irrelevant additive constant. This is the geometric interpretation of (6).
Using standard matrix model technology, the matrix integral (6) can be rewritten as an integral over the eigenvalues $`\{\lambda _a\}`$, $`a=1,\mathrm{},N`$ of $`\varphi `$
(8)
$$Z_{\stackrel{~}{m}}=\frac{1}{N!}_^N\underset{a=1}{\overset{N}{}}d\lambda _a\mathrm{\Delta }(\lambda _a)e^{\frac{1}{g_s}_{a=1}^NW_m(\lambda _a)}.$$
where
$$\mathrm{\Delta }(\lambda _a)=\underset{ab}{}(\lambda _a\lambda _b).$$
The critical points of the classical superpotential are in one to one correspondence with partitions of the form
$$N=N_1+N_2+\mathrm{}+N_n.$$
Each such partition corresponds to a distribution of eigenvalues among the $`n`$ zeros of $`W_m`$. For the purpose of large $`N`$ duality we are interested in the perturbative expansion of the matrix integral around such a critical point in the large $`N`$ limit. The duality predicts a precise relation between this expansion and the perturbative expansion of the closed topological string on a smooth threefold $`X_l`$.
Here we will only explain how duality works for genus zero amplitudes. The genus zero amplitudes of the topological open string are captured by the large $`N`$ planar limit of the matrix integral. This means that we take the limits
$$N\mathrm{},g_s0\text{ while keeping }\mu =Ng_s\text{ fixed}.$$
In this limit, the perturbative expansion of the matrix model free energy about a classical vacuum is encoded in a geometric structure associated to the semiclassical equations of motion. The semiclassical equations of motion can be obtained by applying the variational principle to the effective superpotential
$$W^{scl}=\underset{a=1}{\overset{N}{}}W_m(\lambda _a)g_s\text{log}\mathrm{\Delta }(\lambda _a).$$
The distribution of eigenvalues is characterized by the resolvent
$$\omega (x)=\frac{1}{N}\mathrm{Tr}\left(\frac{1}{x\varphi }\right)$$
which is a rational function on the complex plane with poles at the locations of the eigenvalues. In the $`N\mathrm{}`$ limit, the eigenvalues behave as a continuous one dimensional fluid with density $`\rho (\lambda )`$ normalized so that
$$_{}\rho (\lambda )๐\lambda =1.$$
The large $`N`$ limit of the resolvent is
$$\omega _{\mathrm{}}(x)=_{}\frac{\rho (\lambda )d\lambda }{x\lambda }.$$
Then one can show that the semiclassical vacua at large $`N`$ are in one to one correspondence to solutions to the algebraic equation
(9)
$$\omega _{\mathrm{}}^2\frac{1}{\mu }W_m^{}(x)\omega _{\mathrm{}}=g(x)$$
where $`g(x)`$ is a polynomial of degree smaller or equal to $`n1`$. More precisely, for each choice of $`f(x)`$, one can find a large $`N`$ semiclassical vacuum by solving equation (9). The distribution of eigenvalues in such a semiclassical vacuum is supported on a disjoint union of line segments $`\mathrm{\Gamma }_1,\mathrm{},\mathrm{\Gamma }_n`$ centered around the roots of $`W_m(x)`$. Note that equation (9) determines a hyperelliptic curve, which is a double cover of the complex plane with branch points situated at the endpoints of the segments $`\mathrm{\Gamma }_i`$, $`i=1,\mathrm{},n`$. The line segments $`\mathrm{\Gamma }_i`$, $`i=1,\mathrm{},n`$ are branch cuts for the double cover.
Next we claim that the genus zero free energy can be expressed in terms of period integrals of the meromorphic one-form $`\eta `$ on the hyperelliptic curve. Given the branch cuts $`\mathrm{\Gamma }_i`$, $`i=1,\mathrm{},n`$ we can choose $`2(n1)`$ contours $`(A_1,\mathrm{},A_{n1},B_1,\mathrm{},B_{n1})`$ in the complex plane $``$, which give rise to a symplectic basis of cycles on the hyperelliptic curve. The periods of $`\eta `$ on the $`A`$-cycles
$$s^i=\frac{1}{2\pi i}_{A_i}\omega _{\mathrm{}}(x)๐x$$
determines the filling fractions associated to the cuts $`\mathrm{\Gamma }_1,\mathrm{},\mathrm{\Gamma }_{n1}`$. Note that there are only $`n1`$ independent filling fractions since their sum should be 1. According to , the integrals of the same differential $`\omega _{\mathrm{}}(x)dx`$ compute the amount of energy necessary for moving an eigenvalue from the $`i`$-th branch cut $`\mathrm{\Gamma }_i`$, $`i=1,\mathrm{},n1`$ to $`\mathrm{\Gamma }_n`$. Therefore, if we denote by $`^{\text{op}}`$ the free energy of the matrix model, we have
$$\frac{^{\text{op}}}{s^i}=_{B_i}\omega _{\mathrm{}}(x)๐x,i=1,\mathrm{},n1.$$
The main claim of large $`N`$ duality is that the large $`N`$ limit of the open string theory on $`\stackrel{~}{X}_{\stackrel{~}{m}}`$ is equivalent to a closed string theory on a threefold $`X_l`$ of the form (1) for some $`l(x)`$ of the form $`l(x)=f(x)+(W_m^{}(x))^2`$. The above geometric interpretation of the matrix model free energy makes this correspondence very transparent. To the spectral curve of the matrix model (9) we can associate a noncompact Calabi-Yau threefold $`X_{m,f}`$ determined by the equation
(10)
$$uv+y^2(W_m^{}(x))^2=f(x)$$
where
$$y=2\mu \omega _{\mathrm{}}W_m^{}(x),f(x)=4\mu g(x).$$
This threefold is a fibration in quadrics over the complex plane with coordinate $`x`$. The generic fiber is a smooth affine quadric in $`^3`$, while the fibers over the roots of the polynomial $`(W_m^{}(x))^2+f(x)`$ are singular quadrics with isolated $`A_1`$ singularities. As explained above, one can construct a smooth two cycle homeomorphic to a two-sphere on each smooth fiber. This cycle shrinks to a point on the singular fibers. Therefore $`X_{m,f}`$ can be viewed as the total space of a family of affine quadrics which degenerate to singular $`A_1`$ quadrics at finitely many points. The smooth cycles in the smooth fibers are vanishing cycles with respect to the degenerations.
Each contour $`A_i`$ in the complex plane gives rise to closed three-cycle $`S_i`$ which is swept by the two-cycles in the fibers over the points of $`A_i`$. We can perform the same construction for the contours $`B_i`$. In this case the fiber two-cycles shrink at the endpoints of the contour resulting again in a closed three-cycle $`T_i`$ on $`X_{m,f}`$. With a suitable normalization of the global holomorphic three-form on $`X_{m,f}`$, one now gets that
(11)
$$s^i=_{S_i}\mathrm{\Omega }_{X_{m,f}},\frac{^{\text{op}}}{s^i}=_{T_i}\mathrm{\Omega }_{X_{m,f}}$$
for $`i=1,\mathrm{},n1`$. However these are precisely the defining relations for the special geometry holomorphic prepotential associated to the family of smooth Calabi-Yau threefolds $`๐ณ`$. Therefore we can conclude that the open string free energy $`^{\text{op}}`$ can be interpreted in the large $`N`$ planar limit as the genus zero closed string free energy for a $`๐`$-model on the threefold $`X_{m,f}`$.
## 3 Dijkgraaf-Vafa limits of compact Calabi-Yau spaces
In the previous section we have studied large N duality for a special class of extremal transitions among noncompact Calabi-Yau threefolds. The geometric features of those models enabled us to formulate and give a physical proof of the duality conjecture. In this section we will develop a general geometric framework for geometric transitions by generalizing the essential features of the local models. The first important aspect of Dijkgraaf-Vafa transitions which can be taken as a guiding principle for more general geometric situations is a stratified structure of the moduli space which will be described in detail below.
### 3.1 Stratification of moduli
The general setup for geometric transitions consists of the following data. We take $`๐ณ`$ to be a fixed component of the moduli space of Calabi-Yau threefolds. Let $`๐ด`$ be the subvariety of $`๐ณ`$ parameterizing the singular threefolds with a fixed number $`n`$ of isolated ODPs which admit a crepant projective resolution. We will denote by $`l`$, $`m`$ the points of $`๐ณ`$ and respectively $`๐ด`$ and by $`X_l`$, $`X_m`$ the corresponding Calabi-Yau threefolds. For a fixed generic point $`m๐ด`$, $`X_m`$ may have several distinct projective crepant resolutions related by flops. Therefore the moduli space $`\stackrel{~}{๐ด}`$ of the resolution is a finite cover of $`๐ด`$. We denote by $`\rho :\stackrel{~}{๐ด}๐ด`$ the finite to one map which associates to any smooth crepant resolution the singular threefold obtained by contracting the exceptional locus. We will also denote by $`\stackrel{~}{m}`$ a point in $`\stackrel{~}{๐ด}`$ so that $`\rho (\stackrel{~}{m})=m`$. Then the extremal transition can be represented by a diagram of the form
(12)
where $`X_l`$ is a smooth Calabi-Yau space corresponding to a point $`l๐ณ`$, is a degeneration of $`X_l`$ to a Calabi-Yau variety $`X_m`$, $`m๐ด`$ having $`n`$ ordinary double points, and $`\stackrel{~}{X}_{\stackrel{~}{m}}X_m`$, $`\stackrel{~}{m}\stackrel{~}{๐ด}`$ is a crepant quasi-projective resolution of $`X_m`$.
In the previous section, the connection between holomorphic Chern-Simons theory and matrix models was based on the existence of a maximally degenerate point $`s๐ด`$ where the Calabi-Yau threefold develops a curve of $`A_1`$ singularities. In a general situation, we should be looking for a singular subspace $`๐บ๐ด`$ characterized by the property that the Calabi-Yau threefolds parameterized by points $`s๐บ`$ have curves of singularities. It is not known if such a boundary locus of $`๐ณ`$ exists in general, but from now on we will restrict our considerations to components with this property. We will discuss several examples later in this section. Note that the threefolds $`X_s`$, $`s๐บ`$ have a unique projective crepant resolution, as opposed to Calabi-Yau varieties $`X_m`$ parameterized by generic points $`m๐ด`$. Therefore the moduli space $`\stackrel{~}{๐บ}`$ of the resolution is a subspace of the ramification locus of the map $`\rho :\stackrel{~}{๐ด}๐ด`$ isomorphic to $`๐บ`$. Therefore we obtain a stratified structure of the moduli spaces $`๐ณ`$, $`\stackrel{~}{๐ด}`$ described by the following diagram
(13)
$$\begin{array}{ccccc}\stackrel{~}{๐บ}& & \stackrel{~}{๐ด}& & \\ & & & & \\ ๐บ& & ๐ด& & ๐ณ.\end{array}$$
From now on we will refer to extremal transitions with this structure as stratified extremal transitions. Let us now discuss some examples.
### 3.2 Examples of nested moduli
Several examples of stratified extremal transitions among compact Calabi-Yau threefolds have been constructed in the literature . In all these examples, the Calabi-Yau threefolds are hypersurfaces or complete intersections in weighted projective spaces. Other examples can be obtained using the Borcea-Voisin construction.
Here we will discuss in detail one of the complete intersection models. We consider the moduli space $`๐ณ`$ of complete intersections in $`^5`$ of the form
(14)
$$Q_2(z_i)=0,Q_4(z_i)=0$$
where $`Q_2(z_i),Q_4(z_i)`$ are homogeneous polynomials of degree $`2,4`$ in the projective coordinates $`[z_1:z_2:\mathrm{}:z_6]`$. The stratification we are interested is induced by a stratification of the space of quadric polynomials $`Q_2(z_i)`$. Let us denote by $`Z_2`$, $`Z_4`$ the four dimensional quadric and quartic in $`^5`$ cut by the equations (14). If $`Q_2`$ is a quadric of rank three, then $`Z_2`$ will be singular along a linear subspace $`^2^5`$ which intersects a generic $`Z_4`$ along curve $`\mathrm{\Sigma }`$ of genus $`g=3`$. The resulting Calabi-Yau threefold has $`A_1`$ singularities at the points of $`\mathrm{\Sigma }`$. If $`Q_2`$ is a quadric of rank four, $`Z_2`$ will be singular along a line $`^1^4`$, which intersects a generic quartic $`Z_4`$ in four points. Therefore the complete intersection has four isolated ODPs. If $`Q_2`$ is a quadric of rank greater or equal to $`5`$, the generic complete intersection is smooth. Therefore we obtain a stratified moduli space
$$๐บ๐ด๐ณ$$
where $`๐บ`$, $`๐ด`$, $`๐ณ`$ have dimensions $`83,86`$ and respectively $`89`$. The general point in $`๐บ`$, $`๐ด`$, $`๐ณ`$ is a complete intersection of a generic $`Z_4`$ and a $`Z_2`$ defined by a quadric $`Q_2`$ of rank $`3`$, $`4`$, and $`5`$ respectively. For future reference let us write down equations for the generic threefolds in each stratum. Up to automorphisms of $`^5`$, we can write the equation of a rank three quadric in the form
$$z_1^2z_2z_3=0.$$
The singular locus is cut by the equations $`z_1=z_2=z_3=0`$. The quadrics corresponding to points in $`๐ด`$ can be written as
$$z_1^2z_2z_3+Q_1(z_4,z_5,z_6)^2=0$$
where $`Q_1(z_4,z_5,z_6)`$ is a homogeneous polynomial of degree one in $`z_4,z_5,z_6`$. The singular locus is given in this case by $`z_1=z_2=z_3=Q_1(z_4,z_5,z_6)=0`$. The quadrics corresponding to points in $`๐ณ`$ can be written as
$$z_1^2z_2z_3+Q_2(z_4,z_5,z_6)=0$$
where $`Q_2(z_4,z_5,z_6)`$ is a homogeneous polynomial of degree two in $`z_4,z_5,z_6`$ which is not necessarily a perfect square.
Given a stratified extremal transition among compact Calabi-Yau threefolds, one can obtain a similar transition among noncompact spaces by linearization. This is an algebraic-geometric process which embodies the notion of local limit of a Calabi-Yau space often used in the physics literature. For a stratified extremal transition as above, we would like to perform linearization of the singular threefolds $`X_s`$ along the curves of singularities $`\mathrm{\Sigma }`$. Taking into account the complete intersection presentation of these models, we can achieve this goal by linearizing the ambient projective space $`^5`$ along the subspace $`^2^5`$ cut by the equations $`z_1=z_2=z_3=0`$.
Consider the direct product $`\times ^5`$ regarded as a trivial fibration over the complex line, and perform a blow-up along the subspace $`\{0\}\times ^2`$. The resulting space has the structure of a fibration over $``$ in which the central fiber is reducible and consists of two components. One component $`\overline{P}`$ is isomorphic to the total space of the projective bundle
(15)
$$(๐ช(1)^3๐ช)$$
over $`^2`$. The second component is isomorphic to the blow-up of $`^5`$ along $`^2`$. The two components intersect transversely along the section of $`\overline{P}`$ over $`^2`$ defined by the trivial summand in (15). The linearization of $`^5`$ along $`^2`$ is given by the complement $`P`$ of this section in $`\overline{P}`$. By construction, $`P`$ is isomorphic to the total space of the bundle $`๐ช(1)^3`$ over $`^2`$.
Now let us consider a complete intersection $`X_s`$ of the form (14) with a quadric polynomial of the form $`Q_2(z_i)=z_1^2z_2z_3`$. The curve $`\mathrm{\Sigma }`$ of $`A_1`$ singularities on $`X_s`$ is cut by the equations
$$z_1=z_2=z_3=0,Q_4(z_i)=0.$$
Therefore $`\mathrm{\Sigma }`$ is a quartic in the projective plane $`^2^5`$ defined by
$$P_4(z_4,z_4,z_6)Q_4(0,0,0,z_4,z_5,z_6)=0.$$
The linearization of $`X_s`$ along $`\mathrm{\Sigma }`$ can be accordingly described as a complete intersection in $`P`$. Let $`x_1,x_2,x_3`$ denote the canonical coordinates along the fibers of $`P`$, that is tautological sections of $`๐ช(1)^3`$ pulled back to $`P`$. Then the linearization of $`X_s`$ is given by the following equations in $`P`$
(16)
$$x_1^2x_2x_3=0,P_4(z_4,z_5,z_6)=0.$$
Therefore we obtain a moduli space $`๐บ`$ of singular threefolds isomorphic to the space of quartics in $`^2`$. The higher strata $`๐ด,๐ณ`$ can be described as moduli spaces of smoothings of (16) by deforming the quadric equation as explained above.
Note that in this case the linearized threefolds $`X_s`$, $`s๐บ`$ are isomorphic to a direct product of the form $`\mathrm{\Sigma }\times Y`$ where $`Y`$ is the canonical $`A_1`$ surface singularity. The resolution $`\stackrel{~}{X}_s`$ will therefore be isomorphic to a direct product $`\mathrm{\Sigma }\times \stackrel{~}{Y}`$, just as in the previous section. Moreover, the deformations $`X_m,X_l`$, as well as the resolution $`\stackrel{~}{X}_{\stackrel{~}{m}}`$ have the structure of affine quadric bundles over $`\mathrm{\Sigma }`$. Therefore the local transition obtained by linearization is a direct generalization of the Dijkgraaf-Vafa transitions. This motivates calling such models Dijkgraaf-Vafa limits of compact Calabi-Yau spaces. We will construct below a more general class of extremal transitions among noncompact Calabi-Yau threefolds which exhibit the same properties.
## 4 Linear transitions
The local extremal transition obtained in the previous section by linearization is a special case of a general abstract construction which will be the main focus of this section. These transitions will be called linear transitions because the moduli spaces $`\stackrel{~}{๐ด}`$ and $`๐ณ`$ are vector bundles over the bottom stratum $`๐บ`$ as will become clear below. First we present the geometric construction, and then explain the connection between our noncompact Calabi-Yau threefolds and Hitchinโs integrable system.
Let $`\mathrm{\Sigma }`$ denote a smooth projective curve of genus $`g2`$. Take $`๐บ`$ to be the moduli space of pairs $`(\mathrm{\Sigma },V)`$, where $`V\mathrm{\Sigma }`$ is a semi-stable rank two bundle equipped with a fixed isomorphism $`\mathrm{\Lambda }^2VK_\mathrm{\Sigma }`$. For each such pair $`s=(\mathrm{\Sigma },V)`$, we can construct a singular Calabi-Yau threefold $`X_s`$ as follows. Let $`T_s`$ be the total space of $`V`$ and let $`\xi _s:T_sT_s`$ be the holomorphic involution acting by multiplication by $`(1)`$ on each fiber. Then we take $`X_s`$ to be the quotient $`T_s/(\xi _s)`$, where $`(\xi _s)`$ denotes the finite group with two elements generated by $`\{1,\xi _s\}`$. Note that $`\xi _s`$ fixes the zero section of $`V\mathrm{\Sigma }`$, therefore $`X_s`$ has a curve of $`A_1`$ singularities isomorphic to $`\mathrm{\Sigma }`$. Blowing up $`X_s`$ along its singular locus $`\mathrm{\Sigma }X_s`$ yields a canonical quasi-projective crepant resolution $`\stackrel{~}{X}_sX_s`$ of the singularities of $`X_s`$. The exceptional locus $`S_s`$ of the map $`\stackrel{~}{X}_sX_s`$ is the projectivization of the normal cone of $`\mathrm{\Sigma }X_s`$ and so can be identified with the geometrically ruled surface $`(V)`$. For future reference, note that $`\stackrel{~}{X}_s`$ can be naturally identified with the total space of a line bundle over $`S_s`$, so that the exceptional divisor $`S_s\stackrel{~}{X}_s`$ becomes the zero section of this line bundle. Indeed, this follows by noting the blow-up of the vertex of a two dimensional affine quadric is the total space of $`๐ช_^1(2)`$. Explicitly we get $`\stackrel{~}{X}_s\mathrm{Tot}(๐ช_{S_s}(2))`$.
In order to describe $`๐ด`$ and $`๐ณ`$ we have first to understand the deformation theory of a singular threefold $`X_s`$ as well as the deformation theory of its resolution $`\stackrel{~}{X}_s`$.
### 4.1 Deformations of singular threefolds
For simplicity we will drop the subscript $`s`$ in this subsection, denoting a singular threefold by $`X`$. The small resolution $`\stackrel{~}{X}_s`$ will be denoted by $`\stackrel{~}{X}`$. The space of infinitesimal deformations of $`X`$ is $`\text{Ext}_X^1(\mathrm{\Omega }_X^1,๐ช_X)`$. The local to global spectral sequence yields the following exact sequence
(17) $`0`$ $`H^1(\underset{ยฏ}{\mathrm{Ext}}_X^0(\mathrm{\Omega }_X^1,๐ช_X))\text{Ext}^1(\mathrm{\Omega }_X^1,๐ช_X)H^0(\underset{ยฏ}{\mathrm{Ext}}_X^1(\mathrm{\Omega }_X^1,๐ช_X))`$
$`H^2(\underset{ยฏ}{\mathrm{Ext}}^0(\mathrm{\Omega }_X^1,๐ช_X))`$
The first term of this sequence $`H^1(\underset{ยฏ}{\mathrm{Ext}}_X^0(\mathrm{\Omega }_X^1,๐ช_X))`$ parameterizes equisingular deformations of $`X`$, therefore it is isomorphic to the tangent space $`T_s๐บ`$ to the bottom stratum $`๐บ`$. Our main interest in this section is in the quotient space
(18)
$$\frac{\text{Ext}^1(\mathrm{\Omega }_X^1,๐ช_X)}{H^1(\underset{ยฏ}{\mathrm{Ext}}_X^0(\mathrm{\Omega }_X^1,๐ช_X))}\text{ker}\left(H^0(\underset{ยฏ}{\mathrm{Ext}}_X^1(\mathrm{\Omega }_X^1,๐ช_X))H^2(\underset{ยฏ}{\mathrm{Ext}}^0(\mathrm{\Omega }_X^1,๐ช_X))\right)$$
which parameterizes general deformations of $`X_s`$ modulo equisingular deformations.
We claim that for the singular threefolds $`X=T/(\xi )`$ constructed above, this space is isomorphic to the space of holomorphic quadratic differentials on $`\mathrm{\Sigma }`$
$$H^0(\mathrm{\Sigma },K_\mathrm{\Sigma }^2).$$
In order to prove this claim, it will be useful to find a presentation of $`X`$ as a singular hypersurface in the total space of a holomorphic bundle over $`\mathrm{\Sigma }`$. We will first explain this construction in a simplified situation when $`V`$ is complex vector space of dimension two and $`\xi `$ is a holomorphic involution acting on $`V`$ by multiplication by $`1`$. In this case $`V/(\xi )`$ can be realized as a hypersurface in the complex vector space $`W=S^2(V)`$. The map $`V/(\xi )W`$ is given by the $`\xi `$-invariant polynomials on $`V`$. Explicitly we can describe this map as follows.
The second symmetric tensor power $`S^2(V)`$ is a subspace of $`\text{Hom}(V^{},V)`$. Given a linear map $`\varphi :V^{}V`$, there is an induced determinant map
$$\text{det}(\varphi ):=^2\varphi \mathrm{\Lambda }^2(V^{})\mathrm{\Lambda }^2(V).$$
Let us write an arbitrary element of $`S^2(V)`$ in the form
$$\varphi =ue_1e_1+v(e_1e_2+e_2e_1)+we_2e_2$$
where $`\{e_1,e_2\}`$ is a basis of $`V`$. Let $`\{f^1,f^2\}`$ be the dual basis of $`V^{}`$. Then a straightforward linear algebra computation shows that
$$\text{det}(\varphi )(f^1f^2)=(uwv^2)e_1e_2.$$
Therefore the hypersurface
$$\text{det}(\varphi )=0$$
is isomorphic to the canonical $`A_1`$ singularity $`V/(\xi )`$.
If $`V`$ is a rank two bundle over $`\mathrm{\Sigma }`$, we can perform this construction fiberwise, obtaining a hypersurface in the total space $`W`$ of the second symmetric power $`S^2(V)`$. Let us denote by $`\pi _W:W\mathrm{\Sigma }`$ the projection map. Then the above computation shows that there is a tautological section $`det_W`$ of the pull-back bundle
$$\pi _W^{}\underset{ยฏ}{\mathrm{Hom}}_\mathrm{\Sigma }(\mathrm{\Lambda }^2V^{},\mathrm{\Lambda }^2V)=\pi _W^{}(\mathrm{\Lambda }^2V)^2\pi _W^{}K_\mathrm{\Sigma }^2$$
to the total space $`W`$. The zero locus of this section is a singular hypersurface $`X`$ in $`W`$ isomorphic to $`V/(\xi )`$. The singular locus of $`X`$ is isomorphic to $`\mathrm{\Sigma }`$.
Using this presentation of $`X`$, we can compute the space of normal deformations (18). We have an exact sequence
$$๐ฅ_X/๐ฅ_X^2\mathrm{\Omega }_W^1_{๐ช_W}๐ช_X\mathrm{\Omega }_X^10.$$
Since $`X`$ is a Cartier divisor in $`W`$, it follows that the first map in this sequence is injective. Therefore we obtain an exact sequence
(19)
$$0๐ช_W(X)_{|X}\mathrm{\Omega }_{W|X}^1\mathrm{\Omega }_X^10.$$
Thus the complex $`[๐ช_W(X)_{|X}\mathrm{\Omega }_{W|X}^1]`$ is a resolution of the sheaf of Kรคhler differentials $`\mathrm{\Omega }_X^1`$ by locally free $`๐ช_X`$-modules. The Ext sheaves $`\underset{ยฏ}{\mathrm{Ext}}_X^i(\mathrm{\Omega }_X^1,๐ช_X)`$ are the sheaf cohomology groups of the dual complex
(20)
$$\mathrm{\Theta }_{W|X}\stackrel{f}{}๐ช_W(X)_{|X}$$
obtained from $`[๐ช_W(X)_{|X}\mathrm{\Omega }_{W|X}^1]`$ by applying the functor $`\underset{ยฏ}{\mathrm{Hom}}_X(,๐ช_X)`$. Here $`\mathrm{\Theta }_W:=\underset{ยฏ}{\mathrm{Hom}}(\mathrm{\Omega }_W^1,๐ช_W)`$ denotes the sheaf of holomorphic tangent vectors to $`W`$. Computing locally we see that the first local Ext sheaf ($`=`$ the sheaf cokernel of $`f`$) is given by
(21)
$$\underset{ยฏ}{\mathrm{Ext}}_X^1(\mathrm{\Omega }_X^1,๐ช_X)๐ช_X(X)_{|\mathrm{\Sigma }}.$$
By construction,
$$๐ช_X(X)_{|\mathrm{\Sigma }}K_\mathrm{\Sigma }^2$$
therefore we find that
(22)
$$H^0(X,\underset{ยฏ}{\mathrm{Ext}}_X^1(\mathrm{\Omega }_X^1,๐ช_X))H^0(\mathrm{\Sigma },K_\mathrm{\Sigma }^2)$$
is the space of holomorphic quadratic differentials on $`\mathrm{\Sigma }`$.
In order to finish the computation, we have to determine the kernel of the map
(23)
$$H^0(X,\underset{ยฏ}{\mathrm{Ext}}_X^1(\mathrm{\Omega }_X^1,๐ช_X))H^2(X,\underset{ยฏ}{\mathrm{Ext}}_X^0(\mathrm{\Omega }_X^1,๐ช_X)).$$
Note that the projection $`\pi :X\mathrm{\Sigma }`$ is an affine map, therefore we have
$$H^2(X,)=H^2(\mathrm{\Sigma },\pi _{})=0$$
for any coherent sheaf $``$ on $`X`$. Therefore the map (23) is trivial, and we obtain the desired result.
### 4.2 Deformations of the resolution
We now turn to the deformation theory of the resolution $`\stackrel{~}{X}_s`$. By analogy with the previous subsection, our aim is to understand all deformations of $`\stackrel{~}{X}_s`$ modulo the deformations of the pair $`(\stackrel{~}{X}_s,S)`$. We will keep using the notation of the previous paragraph omitting the subscript $`s`$.
The infinitesimal deformations of the pair $`(\stackrel{~}{X},S)`$ are parameterized by the first hypercohomology group of the two term complex
$$๐ฏ:\mathrm{\Theta }_{\stackrel{~}{X}}i_SN_{S/\stackrel{~}{X}}$$
where the first term is placed in degree zero and $`i_S:S\stackrel{~}{X}`$ is the natural inclusion. The hypercohomology spectral sequence reduces to an exact sequence which reads in part
By construction, the normal bundle $`N_{S/\stackrel{~}{X}}`$ is isomorphic to the canonical bundle $`K_S`$ of $`S`$. Since $`S=(V)=\mathrm{Proj}(S^{}V^{})`$ is a projective bundle over $`\mathrm{\Sigma }`$, we have
(24) $`K_S`$ $`=q^{}K_\mathrm{\Sigma }\omega _{S/\mathrm{\Sigma }}`$
$`q^{}(K_\mathrm{\Sigma }^2V^{})๐ช_S(2)`$
$`=๐ช_S(2)`$
where $`๐ช_S(1)`$ denotes the relative hyperplane bundle of $`S`$ over $`\mathrm{\Sigma }`$, and we have used the relation \[17, Chapter III.8, exercise 8.4\]
$$\omega _{S/\mathrm{\Sigma }}q^{}\mathrm{\Lambda }^2V^{}๐ช_S(2).$$
This shows that
$$H^0(\stackrel{~}{X},N_{S/\stackrel{~}{X}})H^0(S,K_S)=0.$$
In addition, it is not hard to check that the map
(25)
$$^2(\stackrel{~}{X},๐ฏ)H^2(\stackrel{~}{X},\mathrm{\Theta }_{\stackrel{~}{X}})$$
is injective. Indeed, by definition $`\mathrm{\Theta }_{\stackrel{~}{X}}i_SN_{S/\stackrel{~}{X}}`$ is surjective. The kernel sheaf $`\mathrm{\Theta }_{X,S}=\mathrm{ker}[\mathrm{\Theta }_{\stackrel{~}{X}}i_SN_{S/\stackrel{~}{X}}]`$ is locally free and is the sheaf of germs of vector fields on $`\stackrel{~}{X}`$ that at the points of $`S`$ are tangent to $`S`$. Write $`a:XS`$ for the natural projection. As explained at the beginning of section 4, the resolution $`\stackrel{~}{X}`$ of $`X`$ is the total space of the line bundle $`๐ช_S(2)K_SN_{S/\stackrel{~}{X}}`$ on $`S`$. In particular, the vertical tangent bundle $`\mathrm{\Theta }_{\stackrel{~}{X}/S}`$ can be identified with the line bundle $`a^{}K_S`$ and so the tangent sequence for the map $`a:\stackrel{~}{X}S`$ reads:
(26)
$$0a^{}K_S\mathrm{\Theta }_{\stackrel{~}{X}}\stackrel{da}{}a^{}T_S0.$$
Combining (26) with the normal-to-$`S`$ sequence
$$0\mathrm{\Theta }_{\stackrel{~}{X},S}\mathrm{\Theta }_Xi_SN_{S/\stackrel{~}{X}}0$$
we obtain a commutative diagram with exact rows and columns:
(27)
Since $`S\stackrel{~}{X}=\mathrm{Tot}(K_S)`$ is identified with the zero section of $`K_S`$, it follows that the tautological section of $`a_{}K_S`$ vanishes precisely at $`S`$, and so $`a^{}K_S๐ช_{\stackrel{~}{X}}(S)`$. Consequently, the diagram (27) becomes
Looking at the long exact sequences in cohomology for the first two rows we get
and so the map $`^2(๐ฏ)=H^2(\mathrm{\Theta }_{\stackrel{~}{X},S})H^2(\mathrm{\Theta }_{\stackrel{~}{X}})`$ will be injective if the map $`H^2(๐ช_{\stackrel{~}{X}})H^2(a^{}K_S)`$ is injective. Since $`a:\stackrel{~}{X}S`$ is an affine map we have
$$\begin{array}{cc}\hfill H^2(\stackrel{~}{X},๐ช_{\stackrel{~}{X}})& =H^2(S,a_{}๐ช_{\stackrel{~}{X}})=H^2(S,_{i0}K_S^i)\hfill \\ \hfill H^2(\stackrel{~}{X},a^{}K_S)& =H^2(S,a_{}a^{}K_S)=H^2(S,_{i1}K_S^i).\hfill \end{array}$$
The map $`H^2(๐ช_{\stackrel{~}{X}})H^2(a^{}K_S)`$ is induced from the obvious inclusion of sheaves $`_{i0}K_S^i_{i1}K_S^i`$ and so we have an exact sequence
$$H^1(S,_{i1}K_S^i)H^1(S,K_S)H^2(๐ช_{\stackrel{~}{X}})H^2(a^{}K_S).$$
Since the map $`H^1(S,_{i1}K_S^i)H^1(S,K_S)`$ is induced from the projection $`K_S(_{i0}K_S^i)K_S`$, it is clearly surjective and hence $`H^2(๐ช_{\stackrel{~}{X}})H^2(a^{}K_S)`$ must be injective. This implies the injectivity of $`^2(๐ฏ)H^2(\mathrm{\Theta }_{\stackrel{~}{X}})`$ and so we get a short exact sequence of the form
(28)
$$0^1(๐ฏ)H^1(\stackrel{~}{X},\mathrm{\Theta }_{\stackrel{~}{X}})H^1(\stackrel{~}{X},N_{S/\stackrel{~}{X}})0.$$
The first term in (28) parameterizes infinitesimal deformations of the pair $`(\stackrel{~}{X},S)`$, hence it is isomorphic to the tangent space $`T_s\stackrel{~}{๐บ}`$ to $`\stackrel{~}{๐บ}`$ at $`s`$. The middle term parameterizes infinitesimal deformations of $`\stackrel{~}{X}`$ regardless of the behavior of $`S`$. Therefore the quotient
(29)
$$\frac{H^1(\stackrel{~}{X},\mathrm{\Theta }_{\stackrel{~}{X}})}{^1(๐ฏ)}H^1(\stackrel{~}{X},i_SN_{S/\stackrel{~}{X}})$$
parameterizes infinitesimal deformations of $`\stackrel{~}{X}`$ not preserving $`S`$ modulo deformations of the pair$`(\stackrel{~}{X},S)`$. Using the fact that $`K_S=๐ช_S(2)`$ and the Leray spectral sequence, we can easily compute
(30)
$$H^1(\stackrel{~}{X},N_{S/\stackrel{~}{X}})H^1(S,K_S)H^0(\mathrm{\Sigma },K_\mathrm{\Sigma }).$$
### 4.3 Higher strata
Let us now construct the higher strata $`\stackrel{~}{๐ด},๐ณ`$. Note that assuming these spaces exist, the quotient spaces (18), (29) are isomorphic to the fibers of the normal bundle $`N_{๐บ/๐ณ}`$ and respectively $`N_{\stackrel{~}{๐บ}/\stackrel{~}{๐ด}}`$ at $`s`$. In this section we will show that these infinitesimal normal deformations can be integrated to finite linear deformations. More precisely, we will construct families of noncompact Calabi-Yau manifolds parameterized by
$$๐ณ=๐บ\times H^0(\mathrm{\Sigma },K_\mathrm{\Sigma }^2),\stackrel{~}{๐ด}=\stackrel{~}{๐บ}\times H^0(\mathrm{\Sigma },K_\mathrm{\Sigma })$$
together with a quadratic map $`\mathrm{\Pi }:\stackrel{~}{๐ด}๐ณ`$ which form a diagram of the form (13).
We start with the construction of $`\stackrel{~}{๐ด}`$. The main observation here is that the space $`H^1(S,K_S)=H^0(\mathrm{\Sigma },K_\mathrm{\Sigma })`$ parameterizes deformations of the canonical bundle of $`S`$ as an affine bundle over $`S`$. For a fixed $`S`$ there is a linear family of such deformations which can be constructed synthetically as follows.
An element $`\alpha H^0(\mathrm{\Sigma },K_\mathrm{\Sigma })=H^1(S,K_S)\text{Ext}^1(๐ช_S,K_S)`$ determines (up to isomorphism) an extension
(31)
$$0K_SE_\alpha ๐ช_S0,$$
where $`E_\alpha `$ is a locally free sheaf of rank two on $`S`$. Let $`\overline{X}_\alpha `$ be the total space of the $`^1`$ bundle $`(E_\alpha )`$ over $`S`$, and let $`\stackrel{~}{X}_\alpha `$ be the complement of the infinity section $`H_\alpha :=(K_S)(E_\alpha )`$. Then $`\stackrel{~}{X}_\alpha `$ is an affine bundle over $`S`$, or more formally an $`K_S`$-torsor over $`S`$. The threefolds $`\stackrel{~}{X}_\alpha `$ form a linear family
$$\stackrel{~}{๐ณ}\stackrel{~}{๐ด}_s,\stackrel{~}{๐ด}_s:=H^0(\mathrm{\Sigma },K_\mathrm{\Sigma }).$$
of noncompact Calabi-Yau manifolds.
For future reference let us summarize some elementary geometric properties of the generic fiber $`\stackrel{~}{X}_\alpha `$. Assume that $`\alpha `$ has distinct simple zeroes. Let $`\pi _\alpha :E_\alpha \mathrm{\Sigma }`$ denote the projection map to $`\mathrm{\Sigma }`$. By construction, the restriction of $`E_\alpha `$ to a generic fiber $`S_p`$, $`p\text{div}(\alpha )`$ is the unique (up to isomorphism) nontrivial extension
$$0๐ช(2)๐ช(1)๐ช(1)๐ช0$$
of $`๐ช`$ by $`๐ช(2)`$ over $`^1`$. The restriction of $`E_\alpha `$ to a special fiber $`S_p`$ of the ruling with $`p\text{div}(\alpha )`$ is the trivial extension
$$0๐ช(2)๐ช(2)๐ช๐ช0$$
Therefore the generic fibers of $`\pi _\alpha :E_\alpha \mathrm{\Sigma }`$ are isomorphic to the total space of the rank two bundle
$$๐ช(1)๐ช(1)^1,$$
whereas the special fibers are isomorphic to the total space of the rank two bundle
$$๐ช(2)๐ช^1.$$
It follows that the projective bundle $`(E_\alpha )`$ is a projective quadric fibration $`\overline{\pi }_\alpha :(E_\alpha )\mathrm{\Sigma }`$ with generic fibers isomorphic to $`๐ฝ_0=^1\times ^1`$ and special fibers isomorphic to the Hirzebruch surface $`๐ฝ_2`$.
Recall that the noncompact threefold $`\stackrel{~}{X}_\alpha `$ is the complement in $`\overline{X}_\alpha `$ of the section at infinity $`H_\alpha =(K_S)`$. From the point of view of the fibration structure over $`\mathrm{\Sigma }`$, $`H_\alpha `$ intersects the generic $`๐ฝ_0`$ fiber of $`\overline{\pi }_\alpha :(E_\alpha )\mathrm{\Sigma }`$ along a $`(1,1)`$ curve and the special $`๐ฝ^2`$ fibers along a section of $`๐ฝ^2^1`$ of self-intersection $`+2`$. Therefore the noncompact threefold $`\stackrel{~}{X}_\alpha `$ contains $`2g2`$ projective rational curves $`C_1,\mathrm{},C_{2g2}`$ which can be identified with the negative sections of the special $`๐ฝ_2`$ fibers. A straightforward local computation confirms that each of these curves is a $`(1,1)`$ curve on $`\stackrel{~}{X}_\alpha `$. Therefore for generic $`\alpha `$, $`\stackrel{~}{X}_\alpha `$ contains exactly $`2g2`$ isolated $`(1,1)`$ curves as expected.
If $`\alpha `$ is non generic, that is it has multiple zeroes, a similar analysis shows that $`\stackrel{~}{X}_\alpha `$ contains a projective rational curve for each zero of $`\alpha `$. However the curve corresponding to a double zero is a rigid $`(0,2)`$ curve as opposed to a $`(1,1)`$ curve as in the generic case.
Next we will show that one can contract the exceptional curves constructed above on each $`\stackrel{~}{X}_\alpha `$ obtaining a singular threefold $`X_{\alpha ^2}`$ which depends only on $`\alpha ^2H^0(\mathrm{\Sigma },K_\mathrm{\Sigma }^2)`$. We claim that there exists a $`^2`$-bundle $`\overline{\pi }_W:\overline{W}\mathrm{\Sigma }`$, so that for any $`\alpha H^0(\mathrm{\Sigma },K_\mathrm{\Sigma })`$ there exists a canonical map $`\varphi _\alpha :\overline{X}_\alpha \overline{W}`$ which contracts the exceptional curves. The image $`\varphi _\alpha (\overline{X}_\alpha )`$ is a singular hypersurface $`\overline{X}_{\alpha ^2}`$ in $`\overline{W}`$ depending only on $`\alpha ^2`$ sitting in a fixed (independent of $`\alpha `$) linear system on $`\overline{W}`$.
Moreover, there is a preferred hyperplane at infinity $`h_{\mathrm{}}`$ in $`\overline{W}`$ so that the restriction
$$\varphi _\alpha |_{\stackrel{~}{X}_\alpha }:\stackrel{~}{X}_\alpha X_{\alpha ^2}$$
is a small contraction of $`\stackrel{~}{X}_\alpha `$ onto the noncompact nodal Calabi-Yau threefold
$$X_{\alpha ^2}=\overline{X}_{\alpha ^2}\left(\overline{X}_{\alpha ^2}h_{\mathrm{}}\right).$$
To prove this claim, take $`\overline{W}`$ to be the $`^3`$-bundle $`(S^2V๐ช_\mathrm{\Sigma })`$ over $`\mathrm{\Sigma }`$, and take $`h_{\mathrm{}}`$ to be the hyperplane $`(S^2V)`$. In order to construct the map $`\varphi _\alpha :\overline{X}_\alpha \overline{W}`$ it suffices to exhibit a line bundle $`\xi _\alpha `$ on $`\overline{X}_\alpha `$ so that
$$\overline{\pi }_\alpha \xi _\alpha \left(S^2V๐ช_\mathrm{\Sigma }\right)^{}=S^2(V^{})๐ช_\mathrm{\Sigma }.$$
Let $`\xi _\alpha `$ be the relative hyperplane bundle $`\xi _\alpha =๐ช_{\overline{X}_\alpha }(H_\alpha )`$ for $`r_\alpha :\overline{X}_\alpha S`$. The restriction of $`\xi _\alpha `$ to each fiber of $`r_\alpha :\overline{X}_\alpha S`$ is isomorphic to $`๐ช_^1(1)`$ and $`r_\alpha \xi _\alpha E_\alpha ^{}`$. Moreover, the restriction of $`\xi _\alpha `$ to a generic $`๐ฝ_0`$ fiber of the quadric fibration $`\overline{\pi }_\alpha :\overline{X}_\alpha \mathrm{\Sigma }`$ is isomorphic to $`๐ช_{๐ฝ_0}(1,1)`$ whereas the restriction to a special $`๐ฝ_2`$ fiber is isomorphic to $`๐ช_{๐ฝ_2}(\mathrm{\Delta }_0)`$ where $`\mathrm{\Delta }_0`$ denotes the zero section on $`๐ฝ_2`$, $`\mathrm{\Delta }_0^2=2`$.
Let us compute
$$\overline{\pi }_\alpha \xi _\alpha =q_{}(r_\alpha \xi _\alpha )q_{}E_\alpha ^{}.$$
By construction, $`E_\alpha ^{}`$ is an extension of the form
$$0๐ช_SE_\alpha ^{}K_S^10$$
on $`S`$. Taking direct images, we find the following extension on $`\mathrm{\Sigma }`$
$$0๐ช_\mathrm{\Sigma }q_{}E_\alpha ^{}q_{}(K_S^1)0$$
where we have used $`R^1q_{}๐ช_S=0`$. Now $`K_S๐ช_S(2)`$, hence $`q_{}K_S^1S^2V^{}`$. Therefore we have an extension of the form
(32)
$$0๐ช_\mathrm{\Sigma }q_{}E_\alpha ^{}S^2V^{}0$$
on $`\mathrm{\Sigma }`$. In order to construct a contraction map $`\varphi _\alpha `$ we have to prove that this extension splits and we also have to choose a splitting.
First note that if $`V`$ is a stable rank two bundle on $`\mathrm{\Sigma }`$,
$$\text{Ext}^1(๐ช_\mathrm{\Sigma },S^2V)H^0(\mathrm{\Sigma },S^2V^{}K_\mathrm{\Sigma })^{}=0$$
since $`S^2V^{}K_\mathrm{\Sigma }`$ is a stable bundle with trivial determinant. Therefore in that case, the extension (32) splits. If $`V`$ is semistable, but not stable, the extension group $`\text{Ext}^1(๐ช_\mathrm{\Sigma },S^2V)`$ is not necessarily trivial. However, we claim that the extension (32) is still split for an arbitrary semistable bundle $`V`$. This claim is equivalent to the statement that the pushforward map
$$q_{}:H^1(S,K_S)H^1(\mathrm{\Sigma },(q_{}K_S^1)^{})$$
is trivial. Using Serre duality and respectively relative Serre duality for the map $`q:S\mathrm{\Sigma }`$, we obtain a dual map
$$q_{}^{}:H^0(\mathrm{\Sigma },q_{}(K_{S/\mathrm{\Sigma }}^1))H^1(S,๐ช_S)H^1(\mathrm{\Sigma },๐ช_\mathrm{\Sigma })$$
where $`K_{S/\mathrm{\Sigma }}`$ is the relative dualizing sheaf for $`q:S\mathrm{\Sigma }`$. This map is the connecting homomorphism for the short exact sequence of sheaves
(33)
$$0๐ช_\mathrm{\Sigma }q_{}\left(q^{}V๐ช_S(1)\right)q_{}(K_{S/\mathrm{\Sigma }}^1)0.$$
obtained by pushing forward the dual of the relative Euler sequence on $`S`$. We can easily compute the terms in the above exact sequence obtaining
$$0๐ช_\mathrm{\Sigma }V^{}V\text{End}_0(V)0$$
where $`\text{End}_0(V)`$ is the bundle of traceless endomorphisms of $`V`$. However this sequence is canonically split, hence the connecting homomorphism vanishes. Therefore we can conclude that the extension (32) is split. Choosing a splitting we obtain a (non-canonical) isomorphism
$$q_{}(E_\alpha ^{})S^2V๐ช_\mathrm{\Sigma }$$
which defines a map $`\varphi _\alpha :\overline{X}_\alpha \overline{W}`$ as claimed above. Since the divisor $`\overline{H}_\alpha `$ does not intersect the exceptional curves on $`\overline{X}_\alpha `$, it follows that these curves are contracted by $`\varphi _\alpha `$. Therefore $`\varphi _\alpha `$ maps $`\overline{X}_\alpha `$ onto a nodal hypersurface in $`\overline{W}`$.
Next, we will show that the image of $`\varphi _\alpha `$ moves in a fixed linear system on $`\overline{W}`$. Let us fix an $`\alpha H^0(\mathrm{\Sigma },K_\mathrm{\Sigma })`$. By construction, the image of $`\varphi _\alpha `$ must be the zero locus of a section of a line bundle $``$ on $`\overline{W}`$. Since $`\varphi _\alpha :\overline{X}_\alpha \overline{W}`$ is a small contraction, we can apply the adjunction formula obtaining
(34)
$$K_{\overline{X}_\alpha }=\varphi _\alpha ^{}(K_{\overline{W}}).$$
A routine computation using the relative Euler sequence yields
(35)
$$K_{\overline{X}_\alpha }\xi _\alpha ^2,K_{\overline{W}}๐ช_{\overline{W}}(4)\overline{\pi }_W^{}K_\mathrm{\Sigma }^2.$$
By direct substitution in (34), we obtain
$$\xi _\alpha ^2\xi _\alpha ^4\overline{\pi }_\alpha ^{}K_\mathrm{\Sigma }^2\varphi _\alpha ^{}.$$
Since this equation is valid for any value of $`\alpha `$ it follows that
(36)
$$๐ช_{\overline{W}}(2)\overline{\pi }_W^{}K_\mathrm{\Sigma }^2$$
Therefore we can conclude that the hypersurfaces $`\varphi _\alpha (\overline{X}_\alpha )`$ belong to the linear system $`|2h_{\mathrm{}}+\overline{\pi }_W^{}K_\mathrm{\Sigma }^2|`$ for any $`\alpha `$.
Let us compute the space of global sections of $``$. We have
$`\overline{\pi }_{\overline{W}}๐ช_{\overline{W}}(2)`$ $`=S^2(๐ช_\mathrm{\Sigma }S^2V^{})`$
$`=๐ช_\mathrm{\Sigma }S^2V^{}S^2(S^2V^{})`$
$`=๐ช_\mathrm{\Sigma }S^2V^{}S^2(\mathrm{\Lambda }^2V^{})S^4V^{}`$
hence, using the isomorphisms $`\mathrm{\Lambda }^2VK_\mathrm{\Sigma }`$, and $`S^4V^{}K_\mathrm{\Sigma }^2S^4VK_\mathrm{\Sigma }^2`$, we find
(37) $`H^0(\overline{W},๐ช_{\overline{W}}(2)`$ $`\overline{\pi }_W^{}K_\mathrm{\Sigma }^2)=`$
$`H^0(\mathrm{\Sigma },๐ช_\mathrm{\Sigma })H^0(\mathrm{\Sigma },K_\mathrm{\Sigma }^2)H^0(\mathrm{\Sigma },S^2V)H^0(\mathrm{\Sigma },S^4VK_\mathrm{\Sigma }^2).`$
Let us interpret the terms in the right hand side of equation (37). By construction, a non-zero element in $`H^0(\mathrm{\Sigma },๐ช_\mathrm{\Sigma })`$ viewed as a section in $`๐ช_{\overline{W}}(2)\overline{\pi }_W^{}K_\mathrm{\Sigma }^2`$ will vanish exactly along the image $`\varphi _0(\overline{X})`$ of the undeformed threefold $`\overline{X}`$ in $`\overline{W}`$. This is the projective completion of the singular affine hypersurface $`XW`$ constructed in section 4.1 as the zero locus of the determinant $`det_WH^0(W,\pi _W^{}K_\mathrm{\Sigma }^2)`$. We will denote by $`det_{\overline{W}}`$ the section of $``$ whose zero locus is $`\overline{X}`$. The second term parameterizes hypersurface deformations of $`\overline{X}`$ of the form
$$\underset{\overline{W}}{det}\beta =0$$
where $`\beta H^0(\mathrm{\Sigma },๐ช_\mathrm{\Sigma }(2K_\mathrm{\Sigma }))`$ is a quadratic differential on $`\mathrm{\Sigma }`$. By construction, the defining equation of the hypersurface $`\varphi _\alpha (\overline{X}_\alpha )`$ is in the affine subspace
(38)
$$H^0(\mathrm{\Sigma },๐ช_\mathrm{\Sigma })\{\alpha ^2\}H^0(\mathrm{\Sigma },S^2V).$$
All divisors in this affine subspace are isomorphic since any two divisors are related by a global automorphism of $`W`$ which is a translation by a section in $`H^0(\mathrm{\Sigma },S^2V)`$. In fact the affine space (38) also parameterizes the choice of a splitting of the exact sequence (32). Therefore each point $`a`$ in this affine space represents the image of $`\overline{X}_\alpha `$ through a map $`\varphi _{\alpha ,a}`$ which depends on the choice of the splitting. In particular for some point $`a`$ we will obtain a hypersurface in $`\overline{W}`$ with defining equation
$$\underset{\overline{W}}{det}\alpha ^2=0$$
We will denote this hypersurface by $`\overline{X}_{\alpha ^2}`$. Finally, the last term in the right hand side of the decomposition (37) corresponds other deformations of $`\overline{X}`$ which will not be considered in this paper (note that for stable $`V`$ these extra deformations vanish).
This gives rise to the following picture. For a fixed point $`s๐บ\stackrel{~}{๐บ}`$ we obtain a linear deformation space $`๐ณ_s=H^0(\mathrm{\Sigma },K_\mathrm{\Sigma }^2)`$ parameterizing noncompact Calabi-Yau threefolds $`X_\beta `$ determined by equations of the form
(39)
$$\underset{W}{det}\pi _W^{}\beta =0$$
We also have a linear deformation space $`\stackrel{~}{๐ด}_s=H^0(\mathrm{\Sigma },K_\mathrm{\Sigma })`$ parameterizing the threefolds $`\stackrel{~}{X}_\alpha `$. Moreover we have a quadratic map
(40)
$$\mathrm{\Pi }_s:\stackrel{~}{๐ด}_s๐ณ,\alpha \alpha ^2$$
which corresponds to a small contraction of $`\stackrel{~}{X}_\alpha `$. The image of this quadratic map is a singular subvariety in $`H^0(\mathrm{\Sigma },K_\mathrm{\Sigma }^2)`$ isomorphic to $`H^0(\mathrm{\Sigma },K_\mathrm{\Sigma })/(\pm 1)`$. Note that $`๐ด_s,๐ณ_s,\stackrel{~}{๐ด}_s`$ do not depend on the point $`s๐บ`$.
Then we can construct the higher strata $`๐ด,๐ณ,\stackrel{~}{๐ด}`$ as direct products
(41) $`๐ด=๐บ\times H^0(\mathrm{\Sigma },K_\mathrm{\Sigma })/(\pm 1)`$
$`\stackrel{~}{๐ด}=\stackrel{~}{๐บ}\times H^0(\mathrm{\Sigma },K_\mathrm{\Sigma })`$
$`๐ณ=๐บ\times H^0(\mathrm{\Sigma },K_\mathrm{\Sigma }^2).`$
###### Remark 4.1
The explicit geometric description of the non-compact Calabi-Yau spaces $`X_s`$, $`X_l`$ and $`\stackrel{~}{X}_s`$ above can be extended to โlinearโ Calabi-Yau varieties with an arbitrary $`ADE`$ singularity along a curve $`\mathrm{\Sigma }`$. More precisely suppose $`RSL(2,)`$ is a fixed finite subgroup. We now can look at the moduli space $`๐บ`$ of pairs $`(\mathrm{\Sigma },V)`$, where $`\mathrm{\Sigma }`$ is a smooth curve of genus $`g2`$ and $`V\mathrm{\Sigma }`$ is a rank two holomorphic vector bundle which has canonical determinant and is equipped with a fiberwise $`R`$-action. Again for each $`s=(\mathrm{\Sigma },V)`$ we can form the non-compact Calabi-Yau variety $`X_s=\mathrm{tot}(V)/R`$. The variety $`X_s`$ has a curve of singularities if type $`R`$ and canonical minimal crepant resolution $`\stackrel{~}{X}_s`$. We can again look at the moduli spaces $`\stackrel{~}{๐ด}`$ and $`๐ณ`$ of $`\stackrel{~}{X}_s`$ and $`X_s`$ and try to describe them explicitly. A uniform description of these spaces for all groups $`R`$ was given in . Again, it turns out that $`๐ณ`$ and $`\stackrel{~}{๐ด}`$ are total spaces of vector bundles on $`๐บ`$. In Szendrรถi identifies the fibers $`๐ณ_s`$ and $`\stackrel{~}{๐ด}_s`$ over a point $`s=(\mathrm{\Sigma },V)`$ with the vector spaces
$$\begin{array}{cc}\hfill \stackrel{~}{๐ด}_s& =H^0(\mathrm{\Sigma },K_\mathrm{\Sigma }๐ฑ)\hfill \\ \hfill ๐ณ_s& =H^0(\mathrm{\Sigma },(K_\mathrm{\Sigma }๐ฑ)/W),\hfill \end{array}$$
where $`๐ฑ`$ and $`W`$ denote the Cartan algebra and the Weyl group of the complex $`ADE`$ group corresponding to $`R`$ under the McKay correspondence.
Furthermore describes explicitly the universal families of deformations of $`\stackrel{~}{X}_s`$ and $`X_s`$ over $`\stackrel{~}{๐ด}_s`$ and $`๐ณ_s`$ and shows that $`๐ด_s๐ณ_s`$ is naturally isomorphic to the cone
$$H^0(\mathrm{\Sigma },(K_\mathrm{\Sigma }๐ฑ))/WH^0(\mathrm{\Sigma },(K_\mathrm{\Sigma }๐ฑ)/W).$$
We will analyze the large $`N`$ physics of these more general transversal geometries in the forthcoming paper but for now we concentrate on the case $`R=/2`$.
## 5 Intermediate Jacobians and Hitchin Pryms
As we saw in the previous section the (normal to $`๐บ`$) loci $`๐ณ_s๐ณ`$ are isomorphic to the base $`H^0(\mathrm{\Sigma },K_\mathrm{\Sigma }^2)`$ of the $`A_1`$-Hitchin system on $`\mathrm{\Sigma }`$. This raises the question whether there is a more intrinsic geometric connection between our noncompact Calabi-Yau threefolds and Hitchin systems. In this section we will give a positive answer to this question developing an intrinsic geometric relation between Hitchin Pryms and the intermediate Jacobians of the Calabi-Yau threefolds $`X_\beta `$, $`\beta H^0(\mathrm{\Sigma },K_\mathrm{\Sigma }^2)`$. We begin with a brief review of the Hitchin system.
### 5.1 Hitchin integrable systems and Pryms
For simplicity we will consider here only the $`SL(2,)`$ Hitchin system which is relevant for our problem. Recall that an algebraically completely integrable Hamiltonian system (ACIHS) is defined by the following data
a nonsingular complex algebraic variety $`๐ฉ`$ equipped with a non-degenerate global holomorphic $`(2,0)`$ form $`\sigma `$
a projection $`๐:๐ฉ`$ where $``$ is a nonsingular complex algebraic variety so that the fibers $`๐ฉ_\beta `$ of $`๐`$ are abelian varieties satisfying
$`๐ฉ_\beta `$ is a Lagrangian subvariety of $`๐ฉ`$ for any point $`\beta `$.
Let us now recall the construction of the Hitchin integrable system following . A $`SL(2,)`$ Higgs bundle $`(E,\varphi )`$ on $`\mathrm{\Sigma }`$ consists of a rank two holomorphic bundle on $`\mathrm{\Sigma }`$ with trivial determinant and a global section $`\varphi H^0(\mathrm{\Sigma },\mathrm{End}_0(E)K_\mathrm{\Sigma })`$. Here $`\mathrm{End}_0(E)`$ denotes the bundle of traceless endomorphisms of $`E`$. Such a pair is called stable (semistable) if there are no $`\varphi `$-invariant subbundles of $`E`$ that violate the usual slope inequality . In this case, there exists a quasi-projective moduli variety $`:=_{SL(2,)}`$ of semistable Hitchin pairs of complex dimension $`6g6`$. More generally , we can consider $`G`$ Higgs bundles for a general complex reductive group $`G`$. By definition these are semistable pairs $`(P,\varphi )`$ with $`P`$ a principal $`G`$-bundle on $`\mathrm{\Sigma }`$ and $`\varphi H^0(\mathrm{\Sigma },\mathrm{ad}(P)K_\mathrm{\Sigma })`$. Again there is a quasi-projective moduli space of such Higgs bundles and much of the discussion below generalizes to these moduli spaces. For the purposes of this paper we will ignore these more general moduli spaces with one exception, namely the moduli space $`_{GL(2,)}`$ of topologically trivial $`GL(2,)`$ Higgs bundles on $`\mathrm{\Sigma }`$. As we will see below, the spaces $`_{SL(2,)}`$ and $`_{GL(2,)}`$ are closely related. They are the only moduli spaces of Hitchin pairs corresponding to a structure group of type $`A_1`$ and will not reappear in the next section as families of intermediate Jacobians for Calabi-Yaus in the moduli space $`๐ณ`$.
The key element in the construction of the ACIHS is the notion of a spectral cover introduced in . The spectral cover $`p_\beta :\stackrel{~}{\mathrm{\Sigma }}=\stackrel{~}{\mathrm{\Sigma }}_\beta \mathrm{\Sigma }`$ of a pair $`(E,\varphi )`$ is a curve in the total space of the cotangent bundle $`T^{}\mathrm{\Sigma }`$ of $`\mathrm{\Sigma }`$ defined by the eigenvalue equation
(42)
$$det(y\mathrm{id}p^{}\varphi )=0.$$
Here $`p:T^{}\mathrm{\Sigma }\mathrm{\Sigma }`$ is the natural projection and $`yH^0(T^{}\mathrm{\Sigma },p^{}T^{}\mathrm{\Sigma })`$ is the tautological section. Since $`\varphi `$ is a traceless endomorphism of $`E`$, this equation can be rewritten as
(43)
$$y^2\beta =0$$
where $`\beta =det(\varphi )H^0(\mathrm{\Sigma },K_\mathrm{\Sigma }^2)`$ is the determinant of $`\varphi `$. This shows that the spectral cover $`\stackrel{~}{\mathrm{\Sigma }}`$ is smooth reduced and irreducible if and only if $`\beta `$ has distinct simple zeroes. Moreover, $`\stackrel{~}{\mathrm{\Sigma }}`$ is invariant under the holomorphic involution $`\iota :T^{}\mathrm{\Sigma }T^{}\mathrm{\Sigma }`$ which acts by multiplication by $`(1)`$ on each fiber.
According to , the map
(44)
$$๐:H^0(\mathrm{\Sigma },K_\mathrm{\Sigma }^2),(E,\varphi )det(\varphi )$$
is proper and surjective. We will denote by $`B=H^0(\mathrm{\Sigma },K_\mathrm{\Sigma }^2)`$ the space of quadratic differentials on $`\mathrm{\Sigma }`$ and by $`B`$ the open subset consisting of quadratic differentials $`\beta `$ with simple zeroes. Let $`๐ฉ=๐^1()`$ denote the inverse image of $``$ in $``$.
In order to determine the fiber of $`๐`$ at a point $`\beta `$, note that a pair $`(E,\varphi )๐ฉ`$ gives rise to a pair $`(\stackrel{~}{\mathrm{\Sigma }},L)`$ where $`L`$ is a complex holomorphic line bundle on $`\stackrel{~}{\mathrm{\Sigma }}`$ of degree $`2g2`$. $`L`$ is defined by the property that the fiber $`L_\lambda `$ over a point $`\lambda \stackrel{~}{\mathrm{\Sigma }}`$ is the eigenspace of $`\varphi `$ corresponding to the eigenvalue $`\lambda `$. Since $`E`$ is a $`SL(2,)`$ rather than $`GL(2,)`$ bundle, the resulting line bundle $`L\mathrm{\Sigma }_\beta `$ satisfies the equivariance condition
(45)
$$\iota ^{}L=L^{}p_\beta ^{}K_\mathrm{\Sigma }.$$
One can show that there is a one-to-one correspondence between pairs $`(E,\varphi )`$ in $`๐ฉ`$ with fixed spectral cover $`\stackrel{~}{\mathrm{\Sigma }}`$ and line bundles $`L`$ on $`\stackrel{~}{\mathrm{\Sigma }}`$ satisfying condition (45). This correspondence can be extended to singular, reducible or non-reduced spectral covers by allowing $`L`$ to be a torsion free rank one sheaf on $`\stackrel{~}{\mathrm{\Sigma }}`$.
This shows that the fiber $`๐ฉ_\beta =๐^1(\beta )`$ is isomorphic to the subvariety of the Picard variety $`\text{Pic}^{2g2}(\stackrel{~}{\mathrm{\Sigma }}_\beta )`$ of line bundles on $`\stackrel{~}{\mathrm{\Sigma }}`$ of degree $`2g2`$. This subvariety is defined by the equation (45). Equivalently it can be identified with the fiber $`\mathrm{Nm}^1([K_\mathrm{\Sigma }])`$ of the norm map $`\mathrm{Nm}:\mathrm{Pic}^{2g2}(\stackrel{~}{\mathrm{\Sigma }})\mathrm{Pic}^{2g2}(\mathrm{\Sigma })`$ over the canonical point $`[K_\mathrm{\Sigma }]\mathrm{Pic}^{2g2}(\mathrm{\Sigma })`$. According to , the degree zero Picard $`\text{Pic}^0(\stackrel{~}{\mathrm{\Sigma }})`$ decomposes up to isogeny into a direct product of Abelian varieties
$$\text{Pic}^0(\stackrel{~}{\mathrm{\Sigma }})\text{Pic}^0(\stackrel{~}{\mathrm{\Sigma }})^+\times \text{Pic}^0(\stackrel{~}{\mathrm{\Sigma }})^{}$$
where $`\text{Pic}^0(\stackrel{~}{\mathrm{\Sigma }})^+`$ is the fixed locus of the inversion involution $`LL^{}`$ on $`\text{Pic}^0(\stackrel{~}{\mathrm{\Sigma }})`$ and $`\text{Pic}^0(\stackrel{~}{\mathrm{\Sigma }}_\beta )^{}`$ is the anti-invariant part of this involution. The abelian subvariety $`\text{Pic}^0(\stackrel{~}{\mathrm{\Sigma }}_\beta )^{}`$ is usually denoted by $`\text{Prym}(\stackrel{~}{\mathrm{\Sigma }}/\mathrm{\Sigma })`$ and is called the Prym variety of the spectral cover. The natural translation action of $`\text{Pic}^0(\stackrel{~}{\mathrm{\Sigma }})`$ on $`\text{Pic}^{2g2}(\stackrel{~}{\mathrm{\Sigma }})`$ intertwines the inversion involution with the involution $`\iota `$ and realizes the Hitchin fiber $`๐ฉ_\beta \mathrm{Pic}^{2g2}(\stackrel{~}{\mathrm{\Sigma }})`$ as a principal homogeneous space over the Prym variety $`\mathrm{Prym}(\stackrel{~}{\mathrm{\Sigma }}/\mathrm{\Sigma })\mathrm{Pic}^0(\stackrel{~}{\mathrm{\Sigma }})`$. So $`๐ฉ_\beta `$ is (non-canonically) isomorphic to $`\mathrm{Prym}(\stackrel{~}{\mathrm{\Sigma }}/\mathrm{\Sigma })`$.
Since we have an isomorphism $`\mathrm{Pic}^0(\stackrel{~}{\mathrm{\Sigma }})J(\stackrel{~}{\mathrm{\Sigma }})`$ determined by the Abel-Jacobi map, it follows that the Jacobian
$$J(\stackrel{~}{\mathrm{\Sigma }})=H^0(\stackrel{~}{\mathrm{\Sigma }},\mathrm{\Omega }_{\stackrel{~}{\mathrm{\Sigma }}}^1)^{}/H_1(\stackrel{~}{\mathrm{\Sigma }},)$$
also decomposes up to isogeny into a direct sum of invariant and anti-invariant parts. The Abel-Jacobi map maps the Prym to the abelian subvariety of $`J(\stackrel{~}{\mathrm{\Sigma }})^{}J(\stackrel{~}{\mathrm{\Sigma }})`$, given by
(46)
$$J(\stackrel{~}{\mathrm{\Sigma }})^{}=(H^0(\stackrel{~}{\mathrm{\Sigma }},\mathrm{\Omega }_{\stackrel{~}{\mathrm{\Sigma }}}^1)^{})^{}/H_1(\stackrel{~}{\mathrm{\Sigma }},)^{}$$
where the superscript โ$``$โ denotes the anti-invariant part. Note that
$$\begin{array}{cc}\hfill H^0(\stackrel{~}{\mathrm{\Sigma }},K_{\stackrel{~}{\mathrm{\Sigma }}})& =H^0(\stackrel{~}{\mathrm{\Sigma }},p^{}K_\mathrm{\Sigma }^2)\hfill \\ & =H^0(\mathrm{\Sigma },K_\mathrm{\Sigma }^2p_{}๐ช_{\stackrel{~}{\mathrm{\Sigma }}})\hfill \\ & =H^0(\mathrm{\Sigma },K_\mathrm{\Sigma })H^0(\mathrm{\Sigma },K_\mathrm{\Sigma }^2).\hfill \end{array}$$
Under this isomorphism, the space $`H^0(\stackrel{~}{\mathrm{\Sigma }},\mathrm{\Omega }_{\stackrel{~}{\mathrm{\Sigma }}}^1)^{}=H^0(\stackrel{~}{\mathrm{\Sigma }},K_{\stackrel{~}{\mathrm{\Sigma }}})^{}`$ of anti-invariant holomorphic differentials on $`\stackrel{~}{\mathrm{\Sigma }}`$ is identified with $`H^0(\mathrm{\Sigma },K_\mathrm{\Sigma }^2)`$. Therefore $`dim๐ฉ_\beta =dim\text{Prym}(\stackrel{~}{\mathrm{\Sigma }}/\mathrm{\Sigma })`$ equals the dimension $`3g3`$ of the space of quadratic differentials on $`\mathrm{\Sigma }`$.
So far we have constructed a family $`๐:๐ฉ`$ of Abelian varieties over the open subset $`B`$ so that the dimension of the fibers equals the dimension of the base. To complete the data of an ACIHS we have to construct a holomorphic symplectic form $`\sigma `$ on $`๐ฉ`$ so that the fibers $`๐ฉ_\beta `$ are Lagrangian cycles with respect to $`\sigma `$. This can be seen easily by noticing that the cotangent space to the moduli space of rank two stable bundles with trivial determinant at a point $`E`$ is given by the cohomology group $`H^0(\mathrm{\Sigma },\mathrm{End}_0(E)K_\mathrm{\Sigma })`$. In other words the total space of the cotangent bundle to the moduli of stable bundles (of rank two and with trivial determinant) is a Zariski open and dense set in $``$. Since $`๐ฉ`$ is also Zariski open and dense and since cotangent bundles are naturally symplectic, we obtain a holomorphic symplectic form defined on an open dense set in $`๐ฉ`$. Hitchin showed that this form extends to a holomorphic symplectic form on all of $`๐ฉ`$ and argued that all the fibers $`๐ฉ_\beta =๐^1(\beta )`$ are necessarily Lagrangian (see also ).
###### Remark 5.1
The holomorphic symplectic structure $`\sigma `$ admits also an explicit interpretation in terms of the spectral data $`(\stackrel{~}{\mathrm{\Sigma }},L)`$. We recall this interpretation since it has a direct physical significance: it is related to the natural special Kรคhler geometry on the base $``$ of the ACIHS $`๐:๐ฉ`$. Let us denote by $`๐ฑ`$ the vector bundle over $``$ whose sections are vertical vector fields on $`๐ฉ`$ which are constant on each torus fiber, i.e. $`๐ฑ=๐_{}T_{๐ฉ/}`$. Note that the fiber $`๐ฑ_\beta `$ is isomorphic to the space $`(H^0(\stackrel{~}{\mathrm{\Sigma }},\mathrm{\Omega }_{\stackrel{~}{\mathrm{\Sigma }}}^1)^{})^{}`$, which is isomorphic in turn to $`B^{}=H^0(\mathrm{\Sigma },K_\mathrm{\Sigma }^2)^{}`$. On the other hand the tangent space $`๐ฏ_\beta `$ to the base at the point $`\beta `$ is isomorphic to $`B=H^0(\mathrm{\Sigma },K_\mathrm{\Sigma }^2)`$. Therefore we have the isomorphisms
$$๐ฏ_{}^{}๐ฑB^{}๐ช_{}.$$
The integrable structure can be characterized by the cubic criterion of Donagi and Markman . Let us choose a marking of the Abelian varieties $`๐ฉ_\beta `$, i.e. a continuously varying symplectic basis of $`H_1(๐ฉ_\beta ,)`$, $`\beta `$. For example we can choose a symplectic basis of anti-invariant cycles in $`H_1(\stackrel{~}{\mathrm{\Sigma }}_\beta ,)^{}`$, for $`\beta `$.
Then, locally on $``$ the family $`๐ฉ`$ determines a period map $`\varrho :_{3g3}`$ where $`_{3g3}`$ denotes the Siegel upper half space
$$_{3g3}=\{(3g3)\times (3g3)\text{symmetric complex matrices }\text{Z}\text{ with}\text{Im}(Z)>0\}$$
Note that we can identify $`_{3g3}`$ with a subspace of $`S^2B`$ by choosing a basis of holomorphic quadratic differentials on $`\mathrm{\Sigma }`$. Then the following conditions are equivalent \[7, Lemma 7.4\]
There exists a holomorphic symplectic form $`\sigma `$ on $`๐ฉ`$ so that the fibers of $`๐:๐ฉ`$ are Lagrangian, and $`\sigma `$ induces the identity isomorphism
$$\text{Id}\text{Hom}(๐ฏ_{๐ฉ/},๐^{}๐ฏ_{}^{})\text{Hom}(๐^{}๐ฑ,๐^{}๐ฑ).$$
The period map $`\varrho :S^2V`$ can be locally written in $``$ as the Hessian of a holomorphic function on $``$ (the holomorphic prepotential.)
The differential of the period map $`d\varrho \text{Hom}(๐ฏ_{},S^2๐ฑ)๐ฑS^2๐ฑ`$ is a section of $`S^3๐ฑ`$ (the cubic condition).
For the family $`๐ฉ`$ of $`SL(2,)`$-Hitchin Pryms the cubic $`d\varrho `$ can be computed explicitly and is given by
(47)
$$d\varrho _\beta :S^3(๐ฏ_\beta ),\gamma _1\gamma _2\gamma _3\text{Res}^2\left(\frac{\gamma _1\gamma _2\gamma _3}{\beta ^2}\right).$$
where $`\text{Res}^2\left(\frac{\gamma _1\gamma _2\gamma _3}{\beta ^2}\right)`$ is the quadratic residue of a quadratic differential.
For future reference we recast the Hodge theoretic interpretation (46) of the spectral Prym in terms of data on the base curve $`\mathrm{\Sigma }`$. Recall that $`\mathrm{Prym}(\stackrel{~}{\mathrm{\Sigma }}/\mathrm{\Sigma })`$ was naturally identified with the kernel of the norm map $`\mathrm{Nm}:\mathrm{Pic}^0(\stackrel{~}{\mathrm{\Sigma }})\mathrm{Pic}^0(\mathrm{\Sigma })`$ between the degree zero Picard varieties of $`\stackrel{~}{\mathrm{\Sigma }}`$ and $`\mathrm{\Sigma }`$. Topologically we have identifications $`\mathrm{Pic}^0(\stackrel{~}{\mathrm{\Sigma }})=H_1(\stackrel{~}{\mathrm{\Sigma }},)(/)`$ and $`\mathrm{Pic}^0(\mathrm{\Sigma })=H_1(\mathrm{\Sigma },)(/)`$. In fact these identifications can be thought of as isomorphisms of abelian varieties if we endow the right hand sides with the complex structures coming from the Hodge structures on the first homology of $`\stackrel{~}{\mathrm{\Sigma }}`$ and $`\mathrm{\Sigma }`$. This gives a natural topological identification
(48)
$$\begin{array}{cc}\hfill \mathrm{Prym}(\stackrel{~}{\mathrm{\Sigma }}/\mathrm{\Sigma })& =\mathrm{ker}\left[\mathrm{Pic}^0(\stackrel{~}{\mathrm{\Sigma }})\stackrel{\mathrm{Nm}}{}\mathrm{Pic}^0(\mathrm{\Sigma })\right]\hfill \\ & =\mathrm{ker}\left[H_1(\stackrel{~}{\mathrm{\Sigma }},)H_1(\mathrm{\Sigma },)\right]_{}(/)\hfill \\ & =H^1(\mathrm{\Sigma },๐ฆ)_{}(/),\hfill \end{array}$$
where
$$๐ฆ:=\mathrm{ker}\left(p_\beta _{\stackrel{~}{\mathrm{\Sigma }}}\stackrel{\mathrm{Tr}}{}_\mathrm{\Sigma }\right)$$
is the kernel of the natural trace map. In the last step of (48), we used the long exact sequence:
$$0H^0(\mathrm{\Sigma },/n)H^1(\mathrm{\Sigma },๐ฆ)H^1(\stackrel{~}{\mathrm{\Sigma }},)H^1(\mathrm{\Sigma },)$$
which implies that $`H^1(\mathrm{\Sigma },๐ฆ)`$ agrees with $`\mathrm{ker}(H_1(\stackrel{~}{\mathrm{\Sigma }},)H_1(\stackrel{~}{\mathrm{\Sigma }},))`$ up to torsion, which disappears when tensoring with $`/`$.
Alternatively, the sheaf $`๐ฆ`$ can be described as follows. Write $`๐ฉ\mathrm{\Sigma }`$ and $`๐น\stackrel{~}{\mathrm{\Sigma }}`$ for the branch and ramification divisors of $`p_\beta `$. Let $`\mathrm{\Sigma }^o:=\mathrm{\Sigma }๐ฉ`$, $`\stackrel{~}{\mathrm{\Sigma }}^o:=\mathrm{\Sigma }๐น`$, and let $`p_\beta ^o:=\stackrel{~}{\mathrm{\Sigma }}^o\mathrm{\Sigma }^o`$ be the restriction of the projection. If we now denote by $`j:\mathrm{\Sigma }^o\mathrm{\Sigma }`$, then $`๐ฆ`$ can be viewed as $`j_{}`$ of the local system on $`\mathrm{\Sigma }^o`$ of anti-invariant $``$-valued functions on $`\stackrel{~}{\mathrm{\Sigma }}^o`$. More invariantly, we have
(49)
$$๐ฆ=(p_\beta \mathrm{\Lambda }_\mathrm{r})^W=j_{}((p_\beta ^o\mathrm{\Lambda }_\mathrm{r})^W),$$
where $`\mathrm{\Lambda }_\mathrm{r}`$ and $`W`$ denote the root lattice and Weyl group of of $`SL(2,)`$ respectively, and we view the covering involution of $`\stackrel{~}{\mathrm{\Sigma }}\mathrm{\Sigma }`$ as the generator of $`W`$.
The analysis of the moduli space $`_{GL(2,)}`$ of topologically trivial $`GL(2,)`$ Higgs bundles is similar. In fact, the moduli space $``$ determines $`_{GL(2,)}`$. To see this we first note that with any Hitchin pair $`(E,\varphi )`$ consisting of a rank two vector bundle $`E`$ with trivial determinant and a Higgs field $`\varphi :EEK_\mathrm{\Sigma }`$ gives rise to a $`GL(2,)`$ Hitchin pair $`(P_E,\mathrm{ad}_\varphi )`$, where $`P_E`$ is the $`GL(2,)`$-bundle associated with the frame bundle of $`E`$ via the adjoint representation of $`SL(2,)`$. It is easy to check that this procedure preserves semistability and so one gets a well defined morphism $`๐๐
:_{GL(2,)}`$. Furthermore, since a principal $`GL(2,)`$ bundle can be viewed as a $`^2`$ bundle, and since on a curve all projective bundles are projectivizations of vector bundles, we see that the morphism $`_{GL(2,)}`$ is surjective. A more careful analysis (see e.g. ) shows that $`_{GL(2,)}`$ is in fact the quotient of $``$ by the finite group $`H^1(\mathrm{\Sigma },/2)=\mathrm{Pic}^0(\mathrm{\Sigma })[2]`$ of $`2`$-torsion line bundles on $`\mathrm{\Sigma }`$. Here an element $`\xi \mathrm{Pic}^0(\mathrm{\Sigma })[2]`$ acts as $`(E,\varphi )(E\alpha ,\varphi \mathrm{id})`$ and so this action preserves the fibers $`๐ฉ_\beta `$ of the Hitchin map $`๐`$ and the symplectic form on $``$. Thus $``$ and $`_{GL(2,)}`$ are ACIHS which are fibered by Lagrangian tori over the same base space $`B=H^0(\mathrm{\Sigma },K_\mathrm{\Sigma }^2)`$ and related by a finite map that respects the symplectic forms and the Lagrangian fibrations:
In particular, the finite map $`๐๐
`$ gives an explicit identification of the fiber of the $`GL(2,)`$ Hitchin map over $`\beta `$ with the quotient $`๐ฉ_\beta /\mathrm{Pic}^0(\mathrm{\Sigma })[2]`$. Thus the fibers of the $`SL(2,)`$ and $`GL(2,)`$ Hitchin maps over the same point $`\beta `$ are isogenous abelian varieties.
This concludes our review of the $`A_1`$-Hitchin system. Next, we will explain the connection between this ACIHS and the family of noncompact threefolds constructed in the previous section.
### 5.2 Intermediate Jacobians and the Calabi-Yau integrable system
Let us start with some general considerations regarding Jacobian fibrations and integrable systems associated to families of Calabi-Yau threefolds .
Suppose $``$ is a component of an enlarged moduli space of smooth projective Calabi-Yau threefolds supporting a universal family $`๐ณ`$. Recall that the enlarged moduli space parameterizes Calabi-Yau threefolds together with a choice of a nontrivial global holomorphic three-form. The family $`๐ณ`$ determines a complex torus fibration $`๐ฅ`$ whose fiber over a point $`\beta `$ is the intermediate Jacobian
(50)
$$J_3(X_\beta )=F^1H^3(X_\beta ,)^{}/H_3(X_\beta ,).$$
Here
$$F^1H^3(X_\beta )=H^{3,0}(X_\beta ,)H^{2,1}(X_\beta )$$
denotes the first step in the Hodge filtration on the third cohomology of $`X_\beta `$. The intermediate Jacobian has a natural non-degenerate indefinite polarization corresponding to the intersection pairing on $`H_3(X_\beta ,)`$. Therefore $`J_3(X_\beta )`$ are complex analytic tori, but not Abelian varieties. It is also possible to describe the intermediate Jacobian without passing to quotients. Namely we have a natural identification:
$$J_3(X_\beta )=H_3(X_\beta ,)_{}S^1,$$
where $`S^1`$ is the circle $`S^1=/`$. This point of view is particularly good for studying the algebraic properties of $`J_3(X_\beta )`$ as a torus but the complex structure on $`J_3(X_\beta )`$ is disguised in this interpretation.
According to the resulting fibration $`๐ฅ`$ underlies an analytically completely integrable Hamiltonian system. This is very similar to the ACIHS structure encountered in the previous section, except that we have to employ analytic spaces instead of algebraic varieties in the defining properties (i)-(iii) listed on page 5.1. The existence of an analytic integrable structure follows again from the cubic criterion of . The required cubic form is in this case the normalized Yukawa coupling (see for example .)
In the section 4.3 we have constructed a family of noncompact Calabi-Yau threefolds $`๐ณ๐ณ`$ parameterized by $`๐ณ=๐บ\times B`$ where $`๐บ`$ is the moduli space of rank two bundles on $`\mathrm{\Sigma }`$ with canonical determinant, and $`B=H^0(\mathrm{\Sigma },K_\mathrm{\Sigma }^2)`$. The threefold $`X_l`$, $`l=(s,\beta )๐บ\times B`$ is smooth if and only if $`\beta B`$ has distinct simple zeroes, that is if and only if $`\beta `$. Since the dependence on the point $`s๐บ`$ is inessential, throughout this section we will consider the family obtained by restricting $`๐ณ๐ณ`$ to a subspace of the form $`\{s\}\times B๐บ\times B`$. From now on, until the end of this section, we will drop the point $`s`$ from the labeling. Moreover, abusing notation we will denote by $`๐ณ`$ the restriction of the family $`๐ณ`$ to the open subset $`\{s\}\times B`$ parameterizing smooth threefolds.
Our main goal is to establish a connection between the intermediate Jacobian fibration of the family $`๐ณ`$ and the Hitchin integrable system. In contrast with the case of compact threefolds described above, there is no general result concerning the existence of an integrable structure on the intermediate Jacobian fibration of a family of noncompact Calabi-Yau manifolds. In general, these Jacobians are noncompact and may not even have the same dimension as the base of the fibration.
However in our case the fibers of the family $`๐ณ`$ are simple enough so that we can analyze their intermediate Jacobians in detail. Fix a point $`\beta `$ and let $`X_\beta `$ be the corresponding smooth non-compact Calabi-Yau threefold. The intermediate Jacobians of $`X_\beta `$ are Hodge theoretic invariants of the complex structure of $`X_\beta `$. They are generalized tori (= quotients of a vector space by a discrete abelian subgroup) defined in terms of the mixed Hodge structure on the cohomology or the homology of $`X_\beta `$. More precisely, by the work of Deligne we know that the abelian group $`H=H^k(X_\beta ,)`$ (respectively $`H=H_k(X_\beta ,)`$) is equipped with a mixed Hodge structure $`(W_{},F^{})`$, where $`W_{}`$ is an increasing weight filtration on the rational vector space $`H_{}:=H`$ and $`F^{}`$ is a decreasing Hodge filtration on the complex vector space $`H_{}:=H`$.
The weight and Hodge filtrations should be compatible in the sense that $`F^{}`$ induces a Hodge decomposition of weight $`\mathrm{}`$ on the $`\mathrm{}`$-th graded piece of $`\mathrm{gr}_{\mathrm{}}^W=W_{\mathrm{}}/W_\mathrm{}1`$ of the weight filtration. This means that $`\mathrm{gr}_{\mathrm{}}^W=_{p+q=\mathrm{}}H^{p,q}`$, where
$$H^{p,q}=[(F^pW_{\mathrm{}}+W_\mathrm{}1)/W_\mathrm{}1][(\overline{F}^qW_{\mathrm{}}+W_\mathrm{}1)/W_\mathrm{}1].$$
Equivalently the three filtrations $`W_{}`$, $`F^{}`$ and $`\overline{F}^{}`$ should satisfy
$$\mathrm{gr}_F^p\mathrm{gr}_{\overline{F}}^q\mathrm{gr}_{\mathrm{}}^WH=0,\text{ for all }p+q\mathrm{}.$$
Given a mixed Hodge structure $`(H,W_{},F^{})`$ we can consider the smallest interval $`[a,b]`$ such that $`\mathrm{gr}_{\mathrm{}}^W=0`$ for $`\mathrm{}[a,b]`$. The integer $`ba`$ is called length of the mixed Hodge structure and $`a`$ and $`b`$ are the lowest and highest weight respectively. A mixed Hodge structure of length zero is pure of weight $`a=b`$.
With every mixed Hodge structure $`(H,W_{},F^{})`$ one associates a sequence of intermediate Jacobians. If $`(H,W_{},F^{})`$ is a mixed Hodge structure and $`p`$ is any integer satisfying
$$p>\frac{1}{2}(\text{highest weight of }(H,W_{},F^{})),$$
then the level $`p`$ intermediate Jacobian of $`(H,W_{},F^{})`$ is
$$H_{}/(F^pH_{}+H).$$
The condition on $`p`$ here is imposed to ensure that $`H_{}`$ projects to a discrete subgroup in $`H_{}/F^pH`$, i.e. that the Jacobian is a generalized torus.
Since $`X_\beta `$ is non-compact we will have to take extra care in distinguishing the intermediate Jacobians associated with the mixed Hodge structures on $`H^3(X_\beta ,)`$ and $`H_3(X_\beta ,)`$. We will denote these generalized tori by $`J^3(X_\beta )`$ and $`J_3(X_\beta )`$ respectively. Explicitly
(51) $`J^3(X_\beta )`$ $`=H^3(X_\beta ,)/(F^2H^3(X_\beta ,)+H^3(X_\beta ,)),`$
(52) $`J_3(X_\beta )`$ $`=H_3(X_\beta ,)/(F^1H_3(X_\beta ,)+H_3(X_\beta ,)),`$
(53) $`=H^3(X_\beta ,)/(F^2H^3(X_\beta ,)+H_3(X_\beta ,)),`$
where in the formula (53) the inclusion $`H_3(X_\beta ,)/(\text{torsion})H^3(X_\beta ,)`$ is given by the intersection pairing map on three dimensional cycles in $`X_\beta `$. More precisely, by the universal coefficients theorem we can identify $`H^3(X_\beta ,)/(\text{torsion})`$ with the dual lattice $`H_3(X_\beta ,)^{}:=\mathrm{Hom}_{}(H_3(X_\beta ,),)`$. Combining this identification with the intersection pairing on the third homology of $`X_\beta `$ we get a well defined map
which is injective on the free part of $`H_3(X_\beta ,)`$. Combining $`i`$ with the natural inclusion $`H^3(X_\beta ,)/(\text{torsion})H^3(X_\beta ,)`$ we obtain the map appearing in (53). Furthermore since $`i`$ is injective modulo torsion, it follows that the induced surjective map on intermediate Jacobians
(54)
$$J_3(X_\beta )J^3(X_\beta )$$
is a finite isogeny of generalized tori. Note that when $`X_\beta `$ is compact the unimodularity of the Poincare pairing implies that (54) is an isomorphism and so we do not have to worry about the distinction between $`J_3(X_\beta )`$ and $`J^3(X_\beta )`$. In fact, we will see below that for our non-compact $`X_\beta `$, $`\beta `$ the mixed Hodge structure on $`H^3(X,)`$ is actually pure and of weight $`3`$. This implies that
$$\begin{array}{cc}\hfill J_3(X_\beta )& =H_3(X_\beta ,)_{}S^1\hfill \\ \hfill J^3(X_\beta )& =H^3(X_\beta ,)_{}S^1\hfill \end{array}$$
and so the two Jacobians are compact complex tori. Furthermore the isogeny (54) can be identified explicitly as
However we will also check that the map $`i\mathrm{id}`$ is not an isomorphism and has a finite kernel that can be identified explicitly.
To demonstrate the purity of the Hodge structure on $`H^3(X,)`$ we look at the map $`\pi _\beta :X_\beta \mathrm{\Sigma }`$ onto the compact Riemann surface $`\mathrm{\Sigma }`$. As we saw in the previous section, the fibers $`X_{\beta ,t}:=\pi _\beta ^1(t)`$ are smooth affine quadrics for $`t`$ not in the divisor of the quadratic differential $`\beta H^0(\mathrm{\Sigma },K_\mathrm{\Sigma }^2)`$ and are irreducible quadratic cones for those $`t`$ for which $`\beta (t)=0`$. On the other hand, every two dimensional smooth affine quadric $`Q^3`$ is deformation equivalent<sup>2</sup><sup>2</sup>2The deformation equivalence of $`Q`$ and $`\mathrm{tot}(๐ช_^1(2))`$ is realized explicitly by the resolved conifold $`\stackrel{~}{Z}`$ discussed in detail in section 2. to the surface $`\mathrm{tot}(๐ช_^1(2))`$. Thus by Ehresmannโs fibration theorem $`Q`$ is homeomorphic to $`\mathrm{tot}(๐ช_^1(2))`$ which in turn is homotopy equivalent to $`^1`$. Therefore
$$\begin{array}{cc}\hfill H^0(Q,)& =H^2(Q,)=\hfill \\ \hfill H_0(Q,)& =H_2(Q,)=\hfill \end{array}$$
and the rest of the cohomology and homology of $`Q`$ vanishes. Also, under the deformation equivalence $`Q\mathrm{tot}(๐ช_^1(2))`$ the generator $`c`$ of $`H_2(Q,)`$ can be identified with the zero section of $`๐ช_^1(2))`$ and so the intersection form on $`H_2(Q,)`$ is given by $`cc=2`$. Thus the second homology $`H_2(Q,)`$ can be intrinsically identified with the root lattice $`\mathrm{\Lambda }_\mathrm{r}`$ of $`SL(2,)`$. If we now use the universal coefficients theorem to identify $`H^2(Q,)`$ with $`H_2(Q,)^{}`$ we get a natural map
given by the intersection pairing on two cycles. Since $`cc=2`$, it follows that the image of $`H_2(Q,)`$ is a subgroup of index two in $`H^2(Q,)`$. In other words we get a natural isomorphism of $`H^2(Q,)`$ with the weight lattice $`\mathrm{\Lambda }_\mathrm{w}`$ of $`SL(2,)`$. Now fix a base point $`t_0\mathrm{\Sigma }^o=\mathrm{\Sigma }(\text{divisor of }\beta )`$ and identify the fiber $`\pi _\beta ^1(t_0)`$ with $`Q`$. Choose a collection $`\{\gamma _x\}_{x\text{div}(\beta )}`$ of non-intersecting paths in $`\mathrm{\Sigma }`$ connecting $`t_0`$ with each zero $`x`$ of $`\beta `$. Since $`\pi _\beta :X_\beta \mathrm{\Sigma }`$ is a Lefschetz fibration, it follows that the sphere $`c\pi _\beta ^1(t_0)Q`$ vanishes along each $`\gamma _x`$ and that the local monodromy on $`H_2(Q,)`$ is given by the Picard-Lefschetz transformation $`cc+(c,c)c=c2c=c`$. Thus the local monodromy action on $`H^2(Q,)=\mathrm{\Lambda }_\mathrm{w}`$ is naturally equal to the action of the Weyl group $`W`$ on the weight lattice of $`SL(2,)`$. Furthermore, it is not hard to compute the global monodromy
(55)
$$\mathrm{mon}:\pi _1(\mathrm{\Sigma }^o,t_0)\{\pm 1\}=\mathrm{Aut}(H^2(\pi _\beta ^1(t_0),)).$$
Indeed, as explained at the end of section 4.3, the threefold $`X_\beta `$ is given by the equation
(56)
$$\mathrm{det}_W\pi _W^{}\beta =0$$
in the total space $`W`$ of the rank three vector bundle $`S^2V`$. In particular, the global monodromy (56) is the same as the monodromy of a double cover of $`\mathrm{\Sigma }`$ which is branched precisely at the zeroes of $`\beta `$. If we further look at the compactification $`\overline{X}_\beta `$ we can identify the representation (55) with the monodromy on the two families of rulings in the fibers of $`\overline{X}_{\beta |\mathrm{\Sigma }^o}\mathrm{\Sigma }^o`$. However from the equation (56) it is manifest that the covering parameterizing the two families of rulings is given by the equation $`y^2\beta `$ in $`T^{}\mathrm{\Sigma }`$, i.e. is the spectral cover $`p_\beta :\stackrel{~}{\mathrm{\Sigma }}_\beta \mathrm{\Sigma }`$.
This description of the local and global monodromies implies that we have a natural identification
$$R^2\pi _\beta =j_{}((p_\beta ^o\mathrm{\Lambda }_\mathrm{w})^W)=(p_\beta \mathrm{\Lambda }_\mathrm{w})^W,$$
where $`j:\mathrm{\Sigma }^o\mathrm{\Sigma }`$ and $`p_\beta ^o:\stackrel{~}{\mathrm{\Sigma }}^o\mathrm{\Sigma }^o`$ are the maps described at the end of the previous section.
Finally from the Leray spectral sequence for the map $`\pi _\beta :X_\beta \mathrm{\Sigma }`$ we immediately see that $`H^3(\mathrm{\Sigma },)=H^1(\mathrm{\Sigma },R^2\pi _\beta )`$ or equivalently
(57)
$$H^3(\mathrm{\Sigma },)=H^1(\mathrm{\Sigma },(p_\beta \mathrm{\Lambda }_\mathrm{w})^W).$$
Similarly, we have $`R^2\pi _\beta =j_{}(R^2\pi _\beta ^o)`$ and $`H^3(X_\beta ,)=H^1(\mathrm{\Sigma },j_{}(R^2\pi _\beta ^o))`$. Since the Leray filtration is compatible with mixed Hodge structures and for the affine quadric $`Q`$ the cohomology $`H^2(Q,)`$ is spanned by a single class of type $`(1,1)`$, it follows that the Hodge structure on $`H^3(X_\beta ,)`$ is pure of weight three. The last statement follows from the fact that the local system $`R^2\pi _\beta ^o`$ is a variation of pure Hodge structures of Tate type and weight two, and from the fact that for every complex local system $`๐`$ on $`\mathrm{\Sigma }^o`$ we have $`H^1(\mathrm{\Sigma },j_{}๐)=\mathrm{im}\left[H_c^1(\mathrm{\Sigma }^o,๐)H^1(\mathrm{\Sigma }^o,๐)\right]`$. In particular the only non-trivial pieces in the Hodge decomposition on $`H^3(X_\beta ,)`$ are of Hodge types $`(2,1)`$ and $`(1,2)`$. Twisting $`H^3(X_\beta ,)`$ by the Tate Hodge structure of weight $`(2)`$ we get a Hodge decomposition on $`H^3(X_\beta ,)`$ which involves only $`(1,0)`$ and $`(0,1)`$ components. Thus $`J^3(X_\beta )`$ and $`J_3(X_\beta )`$ are both abelian varieties which are dual to each other. Finally, the two intermediate Jacobians come with a canonical isogeny $`J_3(X_\beta )J^3(X_\beta )`$ which combined with the duality gives natural polarizations on the abelian varieties $`J^3(X_\beta )`$ and $`J_3(X_\beta )`$.
With all of this in place, we are now ready to compare the intermediate Jacobians $`J^3(X_\beta )`$ and $`J_3(X_\beta )`$ with the Hitchin fiber $`๐ฉ_\beta =\mathrm{Prym}(\stackrel{~}{\mathrm{\Sigma }}_\beta /\mathrm{\Sigma }_\beta )`$. According to equations (48), (49) and (57) we have
$$\begin{array}{cc}\hfill ๐ฉ_\beta & =H^1(\mathrm{\Sigma },(p_\beta \mathrm{\Lambda }_\mathrm{r})^W)_{}S^1,\hfill \\ \hfill J^3(X_\beta )& =H^1(\mathrm{\Sigma },(p_\beta \mathrm{\Lambda }_\mathrm{w})^W)_{}S^1,\hfill \\ \hfill J_3(X_\beta )& =H^1(\mathrm{\Sigma },(p_\beta \mathrm{\Lambda }_\mathrm{w})^W)^{}_{}S^1,\hfill \end{array}$$
with complex structures all coming from the Hodge decomposition on the space $`H^1(\mathrm{\Sigma },(p_\beta _{\stackrel{~}{\mathrm{\Sigma }}_\beta })^W)`$, where the $`W`$ action on the constant sheaf $`_{\stackrel{~}{\mathrm{\Sigma }}_\beta }\stackrel{~}{\mathrm{\Sigma }}_\beta `$ corresponds to the diagonal action of $`W`$ on $`\stackrel{~}{\mathrm{\Sigma }}_\beta \times `$ which is the covering involution on $`\stackrel{~}{\mathrm{\Sigma }}_\beta `$ and the multiplication by $`(1)`$ on $``$. Therefore the comparison of $`J^3(X_\beta )`$ and $`J_3(X_\beta )`$ with $`๐ฉ_\beta `$ as polarized abelian varieties amounts to a comparison of the torsion free parts of the abelian groups $`H^1(\mathrm{\Sigma },(p_\beta \mathrm{\Lambda }_\mathrm{w})^W)`$ and $`H^1(\mathrm{\Sigma },(p_\beta \mathrm{\Lambda }_\mathrm{r})^W)`$.
To make the comparison more explicit we choose a basis in $`H^1(\stackrel{~}{\mathrm{\Sigma }}_\beta ,)=H_1(\stackrel{~}{\mathrm{\Sigma }}_\beta ,)`$ which is adapted to the double cover $`p_\beta :\stackrel{~}{\mathrm{\Sigma }}_\beta \mathrm{\Sigma }`$. Up to homotopy we can bring all the branch points of $`p_\beta `$ inside a fixed disk $`D\mathrm{\Sigma }`$. Thinking of $`D`$ as a genus zero surface with a single boundary component, we can build the cover $`\stackrel{~}{\mathrm{\Sigma }}_\beta `$ topologically by first taking a double cover of $`D`$ corresponding to branch cuts between pairs of branch points and then attaching two copies of $`\mathrm{\Sigma }D`$ to the two boundary circles of this double cover as depicted on Figure 1 below.
Now choose a basis $`\{\alpha _i\}_{i=1}^{2g}`$ in $`H_1(\mathrm{\Sigma }D,)=H_1(\mathrm{\Sigma },)`$ consisting of loops contained in $`\mathrm{\Sigma }D`$. The inverse image of an $`\alpha _i`$ in $`\stackrel{~}{\mathrm{\Sigma }}_\beta `$ consists of two independent disjoint loops $`\alpha _i^{}`$ and $`\alpha _i^{\prime \prime }`$. Adding to $`\alpha _i^{}`$ and $`\alpha _i^{\prime \prime }`$ the extra loops $`\gamma _j`$ coming from the branch cuts on $`D`$ we get a basis
(58)
$$\left\{\alpha _i^{},\alpha _i^{\prime \prime },\gamma _j\right|i=1,\mathrm{},2g,j=1,\mathrm{},4g6\}$$
of $`H^1(\stackrel{~}{\mathrm{\Sigma }}_\beta ,)`$ for which:
* Under the map $`p_\beta :H^1(\stackrel{~}{\mathrm{\Sigma }}_\beta ,)H^1(\mathrm{\Sigma },)`$ we have
$$p_\beta (\alpha _i^{})=p_\beta (\alpha _i^{\prime \prime })=\alpha _i,\text{and}p_\beta (\gamma _j)=0.$$
* The covering involution $`ฤฑ:\stackrel{~}{\mathrm{\Sigma }}_\beta \stackrel{~}{\mathrm{\Sigma }}_\beta `$ for the cover $`p_\beta `$ transforms this basis as
$$ฤฑ(\alpha _i^{})=\alpha _i^{\prime \prime },ฤฑ(\gamma _j)=\gamma _j.$$
* The classes $`\alpha _i^{}`$, $`\alpha _i^{\prime \prime }`$, and $`\gamma _j`$ do not intersect with each other.
With the basis (58) at our disposal we can now proceed with the comparison of $`H^1(\mathrm{\Sigma },(p_\beta \mathrm{\Lambda }_\mathrm{w})^W)`$ and $`H^1(\mathrm{\Sigma },(p_\beta \mathrm{\Lambda }_\mathrm{r})^W)`$. Choose a generator of $`\mathrm{\Lambda }_\mathrm{r}`$ and identify $`\mathrm{\Lambda }_\mathrm{r}`$ with $``$ and $`\mathrm{\Lambda }_\mathrm{r}`$ with $``$. Using the natural injection $`\mathrm{\Lambda }_\mathrm{r}\mathrm{\Lambda }_\mathrm{w}`$ given by the Cartan form on $`\mathrm{\Lambda }_\mathrm{r}`$ we can identify $`\mathrm{\Lambda }_\mathrm{w}`$ with the sub-lattice $`\frac{1}{2}`$. With this identification the generator of the Weyl group $`W`$ acts as multiplication by $`(1)`$. Now by the calculation we did at the end of section 5.1 we get that
$$\begin{array}{cc}\hfill H^1(\mathrm{\Sigma },(p_\beta \mathrm{\Lambda }_\mathrm{r})^W)& =\left(H^1(\stackrel{~}{\mathrm{\Sigma }}_\beta ,)\mathrm{\Lambda }_\mathrm{r}\right)^W=H^1(\stackrel{~}{\mathrm{\Sigma }}_\beta ,)^{}=\left(_i(\alpha _i^{}\alpha _i^{\prime \prime })\right)\left(_j\gamma _j\right)\hfill \\ \hfill H^1(\mathrm{\Sigma },(p_\beta \mathrm{\Lambda }_\mathrm{w})^W)& =\left(H^1(\stackrel{~}{\mathrm{\Sigma }}_\beta ,)\mathrm{\Lambda }_\mathrm{w}\right)^W=\frac{1}{2}H^1(\stackrel{~}{\mathrm{\Sigma }}_\beta ,)^{}=\frac{1}{2}\left(\left(_i(\alpha _i^{}\alpha _i^{\prime \prime })\right)\left(_j\gamma _j\right)\right).\hfill \end{array}$$
Here $`H^1(\stackrel{~}{\mathrm{\Sigma }}_\beta ,)^{}`$ denotes the anti-invariants of the action of the covering involution $`ฤฑ`$.
From this identification it follows that the abelian varieties
$$\begin{array}{cc}\hfill ๐ฉ_\beta & =H^1(\mathrm{\Sigma },(p_\beta \mathrm{\Lambda }_\mathrm{r})^W)S^1\hfill \\ \hfill J^3(X)& =H^1(\mathrm{\Sigma },(p_\beta \mathrm{\Lambda }_\mathrm{w})^W)S^1\hfill \end{array}$$
are isomorphic but the natural map between them is given by a composition of an isomorphism and a multiplication-by-two map.
Similarly we get that the dual abelian varieties $`๐ฉ_\beta ^{}`$ and $`J_3(X_\beta )`$ are also isomorphic. Explicitly we compute
$$\begin{array}{cc}\hfill H^1(\mathrm{\Sigma },(p_\beta \mathrm{\Lambda }_\mathrm{r})^W)^{}& =\left(H^1(\stackrel{~}{\mathrm{\Sigma }}_\beta ,)\mathrm{\Lambda }_\mathrm{w}\right)_W=\left(\left(_i\frac{\alpha _i^{}\alpha _i^{\prime \prime }}{2}\right)\left(_j\gamma _j\right)\right)\hfill \\ \hfill H^1(\mathrm{\Sigma },(p_\beta \mathrm{\Lambda }_\mathrm{w})^W)^{}& =\left(H^1(\stackrel{~}{\mathrm{\Sigma }}_\beta ,)\mathrm{\Lambda }_\mathrm{r}\right)_W=2\left(\left(_i\frac{\alpha _i^{}\alpha _i^{\prime \prime }}{2}\right)\left(_j\gamma _j\right)\right).\hfill \end{array}$$
Here for a $`W`$-module $`\mathrm{\Gamma }`$ we write $`\mathrm{\Gamma }_W`$ for the group of $`W`$-coinvariants in $`\mathrm{\Gamma }`$. That is, $`\mathrm{\Gamma }_W`$ is the quotient of $`\mathrm{\Gamma }`$ by the additive subgroup in $`\mathrm{\Gamma }`$ generated by all elements of the form $`\gamma w\gamma `$, $`wW`$, $`\gamma \mathrm{\Gamma }`$. In particular we see that $`๐ฉ_\beta ^{}=J_3(X_\beta )`$ is also isomorphic to the quotient of $`๐ฉ_\beta =J^3(X_\beta )`$ by the group $`\mathrm{Pic}^0(\mathrm{\Sigma })[2]`$ of $`2`$-torsion points on the Jacobian of $`\mathrm{\Sigma }`$. On the other hand, as we saw in the previous section (see also ), the quotient $`๐ฉ_\beta /\mathrm{Pic}^0(\mathrm{\Sigma })[2]`$ is the Hitchin fiber for the Hitchin system of the Langlands dual group $`GL(2,)`$.
To summarize: The family of homology intermediate Jacobians associated with the family of non-compact Calabi-Yau manifolds $`๐ณ`$ is an ACIHS isomorphic to the Hitchin system for the group $`GL(2,)`$. The family of cohomology intermediate Jacobians associated with $`๐ณ`$ is an ACIHS isomorphic to the Hitchin system for the group $`SL(2,)`$.
In we will show how to generalize the above statement to an isomorphism of the Calabi-Yau integrable system of the Calabi-Yau varieties described in Remark 4.1 and the Hitchin integrable system for the corresponding $`ADE`$ group.
## 6 Large $`N`$ quantization for linear transitions
In this section we give a physical proof of genus zero large N duality for the linear transitions constructed in section four. The open string side of the duality is constructed by wrapping topological B-branes on the exceptional curves in a threefold $`\stackrel{~}{X}_{\stackrel{~}{m}}`$ corresponding to a generic point in $`\stackrel{~}{๐ด}`$. The resulting topological open-closed string theory is expected to be related to closed topological string theory on a smoothing $`X_l`$, $`l๐ณ`$ of the nodal singularities. We will give a physical proof for this equivalence at genus zero by showing that the large $`N`$ dynamics of topological B-branes in the planar limit is governed by the Hitchin integrable system constructed in section five.
Following the strategy of the previous sections, we will fix a point $`s๐บ`$ and work only along the normal to $`๐บ`$ slice $`๐ณ_s=H^0(\mathrm{\Sigma },K_\mathrm{\Sigma }^2)`$ of the open stratum $`๐ณ`$ of the moduli space. We are interested in topological open-closed string theory on a resolution $`\stackrel{~}{X}_\alpha `$ of a nodal threefold $`X_{\alpha ^2}`$. Assuming $`\alpha `$ to be generic, let us denote by $`C_1,\mathrm{},C_{2g2}`$ the exceptional curves on $`\stackrel{~}{X}_\alpha `$. We construct an open-closed topological string theory by wrapping $`N_i`$ $`๐`$-type branes on the curve $`C_i`$, $`i=1,\mathrm{},2g2`$ so that the net D-brane number $`_{i=1}^{2g2}N_i`$ is zero. This means that on each curve $`C_i`$ we may have either branes or anti-branes depending on the sign of $`N_i`$. We will denote by $`N`$ the total number of branes in the system, which equals the total number of anti-branes i.e.
$$N=\underset{N_i>0}{}N_i=\underset{N_i<0}{}N_i.$$
For a more precise mathematical definition of the boundary B-model, recall that topological B-branes on a Calabi-Yau space should be thought of as derived objects (see also .) The off-shell dynamics of B-branes is captured by a topological string field theory whose action is a graded version of holomorphic Chern-Simons theory on $`\stackrel{~}{X}_\alpha `$. For a physical proof of large $`N`$ duality, we need to solve this theory at least in the large $`N`$ planar limit. We will employ a strategy inspired from . Since we are working with linearized deformations, we can write the holomorphic Chern-Simons theory on $`\stackrel{~}{X}_\alpha `$ as a perturbation of holomorphic Chern-Simons on $`\stackrel{~}{X}`$. The latter can be reformulated in terms of a holomorphic gauge theory on the curve $`\mathrm{\Sigma }`$ by dimensional reduction. The effect of complex structure deformations of $`\stackrel{~}{X}`$ can then be taken into account by a perturbation of the gauge theory on $`\mathrm{\Sigma }`$ which is more tractable than holomorphic Chern-Simons theory on $`\stackrel{~}{X}_\alpha `$.
There is however an important subtlety in this approach. The holomorphic Chern-Simons theory only captures the open string background in a fixed closed string background. The dynamics of the full-open closed topological string theory should be described in terms of holomorphic Chern-Simons theory coupled to Kodaira-Spencer theory. Although we do not have a rigorous justification, we will assume that the Kodaira-Spencer theory decouples from the holomorphic Chern-Simons theory in the genus zero sector of the theory. Therefore if we are only interested in the large $`N`$ planar limit of the theory we can quantize open strings in a fixed closed string background. This assumption will be a posteriori justified by the results modulo a subtle caveat related to the integration measure for the open string theory which will be discussed in section 6.3.
### 6.1 Holomorphic Chern-Simons theory and twisted Higgs complexes
Following the general outline, let us start with holomorphic Chern-Simons theory on $`\stackrel{~}{X}`$. Recall that $`\stackrel{~}{X}`$ contains a ruled surface $`S`$ obtained by resolving the curve $`\mathrm{\Sigma }`$ of $`A_1`$ singularities of $`X`$. We consider a D-brane configuration consisting of $`N`$ branes and $`N`$ anti-branes wrapping fibers of $`S`$. For simplicity let us consider the generic case in which the branes and antibranes wrap distinct fibers of $`S`$. More precisely we specify two divisors
$$D_+=\underset{a=1}{\overset{N}{}}p_a,D_{}=\underset{a=1}{\overset{N}{}}q_a$$
on $`\mathrm{\Sigma }`$ so that the branes are supported on the fibers $`S_{p_a}`$ and the antibranes are supported on the fibers $`S_{q_a}`$. Generically we will have $`p_ap_b`$, $`q_aq_b`$ for any $`a,b=1\mathrm{}N`$, $`ab`$, and $`p_aq_b`$ for any $`a,b=1,\mathrm{},N`$.
The corresponding boundary topological B-model is given by the complex $`๐ฌ=๐ฌ^+๐ฌ^{}[1]`$ where
$$๐ฌ^+=_{a=1}^N๐ช_{S_{p_a}},๐ฌ^{}=_{a=1}^N๐ช_{S_{q_a}}.$$
The boundary chiral ring is isomorphic to the Ext algebra
$$_{k=0}^3\mathrm{Ext}_{\stackrel{~}{\mathrm{X}}}^\mathrm{k}(๐ฌ,๐ฌ)$$
In order to write down a physical action for off-shell fluctuations around this open string background, it is more convenient to work with a locally free resolution $``$ of $`๐ฌ`$. Since $``$ and $`๐ฌ`$ are quasi-isomorphic complexes, $``$ defines an equivalent boundary $`๐`$-model. Then the space of off-shell open string states is given by
$$๐_{\stackrel{~}{X}}=\underset{k=0}{\overset{3}{}}\underset{m,n}{}\mathrm{\Omega }_{\stackrel{~}{X}}^{0,k}(E_m^{}E_n).$$
Note that there is a integral ghost number grading on this vector space defined by $`k+(nm)`$. The physical states are elements of ghost number $`k+(nm)=1`$. The string field theory action for the physical states is a graded version of holomorphic Chern-Simons theory This action is not very tractable for concrete practical applications.
However, since $`\stackrel{~}{X}`$ is isomorphic to the total space of a line bundle over the ruled surface $`S`$ we can find a better starting point for large $`N`$ quantization invoking Koszul duality. Very briefly, in this situation Koszul duality establishes an equivalence between coherent sheaves on $`\stackrel{~}{X}`$ finite over $`S`$ and Higgs sheaves on $`S`$. For convenience, recall that a Higgs sheaf on $`S`$ is a pair $`(๐ฌ,\mathrm{\Phi })`$ where $`๐ฌ`$ is a coherent sheaf on $`S`$ and $`\mathrm{\Phi }:๐ฌ๐ฌK_S`$ is a morphism from $`๐ฌ`$ to $`๐ฌK_S`$. In general $`\mathrm{\Phi }`$ should satisfy an integrability condition which is empty in our particular case.
From a physical point of view, a Higgs sheaf on $`S`$ can be interpreted as a topological B-brane wrapping the surface $`S\stackrel{~}{X}`$ as follows. For simplicity suppose that $`๐ฌ`$ is locally free and denote by $`Q`$ the underlying vector bundle. Then the data $`(S,Q)`$ determines a topological boundary B-model with a nilpotent BRST symmetry. Such models have been analyzed in great detail in . Their results will be very useful in the following. In particular, according to , the spectrum of B-model boundary physical states is realized as the limit of a local to global spectral sequence with second term
$$E_2^{p,q}=H^{0,p}(End(Q)\mathrm{\Lambda }^q(N_{S/X}))$$
which converges to $`\mathrm{Ext}_{\stackrel{~}{\mathrm{X}}}^\mathrm{k}(๐ฌ,๐ฌ)`$, $`k=p+q`$. Koszul duality implies that this spectral sequence collapses at the second term, and it has a canonical split filtration so that
$$\mathrm{Ext}_{\stackrel{~}{X}}^k(๐ฌ,๐ฌ)=_{p+q=k}H^{0,p}(End(Q)\mathrm{\Lambda }^q(N_{S/X})).$$
Since $`N_{S/\stackrel{~}{X}}K_S`$, this shows that instead of working with bundles on $`\stackrel{~}{X}`$, it suffices to consider bundles on the compact surface $`S`$ as long as we take into account the Higgs field data.
It is also helpful to discuss this data from the point of view of holomorphic Chern-Simons theory. Adopting a differential geometric point of view, we can think of $`Q`$ as a $`C^{\mathrm{}}`$ bundle on $`S`$ equipped with a connection $`A`$ satisfying the integrability condition $`F_A^{0,2}=0`$. Then the covariant Dolbeault operator $`\overline{}_A`$ determines a holomorphic structure on $`Q`$. The off-shell open string states of the boundary B-model are elements of the infinite dimensional vector space
(59)
$$๐_S=_{p=0}^2_{q=0}^1\mathrm{\Omega }_{\stackrel{~}{X}}^{0,p}(End(Q)\mathrm{\Lambda }^q(N_{S/X}))$$
In order to construct a string field action for the off-shell fluctuations around this background we define a DG-algebra structure on $`๐`$ as follows. We first construct the $``$-graded superalgebra
(60)
$$\mathrm{\Omega }_S=\left(_{p=0}^2\mathrm{\Omega }_S^{0,p}\right)_{\mathrm{\Omega }_S^{0,0}}\left(_{q=0}^1\mathrm{\Omega }_S^{0,0}(\mathrm{\Lambda }^q(N_{S/\stackrel{~}{X}}))\right)$$
where the two factors are the exterior algebras of the antiholomorphic cotangent bundle of $`S`$ and respectively the holomorphic normal bundle to $`S`$ in $`\stackrel{~}{X}`$. The grading $`\mathrm{deg}:\mathrm{\Omega }_S`$ is defined by
(61)
$$\mathrm{deg}(\omega )=p+q,\mathrm{for}\omega \mathrm{\Omega }_S^{0,p}(\mathrm{\Lambda }^q(N_{S/\stackrel{~}{X}})).$$
Since $`\mathrm{\Lambda }^1(N_{S/\stackrel{~}{X}})\mathrm{\Omega }_S^{2,0}`$, we have an isomorphism of graded vector spaces
(62)
$$\mathrm{\Omega }_S_{p=0}^2_{q=0}^1\mathrm{\Omega }_S^{2q,p}.$$
Using this isomorphism, the superalgebra structure on $`\mathrm{\Omega }_S`$ can be explicitly written in the form
(63)
$$\omega \omega ^{}=(1)^{p^{}q}\omega \omega ^{}$$
where $`\omega \mathrm{\Omega }_S^{2q,p}`$, $`\omega ^{}\mathrm{\Omega }_S^{2q^{},p^{}}`$. Then we give the space $`๐_S`$ defined in (59) a tensor product superalgebra structure of the form
(64)
$$๐_S=\mathrm{\Omega }_S_{\mathrm{\Omega }_S^{0,0}}\mathrm{\Omega }_S^{0,0}(End(Q))$$
where the last factor can be regarded as a superalgebra with trivial odd component. The grading (61) extends trivially to $`๐_S`$.
Next, note that the covariant Dolbeault operator with respect to the background connection $`A`$ defines a degree one differential operator $`\overline{}^{(0)}:๐_S๐_S`$ satisfying the Leibnitz rule
$$\overline{}^{(0)}(\omega \omega ^{})=(\overline{}^{(0)}\omega )\omega ^{}+(1)^{deg(\omega )}\omega (\overline{}^{(0)}\omega ^{})$$
for any $`\omega ,\omega ^{}๐_S`$. Therefore we obtain an associative DG-algebra structure on $`๐_S`$.
In order to write down a holomorphic Chern-Simons action for off-shell fluctuations we also need a trace $`\mathrm{tr}:๐`$, which in this case is given by
$$\mathrm{tr}=_S\mathrm{๐๐๐}$$
where $`\mathrm{๐๐๐}:๐\mathrm{\Omega }_S`$ is the supertrace. The string field action is defined for ghost number one fields $`\varphi ๐_S`$, $`\mathrm{deg}(\varphi )=1`$ which parameterize arbitrary deformations $`\overline{}^{(0)}\overline{}^{(0)}+\varphi `$ of the $`DG`$ structure on $`๐_S`$. More precisely, $`\varphi `$ can be written as a sum of homogeneous elements
(65) $`\varphi =\varphi ^{0,1}+\varphi ^{2,0},`$
where $`\varphi ^{0,1}\mathrm{\Omega }_S^{0,1}(End(Q))`$ is an arbitrary deformation of the background Dolbeault operator $`\overline{}^{(0)}`$. on $`Q`$ and $`\varphi ^{2,0}\mathrm{\Omega }_S^{0,0}(End(Q)N_{S/\stackrel{~}{X}})`$ is a Higgs field on $`S`$. These are the expected off-shell $`C^{\mathrm{}}`$ deformations of a topological $`๐`$-brane supported on $`S`$. Applying the general reasoning of to the present case, it follows that the string field action reduces to a holomorphic Chern-Simons action on $`S`$ of the form
(66)
$$๐_{CS}=_S\mathrm{๐๐๐}\left(\frac{1}{2}\varphi \overline{}^{(0)}\varphi +\frac{1}{3}\varphi ^3\right).$$
Substituting equation (65) in this expression we obtain
(67) $`๐_{CS}`$ $`={\displaystyle _S}\mathrm{Tr}\left(\varphi ^{2,0}(\overline{}^{(0)}\varphi ^{0,1}+\varphi ^{0,1}\varphi ^{0,1})\right)`$
$`={\displaystyle _S}\mathrm{Tr}\left(\varphi ^{2,0}F^{0,2}\right)`$
where $`F^{0,2}`$ is the $`(0,2)`$ component of the curvature of the deformed connection $`A+\varphi ^{0,1}`$. Note that the action $`๐_{CS}`$ is left invariant by gauge transformations of the string field $`\varphi `$ of the form
$$\delta \varphi =\overline{}^{(0)}\lambda +[\varphi ,\lambda ]$$
where $`\lambda ๐`$ is an arbitrary ghost number zero field.
The solutions to the equations of motion for $`\varphi `$ modulo gauge transformations parameterize deformations of the boundary topological B-model. Applying the variational principle to the action (66) yields the Maurer-Cartan equation
$$\overline{}^{(0)}\varphi +\varphi \varphi =0.$$
In components, we obtain
$$\overline{}\varphi ^{2,0}=0,F^{0,2}=0$$
hence the solutions are in one to one correspondence to Higgs bundle structures on a fixed underlying $`C^{\mathrm{}}`$ bundle $`Q`$. Gauge equivalent solutions correspond to isomorphic Higgs bundles, therefore we can conclude that deformations of the boundary B-model are in one to one correspondence to isomorphism classes of Higgs bundles.
In our case, we have to extend the above construction to an open string background specified by a complex $`๐ฌ`$ of coherent sheaves on $`S`$. We first pick a locally free resolution $``$ of $`๐ฌ`$ on $`S`$. Note that $`๐ฌ`$ and $``$ are quasi-isomorphic as complexes of coherent sheaves on $`\stackrel{~}{X}`$, therefore they define equivalent boundary $`๐`$-models. Then we construct a holomorphic Chern-Simons action on $`S`$ for off-shell fluctuations around this open string background following the same steps. The main difference is that we will have to take into account the $``$-grading of the complex $``$, as explained in a similar context in . The discussion is fairly general, so in the following we can take $``$ to be an arbitrary complex of locally free sheaves.
In differential-geometric language, the open string background is specified by a finite sequence of smooth complex bundles and maps
(68)
$$\mathrm{}E_{n1}\stackrel{e_{n,n1}}{}E_n\stackrel{e_{n+1,n}}{}E_{n+1}\mathrm{}$$
The bundles are equipped with background connections $`A_n`$ subject to the integrability condition $`F_{A_n}^{0,2}=0`$ for all $`n`$. Therefore the covariant Dolbeault operators $`\overline{}_n^{(0)}=\overline{}_{A_n}`$ define holomorphic structures on the bundles $`E_n`$. The maps $`e_{n+1,n}`$, $`n`$ are required to be holomorphic with respect to the resulting complex structures and satisfy the condition
$$e_{n+1,n}e_{n,n1}=0$$
for all $`n`$.
In this context, the space of off-shell open string states is given by
$$๐_S=_{p=0}^2_{q=0}^1_{m,n}\mathrm{\Omega }_S^{0,p}(\mathrm{Hom}(E_m,E_n)\mathrm{\Lambda }^q(N_{S/X}))$$
The ghost number grading is defined by $`\mathrm{deg}:๐_S`$,
$$\mathrm{deg}(\varphi )=p+q+(nm)$$
for $`\varphi \mathrm{\Omega }_S^{0,p}(\mathrm{Hom}(E_m,E_n)\mathrm{\Lambda }^q(N_{S/\stackrel{~}{X}}))`$. In order to define the correct superalgebra structure on $`๐_S`$, we have to regard the $``$-graded vector bundle $`\{E_n\}`$ as a $``$-graded supervector bundle $`\{\stackrel{ห}{E}_n\}`$ , where
$$\stackrel{ห}{E}_n=\{\begin{array}{cc}(E_n,0),& \mathrm{for}n\mathrm{even}\\ (0,E_n),& \mathrm{for}n\mathrm{odd}.\end{array}$$
Then we have a superalgebra $`\mathrm{\Omega }_S^{0,0}(End(\stackrel{ห}{E}))`$ where $`\stackrel{ห}{E}`$ is the supervector bundle $`\stackrel{ห}{E}=(E^+,E^{})`$ where
$$E^+=_nE_{2n},E^{}=_nE_{2n+1}.$$
The superalgebra structure on $`๐_S`$ is then defined by writing $`๐_S`$ as a tensor product of superalgebras
(69)
$$๐_S=\mathrm{\Omega }_S_{\mathrm{\Omega }_S^{0,0}}\mathrm{\Omega }_S^{0,0}(End(\stackrel{ห}{E})).$$
Note that the Dolbeault operators $`\{\overline{}_n^{(0)}\}`$ define a degree one differential operator $`\overline{}^{(0)}:๐_S๐_S`$ satisfying the Leibnitz rule with respect to superalgebra multiplication. Moreover the maps $`e_{n+1,n}:E_nE_{n+1}`$ define a degree one element $`e๐_S`$. Then we define the BRST operator $`D^{(0)}:๐_S๐_S`$ in the background specified by the complex (68) to be
$$D^{(0)}=\overline{}^{(0)}+e.$$
Given a field $`\varphi ๐_S`$, we have
$$D^{(0)}\varphi =\overline{}^{(0)}\varphi +[e,\varphi ]$$
where $`[,]`$ is the supercommutator
$$[\varphi ,\varphi ^{}]=\varphi \varphi ^{}(1)^{\mathrm{deg}(\varphi )\mathrm{deg}(\varphi ^{})}\varphi ^{}\varphi .$$
Now we can write down the graded holomorphic Chern-Simons action for ghost number one open string fields $`\varphi ๐_S`$
(70)
$$๐_{CS}=_S\mathrm{๐๐๐}\left(\frac{1}{2}\varphi D^{(0)}\varphi +\frac{1}{3}\varphi ^3\right).$$
This action is left invariant by infinitesimal gauge transformations of the form
(71)
$$\delta \varphi =(D^{(0)}+\varphi )\lambda $$
where $`\lambda `$ is an arbitrary ghost number zero element of $`๐_S`$. Note that the $`D=D^{(0)}+\varphi `$ is an arbitrary off-shell deformation of the BRST operator. The equations of motion derived from the holomorphic Chern-Simons action can be written in compact form
(72)
$$D^{(0)}\varphi +\varphi ^2=0.$$
which is equivalent with the integrability condition
$$D^2=0$$
for the deformed BRST operator. The solutions to these equations modulo gauge transformations parameterize deformations of the open string background specified by the complex (68). In the ungraded case, we identified these deformations with Higgs bundle structures on a fixed $`C^{\mathrm{}}`$ bundle up to isomorphism. The equations (72) yield a generalization of the Higgs bundle conditions which is better understood in the framework of D-brane categories, which we explain next.
So far we have been studying fluctuations around a fixed open string background. In principle one can consider more general situations in which we have several topological D-branes wrapping the surface $`S`$. In that case the algebraic structure of the resulting open string theory is encoded in a triangulated D-brane category . This category is a physical variant of the Bondal-Kapranov construction introduced in the context of cubic string field theory in .
In the present context, we start with a DG-category $`๐`$ given as follows. The objects of $`๐`$ are holomorphic vector bundles $`ES`$. The space of morphisms between two objects $`E,E^{}`$ is given by the complex
$`\mathrm{Hom}_๐(E,E^{})`$ $`=_{p=0}^2_{q=0}^1\mathrm{\Omega }_S^{0,p}(Hom(E,E^{})\mathrm{\Lambda }^q(N_{S/\stackrel{~}{X}}))`$
$`_{p=0}^2_{q=0}^1\mathrm{\Omega }_S^{2q,p}(Hom(E,E^{}))`$
where the grading is given by $`p+q`$ and the differential is given by the covariant Dolbeault operator $`\overline{}^{(0)}`$. Here $`E,E^{}`$ are regarded again as $`C^{\mathrm{}}`$ bundles equipped with $`(0,1)`$-connections.
Next we construct a new DG-category $`\stackrel{~}{๐}`$ by taking the shift completion of $`๐`$. The objects of $`\stackrel{~}{๐}`$ are pairs $`(E,n)`$, where $`E`$ is an object of $`C`$ and $`n`$. The space of morphisms between two objects $`(E,n)`$, $`(E^{},n^{})`$ is the shifted complex
$$\mathrm{Hom}_{\stackrel{~}{๐}}((E,n),(E^{},n^{}))=\mathrm{Hom}_๐(E,E^{})[nn^{}].$$
From a physical point of view, the integer $`n`$ represents the D-brane grading introduced in . For future reference, we should also keep in mind that composition of morphisms in $`\stackrel{~}{C}`$ is given by
(73)
$$(gf)_{\stackrel{~}{๐}}=(1)^{(k+nn^{})(n^{}n^{\prime \prime })}(gf)_๐,$$
for any $`f\mathrm{Hom}_{\stackrel{~}{๐}}^k((E,n),(E^{},n^{}))`$ and any $`g\mathrm{Hom}_{\stackrel{~}{๐}}^l((E^{},n^{}),(E^{\prime \prime },n^{\prime \prime })`$.
The D-brane category is the triangulated category $`\mathrm{Tr}(\stackrel{~}{๐})`$ of twisted complexes over $`\stackrel{~}{๐}`$ defined in . Twisted complexes over $`\stackrel{~}{๐}`$ are finite collections of objects $`\{(E_i,n_i)\}`$ and degree one morphisms $`\mathrm{\Phi }_{ji}\mathrm{Hom}_{\stackrel{~}{๐}}^1((E_i,n_i),(E_j,n_j))`$ satisfying the Maurer-Cartan equation
(74)
$$\overline{}^{(0)}\mathrm{\Phi }_{ji}+\underset{k}{}\mathrm{\Phi }_{jk}\mathrm{\Phi }_{ki}=0.$$
Twisted complexes form a DG-category denoted by $`\text{Pre-Tr}(\stackrel{~}{๐})`$ in . The morphisms between two such objects $`(E_i,n_i,\mathrm{\Phi }_{ji})`$ and $`(E_i^{},n_i^{},\mathrm{\Phi }_{ji}^{})`$ are DG-complexes
(75)
$$\mathrm{Hom}_{\text{Pre-Tr}(\stackrel{~}{๐})}^k((E_i,n_i,\mathrm{\Phi }_{ji}),(E_j^{},n_j^{}.\mathrm{\Phi }_{lj}^{}))=_{i,j}\mathrm{Hom}_{\stackrel{~}{๐}}^k((E_i,n_i),(E_j^{},n_j^{}))$$
The action of the differential on morphisms $`\eta \mathrm{Hom}_{\stackrel{~}{๐}}^l((E_i,n_i),(E_j^{},n_j^{}))`$ is defined by
(76)
$$d\eta =\overline{}^{(0)}\eta +\underset{k}{}\mathrm{\Phi }_{kj}^{}\eta (1)^k\eta \mathrm{\Phi }_{ik}.$$
Then one can obtain a triangulated category with same objects as $`\text{Pre-Tr}(\stackrel{~}{๐})`$ by taking the space of morphisms between two objects to be the degree zero cohomology of the complex (75),(76). For our purposes it is more convenient to work with the $``$ graded category obtained from $`\text{Pre-Tr}(\stackrel{~}{๐})`$ by taking the space of morphisms between two objects to be the full cohomology of the complex (75),(76). We will denote the resulting enriched triangulated category by $`\mathrm{Tr}(\stackrel{~}{๐})`$.
In order to see the connection between twisted complexes and graded holomorphic Chern-Simons theory, note that the solutions to the equation of motion (72) are in fact diagonal twisted complexes characterized by $`n_i=i`$ for all $`i`$.
More precisely, if $`\varphi ๐^1`$ is a solution to (72), one can easily check that $`\mathrm{\Phi }=\varphi +e`$ satisfies the equation
(77)
$$\overline{}^{(0)}\mathrm{\Phi }+\mathrm{\Phi }^2=0.$$
By construction the collection of fields
$$\mathrm{\Phi }_{nm}\mathrm{\Omega }_S^{2q,p}(Hom(E_m,E_n))$$
with $`p+q=1+mn`$ can be regarded as morphisms
$$\mathrm{\Phi }_{nm}\text{Hom}_{\stackrel{~}{๐}}^1(E_m,E_m)\text{Hom}_๐^{1+mn}(E_m,E_n).$$
between the bundles $`E_m,E_n`$ in the category $`\stackrel{~}{๐}`$. Moreover, the equation of motion (77) is identical to the Maurer-Cartan equation (74). In order to check this equivalence, notice that the sign rule (73) for composition of morphisms in $`\stackrel{~}{๐}`$ is compatible with multiplication in the superalgebra (69). Therefore the collection $`\{E_n,\mathrm{\Phi }_{nm}\}`$ is a twisted complex with $`n_i=i`$.
Keeping this correspondence in mind, from now on we will refer to the objects of $`\mathrm{Tr}(\stackrel{~}{๐})`$ as twisted Higgs complexes.
In order to apply this construction to D-branes wrapping fibers of the ruling $`S\mathrm{\Sigma }`$, recall that we have canonical locally free resolutions of divisors on $`S`$
$`0๐ช_S\left({\displaystyle \underset{a=1}{\overset{N}{}}}S_{p_a}\right)\stackrel{e_1}{}๐ช_S_{a=1}^N๐ช_{S_{p_a}}0`$
$`0๐ช_S\left({\displaystyle \underset{a=1}{\overset{N}{}}}S_{q_a}\right)\stackrel{e_2}{}๐ช_S_{a=1}^N๐ช_{S_{q_a}}0.`$
Then we can construct a complex $``$ of locally free sheaves which is quasi-isomorphic to $`๐ฌ`$ and has the form
(78)
The graded holomorphic Chern-Simons action we are searching for given by (70) in which the locally free complex $``$ is taken to be (78).
According to the general approach explained in section 2, we need to understand the moduli space of solutions to the equations of of motion of the Chern-Simons action modulo gauge transformations. In this case the problem can be further simplified by taking dimensional reduction of the action along the fibers of the ruling $`q:S\mathrm{\Sigma }`$. More precisely, we pick up a Kรคhler metric on $`S`$ so that the volume of the fibers is very small compared to the volume of the base. Then we can reduce the Chern-Simons action along the fibers obtaining a two-dimensional field theory. Strictly speaking this procedure is employed in the physics literature only when $`S`$ is a direct product $`S=\mathrm{\Sigma }\times ^1`$. In fact this restriction is too severe. Dimensional reduction can be applied equally well in all cases when the projective bundle $`S=(V)`$, where $`V\mathrm{\Sigma }`$ is a rank $`2`$ holomorphic bundle, having locally constant transition functions . This will be the case if $`V\mathrm{\Sigma }`$ is a polystable holomorphic bundle. For simplicity, we will assume in the following that $`S`$ is a direct product, that is $`V`$ is a trivial rank $`2`$ bundle. If $`V`$ is non-trivial, but polystable, the result remains unchanged.
Let $`=q^{}`$ be the pull-back of a complex of holomorphic vector bundles $``$
(79)
$$\mathrm{}F_{n1}\stackrel{f_{n,n1}}{}F_n\stackrel{f_{n+1,n}}{}F_{n+1}\mathrm{}$$
As above the bundles $`F_n`$ are regarded as $`C^{\mathrm{}}`$ bundles equipped with $`(0,2)`$ connections $`B_n`$ which determine covariant Dolbeault operators $`\overline{}_n`$. In order to perform the dimensional reduction we write the off-shell field $`\varphi =\varphi _{n,m}^{2q,p}๐`$ as
(80)
$$\varphi _{n,m}^{2q,p}=\psi _{n,m}^{2q,p}\eta ^{0,0}+\chi _{n,m}^{2q1,p}\eta ^{1,0}+\chi _{n,m}^{2q,p1}\eta ^{0,1}+\psi _{n,m}^{2q1,p1}\eta ^{1,1}$$
where $`\psi _{n,m}^{k,l},\chi _{n,m}^{k,l}\mathrm{\Omega }_\mathrm{\Sigma }^{k,l}(Hom(F_m,F_n))`$, $`\eta ^{r,s}\mathrm{\Omega }_^1^{r,s}`$. In the right hand side of (80), all forms should be pulled back to $`S`$. Next, we take $`\eta ^{r,s}`$ to be solutions to the linearized equations of motion modulo gauge transformations. Since the background complex is pulled back form $`\mathrm{\Sigma }`$, the linearized equations of motion read
$$\overline{}\eta ^{r,s}=0$$
for any $`r,s=0,1`$. Therefore the space of solutions to the linearized equations of motion modulo gauge transformations is parameterized by $`H^{r,s}(^1)`$. Let us choose generators $`1H^{0,0}(^1)`$, $`[\eta ]H^{1,1}(^1)`$ of the nontrivial cohomology groups. Then the dimensional reduction ansatz (80) reduces to
(81)
$$\varphi _{n,m}^{2q,p}=\psi _{n,m}^{2q,p}+\eta \psi _{n,m}^{2q1,p1}.$$
Therefore we obtain the following space of off-shell fields on $`\mathrm{\Sigma }`$
(82)
$$=_{q=0}^1_{p=0}^1_{m,n}\mathrm{\Omega }_\mathrm{\Sigma }^{q,p}\left(Hom(F_m,F_n)\right)$$
where the ghost number grading is given by $`\text{deg}(\psi _{n,m}^{q,p})=2q+p+(nm)`$. We can give $``$ a $``$-graded superalgebra structure by adopting the construction used below equation (68). Let $`\mathrm{\Omega }_\mathrm{\Sigma }`$ be the $``$-graded superalgebra obtained by reducing $`\mathrm{\Omega }_S`$ along the fibers of the ruling. We have
$$\mathrm{\Omega }_\mathrm{\Sigma }=_{q=0}^1_{p=0}^1\mathrm{\Omega }_\mathrm{\Sigma }^{q,p}$$
where the multiplication is defined by
$$\omega \omega ^{}=(1)^{p^{}q}\omega \omega ^{}.$$
Let $`\stackrel{ห}{F}=(F^+,F^{})`$ be the supervector bundle obtained by rolling the $``$-graded vector space $`_nF_n`$. Then we take
(83)
$$=\mathrm{\Omega }_\mathrm{\Sigma }_{\mathrm{\Omega }_\mathrm{\Sigma }^{0,0}}\mathrm{\Omega }_\mathrm{\Sigma }^{0,0}(End(\stackrel{ห}{F})).$$
The background connections $`B_n`$ together with the maps $`f_{n+1,n}:F_nF_{n+1}`$ define a differential operator $`D^{(0)}=\overline{}^{(0)}+[f,]:`$ satisfying the Leibnitz rule. By dimensional reduction, the Chern-Simons action (70) yields the following action on $`\mathrm{\Sigma }`$
(84)
$$๐=_\mathrm{\Sigma }\mathrm{๐๐๐}\left(\frac{1}{2}\psi D^{(0)}\psi +\frac{1}{3}\psi ^3\right).$$
This is again left invariant by infinitesimal gauge transformations of the form
$$\delta \psi =D\lambda $$
where $`\lambda `$ is an arbitrary ghost number zero element and $`D=D^{(0)}+\psi `$. The equations of motion of this action are again of the form
(85)
$$D^{(0)}\psi +\psi ^2=0.$$
In order to facilitate the construction of the moduli space of solutions to the equations (85) modulo gauge transformations, it is very helpful to rephrase this construction in terms of D-brane categories. One can construct a triangulated category of twisted complexes on $`\mathrm{\Sigma }`$ by performing dimensional reduction on twisted complexes on $`S`$. More precisely, let us start with the DG category $`๐`$ of holomorphic vector bundles $`F\mathrm{\Sigma }`$ so that the space of morphisms between two objects $`F,F^{}`$ is given by the complex
(86)
$$\text{Hom}_๐(F,F^{})=_{q=0}^1_{p=0}^1\mathrm{\Omega }^{q,p}(Hom(F,F^{})).$$
The grading is defined by $`p+2q`$ and the differential is given by the covariant Dolbeault operator as in the previous case. The D-brane category in question is the enriched triangulated category $`\mathrm{Tr}(\stackrel{~}{๐})`$ of twisted complexes on $`\mathrm{\Sigma }`$ associated to the shift completion $`\stackrel{~}{๐}`$. It is a straightforward exercise to check that twisted complexes on $`\mathrm{\Sigma }`$ can be obtained by dimensional reduction of twisted complexes on $`S`$.
This concludes our discussion of holomorphic D-branes on $`\stackrel{~}{X}`$ from a classical point of view. In order to reach our goal we have to understand the quantum dynamics of holomorphic branes on $`\stackrel{~}{X}`$, at least in the large $`N`$ limit.
### 6.2 Quantization and moduli space
As explained in section 2, the quantization of the holomorphic Chern-Simons theory involves a formal integral of a holomorphic measure on a middle dimensional real cycle in the space of fields. Exploiting the topological symmetry of this theory, this functional integral localizes to a finite dimensional integral on a middle dimensional cycle in the moduli space $``$ of solutions to the classical field equations of motion.
We would like to carry out this construction for the holomorphic Chern-Simons action associated to the complex (78). Since this complex is pulled back from $`\mathrm{\Sigma }`$, the classical moduli space can be determined using dimensional reduction and truncation to zero modes. Therefore it suffices to consider the cubic action (84) for a complex of the form
(87)
$$0๐ช_\mathrm{\Sigma }(D_+)\stackrel{\left(\genfrac{}{}{0pt}{}{f_1}{0}\right)}{}๐ช_\mathrm{\Sigma }๐ช_\mathrm{\Sigma }(D_{})\stackrel{(f_2\mathrm{\hspace{0.17em}0})}{}๐ช_\mathrm{\Sigma }0$$
where $`D_+=_{a=1}^Np_a`$, $`D_{}=_{a=1}^Nq_a`$ are disjoint divisors on $`\mathrm{\Sigma }`$ and
$`0๐ช_\mathrm{\Sigma }(D_+)\stackrel{f_1}{}๐ช_\mathrm{\Sigma }๐ช_{D_+}0`$
$`0๐ช_\mathrm{\Sigma }(D_{})\stackrel{f_2}{}๐ช_\mathrm{\Sigma }๐ช_D_{}0`$
are canonical maps. On common grounds the functional integral (4) reduces to an integral over a middle dimensional real cycle on the moduli space $``$ of critical points of the action modulo gauge transformations. The holomorphic measure of the moduli space integral should be determined in principle by integrating out the massive modes, provided that the original measure $`D\psi `$ is at least formally well defined. A direct approach to this problem is beyond the purpose of the present paper so we will employ a different technique. We first determine the moduli space and then find an expression for the measure using holomorphy and physical constraints. In the process we will discover a new aspect of this problem involving spin structures on $`\mathrm{\Sigma }`$.
In order to find the moduli space it is convenient to write the action (84) in terms of the field $`\mathrm{\Psi }=f+\psi \psi ^{0,1}`$ and the Dolbeault operator $`\overline{}=\overline{}^{(0)}+\psi ^{0,1}`$. Then we have
(88)
$$S_{CS}=_S\mathrm{Tr}_s\left(\mathrm{\Psi }_{01}^{1,0}\overline{}\mathrm{\Psi }_{10}^{0,0}+\mathrm{\Psi }_{12}^{1,0}\overline{}\mathrm{\Psi }_{21}^{0,0}+\mathrm{\Psi }_{02}^{1,1}\mathrm{\Psi }_{21}^{0,0}\mathrm{\Psi }_{10}^{0,0}\right).$$
The equations of motion (85) become
(89) $`\overline{}_{10}\mathrm{\Psi }_{10}^{0,0}=0`$ $`\mathrm{\Psi }_{21}^{0,0}\mathrm{\Psi }_{10}^{0,0}=0`$
$`\overline{}_{21}\mathrm{\Psi }_{21}^{0,0}=0`$ $`\mathrm{\Psi }_{01}^{1,0}\mathrm{\Psi }_{10}^{0,0}=0`$
$`\overline{}_{01}\mathrm{\Psi }_{01}^{1,0}+\mathrm{\Psi }_{02}^{1,1}\mathrm{\Psi }_{21}^{0,0}=0`$ $`\mathrm{\Psi }_{10}^{0,0}\mathrm{\Psi }_{01}^{1,0}\mathrm{\Psi }_{12}^{1,0}\mathrm{\Psi }_{21}^{0,0}=0`$
$`\overline{}_{21}\mathrm{\Psi }_{21}^{1,0}+\mathrm{\Psi }_{10}^{0,0}\mathrm{\Psi }_{02}^{1,1}=0`$ $`\mathrm{\Psi }_{21}^{0,0}\mathrm{\Psi }_{12}^{1,0}=0`$
As explained above, our goal is to determine the moduli space of solutions to the equations of motion modulo gauge transformations. A better formulation of the problem can be achieved in the framework of D-brane categories developed in the previous subsection. We have to find the moduli space of three term twisted complexes of the form $`๐ฏ=(F_i,\mathrm{\Psi }_{ji})`$, $`i,j=0,1,2`$ where
$$F_0=๐ช_\mathrm{\Sigma }(D_+),F_1=๐ช_\mathrm{\Sigma }๐ช_\mathrm{\Sigma }(D_{}),F_2=๐ช_\mathrm{\Sigma }$$
as $`C^{\mathrm{}}`$-bundles. By convention we fix the degrees of $`F_0,F_1,F_2`$ to be $`1,0,1`$. Two such twisted complexes are said to be equivalent if there exist morphisms $`\eta H_{\mathrm{Tr}(\stackrel{~}{๐})}^0(๐ฏ,๐ฏ^{})`$, $`\rho H_{\mathrm{Tr}(\stackrel{~}{๐})}^0(๐ฏ^{},๐ฏ)`$ such that $`\rho \eta =๐_๐ฏ`$ and $`\eta \rho =๐_๐ฏ^{}`$.
In order to formulate a well defined moduli problem, we have to translate this data into algebraic-geometric language. We start with the three term twisted complexes $`๐ฏ`$. The fields $`\mathrm{\Psi }_{m,m}^{01,}`$, $`m=0,1,2`$ correspond to integrable complex structures on the bundles $`F_m`$, $`m=0,1,2`$. We will denote by $`_m`$, $`m=0,1,2`$ the associated locally free sheaves. $`\mathrm{\Psi }_{m+1,m}^{0,0}:_m_{m+1}`$ determine morphisms $`\mathrm{\Psi }_{10}:_0_1`$, $`\mathrm{\Psi }_{21}:_1_2`$ of locally free sheaves which compose to zero. Therefore we obtain a complex of locally free sheaves
(90)
$$0_0\stackrel{\mathrm{\Psi }_{10}}{}_1\stackrel{\mathrm{\Psi }_{21}}{}_20.$$
The remaining fields $`\mathrm{\Psi }_{m,m+1}^{1,0}`$, $`m=0,1,2`$, $`\mathrm{\Psi }_{02}^{1,1}`$ can be interpreted as follows. $`\mathrm{\Psi }_{02}^{1,1}`$ determines an extension
(91)
$$0_0๐ช(K_\mathrm{\Sigma })๐ฎ_20.$$
and $`\mathrm{\Psi }_{01}^{1,0}`$, $`\mathrm{\Psi }_{12}^{1,0}`$ determine splittings
(92)
(93)
where $`\mathrm{\Psi }_{12}^{}๐ฎ`$, $`\mathrm{\Psi }_{10}๐ฎ`$ denote the pullback and respectively pushforward extensions. For further reference note that the splitting $`\mathrm{\Psi }_{01}`$ induces a map $`\stackrel{~}{\mathrm{\Psi }}_{21}:_1๐ฎ`$ lifting $`\mathrm{\Psi }_{21}:_1_2`$. The splitting $`\mathrm{\Psi }_{12}`$ induces a canonical projection $`\mathrm{\Pi }_{12}:\mathrm{\Psi }_{10}๐ฎ_1๐ช(K_\mathrm{\Sigma })`$ which induces in turn a map $`\stackrel{~}{\mathrm{\Psi }}_{10}:๐ฎ_1๐ช(K_\mathrm{\Sigma })`$ extending $`\mathrm{\Psi }_{10}๐_{๐ช(K_\mathrm{\Sigma })}:_0๐ช(K_\mathrm{\Sigma })_1๐ช(K_\mathrm{\Sigma })`$.
In order to interpret the remaining constraints
(94) $`\mathrm{\Psi }_{01}^{1,0}\mathrm{\Psi }_{10}^{0,0}=0`$
$`\mathrm{\Psi }_{10}^{0,0}\mathrm{\Psi }_{01}^{1,0}\mathrm{\Psi }_{12}^{1,0}\mathrm{\Psi }_{21}^{0,0}=0`$
$`\mathrm{\Psi }_{21}^{0,0}\mathrm{\Psi }_{12}^{1,0}=0`$
note for example that $`\mathrm{\Psi }_{10}:_0_1`$ induces a splitting
The first equation in (94) is equivalent to the condition
(95)
$$\mathrm{\Psi }_{10}^{}\mathrm{\Psi }_{21}^{}๐ฎ=_0_0๐ช(K_\mathrm{\Sigma })\text{and}\mathrm{\Psi }_{10}^{}\mathrm{\Psi }_{01}=(๐__0,0).$$
Similarly, $`\mathrm{\Psi }_{21}:_1_2`$ induces a splitting
The last equation is equivalent to
(96)
$$\left(\mathrm{\Psi }_{21}๐_{๐ช(K_\mathrm{\Sigma })}\right)_{}\mathrm{\Psi }_{10}๐ฎ=_2_2๐ช(K_\mathrm{\Sigma })\text{and}\left(\mathrm{\Psi }_{21}๐_{๐ช(K_\mathrm{\Sigma })}\right)_{}\mathrm{\Psi }_{12}=(๐__2,0).$$
Finally, in order to find the algebraic interpretation of the second equation in (94) note that we have two induced splittings
Note also that the extensions $`\left(\mathrm{\Psi }_{10}๐_{๐ช(K_\mathrm{\Sigma })}\right)_{}\mathrm{\Psi }_{21}^{}๐ฎ`$ and $`\mathrm{\Psi }_{21}^{}\mathrm{\Psi }_{10}๐ฎ`$ are canonically isomorphic. Then the middle equation in (94) is equivalent to the condition that the two splittings agree
(97)
$$\left(\mathrm{\Psi }_{10}๐_{๐ช(K_\mathrm{\Sigma })}\right)_{}\mathrm{\Psi }_{01}=\mathrm{\Psi }_{21}^{}\mathrm{\Psi }_{12}.$$
To summarize, we have shown that the data given by a three term twisted complex $`๐ฏ=(_i,\mathrm{\Psi }_{ji})`$, $`i,j=0,1,2`$ on $`\mathrm{\Sigma }`$ is equivalent to the following algebraic data
$`i)`$ a three term complex $``$ of locally free sheaves (90),
$`ii)`$ an extension of the form (91), and
$`iii)`$ splittings (92), (93) satisfying conditions (95), (96) and (97).
Next we have to specify equivalence relations between two sets of algebraic data. This will be achieved by finding an algebraic description for the morphism space between three term twisted complexes in $`\mathrm{Tr}(\stackrel{~}{๐})`$. Let $`(_i,\mathrm{\Psi }_{02},\mathrm{\Psi }_{01},\mathrm{\Psi }_{12})`$ and $`(_i^{},\mathrm{\Psi }_{02}^{},\mathrm{\Psi }_{01}^{},\mathrm{\Psi }_{12}^{})`$ $`i=0,1,2`$ be two sets of algebraic data satisfying conditions $`(i)(iii)`$ above.
Given the data $`(_i,\mathrm{\Psi }_{ji})`$ as above, we construct the five term complex $`๐ฆ`$
(98)
$$_0\stackrel{\mathrm{\Psi }_{10}}{}_1\stackrel{\stackrel{~}{\mathrm{\Psi }}_{21}}{}๐ฎ\stackrel{\stackrel{~}{\mathrm{\Psi }}_{10}}{}_1๐ช(K_\mathrm{\Sigma })\stackrel{\mathrm{\Psi }_{21}๐_{๐ช(K_\mathrm{\Sigma })}}{}_2๐ช(K_\mathrm{\Sigma }).$$
in which the first term has degree $`2`$. We also have a similar complex $`๐ฆ^{}`$ for the data $`(_i^{},\mathrm{\Psi }_{ji}^{})`$
(99)
$$_0^{}\stackrel{\mathrm{\Psi }_{10}^{}}{}_1^{}\stackrel{\stackrel{~}{\mathrm{\Psi }}_{21}^{}}{}๐ฎ^{}\stackrel{\stackrel{~}{\mathrm{\Psi }}_{10}^{}}{}_1^{}๐ช(K_\mathrm{\Sigma })\stackrel{\mathrm{\Psi }_{21}^{}๐_{๐ช(K_\mathrm{\Sigma })}}{}_2^{}๐ช(K_\mathrm{\Sigma }).$$
Note that we have short exact sequences of complexes
(100) $`0๐ช(K_\mathrm{\Sigma })[2]๐ฆ0`$
$`0^{}๐ช(K_\mathrm{\Sigma })[2]๐ฆ^{}^{}0`$
Applying the functor $`Hom(,^{})๐ช(K_\mathrm{\Sigma })[2]`$ to the first complex in (100), we obtain the short exact sequence of complexes
(101)
$$0Hom(,^{})๐ช(K_\mathrm{\Sigma })[2]Hom(๐ฆ๐ช(K_\mathrm{\Sigma })[2],^{})Hom(,^{})0.$$
Applying the functor $`Hom(,)`$ to the second complex in (100) we obtain the short exact sequence of complexes
(102)
$$0Hom(,^{})๐ช(K_\mathrm{\Sigma })[2]Hom(,๐ฆ^{})Hom(,^{})0.$$
Therefore we have produced two extensions of the complex $`Hom(,^{})`$ by $`Hom(,^{})๐ช(K_\mathrm{\Sigma })[2]`$. Next we construct a new complex $`C(๐ฆ,๐ฆ^{})`$ by taking the difference of the two extensions. This results in a five term complex whose terms are explicitly computed in appendix A. We claim the $``$ graded space of morphisms between the twisted complexes $`๐ฏ,๐ฏ^{}`$ in $`\mathrm{Tr}(\stackrel{~}{๐})`$ is isomorphic to the hypercohomology of $`C(๐ฆ,๐ฆ^{})`$ on $`\mathrm{\Sigma }`$
(103)
$$H_{\mathrm{Tr}(\stackrel{~}{๐})}^{}(๐ฏ,๐ฏ^{})^{}(C(๐ฆ,๐ฆ^{})).$$
The proof of this assertion reduces to a rather lengthly homological algebra computation which is performed in appendix A.
Using the algebraic formulation we can set up a well defined moduli problem for a three term complex $``$ of the form (87) in the category of twisted complexes. In principle, one can construct a moduli stack associated to this moduli problem, but we will perform such a construction here. For a semiclassical analysis it suffices to identify the irreducible component of the moduli space which contains equivalence classes of complexes of the form $``$ up to birational equivalence. This is an easier task, which can be accomplished as follows.
First note that we can construct a family of equivalence classes of complexes $``$ in the category of twisted complexes by varying the divisors $`D_+,D_{}`$ on $`\mathrm{\Sigma }`$. This gives rise to a map $`\varphi :\text{Sym}^N(\mathrm{\Sigma })^2`$ from $`\text{Sym}^N(\mathrm{\Sigma })^2`$ to the moduli space sending a point $`(D_+,D_{})(\mathrm{\Sigma })^2`$ to a three term complex of the form (87).
Now let us compute the space of infinitesimal first order deformations $`H_{\mathrm{Tr}(\stackrel{~}{๐})}^1(,)`$ of the complex $``$ in the category of twisted complexes. According to equation (103), we have
$$H_{\mathrm{Tr}(\stackrel{~}{๐})}^1(,)^1(C(๐ฆ,๐ฆ))$$
where $`๐ฆ,๐ฆ^{}`$ are the five term complexes defined in (98),(99). Since $`\mathrm{\Psi }_{02}=0`$ in this case, the extension (91) is canonically split. Then one can check that
$$๐ฆ=๐ช(K_\mathrm{\Sigma })[2]$$
and
$$C(๐ฆ,๐ฆ)=End()End()๐ช(K_\mathrm{\Sigma })[2].$$
Therefore we have to evaluate the hypercohomology group
$$^1(End()End()๐ช(K_\mathrm{\Sigma })[2])=\text{Hom}_{D^b(\mathrm{\Sigma })}(,[1])\text{Hom}_{D^b(\mathrm{\Sigma })}(,๐ช(K_\mathrm{\Sigma })[1]).$$
The complex $``$ is quasi-isomorphic to its cohomology $`๐ช_{D_+}๐ช_D_{}[1]`$, hence we have
(104) $`\text{Hom}_{D^b(\mathrm{\Sigma })}(,[1])\text{Hom}_{D^b(\mathrm{\Sigma })}(,๐ช(K_\mathrm{\Sigma })[1])=`$
$`\text{Hom}_{D^b(\mathrm{\Sigma })}(๐ช_{D_+}๐ช_D_{}[1],๐ช_{D_+}[1]๐ช_D_{})`$
$`\text{Hom}_{D^b(\mathrm{\Sigma })}(๐ช_{D_+}๐ช_D_{}[1],๐ช_{D_+}(K_\mathrm{\Sigma })[1]๐ช_D_{}(K_\mathrm{\Sigma })[2])=`$
$`\text{Hom}_{D^b(\mathrm{\Sigma })}(๐ช_{D_+},๐ช_{D_+}[1])\text{Hom}_{D^b(\mathrm{\Sigma })}(๐ช_D_{},๐ช_D_{}[1]).`$
Using Serre duality, one can show that all other terms in the right hand side of the above equation vanish provided that $`D_+,D_{}`$ have disjoint supports. Therefore, if $`\text{Supp}(D_+)\text{Supp}(D_{})=\mathrm{}`$, we find that
$$^1(C(๐ฆ,๐ฆ))=\text{Ext}^1(๐ช_{D_+},๐ช_{D_+})\text{Ext}^1(๐ช_D_{},๐ช_D_{}),$$
which is the space of infinitesimal deformations of the pair $`(D_+,D_{})`$ i.e. the tangent space $`T_{(D_+,D_{})}\text{Sym}^N(\mathrm{\Sigma })^2`$.
Let $`\mathrm{\Delta }_\pm `$ denote the big diagonals in the two factors of the direct product $`\text{Sym}^N(\mathrm{\Sigma })\times \text{Sym}^N(\mathrm{\Sigma })`$; the points in $`\text{Sym}^N(\mathrm{\Sigma })\mathrm{\Delta }_\pm `$ parameterize effective divisors $`D_\pm `$ with distinct simple points. We will also denote by
$$\mathrm{\Delta }\text{Sym}^N(\mathrm{\Sigma })^2,\mathrm{\Delta }=\{(D_+,D_{})\text{Sym}^N(\mathrm{\Sigma })^2|\text{Supp}(D_+)\text{Supp}(D_{})\mathrm{}\}.$$
the big diagonal of the direct product.
The above considerations show that $`\varphi `$ is an isomorphism between the open subset
$$\left(\text{Sym}^N(\mathrm{\Sigma })\times \text{Sym}^N(\mathrm{\Sigma })\right)\left(\mathrm{\Delta }_+\times \text{Sym}^N(\mathrm{\Sigma })\text{Sym}^N(\mathrm{\Sigma })\times \mathrm{\Delta }_{}\mathrm{\Delta }\right)$$
and an open subset of the moduli space $``$. Note that $`\varphi `$ is also well defined and induces an isomorphism of virtual tangent spaces along the divisor
$$\mathrm{\Delta }_+\times \text{Sym}^N(\mathrm{\Sigma })\text{Sym}^N(\mathrm{\Sigma })\times \mathrm{\Delta }_{}.$$
However, $`\varphi `$ does not extend to this divisor as in isomorphism of stacks, since the moduli space $``$ is a higher stack. In principle, it could extend as an isomorphism of stacks in the presence of a suitable stability condition on twisted complexes which would make $``$ an stack. We will not attempt to formulate such a condition here, but we will assume that with right choice of a stability condition, we can identify an open subset of the D-brane moduli space with the open subset
$$_0=[\mathrm{\Sigma }^N/S_N]\times [\mathrm{\Sigma }^N/S_N]\mathrm{\Delta },$$
where $`[\mathrm{\Sigma }^N/S_N]`$ denotes the (stacky) orbifold quotient of the cartesian product $`\mathrm{\Sigma }^N`$ by the action of the symmetric group $`S_N`$.
### 6.3 The measure
The next step in the quantization process requires a holomorphic measure on the moduli space. In the following we will concentrate on the open subset $`_0`$ of the moduli space constructed in the previous subsection. Note that $`_0`$ has a finite cover of the form
(105)
$$_0^{}=\left(\mathrm{\Sigma }^N\times \mathrm{\Sigma }^N\right)\mathrm{\Delta }$$
where $`\mathrm{\Delta }`$ is defined again as the divisor of $`\mathrm{\Sigma }^N\times \mathrm{\Sigma }^N`$ where a point $`p_a`$ coincides with a point $`q_b`$ for some $`a,b=1,\mathrm{},N`$. In fact $`_0`$ is a quotient of $`_0^{}`$ by the obvious action of $`S_N\times S_N`$ which from a physical point of view should be thought of as a residual discrete gauge action.
Following the common practice in gauge theories, we will write the measure on a finite cover of the moduli space of the form and divide the resulting functional integral by $`|S_N\times S_N|=(N!)^2`$.
The restriction of the measure to $`_0^{}`$ should be a holomorphic $`2N`$ differential form. Since $`_0`$ is isomorphic to the complement of the diagonal $`\mathrm{\Delta }`$ in $`\mathrm{\Sigma }^{2N}`$, any such form can be extended to a meromorphic $`2N`$ form $`\mathrm{\Omega }`$ on $`\mathrm{\Sigma }^{2N}`$.
In principle one should be able to derive a formula for $`\mathrm{\Omega }`$ starting from the path integral formulation of holomorphic Chern-Simons theory. However, this approach may be quite cumbersome in practice, so it is more economical to find a formula for $`\mathrm{\Omega }`$ based on holomorphy and physical constraints. The main idea is that physical constraints specify the polar structure of $`\mathrm{\Omega }`$ along the divisor
(106)
$$\left(\mathrm{\Delta }_+\times \mathrm{\Sigma }^N\right)\left(\mathrm{\Sigma }^N\times \mathrm{\Delta }_{}\right)\mathrm{\Delta }.$$
Although these conditions do not determine $`\mathrm{\Omega }`$ uniquely, with some additional physical insight we can write down the measure uniquely up to a scale.
The physical constraints on $`\mathrm{\Omega }`$ are imposed by the universal character of the local effective interactions among branes at very short distances. This means that when the branes are very close to each other the dominant interaction terms are identical to their local counterparts discussed in section two. Therefore we should have Coulomb repulsion between two branes or two anti-branes approaching each other and also Coulomb attraction between a brane anti-brane pair . This means that $`\mathrm{\Omega }`$ should have a zero of order two along each divisor in $`_0^{}`$ given by $`p_a=p_b`$ or $`q_a=q_b`$ for any $`a,b=1,\mathrm{},N`$, $`ab`$ and a pole of order two along divisors given by $`p_a=q_b`$ for all $`a,b=1,\mathrm{},N`$. It is straightforward to check that any such differential must have extra zeroes on $`_0`$. These zeroes do not have a direct physical interpretation.
Although a natural meromorphic form $`\mathrm{\Omega }`$ with these properties does not exist, we can construct such a preferred form once we choose a spin structure on $`\mathrm{\Sigma }`$. More precisely, let us pick a theta-characteristic $`ฯต`$, and take $`\mathrm{\Omega }`$ to be the square of the free fermion correlator
(107)
$$\psi (p_1)\mathrm{}\psi (p_N)\overline{\psi }(q_1)\mathrm{}\overline{\psi }(q_N).$$
where $`\psi `$ is a complex spinor on $`\mathrm{\Sigma }`$ with respect to the spin structure $`ฯต`$. This is a well defined meromorphic top form on $`\mathrm{\Sigma }^N\times \mathrm{\Sigma }^N`$ exhibiting the physical behavior explained in the last paragraph.
Now, one may legitimately ask why this peculiar construction is the correct moduli space measure for the holomorphic Chern-Simons theory. A short answer to this question is that although the physical constraints do not fix the measure uniquely, they do fix the relevant part of the measure in the large $`N`$ planar limit. That is the structure of the semiclassical large $`N`$ vacua of the theory is insensitive to changing the measure by adding a holomorphic top form without zeroes or poles along the divisor (106). This will be manifest in the calculations performed in the next section. Therefore this choice of the measure is as good as any other choice exhibiting identical polar structure along the divisor (106).
However this is not a conceptually satisfying answer since the full quantum theory involves much more than the large $`N`$ planar limit. To gain a different perspective on this problem, recall that so far we have ignored the coupling between holomorphic Chern-Simons theory and Kodaira-Spencer theory. Indeed, although this coupling does not play an important role in the large $`N`$ planar limit, it is certainly expected to play an important role in the full quantum theory. Without getting into too much detail at this point, let us mention that in the present set-up Kodaira-Spencer theory can be described in terms of a chiral boson on the Riemann surface $`\mathrm{\Sigma }`$ by analogy with the examples considered in . It is well known that in order to define the theory of a chiral boson at quantum level, one has to choose a spin structure on $`\mathrm{\Sigma }`$. Moreover, the branes should be regarded as fermion operators in this theory related to the Kodaira-Spencer field by bosonization . This explains the choice of a spin structure. The construction of the measure in terms of fermionic correlators can be justified starting from the identification between the Kodaira-Spencer field and the collective field of the D-branes in the large $`N`$ limit . We will fully develop these ideas elsewhere.
It is also worth noting that the choice of a spin structure has a natural geometric interpretation in the topological closed string theory discussed in section five. Recall that the genus zero structure of the theory was shown to be encoded in an $`A_1`$ Hitchin integrable system $`\varpi :๐ฉ`$. The fiber $`๐ฉ_\beta `$ are torsors over the Prym variety $`\text{Prym}(\stackrel{~}{\mathrm{\Sigma }}_\beta /\mathrm{\Sigma })`$. The choice of a spin structure on $`\mathrm{\Sigma }`$ determines a section of the fibration $`\varpi :๐ฉ`$, hence also an isomorphism between each fiber $`๐ฉ_\beta `$ and the Prym. It would be very interesting to understand the connection between this section and Kodaira-Spencer theory on the open-closed side of the transition.
For computational purposes it is helpful to write the fermionic correlator (107) in terms of $`\vartheta `$-functions and prime forms. To keep the technical complications to a minimum, we will choose $`ฯต`$ so that the corresponding Dirac operator has no zero modes: $`h^0(\mathrm{\Sigma },ฯต)=h^1(\mathrm{\Sigma },ฯต)=0`$. In particular $`ฯต`$ has to be an even spin structure.
Let us briefly recall the construction of the prime form associated to the Riemann surface $`\mathrm{\Sigma }`$. We denote by $`\stackrel{~}{\mathrm{\Sigma }}`$ the universal cover of $`\mathrm{\Sigma }`$ and by $`\stackrel{~}{p},\stackrel{~}{q},\mathrm{}`$ points on the universal cover projecting to $`p,q,\mathrm{}`$ on $`\mathrm{\Sigma }`$. The prime form is a $`(\frac{1}{2},\frac{1}{2})`$ differential on $`\stackrel{~}{\mathrm{\Sigma }}\times \stackrel{~}{\mathrm{\Sigma }}`$ so that $`E(\stackrel{~}{p},\stackrel{~}{q})`$
$$E(\stackrel{~}{p},\stackrel{~}{q})=0p=q,$$
and the order of vanishing along $`p=q`$ is one. One can construct such a form by picking up a nonsingular odd theta characteristic $`\delta `$ with $`h^0(\mathrm{\Sigma },\delta )=1`$. Then we have
(108)
$$E(\stackrel{~}{p},\stackrel{~}{q})=\frac{\vartheta (\stackrel{~}{p}\stackrel{~}{q}+\delta )}{h(\stackrel{~}{p})h(\stackrel{~}{q})}$$
where $`\vartheta `$ is Riemannโs theta function and $`h`$ is the unique (up to multiplication by a nonzero constant) section of $`\delta `$. Note that the function $`E`$ does not depend on the choice of $`\delta `$ but it depends on the choice of a homology basis for $`\mathrm{\Sigma }`$ \[38, pg 17\]. Somewhat more invariantly we can view $`E`$ as a section in a line bundle on the surface $`\mathrm{\Sigma }\times \mathrm{\Sigma }`$. Let $`\theta \mathrm{Pic}^{g1}(\mathrm{\Sigma })`$ be the canonical theta line bundle on the degree $`(g1)`$ Jacobian. Let $`\mathrm{AJ}_\delta :\mathrm{\Sigma }\times \mathrm{\Sigma }\mathrm{Pic}^{g1}(\mathrm{\Sigma })`$ be the Abel-Jacobi map $`\mathrm{AJ}_\delta (p,q)=pq+\delta `$. Now the theta function $`\vartheta (\stackrel{~}{p}\stackrel{~}{q}+\delta )`$ descends to a section of the line bundle $`\mathrm{AJ}_\delta ^{}\theta `$ and so $`E`$ descends to a section (which will be denoted again by $`E`$) in the line bundle $`(\mathrm{AJ}_\delta ^{}\theta )p_1^{}\delta p_2^{}\delta `$. It is easy to check that this section is holomorphic. Furthermore a straightforward application of the see-saw principle shows that $`(\mathrm{AJ}_\delta ^{}\theta )p_1^{}\delta p_2^{}\delta =๐ช_{\mathrm{\Sigma }\times \mathrm{\Sigma }}(\mathrm{๐ฃ๐๐บ๐})`$ and so up to scale $`E`$ is the unique holomorphic section in the line bundle $`๐ช_{\mathrm{\Sigma }\times \mathrm{\Sigma }}(\mathrm{๐ฃ๐๐บ๐})`$ corresponding to the diagonal divisor $`\mathrm{๐ฃ๐๐บ๐}:=\{(x,y)\mathrm{\Sigma }\times \mathrm{\Sigma }|x=y\}`$. Since the self-intersection $`\mathrm{๐ฃ๐๐บ๐}^2=22g`$ of the divisor $`\mathrm{๐ฃ๐๐บ๐}`$ is negative, it follows that $`H^0(\mathrm{\Sigma }\times \mathrm{\Sigma },๐ช(\mathrm{๐ฃ๐๐บ๐}))`$ is one dimensional and so $`E`$ necessarily vanishes along the diagonal.
Now consider the following top degree meromorphic differential on $`\stackrel{~}{\mathrm{\Sigma }}^{2N}`$
(109)
$$\mathrm{\Omega }(\stackrel{~}{p}_a,\stackrel{~}{q}_a)=\frac{\vartheta (_{a=1}^N\stackrel{~}{p}_a_{a=1}^N\stackrel{~}{q}_a+ฯต)^2}{\vartheta (ฯต)^2}\frac{_{a,b=1}^{N}{}_{ab}{}^{}E(\stackrel{~}{p}_a,\stackrel{~}{p}_b)E(\stackrel{~}{q}_a,\stackrel{~}{q}_b)}{_{a,b=1}^NE(\stackrel{~}{p}_a,\stackrel{~}{q}_b)^2}.$$
Using the modular transformation properties of $`E(\stackrel{~}{p},\stackrel{~}{q})`$ and the $`\vartheta `$-function, one can check that $`\mathrm{\Omega }`$ descends to a differential (denoted by the same letter) on $`\mathrm{\Sigma }^{2N}`$. Again we can interpret $`\mathrm{\Omega }`$ as a section in a line bundle on $`\mathrm{\Sigma }^{2N}`$. If we write $`๐บ_ฯต:\mathrm{\Sigma }^{2N}\mathrm{Pic}^{g1}(\mathrm{\Sigma })`$ for the Abel-Jacobi map given by $`๐บ_ฯต((\{p_a\}_{a=1}^N,\{q_a\}_{a=1}^N)):=ฯต+_{a=1}^N(p_aq_a)`$, then $`\mathrm{\Omega }`$ is by definition a meromorphic section in the line bundle $`๐บ_ฯต^{}\theta ๐ช_{\mathrm{\Sigma }^{2N}}(2\mathrm{\Delta }_++2\mathrm{\Delta }_{}2\mathrm{\Delta })`$ with a double pole along the divisor $`\mathrm{\Delta }`$. Again a quick computation with the see-saw principle identifies $`๐บ_ฯต^{}\theta ๐ช_{\mathrm{\Sigma }^{2N}}(2\mathrm{\Delta }_++2\mathrm{\Delta }_{}2\mathrm{\Delta })`$ with the line bundle $`K_{\mathrm{\Sigma }^{2N}}`$. In other words, $`\mathrm{\Omega }`$ is a meromorphic $`2N`$ form on $`\mathrm{\Sigma }^{2N}`$ which has a double pole precisely along the โdiagonalโ divisor $`\mathrm{\Delta }\mathrm{\Sigma }^N\times \mathrm{\Sigma }^N`$. According to , the formula (109) represents the square of the fermionic correlator (107). In the following we will adopt (109) as the measure for the moduli space integral.
For a complete definition of the quantum theory, we have to specify an integration contour in $`\mathrm{\Sigma }^{2N}`$. A priori, there is no canonical choice of contour, and at the moment it is unclear what physical constraints should be imposed on such a contour. In all holomorphic Chern-Simons theories studied so far , the moduli space has a natural antiholomorphic involution $`\tau :`$ so that $`\tau ^{}\mathrm{\Omega }=\overline{\mathrm{\Omega }}`$. In these case the contour $`๐`$ is chosen to be the fixed point set of this involution, which is a special Lagrangian cycle calibrated by the measure $`\mathrm{\Omega }`$. In particular, the restriction of $`\mathrm{\Omega }`$ to $`๐`$ is a real differential form up to multiplication by a nonzero complex number.
This reality condition is important for a semiclassical analysis for the following simple reason. The semiclassical vacua are determined as the critical points of a complex function on $`\mathrm{\Gamma }`$ obtained by dividing $`\mathrm{\Omega }`$ by some reference classical measure $`\mathrm{\Omega }_0`$, which is real. Generically, a complex valued function on the cycle $`๐`$ does not have any critical points. Therefore, if we choose $`๐`$ to be an arbitrary cycle, we will not be able to develop a semiclassical expansion of the theory. While this is not a consistency requirement of a nonperturbative quantum field theory, we will adopt the reality condition as a selection criterion for $`๐`$. The issue of the dependence of the action on the choice of a contour has been raised in the physics literature before. In particular a reality condition appears in , and the subtleties of extending this condition to the context of holomorphic matrix models are analyzed in detail in .
In our case, for a generic $`\mathrm{\Sigma }`$, the moduli space does not admit antiholomorphic involutions, therefore there is no canonical choice for $`๐`$. Following the arguments of the previous paragraph, we will choose $`๐`$ to be any special Lagrangian cycle with respect to the measure $`\mathrm{\Omega }`$. If $`\mathrm{\Sigma }`$ admits an antiholomorphic involution $`\tau :\mathrm{\Sigma }\mathrm{\Sigma }`$, we can construct such a cycle as $`๐=\mathrm{\Gamma }^{2N}`$ where $`\mathrm{\Gamma }`$ is a component of the fixed point set of $`\tau `$ on $`\mathrm{\Sigma }`$. In the absence of a real structure, we will simply assume that we can find a closed contour $`\mathrm{\Gamma }`$ on $`\mathrm{\Sigma }`$ so that $`\mathrm{\Gamma }^{2N}`$ is special Lagrangian with respect to $`\mathrm{\Omega }`$.
### 6.4 Deformations and classical superpotential
So far we have formulated holomorphic Chern-Simons theory on $`\stackrel{~}{X}`$ in terms of a contour integral of a meromorphic top form on the classical moduli space $`_0`$. This is only a first step in our program since we are interested in $`๐`$-branes on a deformed threefold $`\stackrel{~}{X}_\alpha `$.
Recall that the threefold $`\stackrel{~}{X}_\alpha `$ constructed in section four has an affine bundle structure over the ruled surface $`S`$. If $`\alpha `$ is generic, that is
$$\text{div}(\alpha )=v_1+\mathrm{}+v_{2g2}$$
with $`v_1,\mathrm{},v_{2g2}`$ distinct points on $`\mathrm{\Sigma }`$, then only the $`2g2`$ fibers of the ruling $`q:S\mathrm{\Sigma }`$ sitting over the points $`\{v_i\}`$ will lift to holomorphic isolated $`(1,1)`$ curves $`C_1,\mathrm{},C_{2g2}`$ curves in $`\stackrel{~}{X}_\alpha `$.
The classical vacua of this theory consist of configurations of $`N_k^+`$, $`k=1,\mathrm{},r`$ branes wrapped on $`r`$ curves $`C_{s_k}`$, $`k=1,\mathrm{},r`$, and $`N_k^{}`$, $`k=r+1,\mathrm{},2g2`$ antibranes wrapped on the remaining curves $`C_{s_k}`$, $`k=r+1,\mathrm{},2g2`$ for some $`1r2g3`$. The multiplicities are subject to the constraint
(110)
$$\underset{k=1}{\overset{r}{}}N_k^+=\underset{k=r+1}{\overset{2g2}{}}N_k^{}=N.$$
In terms of the holomorphic gauge theory on $`\mathrm{\Sigma }`$, these D-brane configurations correspond to points of the form
(111)
$$D_+=\underset{k=0}{\overset{r}{}}N_k^+v_{s_k},D_{}=\underset{k=r+1}{\overset{2g2}{}}N_k^{}v_{s_k}$$
of the undeformed moduli space. Without loss of generality we can set $`s_k=k`$, $`k=1,\mathrm{}2g2`$ in the following.
Following the strategy of Dijkgraaf-Vafa transitions, holomorphic Chern-Simons theory on $`\stackrel{~}{X}_\alpha `$ can be defined as a superpotential deformation of the $`\alpha =0`$ theory. The superpotential in question should be a (possibly multivalued) holomorphic function on the finite cover $`_0^{}`$ of the moduli space whose critical points are in one-to-one correspondence with classical D-brane configurations.
As explained in section 2, such a function has a natural geometric origin. Indeed, one can construct a transverse holomorphic family $`๐`$ of two-cycles on $`\stackrel{~}{X}_\alpha `$ parameterized by $`\mathrm{\Sigma }`$ including the holomorphic curves $`C_1,\mathrm{},C_{2g2}`$ at the points $`v_1,\mathrm{},v_{2g2}`$. The generic fiber $`C_p`$ of this family over a point $`p\mathrm{\Sigma }`$ is a smooth non holomorphic two-cycle on the affine quadric $`(\stackrel{~}{X}_\alpha )_p`$, similarly to the local situation described in section 2.
This family determines a Donaldson-Thomas superpotential on the moduli space via the Abel-Jacobi map by analogy with the considerations of section 2. The only difference is that in the present case, the Donaldson-Thomas superpotential is multivalued since $`\mathrm{\Sigma }`$ contains nontrivial homology one-cycles. Therefore in order to obtain a single valued expression we have to work on the universal cover $`\stackrel{~}{\mathrm{\Sigma }}^{2N}`$ of the direct product $`\mathrm{\Sigma }^{2N}`$. The superpotential is then given by
(112)
$$W_\alpha (\stackrel{~}{p}_a,\stackrel{~}{q}_a)=\underset{a=1}{\overset{N}{}}_{\stackrel{~}{q}_a}^{\stackrel{~}{p}_a}\alpha .$$
It is a simple exercise to check that the critical points of $`W_\alpha `$ are classical vacua of the form (111).
Summarizing this discussion, the deformed holomorphic Chern-Simons action will be defined by a measure
(113)
$$\mathrm{\Omega }_\alpha (\stackrel{~}{p}_a,\stackrel{~}{q}_a)=\mathrm{\Omega }(\stackrel{~}{p}_a,\stackrel{~}{q}_a)e^{\frac{1}{g_s}W_\alpha (\stackrel{~}{p}_a,\stackrel{~}{q}_a)}$$
on $`\stackrel{~}{\mathrm{\Sigma }}^{2N}`$. To complete the construction, we have to specify a middle dimensional cycle $`๐`$ on $`\stackrel{~}{\mathrm{\Sigma }}^{2N}`$. According to the discussion of the previous subsection, $`๐`$ should be a special Lagrangian cycle with respect to the deformed measure (113). In the following we will assume that the cycle $`๐`$ can be chosen of the form $`๐=\mathrm{\Gamma }^{2N}`$ where $`\mathrm{\Gamma }`$ is a closed one-cycle on $`\mathrm{\Sigma }`$ passing through the zeroes of $`\alpha `$. This assumption is not unreasonable. For instance if $`\mathrm{\Sigma }`$ is a real curve equipped with an antiholomorphic involution $`\tau :\mathrm{\Sigma }\mathrm{\Sigma }`$, and $`\alpha `$ satisfies an appropriate reality condition, $`\mathrm{\Gamma }`$ can be chosen to be a component of the fixed point set of $`\tau `$.
### 6.5 Semiclassical vacua at large $`N`$ and Hitchin systems
In this subsection we determine the semiclassical vacua of the deformed holomorphic Chern-Simons theory in the large $`N`$ limit. The main result is that the semiclassical vacua are in one-to-one correspondence to $`A_1`$ Hitchin spectral covers of $`\mathrm{\Sigma }`$. This establishes a direct connection with the algebraic integrable systems found in the previous section.
In order to derive the semiclassical equations of motion we will rewrite the measure $`\mathrm{\Omega }_\alpha `$ in a more explicit form. Let $`U`$ denote the complement of the divisor $`h=0`$ in $`\stackrel{~}{\mathrm{\Sigma }}`$. We define a local coordinate function $`z:U`$ so that
$$dz(\stackrel{~}{p})=h(\stackrel{~}{p})^2$$
as in \[39, pg. 3.207-3.211\]. The restriction of $`\mathrm{\Omega }`$ to $`U^{2N}`$ can then be written as
(114)
$$\mathrm{\Omega }|_U=\frac{\vartheta (_{a=1}^N\stackrel{~}{p}_a_{a=1}^N\stackrel{~}{q}_a+ฯต)^2}{\vartheta (ฯต)^2}\frac{_{a,b=1}^{N}{}_{ab}{}^{}\vartheta (\stackrel{~}{p}_a\stackrel{~}{p}_b+\delta )\vartheta (\stackrel{~}{q}_a\stackrel{~}{q}_b+\delta )}{_{a,b=1}^N\vartheta (\stackrel{~}{p}_a\stackrel{~}{q}_b+\delta )^2}\underset{a=1}{\overset{N}{}}dz(\stackrel{~}{p}_a)dz(\stackrel{~}{q}_a)$$
The deformed measure $`\mathrm{\Omega }_\alpha `$ gives rise to an effective semiclassical superpotential
(115) $`{\displaystyle \frac{1}{g_s}}W_\alpha ^{\mathrm{eff}}(\stackrel{~}{p}_a,\stackrel{~}{q}_a)=`$ $`{\displaystyle \frac{1}{g_s}}W_\alpha (\stackrel{~}{p}_a,\stackrel{~}{q}_a){\displaystyle \underset{a,b=1}{\overset{N}{}}_{ab}}(\mathrm{log}\vartheta (\stackrel{~}{p}_a\stackrel{~}{p}_b+\delta )+\mathrm{log}\vartheta (\stackrel{~}{q}_a\stackrel{~}{q}_b+\delta ))`$
$`+{\displaystyle \underset{a,b=1}{\overset{N}{}}}\mathrm{log}\vartheta (\stackrel{~}{p}_a\stackrel{~}{q}_b+\delta )2\mathrm{log}\vartheta \left({\displaystyle \underset{a=1}{\overset{N}{}}}\stackrel{~}{p}_a{\displaystyle \underset{a=1}{\overset{N}{}}}\stackrel{~}{q}_a+ฯต\right).`$
Applying the variational principle to $`W_\alpha ^{\mathrm{eff}}(\stackrel{~}{p}_a,\stackrel{~}{q}_a)`$ we obtain the following semiclassical equations of motion
(116)
$$\begin{array}{cc}\hfill \frac{1}{g_s}\alpha (\stackrel{~}{p}_a)+2\underset{b=1}{\overset{N}{}}_{ba}d_{\stackrel{~}{p}_a}\mathrm{log}\vartheta (\stackrel{~}{p}_a\stackrel{~}{p}_b+\delta )& 2\underset{b=1}{\overset{N}{}}_{ba}d_{\stackrel{~}{p}_a}\mathrm{log}\vartheta (\stackrel{~}{p}_a\stackrel{~}{q}_b+\delta )\hfill \\ & +2d_{\stackrel{~}{p}_a}\mathrm{log}\vartheta \left(\underset{b=1}{\overset{N}{}}\stackrel{~}{p}_b\underset{b=1}{\overset{N}{}}\stackrel{~}{q}_b+ฯต\right)=0\hfill \\ \hfill \frac{1}{g_s}\alpha (\stackrel{~}{q}_a)+2\underset{b=1}{\overset{N}{}}_{ba}d_{\stackrel{~}{q}_a}\mathrm{log}\vartheta (\stackrel{~}{q}_a\stackrel{~}{q}_b+\delta )& 2\underset{b=1}{\overset{N}{}}_{ba}d_{\stackrel{~}{q}_a}\mathrm{log}\vartheta (\stackrel{~}{p}_b\stackrel{~}{q}_a+\delta )\hfill \\ & +2d_{\stackrel{~}{q}_a}\mathrm{log}\vartheta \left(\underset{b=1}{\overset{N}{}}\stackrel{~}{p}_b\underset{b=1}{\overset{N}{}}\stackrel{~}{q}_b+ฯต\right)=0.\hfill \end{array}$$
In order to study the large $`N`$ limit of these equations, let us introduce the meromorphic differential
(117)
$$\omega (\stackrel{~}{p})=\frac{1}{N}d_{\stackrel{~}{p}}\underset{a=1}{\overset{N}{}}\mathrm{log}\frac{E(\stackrel{~}{p},\stackrel{~}{p}_a)}{E(\stackrel{~}{p},\stackrel{~}{q}_a)}.$$
It is easy to check that $`\omega `$ is invariant under the action of the fundamental group, therefore it descends to a meromorphic one-form on $`\mathrm{\Sigma }`$. This is an Abelian differential of the third kind with simple poles with residue $`\frac{1}{N}`$ at $`p_a`$, $`a=1,\mathrm{},N`$ and simple poles with residue $`\frac{1}{N}`$ at $`q_a`$, $`a=1,\mathrm{},N`$. In the present context $`\omega `$ plays the same role as the resolvent in the large $`N`$ solution of matrix models. Over $`U`$ we can write
(118)
$$\omega (\stackrel{~}{p})=\frac{1}{N}\underset{a=1}{\overset{N}{}}d_{\stackrel{~}{p}}\mathrm{log}\frac{\vartheta (\stackrel{~}{p}\stackrel{~}{p}_a+\delta )}{\vartheta (\stackrel{~}{p}\stackrel{~}{q}_a+\delta )}.$$
Our goal is to show that in the large $`N`$ limit the equations of motion (116) give rise to an algebraic equation of the form
$$\omega ^2\frac{1}{\mu }\alpha \omega =\text{holomorphic deformation}$$
analogous to the loop equation of matrix models. Here $`\mu =Ng_s`$ is the โt Hooft coupling constant which is kept finite in the large $`N`$ limit. Note that the theta function $`\vartheta (\stackrel{~}{p}\stackrel{~}{p}_a+\delta )`$ has a first order zero at $`\stackrel{~}{p}=\stackrel{~}{p}_a`$ \[39, pg. 311\]. Therefore locally we can write
(119)
$$\vartheta (\stackrel{~}{p}\stackrel{~}{p}_a+\delta )=(z(\stackrel{~}{p})z(\stackrel{~}{p}_a))\stackrel{~}{\vartheta }(\stackrel{~}{p},\stackrel{~}{p}_a)$$
where $`\stackrel{~}{\vartheta }(\stackrel{~}{p},\stackrel{~}{p}_a)`$ is a holomorphic function over $`U`$ non vanishing at $`\stackrel{~}{p}=\stackrel{~}{p}_a`$, $`a=1,\mathrm{},N`$. Then $`\omega `$ can be further rewritten in the form
(120)
$$\omega (\stackrel{~}{p})=\frac{1}{N}\underset{a=1}{\overset{N}{}}\left(\frac{dz(\stackrel{~}{p})}{z(\stackrel{~}{p})z(\stackrel{~}{p}_a)}\frac{dz(\stackrel{~}{p})}{z(\stackrel{~}{p})z(\stackrel{~}{q}_a)}+d_{\stackrel{~}{p}}\mathrm{log}\stackrel{~}{\vartheta }(\stackrel{~}{p},\stackrel{~}{p}_a)d_{\stackrel{~}{p}}\mathrm{log}\stackrel{~}{\vartheta }(\stackrel{~}{p},\stackrel{~}{q}_a)\right).$$
The equations of motion (116) yield
(121) $`{\displaystyle \frac{1}{g_s}}{\displaystyle \underset{a=1}{\overset{N}{}}}{\displaystyle \frac{1}{z(\stackrel{~}{p})z(\stackrel{~}{p}_a)}}{\displaystyle \frac{\alpha (\stackrel{~}{p}_a)}{dz(\stackrel{~}{p}_a)}}+2{\displaystyle \underset{a,b=1}{\overset{N}{}}_{ba}}{\displaystyle \frac{1}{z(\stackrel{~}{p})z(\stackrel{~}{p}_a)}}{\displaystyle \frac{d_{\stackrel{~}{p}_a}\mathrm{log}\vartheta (\stackrel{~}{p}_a\stackrel{~}{p}_b+\delta )}{dz(\stackrel{~}{p}_a)}}`$
$`2{\displaystyle \underset{a,b=1}{\overset{N}{}}_{ba}}{\displaystyle \frac{1}{z(\stackrel{~}{p})z(\stackrel{~}{p}_a)}}{\displaystyle \frac{d_{\stackrel{~}{p}_a}\mathrm{log}\vartheta (\stackrel{~}{p}_a\stackrel{~}{q}_b+\delta )}{dz(\stackrel{~}{p}_a)}}`$
$`+2{\displaystyle \underset{a=1}{\overset{N}{}}}{\displaystyle \frac{1}{z(\stackrel{~}{p})z(\stackrel{~}{p}_a)}}{\displaystyle \frac{d_{\stackrel{~}{p}_a}\mathrm{log}\vartheta (\underset{b=1}{\overset{N}{}}\stackrel{~}{p}_b\underset{b=1}{\overset{N}{}}\stackrel{~}{q}_b+ฯต)}{dz(\stackrel{~}{p}_a)}}=0`$
$`{\displaystyle \frac{1}{g_s}}{\displaystyle \underset{a=1}{\overset{N}{}}}{\displaystyle \frac{1}{z(\stackrel{~}{p})z(\stackrel{~}{q}_a)}}{\displaystyle \frac{\alpha (\stackrel{~}{q}_a)}{dz(\stackrel{~}{q}_a)}}+2{\displaystyle \underset{a,b=1}{\overset{N}{}}_{ba}}{\displaystyle \frac{1}{z(\stackrel{~}{p})z(\stackrel{~}{q}_a)}}{\displaystyle \frac{d_{\stackrel{~}{q}_a}\mathrm{log}\vartheta (\stackrel{~}{q}_a\stackrel{~}{q}_b+\delta )}{dz(\stackrel{~}{q}_a)}}`$
$`2{\displaystyle \underset{a,b=1}{\overset{N}{}}_{ba}}{\displaystyle \frac{1}{z(\stackrel{~}{p})z(\stackrel{~}{q}_a)}}{\displaystyle \frac{d_{\stackrel{~}{q}_a}\mathrm{log}\vartheta (\stackrel{~}{p}_b\stackrel{~}{q}_a+\delta )}{dz(\stackrel{~}{q}_a)}}`$
$`+2{\displaystyle \underset{a=1}{\overset{N}{}}}{\displaystyle \frac{1}{z(\stackrel{~}{p})z(\stackrel{~}{q}_a)}}{\displaystyle \frac{d_{\stackrel{~}{q}_a}\mathrm{log}\vartheta (\underset{b=1}{\overset{N}{}}\stackrel{~}{p}_b\underset{b=1}{\overset{N}{}}\stackrel{~}{q}_b+ฯต)}{dz(\stackrel{~}{q}_a)}}=0.`$
Using (120), (121), a somewhat tedious computation yields
(122) $`\left({\displaystyle \frac{\omega (\stackrel{~}{p})}{dz(\stackrel{~}{p})}}\right)^2{\displaystyle \frac{1}{Ng_s}}{\displaystyle \frac{\omega (\stackrel{~}{p})}{dz(\stackrel{~}{p})}}{\displaystyle \frac{\alpha (\stackrel{~}{p})}{dz(\stackrel{~}{p})}}=`$
$`{\displaystyle \frac{1}{N^2g_s}}{\displaystyle \underset{a=1}{\overset{N}{}}}\left[{\displaystyle \frac{1}{z(\stackrel{~}{p})z(\stackrel{~}{p}_a)}}\left({\displaystyle \frac{\alpha (\stackrel{~}{p})}{dz(\stackrel{~}{p})}}{\displaystyle \frac{\alpha (\stackrel{~}{p}_a)}{dz(\stackrel{~}{p}_a)}}\right){\displaystyle \frac{1}{z(\stackrel{~}{p})z(\stackrel{~}{q}_a)}}\left({\displaystyle \frac{\alpha (\stackrel{~}{p})}{dz(\stackrel{~}{p})}}{\displaystyle \frac{\alpha (\stackrel{~}{q}_a)}{dz(\stackrel{~}{q}_a)}}\right)\right]`$
$`{\displaystyle \frac{1}{N^2g_s}}{\displaystyle \underset{a=1}{\overset{N}{}}}\left({\displaystyle \frac{d_{\stackrel{~}{p}}\mathrm{log}\stackrel{~}{\vartheta }(\stackrel{~}{p},\stackrel{~}{p}_a)d_{\stackrel{~}{p}}\mathrm{log}\stackrel{~}{\vartheta }(\stackrel{~}{p},\stackrel{~}{q}_a)}{dz(\stackrel{~}{p})}}{\displaystyle \frac{\alpha (\stackrel{~}{p})}{dz(\stackrel{~}{p})}}\right)`$
$`+{\displaystyle \frac{1}{N^2}}{\displaystyle \underset{a=1}{\overset{N}{}}}\left[\left({\displaystyle \frac{d_{\stackrel{~}{p}}\mathrm{log}\vartheta (\stackrel{~}{p}\stackrel{~}{p}_a+\delta )}{dz(\stackrel{~}{p})}}\right)^2+\left({\displaystyle \frac{d_{\stackrel{~}{p}}\mathrm{log}\vartheta (\stackrel{~}{p}\stackrel{~}{q}_a+\delta )}{dz(\stackrel{~}{p})}}\right)^2\right]`$
$`+{\displaystyle \frac{2}{N^2}}{\displaystyle \underset{a,b=1}{\overset{N}{}}_{ab}}{\displaystyle \frac{1}{z(\stackrel{~}{p})z(\stackrel{~}{p}_a)}}\left({\displaystyle \frac{d_{\stackrel{~}{p}}\mathrm{log}\stackrel{~}{\vartheta }(\stackrel{~}{p},\stackrel{~}{p}_b)}{dz(\stackrel{~}{p})}}{\displaystyle \frac{d_{\stackrel{~}{p}_a}\mathrm{log}\stackrel{~}{\vartheta }(\stackrel{~}{p}_a,\stackrel{~}{p}_b)}{dz(\stackrel{~}{p}_a)}}\right)`$
$`+{\displaystyle \frac{2}{N^2}}{\displaystyle \underset{a,b=1}{\overset{N}{}}_{ab}}{\displaystyle \frac{1}{z(\stackrel{~}{p})z(\stackrel{~}{q}_a)}}\left({\displaystyle \frac{d_{\stackrel{~}{p}}\mathrm{log}\stackrel{~}{\vartheta }(\stackrel{~}{p},\stackrel{~}{q}_b)}{dz(\stackrel{~}{p})}}{\displaystyle \frac{d_{\stackrel{~}{q}_a}\mathrm{log}\stackrel{~}{\vartheta }(\stackrel{~}{q}_a,\stackrel{~}{q}_b)}{dz(\stackrel{~}{q}_a)}}\right)`$
$`{\displaystyle \frac{2}{N^2}}{\displaystyle \underset{a,b=1}{\overset{N}{}}}{\displaystyle \frac{1}{z(\stackrel{~}{p})z(\stackrel{~}{p}_a)}}\left({\displaystyle \frac{d_{\stackrel{~}{p}}\mathrm{log}\stackrel{~}{\vartheta }(\stackrel{~}{p},\stackrel{~}{q}_b)}{dz(\stackrel{~}{p})}}{\displaystyle \frac{d_{\stackrel{~}{p}_a}\mathrm{log}\stackrel{~}{\vartheta }(\stackrel{~}{p}_a,\stackrel{~}{q}_b)}{dz(\stackrel{~}{q}_a)}}\right)`$
$`{\displaystyle \frac{2}{N^2}}{\displaystyle \underset{a,b=1}{\overset{N}{}}}{\displaystyle \frac{1}{z(\stackrel{~}{p})z(\stackrel{~}{q}_a)}}\left({\displaystyle \frac{d_{\stackrel{~}{p}}\mathrm{log}\stackrel{~}{\vartheta }(\stackrel{~}{p},\stackrel{~}{p}_b)}{dz(\stackrel{~}{p})}}{\displaystyle \frac{d_{\stackrel{~}{q}_a}\mathrm{log}\stackrel{~}{\vartheta }(\stackrel{~}{q}_a,\stackrel{~}{p}_b)}{dz(\stackrel{~}{q}_a)}}\right)`$
$`+{\displaystyle \frac{1}{N^2}}{\displaystyle \underset{a,b=1}{\overset{N}{}}_{ab}}\left({\displaystyle \frac{d_{\stackrel{~}{p}}\mathrm{log}\stackrel{~}{\vartheta }(\stackrel{~}{p},\stackrel{~}{p}_a)}{dz(\stackrel{~}{p})}}{\displaystyle \frac{d_{\stackrel{~}{p}}\mathrm{log}\stackrel{~}{\vartheta }(\stackrel{~}{p},\stackrel{~}{p}_b)}{dz(\stackrel{~}{p})}}+{\displaystyle \frac{d_{\stackrel{~}{p}}\mathrm{log}\stackrel{~}{\vartheta }(\stackrel{~}{p},\stackrel{~}{q}_a)}{dz(\stackrel{~}{p})}}{\displaystyle \frac{d_{\stackrel{~}{p}}\mathrm{log}\stackrel{~}{\vartheta }(\stackrel{~}{p},\stackrel{~}{q}_b)}{dz(\stackrel{~}{p})}}\right)`$
$`{\displaystyle \frac{2}{N^2}}{\displaystyle \underset{a,b=1}{\overset{N}{}}}{\displaystyle \frac{d_{\stackrel{~}{p}}\mathrm{log}\stackrel{~}{\vartheta }(\stackrel{~}{p},\stackrel{~}{p}_a)}{dz(\stackrel{~}{p})}}{\displaystyle \frac{d_{\stackrel{~}{p}}\mathrm{log}\stackrel{~}{\vartheta }(\stackrel{~}{p},\stackrel{~}{q}_b)}{dz(\stackrel{~}{p})}}`$
$`{\displaystyle \frac{2}{N^2}}{\displaystyle \underset{a=1}{\overset{N}{}}}{\displaystyle \frac{1}{z(\stackrel{~}{p})z(\stackrel{~}{p}_a)}}{\displaystyle \frac{d_{\stackrel{~}{p}_a}\mathrm{log}\vartheta (\underset{b=1}{\overset{N}{}}\stackrel{~}{p}_b\underset{b=1}{\overset{N}{}}\stackrel{~}{q}_b+ฯต)}{dz(\stackrel{~}{p}_a)}}`$
$`{\displaystyle \frac{2}{N^2}}{\displaystyle \underset{a=1}{\overset{N}{}}}{\displaystyle \frac{1}{z(\stackrel{~}{p})z(\stackrel{~}{q}_a)}}{\displaystyle \frac{d_{\stackrel{~}{q}_a}\mathrm{log}\vartheta (\underset{b=1}{\overset{N}{}}\stackrel{~}{p}_b\underset{b=1}{\overset{N}{}}\stackrel{~}{q}_b+ฯต)}{dz(\stackrel{~}{q}_a)}}.`$
Now we take the limit $`N\mathrm{}`$, $`g_s0`$ keeping the โt Hooft coupling $`\mu =Ng_s`$ fixed. The behavior of the terms in the equation (122) is determined by their scaling with $`N`$. Terms of the form
$$\frac{1}{N}\underset{a=1}{\overset{N}{}}\mathrm{}\text{and}\frac{1}{N^2}\underset{a,b=1}{\overset{N}{}}\mathrm{}$$
are expected to have finite limit when $`N\mathrm{}`$, whereas terms of the form
$$\frac{1}{N^2}\underset{a=1}{\overset{N}{}}\mathrm{}$$
tend to zero because they are suppressed by an extra power of $`N`$. Using these rules, we can check that the right hand side of equation (122) is a holomorphic function on $`U`$ in the large $`N`$ limit. The polar part of that expression โ that is the terms in the second and last two lines โ scales as $`1/N`$, therefore it vanishes in the large $`N`$ limit. The remaining terms are nonzero, but the poles cancel even at finite $`N`$.
The limit of the left hand side of equation (122) is a quadratic expression of the form
(123)
$$\omega _{\mathrm{}}^2\frac{1}{\mu }\omega _{\mathrm{}}\alpha $$
where $`\omega _{\mathrm{}}`$ is the large $`N`$ limit of the meromorphic form $`\omega `$ which characterizes the distribution of branes on $`\mathrm{\Sigma }`$. We will assume that the limit $`\omega _{\mathrm{}}`$ exists and it is a well defined mathematical object on $`\mathrm{\Sigma }`$. By construction, the expression (123) can only have poles at the locations of the branes on $`\mathrm{\Sigma }`$. However, we have shown in the previous paragraph that these poles are absent from the left hand side of equation (122). Therefore we can conclude that the large $`N`$ limit of equation (122) is of the form
(124)
$$\omega _{\mathrm{}}\frac{1}{\mu }\alpha \omega _{\mathrm{}}=\beta $$
where $`\beta `$ is a global holomorphic quadratic differential on $`\mathrm{\Sigma }`$.
Then $`\omega _{\mathrm{}}`$ must be a multivalued holomorphic differential on $`\mathrm{\Sigma }`$ with branch points located at the zeroes of $`\beta `$. Note that by construction $`\omega _{\mathrm{}}`$ characterizes the distribution of branes on $`\mathrm{\Sigma }`$ in the large $`N`$ limit. For finite $`N`$, $`\omega `$ has poles at the locations of the branes, therefore we would naively expect $`\omega _{\mathrm{}}`$ to have infinitely many poles on $`\mathrm{\Sigma }`$. The only way such a mathematical object can be well defined is if the collection of poles of $`\omega `$ becomes a collection of branch cuts $`\mathrm{\Gamma }_1,\mathrm{},\mathrm{\Gamma }_{2g2}`$ in the large $`N`$ limit, and $`\omega _{\mathrm{}}`$ is a multivalued differential. This is consistent with the large $`N`$ limit (124) of the equation of motion and it also shows that the branch points must be located on the contour $`\mathrm{\Gamma }`$ and the branch cuts must be some line segments contained in $`\mathrm{\Gamma }`$. The filling fraction associated to each branch cut is determined by the period
(125)
$$_{\gamma _s}\mathrm{\Omega }_{\mathrm{}}$$
where $`\gamma _s`$ is a contour on $`\mathrm{\Sigma }`$ surrounding the branch cut $`\mathrm{\Gamma }_s`$, $`s=1,\mathrm{}2g2`$.
To summarize: we have found that the large $`N`$ semiclassical vacuum configurations are in one-to-one correspondence with quadratic holomorphic differentials $`\beta `$. The distribution of branes in a semiclassical vacuum determined by $`\beta `$ is encoded in the multivalued differential $`\omega _{\mathrm{}}`$ which solves (124).
Now the connection with Hitchin spectral covers becomes manifest. Equation (124) is the defining equation of a spectral cover $`\stackrel{~}{\mathrm{\Sigma }}_\beta `$ and $`\omega _{\mathrm{}}`$ is the canonical holomorphic differential on $`\stackrel{~}{\mathrm{\Sigma }}_\beta `$. Each contour $`\gamma _s`$ on $`\mathrm{\Sigma }`$ lifts to an anti-invariant closed one-cycle $`\stackrel{~}{\gamma }_s`$ on $`\mathrm{\Sigma }_\beta `$, and the filling fractions (125) are given by the periods of $`\omega _{\mathrm{}}`$ on the cycles $`\stackrel{~}{\gamma }_s`$.
Taking into account the relation between Hitchin Pryms and homology intermediate Jacobians proved in section 5.2, the filling fractions (125) can be related to periods of the holomorphic three-form $`\mathrm{\Omega }_{X_\beta }`$ on the threefold $`X_\beta `$ constructed in section 4.
In conclusion, we have shown that the large $`N`$ limit of the holomorphic Chern-Simons theory is governed by a Hitchin integrable system, which is in turn isomorphic to the Calabi-Yau integrable system for the universal family of Calabi-Yau threefolds over the of the moduli space $`๐ณ`$. This is a physical proof of large $`N`$ duality at genus zero.
Appendix A Morphisms of Twisted Complexes
In this appendix we prove equation (103) in section 6.2. For convenience, recall that we are given two sets of algebraic data $`(_i,\mathrm{\Psi }_{ji})`$, $`(_i^{},\mathrm{\Psi }_{ji}^{})`$, $`i,j=0,1,2`$ satisfying conditions $`(i)`$-$`(iii)`$ below equation (97). In particular we have two three term complexes $`(,^{})`$ and we construct the extensions
(A.126) $`0๐ช(K_\mathrm{\Sigma })[2]๐ฆ0`$
$`0^{}๐ช(K_\mathrm{\Sigma })[2]๐ฆ^{}^{}0`$
which yield the short exact sequences of complexes
(A.127) $`0Hom(,^{})๐ช(K_\mathrm{\Sigma })[2]Hom(๐ฆ๐ช(K_\mathrm{\Sigma })[2],^{})Hom(,^{})0.`$
$`0Hom(,^{})๐ช(K_\mathrm{\Sigma })[2]Hom(,๐ฆ^{})Hom(,^{})0.`$
More explicitly, using the notation of section 5.2, we have
(A.128) $`_0\stackrel{\mathrm{\Psi }_{10}}{}_1\stackrel{\stackrel{~}{\mathrm{\Psi }}_{21}}{}๐ฎ\stackrel{\stackrel{~}{\mathrm{\Psi }}_{10}}{}_1๐ช(K_\mathrm{\Sigma })\stackrel{\mathrm{\Psi }_{21}๐_{๐ช\left(K_\mathrm{\Sigma }\right)}}{}_2๐ช(K_\mathrm{\Sigma }).`$
$`_0^{}\stackrel{\mathrm{\Psi }_{10}^{}}{}_1^{}\stackrel{\stackrel{~}{\mathrm{\Psi }}_{21}^{}}{}๐ฎ^{}\stackrel{\stackrel{~}{\mathrm{\Psi }}_{10}^{}}{}_1^{}๐ช(K_\mathrm{\Sigma })\stackrel{\mathrm{\Psi }_{21}^{}๐_{๐ช\left(K_\mathrm{\Sigma }\right)}}{}_2^{}๐ช(K_\mathrm{\Sigma }).`$
where the sheaves $`๐ฎ,๐ฎ^{}`$ are given by extensions
(A.129) $`0_0(K_\mathrm{\Sigma })\stackrel{i}{}๐ฎ\stackrel{\rho }{}_20`$
$`0_0^{}(K_\mathrm{\Sigma })\stackrel{i^{}}{}๐ฎ^{}\stackrel{\rho ^{}}{}_2^{}0.`$
Our problem is to construct the difference $`C(๐ฆ,๐ฆ^{})`$ of the two extensions (A.127) and compute its hypercohomology.
Suppose we have two exact sequences of complexes of coherent sheaves on a smooth projective variety
(A.130) $`0ACB0`$
$`0AC^{}B0.`$
The difference $`C^{}C`$ can be constructed in two steps. First take the pull-back extension
(A.131)
where $`\iota ^{}:AAA`$ is the anti diagonal embedding. Then take a second pullback
(A.132)
where $`\iota ^+:BBB`$ is the diagonal embedding.
We have to carry out this construction for the extensions (A.127). In order to keep the formulas short, we will use the notation $`H_{nm}Hom(_m,_n^{})`$. We will first write down explicitly the complexes $`Hom(,^{}),Hom(,^{})๐ช(K_\mathrm{\Sigma })[2]`$ and then write down the complex $`C(๐ฆ,๐ฆ^{})`$ as an extension of complexes, obtaining the following diagrams.
(A.133) $`Hom(,^{}):`$ $`Hom(,^{})๐ช(K_\mathrm{\Sigma })[2]:`$
(A.134)
This determines the terms of degrees $`2,1,3,4`$. The remaining terms can be determined following the steps (A.131)-(A.132) described above. In degree $`0`$, the first step yields
(A.135)
$$\frac{CC^{}}{\iota ^{}(A)}=\frac{Hom(_2,๐ฎ^{})Hom(S,_0^{})(K_\mathrm{\Sigma })}{\iota ^{}(H_{02}(K_\mathrm{\Sigma }))}H_{22}H_{11}^2H_{00}$$
and the final result is
(A.136)
$$C(๐ฆ^{},๐ฆ)^0=\frac{Hom(_2,๐ฎ^{})Hom(S,_0^{})(K_\mathrm{\Sigma })}{\iota ^{}(H_{02}(K_\mathrm{\Sigma }))}H_{11}.$$
In degree one, we obtain at the first step
(A.137) $`{\displaystyle \frac{CC^{}}{\iota ^{}(A)}}`$ $`={\displaystyle \frac{H_{12}(K_\mathrm{\Sigma })Hom(_1,๐ฎ^{})H_{10}H_{21}Hom(๐ฎ,_1^{})(K_\mathrm{\Sigma })H_{01}(K_\mathrm{\Sigma })}{\iota ^{}(H_{12}(K_\mathrm{\Sigma })H_{01}(K_\mathrm{\Sigma }))}}`$
$`={\displaystyle \frac{H_{12}(K_\mathrm{\Sigma })Hom(_1,๐ฎ^{})Hom(๐ฎ,_1^{})(K_\mathrm{\Sigma })H_{01}(K_\mathrm{\Sigma })}{\iota ^{}(H_{12}(K_\mathrm{\Sigma })H_{01}(K_\mathrm{\Sigma }))}}H_{10}H_{21}`$
$`Hom(_1,๐ฎ^{})Hom(๐ฎ,_1^{})(K_\mathrm{\Sigma })H_{10}H_{21}.`$
After the second step we obtain
(A.138)
$$C(๐ฆ,๐ฆ^{})^1Hom(_1,๐ฎ^{})Hom(๐ฎ,_1^{})(K_\mathrm{\Sigma }).$$
The term of degree two can be similarly determined to be
(A.139)
$$C(๐ฆ,๐ฆ^{})^2Hom(_0,๐ฎ^{})\times _{H_{20}}Hom(๐ฎ,_2^{})(K_\mathrm{\Sigma })H_{11}(K_\mathrm{\Sigma }).$$
Now let us determine the differentials. The first and the last differentials are standard, hence we will focus on the remaining ones. We have
(A.140) $`c_{0,1}:H_{12}H_{01}`$ $`{\displaystyle \frac{Hom(_2,๐ฎ^{})Hom(๐ฎ,_0^{})(K_\mathrm{\Sigma })}{\iota ^{}(H_{02}(K_\mathrm{\Sigma }))}}H_{11}`$
$`(s_{12},s_{01})`$ $`([\stackrel{~}{\mathrm{\Psi }}_{21}^{}s_{12},+(s_{01}๐_{K_\mathrm{\Sigma }})\stackrel{~}{\mathrm{\Psi }}_{10}],s_{12}\mathrm{\Psi }_{21}+\mathrm{\Psi }_{10}^{}s_{01})`$
where we use the notation $`[,]`$ for equivalence classes in the quotient $`(Hom(_2,๐ฎ^{})Hom(๐ฎ,_0^{})(K_\mathrm{\Sigma }))/\iota ^{}(H_{02}(K_\mathrm{\Sigma }))`$. Next,
(A.141) $`c_{10}:`$ $`{\displaystyle \frac{Hom(_2,๐ฎ^{})Hom(๐ฎ,_0^{})(K_\mathrm{\Sigma })}{\iota ^{}(H_{02}(K_\mathrm{\Sigma }))}}H_{11}Hom(_1,๐ฎ^{})Hom(๐ฎ,_1^{})(K_\mathrm{\Sigma })`$
$`([u,v],w)(u\mathrm{\Psi }_{21}i^{}v\stackrel{~}{\mathrm{\Psi }}_{21}+\stackrel{~}{\mathrm{\Psi }}_{21}^{}w,\stackrel{~}{\mathrm{\Psi }}_{10}^{}u\rho +(\mathrm{\Psi }_{10}^{}๐_{K_\mathrm{\Sigma }})vw\stackrel{~}{\mathrm{\Psi }}_{10})`$
using the notation of equation (A.129). Let us check that the map $`c_{10}`$ is well defined on equivalence classes $`[u,v]`$. It suffices to show that $`c_{10}([i^{}s,s\rho ],0)=0`$ for any element $`sH_{02}(K_\mathrm{\Sigma })`$. We have
$`c_{10}([i^{}s,s\rho ],0)`$ $`=(i^{}s\mathrm{\Psi }_{21}i^{}s\rho \stackrel{~}{\mathrm{\Psi }}_{21},\stackrel{~}{\mathrm{\Psi }}_{10}^{}i^{}s\rho +(\mathrm{\Psi }_{10}^{}๐_{K_\mathrm{\Sigma }})s\rho )`$
$`=(i^{}s\mathrm{\Psi }_{21}i^{}s\mathrm{\Psi }_{21},(\mathrm{\Psi }_{10}^{}๐_{K_\mathrm{\Sigma }})s\rho +(\mathrm{\Psi }_{10}^{}๐_{K_\mathrm{\Sigma }})s\rho )`$
$`=0`$
where we have used the relations
$$\rho \stackrel{~}{\mathrm{\Psi }}_{21}=\mathrm{\Psi }_{21},\stackrel{~}{\mathrm{\Psi }}_{10}^{}i^{}=\mathrm{\Psi }_{10}^{}๐_{K_\mathrm{\Sigma }}$$
following from diagrams (92),(93). One can similarly check that $`c_{10}c_{0,1}=0`$ using the relations
$$\stackrel{~}{\mathrm{\Psi }}_{10}\mathrm{\Psi }_{21}=(\mathrm{\Psi }_{21}๐_{K_\mathrm{\Sigma }})\stackrel{~}{\mathrm{\Psi }}_{10}=0,\stackrel{~}{\mathrm{\Psi }}_{10}^{}\stackrel{~}{\mathrm{\Psi }}_{21}^{}=\stackrel{~}{\mathrm{\Psi }}_{21}^{}\mathrm{\Psi }_{10}^{}=0$$
following from (A.128). Proceeding in a similar manner, we find
(A.142) $`c_{21}:`$ $`Hom(_1,๐ฎ^{})Hom(๐ฎ,_1^{})(K_\mathrm{\Sigma })Hom(_0,๐ฎ^{})\times _{H_{20}}Hom(๐ฎ,_2^{})(K_\mathrm{\Sigma })H_{11}(K_\mathrm{\Sigma })`$
$`(x,y)(x\mathrm{\Psi }_{10}+(\stackrel{~}{\mathrm{\Psi }}_{21}^{}๐_{K_\mathrm{\Sigma }})yi,(\rho ^{}๐_{K_\mathrm{\Sigma }})x\stackrel{~}{\mathrm{\Psi }}_{10}+(\mathrm{\Psi }_{21}^{}๐_{K_\mathrm{\Sigma }})y,\stackrel{~}{\mathrm{\Psi }}_{10}^{}x+y\stackrel{~}{\mathrm{\Psi }}_{21})`$
and
(A.143) $`c_{32}:Hom(_0,๐ฎ^{})`$ $`\times _{H_{20}}Hom(๐ฎ,_2^{})(K_\mathrm{\Sigma })H_{11}(K_\mathrm{\Sigma })H_{21}(K_\mathrm{\Sigma })H_{10}(K_\mathrm{\Sigma })`$
$`(r,t,z)(t\stackrel{~}{\mathrm{\Psi }}_{21}\mathrm{\Psi }_{21}^{}z,\stackrel{~}{\mathrm{\Psi }}_{10}^{}rz\mathrm{\Psi }_{10})`$
It is a straightforward exercise to check that $`c_{21}`$ is well defined and $`c_{21}c_{10}=c_{32}c_{21}=0`$.
In the remaining part of this section, we will write down the hypercohomology double complex of the sheaf complex $`C(๐ฆ,๐ฆ^{})`$ and prove formula (103). We will regard all locally free sheaves $`_i,_i^{}`$, $`i=0,1,2`$ as $`C^{\mathrm{}}`$ vector bundles $`F_i,F_i^{}`$, $`i=0,1,2`$ equipped with Dolbeault operators and use Dolbeault resolutions. The hypercohomology double complex is if the form
$$^{p,q}=\mathrm{\Omega }^{0,p}(C(๐ฆ,๐ฆ^{})^q),D=\overline{}+(1)^pc$$
where $`c`$ is the differential of $`C(๐ฆ,๐ฆ^{})`$ and it gives rise to a single complex
(A.144)
$$^n=_{p+q=n}\mathrm{\Omega }^{0,p}(C(๐ฆ,๐ฆ^{})^q),D=\overline{}+(1)^pc.$$
For concreteness, we will write down explicit formulas for the subcomplex
$$^1\stackrel{D_{0,1}}{}^0\stackrel{D_{10}}{}^1$$
and show that $`H^0(,D)`$ is isomorphic to the space of degree zero morphisms between three term twisted complexes as defined in section 5.1. One can prove the same result for degree one morphisms by a very similar computation.
Let us write down the terms of the complex (A.144). In degree $`1`$ we have
(A.145) $`\mathrm{\Omega }^{0,0}(C(๐ฆ,๐ฆ^{})^1)\mathrm{\Omega }^{0,1}(C(๐ฆ,๐ฆ^{})^2)\mathrm{\Omega }^{0,0}(F_2,F_1^{})\mathrm{\Omega }^{0,0}(F_1,F_0^{})\mathrm{\Omega }^{0,1}(F_2,F_0^{})`$
where we have used the shorthand notation $`\mathrm{\Omega }^{0,p}(Hom(F_m,F_n^{}))\mathrm{\Omega }^{0,p}(F_m,F_n^{})`$. The Dolbeault operator is the direct sum of the Dolbeault operators for the individual terms in (A.145). In order to write down the degree zero term, note that the extensions (A.129) are split as exact sequences of $`C^{\mathrm{}}`$ bundles. Therefore we have
(A.146) $`\mathrm{\Omega }^{0,0}(C(๐ฆ,๐ฆ^{})^0)\mathrm{\Omega }^{0,1}(C(๐ฆ,๐ฆ^{})^1)`$ $`{\displaystyle \frac{\mathrm{\Omega }^{0,0}(F_2,F_0^{}(K_\mathrm{\Sigma }))\mathrm{\Omega }^{0,0}(F_2,F_0^{}(K_\mathrm{\Sigma }))}{\iota ^{}\mathrm{\Omega }^{0,0}(F_2,F_0^{}(K_\mathrm{\Sigma }))}}`$
$`\mathrm{\Omega }^{0,0}(F_2,F_2^{})\mathrm{\Omega }^{0,0}(F_0,F_0^{})\mathrm{\Omega }^{0,0}(F_1,F_1^{})`$
$`\mathrm{\Omega }^{0,1}(F_2,F_1^{})\mathrm{\Omega }^{0,1}(F_1,F_0^{})`$
$`\mathrm{\Omega }^{0,0}(F_2,F_0^{}(K_\mathrm{\Sigma }))\mathrm{\Omega }^{0,0}(F_2,F_2^{})\mathrm{\Omega }^{0,0}(F_0,F_0^{})`$
$`\mathrm{\Omega }^{0,0}(F_1,F_1^{})\mathrm{\Omega }^{0,1}(F_2,F_1^{})\mathrm{\Omega }^{0,1}(F_1,F_0^{}).`$
The Dolbeault operator
$$\overline{}:\mathrm{\Omega }^{0,0}(C(๐ฆ,๐ฆ^{})^0)\mathrm{\Omega }^{0,1}(C(๐ฆ,๐ฆ^{})^0)$$
is given by
(A.147)
$$\overline{}\left[\begin{array}{c}\lambda _{02}^{1,0}\\ \lambda _{22}^{0,0}\\ \lambda _{00}^{0,0}\\ \lambda _{11}^{0,0}\end{array}\right]=\left[\begin{array}{c}\overline{}\lambda _{02}^{1,0}+\mathrm{\Psi }_{02}^{1,1}\lambda _{22}^{0,0}\lambda _{00}^{0,0}\mathrm{\Psi }_{02}^{1,1}\\ \overline{}\lambda _{22}^{0,0}\\ \overline{}\lambda _{00}^{0,0}\\ \overline{}\lambda _{11}^{0,0}\end{array}\right]$$
where $`\lambda _{n,m}^{p,q}`$ denotes an arbitrary element of $`\mathrm{\Omega }_{n,m}^{p,q}`$. The degree one term is
(A.148) $`\mathrm{\Omega }^{0,0}`$ $`(C(๐ฆ,๐ฆ^{})^1)\mathrm{\Omega }^{0,1}(C(๐ฆ,๐ฆ^{})^0)`$
$`\mathrm{\Omega }^{0,0}(F_1,F_0^{}(K_\mathrm{\Sigma }))\mathrm{\Omega }^{0,0}(F_1,F_2^{})\mathrm{\Omega }^{0,0}(F_0,F_1^{})\mathrm{\Omega }^{0,0}(F_2,F_1^{}(K_\mathrm{\Sigma }))`$
$`\mathrm{\Omega }^{0,1}(F_2,F_0^{}(K_\mathrm{\Sigma }))\mathrm{\Omega }^{0,1}(F_2,F_2^{})\mathrm{\Omega }^{0,1}(F_0,F_0^{})\mathrm{\Omega }^{0,1}(F_1,F_1^{})`$
Now let us compute the differentials. We have
(A.149) $`D_{0,1}:\mathrm{\Omega }^{0,0}(C(๐ฆ,๐ฆ^{})^1)\mathrm{\Omega }^{0,1}(C(๐ฆ,๐ฆ^{})^2)`$ $`\mathrm{\Omega }^{0,0}(C(๐ฆ,๐ฆ^{})^0)\mathrm{\Omega }^{0,1}(C(๐ฆ,๐ฆ^{})^1)`$
$`D_{0,1}\left[\begin{array}{c}\lambda _{12}^{0,0}\\ \lambda _{01}^{0,0}\\ \lambda _{02}^{0,1}\end{array}\right]`$ $`=\left[\begin{array}{c}\lambda _{01}^{0,0}\mathrm{\Psi }_{12}^{1,0}+\mathrm{\Psi }_{}^{}{}_{01}{}^{1,0}\lambda _{12}^{0.0}\\ \mathrm{\Psi }_{}^{}{}_{21}{}^{0,0}\lambda _{12}^{0,0}\\ \lambda _{01}^{0,0}\mathrm{\Psi }_{10}^{0,0}\\ \mathrm{\Psi }_{}^{}{}_{10}{}^{0,0}\lambda _{01}^{0,0}+\lambda _{12}^{0,0}\mathrm{\Psi }_{21}^{0,0}\\ \overline{}\lambda _{12}^{0,0}\mathrm{\Psi }_{10}^{0,0}\lambda _{02}^{0,1}\\ \overline{}\lambda _{01}^{0,0}+\lambda _{02}^{0,1}\mathrm{\Psi }_{21}^{0,0}\end{array}\right]`$
In order to evaluate
$$D_{10}:\mathrm{\Omega }^{0,0}(C(๐ฆ,๐ฆ^{})^0)\mathrm{\Omega }^{0,1}(C(๐ฆ,๐ฆ^{})^1)\mathrm{\Omega }^{0,0}(C(๐ฆ,๐ฆ^{})^1)\mathrm{\Omega }^{0,1}(C(๐ฆ,๐ฆ^{})^0),$$
we need some auxiliary results. The relevant Dolbeault operator has been written down in (A.147). We also have to determine the action of differentials $`c_{0,1}`$, $`c_{10}`$ on Dolbeault elements $`\lambda _{nm}^{p,q}`$. Given the isomorphism (A.146), the differential
$$c_{10}:\mathrm{\Omega }^{0,0}(C(๐ฆ,๐ฆ^{})^0)\mathrm{\Omega }^{0,0}(C(๐ฆ,๐ฆ^{})^1)$$
can be determined by making the substitutions
$$u=\frac{1}{2}\lambda _{02}^{1,0}+\lambda _{22}^{0,0},v=\frac{1}{2}\lambda _{02}^{1,0}+\lambda _{00}^{0,0},w=\lambda _{11}^{0,0}$$
in (A.141). Using the fact that the extensions (A.129) are split, we have
(A.150) $`u\mathrm{\Psi }_{21}i^{}v\stackrel{~}{\mathrm{\Psi }}_{21}+\stackrel{~}{\mathrm{\Psi }}_{21}^{}w`$ $`=\left({\displaystyle \frac{1}{2}}\lambda _{02}^{1,0}+\lambda _{22}^{0,0}\right)\mathrm{\Psi }_{21}^{0,0}{\displaystyle \frac{1}{2}}\lambda _{02}^{1,0}\mathrm{\Psi }_{21}^{0,0}\lambda _{00}^{0,0}\mathrm{\Psi }_{01}^{1,0}+\mathrm{\Psi }_{21}^{0,0}\lambda _{11}^{0,0}`$
$`+\mathrm{\Psi }_{01}^{1,0}\lambda _{11}^{0,0}`$
$`=\lambda _{02}^{1,0}\mathrm{\Psi }_{21}^{0,0}\lambda _{00}^{0,0}\mathrm{\Psi }_{01}^{1,0}+\mathrm{\Psi }_{01}^{1,0}\lambda _{11}^{0,0}\lambda _{22}^{0,0}\mathrm{\Psi }_{21}^{0,0}+\mathrm{\Psi }_{21}^{0,0}\lambda _{11}^{0,0}`$
$`\stackrel{~}{\mathrm{\Psi }}_{10}^{}u\rho +(\mathrm{\Psi }_{10}^{}๐_{K_\mathrm{\Sigma }})vw\stackrel{~}{\mathrm{\Psi }}_{10}`$ $`={\displaystyle \frac{1}{2}}\mathrm{\Psi }^{}\lambda _{02}^{1,0}+\mathrm{\Psi }_{12}^{1,0}\lambda _{22}^{0,0}+\mathrm{\Psi }_{10}^{0,0}\left({\displaystyle \frac{1}{2}}\lambda _{02}^{1,0}+\lambda _{00}^{0,0}\right)\lambda _{11}^{0,0}\mathrm{\Psi }_{10}^{0,0}`$
$`\lambda _{11}^{0,0}\mathrm{\Psi }_{12}^{1,0}`$
$`=\mathrm{\Psi }_{10}^{0,0}\lambda _{02}^{1,0}+\mathrm{\Psi }_{12}^{1,0}\lambda _{22}^{0,0}\lambda _{11}^{0,0}\mathrm{\Psi }_{12}^{1,0}+\mathrm{\Psi }_{10}^{0,0}\lambda _{00}^{0,0}\lambda _{11}^{0,0}\mathrm{\Psi }_{10}^{0,0}`$
Then we obtain
(A.151)
$$D_{10}\left[\begin{array}{c}\lambda _{02}^{1,0}\\ \lambda _{22}^{0,0}\\ \lambda _{00}^{0,0}\\ \lambda _{11}^{0,0}\\ \lambda _{12}^{0,1}\\ \lambda _{01}^{0,1}\end{array}\right]=\left[\begin{array}{c}\mathrm{\Psi }_{01}^{1,0}\lambda _{11}^{0,0}\lambda _{02}^{1,0}\mathrm{\Psi }_{21}^{0,0}\lambda _{00}^{0,0}\mathrm{\Psi }_{01}^{1,0}\\ \mathrm{\Psi }_{21}^{0,0}\lambda _{11}^{0,0}\lambda _{22}^{0,0}\mathrm{\Psi }_{21}^{0,0}\\ \mathrm{\Psi }_{10}^{0,0}\lambda _{00}^{0,0}\lambda _{11}^{0,0}\mathrm{\Psi }_{10}^{0,0}\\ \mathrm{\Psi }_{10}^{0,0}\lambda _{02}^{1,0}+\mathrm{\Psi }_{12}^{1,0}\lambda _{22}^{0,0}\lambda _{11}^{0,0}\mathrm{\Psi }_{12}^{1,0}\\ \overline{}\lambda _{02}^{1,0}\mathrm{\Psi }_{01}^{1,0}\lambda _{12}^{0,1}\lambda _{01}^{0,1}\mathrm{\Psi }_{12}^{1,0}+\mathrm{\Psi }_{02}^{1,1}\lambda _{22}^{0,0}\lambda _{00}^{0,0}\mathrm{\Psi }_{02}^{1,1}\\ \overline{}\lambda _{22}^{0,0}\mathrm{\Psi }_{21}^{0,0}\lambda _{12}^{0,1}\\ \overline{}\lambda _{00}^{0,0}\lambda _{01}^{0,1}\mathrm{\Psi }_{10}^{0,0}\\ \overline{}\lambda _{11}^{0,0}\mathrm{\Psi }_{10}^{0,0}\lambda _{01}^{0,1}\lambda _{12}^{0,1}\mathrm{\Psi }_{21}^{0,0}\end{array}\right]$$
The hypercohomology group $`^0(C(๐ฆ,๐ฆ^{}))`$ is isomorphic to the quotient $`\text{Ker}(D_{10})/\text{Im}(D_{0,1})`$. We would like to compare this space with the space of degree zero morphisms $`H_{\mathrm{Tr}(\stackrel{~}{๐})}^0(๐ฏ,๐ฏ^{})`$. The later is the $`0`$-th cohomology of the complex $`H_{\text{Pre-Tr}(\stackrel{~}{๐})}(๐ฏ,๐ฏ^{})`$ defined in section 5.1, equations (75), (76). Given equations (A.145), (A.146) and (A.148), it is straightforward to check that
(A.152)
$$H_{\text{Pre-Tr}(\stackrel{~}{๐})}^l(๐ฏ,๐ฏ^{})^l$$
for $`l=1,0,1`$. One can also show that the differentials are identical by specializing formulas (76) to the case at hand taking into account (73). Then, for $`\lambda _{n^{},n}^{q,p}H_{\stackrel{~}{๐}}^l((F_n,n1),(F_n^{}^{},n^{}1))`$, $`n,n^{}=0,1,2`$, $`p,q=0,1`$, $`l=2q+p+n^{}n`$, we obtain
(A.153)
$$d\lambda _{n^{},n}^{q,p}=\overline{}\lambda _{n,m}^{q,p}+\underset{m=0}{\overset{2}{}}\underset{r,s=0}{\overset{1}{}}\left[(1)^{p(n^{}m)}\mathrm{\Psi }_{m,n^{}}^{s,r}\lambda _{n^{},n}^{q,p}(1)^{(r+1)(nn^{})+p}\lambda _{n^{},n}^{q,p}\mathrm{\Psi }_{n,m}^{s,r}\right]$$
where by convention $`\mathrm{\Psi }_{m,n^{}}^{s,r}=0`$ unless $`2s+r+mn^{}=1`$ and $`\mathrm{\Psi }_{n,m}^{s,r}=0`$ unless $`2s+r+nm=1`$. It is a simple exercise to check that the differentials (A.153) agree with (A.149) and (A.151). This proves formula (103) in degree zero. The proof in degree one is very similar.
|
warning/0506/cond-mat0506439.html
|
ar5iv
|
text
|
# Chaos, Coherence and the Double-Slit Experiment
## I Introduction
Ever since the inception of quantum theory, questions have been raised related to its connection to classical physics wheeler83 . From a dynamical point of view, it is generally accepted that the Liouville and Schrรถdinger equations deliver the same time-evolution for short enough times, $`t<t_\mathrm{E}`$. In both chaotic and integrable dynamical systems, $`t_\mathrm{E}`$ goes to infinity in the semiclassical limit of large quantum numbers. In chaotic systems, however, the quantum breaktime $`t_\mathrm{E}=\lambda ^1|\mathrm{ln}\mathrm{}_{\mathrm{eff}}|`$ does so only logarithmically slowly in the effective Planckโs constant $`\mathrm{}_{\mathrm{eff}}`$ ($`\lambda `$ is the systemโs Lyapunov exponent) berman78 . For $`t>t_\mathrm{E}`$, the standard view is that external sources of decoherence have to be invoked in order to reestablish the correspondence between quantum and classical mechanics aak ; joos ; caldeira85 ; buttiker86a .
Arguing that the necessity of external degrees of freedom for the quantum to classical transition remains unclear (see for instance Ref.casati95 ), Casati and Prosen recently performed a numerical double-slit experiment casati04 . The set-up they considered is sketched in Fig. 1. One pierces two openings of width $`W`$ in an otherwise closed cavity. Inside the cavity, a particle of mass $`m1`$ is prepared in an initial wavepacket of minimal spread in momentum. The system is considered to be semiclassical, i.e. the ratio of the linear system size and the particleโs de Broglie wavelength is big $`L/\nu =kL/2\pi 1`$. As time goes by, the particle leaks out of the cavity with an average decay time $`\tau _dL^2/(Wv)\tau _\mathrm{f}`$ much larger than the time of flight $`\tau _\mathrm{f}L/v`$ across the cavity, $`v`$ being the particleโs velocity. That is, the particle bounces many times between the cavityโs boundary before exiting. One then records the integrated probability current $`I(x)`$ through the screen, (from now on we set $`\mathrm{}1`$)
$`I(z)`$ $`=`$ $`{\displaystyle _0^{\mathrm{}}}dt\mathrm{Im}[\psi ^{}(z,y;t)_y\psi (z,y;t)]_{y=y_\mathrm{o}}.`$ (1)
Two different situations were considered, where the cavity was either integrable (an isosceles right triangle) or chaotic (where the hypotenuse was replaced by a circular arc). In the integrable case, numerics showed that $`I(x)`$ exhibits the expected interference fringes. Those fringes were however absent in the chaotic case where $`I(x)`$ takes on its classical, structureless shape. These results prompted Casati and Prosen to draw two conclusions: (i) the double-slit set-up provides for a โvivid and fundamental illustration of the manifestation of classical chaos in quantum mechanicsโ, and (ii) dynamical chaos alone (i.e. without any external source of noise, or any coupling to an external bath or environment) can produce sufficient randomization of quantum-mechanical phases resulting in a quantum to classical transition in the semiclassical limit. The reasoning path leading to conclusion (ii) is qualitatively the following. Due to the long lifetime of the particle inside the cavity, the wavepacket must hit the cavity walls many times before exiting. Semiclassically, the wavepacket follows many classical trajectories exiting at different times, and thus accumulating different action phases. In the regular case, because the particleโs initial momentum is well defined, the action phases accumulated along all those trajectories are correlated. In the chaotic case however, the initial momentum uncertainty grows exponentially with time and the classical trajectories have a broad, continuous distribution of duration. Hence they acquire a random distribution of action phases. Based on this observation, Casati and Prosen concluded that this phase randomization prevents interference fringes to occur, in agreement with their numerical calculation. It is important to realize at this point that at any given point and time, the phase of the wavefunction is uniquely defined, and can in principle be deterministically obtained from the initial condition.
There is no controversy related to conclusion (i). Conclusion (ii) however, not only challenges the standard view according to which long-time quantum-classical correspondence requires coupling to external degrees of freedom, but has to be reconciled with well established mesoscopic physics results textbooks . It is indeed well known that both transport aronov87 ; webb85 ; yacoby95 and thermodynamical cheung89 ; vonoppen91 ; levy90 properties of multiply connected mesoscopic samples threaded by a magnetic flux display coherent flux-periodic oscillations of a purely quantal origin. It is doubtful that all experimentally investigated systems are integrable. From a theoretical point of view, such oscillations have been moreover predicted for disordered, diffusive samples with point-like impurities which are arguably as good โphase-randomizersโ (in the sense given above) as deterministic chaos. The even flux-harmonics (those having a period in the applied flux $`\varphi `$ of $`\varphi _0/2n`$, with $`n`$ a positive integer and $`\varphi _0=h/e`$ the flux quantum) of these oscillations even survive disorder averaging aronov87 ; vonoppen91 , and in the case of transport experiments โร la Sharvin and Sharvinโ, the amplitude of the Aharonov-Bohm oscillations of the conductance are mostly insensitive to the amount of disorder aronov87 . Clearly, conductance is insensitive to the โdynamical decoherenceโ scenario of conclusion (ii). The purpose of this article is to reconcile the numerical experiment of Ref. casati04 with the established theoretical and experimental wisdom of mesoscopic physics, as well as to investigate dephasing in ballistic mesoscopic systems.
We will present a comparative semiclassical calculation of the outgoing probability current in the Aharonov-Bohm two-slit set-up of Fig. 2(a) (similar to the set-up of Ref. casati04 , see Fig. 1) and of the conductance in the set-up of Fig. 2(b). In both cases, a cavity is connected to two intermediate left (L) and right (R) leads carrying $`N_{\mathrm{R},\mathrm{L}}1`$ transport channels. These intermediate leads eventually merge and the loop they form is threaded by a magnetic flux $`\varphi `$. In the transport set-up of Fig. 2(b), the cavity is in addition connected to a current-injecting lead carrying $`N_\mathrm{B}`$ transport channels. In the two instances, we consider ideally connected, i.e. non reflecting leads, and will restrict ourselves to the situation where the number $`N_\mathrm{T}`$ of outgoing channels obeys $`N_\mathrm{T}N_\mathrm{L}+N_\mathrm{R}`$. One can then neglect processes where particles circulate several times around the Aharonov-Bohm loop. As but one consequence, our semiclassical treatment is fully unitary, but flux-dependent weak localization corrections are absent. Such corrections have been considered in a different ballistic set-up in Ref. see03 . The set-up of Fig. 2(b) in the diffusive regime has been considered in Ref. mirlin04 . A nonunitary semiclassical treatment of the set-up of Fig. 2(b) considering backscattering due to pairs of time-reversed paths has been presented in Ref. kawabata96 .
Our conclusion is that, while there is nothing wrong with most of the reasonings and the numerical results of Ref. casati04 , decoherence cannot be claimed to occur when one observable does not display interference patterns, but when this is the case for all possible observables. The conductance experiment of Fig. 2(b) will be shown to exhibit sample-dependent Aharonov-Bohm oscillations in both cases of an integrable and a chaotic cavity. We will see how these oscillations disappear as dephasing is introduced. Our results support the standard wisdom according to which the quantum to classical crossover requires a coupling to external degrees of freedom.
The paper is organized as follows. In Section II we present a semiclassical calculation for an Aharonov-Bohm set-up similar to the two-slit experiment considered in Ref. casati04 . This calculation is extended to the calculation of the conductance in an Aharonov-Bohm transport set-up in Section III. In Section IV we introduce dephasing by means of a fluctuating electric potential in one arm of the Aharonov-Bohm loop, and investigate the associated disappearance of flux-dependent interference fringes. In the final Section V we will summarize our findings and discuss future directions and open questions.
## II Two-Slit Set-up
We first consider the Aharonov-Bohm two-slit set-up of Fig. 2(a), where an initial wavepacket is prepared inside the cavity. The latter is connected to two outgoing leads carrying many transverse channels. The leads eventually merge, forming a loop threaded by a magnetic flux $`\varphi `$. Once one integrates over a cross-section of the outgoing lead, the situation is fully similar to Ref. casati04 , with $`\varphi `$ playing the role of the coordinate $`z`$ along the screen (see Fig. 1). We consider an initial Gaussian wavepacket $`\psi _0(๐ซ_1)=(\pi \nu ^2)^{d/4}\mathrm{exp}[i๐ฉ_0(๐ซ_1๐ซ_0)|๐ซ_1๐ซ_0|^2/2\nu ^2]`$, and approximate its time-evolution semiclassically by ($`H=v^2/2`$; remember that we set $`m1`$)
$`๐ซ|\mathrm{exp}(iHt)|\psi _0={\displaystyle ๐๐ซ_1\underset{s}{}K_s(๐ซ,๐ซ_1;t)\psi _0(๐ซ_1)},`$ (2a)
$`K_s^H(๐ซ,๐ซ_1;t)=C_s^{1/2}\mathrm{exp}[iS_s(๐ซ,๐ซ_1;t)i\pi \mu _s/2].`$ (2b)
Compared to Ref. casati04 , the Heisenberg uncertainty is evenly distributed between momentum and spatial coordinates in our choice of an initial state. This should not matter in a chaotic cavity, but may affect the outcome of the experiment in a regular cavity. The semiclassical propagator (2b) is expressed as a sum over classical trajectories (labeled $`s`$) connecting $`๐ซ`$ and $`๐ซ_1`$ in the time $`t`$. For each $`s`$, the partial propagator contains the action integral $`S_s^H(๐ซ,๐ซ_1;t)`$ along $`s`$, a Maslov index $`\mu _s`$, and the determinant $`C_s`$ of the stability matrix. Because of the cavity openings, if $`๐ซ`$ in Eqs. (2) is inside the cavity, then the sum runs only over those classical trajectories that have not yet escaped at time $`t`$, whereas if $`๐ซ`$ lies somewhere in a lead, it runs over the trajectories that went exactly once through either of the openings to reach $`๐ซ`$. Here, we are concerned with this latter case, putting $`๐ซ=(x,y=y_\mathrm{o})`$ at the horizontal position $`x`$ on a cross-section $`๐`$ of the outgoing lead defined by $`y=y_\mathrm{o}`$ \[see Fig. 2(a)\]. Later on, we will integrate over $`x`$.
The semiclassical expression for the time-integrated probability current (1) is given by
$`I(x,\varphi )`$ $`=`$ $`{\displaystyle \frac{v}{(\pi \nu ^2)^{d/2}}}{\displaystyle _0^{\mathrm{}}}dt{\displaystyle d๐ซ_1d๐ซ_2}`$ (3)
$`\times `$ $`{\displaystyle \underset{s(t),s^{}(t)}{}}[C_sC_s^{}]^{1/2}\mathrm{cos}\theta _s`$
$`\times `$ $`\mathrm{exp}\left[i\{S_s(x,y_\mathrm{o};๐ซ_1;\varphi ,t)S_s^{}(x,y_\mathrm{o};๐ซ_2;\varphi ,t)\}\right]`$
$`\times `$ $`\mathrm{exp}\left[i\pi (\mu _s^{}\mu _s)/2\right]\mathrm{exp}\left[i๐ฉ_0(๐ซ_1๐ซ_2)\right]`$
$`\times `$ $`\mathrm{exp}\left[(|๐ซ_1๐ซ_0|^2+|๐ซ_2๐ซ_0|^2)/2\nu ^2\right],`$
where we used $`_yS_s=v_{y,s}=v\mathrm{cos}\theta _s`$, with $`v_{y,s}`$ the velocity in $`y`$-direction and thus $`\theta _s`$ the angle of incidence, as the path $`s`$ crosses $`๐`$ at time $`t`$.
The first step in the calculation of $`I(x,\varphi )`$ is to linearize $`S_s(x,y_\mathrm{o};๐ซ_1;\varphi ,t)S_s(x,y_\mathrm{o};๐ซ_0;\varphi ,t)๐ฉ_s(๐ซ_1๐ซ_0)`$, with $`๐ฉ_s`$ the initial momentum on path $`s`$. This is justified by our choice of a narrow initial wavepacket. One it then left with Gaussian integrals over $`๐ซ_{1,2}`$. Enforcing a stationary phase condition, the dominant, classical contributions to $`I(x,\varphi )`$ are identified as those with $`s=s^{}`$. Under our assumption of a final number of transport channels $`N_\mathrm{T}N_\mathrm{L}+N_\mathrm{R}`$ roughly equal or somehow larger than the sum of transport channels in the intermediate leads forming the Aharonov-Bohm loop, single trajectories do not enclose any flux. Diagonal contributions with $`s=s^{}`$ are thus flux independent. Writing $`I(x,\varphi )=I_0(x)+I_\varphi (x)`$, one has
$`I_0(x)`$ $`=`$ $`v(4\pi \nu ^2)^{d/2}`$
$`\times `$ $`{\displaystyle _0^{\mathrm{}}}dt{\displaystyle \underset{s(t)}{}}C_s\mathrm{cos}\theta _s\mathrm{exp}[\nu ^2|๐ฉ_0๐ฉ_s|^2].`$
The stationary phase leading to the diagonal approximation $`s=s^{}`$ is justified once one averages over an interval of energy $`\delta E`$ which is classically small (i.e. which does not modify the trajectories) but quantum-mechanically large (i.e. such that $`\delta E\tau _d1`$). This is indicated by brackets in Eq. (II). The average value $`I_0(x)`$ is calculated under the assumption that the cavity is ergodic, in particular that the wavefunction will eventually leak out of it completely. That is, the time-integrated current through $`๐`$ must be equal to one and one has
$`{\displaystyle \frac{2v}{\pi }}{\displaystyle _0^W}dx{\displaystyle _0^{\mathrm{}}}dt|\psi (x,y_\mathrm{o};t)|^2`$ $`=1,`$ (5)
where a factor $`2/\pi `$ originated from averaging the incidence angle on $`๐`$ in the interval $`[\pi /2,\pi /2]`$. This provides the semiclassical sum rule
$`v(4\pi \nu ^2)^{d/2}{\displaystyle _0^W}dx{\displaystyle _0^{\mathrm{}}}dt`$
$`\times {\displaystyle \underset{s(t)}{}}C_s\mathrm{exp}[\nu ^2`$ $`|๐ฉ_0๐ฉ_s|^2]=\pi /2.`$ (6)
The classical time-integrated current through $`๐`$ is then obtained as
$`{\displaystyle _0^W}dxI_0(x)`$ $`=`$ $`1.`$ (7)
In the limit of a wide outgoing lead, $`W\nu `$, the probability current is ergodically distributed over $`๐`$ so that the average current per unit length is given by $`I_0(x)W^1`$.
After this warm-up calculation we turn our attention to the flux-dependent part $`I_\varphi (x)`$. It correspond to pairs of paths $`s`$ and $`s^{}`$ in Eq. (3) exiting through different arms of the AB ring, and evidently they are not included in the diagonal approximation $`s=s^{}`$. Furthermore, no stationary phase approximation can be systematically enforced to identify them, which reflects the fact that they vanish on average. That is to say $`I_\varphi (x)=0`$, once it is averaged over different initial conditions, a sufficiently large energy interval or an ensemble of different cavities. This is but one consequence of our choice of forward scattering processes only at the merging point of the intermediate leads.
To investigate the behavior of $`I_\varphi (x)`$ for a given cavity and/or initial wavepacket preparation, we proceed to calculate $`I_\varphi ^2(x)`$, the squareroot of which gives the value of the flux-dependent part of $`I(x)`$ for a typical experimental realization. Our approach is similar in spirit to the one followed in Ref. cheung89 in the context of persistent currents. A similar sum rule as (II) is helpful in computing $`I_\varphi ^2(x)`$, and with little extra work we will see that $`I_\varphi ^2(x)(1\mathrm{exp}[\tau _{\mathrm{erg}}/\tau _d])^2(kL)^1(\tau _{\mathrm{erg}}/\tau _d)^2(kL)^1`$, where $`\tau _{\mathrm{erg}}`$ is the ergodic time. In a chaotic cavity, it is generally given by few times the time of flight across the cavity, so that $`\tau _{\mathrm{erg}}/\tau _dW/L`$. In the numerical experiment of Ref. casati04 , both the ratio of the width of the openings to the linear system size and the inverse semiclassical parameter $`kL`$ are much smaller than one, inducing the disappearance of the interference fringes.
Noting that $`S_s(x,y_\mathrm{o};๐ซ_0;\varphi ,t)=S_s(x,y_\mathrm{o};๐ซ_0;t)\pm \pi \varphi /\varphi _0`$, where the $`{}_{}{}^{\prime \prime }+_{}^{\prime \prime }`$ and $`{}_{}{}^{\prime \prime }_{}^{\prime \prime }`$ signs correspond to trajectories going through the right and the left intermediate lead respectively, linearizing in $`๐ซ_{1,2}๐ซ_0`$ and performing the resulting Gaussian integrals over $`๐ซ_{1,2}`$ as above, one has
$`I_\varphi (x)I_\varphi (x^{})`$ $`=`$ $`4v^2(4\pi \nu ^2)^d\mathrm{cos}^2[2\pi \varphi /\varphi _0]{\displaystyle _0^{\mathrm{}}}dt_1{\displaystyle _0^{\mathrm{}}}dt_2{\displaystyle \underset{s_1,s_3L}{}}{\displaystyle \underset{s_2,s_4R}{}}\left({\displaystyle \underset{i=1}{\overset{4}{}}}C_i\right)^{1/2}\mathrm{cos}\theta _1\mathrm{cos}\theta _4`$ (8)
$`\times `$ $`\mathrm{exp}\left[i\left\{S_1(x,y_\mathrm{o};๐ซ_0;t_1)S_2(x,y_\mathrm{o};๐ซ_0;t_1)S_3(x^{},y_\mathrm{o};๐ซ_0;t_2)+S_4(x^{},y_\mathrm{o};๐ซ_0;t_2)\right\}\right]`$
$`\times `$ $`\mathrm{exp}\left[\nu ^2\left\{|๐ฉ_0๐ฉ_1|^2+|๐ฉ_0๐ฉ_2|^2+|๐ฉ_0๐ฉ_3|^2+|๐ฉ_0๐ฉ_4|^2\right\}/2\right],`$
where we shortened the notation, i.e. $`\theta _i=\theta _{s_i}`$, $`S_i=S_{s_i}`$ and so forth. It is important to keep in mind that $`s_1`$ and $`s_2`$ exit the cavity after a time $`t_1`$, while the time of escape is $`t_2`$ for the other two trajectories $`s_3`$ and $`s_4`$. Because trajectories exit via two different arms of the AB loop, the only stationary phase condition that can be satisfied is to set $`s_1=s_3`$ and $`s_2=s_4`$, which then requires to set $`x=x^{}`$ with accuracy $`\nu `$, and $`t_1t_2`$, with an accuracy given by the time $`\tau ^{}\mathrm{}/E`$ necessary for the classical ballistic flow at energy $`E`$ to accumulate an action $`\mathrm{}`$. We substitute $`dt_1dt_2\tau ^{}dt_1`$ to get
$$I_\varphi (x)I_\varphi (x^{})\delta _\nu (xx^{})\mathrm{\hspace{0.33em}4}v^2\tau ^{}(4\pi \nu ^2)^d\mathrm{cos}^2[2\pi \varphi /\varphi _0]_0^{\mathrm{}}dt_1\underset{s_1L}{}\underset{s_2R}{}\underset{i=1}{\overset{2}{}}\left\{C_i\mathrm{cos}\theta _i\mathrm{exp}\left[\nu ^2|๐ฉ_0๐ฉ_1|^2\right]\right\},$$
(9)
where $`\delta _\nu (xx^{})`$ enforces the condition $`x=x^{}`$ with an accuracy $`๐ช(\nu )`$. Because there is only one time integral but two summations over classical paths, one cannot use Eq. (II) directly. Assuming that the system is ergodic, which means in particular that for times long enough, $`t\tau _{\mathrm{erg}}\tau _f`$, spatial averages equal time averages, one writes
$`I_\varphi ^2(x)`$ $``$ $`4v^2(4\pi \nu ^2)^d\mathrm{cos}^2[2\pi \varphi /\varphi _0][\tau ^{}{\displaystyle _0^{\tau _{\mathrm{erg}}}}\mathrm{d}t_1{\displaystyle \underset{s_1L}{}}{\displaystyle \underset{s_2R}{}}{\displaystyle \underset{i=1}{\overset{2}{}}}\left\{C_i\mathrm{cos}\theta _i\mathrm{exp}[\nu ^2|๐ฉ_0๐ฉ_1|^2]\right\}`$ (10)
$`+`$ $`{\displaystyle _{\tau _{\mathrm{erg}}}^{\mathrm{}}}\mathrm{d}t_1\underset{T\mathrm{}}{lim}{\displaystyle \frac{\tau ^{}}{T}}{\displaystyle _{\tau _{\mathrm{erg}}}^T}\mathrm{d}t_2{\displaystyle \underset{s_1L}{}}{\displaystyle \underset{s_2R}{}}{\displaystyle \underset{i=1}{\overset{2}{}}}\left\{C_i\mathrm{cos}\theta _i\mathrm{exp}[\nu ^2|๐ฉ_0๐ฉ_1|^2]\right\}].`$
Here, the second term inside the brackets corresponds to trajectories $`s_1(t_1)`$ and $`s_2(t_2)`$ exiting at different times. Its contribution to the integrated current $`๐xI_\varphi ^2(x)`$ can be calculated using the sum rule (II) and making the assumption that the current is homogeneously distributed on $`๐`$. We find that it vanishes $`\mathrm{lim}_T\mathrm{}\tau ^{}/T`$. The first, pre-ergodic term is highly non-universal and we cannot calculate it generically. We can however give an estimate to its amplitude using
$`{\displaystyle _0^{\tau _{\mathrm{erg}}}}dt_1f(t_1)g(t_1)\tau _{\mathrm{erg}}^1{\displaystyle _0^{\tau _{\mathrm{erg}}}}dt_1dt_2f(t_1)g(t_2)\tau _{\mathrm{erg}}^1(1\mathrm{exp}[\tau _{\mathrm{erg}}/\tau _d])^2{\displaystyle _0^{\mathrm{}}}dt_1dt_2f(t_1)g(t_2).`$ (11)
The first relation results from removing the requirement that both trajectories $`s_1`$ and $`s_2`$ in Eq. (10) exit at the same time, and to obtain the second one, we used the measure of pre-ergodic trajectories in an open chaotic cavity $`\stackrel{~}{\rho }(t\tau _{\mathrm{erg}})=\tau _d^1_0^{\tau _{\mathrm{erg}}}dt\mathrm{exp}[t/\tau _d]`$, where $`\rho (t)=\tau _d^1\mathrm{exp}[t/\tau _d]`$ is the distribution of dwell times through a chaotic system Bauer90 . Using $`\tau ^{}/\tau _{\mathrm{erg}}(kL)^1`$, and assuming again an homogeneous distribution of $`I(x)`$ on $`๐`$, we finally get the typical flux-dependent probability current as
$`I_\varphi ^2(x)^{1/2}`$ $``$ $`\mathrm{cos}[2\pi \varphi /\varphi _0](1\mathrm{exp}[\tau _{\mathrm{erg}}/\tau _d])\sqrt{{\displaystyle \frac{\tau ^{}}{\tau _{\mathrm{erg}}}}}I_0(x).`$ (12)
We believe that Eq. (12) gives an upper bound for the typical flux-dependent part of the probability current in the case of a chaotic cavity. One sees that, compared to $`I_0`$, $`I_\varphi ^2^{1/2}`$ is suppressed by a prefactor $`(1\mathrm{exp}[\tau _{\mathrm{erg}}/\tau _d])(kL)^{1/2}`$. In the chaotic configuration of Ref. casati04 , the dwell time is approximately several hundreds of times larger than the ergodic time. Together with $`kL=180`$, this leads to the suppression of the flux oscillations in a given sample by a relative factor of at least $`(\tau _{\mathrm{erg}}/\tau _d)(kL)^{1/2}10^3`$ compared to the average current value.
While it is always risky to make generic statistical statements on regular systems, it is reasonable to expect that in this case, the pre-ergodic terms in Eq. (9) provide for most of the contributions to $`I_\varphi ^2(x)`$. This is so, since for regular systems, $`\tau _{\mathrm{erg}}`$ is much larger than in a chaotic system, and even diverges in most instances, regular systems being usually not ergodic. Moreover, integrable systems exhibit periodicities and quasiperiodicities and a persistence of correlations over very large times. Starting from Eq. (8), one may thus pair trajectories either with $`\tau ^{}\mathrm{}/E`$, or even completely relaxing the restriction $`|t_1t_2|\tau ^{}`$. One then gets the best case scenario result that
$`I_\varphi ^2(x)_{\mathrm{reg}}^{1/2}`$ $``$ $`\mathrm{cos}[2\pi \varphi /\varphi _0]I_0(x),`$ (13)
i.e. the flux-dependent probability current is of the same magnitude as its classical part $`I_0(x)`$. This is also in agreement with Ref. casati04 . One should stress however that the result (13) cannot be expected to hold generically. In particular, we believe that the choice made in Ref. casati04 of an initial state with narrowest momentum spread is necessary to get interference fringes satisfying (13). Presumably the choice of direction of momentum also plays a role.
To summarize this section, we have shown why the interference fringes disappear for a two-slit experiment out of a chaotic cavity. The main result of this section, Eq. (12), can be checked numerically by increasing the width $`W`$ of the slits or varying $`kL`$, or both, in the numerical experiment of Ref. casati04 . More qualitatively, we argued that in well chosen situations, the interference fringes have a magnitude comparable to the classical probability current if the cavity is regular.
## III Transport Set-up
We next focus on the transport set-up shown in Fig. 2(b). We write the conductance as a sum of a classical nd a flux-dependent part, $`g=g_0+g(\varphi )`$. We use the semiclassical framework developed in Ref. Bar93 . We start from the scattering approach which relates transport properties to the systemโs scattering matrix scatg
$`๐ฎ=\left(\begin{array}{cc}๐ซ\hfill & ๐ญ^{}\hfill \\ ๐ญ\hfill & ๐ซ^{}\hfill \end{array}\right).`$ (16)
For the two terminal geometry we consider, $`๐ฎ`$ is a 2-block by 2-block matrix, written in terms of transmission ($`๐ญ`$ and $`๐ญ^{}`$) and reflection ($`๐ซ`$ and $`๐ซ^{}`$) matrices. From $`๐ฎ`$, the systemโs conductance is given by $`g=\mathrm{Tr}(๐ญ^{}๐ญ)`$ ($`g`$ is expressed in units of $`e^2/h`$).
From Ref. Bar93 , the matrix elements $`t_{mn}`$ of the transmission matrix $`๐ญ`$ are written as
$$t_{mn}=\sqrt{\frac{\pi \mathrm{}}{2W_\mathrm{B}W_\mathrm{T}}}\underset{s}{}\frac{\mathrm{\Phi }_s\mathrm{exp}[iS_s(๐ซ_\mathrm{B},๐ซ_\mathrm{T};E)]}{|\mathrm{cos}\theta _\mathrm{B}^{(m)}\mathrm{cos}\theta _\mathrm{T}^{(n)}M_{21}^s|^{1/2}}.$$
(17)
The sum runs over all classical scattering trajectories entering the cavity with an angle $`\pm \theta _\mathrm{B}^{(m)}`$ at any point $`๐ซ_\mathrm{B}=(x,y_\mathrm{i})`$ on a cross-section $`๐_\mathrm{B}`$ of the bottom lead (of geometric width $`W_\mathrm{B}`$) and exiting it with an angle $`\pm \theta _\mathrm{T}^{(n)}`$ at any point $`๐ซ_\mathrm{T}=(x^{},y_\mathrm{o})`$ on a cross-section $`๐_\mathrm{T}`$ of the top lead (of geometric width $`W_\mathrm{T}`$). The channel indices $`(m,n)`$ specify the entrance and exit angles as $`\mathrm{sin}\theta _\mathrm{B}^{(m)}=\pi \overline{m}/k_FW_\mathrm{B}`$ and $`\mathrm{sin}\theta _\mathrm{T}^{(n)}=\pi \overline{n}/k_FW_\mathrm{T}`$, $`\overline{m}=\pm m`$, $`\overline{n}=\pm n`$, while $`S_s(๐ซ_\mathrm{B},๐ซ_\mathrm{T};E)`$ gives the classical action accumulated along $`s`$. Finally $`M_{21}^s=v_{}/dq_{}`$ is an element of the monodromy matrix (the $``$-direction is normal to the cross-sections), and there is a phase factor $`\mathrm{\Phi }_s=\mathrm{sgn}(\overline{m})\mathrm{sgn}(\overline{n})\mathrm{exp}[i\pi (\overline{m}x_\mathrm{B}/W_\mathrm{B}\overline{n}x_\mathrm{T}/W_\mathrm{T}\mu _s/2+1/4)]`$.
All one needs to calculate the average conductance of a chaotic cavity is the following sum rule, valid in the regime of classical ergodicity Bar93
$`{\displaystyle \underset{s(x_\mathrm{B},x_\mathrm{T};\theta _\mathrm{B}^{(m)},\theta _\mathrm{T}^{(n)})}{}}{\displaystyle \frac{\delta (\tau \tau _s)}{|M_{21}^s|}}{\displaystyle \frac{\mathrm{cos}\theta _\mathrm{B}^{(m)}\mathrm{cos}\theta _\mathrm{T}^{(n)}}{\mathrm{\Sigma }(E)}}\delta x_\mathrm{B}\delta x_\mathrm{T}\stackrel{~}{\rho }(\tau ).`$ (18)
In contrast to Eq. (17), the sum in Eq. (18) is restricted to phase-space trajectories with a well resolved position and momentum direction on $`๐_\mathrm{B}`$ and $`๐_\mathrm{T}`$, up to uncertainties $`\delta x_\mathrm{B},\delta x_\mathrm{T}\nu `$. Here, $`\mathrm{\Sigma }(E)=2\pi A`$ gives the volume of phase space that can be visited by an ergodic particle of energy $`E`$ in a cavity of area $`A`$, and $`\stackrel{~}{\rho }(\tau )=_\tau ^{\mathrm{}}\rho (t)dt=\mathrm{exp}[\tau /\tau _d]`$ gives the survival probability that a particle remains inside an open chaotic system for a time longer than, or equal to $`\tau `$. The meaning of the sum rule (18) is that at any time, surviving classical trajectories have a probability to exit the cavity given by the fraction of phase-space volume covered by the leads to the total accessible volume of phase-space.
From Eqs. (17) and (18), together with the relation $`\tau _d=\pi A/[v(W_\mathrm{B}+W_\mathrm{T})]`$, it is straightforward to calculate the average conductance within the diagonal approximation. One ends up with the classical conductance
$`g`$ $`=`$ $`{\displaystyle \underset{m,n}{}}{\displaystyle \frac{\pi \mathrm{}}{2W_\mathrm{B}W_\mathrm{T}}}{\displaystyle \underset{s}{}}\left|\mathrm{cos}\theta _\mathrm{B}^{(m)}\mathrm{cos}\theta _\mathrm{T}^{(n)}M_{21}^s\right|^1={\displaystyle \frac{N_\mathrm{B}N_\mathrm{T}}{N_\mathrm{B}+N_\mathrm{T}}},`$ (19)
where we used the relation between lead width and channel number $`N=\mathrm{Int}[k_\mathrm{F}W/\pi ]`$. As was the case for the probability current, the average conductance has no flux dependence since diagonally paired trajectories do not enclose any flux.
Following the procedure we applied to $`I^2(\varphi )`$, it is straightforward to calculate the squared typical value of the flux-dependent part of the conductance $`g^2(\varphi )`$ using Eqs. (17) and (18), and $`S_s(x,y_\mathrm{o};๐ซ_0;\varphi ,t)=S_s(x,y_\mathrm{o};๐ซ_0;t)\pm \pi \varphi /\varphi _0`$. One then has
$`g^2(\varphi )`$ $`=`$ $`{\displaystyle \frac{16\pi ^2\mathrm{}^2N_\mathrm{B}N_\mathrm{T}}{\mathrm{\Sigma }^2(E)}}\left({\displaystyle _0^{\mathrm{}}}dt\stackrel{~}{\rho }(t)\right)^2\mathrm{cos}^2(2\pi \varphi /\varphi _0)`$ (20)
$`=`$ $`4{\displaystyle \frac{N_\mathrm{B}N_\mathrm{T}}{(N_\mathrm{B}+N_\mathrm{T})^2}}\mathrm{cos}^2(2\pi \varphi /\varphi _0).`$
Compared to the square of Eq. (19), one sum over pairs of channel indices disappeared from Eq. (20) because of the stationary phase condition we enforced on each of the two pairs of orbits going through the left and right intermediate lead respectively.
Eq. (20) is the main result of this section. It shows the universality of the typical Aharonov-Bohm response of the conductance in our set-up in the chaotic case. For $`N_\mathrm{B}`$ and $`N_\mathrm{T}`$ not too different from each other, $`g^2(\varphi )`$ is independent on $`g`$. The survival of interference fringes in the transport set-up is a direct consequence of the fact that to extract the conductance, one works in energy representation. Once one writes the scattering matrix in time representation, the squared typical conductance is given by an expression similar to Eq. (3), with however two time integrals. This makes it much easier to extract stationary phase conditions, without going through the ergodicity tricks that were needed to go from Eq. (8) to Eq. (12), and explains the ease of calculation with which (20) is derived compared to its probability current counterpart of Eq. (12).
As was the case in the previous section for the probability current, we cannot calculate $`g^2(\varphi )`$ in the integrable case without relying on assumptions which are not necessarily well controlled. In particular, there is, to the best of our knowledge, no sum rule such as (18) for regular systems. As is the case for persistent currents in ballistic systems however vonOp93 , one expects a significantly increased magnetic response, well above the chaotic value (20), because in a regular system, the dwell time distribution is not exponential, but power-law $`\rho (\tau )\tau ^\beta `$ Bauer90 . In the best case scenario, one can expect a response given by the coherent sum of $`N`$ responses \[$`N=\mathrm{min}(N_\mathrm{B},N_\mathrm{T})`$\], leading to a flux dependence of a similar amplitude as the conductance itself. Here, further numerical experiments are needed to clarify the situation.
## IV Dephasing by a Fluctuating Potential
The results (12) and (20) derived above follow from a stationary phase condition. To satisfy the latter, one relies on the exact pairing of trajectories, i.e. setting $`s=s^{}`$ where applicable, and in this way, all accumulated action phases cancel two by two. This is no longer the case in the presence of an external dephasing source. In this case, phase differences inevitably occur in pairs of contracted trajectories due to the interaction with the external source of noise at different times along the trajectory. In this section, we finally discuss this occurrence and how dephasing destroys the Aharonov-Bohm interference fringes.
Following Ref. see01 , we consider that our system as a whole, including charged gates defining the cavity and the Aharonov-Bohm ring, is electrically neutral, as sketched in Fig. 3. This does not prevent local charge fluctuations to occur, which in their turn induce fluctuations of the electric potential felt by the electrons. This is a specific example of dephasing induced by an external source, in this case the electric charges on the gates defining the system, which must fluctuate to ensure that the fluctuations inside the circuit are compensated to make the whole system electrically neutral. These fluctuations result in dephasing, and without loss of generality, we will assume that they affect only electrons passing through one, say the left intermediate lead, during the traversal time $`\tau _\mathrm{L}=L_\mathrm{L}/v`$ through that lead.
We consider the case of weak coupling, where the trajectories are unaffected by the coupling to external degrees of freedom. Dephasing is introduced in our calculation via the substitution
$$S_s(x,y_\mathrm{o};๐ซ_0;t)S_s(x,y_\mathrm{o};๐ซ_0;t)+_0^{\tau _\mathrm{L}}๐t\phi _s(t).$$
(21)
Here $`\phi _s(t)`$ gives the additional action phase accumulated by an electron traveling on path $`s`$ and interacting with the dephasing source at time $`t`$.
Using the central limit theorem, Eqs. (12) and (20) have then to be multiplied by
$$\mathrm{exp}[_0^{\tau _L}๐t_1_0^{\tau _L}๐t_2\phi _s(t_1)\phi _s(t_2)_s/2],$$
(22)
where $`\mathrm{}_s`$ denotes an average over the distribution of phases on different classical trajectories. Further assuming an exponential decay of the phase correlator $`\phi _s(t_1)\phi _s(t_2)_s=\phi _s^2(0)_s\mathrm{exp}[|t_1t_2|/\tau _c]`$, one gets, in the limit $`\tau _c\tau _\mathrm{L}`$, an exponential suppression of the flux response
$`g^2(\varphi )`$ $`=`$ $`{\displaystyle \frac{N_\mathrm{B}N_\mathrm{T}}{(N_\mathrm{B}+N_\mathrm{T})^2}}\mathrm{cos}^2(2\pi \varphi /\varphi _0)e^{\tau _\mathrm{L}/\tau _\phi },`$ (23)
where $`\tau _\phi ^1=2\tau _c\phi _s^2(0)_s`$. In the limit of Nyquist noise, a self-consistent calculation of the phase correlator has been performed in Ref. see01 , within the one-potential approximation, i.e. assuming that the fluctuations of the electric potential are spatially homogeneous inside one arm. A linear temperature dependence of the dephasing rate was obtained, which in our case translates into
$$\tau _\phi ^1=2\tau _c\phi _s^2(0)_s=8\pi \gamma _\mathrm{L}^2k_BT/N_\mathrm{L}.$$
(24)
Here, $`\gamma _\mathrm{L}1`$ stands for the ratio between the electrochemical and the electrical capacitance of the left arm see01 . In the weak coupling limit we are considering, one has $`\gamma _\mathrm{L}1`$. Both the exponential damping of the Aharonov-Bohm flux and the linear temperature dependence of the dephasing rate are in agreement with the experimental results of Ref. han01 on Aharonov-Bohm conductance oscillations in few-channel ballistic systems. Our results (22)-(24) extend those of Ref. see01 to the many-channel case.
As a side remark, we note that in the other limit $`\tau _c\tau _\mathrm{L}`$, one gets a Gaussian suppression of the flux response in the traversal time $`\tau _\mathrm{L}`$,
$`g^2(\varphi )`$ $`=`$ $`{\displaystyle \frac{N_\mathrm{B}N_\mathrm{T}}{(N_\mathrm{B}+N_\mathrm{T})^2}}\mathrm{cos}^2(2\pi \varphi /\varphi _0)e^{\tau _\mathrm{L}^2/\tau _\phi \tau _c},`$ (25)
with the same dephasing time as above. This Gaussian damping has not been obtained previously. Indeed, previous works always assumed $`\delta `$-correlated phases, $`\phi _s(t_1)\phi _s(t_2)_s\delta (t_1t_2)`$, meaning $`\tau _c/\tau _\mathrm{L}0`$.
To close this chapter, we remark that the same dephasing behavior will occur in regular systems as long as the phase correlator decay fast enough. While in that case, an exponential decay is not at all obvious from a dynamical point of view, we stress that, in the limit of long traversal times $`\tau _\mathrm{L}\tau _c`$, the minimal requirement for an exponential damping as in Eq. (23) is a power law decay of the phase correlator $`\phi (t_1)\phi (t_2)=\phi (0)\phi (0)[\tau _c/(\tau _c+|t_1t_2|)]^\alpha `$ with $`\alpha >1`$.
## V Conclusion
We have presented a semiclassical calculation of the flux dependence of the probability current and the conductance in two distinct Aharonov-Bohm set-ups (see Fig. 2). We have shown how the interference fringes in the probability current disappear in chaotic systems in the case of cavities with large dwell times, whereas they may persist in the case of a regular cavity. This is in agreement with and sheds light on the numerical results of Ref. casati04 . Simultaneously, we showed how the situation is completely different in the transport set-up, where the flux response of the conductance becomes universal in the chaotic case. This universality is lost in the case of integrable cavities, where we conjectured that the flux response may be of the same order as the conductance itself.
In the transport set-up, we argued that dephasing from external degrees of freedom is necessary to wash out the flux-periodic interference fringes in the conductance. We introduced dephasing in a similar way as in Refs. see03 ; see01 and found that flux-dependent interference fringes in the conductance vanish exponentially, $`\mathrm{exp}[\tau _\mathrm{L}/\tau _\phi ]`$. Both this exponential damping and the linear temperature-dependence of the dephasing time (24) are in agreement with transport experiments on ballistic Aharonov-Bohm systems han01 . Our results confirm the standard view that external sources of decoherence are generally required to induce a complete quantumโclassical correspondence.
Our semiclassical treatment extends the results of Refs. see03 ; see01 to the many-channel case. Still, the dephasing behavior of Eqs. (22)-(23) relies on the one-potential approximation giving the linear temperature dependence of the phase correlator, Eq. (24). Because Ref. casati04 considered the other limit of sub-wavelength slits, it is likely that diffraction effects play a role there that was neglected here. However, we do not expect diffraction to alter the situation qualitatively.
One of our motivations was to reconcile the results of Ref. casati04 with well-known mesoscopic physics theoretical and experimental results. That is why we deliberately made the hypothesis of forward scattering only, that particles entering one of the intermediate leads (indicated by $`L`$ and $`R`$ in Fig. 2) are transferred to the outgoing lead with probability one. This is justified in the case where that latter lead is somehow wider than the two intermediate leads together, $`N_\mathrm{T}N_\mathrm{R}+N_\mathrm{L}`$. It would be interesting to lift that hypothesis, and consider the emergence of higher flux harmonics and of flux-dependent weak localization corrections to the average conductance, and the influence that dephasing has on them. We expect that the presence of weak-localization corrections would result in the usual Lorentzian damping of the amplitude of Aharonov-Bohm interference fringes in the disorder-averaged conductance (as opposed to the typical conductance calculated here). Further investigations are however necessary to confirm this.
## Acknowledgments
This work has been supported by the Swiss National Science Foundation. It is a pleasure to thank M. Bรผttiker for drawing our attention to Ref.casati04 and for several interesting discussions and comments.
|
warning/0506/math0506019.html
|
ar5iv
|
text
|
# Uniform almost everywhere domination
## 1. Introduction
### 1.1. Domination
Fast growing functions have been investigated in mathematics for over 90 years. Set theorists, for example, have investigated the structure $`\omega ^\omega /\text{Fin}`$ and the associated invariants of the continuum ever since Hausdorff constructed his $`(\omega _1,\omega _1^{})`$-gap \[Hausdorff:gap\]; today, this structure has a role to play in modern descriptive set theory.
Fast growing functions have deep connections with computability. A famous early example is that of Ackermannโs function, defined in 1928 \[Ackermann\]. This is a computable function that grows faster than any primitive recursive function. This example was useful in elucidating the mathematical concept of computability, an understanding reflected in Churchโs Thesis.
In the 1960s, computability theorists became interested in functions that grow faster than all computable functions.
###### Definition 1.1.
Let $`f,g:\omega \omega `$. The function $`f`$ *majorizes* $`g`$ if $`f(n)g(n)`$ for all $`n\omega `$. If $`f(n)g(n)`$ for all but finitely many $`n`$, then $`f`$ *dominates* $`g`$. These are written as $`fg`$ and $`f^{}g`$, respectively. We call $`f`$ *dominant* if it dominates all (total) computable functions.
Dominant functions were explored in conjunction with Postโs Program. The goal of Postโs Program was to find a โsparsenessโ property of the complement of a c.e. set $`A`$ that would ensure that $`A`$ is incomplete. Yates:65 proved that even maximal c.e. sets, which have the sparsest possible compliments among coninfinite c.e. sets, can be complete. This put an end to Postโs Program, but not to the study of sparseness properties.
Let $`p_A(n)`$ be the $`n^{\text{th}}`$ element of the complement of $`A`$. Having $`p_A`$ dominant would certainly imply that the complement of $`A`$ is sparse. On the other hand, Tennenbaum:63 and Martin:66\*1 showed that if $`A`$ is maximal, then $`p_A`$ is dominant. Furthermore, Martin characterized the Turing degrees of both the dominant functions and the maximal c.e. sets. He showed that there is dominant function of degree $`๐`$ iff $`๐`$ is high (i.e., $`\mathrm{๐}^{\prime \prime }๐^{}`$), and that every high c.e. degree contains a maximal set. Together, these results revealed a surprising connection between the structure of c.e. sets, the place of their Turing degree within the jump hierarchy, and domination properties of functions. Later research explored further connections between domination properties, algebraic properties and computational power.
In this paper, we consider the interaction between Lebesgue measure and domination. Motivated by results on dominating functions in generic extensions of set theory, MR2078930 introduced the notion of a *uniformly almost everywhere* (*a.e.*) *dominating* degree: a Turing degree $`๐`$ that computes a function $`f:\omega \omega `$ such that
$$\mu \{Z2^\omega :(g\omega ^\omega )[g_TZg^{}f]\}=1.$$
(Here $`\mu `$ denotes the Lebesgue measure on $`2^\omega `$.) We also call such a function $`f`$ *uniformly a.e. dominating*.
A natural goal is to characterize those Turing degrees that are uniformly a.e. dominating. A function of degree $`\mathrm{๐}^{}`$ that dominates almost all degrees was first constructed by Kurtz:81. (Kurtz used this result to exhibit a difference between the $`1`$-generic and the (weakly) $`2`$-generic degrees: the upward closure of the $`1`$-generic degrees has measure one \[Kurtz:81, Theorem 4.1\], while the upward closure of the (weakly) $`2`$-generic degrees has measure zero \[Kurtz:81, Corollary 4.3a\].) Since the collection of uniformly a.e. dominating degrees is closed upwards, Kurtzโs result implies that every degree $`\mathrm{๐}^{}`$ is in the class. On the other hand, a uniformly a.e. dominating function is dominant, and so by Martinโs result, every uniformly a.e. dominating degree is high. Thus, Dobrinen and Simpson asked whether either the class of complete degrees (degrees above $`\mathrm{๐}^{}`$) or the class of high degrees is identical to the class of uniformly a.e. dominating degrees.
Unfortunately, the truth lies somewhere in the middle. Binns, Kjos-Hanssen, Lerman and Solomon \[Binns.KjosHanssen.ea:nd\] showed that not every high degree is uniformly a.e. dominating, or even *a.e. dominating*, an apparently weaker notion also introduced by MR2078930. They gave two proofs. First, by a direct construction, they produced a high c.e. degree that is not a.e. dominating. (A similar result was independently obtained by Greenberg and Miller, although their example was $`\mathrm{\Delta }_2^0`$, not c.e.)
Second, Binns.KjosHanssen.ea:nd showed that if $`A`$ has a.e. dominating degree, then every set that is $`1`$-random over $`A`$ is $`2`$-random. If $`A`$ is also $`\mathrm{\Delta }_2^0`$, then by Nies:nd\*3, $`\mathrm{}^{}`$ is $`K`$-trivial over $`A`$ and so $`A`$ is *super-high* (i.e., $`A^{}_{tt}\mathrm{}^{\prime \prime }`$). By an index set calculation, there is a c.e. set that is high but not *super-high*, hence not a.e. dominating. It is open whether $`\mathrm{}^{}`$ being $`K`$-trivial over $`๐\mathrm{๐}^{}`$ implies that $`๐`$ is (uniformly) a.e. dominating; Kjos-Hanssen has some related results.
We prove that Dobrinen and Simpsonโs other suggested characterization of the uniformly a.e. dominating degrees also fails.
###### Theorem 1.2.
There is an incomplete (c.e.) uniformly a.e. dominating degree.
We provide two proofs of this result, although only one produces a c.e. degree. In Section 2 we use a priority argument to construct an incomplete c.e. uniformly a.e. dominating degree and in Section 4 we present a more flexible forcing construction of an incomplete uniformly a.e. dominating degree.
### 1.2. Domination and Reverse Mathematics
As observed by MR2078930, uniformly a.e. dominating degrees play a role in determining the reverse mathematical strength of the fact that the Lebesgue measure is regular. For an introduction to reverse mathematics, the reader is directed to Simpson:98.
Regularity means that for every measurable set $`P`$ there is a $`G_\delta `$ set $`QP`$ and an $`F_\sigma `$ set $`SP`$ such that $`\mu (S)=\mu (P)=\mu (Q)`$, where a $`G_\delta `$ set is the intersection of countably many open sets and an $`F_\sigma `$ set is the union of countably many closed sets. Hence the following principle is implied by the regularity of the Lebesgue measure.
###### $`G_\delta \text{-}\mathrm{๐ฑ๐ค๐ฆ}`$.
For every $`G_\delta `$ set $`Q2^\omega `$ there is an $`F_\sigma `$ set $`SQ`$ such that $`\mu (S)=\mu (Q)`$.
Recall that the $`G_\delta `$ sets are exactly those that are $`\mathrm{\Pi }_2^0`$ in a real parameter (that is, boldface $`๐ท_2^0`$), and the $`F_\sigma `$ sets are exactly the $`๐บ_2^0`$ sets. Hence we can consider $`G_\delta \text{-}\mathrm{๐ฑ๐ค๐ฆ}`$ as a statement of second order arithmetic. We will see that $`G_\delta \text{-}\mathrm{๐ฑ๐ค๐ฆ}`$, which appears to be a natural mathematical statement, does not fall in line with the commonly occurring systems of reverse mathematics. In particular, we examine the chain
$$\mathrm{๐ฑ๐ข๐ }_0\mathrm{๐ฃ๐ญ๐ฑ}_0\mathrm{๐ถ๐ถ๐ช๐ซ}_0\mathrm{๐ถ๐ช๐ซ}_0\mathrm{๐ ๐ข๐ }_0.$$
Here $`\mathrm{๐ฑ๐ข๐ }_0`$ is the standard base system that all of the other systems extend; $`\mathrm{๐ถ๐ช๐ซ}_0`$ is $`\mathrm{๐ฑ๐ข๐ }_0`$ plus weak Kรถnigโs lemma; and $`\mathrm{๐ ๐ข๐ }_0`$ is $`\mathrm{๐ฑ๐ข๐ }_0`$ plus the scheme of arithmetic comprehension. These systems are studied extensively in \[Simpson:98\]. The system $`\mathrm{๐ถ๐ถ๐ช๐ซ}_0`$ is somewhat less standard. It consists of $`\mathrm{๐ฑ๐ข๐ }_0`$ plus โweak weak Kรถnigโs lemma", which is introduced in YuSimpson:90. A large amount of basic measure theory can be proved in $`\mathrm{๐ถ๐ถ๐ช๐ซ}_0`$, so it is a natural system for us to be concerned with. The final system, $`\mathrm{๐ฃ๐ญ๐ฑ}_0`$, is less natural from a proof-theoretic standpoint but very natural for computability theorists. It is $`\mathrm{๐ฑ๐ข๐ }_0`$ plus the existence of a function that is diagonally non-recursive; see GiustoSimpson and Jockusch:NoFixedPoints.
Kurtzโs result that $`\mathrm{๐}^{}`$ is uniformly everywhere dominating essentially shows that $`G_\delta \text{-}\mathrm{๐ฑ๐ค๐ฆ}`$ follows from $`\mathrm{๐ ๐ข๐ }_0`$. This relies on the following:
###### Theorem 1.3 (Theorem 3.2 of MR2078930).
A Turing degree $`๐`$ is of uniformly a.e. dominating degree iff for every $`\mathrm{\Pi }_2^0`$ set $`Q2^\omega `$ there is a $`\mathrm{\Sigma }_2^0(๐)`$ set $`SQ`$ such that $`\mu (S)=\mu (Q)`$.
Dobrinen and Simpson conjectured that $`G_\delta \text{-}\mathrm{๐ฑ๐ค๐ฆ}`$ and $`\mathrm{๐ ๐ข๐ }_0`$ are equivalent over $`\mathrm{๐ฑ๐ข๐ }_0`$ (\[MR2078930, Conjecture 3.1\]). This is not true; in fact, there is an $`\omega `$-model of $`G_\delta \text{-}\mathrm{๐ฑ๐ค๐ฆ}`$ that omits $`\mathrm{๐}^{}`$ and hence is not a model of $`\mathrm{๐ ๐ข๐ }_0`$. This was discovered by B. Kjos-Hanssen after the circulation of the priority-method proof of Theorem 1.2. This proof appears to be too rigid to allow us to obtain a version with cone avoidance, but Kjos-Hanssen found a clever way to build the $`\omega `$-model without such a result. His construction is presented in Section 3.
The forcing construction is flexible enough to prove cone avoidance and more. We can thus improve Kjos-Hanssenโs result by showing that $`G_\delta \text{-}\mathrm{๐ฑ๐ค๐ฆ}`$ does not imply even systems much weaker than $`\mathrm{๐ ๐ข๐ }_0`$:
###### Theorem 1.4.
$`\mathrm{๐ฑ๐ข๐ }_0+G_\delta \text{-}\mathrm{๐ฑ๐ค๐ฆ}`$ does not imply $`\mathrm{๐ฃ๐ญ๐ฑ}_0`$.
But although $`G_\delta \text{-}\mathrm{๐ฑ๐ค๐ฆ}`$ seems to lack proof-theoretic strength, none of the traditional systems below $`\mathrm{๐ ๐ข๐ }_0`$ are strong enough to prove it:
###### Proposition 1.5 (Remark 3.5 of MR2078930).
$`\mathrm{๐ถ๐ช๐ซ}_0`$ does not imply $`G_\delta \text{-}\mathrm{๐ฑ๐ค๐ฆ}`$.
The proposition follows easily from the fact that there is an $`\omega `$-model of $`\mathrm{๐ถ๐ช๐ซ}_0`$ that consists of low sets; by formalizing Theorem 1.3, every $`\omega `$-model of $`G_\delta \text{-}\mathrm{๐ฑ๐ค๐ฆ}`$ must include uniformly a.e. dominating degrees, which by Martinโs result are high.
Furthermore, $`G_\delta \text{-}\mathrm{๐ฑ๐ค๐ฆ}`$ seems to be โorthogonal" to the traditional systems in that its strength is insufficient to lift one such system to the system above it:
###### Theorem 1.6.
$`\mathrm{๐ถ๐ช๐ซ}_0+G_\delta \text{-}\mathrm{๐ฑ๐ค๐ฆ}`$ does not imply $`\mathrm{๐ ๐ข๐ }_0`$; $`\mathrm{๐ถ๐ถ๐ช๐ซ}_0+G_\delta \text{-}\mathrm{๐ฑ๐ค๐ฆ}`$ does not imply $`\mathrm{๐ถ๐ช๐ซ}_0`$.
It remains open whether $`\mathrm{๐ฃ๐ญ๐ฑ}_0+G_\delta \text{-}\mathrm{๐ฑ๐ค๐ฆ}`$ implies $`\mathrm{๐ถ๐ถ๐ช๐ซ}_0`$.
### 1.3. Notation, conventions and other technicalities
Our computability theoretic notation is not always classical or consistent, but hopefully completely understandable. Thus, $`\phi _e_{e\omega }`$ is an effective list of all Turing functionals with oracle, and we write $`\phi _e^f(x),\phi _{e,s}^f(x)`$, etc. This notation will be used when we try to diagonalize against some oracle $`f`$ (so $`\phi _e:\omega ^\omega \omega ^\omega `$). On the other hand, for domination purposes, we write Turing functionals as $`\mathrm{\Phi }(Z;x)`$ and $`\mathrm{\Phi }(Z;x)[s]`$. In fact, we only need to consider a single $`\mathrm{\Phi }`$:
###### Lemma 1.7.
There is a partial computable functional $`\mathrm{\Phi }:2^\omega \omega ^\omega `$ such that if
$$\mu \{Z2^\omega :\text{if }\mathrm{\Phi }(Z)\text{ is total, then }\mathrm{\Phi }(Z)^{}f\}=1,$$
then $`f`$ is uniformly a.e. dominating.
###### Proof.
Let $`\mathrm{\Psi }_i_{i\omega }`$ be an effective list of partial computable functionals $`2^\omega \omega ^\omega `$ and define $`\mathrm{\Phi }(0^i1Z)=\mathrm{\Psi }_i(Z)`$. โ
We assume that $`\mathrm{\Phi }`$ has the following (standard) properties (for every $`s,n\omega `$ and $`Z2^\omega `$):
1. $`\mathrm{\Phi }(Z;n)[s]`$ implies $`\mathrm{\Phi }(Z;n)[s]s`$.
2. $`\mathrm{\Phi }(Z;n)[s]`$ implies $`(m<n)\mathrm{\Phi }(Z;m)[s]`$.
We let $`dom\mathrm{\Phi }`$ be the collection of $`Z`$ such that $`\mathrm{\Phi }(Z)`$ is total. For $`n\omega `$, we let $`D_n=\{Z2^\omega :\mathrm{\Phi }(Z;n)\}`$. For a stage $`s\omega `$, $`D_n[s]`$ is given the obvious meaning. For $`g\omega ^\omega `$, let
$$D_{[n,m)}[g]=\{Z2^\omega :(k[n,m))\mathrm{\Phi }(Z;k)[g(k)]\},$$
(including the case where $`m=\mathrm{}`$). It follows from condition (1) that if $`ZD_{[n,m)}[g]`$, then $`g`$ majorizes $`\mathrm{\Phi }(Z)`$ on the interval $`[n,m)`$.
## 2. A proof of Theorem 1.2 via a priority construction
In this section we prove Theorem 1.2. We build $`f:\omega \omega `$ by giving a computable sequence of approximations $`f_s_{s\omega }`$. Assuming the limit exists, $`f=limf_s`$ is $`\mathrm{\Delta }_2^0`$. To ensure that $`f`$ has c.e. degree, it is enough to require that $`f`$ is approximated from below. Formally, $`(n)(s)f_s(n)f_{s+1}(n)`$. This means that $`W=\{n,m:f(n)m\}`$ is a c.e. set; it is clear that $`f_TW`$.
To ensure that $`f`$ is incomplete we will enumerate a c.e. set $`B`$ and meet the requirement
$`R_e:\phi _e^fB`$,
for each $`e\omega `$. These requirements will be handled by incompleteness strategies. The same strategies are responsible for assigning values to $`f`$, which essentially means that they must make $`f`$ large enough to be uniformly a.e. dominating. This can be accomplished if they are supplied with appropriate approximations to the measure of $`dom\mathrm{\Phi }`$. These approximations are given by measure guessing strategies.
We describe the incompleteness and measure guessing strategies first, in relative isolation. Then we explain the priority tree and the full construction.
### 2.1. Incompleteness Strategy
Let $`\sigma `$ be an agent assigned the goal of ensuring that $`\phi _e^fB`$, for some index $`e=e(\sigma )`$. When $`\sigma `$ is initialized, it chooses a *follower* $`x=x(\sigma )`$ that has not been used before in the construction. A typical incompleteness strategy would wait for a computation $`\phi _e^f(x)=0`$, preserve $`f`$ on the use of this computation, and enumerate $`x`$ into $`B`$. The main difference is that our incompleteness strategy will be proactive: it is permitted to change the values of $`f`$ to make $`\phi _e^f(x)=0`$. Indeed, only the incompleteness agents change the values of $`f`$ at all, so they are not only permitted to make these changes, it is crucial that they do so.
Three restrictions are placed on $`\sigma `$โs ability to change the values of $`f`$. First, as already mentioned, it cannot decrease the current values of $`f`$. Second, higher priority agents (who wish to preserve diagonalizing computations) impose restraint $`N=N(\sigma )`$; $`\sigma `$ is not allowed to change $`fN`$. The third restriction (which ensures that eventually $`f`$ will be dominating) involves a rational parameter $`\epsilon =\epsilon (\sigma )`$. For $`\sigma `$ to permanently protect a computation $`\phi _e^f(x)=0`$ with use $`r`$, it must be the case that
($`\mathrm{}`$)
$$\mu \left(dom\mathrm{\Phi }D_{[N,r)}[f]\right)\epsilon .$$
In other words, $`\sigma `$โs action (in protecting $`fr`$) prevents $`f`$ from majorizing $`\mathrm{\Phi }(Z)`$ above $`N`$ for no more than $`\epsilon `$ of all $`Z2^\omega `$. This is the restriction that forces $`\sigma `$ to increase the values of $`f`$.
The first two restrictions place no significant burden on $`\sigma `$, but the third is more demanding. In fact, $`\sigma `$ cannot hope to meet the third restriction without help because it does not know what $`dom\mathrm{\Phi }`$ is. To approximate it, we supply $`\sigma `$ with two useful pieces of information: a rational $`q=q(\sigma )`$ and a natural number $`M=M(\sigma )`$ such that:
1. $`q\mu (dom\mathrm{\Phi })`$.
2. $`\mu (D_M)q+\epsilon /2`$.
In the full construction, these parameters are provided by a measure guessing agent. If $`\sigma `$ is on the true path, then the values of $`q`$ and $`M`$ that are supplied to $`\sigma `$ will meet conditions (1) and (2).
We are now ready to describe the behavior of $`\sigma `$. The possible states of $`\sigma `$ are active, meaning that it is currently imposing restraint to protect a computation $`\phi _e^f(x)`$, and passive. When $`\sigma `$ is initialized, it is passive and it has restraint $`r(\sigma )=0`$. If $`\sigma `$ ever becomes active, it will remain so unless it is reset. This happens if the execution ever moves left of $`\sigma `$, or if condition (2) proves to be false for either $`\sigma `$ or a higher priority active agent. The details of the full construction are below.
Say that $`\sigma `$ is visited at stage $`s\omega `$. If either $`\sigma `$ or a higher priority agent for $`R_e`$ is currently active, then there is nothing to do. Otherwise, $`\sigma `$ searches for a string $`gs^{<s}`$ that has the following (computable) properties:
1. $`gf_sN`$;
2. $`(n[N,|g|))f_s(n)g(n)`$;
3. $`\mu (D_{[N,|g|)}[g])>q\epsilon /2`$; and
4. $`\phi _{e,s}^g(x)=0`$.
If there is such a string $`g`$, then $`\sigma `$ lets $`f_{s+1}g`$ and $`r(\sigma )=|g|`$. It enumerates $`x`$ into $`B`$ and declares itself active. If there is no such $`g`$, then $`\sigma `$ does nothing and remains passive.
This completes the description of the incompleteness strategy. We prove below that if $`\sigma `$ is on the true path and it ever becomes satisfied, then ($`\mathrm{}`$2.1) holds. Because agents that are not on the true path might also attempt to protect computations, what we actually prove is stronger: if an agent ever becomes active (hence is imposing restraint), either ($`\mathrm{}`$2.1) holds or the agent is eventually reinitialized (so that its restraint is removed).
###### Remark 2.1.
Unlike many tree constructions, it is important that at most one node on each level (i.e. at most one node per requirement) imposes restraint. Say a node at level $`e`$ ensures that $`f`$ dominates except for a set of size at most $`\epsilon _e`$. We will argue that $`f`$ dominates almost everywhere, using the fact that $`lim_e\mathrm{}_{e^{}>e}\epsilon _e^{}=0`$. If several nodes on the same level $`e`$ were to impose restraint, then $`\epsilon _e`$ must be counted more than once, making the calculation incorrect. This is why we stipulated that if $`\sigma `$ is visited at some stage $`s`$ and if at the same stage, some $`\sigma ^{}<_L\sigma `$ on the same level is active, then $`\sigma `$ does not act. Of course, we are making use of the fact that $`\sigma ^{}`$โs success is also $`\sigma `$โs.
### 2.2. Measure Guessing Strategy
Measure guessing agents change neither $`f`$ nor $`B`$ and they impose no restraint on other agents. Their only function is to provide the values of $`q`$ and $`M`$ to the incompleteness agents at the next higher level. A measure guessing agent $`\tau `$ is initialized with a rational parameter $`\delta =\delta (\tau )`$. Its primary job is to find a rational $`q`$ that approximates the measure of $`dom\mathrm{\Phi }`$ from below to within $`\delta `$. This is done as follows. Divide the interval $`[0,1]`$ into subintervals of length $`\delta `$. When $`\tau `$ is visited at stage $`s`$, it compares, for each $`ns`$, the measure of $`D_n[s]`$ with that of $`D_n[t]`$, where $`t`$ was the previous stage at which $`\tau `$ was visited. If the measure of some $`D_n`$ has crossed the threshold from one subinterval $`I^{}`$ to one on its right $`I`$, then (for the least such $`n`$) $`\tau `$ guesses that $`q=\mathrm{min}I`$ approximates the measure of $`dom\mathrm{\Phi }`$. Assume that $`\tau `$ is visited infinitely often and $`\mathrm{min}I`$ is the largest approximation guessed infinitely often. Then $`\mu (D_n)\mathrm{min}I`$ for all $`n\omega `$ and $`\mu (D_n)>\mathrm{max}I`$ for finitely many $`n`$. Therefore, $`\mu (dom\mathrm{\Phi })I`$.
We give the details. Let $`d=1/\delta `$. The outcomes of $`\tau `$ will be of the form $`q,M\times \omega `$, where $`q\{0,\delta ,2\delta ,\mathrm{},d\delta \}`$. When $`\tau `$ is first initialized, its outcome is $`0,0`$. Say that $`\tau `$ is visited at stage $`s\omega `$ and that the previous visit occurred at stage $`t<s`$. To provide a guess, $`\tau `$ looks for $`ns`$ and $`bd`$ such that $`\mu (D_n[t])<b\delta `$ but $`\mu (D_n[s])b\delta `$. For the greatest such $`b`$ (or equivalently, the $`b`$ corresponding to the least such $`n`$), $`\tau `$ lets $`q=b\delta `$. Otherwise, $`\tau `$ lets $`q=0`$. Finally, $`\tau `$ takes the least $`M`$ such that $`\mu (D_M[s])<q+\delta `$. Because $`\mu (D_n[s])`$ is monotonically decreasing as a function of $`n`$, for all $`nM`$ we also have $`\mu (D_n[s])<q+\delta `$. The outcome of $`\tau `$ at stage $`s`$ is $`q,M`$.
###### Remark 2.2.
Suppose that $`\tau `$ has outcome $`q,M_0`$ at stage $`s_0`$ and outcome $`q,M_1`$ at $`s_1>s_0`$. Further suppose that whenever $`\tau `$ is visited at a stage $`t`$ between $`s_0`$ and $`s_1`$, its outcome at $`t`$ is of the form $`q^{},M^{}`$ with $`q^{}q`$. Then $`M_1=M_0`$.
### 2.3. The Priority Tree
As usual, agents are organized on a tree, with the children of an agent representing its potential outcomes. Write $`\alpha \beta `$ to mean that $`\beta `$ is a proper extension of $`\alpha `$. Each agent comes with a linear ordering $`<_L`$ on its children. We extend $`<_L`$ to other nodes as follows: say that $`\alpha `$ is *to the left* of $`\beta `$ and write $`\alpha <_L\beta `$ if there are $`\rho \alpha `$ and $`\nu \beta `$ such that $`\rho `$ and $`\nu `$ have the same parent and $`\rho <_L\nu `$. Write $`\alpha <\beta `$ if either $`\alpha \beta `$ or $`\alpha <_L\beta `$. This is the total ordering lexicographically induced on the tree by the ordering we impose on the children of agents. If $`\alpha <\beta `$, then we say that $`\alpha `$ *has higher priority* than $`\beta `$.
The even levels of the priority tree are devoted to measure guessing agents and the odd levels to incompleteness agents. A measure guessing agent $`\tau `$ at level $`2k`$ is supplied with the parameter $`\delta (\tau )=3^k/2`$. As described above, its outcomes have the form $`q,M\times \omega `$, where $`q`$ is restricted to rationals of the form $`b\delta (\tau )`$. The outcomes are ordered first by $`q`$ and then by $`M`$, with larger numbers *to the left of* smaller numbers.
An incompleteness agent $`\sigma =\tau {}_{}{}^{}q,M`$ at level $`2k+1`$ has parameters $`e(\sigma )=k`$ and $`\epsilon (\sigma )=3^k=2\delta (\tau )`$. We also obviously set $`q(\sigma )=q`$ and $`M(\sigma )=M`$. The two final parameters, the follower $`x(\sigma )`$ and the restraint $`N(\sigma )`$ imposed by stronger nodes, are determined when $`\sigma `$ is *initialized*. To initialize $`\sigma `$ at stage $`s\omega `$, set its state to passive, let the restraint $`\sigma `$ imposes $`r(\sigma )=0`$ and choose a follower $`x(\sigma )\omega `$ that has not yet been assigned in the construction. Furthermore, set
$$N(\sigma )=\mathrm{max}\{r(\sigma ^{}):\sigma ^{}<_L\sigma \text{ is active at stage }s\}.$$
The children of $`\sigma `$ are $`\sigma {}_{}{}^{}\text{active}<_L\sigma {}_{}{}^{}\text{passive}`$.
### 2.4. Full Construction
Let $`f_0(n)=0`$ for all $`n\omega `$. The construction proceeds in stages. The preliminary phase of stage $`s\omega `$ involves reevaluating, and possibly *resetting*, currently active incompleteness agents. Reset agents must be reinitialized the next time they are visited. Say that $`\sigma =\tau {}_{}{}^{}q,M`$ is active at stage $`s`$. If $`\mu (D_M[s])>q+\epsilon (\sigma )/2`$, then $`\sigma `$ acted based on a false assumption and it could be the case that $`\sigma `$ is forcing $`fr(\sigma )`$ to remain prohibitively small. Therefore, we reset $`\sigma `$. We also reset all previously initialized incompleteness agents of lower priority than $`\sigma `$ (to allow them to recompute their restraints the next time they are visited).
###### Remark 2.3.
Suppose that $`\tau `$ lies on the true path and that $`\sigma =\tau {}_{}{}^{}q,M`$ is active at stage $`s`$. Further suppose that $`\tau `$โs guess is found to be incorrect at $`s`$ (in other words, $`\mu (D_M[s])>q+\delta (\tau )`$). Then the next time that $`\tau `$ is accessible, its new outcome lies to the left of $`\sigma `$ and so $`\sigma `$ is reset. It would seem that this mechanism would suffice and that explicit resetting is unnecessary. However, unlike many tree constructions, we need to be concerned with the restraint imposed by nodes that lie to the left of the true path. Such unwarranted restraint may prevent $`f`$ from sufficiently dominating, and so needs to be reset when found incorrect.
During the main phase of stage $`s`$, we execute the strategies of finitely many agents on the priority tree, following a path of length at most $`s`$. This is done in substages $`ts`$. We begin at substage $`t=0`$ by visiting the root node $`\alpha _0=\lambda `$. Say that we are visiting an agent $`\alpha _t`$ at substage $`t`$. First, reset any incompleteness agents $`\sigma `$ such that $`\alpha _t<_L\sigma `$. (Note that if $`\sigma `$ is reset and $`\sigma <\sigma ^{}`$, then $`\alpha _t<_L\sigma ^{}`$, so $`\sigma ^{}`$ is also reset.)
*Case $`1:`$ $`\alpha _t`$ is a measure guessing agent.* If the outcome of $`\alpha _t`$ at stage $`s`$ is $`q,M`$, then let $`\alpha _{t+1}=\alpha _t{}_{}{}^{}q,M`$ and end the substage.
*Case $`2:`$ $`\alpha _t`$ is an incompleteness agent.* If $`\alpha _t`$ has never been visited before or has been reset since the last time it was visited, then it is initialized. If $`\alpha _t`$ is currently active, then end the substage and set $`\alpha _{t+1}=\alpha _t{}_{}{}^{}\mathrm{๐บ๐ผ๐๐๐๐พ}`$. Similarly, if there is a higher priority agent for $`R_e`$ that is active at stage $`s`$, then set $`\alpha _{t+1}=\alpha _t{}_{}{}^{}\mathrm{๐๐บ๐๐๐๐๐พ}`$ and end the substage. Otherwise, execute the incompleteness strategy for $`\alpha _t`$ at stage $`s`$. If $`\alpha _t`$ becomes active (so that changes are made to $`f`$ and $`B`$), then end stage $`s`$ entirely. Otherwise, let $`\alpha _{t+1}=\alpha _t{}_{}{}^{}\mathrm{๐๐บ๐๐๐๐๐พ}`$ and end the substage.
This continues until substage $`t=s`$ is completed or until stage $`s`$ is explicitly ended because an incompleteness agent becomes active. Finally, for any $`x<domf_s`$, if not expressly altered by us during the stage, we let $`f_{s+1}(x)=f_s(x)`$. This completes the construction.
### 2.5. Verification
Inductively define the *true path* to be the leftmost path visited infinitely often. In particular:
* The root node $`\lambda `$ is on the true path.
* If $`\rho `$ is on the true path and $`\nu `$ is the leftmost child of $`\rho `$ that is visited infinitely often (if such exists), then $`\nu `$ is on the true path.
It is clear that if $`\rho `$ is on the true path, then there is a stage $`s\omega `$ after which no agent left of $`\rho `$ is ever visited.
###### Claim 2.4.
If $`\sigma `$ is an incompleteness agent on the true path, then there is a stage $`s\omega `$ at which $`\sigma `$ is initialized and after which it will never be reset.
###### Proof.
Take a stage $`t\omega `$ large enough that no agent left of $`\sigma `$ will ever again be visited. By induction, we may also assume that $`t`$ is large enough that the agents $`\sigma ^{}\sigma `$ have all been initialized for the final time (and will never be reset). None of these $`\sigma ^{}`$ can become active after stage $`t`$, or else the execution would move left of $`\sigma `$.
Although no $`\sigma ^{}<_L\sigma `$ can become active after stage $`t`$, they can be reset in the preliminary phase of the construction and this will reset $`\sigma `$. But only active agents become reset and only finitely many $`\sigma ^{}<_L\sigma `$ are active at stage $`t`$. Therefore, there is a stage $`t^{}t`$ after which no agents left of $`\sigma `$ are ever reset.
This leaves only one way that $`\sigma =\tau {}_{}{}^{}q,M`$ can be reset at any stage $`t^{\prime \prime }t^{}`$: if $`\sigma `$ is active at stage $`t^{\prime \prime }`$ and $`\mu (D_M[t^{\prime \prime }])>q+\epsilon (\sigma )/2`$. But if this is the case, then $`q,M`$ cannot be the outcome of $`\tau `$ after stage $`t^{\prime \prime }`$, contradicting the fact that $`\sigma `$ is on the true path. Therefore, $`\sigma `$ is never reset after stage $`t^{}`$. But $`\sigma `$ is visited infinitely often, so there is a stage $`s\omega `$ at which $`\sigma `$ is initialized and after which it will never be reset. โ
###### Claim 2.5.
The true path is infinite.
###### Proof.
We prove that there is no last node on the true path. First, consider an incompleteness agent $`\sigma `$ on the true path. By Claim 2.4, there is a last stage $`t`$ at which $`\sigma `$ is initialized. After stage $`t`$, $`\sigma `$ may become active at most once, so one of the outcomes of $`\sigma `$ is eventually permanent.
Now consider a measure guessing agent $`\tau `$ on the true path. The first coordinate of the outcome of $`\tau `$ is taken from the finite set $`Q=\{b\delta (\tau ):\mathrm{\hspace{0.17em}0}b1/\delta (\tau )\}`$. Let $`q`$ be the greatest element of $`Q`$ that occurs as the first coordinate of the outcome infinitely often. Assume that no greater first coordinate occurs after stage $`s\omega `$. Let $`q,M`$ be the outcome of $`\tau `$ at some stage $`s`$. By Remark 2.2, if $`q^{},M^{}`$ is the outcome of $`\tau `$ at some other stage $`s`$, then either $`q^{}<q`$ or $`q^{}=q`$ and $`M^{}=M`$. Therefore, either $`q,M<_Lq^{},M^{}`$ or $`q,M=q^{},M^{}`$, and the second case occurs infinitely often. Hence $`\tau {}_{}{}^{}q,M`$ is on the true path. โ
###### Remark 2.6.
Let $`\tau `$ be a measure guessing node, and suppose that $`\tau `$ and $`\tau {}_{}{}^{}q,M`$ are on the true path. Then $`\mu (D_n)q`$ for all $`n\omega `$ and $`\mu (D_M)q+\delta (\tau )`$. Therefore, $`\mu (dom\mathrm{\Phi })[q,q+\delta (\tau )]`$.
We are primarily interested in the incompleteness agents that are eventually permanently active. Let the set of all such agents be $`G=\{\sigma _0,\sigma _1,\mathrm{}\}`$, with $`\sigma _0<\sigma _1<\mathrm{}`$. The fact that we can thus enumerate $`G`$ relies on the following:
###### Fact 2.7.
The collection of nodes that lie either on, or to the left of the true path that are ever visited has order type $`\omega `$ under $`<`$. This is because for each node $`\alpha `$ on the true path, only finitely many nodes to the left of $`\alpha `$ are ever visited.
###### Claim 2.8.
Assume that $`\sigma `$ is initialized at stage $`s\omega `$ and is never reset after stage $`s`$. Suppose that $`\sigma ^{}<\sigma `$. Then if $`\sigma ^{}`$ is active at $`s`$, it remains permanently so (hence $`\sigma ^{}G`$); otherwise, $`\sigma ^{}`$ never becomes active after $`s`$ (hence $`\sigma ^{}G`$).
###### Proof.
First assume that $`\sigma ^{}`$ is active at stage $`s`$. If $`\sigma ^{}`$ is ever reset, then every lower priority agent is reset, including $`\sigma `$. But this never happens, so $`\sigma ^{}G`$.
Now suppose that $`\sigma ^{}`$ is not active at stage $`s`$. It follows that $`\sigma ^{}{}_{}{}^{}\mathrm{๐๐บ๐๐๐๐๐พ}<\sigma `$ (as $`\sigma `$ is accessible at stage $`s`$). If $`\sigma ^{}`$ becomes active at some later stage, then $`\sigma ^{}{}_{}{}^{}\mathrm{๐บ๐ผ๐๐๐๐พ}`$ would be accessible. But this would reset $`\sigma `$ because $`\sigma ^{}{}_{}{}^{}\mathrm{๐บ๐ผ๐๐๐๐พ}`$ lies to the left of $`\sigma `$. โ
For all $`i\omega `$, let $`N_i`$ and $`r_i`$ denote the final values of $`N(\sigma _i)`$ and $`r(\sigma _i)`$, respectively.
###### Claim 2.9.
For all $`i\omega `$:
* Once $`\sigma _i`$ becomes permanently active, $`f`$ cannot change below $`r_i`$.
* $`N_{i+1}=r_i`$.
###### Proof.
(a) Assume that $`\sigma _i`$ is permanently active after stage $`s\omega `$. From $`s`$ onwards, $`\sigma _i`$ imposes restraint $`r_i`$ on weaker agents, so such agents do not change $`fr_i`$. Any action by a stronger agent is impossible after the last stage $`s_i`$ at which $`\sigma _i`$ is initialized, and $`s_i<s`$.
(b) At stage $`s_i`$, the agents $`\sigma <\sigma _i`$ that are active are exactly $`\sigma _0,\mathrm{},\sigma _{i1}`$, and their restraints have reached their final values. Thus $`\sigma _i`$ defines $`N_i=\mathrm{max}\{r_j:j<i\}`$ at stage $`s_i`$. When $`\sigma _i`$ later becomes active, it imposes a permanent restraint $`r_i`$, which is greater than $`N_i`$. It follows that $`r_0<r_1<\mathrm{}`$, and so $`N_{i+1}=r_i`$. โ
###### Claim 2.10.
$`G`$ is infinite.
###### Proof.
We can enumerate $`\phi _e_{e\omega }`$ in such a way that there are infinitely many $`e`$ such that for all $`t\omega `$, for all $`xt`$ and all $`g(t+1)^t`$ we have $`\phi _{e,t}^g(x)=0`$; we retroactively assume that we used such an enumeration. We will show that for each such $`e`$, $`G`$ contains an agent working for $`R_e`$.
Pick such an $`e`$ and let $`\sigma =\tau {}_{}{}^{}q,M`$ be the agent of length $`2e+1`$ on the true path. Assume that the final initialization of $`\sigma `$ occurs at stage $`s\omega `$.
*Case $`1:`$ An agent $`\sigma ^{}<_L\sigma `$ for $`R_e`$ is active at stage $`s`$.* If $`\sigma ^{}`$ is ever reset, then $`\sigma `$ would also be reset. This is impossible, so $`\sigma ^{}G`$.
*Case $`2:`$ No such $`\sigma ^{}`$ exists.* No $`\sigma ^{}<_L\sigma `$ becomes active after stage $`s`$, so as long as $`\sigma `$ remains passive, its full strategy will be executed every time it is visited. At stage $`s`$, a follower $`x`$ is chosen and the final restraint $`N`$ is determined. By Claim 2.9, $`fN`$ is fixed after stage $`s`$.
We know that $`q,M`$ is the correct outcome of $`\tau `$, so $`(n)\mu (D_n)q`$ (recall that $`D_n`$ is a decreasing sequence.) Let $`v=\mathrm{max}\{N,M\}`$. There is a $`t_0`$ such that $`\mu (D_v[t_0])>q\epsilon (\sigma )/2`$. For any string $`g\omega ^{v+1}`$ extending $`fN`$ such that $`g(n)t_0`$ for all $`n[N,v+1)`$, we have $`\mu (D_{[N,|g|)}[g])>q\epsilon (\sigma )/2`$.
Consider a stage $`t\mathrm{max}\{t_0,x,v+1\}`$ at which $`\sigma `$ is accessible. Let $`g=fN{}_{}{}^{}t_{}^{v+1N}`$. Of course $`f_t(n)t`$ for all $`n`$, so by the assumptions on $`e`$, $`\phi _{e,t}^g(x)=0`$. Thus $`g`$ satisfies all the conditions that make it eligible to be picked as a new initial segment of $`f`$. It follows that if $`\sigma `$ did not act before stage $`t`$, then it does so and becomes permanently active. โ
###### Claim 2.11.
$`f=lim_sf_s`$ exists.
###### Proof.
Combining Claims 2.9(b) and 2.10, the intervals $`\{[N_i,r_i)\}_{i\omega }`$ partition $`\omega `$. Furthermore, by Claim 2.9(a), $`f`$ is stable on $`[0,r_i)`$ once $`\sigma _i`$ becomes permanently active. Therefore, $`lim_sf_s(n)`$ converges for all $`n\omega `$. โ
###### Claim 2.12.
For all $`i`$, $`\mu \left(dom\mathrm{\Phi }D_{[N_i,r_i)}[f]\right)\epsilon (\sigma _i)`$.
###### Proof.
Assume for a contradiction that $`\mu \left(dom\mathrm{\Phi }D_{[N_i,r_i)}[f]\right)>\epsilon (\sigma _i)`$. Let $`\sigma _i=\tau {}_{}{}^{}q,M`$. Take $`s\omega `$ to be the stage at which $`\sigma _i`$ becomes permanently active and let $`g\omega ^{<\omega }`$ be the string that was used at that activation. So $`r_i=|g|`$ and $`gf`$. This implies that $`D_{[N_i,r_i)}[g]=D_{[N_i,r_i)}[f]`$. But of course, $`dom\mathrm{\Phi }D_M`$. Therefore, $`\mu \left(D_MD_{[N_i,r_i)}[g]\right)>\epsilon (\sigma _i)`$.
By the definition of the incompleteness strategy, $`\mu (D_{[N_i,r_i)}[g])>q\epsilon (\sigma _i)/2`$. Also $`r_i>M`$, so $`D_{[N_i,r_i)}[g]D_M`$. Together with the conclusion of the previous paragraph, we have $`\mu (D_M)>q+\epsilon (\sigma _i)/2`$. But then $`\mu (D_M[t])>q+\epsilon (\sigma _i)/2`$, for any sufficiently large $`t\omega `$. Therefore, $`\sigma _i`$ would be reset at the first phase of stage $`t`$, which is a contradiction. โ
###### Claim 2.13.
$`f`$ is uniformly a.e. dominating.
###### Proof.
Fix $`e\omega `$. The construction ensures that at most one incompleteness agent at each level can be active at a time; hence at most one can belong to $`G`$. Thus there is an $`i\omega `$ large enough that $`(ji)|\sigma _j|2e+1`$. Furthermore, $`_{ji}\epsilon (\sigma _j)_{ke}3^k=3^{e+1}/2`$ for this choice of $`i`$. By Claims 2.9(b) and 2.10, the intervals $`\{[N_j,r_j)\}_{ji}`$ partition $`[N_i,\mathrm{})`$. Therefore, if $`Z_{ji}D_{[N_j,r_j)}`$, then $`f`$ majorizes $`\mathrm{\Phi }(Z)`$ above $`N_i`$. By Claim 2.12,
$$\mu \left(dom\mathrm{\Phi }\underset{ji}{}D_{[N_j,r_j)}[f]\right)\underset{ji}{}\mu \left(dom\mathrm{\Phi }D_{[N_j,r_j)}[f]\right)\frac{3^{e+1}}{2}.$$
In other words, the set of $`Zdom\mathrm{\Phi }`$ such that $`f`$ fails to dominate $`\mathrm{\Phi }(Z)`$ has measure at most $`3^{e+1}/2`$. But $`e\omega `$ was arbitrary, so $`f`$ is uniformly a.e. dominating. โ
###### Claim 2.14.
$`f<_T\mathrm{๐}^{}`$.
###### Proof.
It is sufficient to prove that $`Bโฉฝฬธ_Tf`$. Fix an index $`e\omega `$.
*Case $`1:`$ There is an $`R_e`$ agent $`\sigma _iG`$.* Let $`s\omega `$ be the last stage at which $`\sigma _i`$ becomes active and let $`x_i=x(\sigma _i)[s]`$. By Claim 2.9(a), this is done via $`g=f_{s+1}r_i=fr_i`$. Because $`\sigma _i`$ is activated, we know that $`x_iB`$ and $`\phi _{e,s}^g(x_i)=0`$. Therefore, $`\phi _e^f(x_i)=0B(x_i)`$.
*Case $`2:`$ There is no agent for $`R_e`$ in $`G`$.* Let $`\sigma =\tau {}_{}{}^{}q,M`$ be the incompleteness agent of length $`2e+1`$ on the true path. Assume that $`\sigma `$ is initialized for the last time at stage $`s\omega `$. Let $`x=x(\sigma )[s]`$ and $`N=N(\sigma )[s]`$. Note that $`xB`$, because $`\sigma `$ does not become active after stage $`s`$ (else $`\sigma G`$, so we would be in Case 1). Assume, for a contradiction, that $`\phi _e^f(x)=0`$. Take $`g\omega ^{<\omega }`$ such that $`|g|>\mathrm{max}\{M,N\}`$, $`g`$ is an initial segment of $`f`$, and $`\phi _e^g(x)=0`$.
By Claim 2.8, $`\sigma ^{}<\sigma `$ is active at stage $`s`$ iff $`\sigma ^{}G`$. Choose $`i\omega `$ such that $`\sigma _{i1}<\sigma <\sigma _i`$. In particular, $`N=N_i`$. Since $`\sigma `$ is on the true path, we have $`\sigma \sigma _j`$ for all $`ji`$. This shows that $`(ji)|\sigma _j|>2e+1`$. Now take $`m\omega `$ large enough that $`r_m|g|`$. Then, $`_{j[i,m]}\epsilon (\sigma _j)<_{k>e}3^k=3^e/2`$. By the same argument as given in Claim 2.13, $`\mu \left(dom\mathrm{\Phi }D_{[N,r_m)}[f]\right)<3^e/2`$. Therefore, $`\mu \left(dom\mathrm{\Phi }D_{[N,|g|)}[g]\right)<3^e/2`$. We know that $`\mu (dom\mathrm{\Phi })q`$. This proves that $`\mu (D_{[N,|g|)}[g])>q3^e/2`$.
Let $`t>s`$ be a stage at which $`\sigma `$ is accessible that is large enough so that $`\phi _{e,t}^g(x)=0`$. There is nothing stopping $`\sigma `$ from acting at stage $`t`$, which is the desired contradiction. โ
## 3. Reverse mathematics I: avoiding cone avoidance
Although the above c.e. construction (Section 2) does not seem to generalize to yield a cone avoidance result, Kjos-Hanssen showed that it does have a reverse mathematical consequence.
###### Theorem 3.1 (Kjos-Hanssen).
There is an $`\omega `$-model of $`\mathrm{๐ฑ๐ข๐ }_0+G_\delta \text{-}\mathrm{๐ฑ๐ค๐ฆ}`$ that does not contain $`\mathrm{๐}^{}`$. Hence $`G_\delta \text{-}\mathrm{๐ฑ๐ค๐ฆ}`$ does not imply $`\mathrm{๐ ๐ข๐ }_0`$ over $`\mathrm{๐ฑ๐ข๐ }_0`$.
###### Proof.
We construct an ideal of Turing degrees that (as an $`\omega `$-model) satisfies $`G_\delta \text{-}\mathrm{๐ฑ๐ค๐ฆ}`$ but does not contain $`\mathrm{๐}^{}`$. The ideal is the downward closure of an increasing sequence $`๐_1<๐ก_1<๐_2<๐ก_2<\mathrm{}`$. We let $`๐_1=\mathrm{๐}`$ and let $`๐ก_1`$ be the c.e. degree given by Theorem 1.2. The degree $`๐ก_1`$ is high. In the structure $`๐[๐ก_1,๐ก_1^{}]`$ we can find some $`๐_2`$ that is $`\text{low}(๐ก_1)`$ and that joins $`\mathrm{๐}^{}`$ to $`๐ก_1^{}=\mathrm{๐}^{\prime \prime }`$ (PosnerRobinson). Now in the structure $`๐[๐_2,๐_2^{}]=๐[๐_2,\mathrm{๐}^{\prime \prime }]`$, a relativized version of Theorem 1.2 yields a degree $`๐ก_2<\mathrm{๐}^{\prime \prime }`$ that is uniformly almost everywhere dominating over $`๐_2`$. We cannot have $`๐ก_2\mathrm{๐}^{}`$ because $`๐ก_2๐_2`$ and $`๐ก_2<\mathrm{๐}^{\prime \prime }`$.
We now repeat. Again, using a relativized version of \[PosnerRobinson\], we get an $`๐_3๐[๐ก_2,\mathrm{๐}^{\prime \prime \prime }]`$ that is $`\text{low}(๐ก_2)`$ and joins $`\mathrm{๐}^{\prime \prime }`$ to $`\mathrm{๐}^{\prime \prime \prime }`$; and an $`๐ก_3๐[๐_3,\mathrm{๐}^{\prime \prime \prime })`$ that is uniformly a.e. dominating over $`๐_3`$. As before, $`๐ก_3`$ is not above $`\mathrm{๐}^{\prime \prime }`$. But as $`๐ก_3๐_2`$ and $`\mathrm{๐}^{}๐_2=\mathrm{๐}^{\prime \prime }`$ we cannot have $`๐ก_3\mathrm{๐}^{}`$. The process now repeats itself to get the rest of the sequence. โ
## 4. A proof of Theorem 1.2 via a forcing construction
In this section we introduce a forcing notion that produces a uniformly a.e. dominating function and that allows us to obtain cone avoidance and more.
### 4.1. The notion of forcing
We approximate a function $`f^G`$. A *condition* is a pair $`f,\epsilon `$ where $`f\omega ^{<\omega }`$ and $`\epsilon `$ is a positive rational. The idea is that $`๐ฉ=f,\epsilon `$ states that $`f`$ is an initial segment of $`f^G`$ and further $`๐ฉ`$ makes an *$`\epsilon `$-promise*: the collection of $`Zdom\mathrm{\Phi }`$ such that $`f^G`$ fails to majorize $`\mathrm{\Phi }(Z)`$ from $`|f|`$ onwards has size $`<\epsilon `$. Thus, an extension $`gf`$ *respects the $`\epsilon `$-promise* if
$$\mu \left(dom\mathrm{\Phi }D_{[|f|,|g|)}[g]\right)<\epsilon .$$
However, this is not a good definition of a partial ordering on the conditions; we can have $`g`$ keep the $`\epsilon `$-promise of $`f,\epsilon `$ and $`h`$ keep the $`\delta `$-promise of $`g,\delta `$ but fail to respect the $`\epsilon `$-promise of $`f,\epsilon `$. Thus, the relation would not be transitive. A simple modification ensures that every $`h`$ that keeps the $`\delta `$-promise of $`g,\delta `$ also keeps the $`\epsilon `$-promise of $`f,\epsilon `$. We say that a condition $`g,\delta `$ *extends* another condition $`f,\epsilon `$ if $`fg`$, $`\delta \epsilon `$ and further, if $`fg`$, then
$$\mu \left(dom\mathrm{\Phi }D_{[|f|,|g|)}[g]\right)+\delta <\epsilon .$$
###### Lemma 4.1.
The extension relation is transitive.
###### Proof.
Suppose that $`g,\delta `$ extends $`f,\epsilon `$ and is extended by $`h,\gamma `$; we show that $`h,\gamma `$ extends $`f,\epsilon `$. If either $`f=g`$ or $`g=h`$, then this is easy. Otherwise, the point is that
$$D_{[|f|,|h|)}[h]=D_{[|f|,|g|)}[g]D_{[|g|,|h|)}[h]$$
and so
$$\begin{array}{c}\mu \left(dom\mathrm{\Phi }D_{[|f|,|h|)}[h]\right)\mu \left(dom\mathrm{\Phi }D_{[|f|,|g|)}[g]\right)+\mu \left(dom\mathrm{\Phi }D_{[|g|,|h|)}[h]\right)\hfill \\ \hfill (\epsilon \delta )+(\delta \gamma )=\epsilon \gamma ,\end{array}$$
as required. โ
###### Notation.
We let $``$ be the collection of all conditions. For a condition $`๐ฉ=f,\epsilon `$ we write $`f^๐ฉ=f`$ and $`\epsilon ^๐ฉ=\epsilon `$. We also let $`n^๐ฉ=|f^๐ฉ|`$.
###### Lemma 4.2.
For all $`n<\omega `$, the set $`\{๐ฉ:n^๐ฉ>n\}`$ is dense in $``$.
###### Proof.
Let $`๐ฉ`$. Let $`n>n^๐ฉ`$. For large enough $`s`$,
$$\mu (D_nD_n[s])<\epsilon ^๐ฉ.$$
Now take $`g\omega ^n`$ extending $`f`$ such that $`D_n[s]D_{[n^๐ฉ,n)}[g]`$ (for example, by defining $`g(m)=s`$ for $`mn^๐ฉ`$). As $`dom\mathrm{\Phi }D_n`$, we get that $`\mu \left(dom\mathrm{\Phi }D_{[n^๐ฉ,n)}[g]\right)<\epsilon ^๐ฉ`$. We can then pick some small $`\delta `$ so that $`g,\delta `$ extends $`๐ฉ`$. โ
If $`G`$ is generic (from now, by the word โgeneric" we mean, โsufficiently generic for the given argument"), then we let
$$f^G=\underset{๐ฉG}{}f^๐ฉ.$$
The following is a corollary of Lemma 4.2:
###### Corollary 4.3.
If $`G`$ is generic, then $`f^G\omega ^\omega `$.
We now show that the $`\epsilon `$-promises are kept.
###### Lemma 4.4.
Let $`๐ฉ`$, and suppose that $`๐ฉG`$ and that $`G`$ is generic. Then
$$\mu \left(dom\mathrm{\Phi }D_{[n^๐ฉ,\omega )}\left[f^G\right]\right)\epsilon ^๐ฉ.$$
###### Proof.
The sequence $`D_{[n^๐ฉ,m)}[f^G]_{m>n^๐ฉ}`$ decreases with $`m`$ and
$$D_{[n^๐ฉ,\omega )}\left[f^G\right]=\underset{m>n^๐ฉ}{}D_{[n^๐ฉ,m)}\left[f^G\right].$$
So it is enough to prove that $`\mu \left(dom\mathrm{\Phi }D_{[n^๐ฉ,m)}\left[f^G\right]\right)\epsilon ^๐ฉ`$, for all $`m>n^๐ฉ`$. For any $`m`$, there is a $`๐ชG`$ extending $`๐ฉ`$ such that $`n^๐ชm`$. By the definition of our partial ordering,
$$\mu \left(dom\mathrm{\Phi }D_{[n^๐ฉ,n^๐ช)}\left[f^๐ช\right]\right)<\epsilon ^๐ฉ.$$
But $`D_{[n^๐ฉ,n^๐ช)}\left[f^๐ช\right]D_{[n^๐ฉ,m)}\left[f^G\right]`$ because $`f^๐ชf^G`$, which completes the proof. โ
The following is immediate.
###### Lemma 4.5.
For all $`\epsilon >0`$, the set $`\{๐ฉ:\epsilon ^๐ฉ<\epsilon \}`$ is dense in $``$. โ
As a corollary,
###### Corollary 4.6.
If $`G`$ is generic, then $`f^G`$ is uniformly almost everywhere dominating.
### 4.2. Cone avoidance, etc.
We show that if $`G`$ is generic, then indeed $`f^G`$ has no special properties beyond domination. The following is the crucial technical lemma. Consider the proof that if $`g`$ is Cohen generic over $`A`$ and $`A`$ is not computable, then $`g`$ does not compute $`A`$. If some condition $`\tau 2^{<\omega }`$ forces that $`\phi _e^g=A`$ (and in particular is total), then $`A=_{\sigma \tau }\phi _e^\sigma `$ is computable because the collection of extensions of $`\tau `$ is computable. We would like to do the same, but our partial ordering is not computable. This difficulty is overcome as follows: given $`๐ฉ`$, we can make a promise $`\epsilon ^{}`$ much tighter than $`\epsilon ^๐ฉ`$ and find a rational $`q`$ sufficiently close to $`dom\mathrm{\Phi }`$ such that every sufficiently long string $`gf^๐ฉ`$ respecting the $`\epsilon ^{}`$-promise satisfies $`\mu (D_{[n^๐ฉ,|g|)}[g])>q`$ and every string satisfying the latter (computable) condition respects the $`\epsilon ^๐ฉ`$-promise. We can now imitate the diagonalization argument (and more): if $`๐ฉ`$ forces that $`\phi _e^{f^G}=A`$, then we compute $`A`$ by examining $`\phi _e^g`$ for strings $`g`$ satisfying the middle condition above. We argue that this must give us all of $`A`$, for otherwise we could extended $`๐ฉ`$ to keep the $`\epsilon ^{}`$-promise and avoid $`\phi _e^{f^G}=A`$.
###### Lemma 4.7.
Let $`๐ฉ`$. Then there is a c.e. set
$$S\{f^๐ช:๐ช๐ฉ\}$$
and a $`๐ฉ^{}๐ฉ`$ such that $`\{๐ช๐ฉ^{}:f^๐ชS\}`$ is dense below $`๐ฉ^{}`$.
###### Proof.
Find some $`n>n^๐ฉ`$ such that $`\mu \left(D_ndom\mathrm{\Phi }\right)<\epsilon ^๐ฉ/2`$; also find a rational $`q<\mu (dom\mathrm{\Phi })`$ such that $`\mu (dom\mathrm{\Phi })q<\epsilon ^๐ฉ/2`$. Let
$$S=\{g\omega ^{<\omega }:gf^๐ฉ\text{}|g|>n\text{, and }\mu (D_{[n^๐ฉ,|g|)}[g])>q\}$$
It is clear that $`S`$ is c.e. Let $`gS`$; we show that for some $`๐ช๐ฉ`$ we have $`f^๐ช=g`$. We have $`\mu (D_{[n^๐ฉ,|g|)}[g])>\mu (dom\mathrm{\Phi })\epsilon ^๐ฉ/2`$ and $`\mu (D_{|g|})<\mu (dom\mathrm{\Phi })+\epsilon ^๐ฉ/2`$; together we get $`\mu \left(D_{|g|}D_{[n^๐ฉ,|g|)}[g]\right)<\epsilon ^๐ฉ`$. Of course, $`dom\mathrm{\Phi }D_{|g|}`$ and so $`\mu \left(dom\mathrm{\Phi }D_{[n^๐ฉ,|g|)}[g]\right)<\epsilon ^๐ฉ`$.
Next, let $`๐ฉ^{}=f^๐ฉ,\delta `$ where $`\delta <\epsilon ^๐ฉ`$ (so $`๐ฉ^{}๐ฉ`$) and $`\delta <\mu (dom\mathrm{\Phi })q`$. Suppose that $`๐ช๐ฉ^{}`$ and $`n^๐ช>n`$. Then from
$$\mu \left(dom\mathrm{\Phi }D_{[n^๐ฉ,n^๐ช)}\left[f^๐ช\right]\right)<\delta $$
we can conclude that $`\mu (D_{[n^๐ฉ,n^๐ช)}[f^๐ช])>q`$, so $`f^๐ชS`$. โ
###### Lemma 4.8.
If $`A`$ is noncomputable and $`G`$ is generic over $`A`$, then $`f^Gโฉพฬธ_TA`$.
###### Proof.
Let $`\mathrm{\Psi }:\omega ^\omega 2^\omega `$ be a Turing functional. We show that the union of
$`E_0`$ $`=\{๐ฉ:\mathrm{\Psi }(f^๐ฉ)A\}\text{and}`$
$`E_1`$ $`=\{๐ฉ:(x)(๐ช๐ฉ)\mathrm{\Psi }(f^๐ช,x)\}`$
is dense in $``$. Of course if $`G(E_0E_1)\mathrm{}`$, then $`\mathrm{\Psi }(f^G)A`$.
Let $`๐ฉ`$, and take $`S`$ and $`๐ฉ^{}`$ given by Lemma 4.7. If there are $`g,g^{}S`$ such that $`\mathrm{\Psi }(g)\mathrm{\Psi }(g^{})`$, then one of them is incompatible with $`A`$, so $`๐ฉ`$ has an extension in $`E_0`$. If $`_{gS}\mathrm{\Psi }(g)`$ is total, then it is computable, hence different from $`A`$. Again, $`๐ฉ`$ has an extension in $`E_0`$.
Otherwise, for some $`x\omega `$, we have $`\mathrm{\Psi }(g,x)`$ for all $`gS`$. This implies that $`๐ฉ^{}E_1`$: for all $`๐ช๐ฉ^{}`$, $`f^๐ช`$ has an extension in $`S`$, and so $`\mathrm{\Psi }(f^๐ช,x)`$. โ
###### Lemma 4.9.
If $`G`$ is generic, then $`f^G`$ does not have PA-degree.
###### Proof.
Let $`\psi :\omega 2`$ be a partial computable function that has no total computable extension. We show that $`f^G`$ does not compute a 0-1 valued total extension of $`\psi `$.
Let $`\mathrm{\Theta }:\omega ^\omega 2^\omega `$ be a Turing functional. We show that the union of
$`E_0`$ $`=\{๐ฉ:(xdom\psi )\mathrm{\Theta }(f^๐ฉ,x)\psi (x)\}\text{and}`$
$`E_1`$ $`=\{๐ฉ:(x)(๐ช๐ฉ)\mathrm{\Theta }(f^๐ช,x)\}`$
is dense in $``$. Of course if $`G(E_0E_1)\mathrm{}`$, then $`\mathrm{\Theta }(f^G)`$ is not a total extension of $`\psi `$.
Let $`๐ฉ`$; take $`S`$ and $`๐ฉ^{}`$ given by Lemma 4.7. If there is a $`gS`$ such that $`\mathrm{\Theta }(g)\psi `$, then $`๐ฉ`$ has an extension in $`E_0`$. If for some $`x`$, $`\mathrm{\Theta }(g,x)`$ for all $`gS`$, then $`๐ฉ^{}E_1`$.
One of the above must be the case; otherwise, we could compute a completion of $`\psi `$ as follows: for each $`x`$, search for a $`gS`$ such that $`\mathrm{\Theta }(g,x)`$. For the first such $`g`$ found, let $`h(x)=\mathrm{\Theta }(g,x)`$. Then $`h`$ is computable, and must extend $`\psi `$. โ
In fact, the same proof gives us somewhat more:
###### Lemma 4.10.
If $`G`$ is generic, then $`f^G`$ does not have DNR-degree.
###### Proof.
Let $`\phi _e_{e\omega }`$ be an enumeration of all partial computable functions from $`\omega `$ to $`\omega `$.
Let $`\mathrm{\Psi }:\omega ^\omega \omega ^\omega `$ be a Turing functional. We show that the union of
$`E_0`$ $`=\{๐ฉ:(e)\mathrm{\Psi }(f^๐ฉ,e)=\phi _e(e)\}\text{and}`$
$`E_1`$ $`=\{๐ฉ:(x)(๐ช๐ฉ)\mathrm{\Psi }(f^๐ช,x)\}`$
is dense in $``$. Of course if $`G(E_0E_1)\mathrm{}`$, then $`\mathrm{\Psi }(f^G)`$ is not DNR.
Let $`๐ฉ`$, and take $`S`$ and $`๐ฉ^{}`$ given by Lemma 4.7. If there is a $`gS`$ and $`e\omega `$ such that $`\mathrm{\Psi }(g,e)=\phi _e(e)`$, then $`๐ฉ`$ has an extension in $`E_0`$. If there is an $`x`$ such that $`\mathrm{\Psi }(g,x)`$ for all $`gS`$, then $`๐ฉ^{}E_1`$.
Otherwise, define a total function $`h:\omega \omega `$ as follows: for each $`x`$, search for a $`gS`$ such that $`\mathrm{\Psi }(g,x)`$. Let $`h(x)=\mathrm{\Psi }(g,x)`$ for the first such $`g`$ discovered. Then $`h`$ is computable and DNR, which is impossible. โ
### 4.3. Relativization
Let $`B\omega `$. All the results of this section relativize to working above $`B`$. Namely, we can define a notion of forcing $`_B`$; all is exactly as above, except that instead of $`D_n`$ and $`dom\mathrm{\Phi }`$ we use $`\{Z:\mathrm{\Phi }(BZ,n)\}`$ and $`\{Z:\mathrm{\Phi }(BZ)\text{ is total}\}`$. With exactly the same proofs, we see that a generic yields a function $`f^G`$ that is uniformly almost everywhere dominating over $`B`$. Lemma 4.7 now becomes the following:
###### Lemma 4.11.
Let $`๐ฉ_B`$. Then there is a set $`S`$, c.e. in $`B`$, such that $`S\{f^๐ช:๐ช๐ฉ\}`$ and some $`๐ฉ^{}๐ฉ`$ such that $`\{๐ช๐ฉ^{}:f^๐ชS\}`$ is dense below $`๐ฉ^{}`$.
These are the analogous corollaries:
###### Lemma 4.12.
Suppose that $`Aโฉฝฬธ_TB`$ and that $`G_B`$ is generic over $`A`$. Then $`Bf^Gโฉพฬธ_TA`$.
###### Lemma 4.13.
Suppose that $`B`$ does not have PA-degree and that $`G_A`$ is generic over $`B`$. Then $`Bf^G`$ does not have PA-degree.
###### Lemma 4.14.
Suppose that $`B`$ is not DNR, and that $`G_B`$ is generic. Then $`Bf^G`$ is not DNR.
## 5. Reverse mathematics II
The above forcing argument directly yields the results concerning the proof-theoretic strength of $`G_\delta \text{-}\mathrm{๐ฑ๐ค๐ฆ}`$.
Recall from Simpson:98 that $`M2^\omega `$ is an $`\omega `$-model of $`\mathrm{๐ฑ๐ข๐ }_0`$ iff it forms an ideal in the Turing degrees, and it is an $`\omega `$-model of $`\mathrm{๐ถ๐ช๐ซ}_0`$ iff it is a *Scott system*: i.e., a Turing ideal such that for all $`AM`$, there is a $`BM`$ of PA-degree relative to $`A`$. Similarly, YuSimpson:90 proved that a Turing ideal $`M2^\omega `$ is an $`\omega `$-model of $`\mathrm{๐ถ๐ถ๐ช๐ซ}_0`$ iff for all $`AM`$, there is a $`BM`$ that is sufficiently random over $`A`$ (it is enough that $`B`$ is $`1`$-random relative to $`A`$ by a result of Kucera:85).
###### Proof of Theorem 1.4.
An ideal of Turing degrees that models $`G_\delta \text{-}\mathrm{๐ฑ๐ค๐ฆ}`$ but does not include any DNR degrees is easily built using Lemma 4.14. โ
###### Proof of Theorem 1.6.
For the first part, we can inductively construct an $`\omega `$-model of $`\mathrm{๐ถ๐ช๐ซ}_0+G_\delta \text{-}\mathrm{๐ฑ๐ค๐ฆ}`$ that avoids $`\mathrm{๐}^{}`$ by alternatively appealing to Lemma 4.12 and to the fact that a similar cone avoidance lemma holds for obtaining paths through trees, hence for PA-degrees (Jockusch.Soare:72\*1).
For the second part, a similar construction yields an $`\omega `$-model of $`\mathrm{๐ถ๐ถ๐ช๐ซ}_0+G_\delta \text{-}\mathrm{๐ฑ๐ค๐ฆ}`$ that does not satisfy $`\mathrm{๐ถ๐ช๐ซ}_0`$, this time using Lemma 4.13 and the following claim, which essentially appears in YuSimpson:90.
###### Claim 5.1.
Suppose that $`B`$ does not have PA-degree. If $`A`$ is sufficiently random over $`B`$, then $`AB`$ does not have PA-degree.
###### Proof.
This is from \[YuSimpson:90, Page 172\]. Let $`E`$ and $`F`$ be disjoint c.e. sets that cannot be separated by any set computable in $`B`$. By relativizing a result from JockuschSoare:DegreesOfTheories, the measure of
$$S=\{Z:(Y_TZB)EYFY=\mathrm{}\}$$
is zero. This is the collection of sets $`Z`$ such that $`ZB`$ computes a separator of $`E`$ and $`F`$. If $`A`$ is sufficiently random over $`B`$, then $`AS`$, meaning that it satisfies the claim. (In fact, since $`S`$ is a $`\mathrm{\Sigma }_3^0(B)`$-class, it suffices for $`A`$ to be (weakly) $`2`$-random relative to $`B`$ \[Kurtz:81\].) โ
|
warning/0506/math0506030.html
|
ar5iv
|
text
|
# On Graded Bialgebra Deformations
## 1. Introduction
The classification of finite-dimensional pointed Hopf algebras is a basic problem in the theory of Hopf algebras. It is well-known that any pointed Hopf algebra $`H`$ has a coradical filtration, with respect to which one associates a coradically-graded Hopf algebra $`\mathrm{gr}H`$. Following Andruskiewitsch and Schneider, the classification problem can be divided into two parts. One is the classification of all coradically-graded pointed Hopf algebras. The other is to find all possible pointed Hopf algebras $`H`$ with $`\mathrm{gr}H`$ isomorphic to a given coradically-graded pointed Hopf algebra. The second part is just the lifting method in and . One of our motivations is to relate the lifting method with certain bialgebra deformation theory.
The deformation theory for algebras is initiated by Gerstenhaber in , and its analogue for bialgebras appeared first in (also see and ). Inspired by the graded algebra deformation theory in and , we develop in this paper the theory of graded bialgebra deformations and their corresponding cohomology groups. Moreover this deformation theory can be used to explain Andruskiewitsch-Schneiderโs lifting method.
The paper is organized as follows. In section 2, first we recall the notion of liftings and introduce the graded bialgebra deformations, and we show that the lifting is just the same as the graded bialgebra deformation in the sense of Theorem 2.2. The graded-rigid bialgebras are also studied, see Corollary 2.3 and Corollary 2.4. In section 3, we introduce the notion of graded โhatโ bialgebra cohomology groups for graded bialgebras, which controls the graded bialgebra deformations, see Theorem 3.3.
## 2. Liftings and graded bialgebra deformations
We will work on a base field $`๐`$. All unadorned tensors are over $`๐`$. We refer the notion of graded bialgebras and filtered bialgebras to , the notion of graded linear maps to and .
### 2.1.
Let us recall Andruskiewitsch-Schneiderโs liftings method, for more details, see . Note that the lifting defined here is a slight generalization.
Throughout, $`B=_{i0}B_{(i)}`$ will be a graded bialgebra over $`๐`$, with identity element $`1_B`$, multiplication map $`m`$, counit $`\epsilon `$, and comultiplication $`\mathrm{\Delta }`$. Then $`B`$ has a natural bialgebra filtration
$`B_0B_1\mathrm{}B_i\mathrm{},`$
where $`B_i=_{ji}B_{(j)}`$ for any $`n0`$.
A *lifting* of the graded bialgebra $`B`$ is a filtered bialgebra structure, denoted by $`U`$, on the underlying filtered vector space $`B`$ with the above filtration such that
$`\mathrm{gr}U=B`$
as graded bialgebras, where $`\mathrm{gr}U`$ is the graded bialgebra associated to the filtered bialgebra $`U`$ (, p.226). (By $`\mathrm{gr}U=B`$, we use the natural identification of the underlying space $`\mathrm{gr}U`$ with $`B`$, that is $`\mathrm{gr}U_{(i)}=B_i/B_{i1}B_{(i)}`$ for each $`i0`$.)
For any lifting $`U`$ of the graded bialgebra $`B`$, it follows from the definition that $`U`$ and $`B`$ have the same identity element and the counit. Therefore, to give a lifting $`U`$, we just need to define the multiplication $`m_U`$ and comultiplication $`\mathrm{\Delta }_U`$ .
Two liftings $`U`$, $`V`$ of the graded bialgebra $`B`$ are said to be *equivalent*, if there is filtered bialgebra isomorphism $`\theta :UV`$ such that $`\mathrm{gr}\theta =\mathrm{Id}_B`$, where $`\mathrm{gr}\theta `$ is the graded morphism associated to $`\theta `$, and here again we use the identifications $`\mathrm{gr}U=B`$ and $`\mathrm{gr}V=B`$ (as graded bialgebras).
Denote by
$`\text{Lift}(B)`$
the set of equivalent classes of all the liftings of the graded bialgebra $`B`$.
### 2.2.
In this subsection, we will study graded bialgebra deformations of the graded bialgebra $`B=_{i0}B_{(i)}`$.
Let $`l\{+\mathrm{}\}`$. Consider the space $`B[t]/(t^{l+1})`$, which is viewed as a free module over $`๐[t]/(t^{l+1})`$, and also a graded $`๐`$-space with $`\mathrm{deg}t=1`$ and $`\mathrm{deg}b=n`$, if $`bB_{(n)}`$. If $`l=+\mathrm{}`$, then $`B[t]/(t^{l+1})`$ means $`B[t]`$ and $`๐[t]/(t^{l+1})`$ means $`๐[t]`$.
An *$`l`$-th level graded bialgebra deformation* of $`B`$ consists of
$`m_t^l:(BB)[t]/(t^{l+1})B[t]/(t^{l+1})`$
and
$`\mathrm{\Delta }_t^l:B[t]/(t^{l+1})(BB)[t]/(t^{l+1})B[t]/(t^{l+1})_{๐[t]/(t^{l+1})}B[t]/(t^{l+1})`$
which are $`๐[t]/(t^{l+1})`$-linear and homogeneous maps of degree zero such that
1. $`B[t]/(t^{l+1})`$ is a bialgebra over $`๐[t]/(t^{l+1})`$ with identity element $`1_B`$, multiplication $`m_t^l`$, counit $`\epsilon _t^l`$ and comultiplication $`\mathrm{\Delta }_t^l`$, where the counit $`\epsilon _t^l:B[t]/(t^{l+1})๐[t]/(t^{l+1})`$ is given by $`\epsilon _t^l(bt^j)=\epsilon (b)t^j`$, $`bB`$, $`0jl`$;
2. $`m_t^lm\mathrm{Id}_{๐[t]/(t^{l+1})}`$ and $`\mathrm{\Delta }_t^l\mathrm{\Delta }\mathrm{Id}_{๐[t]/(t^{l+1})}`$ mod$`(t)`$, where $`m`$ and $`\mathrm{\Delta }`$ are the multiplication and comultiplication of $`B`$, respectively.
Denote by $`(B[t]/(t^{l+1}),m_t^l,\mathrm{\Delta }_t^l)`$ the above $`l`$-th level graded bialgebra deformation.
From now on, we will abbreviate $`l`$-th level graded bialgebra deformations as $`l`$-deformations, and $`+\mathrm{}`$-deformations will be referred simply as deformations. Denote by $`^l(B)`$ the set of all $`l`$-deformations of the graded bialgebra $`B`$, and $`^+\mathrm{}(B)`$ is written as $`(B)`$. Elements of $`(B)`$ will be written as $`(B[t],m_t,\mathrm{\Delta }_t)`$.
Two $`l`$-deformations $`(B[t]/(t^{l+1}),m_t^l,\mathrm{\Delta }_t^l)`$ and $`(B[t]/(t^{l+1}),m_{}^{}{}_{t}{}^{l},\mathrm{\Delta }_{}^{}{}_{t}{}^{l})`$ are said to be *isomorphic*, if there exists an isomorphism of $`๐[t]/(t^{l+1})`$-bialgebras
$`\varphi :(B[t]/(t^{l+1}),m_t^l,\mathrm{\Delta }_t^l)(B[t]/(t^{l+1}),m_{}^{}{}_{t}{}^{l},\mathrm{\Delta }_{}^{}{}_{t}{}^{l})`$
such that $`\varphi `$ is homogeneous of degree zero and
$`\varphi \mathrm{Id}_B\mathrm{Id}_{๐[t]/(t^{l+1})}\text{mod}(t).`$
Denote by
$`iso^l(B)`$ (*resp*. $`iso(B)`$)
the set of isoclasses of $`l`$-deformations (*resp*. deformations) of the graded bialgebra $`B`$, for $`l`$.
### 2.3.
Use the notation as above. Consider an element $`(B[t]/(t^{l+1}),m_t^l,\mathrm{\Delta }_t^l)`$ of $`^l(B)`$. By definition, we can write
(2.1) $`m_t^l(ab)`$ $`={\displaystyle \underset{0sl}{}}m_s(ab)t^s,`$
and
(2.2) $`\mathrm{\Delta }_t^l(c)`$ $`={\displaystyle \underset{0sl}{}}\mathrm{\Delta }_s(c)t^s,`$
where $`a,b,cB`$, and $`m_s:BBB`$ and $`\mathrm{\Delta }_s:BBB`$ are homogeneous of degree $`s`$. Note that $`m_0=m`$ and $`\mathrm{\Delta }_0=\mathrm{\Delta }`$.
It is easy to check that the associativity of $`m_t^l`$, the compatibility of $`m_t^l`$ and $`\mathrm{\Delta }_t^l`$, and the coassociativity of $`\mathrm{\Delta }_t^l`$ are equivalent to the following identities, respectively, for each $`1nl`$,
(2.3) $`am_n(bc)m_n(abc)+m_n(abc)m_n(ab)c`$
$`=`$ $`{\displaystyle \underset{1sn1}{}}m_s(m_{ns}(ab)c)m_s(am_{ns}(bc)),`$
(2.4) $`m_n(a_{(1)}b_{(1)})a_{(2)}b_{(2)}\mathrm{\Delta }(m_n(ab))+a_{(1)}b_{(1)}m_n(a_{(2)}b_{(2)})`$
$`+a_{(1)}b_la_{(2)}b_r\mathrm{\Delta }_n(ab)+a_lb_{(1)}a_rb_{(2)}`$
$`=`$ $`{\displaystyle \underset{0s,r,s^{},r^{}n1,s+s^{}+r+r^{}=n}{}}(m_rm_r^{})\tau _{23}(\mathrm{\Delta }_s\mathrm{\Delta }_s^{})(ab)`$
$`+{\displaystyle \underset{1sn1}{}}\mathrm{\Delta }_s(m_{ns}(ab)),`$
and
(2.5) $`c_{(1)}\mathrm{\Delta }_n(c_{(2)})(\mathrm{\Delta }\mathrm{Id})\mathrm{\Delta }_n(c)+(\mathrm{Id}\mathrm{\Delta })\mathrm{\Delta }_n(c)\mathrm{\Delta }_n(c_{(1)})c_{(2)}`$
$`=`$ $`{\displaystyle \underset{1sn1}{}}(\mathrm{\Delta }_{ns}\mathrm{Id})\mathrm{\Delta }_s(c)(\mathrm{Id}\mathrm{\Delta }_{ns})\mathrm{\Delta }_s(c),`$
where we use Sweedlerโs notation $`\mathrm{\Delta }(a)=a_{(1)}a_{(2)}`$, $`aB`$, and in the second identity we use the notation $`\mathrm{\Delta }_n(a)=a_la_r`$ and $`\mathrm{\Delta }_n(b)=b_lb_r`$, and the map $`\tau _{23}`$ is the canonical flip map at the second and third positions.
Let $`(B[t]/(t^{l+1}),m_t^l,\mathrm{\Delta }_t^l)`$ and $`(B[t]/(t^{l+1}),m_{}^{}{}_{t}{}^{l},\mathrm{\Delta }_{}^{}{}_{t}{}^{l})`$ be two $`l`$-deformations with the maps $`m_s`$, $`\mathrm{\Delta }_s`$ and $`m_s^{}`$, $`\mathrm{\Delta }_s^{}`$ as in (2.1) and (2.2). An isomorphism $`\varphi `$ between these deformations is given by
(2.6) $`\varphi (a)={\displaystyle \underset{0sl}{}}\varphi _s(a)t^s,aB,`$
where $`\varphi _s:BB`$ is a homogeneous map of degree $`s`$. Note that $`\varphi _0=\mathrm{Id}_B`$. The fact that $`\varphi `$ is a morphism of $`๐[t]/(t^{l+1})`$-bialgebras implies $`\varphi `$ preserves the identity element $`1_B`$ and the counit $`\epsilon _t^l`$, and it satisfies, for each $`1nl`$,
(2.7) $`(m_nm_n^{})(ab)=a\varphi _n(b)\varphi _n(ab)+\varphi _n(a)b`$
$`+{\displaystyle \underset{0<s<n}{}}\{\varphi _s(a)\varphi _{ns}(b)\varphi _s(m_{ns}(ab))+{\displaystyle \underset{r+r^{}=ns}{}}m_s^{}(\varphi _r(a)\varphi _r^{}(b))\}`$
and
(2.8) $`(\mathrm{\Delta }_n\mathrm{\Delta }_n^{})(c)=\mathrm{\Delta }(\varphi _n(c))c_{(1)}\varphi _n(c_{(2)})\varphi _n(c_{(1)})c_{(2)}`$
$`+{\displaystyle \underset{0<s<n}{}}\{\mathrm{\Delta }_s^{}(\varphi _{ns}(c))(\varphi _s\varphi _{ns})(\mathrm{\Delta }(c)){\displaystyle \underset{r+r^{}=ns}{}}(\varphi _r\varphi _r^{})(\mathrm{\Delta }_s(c))\},`$
for all $`a,b,cB`$. Note that above discussion works for all $`l\{+\mathrm{}\}`$.
The analogue of the following lemma is well-known in classical deformation theory.
###### Lemma 2.1.
There exist restriction maps $`r_{l,l^{}}:^l(B)^l^{}(B)`$ for every $`l>l^{}`$, and maps $`r_l:(B)^l(B)`$ such that
$`(B)=\underset{}{\mathrm{lim}}_l^l(B).`$
Proof. The restriction map $`r_{l,l^{}}`$ is given as follows: given $`(B[t]/(t^{l+1}),m_t^l,\mathrm{\Delta }_t^l)`$ in $`^l(B)`$ with the maps $`m_s`$ and $`\mathrm{\Delta }_s`$ defined in (2.1) and (2.2), just define $`m_t^l^{}:=_{0sl^{}}m_st^s`$ and $`\mathrm{\Delta }_t^l^{}:=_{0sl^{}}\mathrm{\Delta }_st^s`$; it is direct to check that $`(B[t]/(t^{l^{}+1}),m_t^l^{},\mathrm{\Delta }_t^l^{})`$ is the desired element in $`^l^{}(B)`$. The map $`r_l`$ is defined in a similar way, and then the result is obvious. $`\mathrm{}`$
A graded bialgebra $`B=_{i0}B_{(i)}`$ is called *graded-rigid* if the set $`iso(B)`$ has only one element, i.e., any deformation of $`B`$ is isomorphic to the trivial one.
### 2.4.
We have the following observation, which says that the graded bialgebra deformations coincide with the liftings.
###### Theorem 2.2.
Let $`B=_{i0}B_{(i)}`$ be a graded bialgebra. There exists a natural bijection
$`Lift(B)iso(B).`$
Proof. We will construct a map $`F:Lift(B)iso(B)`$. Given a lifting $`U`$ of $`B`$. Denote by $`m_U`$ and $`\mathrm{\Delta }_U`$ the multiplication and comultiplication maps of $`U`$. Since $`U`$ is a filtered bialgebra, we have
$$m_U:B_iB_jB_{i+j}\text{and}\mathrm{\Delta }_U:B_n\underset{i+j=n}{}B_iB_j.$$
Therefore, for any $`s0`$, there uniquely exist homogeneous maps of degree $`s`$, say $`m_s:BBB`$ and $`\mathrm{\Delta }_s:BBB`$, such that
$$m_U(ab)=\underset{s0}{}m_s(ab)\text{and}\mathrm{\Delta }_U(c)=\underset{s0}{}\mathrm{\Delta }_s(c).$$
By $`\mathrm{gr}U=B`$ as graded bialgebras, we have $`m_0=m`$ and $`\mathrm{\Delta }_0=\mathrm{\Delta }`$.
Now Define $`F(U)=(B[t],m_t,\mathrm{\Delta }_t)`$ as follows
$`m_t(ab):={\displaystyle \underset{s0}{}}m_s(ab)t^s\text{and }\mathrm{\Delta }_t(c):={\displaystyle \underset{s0}{}}\mathrm{\Delta }_s(c)t^s.`$
It is direct to check that $`F(U)`$ is a deformation.
$`F`$ is well-defined, i.e., it maps equivalent liftings to isomorphic deformations. In fact, for given liftings $`U`$ and $`V`$, an equivalence $`\theta `$ of $`U`$ and $`V`$ is a filtered isomorphism, hence for any $`s0`$, there determines a unique homogeneous map $`\varphi _s:BB`$ of degree $`s`$ such that
$`\theta (a)={\displaystyle \underset{s0}{}}\varphi _s(a),aB.`$
Then define a $`๐[t]`$-linear map $`\varphi :B[t]B[t]`$ such that $`\varphi (a)=_{s0}\varphi _s(a)t^s`$. Hence $`\varphi `$ is an isomorphism between the deformations $`F(U)`$ and $`F(V)`$.
On the other hand, by seeing (2.1) and (2.2), one obtains that $`F`$ is a bijection. This completes the proof. $`\mathrm{}`$
An immediate consequence of Theorem 2.2 is
###### Corollary 2.3.
Let $`B=_{i0}B_{(i)}`$ be a graded bialgebra. Then $`B`$ is graded-rigid implies that, for any filtered bialgebra $`U`$ such that $`\mathrm{gr}UB`$ as graded bialgebras, we have $`UB`$ as bialgebras.
If we assume that $`B`$ is coradically-graded, the converse is also true.
Proof. By Theorem 2.2, $`B`$ is graded-rigid if and only if $`Lift(B)`$ is a single element set, i.e., every lifting of $`B`$ is trivial.
For the first statement, such a filtered bialgebra $`U`$ with $`\mathrm{gr}UB`$ gives rise to a lifting on $`B`$, denoted by $`U^{}`$, such that $`UU^{}`$ (as bialgebras). Since $`B`$ is graded-rigid, we get $`U^{}B`$, thus we are done.
For the second one, assume $`B`$ is coradically-graded. Let $`U`$ be a lifting of $`B`$. Thus by the assumption, there exists an isomorphism $`\theta :UB`$. Note that $`\theta `$ preserves the coradical filtration, thus $`\mathrm{gr}\theta `$ can be viewed as a graded automorphism of $`B`$. Thus take $`\theta ^{}=(\mathrm{gr}\theta )^1\theta :UB`$. So $`\theta ^{}`$ realizes an equivalence between the lifting $`U`$ and the trivial lifting. This proves that $`B`$ is graded-rigid. $`\mathrm{}`$
### 2.5.
In this subsection, we assume that the base field $`๐`$ is algebraically closed of characteristic zero. One can define the variety $`\mathrm{Bialg}_n`$ of the bialgebra structures on $`n`$-dimensional spaces, which carries a natural $`GL_n(๐)`$-action by base changes, see and . Recall that a bialgebra $`B`$ is called rigid if $`GL_n(๐)`$-orbit of $`\mathrm{Bialg}_n`$ containing $`B`$ is Zariski open. In fact, we have
###### Corollary 2.4.
Let $`๐`$ be an algebraically closed field of characteristic zero, $`B=_{i0}B_{(i)}`$ a finite dimensional graded bialgebra over $`๐`$. If $`B`$ is rigid and coradically-graded, then $`B`$ is graded-rigid in the sense of 2.3.
Proof. By Corollary 2.3, we only need to show that every filtered bialgebra $`U`$ with $`\mathrm{gr}UB`$ is isomorphic to $`B`$. Assume the dimension of $`B`$ is $`n`$. By Theorem 3.4 in , $`B`$ is a degeneration of $`U`$, i.e., lies the closure of the orbit of $`U`$ (in the variety $`\mathrm{Bialg}_n`$). However the $`GL_n(๐)`$-orbit of $`B`$ is open, we obtain that $`B`$ and $`U`$ belong to the same $`GL_n(๐)`$-orbit, i.e., $`BU`$ as bialgebras, finishing the proof. $`\mathrm{}`$
## 3. Graded bialgebra cohomology
In this section we will relate the graded bialgebra deformations with corresponding cohomology groups, which will be a graded (and normalized) version of โhatโ bialgebra cohomology groups introduced in (also see ).
### 3.1.
Let $`(B,m,e,\mathrm{\Delta },\epsilon )`$ be a bialgebra. Again we will use Sweedlerโs notation $`\mathrm{\Delta }(a)=a_{(1)}a_{(2)}`$, $`aB`$.
Let us recall the bicomplex in or , p.619. For this end, we need the following maps, where $`p,q1`$ and all $`b`$โs are in $`B`$, $`\lambda ^p:B^{p+1}B^p`$ and $`\rho ^p:B^{p+1}B^p`$ are given by
$`\lambda ^p(b^1\mathrm{}b^{p+1})`$ $`=b_{(1)}^1b^2\mathrm{}b_{(p)}^1b^{p+1},`$
$`\rho ^p(b^1\mathrm{}b^{p+1})`$ $`=b^1b_{(1)}^{p+1}\mathrm{}b^pb_{(p)}^{p+1}.`$
Dually, the maps $`\sigma ^q:B^qB^{q+1}`$ and $`\tau ^q:B^qB^{q+1}`$ are given by
$`\sigma ^q(b^1\mathrm{}b^q)`$ $`=(b_{(1)}^1\mathrm{}b_{(1)}^q)b_{(2)}^1\mathrm{}b_{(2)}^q,`$
$`\tau ^q(b^1\mathrm{}b^q)`$ $`=b_{(1)}^1\mathrm{}b_{(1)}^q(b_{(2)}^1\mathrm{}b_{(2)}^q).`$
In addition, we need $`\mathrm{\Delta }_i^p:B^pB^{p+1}`$ and $`\mu _j^q:B^{q+1}B^q`$, $`1ip`$ and $`1jq`$, which are given by
$`\mathrm{\Delta }_i^p(b^1\mathrm{}b^p)=b^1\mathrm{}b_{(1)}^ib_{(2)}^i\mathrm{}b^p,`$
$`\mu _i^q(b^1\mathrm{}b^{q+1})=b^1\mathrm{}b^ib^{i+1}\mathrm{}b^{q+1}.`$
Let $`C^{p,q}=\mathrm{Hom}_๐(B^q,B^p)`$, $`p,q1`$. Define
$`\delta _h^{p,q}:C^{p,q}C^{p,q+1}\text{ and }\delta _c^{p,q}:C^{p,q}C^{p+1,q}`$
which are given by
$`\delta _h^{p,q}(f)`$ $`=\lambda ^p(\mathrm{Id}f)+{\displaystyle \underset{i=1}{\overset{q}{}}}(1)^if\mu _i^q+(1)^{q+1}\rho ^p(f\mathrm{Id})`$
$`\delta _c^{p,q}(f)`$ $`=(\mathrm{Id}f)\sigma ^q+{\displaystyle \underset{j=1}{\overset{p}{}}}(1)^j\mathrm{\Delta }_j^pf+(1)^{p+1}(f\mathrm{Id})\tau ^q`$
for $`fC^{p,q}`$, where $`\mathrm{Id}`$ denotes the identity map of $`B`$.
It is direct to check that $`(C^{p,q},\delta _h^{p,q},\delta _c^{p,q})`$ is a bicomplex (see , p.619), i.e.,
$`\delta _h^{p,q+1}\delta _h^{p,q}=0,\delta _c^{p,q+1}\delta _h^{p,q}=\delta _h^{p+1,q}\delta _c^{p,q},\delta _c^{p+1,q}\delta _c^{p,q}=0.`$
We will introduce a sub-bicomplex of the above bicomplex. Let $`\text{m}=\mathrm{Ker}\epsilon `$. Denote by $`\text{i}:\text{m}B`$ the inclusion map, and $`\pi :B\text{m}`$ is given by $`\pi (b)=b\epsilon (b)1_B`$, $`bB`$. Set $`D^{p,q}=\mathrm{Hom}_๐(\text{m}^q,\text{m}^p)`$, $`p,q1`$. Note that we have a natural embedding $`D^{p,q}C^{p,q}`$ by identifying
$`fD^{p,q}\text{ with }\text{i}^pf\pi ^qC^{p,q}.`$
We have the following observation
###### Lemma 3.1.
Use the above notation. Then $`\delta _h^{p,q}(D^{p,q})D^{p,q+1}`$ and $`\delta _c^{p,q}(D^{p,q})D^{p+1,q}`$.
Proof. Just note that $`fC^{p,q}`$ lies in $`D^{p,q}`$ if and only if
$`(\mathrm{Id}^{j1}\epsilon \mathrm{Id}^{pj})f=0`$
and
$`f(b^1\mathrm{}b^{i1}1b^{i+1}\mathrm{}b^q)=0`$,
for any $`1iq`$, $`1jp`$ and any $`b^iB`$. Then the lemma follows from the definition of $`\delta _h^{p,q}`$ and $`\delta _c^{p,q}`$ immediately. $`\mathrm{}`$
### 3.2.
From now on $`B=_{i0}B_{(i)}`$ will be a graded bialgebra. In this case $`\text{m}B`$ is a graded subspace. Consider $`D_{(l)}^{p,q}:=\mathrm{Hom}_๐(\text{m}^q,\text{m}^p)_{(l)}`$ , $`l`$, whose elements are homogeneous maps from $`\text{m}^q`$ to $`\text{m}^p`$ of degree $`l`$. Note that $`D_{(l)}^{p,q}D^{p,q}C^{p,q}`$. We have the following
###### Lemma 3.2.
$`\delta _h^{p,q}(D_{(l)}^{p,q})D_{(l)}^{p,q+1}`$ and $`\delta _c^{p,q}(D_{(l)}^{p,q})D_{(l)}^{p+1,q}`$ for each $`l`$, $`p,q1`$.
Proof. Set $`C_{(l)}^{p,q}=\mathrm{Hom}_๐(B^q,B^p)_{(l)}`$. Clearly, $`D_{(l)}^{p,q}=D^{p,q}C_{(l)}^{p,q}`$. From the definition of $`\delta _h^{p,q}`$ and $`\delta _c^{p,q}`$, one sees that they preserve the degrees, i.e., $`\delta _h^{p,q}(C_{(l)}^{p,q})C_{(l)}^{p,q+1}`$ and $`\delta _c^{p,q}(C_{(l)}^{p,q})C_{(l)}^{p+1,q}`$. Now the result follows from Lemma 3.1. $`\mathrm{}`$
Denote by $`\delta _{h,(l)}^{p,q}`$ (*resp*. $`\delta _{c,(l)}^{p,q}`$) the restriction of the maps $`\delta _h^{p,q}`$ (*resp*. $`\delta _c^{p,q}`$) to the subspace $`D_{(l)}^{p,q}`$. Thus by Lemma 3.2, we get a bicomplex $`(D_{(l)}^{p,q},\delta _{h,(l)}^{p,q},\delta _{c,(l)}^{p,q})`$ for each $`l`$.
There is a canonical way to construct a complex from a given bicomplex. Set
$`\widehat{D}_{(l)}^n={\displaystyle \underset{p+q=n+1,p,q1}{}}D_{(l)}^{p,q},n1;`$
define $`_{(l)}^n:\widehat{D}_{(l)}^n\widehat{D}_{(l)}^{n+1}`$ by
$`_{(l)}^n|_{D_{(l)}^{n+1q,q}}:=\delta _{h,(l)}^{p,q}+(1)^q\delta _{c,(l)}^{p,q},1qn.`$
Hence, for each $`l`$, we get a complex
$`0\widehat{D}_{(l)}^1\stackrel{_{(l)}^1}{}\widehat{D}_{(l)}^2\stackrel{_{(l)}^2}{}\widehat{D}_{(l)}^3\stackrel{_{(l)}^3}{}\widehat{D}_{(l)}^4\mathrm{}`$
We define the n-th cohomology group of the above complex to be the *n-th graded โhatโ bialgebra cohomology of degree $`l`$* of the graded bialgebra $`B`$, which will be denoted by $`\widehat{h}_b^n(B)_{(l)}`$, $`n1`$, $`l`$.
It is very useful to write out $`\widehat{h}_b^2(B)_{(l)}`$ and $`\widehat{h}_b^3(B)_{(l)}`$ explicitly from the definition. In what follows, we will use the maps $`\delta _h^{p,q}`$ and $`\delta _c^{p,q}`$, instead of $`\delta _{h,(l)}^{p,q}`$ and $`\delta _{c,(l)}^{p,q}`$ for simplicity. We have the following facts.
1. The cohomology group $`\widehat{h}_b^2(B)_{(l)}`$ consists of all pairs $`(f,g)`$, where $`f:\text{m}\text{m}\text{m}`$ and $`g:\text{m}\text{m}\text{m}`$ are homogeneous maps of degree $`l`$, satisfying the following relations:
$`\delta _h^{1,2}(f)=0,\delta _c^{1,2}(f)+\delta _h^{2,1}(g)=0,\delta _c^{2,1}(g)=0,`$
i.e., for any $`a,b,c\text{m}`$, we have
(3.1) $`af(bc)f(abc)+f(abc)f(ab)c=0,`$
(3.2) $`f(a_{(1)}b_{(1)})a_{(2)}b_{(2)}\mathrm{\Delta }(f(ab))+a_{(1)}b_{(1)}f(a_{(2)}b_{(2)})`$
$`+a_{(1)}g(b)_la_{(2)}g(b)_rg(ab)+g(a)_lb_{(1)}g(a)_rb_{(2)}=0,`$
(3.3) $`c_{(1)}g(c_{(2)})(\mathrm{\Delta }\mathrm{Id})(g(c))+(\mathrm{Id}\mathrm{\Delta })(g(c))g(c_{(1)})c_{(2)}=0,`$
where we write $`g(b)=g(b)_lg(b)_r`$, $`bB`$.
Two pairs $`(f,g)=(f^{},g^{})`$ in $`\widehat{h}_b^2(B)_{(l)}`$ if and only if there exists a homogeneous map $`\theta :\text{m}\text{m}`$ of degree $`l`$ such that, for any $`a,b,c\text{m}`$,
(3.4) $`(ff^{})(ab)=a\theta (b)\theta (ab)+\theta (a)b,`$
(3.5) $`(gg^{})(c)=\mathrm{\Delta }(\theta (c))c_{(1)}\theta (c_{(2)})\theta (c_{(1)})c_{(2)}.`$
2. The group $`\widehat{h}_b^3(B)_{(l)}`$ consists of all triples $`(F,H,G)`$, where
$`F:\text{m}\text{m}\text{m}\text{m},H:\text{m}\text{m}\text{m}\text{m},G:\text{m}\text{m}\text{m}\text{m}`$
are homogeneous maps of degree $`l`$, subject to the relations:
$`\delta _h^{1,3}(F)=0,\delta _h^{2,2}(F)=\delta _c^{1,3}(H),\delta _c^{2,2}(H)=\delta _h^{1,3}(G),\delta _c^{3,1}(G)=0.`$
Note that $`(F,H,G)=0`$ in $`\widehat{h}_b^3(B)_{(l)}`$ if and only if there exists $`(f,g)\widehat{D}_{(l)}^2`$ such that
(3.6) $`(F,H,G)=_{(l)}^2((f,g)),`$
which can be written out explicitly by the definition of $`_{(l)}^2`$.
### 3.3.
Now we are at the position to present our main observations, which relate the graded bialgebra deformations of the graded bialgebra $`B`$ with the cohomology groups $`\widehat{h}_b^2(B)_{(l)}`$ and $`\widehat{h}_b^3(B)_{(l)}`$(compare , Section 5).
###### Theorem 3.3.
Let $`B=_{i0}B_{(i)}`$ be a graded bialgebra. Use the notation as above. Then
1. There is a bijection between $`iso^1(B)`$ and $`\widehat{h}_b^2(B)_{(1)}`$.
2. If $`\widehat{h}_b^2(B)_{(l)}=0`$ for each $`l1`$, then the graded bialgebra $`B`$ is graded-rigid.
3. The obstruction to extend an element of $`^l(B)`$ to $`^{l+1}(B)`$ lies in $`\widehat{h}_b^3(B)_{(l1)}`$, $`l1`$. In particular, if $`\widehat{h}_b^3(B)_{(l1)}=0`$, one can extend any element of $`^l(B)`$ to $`^{l+1}(B)`$.
Proof. (1). Recall from 2.2 that an element in $`^1(B)`$ is just given by $`(B[t]/(t^2),m_t^1,\mathrm{\Delta }_t^1)`$. As in 2.3, write
$`m_t^1(ab)=ab+f(ab)t,\mathrm{\Delta }_t^1(c)=\mathrm{\Delta }(c)+g(c)t,`$
where $`f:BBB`$ and $`g:BBB`$ are homogeneous of degree $`1`$. Note that $`1_B`$ is the identity element of $`B[t]/(t^2)`$, hence $`f(1_Bb)=f(b1_B)=0`$ for all $`bB`$. Moreover, for $`a,b\text{m}`$, $`\epsilon _t^1(m_t^1(ab))=0`$ implies that $`\epsilon _t^1(ab+f(ab)t)=0`$, i.e., $`f(ab)\text{m}`$. Thus we may view $`f`$ belongs to $`D_{(1)}^{1,2}`$. Dually one can show that $`gD_{(1)}^{2,1}`$.
Note that $`m_t^1`$ is an associative multiplication on $`B[t]/(t^2)`$, thus we get
$`f(ab)cf(abc)+f(abc)af(bc)=0,a,b,cB.`$
Therefore we get equation (3.1). Similarly, the fact that $`\mathrm{\Delta }_t^1`$ is an algebra morphism (resp. that $`\mathrm{\Delta }_t^1`$ is an coassociative comultiplication) gives us equation (3.2)(*resp*. equation (3.3)), i.e., $`(f,g)`$ can be viewed as an element in $`\widehat{h}_b^2(B)_{(1)}`$.
Suppose that $`(B[t]/(t^2),m_t^1,\mathrm{\Delta }_t^1)`$ and $`(B[t]/(t^2),m_{}^{}{}_{t}{}^{1},\mathrm{\Delta }_{}^{}{}_{t}{}^{1})`$ are two isomorphic deformations, with $`(f,g)`$ and $`(f^{},g^{})`$ defined as above, respectively. Let $`\varphi `$ (see also 2.3) be the isomorhism. We may write
$`\varphi (a)=a+\theta (a)t,aB,`$
for some homogeneous map $`\theta :BB`$ of degree $`1`$ (note that the map $`\theta `$ may be viewed as a map from m to m). Now it is direct to check that $`\theta `$ realizes an equivalence of $`(f,g)`$ and $`(f^{},g^{})`$ in $`\widehat{h}_b^2(B)_{(1)}`$. Now we have obtained a map from $`^1(B)`$ to $`\widehat{h}_b^2(B)_{(1)}`$, sending $`(B[t]/(t^2),m_t^1,\mathrm{\Delta }_t^1)`$ to $`(f,g)`$. One can easily see that the correspondence is bijective, as required.
(2). To prove that $`B`$ is graded-rigid, we just need to show that $`iso(B)`$ is a single-element set.
Let $`(B[t],m_t,\mathrm{\Delta }_t)`$ be an element in $`(B)`$. As before, write
$`m_t(ab)={\displaystyle \underset{s=0}{\overset{\mathrm{}}{}}}m_s(ab)t^s\text{and}\mathrm{\Delta }_t(c)={\displaystyle \underset{s=0}{\overset{\mathrm{}}{}}}\mathrm{\Delta }_s(c)t^s.`$
Note that $`m_0=m`$ and $`\mathrm{\Delta }_0=\mathrm{\Delta }`$, and $`m_s`$ and $`\mathrm{\Delta }_s`$ are homogeneous maps of degree $`s`$. By a similar argument as (1), we may view $`m_sD_{(s)}^{1,2}`$ and $`\mathrm{\Delta }_sD_{(s)}^{2,1}`$. Moreover, from (1), we see that $`(m_1,\mathrm{\Delta }_1)`$ can be viewed as an element in $`\widehat{h}_b^2(B)_{(1)}`$. Now by the assumption, there exists a homogeneous map $`\theta _1:\text{m}\text{m}`$ of degree $`1`$, such that (see (3.4) and (3.5))
$`m_1(ab)=a\theta _1(b)\theta _1(ab)+\theta _1(a)b,`$
$`\mathrm{\Delta }_1(c)=\mathrm{\Delta }(\theta _1(c))c_{(1)}\theta _1(c_{(2)})\theta _1(c_{(1)})c_{(2)}.`$
Take $`\varphi _1:B[t]B[t]`$ to be a $`๐[t]`$-linear map such that
$`\varphi _1(a)=a+\theta _1(a)t,aB.`$
Note that $`\varphi _1`$ is a bijective map preserving the identity $`1_B`$ and the counit $`\epsilon _t`$. Consider the deformation
$`(B[t],m_{t}^{}{}_{}{}^{}=\varphi _1m_t(\varphi _1^1\varphi _1^1),\mathrm{\Delta }_{t}^{}{}_{}{}^{}=(\varphi _1\varphi _1)\mathrm{\Delta }_t\varphi _1^1).`$
We have
$`m_{t}^{}{}_{}{}^{}(ab)`$ $`=ab+m_{2}^{}{}_{}{}^{}(ab)t^2+m_{3}^{}{}_{}{}^{}(ab)t^3+\mathrm{},`$
$`\mathrm{\Delta }_{t}^{}{}_{}{}^{}(c)`$ $`=\mathrm{\Delta }(c)+\mathrm{\Delta }_{2}^{}{}_{}{}^{}(c)t^2+\mathrm{\Delta }_{2}^{}{}_{}{}^{}(c)t^3+\mathrm{}`$
where $`m_{s}^{}{}_{}{}^{}`$ and $`\mathrm{\Delta }_{s}^{}{}_{}{}^{}`$ are homogeneous maps of degree $`s`$, $`s2`$. Now by comparing (2.3-5) and (3.1-3), we see that $`(m_{2}^{}{}_{}{}^{},\mathrm{\Delta }_{2}^{}{}_{}{}^{})`$ can be viewed as an element in $`\widehat{h}_b^2(B)_{(2)}`$. Hence there exists a homogeneous map $`\theta _2:\text{m}\text{m}`$ of degree $`2`$, such that (again see (3.4) and (3.5))
$`m_{2}^{}{}_{}{}^{}(ab)=a\theta _2(b)\theta _2(ab)+\theta _2(a)b,`$
$`\mathrm{\Delta }_{2}^{}{}_{}{}^{}(c)=\mathrm{\Delta }(\theta _2(c))c_{(1)}\theta _2(c_{(2)})\theta _2(c_{(1)})c_{(2)}.`$
Take $`\varphi _2:B[t]B[t]`$ to be a $`๐[t]`$-linear map such that
$`\varphi _2(a)=a+\theta _2(a)t^2,aB.`$
Now consider the following deformation
$`(B[t],m_{t}^{}{}_{}{}^{\prime \prime }=\varphi _2m_{t}^{}{}_{}{}^{}(\varphi _2^1\varphi _2^1),\mathrm{\Delta }_{t}^{}{}_{}{}^{\prime \prime }=(\varphi _2\varphi _2)\mathrm{\Delta }_{t}^{}{}_{}{}^{}\varphi _2^1),`$
whose coefficients of $`t`$ and $`t^2`$ vanishes. In other words,
$`m_{t}^{}{}_{}{}^{\prime \prime }(ab)`$ $`=ab+m_{3}^{}{}_{}{}^{\prime \prime }(ab)t^3+m_{3}^{}{}_{}{}^{\prime \prime }(ab)t^4+\mathrm{},`$
$`\mathrm{\Delta }_{t}^{}{}_{}{}^{\prime \prime }(c)`$ $`=\mathrm{\Delta }(c)+\mathrm{\Delta }_{3}^{}{}_{}{}^{\prime \prime }(c)t^3+\mathrm{\Delta }_{2}^{}{}_{}{}^{\prime \prime }(c)t^4+\mathrm{}`$
Similarly, we may view that $`(m_{3}^{}{}_{}{}^{\prime \prime },\mathrm{\Delta }_{3}^{}{}_{}{}^{\prime \prime })`$ lies in $`\widehat{h}_b^2(B)_{(3)}`$. By assumption and comparing (3.1-3), we have a homogeneous map $`\theta _3:\text{m}\text{m}`$ such that
$`m_{3}^{}{}_{}{}^{\prime \prime }(ab)=a\theta _3(b)\theta _3(ab)+\theta _3(a)b,`$
$`\mathrm{\Delta }_{3}^{}{}_{}{}^{\prime \prime }(c)=\mathrm{\Delta }(\theta _3(c))c_{(1)}\theta _3(c_{(2)})\theta _3(c_{(1)})c_{(2)}.`$
Now define $`\varphi _3:B[t]B[t]`$ to be a $`๐[t]`$-linear map such that
$`\varphi _3(a)=a+\theta _3(a)t^3,aB.`$
Thus we get the following deformation
$`(B[t],m_{t}^{}{}_{}{}^{\prime \prime \prime }=\varphi _3m_{t}^{}{}_{}{}^{\prime \prime }(\varphi _3^1\varphi _3^1),\mathrm{\Delta }_{t}^{}{}_{}{}^{\prime \prime \prime }=(\varphi _3\varphi _3)\mathrm{\Delta }_{t}^{}{}_{}{}^{\prime \prime }\varphi _3^1),`$
whose coefficients of $`t`$, $`t^2`$ and $`t^3`$ vanishes. Now one can define $`\theta _4`$ and $`\varphi _4`$, and so on.
Finally, define the infinite composition $`\mathrm{}\varphi _3\varphi _2\varphi _1`$ to be $`\varphi `$. Note that the $`๐[t]`$-linear isomorphism $`\varphi :B[t]B[t]`$ is well-defined on every $`aB`$, which preserves the identity $`1_B`$ and the counit $`\epsilon _t`$. (In fact, $`\varphi _s(a)=a+\theta _s(a)t^s`$ where $`\theta _s:\text{m}\text{m}`$ is homogeneous of degree $`s`$, hence, for each fixed $`aB_{(i)}`$, $`\varphi _s(a)=a`$ for $`si`$. Consequently, $`\varphi (a)`$ has only nonzero coeffecients of $`t^s`$ for $`0si`$.) By the construction of each map $`\varphi _s`$, we obtain that the deformation $`(B[t],\varphi m_t(\varphi ^1\varphi ^1),(\varphi \varphi )\mathrm{\Delta }_t\varphi ^1)`$ is trivial, which is also equivalent to the given deformation. Thus we have proved (2).
(3). Let $`(B[t]/(t^{l+1}),m_t^l,\mathrm{\Delta }_t^l)`$ be an element in $`^l(B)`$. Write
$`m_t^l(ab)={\displaystyle \underset{0sl}{}}m_s(ab)t^s\text{and}\mathrm{\Delta }_t^l(c)={\displaystyle \underset{0sl}{}}\mathrm{\Delta }_s(c)t^s,`$
where $`m_s`$ and $`\mathrm{\Delta }_s`$ are homogeneous maps of degree $`s`$. By the same argument as above, one can show that $`m_s`$ (*resp*. $`\mathrm{\Delta }_s`$) can be viewed as maps from $`\text{m}\text{m}`$ to m (*resp*. from m to $`\text{m}\text{m}`$).
To extend $`(B[t]/(t^{l+1}),m_t^l,\mathrm{\Delta }_t^l)`$ to some element in $`^{l+1}(B)`$, we just need to find some homogeneous maps $`f:\text{m}\text{m}\text{m}`$ and $`g:\text{m}\text{m}\text{m}`$ of degree $`(l+1)`$ such that $`(B[t]/(t^{l+2}),m_t^l+t^{l+1}f,\mathrm{\Delta }_t^l+t^{l+1}g)`$ is an bialgebra over $`๐[t]/(t^{l+2})`$.
The associativity of $`m_t^l+t^{l+1}f`$ is equivalent to
$`(m_t^l+t^{l+1}f)(((m_t^l+t^{l+1}f)(ab))c))=(m_t^l+t^{l+1}f)(a((m_t^l+t^{l+1}f)(bc))),`$
for all $`a,b,cB`$. Since $`m_t^l`$ is associative, then the above identity holds if and only if the two-sides have the same coefficients of the term $`t^{l+1}`$. Thus by direct computation, we get
$`F(abc):`$ $`={\displaystyle \underset{s=1}{\overset{l}{}}}m_s(m_{l+1s}(ab)c)m_s(am_{l+1s}(bc))`$
$`=af(bc)f(abc)+f(abc)f(ab)c`$
$`=\delta _h^{1,2}(f)(abc).`$
Similarly, one obtains that the compatibility of the multiplication $`m_t^l+t^{l+1}f`$ and comultiplication $`\mathrm{\Delta }_t^l+t^{l+1}g`$, and the coassociativity of $`\mathrm{\Delta }_t^l+t^{l+1}g`$ are equivalent to the following two identities, respectively,
$`H(ab):`$ $`={\displaystyle \underset{s=1}{\overset{l}{}}}\mathrm{\Delta }_s(m_{l+1s}(ab)){\displaystyle \underset{s+r+s^{}+r^{}=l+1}{}}(m_s^{}m_r^{})\tau _{23}(\mathrm{\Delta }_s(a)\mathrm{\Delta }_r(b))`$
$`=(\delta _c^{1,2}(f)+\delta _h^{2,1}(g))(ab),`$
and
$`G(c):`$ $`={\displaystyle \underset{s=1}{\overset{l}{}}}(\mathrm{\Delta }_s\mathrm{Id})\mathrm{\Delta }_{l+1s}(c)(\mathrm{Id}\mathrm{\Delta }_s)\mathrm{\Delta }_{l+1s}(c)`$
$`=c_{(1)}g(c_{(2)})(\mathrm{\Delta }\mathrm{Id})(g(c))+(\mathrm{Id}\mathrm{\Delta })(g(c))g(c_{(1)})c_{(2)}`$
$`=\delta _c^{2,1}(g)(c),`$
where $`a,b,c\text{m}`$, and $`\tau _{23}`$ is the flip map with respect to the second and third positions.
Now it is direct to check that the element $`(F,H,G)\widehat{D}_{(l1)}`$ is a cocycle (exactly as in the case of algebras and in the case of non-graded bialgebras), i.e., it lies in the kernel of the differential $`_{(l1)}^3`$ from (2.3-5), therefore, it can be viewed as an element in the cohomology group $`\widehat{h}_b^3(B)_{(l1)}`$. Now by comparing the above three identities with (3.6), we obtain that if $`\widehat{h}_b^3(B)_{(l1)}=0`$, then such maps $`f,g`$ always exist, i.e., we can extend $`(B[t]/(t^{l+1}),m_t^l,\mathrm{\Delta }_t^l)`$ to $`(B[t]/(t^{l+2}),m_t^l+t^{l+1}f,\mathrm{\Delta }_t^l+t^{l+1}g)`$. Note that by the above three equivalences, one deduces that $`(B[t]/(t^{l+2}),m_t^l+t^{l+1}f,\mathrm{\Delta }_t^l+t^{l+1}g)`$ belongs to $`^{l+1}(B)`$. This completes the proof. $`\mathrm{}`$
Acknowledgement: We thank the referees for many valuable suggestions.
|
warning/0506/physics0506106.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
Given the current political climate around the world, and the rise of extreme ideologies everywhere, from the Middle East to Africa and Western Europe, models that can provide insight into how such idealogies may spread in a society are clearly of great interest. In particular, given that, (1) the phenomenon of globalization has made interaction between people of various nations around the globe much easier than two decades ago, and (2) the fact that although extreme ideologies are usually advocated by very small fringe groups, but yet they continue to survive, it is important to understand the role of these two factors on the opinion about such antisocial behavior as terrorism. The goal of the present paper is to suggest a model to study this problem, and understand its implications.
Some simple models for terrorism or extreme opinions appeared years ago in the physics and sociological literatures. The present work was motivated by an article on bioterrorism , but the methods that we describe and utilize can also apply to opinion dynamics regarding, for example, the latest โStar Warsโ movie, fashion, a political candidate running for office, or other questions and opinions with varying degrees of enthusiasm. Thus, we do not even try to define โterrorismโ here, as the model that we consider is generic.
The population in our model consists of four parts, $`G,S,E,`$ and $`F`$ corresponding, respectively, to the general, susceptible, excited, and fanatic parts of the population. For simplicity, hereafter we use the same letters to denote the fractions of the total population belonging to each group. Members of the population can be convinced by acquaintances from the $`S,E,`$ and $`F`$ groups to move from the $`G`$ group to $`S`$; from there by the $`E`$ and $`F`$ groups to change to $`E`$, and from there by $`F`$ to change to $`F`$. Moreover, members of each of the three groups $`S,E,`$ and $`F`$ can change their status and go back directly to the $`G`$ group. The dynamics of a model based on such a partitioning of a population has been treated in the continuum limit by deterministic nonlinear differential equations, depending only on the total fractions $`G,S,E,`$ and $`F`$. The continuum model can provide mathematically sufficient conditions for terrorism, or any other opinion about a certain subject, to die out at long times, implying that in the long-time limit everybody will belong to the $`G`$ group , while the fractions $`S,E,`$ and $`F`$ shrink to zero which, when it comes to terrorism, is a good omen for the world.
However, as is well-known in the statistical physics of complex systems, deterministic continuum models represent mean-field approximations to the actual problem which, although they allow for development of mathematical proofs for the existence or nonexistence of certain phenomena and provide us with a first guide, they are also unreliable. Such models cannot take into account the effect of fluctuations on the phenomena. In addition, such models cannot take into account the effect of the internet, fax machines, and satellite television which have made long-range interactions between people on very large scale possible. For example, in a phenomenon somewhat close to the present problem, deterministic differential equations predicted extinction, whereas proper discrete simulations on a square lattice did not predict the same phenomenon.
Therefore, the goal of the present paper is to carry out extensive simulation of a discrete model opinion dynamics which, in a certain continuum limit, becomes similar to a deterministic model proposed by Ref.. We utilize Monte Carlo simulations of a population of individuals. Such simulations may be called agent-based outside physics, but are used in physics since half a century.
The plan of the paper is as follows. We first describe the deterministic continuum model which is based on a set of nonlinear differential equations. We then describe the discrete model which is developed by putting individuals on a two-dimensional lattice, but the individuals can still be influenced by all other individuals. Then, we restrict the influence to nearest neighbours. Finally, we replace the regular 2D lattice by a scale-free network of Barabรกsi-Albert type . The main point of the paper is not testing whether the model can provide quantitative predictions; rather, we deal only with the methods and how to implement them realistically. In particular we follow in assuming that if $`S=E=F=0`$ at some moment, then these three quantities stay zero forever. Thus, one simulation corresponds to the opinion dynamics following one external event and does not include new external events to cause $`S,E,`$ and $`F`$ to become non-zero again.
## 2 The Deterministic Continuum Model
The fractions $`G,S,E,`$ and $`F`$ in the population of agents having the corresponding opinions, with $`C=S+E+F=1G`$, change with time $`t`$ as:
$$dS(t)/dt=\beta _1CG\beta _2S(E+F)/C\gamma _1S$$
$`(1a)`$
$$dE(t)/dt=\beta _2S(E+F)/C\beta _3EF/C\gamma _2E$$
$`(1b)`$
$$dF(t)/dt=\beta _3EF/C\gamma _3E.$$
$`(1c)`$
Without loss of generality we set $`\beta _1=1`$ since, otherwise, it can be absorbed in the time scale. We also set $`\gamma _2=\gamma _1`$. Nevertheless, we still have not only the four parameters $`\beta _2,\beta _3,\gamma _1,`$ and $`\gamma _3`$, but also the three initial concentrations $`E(0),S(0),`$ and $`F(0)`$, which are relevant due to the nonlinearity of the continuum model. We use, in general, one million people and $`\beta _2=0.5,\beta _3=0.5,`$ and $`\gamma _3=0.20`$, starting with $`S=E=F=0.2`$, which will be used throughout the paper. Then, we check for changes in the behaviour when we vary $`\gamma _1`$.
Figure 1 illustrates the behaviour: As predicted in , for $`\gamma _1>\beta _1(=1`$ here), only the general population remains; for decreasing $`\gamma _1`$ first, $`S`$ also survives, then does $`E`$. Finally, for $`\gamma _1=0.25`$ $`F`$ also survives, such that all the four groups, $`G,S,E,`$ and $`F`$ remain present in the population.
## 3 Averaged Lattice
Now we put the agents onto a $`1000\times 1000`$ square lattice, half of whom is selected randomly to be fixed as empty. The fractions $`G,S,E,`$ and $`F`$ now refer to the filled half of the lattice, i.e., they are fractions of the population and not of the lattice size and, thus, still add up to unity. We try to follow closely the above set of equations, Eqs. (1), by the following rules for each time step $`tt+1`$ (simultaneous updating of all agents):
$`G`$ becomes $`S`$ with probability $`\beta _1C`$;
$`F`$ becomes $`G`$ with probability $`\gamma _3`$;
$`E`$ becomes $`G`$ with probability $`\gamma _2`$ and $`F`$ with probability $`\beta _3F/C`$, and
$`S`$ becomes $`G`$ with probability $`\gamma _1`$ and $`E`$ with probability $`\beta _2(E+F)/C`$.
These changes are simulated by first looking at the decay through $`\gamma `$ and then at the radicalisation through $`\beta `$. Therefore, it is possible that, e.g., an $`E`$ first becomes $`G`$ and immediately thereafter changes opinion to $`F`$.
Again we set $`\beta _1=1,\gamma _2=\gamma _1`$. Figure 2 looks similar to Fig. 1 which is not surprising since each agent is affected by all other agents, which is the limit that in statistical physics is described by a mean-field approximation. However, we always find for probability $`\gamma _1<1`$ some susceptible people, which is in contrast to their extinction by continuum model. Only in the unrealistic limit $`\gamma _1=1`$ do they die out.
The reason for this persistence of susceptibles can be understood as follows: For large enough $`\gamma _3`$, opinions $`E`$ and $`F`$ die out soon. Then, we have the differential equation for $`S=1G`$ as a simplification of Eq. (1a):
$$dS(t)/dt=\beta _1(1S)S\gamma _1S(\beta _1\gamma _1)S$$
$`(2)`$
for small $`S`$, giving an exponential decay towards zero for $`\gamma _1>\beta _1`$. For the Monte Carlo approach, $`G`$ becomes $`S`$ with probability $`\beta _1S`$ and $`S`$ becomes $`G`$ with probability $`\gamma _1`$. Equilibrium thus requires, for small $`S`$ and thus $`G`$ near unity, that:
$$\gamma _1S=G\beta _1S=\beta _1(1S)S\mathrm{or}1S=\gamma _1/\beta _1$$
$`(3)`$
which gives $`S=1\gamma _1`$ for our choice $`\beta _1=1`$. Only for $`\gamma _1>\beta _1`$ would $`S`$ become zero, which is not possible if $`\beta _1=1`$ since $`\gamma _1`$ is a probability for the Monte Carlo approach and no longer a rate which could also be larger than one. Putting back $`\beta _1`$ as a free parameter set equal to 0.5, everybody soon returns to the general population, $`S=E=F=0`$, for $`\gamma _1=0.6`$.
## 4 Nearest Neighbour Interactions
Now, we simulate a proper lattice population where only nearest neighbours influence each other. Thus at each time step every agent selects randomly one of the four nearest neighbours as a discussion partner and then follows these rules (again $`C=E+S+F`$):
$`G`$ becomes $`S`$ with probability $`\beta _1`$, if neighbour is $`S,E`$, or $`F`$;
$`F`$ becomes $`G`$ with probability $`\gamma _3`$;
$`E`$ becomes $`G`$ with probability $`\gamma _2`$, and
it becomes $`F`$ with probability $`\beta _3/C`$, if neighbour is $`F`$;
$`S`$ becomes $`G`$ with probability $`\gamma _1`$, and
it becomes $`E`$ with probability $`\beta _2/C`$, if neighbour is $`E`$ or $`F`$.
Thus, no agent is convinced by an empty neighbour to change opinion. Again we set $`\beta _1=1,\gamma _2=\gamma _1`$. Since the behaviour of the population now depends on the single opinions and not only on their sum over all lattice sites, ordered sequential updating with helical boundary conditions was used.
Figure 3 shows that, differently from Figs.1 and 2, both $`E`$ and $`F`$, and not only $`F`$, decay rapidly to zero; the susceptibles remain.
## 5 Networks
Human beings are not trees in an orchard, sitting on a square lattice and having $`k=4`$ neighbours each. The previous section used a half-empty square lattice such that the number of neighbours varied between $`k=0`$ and $`k=4`$. In reality, some people have many friends and some only few. Such social relationships are much better described by scale-free networks of Barabรกsi-Albert type, where the number of people having $`k`$ neighbours decays as $`1/k^3`$ for not too small $`k`$. Moreover, real terrorism obeys a power law .
To grow such a network we start with four people all connected with each other and with themselves. Then, one after the other more sites are added, and each new site selects four people of the existing network as neighbours from whom to take advice. This selection is not random but proportional to the current number $`k`$ of neighbours of that person: Powerful people attract more โfriendsโ than powerless people.
In this standard network, we then use directed opinion links. This means each person takes advice only from those four people whom the person selected when joining the network; the same person ignores the opinions of those who joined later and selected this person as advisor. Directed networks have been used before for, e.g., the Ising models and opinion dynamics . A computer program was listed in . Again, ordered sequential updating was used.
Figure 4 shows that in the present model everyone becomes normal: $`S=E=F=0`$ after sufficiently long time, differently from the results obtained with the square lattice of the previous section. In contrast to the square lattices for $`50\times 50`$ up to $`20,000\times 20,000`$, larger networks needed a slightly longer time for $`E`$ and $`F`$ to decay; see Fig. 5. With different parameters, also quite complicated dynamics can be found; see Fig. 6.
If we use symmetric instead of directed connections in these networks, now containing only 100,000 people, then for $`\gamma _1=0.1`$ all four groups survive; for $`\gamma _1=0.4`$ the $`F`$ die out; for $`\gamma _1=0.6`$ also the $`E`$ and for $`\gamma _1=0.9`$ we end up with $`S=E=F=0`$ (not shown). Similar effects are seen if, initially, the first four people are fanatic and all others have opinion $`G`$.
Thus, similarly to for Ising magnets and differently from for opinion dynamics, the directed network structure gives very different results compared to the undirected case.
## 6 Conclusions
In summary, the model of depends somewhat on the various changes in the underlying connections which we had introduced. Overall we feel that we confirmed the main conclusions from differential equations : Depending on parameters like $`\gamma _1`$, fanatics and/or excited agents survive or die out.
DS thanks F. Bagheri-Tar for crucial help to survive in Los Angeles.
|
warning/0506/physics0506224.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
There have been many attempts to construct models of financial markets and to understand the key statistical features of financial time series. It remains a great challenge, however, due the inherent complexity of the financial market, to develop a parsimonious market model that can reproduce all key โstylizedโ facts observed in real financial data and provide insights into the market mechanism for the emergence of these stylized facts. One of the promising approaches is agent-based modeling, which has reproduced and explained the emergence of some of the โstylizedโ facts. The agent-based modeling provides an ideal framework for investigating the impact of investorsโ behaviors on the price dynamics from many different perspectives; it has become an indispensable tool for understanding the price dynamics of financial markets<sup>1</sup><sup>1</sup>1LeBaron(2005) is a good recent survey on the field. With agent-based modeling, one can model and investigate, for example, how investors make their price forecasts and how their price forecasts influence the price fluctuation (Arthur, Holland, LeBaron, Palmer, and Tayler (1997), DeLong, Shleifer, Summers and Waldmann(1990), Levy, Levy and Solomon (2000)), how investors respond to price change (Lux and Marchesi(1999), Caldarelli, Marsli and Zhang (1997)), and how investors form and change their market beliefs (Brock and LeBaron (1996), Barberis, Shleifer and Vishny (1998)). In this paper we use agent-based models to study how investorsโ fluctuating risk preferences affect the price dynamics. This important issue has not been fully explored.
Financial markets present many important and challenging problems. First, the market consists of intelligent, competing and heterogeneous agents who, with different beliefs in the market, different abilities to acquire and process market information, and mutually conflicting interests, try to make investment decision for their own benefits. Second, each agentโs decision depends on his estimate of price expectations of other agents who also make their own estimates; this precludes expectations being formed by deductive means and leaves inductive reasoning as the only choice (Arthur et al 1997). A market of agents employing inductive reasoning often exhibits irrational herding behavior (Bak, Paczuski and Shubik (1997), Cont and Bouchaud (2001)), resulting in excessive price fluctuations or sometimes market bubbles and crashes. In such market agentsโ sentiment and degrees of risk aversion play a critical role in determining its price dynamics. Third, agents can learn and therefore adapt their strategies dynamically to improve their performance; this exacerbates the unpredictability of the markets. The change in an investorโs strategy or behavior can be the cause or the result of the investorโs changing sentiment (Barberis and Shleifer 1998), represented by (pessimistic) under-reaction or (optimistic) over-reaction to market dynamics driven by arrival of new information and changing market macro/micro-environment. These characteristics of financial market (competing with different beliefs and conflicting interests, interdependence of price expectations, and unpredictable changes of risk aversion) may lead to the formation of so-called noise traders behaviors. DeLong et al (1990) have found this a (behavioral) source for price to diverge significantly from the fundamental value in financial markets โ the so called โnoise trading effectโ. A good model, needless to say, must successfully address these challenges; more importantly, it must be able to produce simulation time series that can capture the key โstylizedโ empirical facts observed in real financial time series. This paper reports our efforts in constructing such a model. The key ingredient of our model is the inclusion of investorsโ changing degrees of risk aversion. We show that this is the main cause of excess stock price fluctuations and the associated volatility clustering.
We outline here our model of interacting heterogeneous agents. We first consider a baseline model, in which the agents use past price information to form their sets of future price expectations. The agents are adaptive as their price expectations are not based on one particular estimator but are determined by their best-performance estimators which may change from time to time. The agents also use their erroneous stochastic beliefs (DeLong et al.(1990)) in the market to make price adjustment on their best forecasts, which are assumed to be normally distributed.
In our baseline model, we assume that the agents have a decreasing absolute, but constant relative, risk aversion (DARA and CRRA) utility function, $`U(c)=(c^{1\gamma }1)/(1\gamma )`$. The existing agent based models of stock market, such as the Santa Fe artificial stock market model, typically use a constant absolute risk aversion (CARA) utility function, $`U(c)=e^{\lambda c}`$, for the price setting equation can be easily derived under such utility. As the focus of our paper is on risk aversion, we choose to use the well accepted (DARA) power utility function. Although a simple analytic formula for the demand function under this utility is not available, we have derived a general functional form and a rather good approximation for the demand function. Like the SFI market model, our baseline model market with a choice of the parameters corresponding to normal market conditions, exhibits some excess volatility, but not to the extent of the volatility observed in real markets. In addition, there is little enhancement of volatility clustering at high volatility regime, which is observed in real market data (Chen, Jayaprakash and Yuan (2005)).
By simply allowing investors to change their risk aversion attitudes, we obtain excess volatility and volatility clustering in very good agreement with real market data. The implication is clear: dynamic risk aversion (instead of fixed constant risk aversion) is directly responsible for excess volatility and the associated clustering. Specifically, in our DRA model, which is built from our baseline model, all agents have the power utility functions ($`(c^{1\gamma }1)/(1\gamma )`$) but with different and time varying risk aversion indices (degrees), $`\gamma _{i,t}`$, which we assume to follow an independent bounded random walk with a variance $`\delta ^2`$. We will show that the magnitude of excess volatility is directly related to $`\delta ^2`$. With such DRA our model market exhibits most of the important statistical characterization of real financial data, such as the โstylizedโ facts related to excess volatility (Mandelbrot (1963), Fama (1963), Bouchaud and Potters(2000), Mantegna and Stanley (1999), Cont (2001)) and volatility clustering (Mandelbrot (1963), Fama (1965), Engle (1982), Baillie, Bollerslev, and Mikkelsen (1996), Chou (1988), Schwert (1989), Poterba and Summers (1986), Chen et al. (2005)).
We have also studied the impacts of the dynamic risk aversion on the market dynamics using a few other baseline models, including the SFI market model, and we found similar results. This suggests that our results on dynamic risk aversion are rather generic.
The next section contains a derivation of the price equation under the power utility function for risk aversion. Section III describes our baseline model with a fixed constant risk aversion. Section IV introduces our DRA model. Section V reports the results from numerical simulations of our model. Section VI considers a DRA model built with the SFI market model as the baseline model. The last section summarizes.
## 2 Demand Function and Price Setting Under the Power Utility Function
We consider a market of $`N`$ heterogeneous agents who form their subjective expectations inductively and independently based on their investment strategies. There are two assets, a risky stock paying a stochastic dividend with a limited supply of $`N`$ shares<sup>2</sup><sup>2</sup>2In practice the number of shares is never the same as the number of agents; here we set the two numbers the same for the sake of convenience and setting them different does not change the results, and a risk-free bond paying a constant interest rate, $`r`$, with infinite supply. All agents have the same form of power utility function, $`U(c_t;\gamma _{i,t})=\frac{c_t^{1\gamma _{i,t}}1}{1\gamma _{i,t}}`$, but they have their own time-dependent risk aversion indices denoted by $`\gamma _{i,t}`$. At each time step $`t`$, every agent decides how to allocate his wealth between the risk-free bond and the risky stock. Since the values for both the dividend payment and the stock price at the next period $`t+1`$ are unknown random variables, the investors can only estimate the probability of various outcomes. Assume each agentโs estimation at time $`t`$ of the next stepโs price and dividend is normally distributed with the (conditional) mean and variance, $`E_{i,t}[p_{t+1}+d_{t+1}]`$ and $`\sigma _{i,t}^2(i=1,2,\mathrm{}N)`$ respectively. It can be shown, by optimizing the total utility, that the demand of agent $`i`$ for holding the share of the risky stock is approximately,
$$D_{i,t}=\frac{E_{i,t}[p_{t+1}+d_{t+1}]p_t(1+r)}{\gamma _{i,t}\sigma _{i,t}^2p_t(1+r)}$$
(1)
where $`p_t`$ is the stock price at time $`t`$, $`\gamma _{i,t}`$ is agent iโs index (degree) of risk aversion, and $`\sigma _{i,t}^2`$ the conditional variance of price estimation.
The market clearing condition: $`_i^ND_{i,t+\tau }=_i^ND_{i,t}=N`$ can be used to determine the current market price and relate the price at time $`t+\tau `$, $`p_{t+\tau }`$ to the price at time $`t`$, $`p_t`$:
$$p_t=\frac{_i^N\frac{E_{i,t}[p_{t+1}+d_{t+1}]}{\gamma _{i,t}\sigma _{i,t}^2(1+r)}}{N+_i^N\frac{1}{\gamma _{i,t}\sigma _{i,t}^2}}.$$
(2)
and
$$p_{t+\tau }=\frac{_i^N\frac{E_{i,t+\tau }[p_{t+\tau +1}+d_{t+\tau +1}]}{\gamma _{i,t+\tau }\sigma _{i,t+\tau }^2(1+r)}_i^N\frac{1}{\gamma _{i,t+\tau }\sigma _{i,t+\tau }^2}}{_i^N\frac{E_{i,t}[p_{t+1}+d_{t+1}]}{\gamma _{i,t}\sigma _{i,t}^2(1+r)}_i^N\frac{1}{\gamma _{i,t}\sigma _{i,t}^2}}p_t.$$
(3)
It can be seen from the demand and price equations that the degree of the agentโs risk aversion plays an important role.
We now show the derivation of the above equations. Assume at time $`t`$ agent $`i`$โs consumption is $`c_{i,t}`$, and he invests a portion $`x`$ of his current consumption in the risky asset. His total utility function defined over the current and future values of consumption is:
$$U(c_{i,t},c_{i,t+1},\gamma _{i,t})=U(c_{i,t},\gamma _{i,t})+U(\frac{c_{i,t+1}(x)}{R_f},\gamma _{i,t}),$$
(4)
where the consumption at time $`t+1`$ can be written as
$$c_{i,t+1}(x)=c_{i,t}[(1x)R_f+x\stackrel{~}{R}_{i,t+1}].$$
(5)
Here $`R_f=1+r`$ is the gross risk-free return and $`\stackrel{~}{R}_{i,t+1}`$ the gross return on the risky asset. Agent $`i`$ determines the amount of his investment on the risky asset, $`x`$, by maximizing his total utility, Eqn (4). The maximization problem can be written as
$$\underset{x}{\mathrm{max}}E_t[U(c_{i,t},\gamma _{i,t})+U(\frac{c_{i,t+1}(x)}{R_f},\gamma _{i,t})]\underset{x}{\mathrm{max}}E_t[U(\frac{c_{i,t+1}(x)}{R_f},\gamma _{i,t})],$$
(6)
The last equality follows because the utility $`U(c_{i,t},\gamma _{i,t})`$ is known at time $`t`$ and it does not contain $`x`$.
The power utility function is given by,
$$U(c_{i,t};\gamma _{i,t})=\frac{c_{i,t}^{1\gamma _{i,t}}1}{1\gamma _{i,t}}$$
(7)
Substituting Eqn (5) into Eqn (7) and then inserting it back to Eqn (6), the maximization is now given by
$$\underset{x}{\mathrm{max}}\frac{c_{i,t}^{1\gamma _{i,t}}}{1\gamma _{i,t}}E_t\{[1+\stackrel{~}{r}_{i,t+1}x]^{1\gamma _{i,t}}\}$$
(8)
where $`\stackrel{~}{r}_{i,t+1}=\frac{\stackrel{~}{R}_{i,t+1}R_f}{R_f}`$ is the present (discounted) value of the net return at next time step $`t`$+1. Assuming that agent $`i`$โs prediction errors of $`\stackrel{~}{r}_{i,t+1}`$ are (conditionally) normally distributed:
$$\stackrel{~}{r}_{i,t+1}=E_t(\stackrel{~}{r}_{i,t+1})+z_{i,t}=r_{i,t}^e+z_{i,t}$$
(9)
where $`r_{t,i}^e=E_t(\stackrel{~}{r}_{i,t+1})`$ is the conditional expected net return, at time $`t`$, of the next time step; and the error of estimation is $`z_{i,t}N(0,\sigma _{i,t}^2)`$. The maximization becomes:
$$\underset{x}{\mathrm{max}}E_t\{(1+xr_{i,t}^e+xz_{i,t})^{1\gamma _{i,t}}\}=\underset{x}{\mathrm{max}}\{\frac{1}{\sqrt{2\pi \sigma _{i,t}^2}}_{\mathrm{}}^{\mathrm{}}e^{\frac{z_{i,t}^2}{2\sigma _{i,t}^2}}f(z_{i,t};x,\gamma _{i,t})๐z_{i,t}\},$$
(10)
where $`f(z_{i,t};x,\gamma _{i,t})=(1+xr_{i,t}^e+xz_{i,t})^{1\gamma _{i,t}}`$. In writing down the above equation we implicitly assumed $`(1+xr_{i,t}^e+xz_{i,t})>0`$, which is a necessary requirement for the power function to be a valid utility measure. The above integral can be further simplified to
$$\frac{1}{\sqrt{\pi }}_{\mathrm{}}^{\mathrm{}}e^{z_{i,t}^2}f(\sqrt{2}\sigma _{i,t}z_{i,t};x,\gamma _{i,t})๐z_{i,t}$$
(11)
This Gaussian integral cannot be evaluated analytically, but it can be approximated by an expansion based on the roots of Hermite Polynomial $`H^{(n)}(\xi )`$ as:
$$_{\mathrm{}}^{\mathrm{}}e^{z_{i,t}^2}f(\sqrt{2}\sigma _{i,t}z_{i,t};x,\gamma _{i,t})๐z_{i,t}=\underset{k=1}{\overset{n}{}}\lambda _k^{(n)}f(\sqrt{2}\sigma _{i,t}\xi _k^{(n)};x,\gamma _{i,t})$$
(12)
where $`\lambda _k^{(n)}(k=1,2,\mathrm{},n)`$ are the coefficients of the summation, and $`\xi _k^{(n)}`$ are the roots of $`n`$th Hermite polynomial $`H^{(n)}(\xi )`$. Performing the maximization by setting the derivative with respect to $`x`$ equal to zero, we obtain the following equation:
$$\underset{k=1}{\overset{n}{}}\lambda _k^{(n)}[1+(r_{i,t}^e+\sqrt{2}\sigma _{i,t}\xi _k^{(n)})x]^{\gamma _{i,t}}\times (r_{i,t}^e+\sqrt{2}\sigma _{i,t}\xi _k^{(n)})=0$$
(13)
Since $`|r_{i,t}^e|1`$, $`\xi _k^{(n)}N(1)`$, and $`\sigma _{i,t}1`$ for the typical time step of one day or shorter, the above can be approximated as
$$\underset{k=1}{\overset{n}{}}\lambda _k^{(n)}[1\gamma _{i,t}(r_{i,t}^e+\sqrt{2}\sigma _{i,t}\xi _k^{(n)})x]\times [r_{i,t}^e+\sqrt{2}\sigma _{i,t}\xi _k^{(n)}]=0.$$
(14)
Here we consider an approximation with $`n=2`$. Note that $`\lambda _1^{(2)}=\lambda _2^{(2)}=\lambda `$, $`\sqrt{2}\xi _1^{(2)}=\sqrt{2}\xi _2^{(2)}(=1)`$, the optimal demand of agent $`i`$ of risky stock can then be obtained as
$$D_{i,t}=x_{i,t}=\frac{r_{i,t}^e}{\gamma _{i,t}[(r_{i,t}^e)^2+\sigma _{i,t}^2]}\frac{r_{i,t}^e}{\gamma _{i,t}\sigma _{i,t}^2}=\frac{E_{i,t}[p_{t+1}+d_{t+1}](1+r)p_t}{\gamma _{i,t}\sigma _{i,t}^2(1+r)p_t},$$
(15)
which is the Eqn. (1). In writing down the above approximation we assume $`(r_{i,t}^e)^2\sigma _{i,t}^2`$, which is certainly true when the time step is one day or shorter.
To get a more accurate approximation of the demand function, higher order Hermite polynomial roots are needed in the summation approximation to the integral in Eqn. (11). It can be shown that, the demand function $`x_i`$ in a higher order approximation is exactly the same as Eqn. (15), except for an overall constant factor.
## 3 The Baseline Model
### 3.1 Price prediction
To use the demand and price setting function derived in the previous section, one still need to incorporate each agentโs prediction of the payoff at the next time step, $`E_t(p_{t+1}+d_{t+1})`$. We assume all agents use the past price information for price forecasting. The simplest way is to calculate a moving average of the available past prices and use it as a proxy of price forecasting for the next time step $`t+1`$. Since investors may have different investment horizons and evaluation strategies, they may use different time lags for their calculation of the moving average of past prices, implying that they have heterogeneous memory lengths (Levy et al. (1994)). We consider each agent has his own $`M`$ sets of predictors so that he can choose the best one for forecasting the price at the next time step. Each price predictor $`E_{i,j}(p_{t+1}+d_{t+1}),(i=1,2,\mathrm{},N;j=1,2,\mathrm{},M)`$ is made of a moving average of past $`L_{i,j}`$ prices with a subjective erroneous stochastic adjustment:
$$E_{i,j,t}(p_{t+1}+d_{t+1})=MA_{i,j,t}=MA_{i,j,t1}(1\frac{1}{L_{i,j}})+\frac{1}{L_{i,j}}(p_t+d_t)+\epsilon _{i,j}$$
(16)
where $`\epsilon _{i,j}N(0,\sigma _{p+d})`$ is Gaussian random variable.
The conditional variance of the estimation, $`\sigma _{i,t}^2`$, is assumed to update with a moving average of the squared forecast error:
$$\sigma _{i,j,t}^2=(1\theta )\sigma _{i,j,t1}^2+\theta [(p_t+d_t)E_{i,j,t1}(p_t+d_t)]^2,$$
(17)
where $`\theta (0<\theta 1)`$ is a weighting constant.
### 3.2 Dividend process
The dividend process is assumed to be a random walk:
$$d_t=d_{t1}+r_d+ฯต_t,$$
(18)
where $`ฯต_t`$ is an i.i.d. Gaussian with zero mean and variance $`\sigma _d`$; $`r_d`$ is the average dividend growth rate. Note that the dividend process in a real stock market may be more complicated than what we assumed here and it may vary from stock to stock. But our results are not sensitive to the choice of a dividend process.
### 3.3 The price setting equation of the baseline model
For heterogeneous agents with fixed constant risk aversion, the demand function Eqn. (1) and price setting Eqn. (2) and (3) can be written as:
$$D_{i,t}=\frac{E_{i,t}[p_{t+1}+d_{t+1}]p_t(r+1)}{\gamma _i\sigma _{i,t}^2(1+r)p_t}$$
(19)
and
$$p_t=\frac{_i^N\frac{E_{i,t}[p_{t+1}+d_{t+1}]}{\gamma _i\sigma _{i,t}^2(1+r)}}{N+_i^N\frac{1}{\gamma _i\sigma _{i,t}^2}}$$
(20)
$$p_{t+\tau }=\frac{_i^N\frac{1}{\gamma _i}(\frac{E_{i,t+\tau }[p_{t+\tau +1}+d_{t+\tau +1}]}{\sigma _{i,t+\tau }^2(1+r)}\frac{1}{\sigma _{i,t+\tau }^2})}{_i^N\frac{1}{\gamma _i}(\frac{E_{i,t}[p_{t+1}+d_{t+1}]}{\sigma _{i,t}^2(1+r)}\frac{1}{\sigma _{i,t}^2})}p_t.$$
(21)
It can be see that the risk aversion indices of the agents play the role of weighting factors in the price setting equations.
If the agents are homogeneous in risk aversion, the above can be further simplified to:
$$D_{i,t}=\frac{E_{i,t}[p_{t+1}+d_{t+1}]p_t(r+1)}{\gamma \sigma _{i,t}^2(1+r)p_t}$$
(22)
and
$$p_t=\frac{_i^N\frac{E_{i,t}[p_{t+1}+d_{t+1}]}{\sigma _{i,t}^2(1+r)}}{N\gamma +_i^N\frac{1}{\sigma _{i,t}^2}}$$
(23)
$$p_{t+\tau }=\frac{_i^N(\frac{E_{i,t+\tau }[p_{t+\tau +1}+d_{t+\tau +1}]}{\sigma _{i,t+\tau }^2(1+r)}\frac{1}{\sigma _{i,t+\tau }^2})}{_i^N(\frac{E_{i,t}[p_{t+1}+d_{t+1}]}{\sigma _{i,t}^2(1+r)}\frac{1}{\sigma _{i,t}^2})}p_t.$$
(24)
From the above equations we see that the risk aversion index $`\gamma `$ only affects the overall level of demand but not its fluctuations. It also does not contribute to the price fluctuations (between the time $`t`$ and time $`t+\tau `$). Thus in the case of homogeneous and constant risk averse agents, the main source of the price fluctuations is from the investorsโ price forecasting. The baseline model does not incorporate investorโs changing sentiment. As a consequence, the price fluctuation is expected to be very limited and we will subsequently show that this is indeed the case.
## 4 Model with Dynamic Risk Aversion
### 4.1 Heterogeneous and dynamic risk aversion
To extend the baseline model, we allow agents to have heterogeneous risk aversion indices (degrees), which vary with time. This reflects the fact that in a real financial market investors have different risk attitudes and the investorsโ sentiment change with time. We assume that the risk aversion index of each agent follows an independent bounded random walk with a constant variance $`\delta ^2`$:
$$\gamma _{i,t}=\gamma _{i,t1}+\delta z_{i,t},\gamma _{i,t}[\gamma _0,\gamma _u]$$
(25)
where $`z_{i,t}`$ is an i.i.d. Gaussian variable with mean zero and unit variance for agent $`i`$, $`\gamma _0(>0)`$ is the lower boundary, and $`\gamma _u(>\gamma _0)`$ the upper boundary. Itโs easy to relate the value of the index at time $`t+\tau `$ to the value at time $`t`$:
$$\gamma _{i,t+\tau }=\gamma _{i,t}+\delta \underset{t=1}{\overset{\tau }{}}z_{i,t}=\gamma _{i,t}+\delta S_{i,\tau }$$
(26)
where $`S_{i,\tau }=_{t=1}^\tau z_{i,t}`$ is the change of risk aversion index of agent $`i`$ from time $`t`$ to time $`t+\tau `$, which can be either positive or negative. Itโs worthwhile to note that in real markets, the dynamics of investorsโ risk aversion attitudes may be more complicated than a simple random walk process we assume here. However, simplifying and idealizing of the real situation helps us to stay focused on the main purpose of investigating the impact of investorsโ fluctuating risk aversion on the price dynamics.
### 4.2 Price setting equation with dynamic risk aversion
Upon substitution of Eqn. (26) into Eqn. (3), we have:
$`p_{t+\tau }={\displaystyle \frac{_i^N\frac{1}{(\gamma _{i,t}+\delta S_{i,\tau })}(\frac{E_{i,t+\tau }(p_{t+\tau +1}+d_{t+\tau +1})}{\sigma _{i,t+\tau }^2(1+r)}\frac{1}{\sigma _{i,t+\tau }^2})}{_i^N\frac{1}{\gamma _{i,t}}(\frac{E_{i,t}(p_{t+1}+d_{t+1})}{\sigma _{i,t}^2(1+r)}\frac{1}{\sigma _{i,t}^2})}}p_t`$ (27)
Comparing Eqn.(27) to Eqn. (21) ($`\gamma _{i,t}=\gamma _{i,t+\tau }=\gamma _i`$) for the case of fixed constant risk aversion, we see that there is an extra term, $`\delta S_{i,\tau }`$, in the price setting equation in the case of DRA. Since $`S_{i,\tau }`$ can be either positive or negative and its value changes with time, $`\gamma _{i,t}+\delta S_{i,\tau }`$ deviates from $`\gamma _{i,t}`$ and fluctuates with time. This fluctuating weighting factor (representing agentโs fluctuating risk aversion) acts like an โamplifierโ of the price deviation induced by the error in agentsโ price estimation, and therefore results in excess price fluctuation. $`|S_{i,\tau }|N(0,\sqrt{\tau })`$, for $`\sqrt{\tau }\delta 1`$, $`\gamma _{i,t+\tau }(\gamma _{i,t}+\delta S_{i,\tau })`$ and $`\gamma _{i,t}`$ can differ substantially, resulting in a large deviation of $`p_{t+\tau }`$ from $`p_t`$. Our numerical results, presented in the next session, clearly show that itโs this risk aversion dynamics that gives rise to the excessive price fluctuations and the associated volatility clustering.
### 4.3 The range of DRA indices
We now examine the range of possible relative risk aversion indices. The choice of the range is important for modeling investorsโ decision-making; it has big impact on the price dynamics, as it directly affects investorsโ demand of the risky asset. The lower the index, the less risk-averse the investor is (thus the higher the demand of risky asset); and vice versa. Thus the risk-aversion attitude has great impact on the price dynamics through its influence on the demand. Some empirical and experimental studies reported that for a โtypicalโ investor, the value of the risk-aversion index $`\gamma `$ is in the range of 0-2 (Mehra and Prescott (1985), Friend and Blume (1975), Levy, Levy and Solomon (2000)). Mehra and Prescott (1985) used a value of risk aversion index with an upper limit of $`10`$ in their treatment of the issue of the โEquity Premium Puzzleโ. However, to โexplainโ the โEquity Premium Puzzleโ of NYSE over 50 years of U.S. postwar period, one needs a relative risk aversion index of 250 if a consumption-based model is used(Cochrane (2005))! These empirical results show that it is better to model the risk aversion with a range of indices, instead of a fixed value. The range we specified consists of an upper bound and a lower bound for the random walk describing DRA indices.
## 5 The Simulation Results and Analysis
### 5.1 The setup
In our simulation we choose the number of agents $`N=100`$, the number of predictors each agent has, $`M=2`$. Setting different number of agents produces similar results. The initial risk aversion indices $`\gamma _{i,0}`$ are all set to 1.0 for the baseline model and are set to $`\gamma _{i,0}[0.2,4]`$ for the model with DRA. The bounds for the index of DRA are $`\gamma _{i,t}[10^5,20]`$, the risk-free interest rate is $`r=5\%`$, the dividend growth rate is $`r_d=2\%`$, and the weighting coefficients for the variance of estimation is $`\theta =1/250`$. The lags used in the price estimators are $`L_{i,j}[2,250]`$, and we set $`\sigma _{p+d}`$=1% for all agents.
### 5.2 Simulation price and trading volume
Letโs first take a look at how excess price fluctuations emerge from a dynamic risk aversion process. Fig.1 shows the simulation time series of the price and the trading volume generated from the model with fixed constant risk aversion (CRA) and the model with dynamic risk aversion (DRA).
From the figure we see clearly that the DRA leads to increased fluctuations in both the price and the trading volume. To have both qualitative and quantitative picture of the impact of the DRA on the price dynamics, we examine the key stylized facts in the following subsections.
### 5.3 Autocorrelation function
One of the stylized facts observed in real financial data is that their autocorrelation functions (ACF) usually start with a low value (from $`\rho _1`$) and decay very slowly with increase of time step for the squared or absolute-valued returns. For an almost-Gaussian process, the values of its ACFs for absolute-valued return are close to zero and independent of the time steps. In Fig.2, we compare the ACFs for absolute-valued returns for the series generated by our baseline model (with CRA) and the DRA model, the series of real data (DJIA and SP500 Index), and Gaussian process.
The figure clearly shows that while the ACFs generated from our baseline model (CRA) is close to that of Gaussian process (Gauss), the results generated from our DRA model are very close to the real financial data (DJIA, SP500).
### 5.4 Excess volatility
The second key stylized facts we examine is the excess volatility (or fat-tails) of returns, which measures the price fluctuation of real financial series. In Fig.3 we plot the distributions of returns (in different time-steps) from our baseline model and the DRA model with the parameters set according to a normal market condition. For comparison, we also plot the return distribution of DJIA and the return distribution generated by a simple Gaussian process.
These plots show that, for all different time periods, the return distributions from our baseline model are very close to those of the Gaussian process; in contrast, the results from the model with DRA are close to the real DJIA data. In the context of our model, it is clear that the dynamic risk aversion leads to excess volatility or a fat-tail in the return distribution, which is one of the most important characterization of real financial time series (Mandelbrot (1963), Fama (1965)).
To further examine the fat tail of the distribution, we plot, in Fig.4, the Kurtosis as a function of the squre root of variance of risk aversion, $`\delta `$. From these plots, we see that the risk aversion dynamics can change the return distribution significantly from a Gaussian distribution (which has $`K`$=3.0). In addition, the smaller the lag $`\tau `$, the larger the Kurtosis generated; this is consistent with the empirical observations in real financial data. These values of the Kurtosis, together with the standard deviation and skewness are listed in Table 1. Note that the statistics generated from our DRA model are quite close to that from DJIA daily data.
### 5.5 Volatility Clustering
Volatility clustering is another important characteristics of financial time series. Here we use the conditional probability measure developed recently by Chen et al. (2005) to examine the volatility clustering. The method uses the return distribution conditional on the absolute return in the previous period to describe a functional relation between the variance of the current return and the absolute return in the previous period. If the volatility in asset returns is clustered, it will be proportional to the volatility in the previous period, the proportionality constant reflects the strength of volatility clustering.
Fig.5 shows how the current volatility depends on the volatility of the previous period for a random walk process, the baseline market model, DRA model, and DJIA daily data. From the figures we see clearly that the baseline market model with fixed constant risk aversion (CRA) produces very low volatility clustering (the curve is flat if there is no volatility clustering, such as the case with the random walk model (Gauss)). In contrast, the volatility clustering from the DRA model is significantly higher, and it is very close to the one generated from DJIA daily prices. The plots also show that, for our DRA model, the smaller the time period (step) of the return, the stronger the volatility clustering; and vice versa; this is consistent with the empirical observations of real market data.
## 6 The SFI market model with dynamic risk aversion
### 6.1 Brief introduction to SFI market
To test the impact of DRA on other baseline model, we use the well-known Santa Fe model of artificial market (Arthur et al. (1997), LeBaron, Arthur and Palmer (1999)). We first give a brief summary of the SFI market below. In the SFI market a constant absolute risk aversion (CARA) utility function ($`U(c_t,\gamma )=e^{c_t\gamma }`$) is assumed for each agent, the demand and price setting equations in fact have similar forms as in Eqn. (2) and (3).
The dividend process is assumed to be an AR(1) process:
$$d_t=\overline{d}+\rho (d_t\overline{d})+\epsilon _t$$
(28)
where $`ฯต_t`$ is an i.i.d. Gaussian with zero mean and variance $`\sigma _e`$
The SFI market assumes that each of the $`N`$ (=25) agents at any time possesses $`M`$ (=100) linear predictors and uses those that best matches the current market state and have recently proved most accurate. Each predictor is a linear regressor of the previous price and dividend, $`E(p_{t+1}+d_{t+1})=a(p_t+d_t)+b`$; it uses a market state โrecognizerโ vector consisting of $`J`$ (=12) elements, each taking a value of either $`0`$, $`1`$ or #(match any market states). The market status is described by a state vector consisting of $`J`$ binary elements, each taking value of either $`1`$ (its specified market condition exists) or $`0`$ (the market condition does not exist). The elements of market state can represent any important market discriminative information, including macro-/micro- economic environment, summary of fundamentals, and market temporal trends, etc. At each time step, only those predictors which match all their $`J`$ elements to the corresponding $`J`$ elements of market status are eligible to be used and are called โactiveโ predictors.
The variance of estimation $`\sigma _{i,t}^2`$ for each agent is assumed to update with a moving average of squared forecast error, defined in Eqn. (17).
Agents learn to improve their performance by discarding the worst (20%) predictors and developing new predictors via a genetic algorithm. This ensures the market some dynamics.
In the SFM, the market conditions are specified as:
1-6 elements represent โcurrent price $`\times `$ interest rate / dividend $`>`$ 0.25, 0.5, 0.75, 0.875, 1.0, 1.125.
7-10 elements describe โCurrent price $`>`$ 5-period moving average of past prices (5-period MA), 10-period MA, 100-period MA, 500-period MA.
11th element always 1;
12th element always 0;
The regressorโs parameters $`\{a,b\}`$ are set to be randomly and uniformly distributed within the ranges: $`a(0.7,1.2)`$ and $`b(10,19.002)`$.
The risk-free interest rate is set to 5%, and for the dividend process: the auto-regression coefficient $`\rho `$ = 0.95, $`\overline{d}=10`$, and $`\sigma _e=0.0745`$.
The weighting coefficients for the variance of estimation, $`\theta =1/75,1/150`$ for faster and slower learning respectively. The genetic algorithm is invoked (on average) every $`T_e=250`$ periods (faster learning) and 1000 periods (slower learning). For more detailed justification for choosing the parameter values, see LeBaron et al.(1999).
With these setups, we have checked that the simulation stock price time series and its statistical properties generated are in fact similar to those of our baseline model.
### 6.2 Numerical results of SFI market model with DRA
The dynamics of risk aversion can be similarly incorporated into the SFI market model. Fig.6 plots the simulation price time series for different DRA variance $`\delta ^2`$. These plots show clearly the impact on price fluctuations from the DRA. To have a quantitative picture on how the excess volatility emerges from the DRA, we plot in Fig.7 the functional relation of the Kurtosis vs variable $`\delta `$. The results are very similar to those plotted in Fig. 4, which were obtained with our much simpler baseline model.
We have checked other key features and found that the SFI-DRA model gives the similar results as our DRA model which is based on a much simpler baseline model. This suggests that DRA is the key mechanism for the emergence of the key stylized facts, and the impact of DRA does not depend on the structures of the baseline models. Therefore the price impact of investorsโ DRA we have studied is generic.
## 7 Summary
We have presented a simple multi-agent model of a financial market which incorporates the dynamics of risk aversions. We assume that the index of DRA follows a simple independent bounded random walk with a constant variance $`\delta ^2`$. We demonstrate that such dynamics is directly responsible for excess volatility and the associated volatility clustering. We compare the numerical results from our model with the results obtained by analyzing the DJIA daily data and show that the simulation data reproduce most of the โstylizedโ facts, such as excess volatility (measured by fat tail and high peak of return distribution), volatility clustering measured by conditional return distribution. We have also tested the DRA on the Santa Fe market model and obtain similar results. This suggests that the impact of DRA among heterogeneous agents we introduced here does not depend on the structure of the particular baseline model used. The degree of excess volatility is essentially controlled by the parameter $`\delta `$. Thus $`\delta `$ can be used as a key market sentiment parameter, in conjunction with the other market indicators such as average return $`r`$ and the average volatility $`\sigma _0`$, to characterize the financial market. We hope that our results presented here will provide new insights into the dynamics of asset price fluctuations governed by investorsโ fluctuating sentiments.
Acknowledgment
Baosheng Yuan is deeply grateful to Blake LeBaron for his very helpful and illuminating suggestions at several points in the research.
|
warning/0506/astro-ph0506170.html
|
ar5iv
|
text
|
# Swift, INTEGRAL, RXTE, and Spitzer reveal IGR J16283โ4838
## 1 Introduction
Star formation in our Galaxy takes place mainly in the dense regions of the spiral arms. These regions host massive molecular clouds and also the majority of the single and binary neutron stars ($`10^9`$) and black holes ($`10^8`$) in the Milky Way. The dense molecular clouds lead to strong star formation activity, which also results in the formation of binary systems, and subsequently to X-ray binary systems. These objects show X-ray flares and outbursts because of accretion processes onto the compact object. At the same time, the gas and dust of the spiral arms absorb most of the emission in the optical to soft X-ray regime below 10 keV. In addition, dense absorbing atmospheres around the object make the detection of these sources even more difficult. The hard X-ray and soft gamma-ray mission INTEGRAL Winkler et al. (2003) operates at energies above 20 keV. With the large field of view of the main instruments, the imager IBIS (Ubertini et al. 2003; $`19^{}\times 19^{}`$, partially coded FOV) and the spectrograph SPI (Vedrenne et al. 2003; $`35^{}\times 35^{}`$, PCFOV), and its observing program focussed on the Galactic plane and center, INTEGRAL is a powerful tool to discover highly absorbed sources ($`N_\mathrm{H}>10^{23}\mathrm{cm}^2`$) in the Galactic plane. So far a handful of those enigmatic objects has been found since the launch of INTEGRAL in October 2002<sup>1</sup><sup>1</sup>1for a list of all sources found by INTEGRAL see http://isdc.unige.ch/$``$rodrigue/html/igrsources.html. Six of those sources have been published so far: IGR J16318โ4848 Walter et al. (2003) with an absorption of $`N_\mathrm{H}19\times 10^{23}\mathrm{cm}^2`$ Matt & Guainazzi (2003), IGR J19140+0951 ($`N_\mathrm{H}=0.31.0\times 10^{23}\mathrm{cm}^2`$; Rodriguez et al. 2005), IGR J16320โ4751 ($`N_\mathrm{H}2\times 10^{23}\mathrm{cm}^2`$; Rodriguez et al. 2003), IGR J16393โ4643 ($`N_\mathrm{H}10^{23}\mathrm{cm}^2`$, Combi et al. 2004), IGR J16358โ4726 ($`N_\mathrm{H}4\times 10^{23}\mathrm{cm}^2`$, Patel et al. 2004), and IGR J16479โ4514 ($`N_\mathrm{H}>5\times 10^{23}\mathrm{cm}^2`$, Walter et al. 2004). While the nature of the latter source is still unknown, the other sources appear to be HMXBs, probably hosting a neutron star as the compact object. Most, if not all, of these sources show variable absorption. In this paper we report the discovery and analysis of another highly absorbed source, IGR J16283โ4838 Soldi et al. (2005). This work makes the first use of the combined data of INTEGRAL, Swift, RXTE, and Spitzer.
## 2 Observations of IGR J16283โ4838
All observations discussed in this Section are summarized in Table 1.
### 2.1 Discovery by INTEGRAL
IGR J16283โ4838 was discovered Soldi et al. (2005) during the observation of the Norma arm region by the imager IBIS/ISGRI Lebrun et al. (2003) on-board INTEGRAL. The observation lasted from April 7, 2005, 13:57 U.T. until April 9, 4:44 U.T. with an effective ISGRI exposure time of 126 ksec. The source position is $`R.A.=16^\mathrm{h}28.3^{}`$, $`DEC=48^{}38^{}`$ (J2000.0) with 3 arcmin uncertainty. The source showed a flux of $`f_X=(4.8\pm 0.8)\times 10^{11}\mathrm{erg}\mathrm{cm}^2\mathrm{s}^1`$ in the 20 - 60 keV band. No emission was detectable above 60 keV. From the analysis of another ISGRI observation with similar exposure time we estimate the $`3\sigma `$ upper limit in the $`60200\mathrm{keV}`$ band $`f_X<1.2\times 10^{10}\mathrm{erg}\mathrm{cm}^2\mathrm{s}^1`$. The analysis of the data prior to the discovery lasting from April 4, 01:55 U.T. until April 6, 11:24 U.T., with an exposure time of 192 ksec resulted in a $`3\sigma `$ upper limit of $`f_{2060\mathrm{keV}}=1.7\times 10^{11}\mathrm{erg}\mathrm{cm}^2\mathrm{s}^1`$. The source showed significant brightening during an INTEGRAL observation starting on April 10, 1:26 U.T. Even though IGR J16283โ4838 was in the partially coded field of view of IBIS, the analysis gave a $`11.6\sigma `$ detection within 96 ksec with a flux of $`f_{2060\mathrm{keV}}=(11.3\pm 1.0)\times 10^{11}\mathrm{erg}\mathrm{cm}^2\mathrm{s}^1`$ Paizis et al. (2005). The low flux level of the source did not allow the extraction of a spectrum from the ISGRI data and no simultaneous soft-X and optical observations are available as IGR J16283โ4838 was always outside the field of view of INTEGRALโs X-ray monitor JEM-X and of the optical monitor OMC. No further INTEGRAL observations of the source were obtained.
### 2.2 X-ray follow-up observations
After the discovery of IGR J16283โ4838 a Swift follow-up observation was requested in order to obtain an X-ray spectrum and an optical measurement. The Swift mission Gehrels et al. (2004) is a multiwavelength observatory for gamma-ray-burst astronomy. The payload combines a gamma-ray instrument (Burst Alert Telescope, 15 - 150 keV; Barthelmy et al. 2005), an X-ray telescope (XRT; Burrows et al. 2005), and a UV-optical telescope (UVOT; Roming et al. 2005). The XRT is a focussing X-ray telescope with a $`110\mathrm{cm}^2`$ effective area, 23 arcmin FOV, 18 arcsec resolution, and 0.2-10 keV energy range. The UVOT design is based on the Optical Monitor (OM) on-board ESAโs XMM-Newton mission, with a field of view of $`17\times 17\mathrm{arcmin}`$ and an angular resolution of 2 arcsec.
Two Swift observations took place 3 and 5 days after the last INTEGRAL observation. The first one started on April 13, 14:02 U.T. with an exposure time of 2.5 ksec, which resulted in an effective Swift/XRT exposure of 550 sec. A preliminary analysis of the XRT data refined the position of IGR J16283โ4838 to $`R.A.=16^\mathrm{h}28^{}10.7^{\prime \prime }`$, $`DEC=48^{}38^{}55^{\prime \prime }`$ (J2000.0) with an estimated uncertainty of 5 arcsec radius Kennea et al. (2005). A second observation was performed on April 15, 00:16 U.T. with 2600 sec effective XRT exposure time.
For our analysis of the Swift data we used the calibration files which have been released on April 5, 2005 and the software provided by the Swift Science Center. These tools are included in the release of HEAsoft 6.0 as of April 12, 2005. Applying a centroid algorithm to the data of April 15 gives a refined position for the source of $`R.A.=16^\mathrm{h}28^{}10.56^{\prime \prime }`$, $`DEC=48^{}38^{}56.4^{\prime \prime }`$ with an uncertainty of 6 arcsec radius, consistent with both the preliminary analysis and the INTEGRAL measurement. The spectrum extracted from the XRT data of April 15 is shown in Figure 1. The spectral fitting was done using version 11.3.2 of XSPEC Arnaud (1996). Both XRT spectra are well represented by an absorbed power law with the same photon index ($`\mathrm{\Gamma }=1.12\pm 0.35`$), but different absorption column density. The observation of April 13 shows a less absorbed ($`N_\mathrm{H}=0.6\genfrac{}{}{0pt}{}{+0.4}{0.2}\times 10^{23}\mathrm{cm}^2`$) spectrum with a lower flux ($`f_{210\mathrm{keV}}=(3.9\pm 0.3)\times 10^{11}\mathrm{erg}\mathrm{cm}^2\mathrm{s}^1`$) than the April 15 one. The latter data show $`N_\mathrm{H}=1.7\genfrac{}{}{0pt}{}{+0.5}{0.4}\times 10^{23}\mathrm{cm}^2`$ and a flux in the 2 โ 10 keV band of $`f_{210\mathrm{keV}}=(2.7\pm 0.3)\times 10^{11}\mathrm{erg}\mathrm{cm}^2\mathrm{s}^1`$. The data are equally well fit by an absorbed black body with $`N_\mathrm{H}=0.3/1.4\times 10^{23}\mathrm{cm}^2`$ (April 13/15) with a temperature of $`kT=2.0\pm 0.3\mathrm{keV}`$. Adding a Gaussian line to the fit does not improve the results significantly. The $`3\sigma `$ upper limit for the Fe K$`\alpha `$ line at 6.4 keV is $`3\times 10^4\mathrm{ph}\mathrm{cm}^2\mathrm{s}^1`$ with an equivalent width of $`EW<600\mathrm{eV}`$. Because of the short exposure time the source was not detected by the BAT instrument.
IGR J16283โ4838 was then also observed twice by RXTE using the Proportional Counter Array (PCA). The first observation starting on April 14, at 0:46 U.T. lasted 3.6 ksec, the second on April 15, at 16:07 U.T. lasted 2.9 ksec Markwardt et al. (2005). During both observations the PCA pointing was offset by 45 arcmin to avoid the nearby bright low mass X-ray binary 4U 1624โ490. The RXTE/PCA has a large field of view ($`2^{}`$ FWZM). For targets near the Galactic plane, a significant amount of Galactic diffuse emission enters the PCA aperture, which is considered background. This background was modeled by taking a nearby observation of the Galactic plane (observation 91409-01-02-00, $`l=341.4^{}`$, $`b=0.6^{}`$). This observation is at a similar latitude as IGR J16283โ4838, so the diffuse emission should have nearly the same spectrum. The background observation was modeled as thermal bremsstrahlung with a temperature of $`kT=7.4`$ keV, plus line emission at $`6.5`$ keV with an equivalent width of 600 eV. The shape of the background template was fixed and added to the spectral model of the two PCA observations of IGR J16283-4838; only the total normalization of the template was allowed to vary. The fluxes are collimator corrected after background subtraction. The best fit models for the source are shown in Table 1.
No pulsations are detectable in the PCA data.
### 2.3 Infrared and optical data
Within the 6 arcsec error radius around the refined position determined from the Swift/XRT data the infrared source 2MASS J16281083โ4838560 is located at 2.7 arcsec distance (Rodriguez & Paizis 2005). This source has K, J, and H band magnitudes of $`K=(13.95\pm 0.06)\mathrm{mag}`$, $`H>15.8\mathrm{mag}`$, and $`J>16.8\mathrm{mag}`$ ($`95\%`$ lower limits).
The Galactic Legacy Infrared Midplane Survey Extraordinaire (GLIMPSE<sup>2</sup><sup>2</sup>2publicly available at http://irsa.ipac.caltech.edu/data/SPITZER/GLIMPSE/; Benjamin et al. 2003) data show the source SSTGLMC G335.3268+00.1016 at 2.9 arcsec distance to the XRT position, consistent with the 2MASS detection. GLIMPSE is a 4-band near- to mid-infrared survey by Spitzer Werner et al. (2004) of the inner two-thirds of the Galactic disk with a spatial resolution of $`2^{\prime \prime }`$. The Infrared Array Camera Fazio et al. (2004) imaged 220 square degrees at wavelengths centered on 3.6, 4.5, 5.8, and 8.0 $`\mu m`$ in the Galactic longitude range $`10^{}`$ to $`65^{}`$ on both sides of the Galactic center and in Galactic latitude $`\pm 1^{}`$. The Spitzer/GLIMPSE data show a clear detection in all four energy bands (Tab. 1). Another observation in the K-band was performed with the 6.5m Magellan-Baade telescope on April 21, 2005. This observation indicates that the source seen in the 2MASS is a blend of point sources, with the brightest showing $`K=14.1\mathrm{mag}`$ Steeghs et al. (2005). Therefore the identification with the Spitzer source is tentative. In case the infrared source is not the counterpart to the hard X-ray source, the data presented here would be upper limits for the near and mid-infrared emission.
Within the error radius of IGR J16283โ4838 no optical counterpart is detectable on the POSS-II plates of the Digitised Sky Survey. During the observations by Swift on April 13 and on April 15 the UVOT took an image in the V-band. No source is detected within the error radius down to a magnitude of $`V>20\mathrm{mag}`$. The image extracted from the Swift/UVOT data on April 15 is shown in Figure 2. The contours indicate the XRT count map and the cross gives the position of the mid-infrared counterpart.
## 3 Spectral Energy Distribution
The spectral energy distribution (SED) of IGR J16283โ4838 is shown in Figure 3. In the chosen diagram a single power law with photon index $`\mathrm{\Gamma }=2`$ would appear as an even, horizontal line. No error bars have been included for the Swift/XRT data and only the XRT data of April 15 are shown for better visibility. From the comparison of the XRT data points with the measurements by RXTE/PCA it is apparent that both were taken during a similar high state of the source, while the two INTEGRAL/ISGRI measurements describe a lower flux state. Unfortunately the $`60200\mathrm{keV}`$ upper limit does not constrain the SED significantly.
Note that we display in the SED the absorbed X-ray fluxes as they are measured at the observer, as most of the absorption appears to be intrinsic to the source. The situation is different in the optical where the flux is absorbed already significantly by material in the line of sight. The hydrogen column density in the direction of the source is $`N_\mathrm{H}=2.2\times 10^{22}\mathrm{cm}^2`$. This leads to an extinction of $`A_V=N_\mathrm{H}/1.79\times 10^{21}\mathrm{cm}^2=12.3\mathrm{mag}`$ Predehl & Schmitt (1995). Therefore the unabsorbed optical limit is $`V>7.7\mathrm{mag}`$ and outside the displayed range of Figure 3. The absorption has a lower effect on the near-infrared fluxes. With $`A_K=0.112A_V`$, $`A_H=0.176A_V`$, and $`A_J=0.276A_V`$ (Schlegel, Finkbeiner, & Davis 1998) the unabsorbed flux values are $`K=12.7\mathrm{mag}`$, $`H>13.6\mathrm{mag}`$, and $`J>13.4\mathrm{mag}`$. The extinction in the mid-infrared range is negligible.
## 4 Discussion
The new hard X-ray source IGR J16283โ4838 exhibits several characteristic features. IGR J16283โ4838 is located at Galactic longitude $`l=335.3^{}`$ and latitude $`+6.1`$ arcmin in the Norma arm region. It shows a strong flare within a time scale of days. The absorption is of the order of $`N_\mathrm{H}=0.41.7\times 10^{23}\mathrm{cm}^2`$ with variations by a factor of 4 within one day and a flat X-ray spectrum ($`\mathrm{\Gamma }1`$) during all observations. The bimodal spectral energy distribution has one peak probably in the near-infrared and the other in the hard X-rays. The equivalent width of the iron K$`\alpha `$ line is $`EW<600\mathrm{eV}`$ ($`f_{\mathrm{K}\alpha }<3\times 10^4\mathrm{ph}\mathrm{cm}^2\mathrm{s}^1`$) and no optical counterpart is detectable ($`V>20\mathrm{mag}`$), probably because of absorption in the line of sight.
Combining this information enables us to put constraints on the nature of the source. The position within the Galactic plane at only $`+6.1`$ arcmin makes a Galactic origin of the source likely, even though some AGN have been seen through the plane by INTEGRAL, like the Seyfert 1 galaxy GRS 1734-292 (Marti et al., 1998; Sazonov et al., 2004). Strong variability as observed in IGR J16283โ4838 has been seen in the X-ray spectra of Seyfert galaxies (e.g. Dewangan et al. 2002), but the X-ray spectrum is too flat ($`\mathrm{\Gamma }1`$) for a Seyfert galaxy. This would still leave the possibility of a blazar as counterpart. But the absorption by the Galaxy in the direction of the source ($`N_\mathrm{H}=2.2\times 10^{22}\mathrm{cm}^2`$) is not high enough to explain the intrinsic absorption of $`1.7\times 10^{23}\mathrm{cm}^2`$, and thus intrinsic strong absorption in the blazar would be required to explain the Swift/XRT spectrum, but this has not been seen so far in blazar spectra.
If we consider IGR J16283โ4838 to be a Galactic source, mainly two types of bright and variable hard X-ray emitters are likely to be a counterpart: Low Mass and High Mass X-ray Binaries, LMXBs and HMXBs, respectively. The hard X-ray spectrum with strong absorption indicates the presence of a HMXB in which no pulsation have been detected so far Markwardt et al. (2005). Also the bright infrared emission, if connected to the X-ray source, would indicate a massive star as the companion of the compact object.
For a HMXB it is likely that IGR J16283โ4838 is located close to a star forming region in a Galactic spiral arm. Along the line of sight towards the source several arms are located Russeil (2003): the Sagittarius-Carina arm (0.7 kpc), the Scutum-Crux arm (3.2 kpc), the Norma-Cygnus arm (4.8 kpc), a star-forming region (7 kpc), and the Perseus arm (10.8 kpc). The luminosity of the object during the flare can be estimated by taking the brightest stage during the RXTE observation and assuming a distance to the object between 1 and 10 kpc. The unabsorbed flux is in this case only 20% larger than the absorbed one, because the significant part of the luminosity is emitted in the hard X-rays. The bolometric luminosity is then in the range $`\mathrm{log}L_{\mathrm{burst}}=34.036.5`$ (where $`L`$ is in units of $`\mathrm{erg}\mathrm{s}^1`$). The quiescent luminosity of the system is at least a factor of $`20`$ lower with $`\mathrm{log}L_\mathrm{q}<3335.2`$. This range of values is consistent with measurements from known Be/X-ray binaries with a neutron star as the compact object Negueruela (1998). In any case the luminosity is far below the Eddington luminosity of a neutron star of $`1.4M_{}`$ ($`L=1.8\times 10^{38}\mathrm{erg}\mathrm{s}^1`$).
The properties of IGR J16283โ4838 are similar to those of a number of highly absorbed sources ($`N_\mathrm{H}=120\times 10^{23}\mathrm{cm}^2`$) found in the Galactic plane, especially in the Norma arm region Walter et al. (2004). The HMXB IGR J19140+0951 shows also strong variable absorption Rodriguez et al. (2005), indicating intrinsic absorption in the source. The observed properties of IGR J16283โ4838 are consistent with those of IGR J19140+0951 in the bright state, where the iron line flux decreased to $`4\times 10^4\mathrm{ph}\mathrm{cm}^2\mathrm{s}^1`$, which is at the upper limit for the Swift/XRT measurement in our case. The (non-)variability of the absorption in IGR J16318โ4848 is still under discussion, as Walter et al. (2003) claim constant absorption, whereas Revnivtsev (2003) discovered variable absorption which could be connected with the orbital phase of the binary system. Only one of the newly detected highly absorbed sources has been claimed so far not to be a HMXB. Patel et al. (2004) observed IGR J16358โ4726 with Chandra. From the X-ray data they favour the source to be a millisecond pulsar LMXB even though the HMXB interpretation cannot be ruled out completely, though it would require some unknown kind of spin-down torque to prevent the neutron star from spinning up in this particular case.
X-ray binaries with strong intrinsic absorption have been known already before INTEGRAL, for example in 4U 1700-377, GX 301-2, Vela X-1, and CI Cam. Except for the latter one, where the nature of the source is unclear to date, these sources are also HMXBs, likely to host a neutron star as the compact object. Vela X-1 shows variable absorption from a negligible value up to $`7\times 10^{23}\mathrm{cm}^2`$ Pan et al. (1994). GX 301-2 shows strong absorption variation (up to $`12\times 10^{23}\mathrm{cm}^2`$; White & Swank 1984), and so does CI Cam ($`(0.025)\times 10^{23}\mathrm{cm}^2`$; Boirin et al. 2002). In 4U 1700-377 the absorption is linked to the state of the HMBX system and varies by a factor of 2 between 0.9 and $`2.0\times 10^{23}\mathrm{cm}^2`$ Boroson et al. (2003). It appears that variable absorption is a common feature in highly absorbed HMXBs. This could mean that the absorbing material is linked to the existence of a high mass donor in the binary system. In this case a strong and dense stellar wind ($`10^7`$ to $`10^5M_{}\mathrm{yr}^1`$) from the early-type stellar companion will probably cause the absorption in the system. The fact that all the absorbed sources so far have shown to be HMXBs (Kuulkers, 2005; Walter et al., 2004) containing neutron stars does not rule out significant contribution of HMXBs with a black hole as the compact object. But these systems are expected to be less numerous than the neutron star HMXBs by a factor of 10 to 100, making the detection of a black hole binary within a sample of only about 10 detected highly absorbed HMXBs unlikely. These absorbed binary systems might provide a significant contribution to the Galactic hard X-ray background at energies above 10 keV (Lebrun et al., 2004; Valinia et al., 2000).
## 5 Conclusions
The newly discovered hard X-ray source IGR J16283โ4838, located in the Norma arm region, is likely to be a HMXB containing a neutron star as the compact object. It is located in the Galactic Plane in the direction of star forming regions in the spiral arms and shows a large flare, which makes an extragalactic origin unlikely. The spectrum is hard ($`\mathrm{\Gamma }1`$) and strongly absorbed during the flare, which indicates a HMXB rather than a LMXB. The luminosity is comparably low ($`L<10^{37}\mathrm{erg}\mathrm{s}^1`$) which is typical for a neutron star HMXB. The strong and variable absorption ($`N_\mathrm{H}=0.41.7\times 10^{23}\mathrm{cm}^2`$) indicates that IGR J16283โ4838 belongs to the class of highly absorbed HMXBs discovered by INTEGRAL along the Galactic plane. Bright and absorbed sources like IGR J16283โ4838 could contribute significantly to the Galactic hard X-ray background in the 10โ200 keV band.
It has to be pointed out that the discovery and classification of IGR J16283โ4838 would not be possible without combining the observations of the recent observatories in space, like INTEGRAL, Swift, RXTE, and Spitzer. Combined efforts from these missions should lead to deeper insights into the nature of the hard X-ray source population in our Galaxy in the near future.
We like to thank John Greaves for pointing out the GLIMPSE data. This work is based in part on observations made with the Spitzer Space Telescope, which is operated by the Jet Propulsion Laboratory, California Institute of Technology under NASA contract 1407. This research has made use of the SIMBAD Astronomical Database which is operated by the Centre de Donnรฉes astronomiques de Strasbourg. This work was supported in part by NASA contract NAS5-00136.
|
warning/0506/cond-mat0506283.html
|
ar5iv
|
text
|
# Holographic assembly of quasicrystalline photonic heterostructures
(May 31, 2005)
## Abstract
Quasicrystals have a higher degree of rotational and point-reflection symmetry than conventional crystals. As a result, quasicrystalline heterostructures fabricated from dielectric materials with micrometer-scale features exhibit interesting and useful optical properties including large photonic bandgaps in two-dimensional systems. We demonstrate the holographic assembly of two-dimensional and three-dimensional dielectric quasicrystalline heterostructures, including structures with specifically engineered defects. The highly uniform quasiperiodic arrays of optical traps used in this process also provide model aperiodic potential energy landscapes for fundamental studies of transport and phase transitions in soft condensed matter systems.
Quasicrystals have long-ranged orientational order even though they lack the translational periodicity of crystals. Not limited by conventional spatial point groups, they can adopt rotational symmetries that are forbidden to crystals. The resulting large number of effective reciprocal lattice vectors endows quasicrystalsโ effective Brillouin zones with an unusually high degree of rotational and point inversion symmetry burkov92 . These symmetries, in turn, facilitate the the development of photonic band gaps (PBG) joannopoulos95 for light propagating through quasicrystalline dielectric heterostructures chan98 ; cheng99b ; Zhang01 , even when the dielectric contrast among the constituent materials is low. Photonic band gaps have been realized in one- hattori94 and two-dimensional zoorob00 lithographically defined quasiperiodic structures. Here we demonstrate rapid assembly of arbitrary materials into two- and three-dimensional quasicrystalline heterostructures with features suitable for photonic device applications.
Our approach is based on the holographic optical trapping technique dufresne98 ; grier03 ; polin05 in which computer-generated holograms are projected through a high-numerical-aperture microscope objective lens to create large three-dimensional arrays of optical traps. In our implementation, light at 532 nm from a frequency-doubled diode-pumped solid state laser (Coherent Verdi) is imprinted with phase-only holograms using a liquid crystal spatial light modulator (SLM) (Hamamatsu X8267 PPM). The modified laser beam is relayed to the input pupil of a $`100\times `$ NA 1.4 SPlan Apo oil immersion objective mounted in an inverted optical microscope (Nikon TE-2000U), which focuses it into traps. The same objective lens is used to form images of trapped objects, using the microscopeโs conventional imaging train polin05 .
We used this system to organize colloidal silica microspheres 1.53 $`\mu \mathrm{m}`$ in diameter (Duke Scientific Lot 5238) dispersed in an aqueous solution of $`180:12:1`$ (wt/wt) acrylamide, $`N,N^{}`$-methylenebisacrylamide and diethoxyacetophenone (all Aldrich electrophoresis grade). This solution rapidly photopolymerizes into a transparent polyacrylamide hydrogel under ultraviolet illumination, and is stable otherwise. Fluid dispersions were imbibed into 30 $`\mu \mathrm{m}`$ thick slit pores formed by bonding the edges of #1 coverslips to the faces of glass microscope slides. The sealed samples were then mounted on the microscopeโs stage for processing and analysis.
Silica spheres are roughly twice as dense as water and sediment rapidly into a monolayer above the coverslip. A dilute layer of spheres is readily organized by holographic optical tweezers into arbitrary two-dimensional configurations, including the quasicrystalline examples in Fig. 1. Figures 1(a), (b) and (c) show planar pentagonal, heptagonal and octagonal quasicrystalline domains suck , respectively, each consisting of more than 100 particles. Highlighted spheres emphasize each domainโs symmetry. These structures all have been shown to act as two-dimensional PBG materials in microfabricated arrays of posts and holes chan98 ; jin99 ; bayindir01 ; escuti04 . As a soft fabrication technique, holographic assembly requires substantially less processing than conventional methods such as electron-beam lithography, and can be applied to a wider range of materials. Unlike complementary optical fabrication techniques such as multiple-beam holographic photopolymerization escuti04 ; gauthier04 ; gauthier05 ; Wang03 , assembly with holographic optical traps lends itself to creating nonuniform architectures with specifically engineered features, such as the channel embedded in the octagonal domain in Fig. 1(d). Similar structures of comparable dimensions have been shown to act as narrow-band waveguides and frequency-selective filters for visible light jin99 ; bayindir01 ; chen99 ; jin00 .
Holographic trappingโs ability to assemble free-form heterostructures extends also to three dimensions. The sequence of images of a rolling icosahedron in Fig. 2 shows how the colloidal spheresโ appearance changes with distance from the focal plane. This sequence also recalls earlier reports leach04 ; sinclair04 that holographic traps can successfully organize spheres into vertical stacks along the optical axis, while maintaining one sphere in each trap.
The icosahedron itself is the fundamental building block of a class of three-dimensional quasicrystals, such as the example in Fig. 3. Building upon our earlier work on holographic assembly korda02 , we assemble a three-dimensional quasicrystalline domain by first creating a two-dimensional arrangement of spheres corresponding to the planar projection of the planned quasicrystalline domain, Fig. 3(a). Next, we translate the spheres along the optical axis to their final three-dimensional coordinates in the quasicrystalline domain, as shown in Fig. 3(b). One icosahedral unit is highlighted in Figs. 3(a) and (b) to clarify this process. Finally, the separation between the traps is decreased in Fig. 3(c) to create an optically dense structure. This particular domain consists of 173 spheres in roughly 7 layers, with typical inter-particle separations of 3 $`\mu \mathrm{m}`$.
The completed quasicrystal was gelled and its optical diffraction pattern recorded at a wavelength of 632 nm by illuminating the sample with a collimated beam from a HeNe laser, collecting the diffracted light with the microscopeโs objective lens and projecting it onto a charge-coupled device (CCD) camera with a Bertrand lens. The well-defined diffraction spots clearly reflect the quasicrystalโs five-fold rotational symmetry in the projected plane.
Holographic assembly of colloidal silica quasicrystals in water is easily generalized to other materials and solvents. Deterministic organization of disparate components under holographic control can be used to embed gain media in PBG cavities, to install materials with nonlinear optical properties within waveguides to form switches, and to create domains with distinct chemical functionalization. The comparatively small domains we have created can be combined into larger heterostructures through sequential assembly and spatially localized photopolymerization. In all cases, this soft fabrication process results in mechanically and environmentally stable materials that can be integrated readily into larger systems.
Beyond the immediate application of holographic trapping to fabricating quasicrystalline materials, the ability to create and continuously optimize such structures provides new opportunities for studying the dynamics polin05 and statistical mechanics denton98 of colloidal quasicrystals. The optically generated quasiperiodic potential energy landscapes developed for this study also should provide a flexible model system for experimental studies of transport korda02b through aperiodically modulated environments.
We are grateful to Paul Steinhardt, Paul Chaikin and Weining Man for illuminating conversations. Support was provided by the National Science Foundation through Grant Number DMR-0451589.
|
warning/0506/math-ph0506063.html
|
ar5iv
|
text
|
# Reduced Gutzwiller formula with symmetry: case of a finite group
## 1. Introduction
The purpose of this work is to give a Gutzwiller trace formula for a reduced quantum Hamiltonian in the framework of symmetries given by a finite group $`G`$ of linear applications of the configuration space $`^d`$. This semi-classical trace formula will link the reduced spectral density to periodic orbits of the dynamical system in the classical reduced space, i.e. the space of $`G`$-orbits.<sup>1</sup><sup>1</sup>1Results of this paper were published without proof in a Note aux Comptes Rendus (see ).
The role that symmetry plays in quantum dynamics was obvious since the beginning of the theory, and emphasized by Hermann Weyl in the book: โThe theory of groups and quantum mechanicsโ (). Pioneering physical results were given for models having a lot of symmetries. In the mathematical domain, first systematical investigations were done in 1978-79, mainly for the eigenvalues counting function of the Laplacian on a Riemannian compact manifold simultaneously by Donnelly and Brรผning & Heintze (see and ). Later, Guillemin and Uribe described the relation with closed trajectories in and . In $`^d`$, a general study was done in the early 80โs for globally elliptic pseudo-differential operators, both in cases of compact finite and Lie groups, by Helffer and Robert (see , ) for high energy asymptotics, and later by El Houakmi and Helffer in the semi-classical setting (see , ). Main results were then given in terms of reduced asymptotics of Weyl type for a counting function of eigenvalues of the operator. Here, in a semi-classical study with a finite group of symmetry, we want to go one step beyond Weyl formulae, investigating oscillations of the spectral density, and establishing a Gutzwiller formula for the reduced quantum Hamiltonian. The case of a compact Lie group will be carried out in another paper (see and ).
Without symmetry, in 1971, M.C. Gutzwiller published for the first time his trace formula linking semi-classically the spectrum of a quantum Hamiltonian $`\widehat{H}`$ near an energy $`E`$, to periodic orbits of the classical Hamiltonian system of $`H`$ on $`^{2d}`$, lying in the energy shell $`\mathrm{\Sigma }_E:=\{H=E\}`$. This was one of the strongest illustrations of the so-called โcorrespondence principleโ. Later, rigorous mathematical proofs were given (see for example , , , ), using various techniques like wave equation, heat equation, microlocal analysis, and more recently wave packets (see ).
Coming back to classical dynamics, let $`H:^{2d}`$ be a smooth Hamiltonian with a finite group of symmetry $`G`$, such that $`H`$ is $`G`$-invariant, i.e. suppose that there is an action $`M`$ from $`G`$ into $`Sp(d,)`$, the group of symplectic matrices of $`^{2d}`$, such that:
(1.1)
$$H(M(g)z)=H(z),gG,z^{2d}.$$
The Hamiltonian system associated to $`H`$ is:
(1.2)
$$\dot{z}_t=JH(z_t),\text{ where }J=\left(\begin{array}{cc}0& I_d\\ I_d& 0\end{array}\right).$$
In the framework of symmetry, specialists in classical dynamics are used to investigate this system in the space of $`G`$-orbits : $`^{2d}/G`$, also called the reduced space.
Here, for a quantum study with symmetry, it is therefore natural to expect a reduced Gutzwiller formula, linking semi-classically the spectrum of the reduced quantum Hamiltonian near the energy $`E`$ to periodic orbits of the reduced classical dynamical system on $`\mathrm{\Sigma }_E/G`$.
We now briefly describe our main result. First, we introduce our quantum reduction. We follow the same setting as in articles of Helffer and Robert , : let $`H:^{2d}`$ be a smooth Hamiltonian and $`G`$ a finite subgroup of the linear group $`Gl(d,)`$. If $`gG`$, we set:
(1.3)
$$M(g)(x,\xi ):=(gx,^tg^1\xi )$$
and we assume that $`H`$ is $`G`$-invariant as in (1.1). As usual, we make suitable assumptions -see (3.5)- to have nice properties for the Weyl quantization of $`H`$ (as functional calculus), which is defined as follows : for $`u๐ฎ(^d)`$,
(1.4)
$$Op_h^w(H)u(x)=(2\pi h)^d_^d_^de^{\frac{i}{h}(xy)\xi }H(\frac{x+y}{2},\xi )u(y)๐y๐\xi .$$
In particular, $`Op_h^w(H)`$ is essentially selfadjoint on $`๐ฎ(^d)`$ and we denote by $`D(\widehat{H})`$, $`\widehat{H}`$ its selfadjoint extension.
$`G`$ acts on the quantum space $`L^2(^d)`$ by $`\stackrel{~}{M}`$ defined for $`gG`$ by :
(1.5)
$$\stackrel{~}{M}(g)(f)(x)=f(g^1x),fL^2(^d),x^d.$$
If $`\chi `$ is an irreducible character of $`G`$, we set $`d_\chi :=\chi (Id)`$. Then, we define the symmetry subspace $`L_\chi ^2(^d)`$ associated to $`\chi `$, by the image of $`L^2(^d)`$ by the projector:
(1.6)
$$P_\chi :=\frac{d_\chi }{|G|}\underset{gG}{}\overline{\chi (g)}\stackrel{~}{M}(g),$$
$`L^2(^d)`$ splits into a Hilbertian sum of $`L_\chi ^2(^d)`$โs (Peter-Weyl decomposition), and the property (1.1) implies that each $`L_\chi ^2(^d)`$ is stable by $`\widehat{H}`$. Our goal is to give semi-classical trace formulae for the restriction $`\widehat{H}_\chi `$ of $`\widehat{H}`$ to $`L_\chi ^2(^d)`$, which will be called the reduced quantum Hamiltonian. We define the following reduced regularized spectral density :
(1.7)
$$๐ข_\chi (h):=\text{Tr}\left(\psi (\widehat{H}_\chi )f\left(\frac{E\widehat{H}_\chi }{h}\right)\right),$$
where $`\psi `$ is smooth, compactly supported in a neighbourhood $`]E\delta E,E+\delta E[`$ of $`E`$ ($`\delta E>0`$) such that $`H^1([E\delta E,E+\delta E])`$ is compact ($`\psi (\widehat{H}_\chi )`$ is an energy cut-off which is trace class), $`f`$ is smooth and $`\widehat{f}`$ (the Fourier transform of $`f`$) is compactly supported in $``$. The case where $`\text{Supp}(\widehat{f})`$ is localised near zero is the one that leads to Weyl formulae, and gives an asymptotic expansion of the counting function of $`\widehat{H}_\chi `$ (see Theorem 4.5, Corollary 4.6). Here we want to focus on the oscillating part of $`๐ข_\chi (h)`$. Thus we suppose that $`0\text{Supp}(\widehat{f})`$.
In order to state the theorem in terms of the reduced space, we need a smooth structure on $`\mathrm{\Sigma }_E/G`$, and thus we suppose that the group acts freely on $`\mathrm{\Sigma }_E`$, so that dynamics of $`H`$ on $`\mathrm{\Sigma }_E`$ would descend to the quotient. Note that this is not an essential assumption, since we have proved the asymptotic without this hypothesis (see Theorem 4.7). The following result involves the quantity $`\chi (g_{\overline{\gamma }}^n)`$, defined as follows: if $`\pi `$ denotes the projection on the quotient and $`\overline{\gamma }`$ is a periodic orbit in $`\mathrm{\Sigma }_E/G`$, if $`\pi (\gamma )=\overline{\gamma }`$, then, there is only one $`g_\gamma `$ in $`G`$ such that, $`z\gamma `$, $`M(g_\gamma )\mathrm{\Phi }_{T_{\overline{\gamma }}^{}}(z)=z`$, where $`T_{\overline{\gamma }}^{}`$ is the primitive period of $`\overline{\gamma }`$. If $`\pi (\gamma _1)=\pi (\gamma _2)`$ then $`g_{\gamma _1}`$ and $`g_{\gamma _2}`$ are conjugate elements of $`G`$, and we denote by $`\chi (g_{\overline{\gamma }})`$ the quantity $`\chi (g_{\gamma _1})=\chi (g_{\gamma _2})`$.
In order to have a finite number of periodic orbits of the reduced space involved in the trace formula, we will suppose that periodic orbits of $`\mathrm{\Sigma }_E/G`$ are non-degenerate, in the following sense : If $`\overline{\gamma }`$ is a periodic orbit of $`\mathrm{\Sigma }_E/G`$, with primitive period $`T_{\overline{\gamma }}^{}`$, and if $`n^{}`$ is such that $`nT_{\overline{\gamma }^{}}\text{Supp}(\widehat{f})`$, then $`1`$ is not an eigenvalue of the differential of the Poincarรฉ map in $`\mathrm{\Sigma }_E/G`$ at $`nT_{\overline{\gamma }^{}}`$: $`\mathrm{ker}[(dP_{\overline{\gamma }})^nId]=\{0\}`$. Then we have the following result:
###### Theorem 1.1.
Under previous assumptions, suppose that the group $`G`$ acts freely on $`\mathrm{\Sigma }_E`$ and that periodic orbits of $`\mathrm{\Sigma }_E/G`$ are non-degenerate in the sense given above. We then have a complete asymptotic expansion of $`๐ข_\chi (h)`$ in powers of $`h`$, modulo an oscillating factor of the form $`e^{i\frac{\alpha }{h}}`$ as $`h0^+`$ (see Theorem 4.7 for details). The first term is given by:
$$๐ข_\chi (h)=d_\chi \psi (E)\underset{\begin{array}{c}\overline{\gamma }\text{ periodic }\\ \text{orbit of }\mathrm{\Sigma }_E/G\end{array}}{}\underset{\begin{array}{c}n^{}\text{ s.t. }\\ nT_{\overline{\gamma }}^{}\text{Supp}\widehat{f}\end{array}}{}\widehat{f}(nT_{\overline{\gamma }}^{})\overline{\chi (g_{\overline{\gamma }}^n)}e^{\frac{i}{h}nS_{\overline{\gamma }}}\frac{T_{\overline{\gamma }}^{}e^{i\frac{\pi }{2}\sigma _{\overline{\gamma },n}}}{2\pi |det((dP_{\overline{\gamma }})^nId)|^{\frac{1}{2}}}+O(h).$$
where $`S_{\overline{\gamma }}:=_0^{T_{\overline{\gamma }}^{}}p_s\dot{q}_s๐s`$, $`P_{\overline{\gamma }}`$ is the Poincarรฉ map of $`\overline{\gamma }`$ in $`\mathrm{\Sigma }_E/G`$, and $`\sigma _{\overline{\gamma },n}`$. The other terms are distributions in $`\widehat{f}`$, with support in the set of periods of orbits in $`\mathrm{\Sigma }_E/G`$.
Remark 1: the case with $`0\text{Supp}(\widehat{f})`$ could have been included in the preceding theorem, and we would get a Weyl term in addition to this oscillating part. This term was already described by El Houakmi (see ) for the leading contribution. We obtain here slightly more detailled asymptotics for the Weyl part, by calculating the contribution of each $`gG`$ : see Theorem 4.5.
Remark 2: one could also consider a symmetry directly given in phase space $`^d\times ^d`$, and set $`G`$ as a finite subgroup of $`Sp(d,)`$. Then we would have to suppose that there is a unitary action $`\stackrel{~}{M}:G(L^2(^d))`$ which is metaplectic, i.e. satisfies:
(1.8)
$$\stackrel{~}{M}(g)^1Op_h^w(H)\stackrel{~}{M}(g)=Op_h^w(Hg),\text{ for all }g\text{ in }G.$$
For a fixed $`g`$, there is always some $`\stackrel{~}{M}(g)`$ satisfying (1.8), but it is not unique (multiply $`\stackrel{~}{M}(g)`$ by a complex of modulus $`1`$). The difficulty is to find a $`\stackrel{~}{M}`$ that is also a group homomorphism.
The method used is close to the one of : unlike articles previously quoted, which used an approximation of the propagator $`\mathrm{exp}(i\frac{t}{h}\widehat{H})`$ by some FIO following the WKB method, we will use here the work of Combescure and Robert on the propagation of coherent states. This method avoids problems of caustics and looks simpler to us. Moreover, the symmetry behaves well with coherent states, and we get very pleasant formulae. Thanks to these wave packets, we first reduce the problem to an application of the generalised stationary phase theorem (section 3). Then we find minimal hypotheses for the critical set to be a smooth manifold, and to ensure that the transverse Hessian of the phase is non-degenerate. These hypotheses will be called โ$`G`$-clean flow conditionsโ, and we get a theorical asymptotic expansion of $`๐ข_\chi (h)`$ under these assumptions (Theorem 4.4). Finally, as particular cases, we will show that these conditions are fulfilled on the one hand when $`\widehat{f}`$ is supported near zero (โWeyl termโ Theorem 4.5), and on the other hand when periodic orbits are non-degenerate (โOscillating termโ Theorem 4.7). In both cases, we calculate geometrically first terms of the asymptotic expansion, to make quantities of the reduced classical dynamics appear, as the energy level, periodic orbits and the Poincarรฉ map. The symmetry of periodic orbits plays an important part in the result.
Aknowledgements: We found strong motivation in the work of physicists B. Lauritzen, J.M. Robbins, and N.D. Whelan (, , ). I am deeply grateful to Didier Robert for his help, comments and suggestions. Part of this work was made with the support received from the ESF (program SPECT). I also thank Ari Laptev for many stimulating conversations.
## 2. Details on quantum reduction
### 2.1. Symmetry subspaces
We recall some basic facts on representations (see , or ): a representation $`\rho :GGl(E)`$ of the group $`G`$ on a finite dimensional complex vector space $`E`$ is said to be irreducible if there is no non-trivial subspace of $`E`$ stable by $`\rho (g)`$, for all $`g`$ in $`G`$. The character $`\chi _\rho :G`$ of a representation is defined by $`\chi _\rho (g):=Tr(\rho (g))`$, for $`gG`$. The degree of the representation $`\rho `$ is denoted by $`d_{\chi _\rho }`$ and is the dimension of $`E`$. Two such representations are isomorphic if and only if they have the same character. We will denote by $`\widehat{G}`$ the set of all irreducible characters, that is the set of characters of irreducible representations. Moreover, $`G`$ finite implies $`\widehat{G}`$ finite.
A representation $`\stackrel{~}{M}`$ of $`G`$ on a Hilbert space is said to be unitary if each $`\stackrel{~}{M}(g)`$ is a unitary operator. This is the case of our representation $`\stackrel{~}{M}`$ on the Hilbert space $`L^2(^d)`$ defined by (1.5) since $`|det(g)|=1`$. One can easily check that $`\stackrel{~}{M}`$ is strongly continuous. Then, the Peter-Weyl theorem (see or ) says that if one set $`L_\chi ^2(^d):=P_\chi (L^2(^d))`$, where $`P_\chi `$ is defined by (1.6), then the $`P_\chi `$โs are orthogonal projectors of sum identity, and we have the Hilbertian decomposition:
(2.1)
$$L^2(^d)=\underset{\text{ }\chi \widehat{G}\text{ }}{\overset{}{}}L_\chi ^2(^d).$$
Furthermore, if $`\chi \widehat{G}`$, then any irreducible sub-representation of $`\stackrel{~}{M}`$ in $`L_\chi ^2(^d)`$ is of character $`\chi `$, and a decomposition having such a property is unique. These $`L_\chi ^2(^d)`$โs will be called here the symmetry subspaces.
One has to think of them as a certain class of functions of $`L^2(^d)`$ having a certain symmetry linked to $`G`$ and $`\chi `$. For example, if $`G=\{\pm Id_^d\}`$, then we have two irreducible characters $`\chi _+`$ and $`\chi _{}`$ such that $`L_{\chi _+}^2(^d)`$ is the set of even functions of $`L^2(^d)`$, and $`L_\chi _{}^2(^d)`$ is the set of odd functions. More generally, if $`\chi `$ is a character of degree $`1`$, then $`\chi `$ is multiplicative, and we have:
$$L_\chi ^2(^d)=\{fL^2(^d):gG,\stackrel{~}{M}(g)f=\chi (g)f\}.$$
This is in particular the case for abelian groups. If $`G`$ $`\sigma _d`$ is the symmetric group of permutation matrices acting on $`^d`$, then there is at least two characters of degree $`1`$: $`\chi _0`$, the trivial character (always equal to $`1`$), and the signature $`\epsilon `$. Thus we get:
* $`L_{\chi _0}^2(^d)=\{fL^2(^d):\sigma G,f(x_{\sigma (1)},\mathrm{},x_{\sigma (d)})=f(x_1,\mathrm{},x_d)\}`$.
* $`L_\epsilon ^2(^d)=\{fL^2(^d):\sigma G,f(x_{\sigma (1)},\mathrm{},x_{\sigma (d)})=\epsilon (\sigma )f(x_1,\mathrm{},x_d)\}`$.
### 2.2. Reduced Hamiltonians
It is easy to check on the formula (1.4) that we have on $`๐ฎ(^d)`$:
(2.2)
$$\stackrel{~}{M}(g)^1Op_h^w(H)\stackrel{~}{M}(g)=Op_h^w(HM(g)),gG.$$
Thus we see that the property of $`G`$-invariance (1.1) is equivalent to the commutation of $`\widehat{H}`$ with all $`\stackrel{~}{M}(g)`$. In particular, it implies that $`\widehat{H}`$ commutes with all $`P_\chi `$โs, and thus, $`L_\chi ^2(^d)`$ is stable by $`\widehat{H}`$. We can then define the operator that we plan to study: if $`\chi \widehat{G}`$, set:
$$D(\widehat{H}_\chi ):=L_\chi ^2(^d)D(\widehat{H}),$$
The restriction $`\widehat{H}_\chi `$ of $`\widehat{H}`$ to $`L_\chi ^2(^d)`$ is called the reduced quantum Hamiltonian, and is a selfadjoint operator on the Hilbert space $`L_\chi ^2(^d)`$. If $`f:`$ is borelian, then we have:
$$[f(\widehat{H}),P_\chi ]=0,D(f(\widehat{H}_\chi ))=D(f(\widehat{H}))L_\chi ^2(^d),f(\widehat{H})=\underset{\chi \widehat{G}}{}f(\widehat{H}_\chi )P_\chi $$
$`f(\widehat{H}_\chi )`$ is the restriction of $`f(\widehat{H})`$ to $`L_\chi ^2(^d)`$. Lastly, if $`\sigma (.)`$ denotes the spectrum of an operator, then we have: $`\sigma (\widehat{H})=\underset{\chi \widehat{G}}{}\sigma (\widehat{H}_\chi )`$ (for details, see ).
One trace formula will be essential for the rest of this article:
###### Lemma 2.1.
If $`f:`$ is borelian, and if $`f(\widehat{H})`$ is trace class on $`L^2(^d)`$, then, for all $`\chi \widehat{G}`$, $`f(\widehat{H}_\chi )`$ is trace class on $`L_\chi ^2(^d)`$ and:
(2.3)
$$\text{Tr}(f(\widehat{H}_\chi ))=\text{Tr}(f(\widehat{H})P_\chi ).$$
Indeed, we have to show that $`|f(\widehat{H}_\chi )|^{\frac{1}{2}}`$ is Hilbert-Schmidt and $`|f(\widehat{H}_\chi )|^{\frac{1}{2}}_{HS}|f(\widehat{H})|^{\frac{1}{2}}_{HS}`$, which is clear by completing an Hilbertian basis of $`L_\chi ^2(^d)`$ in an Hilbertian basis of $`L^2(^d)`$. Then one writes:
$$Tr(f(\widehat{H}_\chi ))=\underset{\lambda \sigma (f(\widehat{H}_\chi ))\{0\}}{}dim(Ker[f(\widehat{H}_\chi )\lambda ])\lambda $$
$$Tr(f(\widehat{H})P_\chi )=\underset{\lambda \sigma (f(\widehat{H})P_\chi )\{0\}}{}dim(Ker[f(\widehat{H})P_\chi \lambda ])\lambda .$$
Furthermore, if $`\lambda 0`$, then $`Ker(f(\widehat{H})P_\chi \lambda )=Ker(f(\widehat{H}_\chi )\lambda )`$, and we get (2.3).
### 2.3. Interpretation of the symmetry
The investigation of $`\widehat{H}_\chi `$ provides informations on the spectrum of $`\widehat{H}`$:
###### Lemma 2.2.
If $`\chi \widehat{G}`$ then eigenvalues of $`\widehat{H}_\chi `$ have a multiplicity proportional to $`d_\chi `$.
Indeed, if $`FL_\chi ^2(^d)`$ is an eigenspace of $`\widehat{H}_\chi `$, then it is $`\stackrel{~}{M}`$-invariant. One can decompose it into irreducible representations. By the Peter-Weyl theorem, the only irreducible representation appearing is the one of character $`\chi `$, and thus is of dimension $`d_\chi `$. In particular, the operator $`\widehat{H}_\chi `$ provides a lower band for the multiplicity of some eigenvalues of $`\widehat{H}`$.
Another remark: by splitting an eigenfunction of $`\widehat{H}`$ on the symmetry subspaces, we get at least an eigenvector in one $`L_\chi ^2(^d)`$. This means that each eigenspace of $`\widehat{H}`$ contains an eigenvector having a certain symmetry. As it is well know for the double well potential ($`G=\{\pm Id\}`$), where eigenspaces are of dimension $`1`$, this leads to an alternance of even/odd eigenspaces and to tunneling effect.
If $`N_\chi (I)`$ denotes the number of eigenvalues of $`\widehat{H}_\chi `$ (with multiplicity) in an interval $`I`$ of $``$, and $`N(I)`$ the one of $`\widehat{H}`$, then the quantity $`N_\chi (I)/N(I)`$ can be thought as the proportion of eigenfunctions of symmetry $`\chi `$ among those corresponding to eigenvalues of $`\widehat{H}`$.
### 2.4. Examples
We give a few examples of Schrรถdinger Hamiltonians with a finite group of symmetry:
$$H(x,\xi ):=|\xi |^2+V(x).$$
1. $`G=\{\pm Id\}`$ : double well: $`V(x)=(x^21)^2`$, harmonic or quartic oscillator: $`V(x)=x^2`$ or $`x^4`$, โthe well on the islandโ: $`V(x)=(x^2+a)e^{x^2}`$ ($`a>0`$). For the two first examples, $`V(x)\underset{+\mathrm{}}{\overset{}{}}+\mathrm{}`$, so $`\widehat{H}`$ is essentially selfadjoint on $`๐ฎ()`$ and with compact resolvant.
2. $`G`$$`\sigma _2`$, $`d=2`$: any potential satisfying $`V(x,y)=V(y,x)`$.
3. Group of isometries of the triangle, $`d=2`$: $`V(x,y)=\frac{1}{2}(x^2+y^2)^2xy^2+\frac{1}{3}x^3`$, which in polar coordinates is $`\stackrel{~}{V}(r,\theta )=V(r\mathrm{cos}\theta ,r\mathrm{sin}\theta )=\frac{1}{2}r^2+\frac{1}{3}r^3\mathrm{cos}(3\theta )`$ (see also the Hรฉnon-Heiles potential: $`V(x,y)=\frac{1}{2}(x^2+y^2)xy^2+\frac{1}{3}x^3`$, but one has to look for the selfadjointness of this operator).
4. Group of isometries of the square, $`d=2`$: $`V(x,y)=\frac{1}{2}x^2y^2`$.
5. $`G(/2)^d`$: harmonic oscillator with distinct frequencies: $`V(x)=<Sx,x>_^d`$, with $`S`$ symmetric positive definite matrix with eigenvalues pairwise distincts. In this case, $`\widehat{H}`$ is still essentially selfadjoint on $`๐ฎ(^d)`$ and with compact resolvant. This is one of the few cases where we can calculate periodic orbits of the dynamical system.
## 3. Reduction of the proof by coherent states
We adapt here the method of . The essential tool is the use of coherent states.<sup>2</sup><sup>2</sup>2More details on the proof can be found in . We refer to the Appendix where we recall basic things about it (se also , , or ). Note that, by an averaging argument (see section 4.2), we could already restrict ourselves to a group of isometries. For the moment, we still use the general expression of (1.3), to keep in mind the symplectic form of $`M(g)`$. We suppose that $`\psi `$ and $`f`$ are in $`๐ฎ()`$ such that $`\text{Supp}(\psi )]E\delta E,E+\delta E[`$ and the Fourier transform $`\widehat{f}`$ of $`f`$ is with compact support. We know from , , that, under hypothesis (3.5), $`\psi (\widehat{H})`$ is trace class for little $`h`$โs, and, by formula (2.3), we have:
$$๐ข_\chi (h)=Tr\left(\psi (\widehat{H}_\chi )f\left(\frac{E\widehat{H}_\chi }{h}\right)\right)=\frac{d_\chi }{|G|}\underset{gG}{}\overline{\chi (g)}I_g(h),$$
where:
(3.1)
$$I_g(h):=Tr\left(\psi (\widehat{H})f\left(\frac{E\widehat{H}}{h}\right)\stackrel{~}{M}(g)\right).$$
Then, by Fourier inversion, we make the $`h`$-unitary quantum propagator $`U_h(t):=e^{i\frac{t}{h}\widehat{H}}`$ appear, and write:
(3.2)
$$I_g(h)=\frac{1}{2\pi }_{}e^{i\frac{tE}{h}}.\widehat{f}(t).\text{Tr}\left(\psi (\widehat{H})U_h(t)\stackrel{~}{M}(g)\right)dt.$$
Then we use the trace formula with coherent states โ see (5.4) โ to write:
(3.3)
$$I_g(h)=\frac{(2\pi h)^d}{2\pi }_{}_{_\alpha ^{2d}}e^{i\frac{tE}{h}}.\widehat{f}(t).m_h(\alpha ,t,g)d\alpha dt.$$
where
(3.4)
$$m_h(\alpha ,t,g):=<U_h(t)\phi _\alpha ;\stackrel{~}{M}(g)^1\psi (\widehat{H})\phi _\alpha >_{L^2(^d)}.$$
With exactly the same proof as in , we get the following lemma:
###### Lemma 3.1.
There exists a compact set $`K`$ in $`^{2d}`$ such that:
$$_{^{2d}K}|m_h(\alpha ,t,g)|๐\alpha =O(h^+\mathrm{}).$$
uniformly with respect to $`gG`$ and $`t`$.
We can then suppose that $`\mathrm{\Sigma }_E:=\{H=E\}`$ is included in $`K`$, and choose a real cut-off function $`\chi _1`$, compactly supported in $`^{2d}`$ and equal to $`1`$ on $`K`$. We can write $`1=\chi _1+(1\chi _1)`$ in (3.3), and settle problems at infinity in $`\alpha `$. Besides, we want to use the functional calculus of Helffer and Robert (, ) for the description of $`\psi (\widehat{H})`$. Thus we make the following hypothesis: $`C>0`$, $`C_\alpha >0`$, $`m>0`$ such that:
(3.5)
$$\{\begin{array}{c}<H(z)>C<H(z^{})>.<zz^{}>^m,z,z^{}^{2d}.\hfill \\ |_z^\alpha H(z)|C_\alpha <H(z)>,z^{2d},\alpha ^{2d}.\hfill \\ H\text{ has a lower band on }^{2d}.\hfill \end{array}$$
Then, we can write for $`N_0`$:
(3.6)
$$\psi (\widehat{H})=\underset{j=0}{\overset{N_0}{}}h^jOp_h^w(a_j)+h^{N_0+1}.R_{N_0+1}(h).$$
where $`\text{Supp}(a_j)H^1(]E\delta E,E+\delta E[)`$, $`a_0(z)=\psi (H(z))`$, with $`\underset{0<h1}{\text{Sup}}R_{N_0+1}(h)_{\text{Tr}}C.h^d.`$
We obtain:
(3.7)
$$I_g(h)=\underset{j=0}{\overset{N_0}{}}h^jI_g^j(h)+O(h^dh^{N_0+1}).$$
Now, we must get a complete asymptotic expansion for a fixed $`j_0`$ in $``$ of the quantity:
(3.8)
$$I_g^{j_0}(h)=\frac{(2\pi h)^d}{2\pi }_{}_{_\alpha ^{2d}}e^{i\frac{tE}{h}}.\widehat{f}(t)\chi _1(\alpha )m_h^{j_0}(\alpha ,t,g)d\alpha dt,$$
with
(3.9)
$$m_h^{j_0}(\alpha ,t,g):=<U_h(t)\phi _\alpha ;\stackrel{~}{M}(g)^1Op_h^w(a_{j_0})\phi _\alpha >_{L^2(^d)}.$$
For the right term of the bracket in (3.9), we expand $`Op_h^w(a_{j_0})\phi _\alpha `$ in powers of $`h`$, by Lemma 3.1 of . Thanks to (2.2), since $`๐ฏ_h(\alpha )=Op_h^w(\mathrm{exp}(\frac{i}{h}(pxq\xi ))`$ โ see Appendix โ we can write:
$$\stackrel{~}{M}(g)^1๐ฏ_h(\alpha )=๐ฏ_h(M(g^1)\alpha )\stackrel{~}{M}(g)^1.$$
For the left term of the bracket in (3.9), we use the theorem of propagation of coherent states given by Combescure and Robert (, or ). If $`M`$, then we have:
$$U_h(t)\phi _\alpha e^{i\frac{\delta (t,\alpha )}{h}}๐ฏ_h(\alpha _t)\mathrm{\Lambda }_h[\underset{j=0}{\overset{M}{}}h^{\frac{j}{2}}b_j(t,\alpha )(x).e^{\frac{i}{2}<M_0x,x>}]_{L^2(^d)}C_{M,T}(\alpha ).h^{\frac{M+1}{2}}$$
where $`\alpha _t=\mathrm{\Phi }_t(\alpha )`$ is the solution of the system (1.2) with initial condition $`\alpha `$ (see Appendix for other notations). After all, since there is no problem of control for $`\alpha `$ at infinity, we get:
(3.10)
$$m_h^{j_0}(\alpha ,t,g)=\underset{k=0}{\overset{2N}{}}\underset{j=0}{\overset{2Nk}{}}h^{\frac{j}{2}}h^{\frac{k}{2}}\underset{|\gamma |=k}{}\frac{^\gamma a_{j_0}(\alpha )}{\gamma !}e^{i\frac{\delta (t,\alpha )}{h}}\mathrm{{\rm Y}}_{j,\gamma }(\alpha ,t,g,h)+O(h^dh^{N+\frac{1}{2}}),$$
with:
$$\mathrm{{\rm Y}}_{j,\gamma }(\alpha ,t,g,h):=<๐ฏ_h(\alpha _t)\mathrm{\Lambda }_hb_j(t,\alpha )e^{\frac{i}{2}<M_0x,x>};๐ฏ_h(M(g^1)\alpha )\mathrm{\Lambda }_h\stackrel{~}{M}(g)^1Q_\gamma \stackrel{~}{\psi }_0>$$
where $`Q_\gamma `$ is the polynomial in $`d`$ variables such that:
(3.11)
$$Op_1^w(z^\gamma )\stackrel{~}{\psi }_0=Q_\gamma .\stackrel{~}{\psi }_0$$
We have: $`\mathrm{\Lambda }_h^{}๐ฏ_h(M(g^1)\alpha )๐ฏ_h(\alpha _t)\mathrm{\Lambda }_h=e^{\frac{i}{2h}<M(g^1)\alpha ,J\alpha _t>}๐ฏ_1\left(\frac{\alpha _tM(g^1)\alpha }{\sqrt{h}}\right)`$ (see Appendix). Thus:
$$\mathrm{{\rm Y}}_{j,\gamma }(\alpha ,t,g,h)=e^{\frac{i}{2h}<M(g^1)\alpha ,J\alpha _t>}<๐ฏ_1\left(\frac{\alpha _tM(g^1)\alpha }{\sqrt{h}}\right)b_j(t,\alpha )e^{\frac{i}{2}<M_0x,x>},\stackrel{~}{M}(g)^1Q_\gamma \stackrel{~}{\psi }_0>_{L^2}.$$
We will use the notation:
$$\alpha =(q,p)^d\times ^dand(q_t,p_t):=\alpha _t=\mathrm{\Phi }_t(\alpha ).$$
Make the change of variable: $`g^1y:=x(q_tg^1q)/\sqrt{h}`$ in the previous $`<;>_{L^2}`$. Since $`G`$ is compact, $`|det(g)|=1`$, and we obtain after calculation:
$$\mathrm{{\rm Y}}_{j,\gamma }(\alpha ,t,g,h)=\pi ^{\frac{d}{4}}e^{\frac{i}{h}[\frac{qp+q_tp_t}{2}^tgpq_t]+\frac{i}{2}|gq_tq|^2}_^de^{\frac{1}{2}<Ay,y>+\beta y}Q_\gamma \left(y+\frac{gq_tq}{\sqrt{h}}\right)b_j(\alpha ,t)(y)๐y,$$
where
(3.12)
$$A:=I^tg^1M_0g^1,and\beta :=\frac{i}{\sqrt{h}}[(qgq_t)+i(^tg^1p_tp)].$$
Then we set:
$$Q_\gamma (x)=:\underset{|\mu ||\gamma |}{}\kappa _{\mu ,\gamma }x^\mu \text{ and }b_j(t,\alpha )(x)=:\underset{|\nu |3j}{}c_{\nu ,j}(t,\alpha )x^\nu $$
(where $`c_{\nu ,j}`$ is smooth in $`t,\alpha `$). For the same reasons as in (parity of $`Q_\gamma `$ and $`b_j(t,\alpha )`$), only entire powers of $`h`$ have non-zero coefficients. Then, we can expend $`Q_\gamma `$ and $`b_j(t,\alpha )`$ and use the following calculus of the Gaussian:
###### Lemma 3.2.
Let $`AM_d()`$ such that $`{}_{}{}^{t}A=A`$, and that $`\mathrm{}A`$ is a positive definite matrix, $`\beta ^d`$ and $`\alpha ^d`$. Then $`A`$ is invertible and
$$_{_x^d}e^{\frac{1}{2}<Ax,x>+\beta x}x^\alpha ๐x=(2\pi )^{\frac{d}{2}}\text{det}_+^{\frac{1}{2}}(A)e^{\frac{1}{2}<A^1\beta ,\beta >}\underset{\eta \alpha }{}(A^1\beta )^\eta P_\eta (A),$$
where $`P_\eta (A)`$ doesnโt depend on $`\beta `$, and $`P_0(A)=1`$ (for a precise definition of $`\text{det}_+^{\frac{1}{2}}`$, see ).
We get: $`e^{i\frac{t}{h}E}e^{\frac{i}{h}\delta (t,\alpha )}\mathrm{{\rm Y}}_{j,\gamma }(\alpha ,t,g,h)=`$
$$\underset{|\nu |3j}{}\underset{|\mu ||\gamma |}{}\kappa _{\mu ,\gamma }c_{\nu ,j}(t,\alpha )\underset{\eta \mu }{}\left(\begin{array}{c}\mu \\ \nu \end{array}\right)(2\pi )^{\frac{d}{2}}\text{det}_+^{\frac{1}{2}}(Ii^tg^1M_0g^1)\underset{\sigma \mu \eta +\nu }{}(gq_tq)^\eta $$
$`\times \left[(Ii^tg^1M_0g^1)^1(\beta _0)\right]^\sigma P_\sigma (A)h^{\frac{1}{2}(|\sigma |+|\eta |)}\mathrm{exp}\left({\displaystyle \frac{i}{h}}\phi _E(t,\alpha ,g)\right),`$
where $`\beta _0:=\sqrt{h}\beta `$, and:
(3.13)
$$\phi _E(t,\alpha ,g)=tE+S(t,\alpha )+qp^tgpq_t+\frac{i}{2}|gq_tq|^2\frac{i}{2}<A^1\beta _0,\beta _0>.$$
Thus, (3.8) and (3.9) give:
$$I_g^{j_0}(h)=\frac{(2\pi h)^d}{2\pi }\underset{k=0}{\overset{2N}{}}\underset{j=0}{\overset{2Nk}{}}h^{\frac{j+k}{2}}\underset{|\gamma |=k}{}\underset{|\nu |3j}{}\underset{|\mu ||\gamma |}{}\frac{\kappa _{\mu ,\gamma }}{\pi ^{\frac{d}{4}}\gamma !}(2\pi )^{\frac{d}{2}}\underset{\eta \mu }{}\left(\begin{array}{c}\mu \\ \nu \end{array}\right)L_{\eta ,\nu ,\mu ,\gamma ,j}(h)+O(h^{N+\frac{1}{2}d}),$$
with:
(3.14)
$$L_{\eta ,\nu ,\mu ,\gamma ,j}(h):=\underset{\sigma \mu \eta +\nu }{}h^{\frac{1}{2}(|\sigma |+|\eta |)}__t_{_\alpha ^{2d}}\mathrm{exp}\left(\frac{i}{h}\phi _E(t,\alpha ,g)\right)\widehat{f}(t)^\gamma a_{j_0}(\alpha )\chi _1(\alpha )D_{\sigma ,\eta ,\nu ,j}(t,\alpha ,g)d\alpha dt.$$
where: $`D_{\sigma ,\eta ,\nu ,j}(t,\alpha ,g):=`$
(3.15)
$$c_{\nu ,j}(t,\alpha )\text{det}_+^{\frac{1}{2}}(Ii^tg^1M_0g^1)P_\sigma (A)\left[A^1[(qgq_t)+i(^tg^1p_tp)]\right]^\sigma (gq_tq)^\eta .$$
A tiresome but straightforward computation gives from (3.13) and (3.12):
$$\phi _E=\phi _1+i\phi _2.$$
(3.16)
$$\{\begin{array}{c}\phi _1(t,\alpha ,g):=(EH(\alpha ))t+\frac{1}{2}<M(g)^1\alpha ,J\alpha >\frac{1}{2}_0^t(\alpha _tM(g^1)\alpha )J\dot{\alpha _s}๐s\hfill \\ \phi _2(t,\alpha ,g):=\frac{i}{4}<(I\widehat{W_t})(M(g)\alpha _t\alpha );(M(g)\alpha _t\alpha )>.\hfill \end{array}$$
where $`\widehat{W_t}:=\left(\begin{array}{cc}W_t& iW_t\\ iW_t& W_t\end{array}\right)`$ with $`\frac{1}{2}(I+W_t):=(Ii^tg^1M_0g^1)^1`$.
###### Lemma 3.3.
We have: $`W_t_{(^d)}<1.`$
Proof: we introduce the Siegel half-plane:
$$\mathrm{\Sigma }_d:=\{ZM_d():^tZ=Z,\text{ and }\mathrm{}Z\text{ is positive definite }\}.$$
We know from pp.202, 203 that if $`Z\mathrm{\Sigma }_d`$, then $`(IiZ)^1(I+iZ)_{(^d)}<1`$. Now, we can take $`Z=^tg^1M_0g^1`$. Indeed $`M_0`$ is symmetric, and, since $`F_\alpha (t)`$ is symplectic, we have:
$$X^d,\mathrm{}(^tX.M_0X)=|(A+iB)^1X|_^d^2.$$
Thus $`Z\mathrm{\Sigma }_d`$. The proof is clear if we note that $`(Ii^tg^1M_0g^1)^1(I+i^tg^1M_0g^1)=W_t`$.$`\mathrm{}`$
We are led to solve a stationary phase problem to get an expansion of each $`L_{\eta ,\nu ,\mu ,\gamma ,j}(h)`$ in powers of $`h`$.
Remark: Note that the term $`D_{\sigma ,\eta ,\nu ,j}`$ โ (3.15) โ and its derivatives will be vanishing on the critical set of the phase for derivatives up to $`|\sigma |+|\eta |`$ (see (3.15) and (4.1)). Therefore, the asymptotic of $`__t_{_\alpha ^{2d}}\mathrm{}๐t๐\alpha `$ will be shifted of $`h`$ to the power $`\frac{1}{2}(|\sigma |+|\eta |)`$. This fact compensates for the term in $`h^{\frac{1}{2}(|\sigma |+|\eta |)}`$, at the beginning of the expression of $`L_{\eta ,\nu ,\mu ,\gamma ,j}(h)`$ in (3.14).
## 4. The stationary phase problem
Now, we fix $`g`$ in $`G`$ and we want to find the conditions under which we will be able to apply the stationary phase theorem under the form of (Theorem 3.3) on $`L_{\eta ,\nu ,\mu ,\gamma ,j}(h)`$. A necessary and sufficient condition will be called โ$`g`$-clean flowโ. Then we will give particular cases for which this criterium is satisfied (see sections 4.2 and 4.3). Our method will first consist in calculating the critical set of the phase $`\phi _E`$ and its Hessian. Then we will calculate the kernel of this Hessian, and, under assumption of smoothness of the critical set, we will describe the conditions for this kernel to be equal to the tangent space of the critical set. In this section, since $`g`$ is fixed in $`G`$, we will denote $`\phi _E(t,z,g)`$ by $`\phi _{E,g}(t,z)`$, for $`z^{2d}`$ and $`t`$.
### 4.1. Computations and $`g`$-clean flow
$``$ Computation of the critical set
$$\text{Let }๐_{E,g}:=\{a\times ^{2d}:\mathrm{}(\phi _{E,g}(a))=0,\phi _{E,g}(a)=0\}.$$
###### Proposition 4.1.
The critical set is:
(4.1)
$$๐_{E,g}=\{(t,z)\times ^{2d}:z\mathrm{\Sigma }_E,M(g)\mathrm{\Phi }_t(z)=z\}.$$
where $`(t,z)\mathrm{\Phi }_t(z)`$ is the flow of the system (1.2).
Proof :
$$\mathrm{}\phi _E(t,z,g)=\mathrm{}\phi _2(t,z,g)=\frac{1}{4}|z_tM(g^1)z|^2\frac{1}{4}\mathrm{}<\widehat{W_t}(M(g)z_tz);M(g)z_tz>_{^{2d}}.$$
We note that, if $`a`$ and $`b`$ are in $`^d`$, then:
$$<\widehat{W_t}(a,b);(a,b)>_{^{2d}}=<W_t(aib);(aib)>_^d.$$
Thus,
$$\mathrm{}\phi _E(t,z,g)=0|z_tM(g^1)z|^2=\mathrm{}<W_t\beta ,\beta >_^d=\mathrm{}<W_t\beta ,\overline{\beta }>_^d,$$
where
$$\beta :=(gq_tq)i(^tg^1p_tp).$$
Therefore, by lemma 3.3, we have: $`\mathrm{}\phi _E(t,z,g)=0\mathrm{\Phi }_t(z)=M(g^1)z`$.
Computation of the gradient of $`\phi _1`$ :
$`\{\begin{array}{c}_t\phi _1(t,z,g)=EH(z)\frac{1}{2}<(z_tM(g^1)z);J\dot{z}_t>\\ _z\phi _1(t,z,g)=\frac{1}{2}(^tM(g^1)+^tF_z(t))J(z_tM(g^1)z)\end{array}`$
Computation of the gradient of $`\phi _2`$ :
$`\{\begin{array}{c}4_t\phi _2(t,z,g)=2<(I\widehat{W_t})(M(g)z_tz);M(g)\dot{z}_t><_t(\widehat{W_t})(M(g)z_tz);(M(g)z_tz)>\\ 4_z\phi _2(t,z,g)=2(^tF_z(t)^tM(g)I)(I\widehat{W_t})(M(g)z_tz)^t[_z(\widehat{W_t})(M(g)z_tz)](M(g)z_tz)\end{array}`$
Thus, we see that $`(t,z,g)๐_{E,g}`$ if and only if $`\mathrm{\Phi }_t(z)=M(g^1)z`$ et $`H(z)=E`$. $`\mathrm{}`$
$``$ Computation of the Hessian $`\text{Hess}\phi _{E,g}(t,z)`$
We first need some formulae coming from the symmetry that will be helpful for the computation: We recall that $`F_z(t)=_z(\mathrm{\Phi }_t(z))`$. By differentiating formula (1.1), we get:
(4.2)
$$H(M(g)z)=^tM(g^1)H(z),z^{2d},gG.$$
This formula implies that we have also:
(4.3)
$$\mathrm{\Phi }_t(M(g)z)=M(g)\mathrm{\Phi }_t(z),z^{2d},gG,t\text{ such that the flow exists at time }t.$$
Moreover we recall that, since $`M(g)`$ is symplectic, we have:
(4.4)
$$JM(g)=^tM(g^1)J\text{ and }M(g)J=J^tM(g^1).$$
Finally, if $`t`$ and $`z`$ are such that $`M(g)\mathrm{\Phi }_t(z)=z`$, then we have:
(4.5)
$$(M(g)F_z(t)I)JH(z)=0\text{ and }(^tF_z(t)^tM(g)I)H(z)=0.$$
The second identity comes from the first since $`M(g)F_z(t)`$ is symplectic. For this first relation, one can differentiate at $`s=t`$ the equation:
$$\mathrm{\Phi }_t(M(g)\mathrm{\Phi }_s(z))=\mathrm{\Phi }_s(z).$$
With these formulae, it is easy to find that:
###### Proposition 4.2.
$`\text{Hess}\phi _{E,g}(t,z)=`$
$$\left(\begin{array}{cc}\frac{i}{2}<(I\widehat{W_t})JH(z);JH(z)>& ^tH(z)\\ & +\frac{i}{2}^t[(^tF_z(t)^tM(g)I)(I\widehat{W_t})JH(z)]\\ & \\ H(z)& \frac{1}{2}[JM(g)F_z(t)^t(M(g)F_z(t))J]\\ +\frac{i}{2}(^tF_z(t)^tM(g)I)(I\widehat{W_t})JH(z)& +\frac{i}{2}(^tF_z(t)^tM(g)I)(I\widehat{W_t})(M(g)F_z(t)I)\end{array}\right).$$
$``$ Computation of the real kernel of the Hessian
If $`AM_n()`$, then we define $`\mathrm{ker}_{_{}}(A):=\{x^n:A(x)=0\}=\mathrm{ker}(\mathrm{}(A))\mathrm{ker}(\mathrm{}(A))`$.
###### Proposition 4.3.
Let $`(t,z)๐_{E,g}`$. Then the real kernel of the Hessian is : $`\mathrm{ker}_{_{}}\text{Hess}\phi _{E,g}(t,z)=`$
(4.6)
$$\{(\tau ,\alpha )\times ^{2d}:\alpha H(z),\tau JH(z)+(M(g)F_z(t)Id)\alpha =0\}.$$
Proof : Let $`\tau `$ and $`\alpha ^{2d}`$. We set:
$$x:=\tau JH(z)+(M(g)F_z(t)I)\alpha .$$
Let us denote by $`\widehat{W_1}`$ and $`\widehat{W_2}`$ the real and imaginary part of $`\widehat{W_t}`$. Then, $`(\tau ,\alpha )\mathrm{ker}_{_{}}\text{Hess}\phi _{E,g}(t,z)`$ if and only if:
(4.7)
$$<\widehat{W_2}JH(z);x>=2<H(z);\alpha >.$$
(4.8)
$$<(I\widehat{W_1})JH(z);x>=0.$$
(4.9)
$$(^tF_z(t)^tM(g)I)(I\widehat{W_1})x=0.$$
and
$$2\tau H(z)+[JM(g)F_z(t)^t(M(g)F_z(t))J]\alpha +(^tF_z(t)^tM(g)I)\widehat{W_2}x=0.$$
We multiply this last identity by $`(M(g)F_z(t))J`$, we note that $`\widehat{W_2}=J\widehat{W_1}`$ and recall that $`M(g)F_z(t)`$ is symplectic to obtain the equivalent identity:
(4.10)
$$(M(g)F_z(t)I)(\widehat{W_1}I)x=2x.$$
Now, if $`(\tau ,\alpha )\mathrm{ker}_{_{}}\text{Hess}\phi _{E,g}(t,z)`$, then, by (4.10) and (4.9), we have:
$$<x,(I\widehat{W_1})x>=0,\text{ i.e. }|x|^2=<\widehat{W_1}x,x>.$$
By lemma 3.3, $`\widehat{W_1}_{(^{2d})}<1`$, thus $`x=0`$, and by (4.7), $`H(z)\alpha `$.
Conversely, if $`x=0`$ and $`H(z)\alpha `$, then, we have (4.7), (4.8), (4.9) and (4.10). Thus $`(\tau ,\alpha )\mathrm{ker}_{_{}}\text{Hess}\phi _{E,g}(t,z)`$. $`\mathrm{}`$
We are now able to describe the conditions under which we can apply the generalised stationary phase theorem on $`L_{\eta ,\nu ,\mu ,\gamma ,j}(h)`$: we easily check the positivity of the imaginary part of the phase $`\phi _{E,g}`$ by lemma 3.3. Moreover, if $`๐_{E,g}`$ is a union of smooth submanifolds of $`\times ^{2d}`$, if $`X๐_{E,g}`$, then the Hessian of $`\phi _{E,g}(X)`$ is non-degenerate on the normal space $`N_X๐_{E,g}`$ if and only if $`\mathrm{ker}_{_{}}\text{Hess}\phi _{E,g}(X)T_X๐_{E,g}`$, the tangent space of $`๐_{E,g}`$ at $`X`$. Besides, note that, by the non-stationary phase theorem, we can restrict this hypothesis to points $`X`$ in $`\text{Supp}(\widehat{f})\times \text{Supp}(a_{j_0})`$.
Definition: let $`gG`$, $`T>0`$, such that $`\text{Supp}(\widehat{f})]T,T[`$, and $`\mathrm{\Psi }_g:=\{\begin{array}{c}]T,T[\times \mathrm{\Sigma }_E^{2d}\hfill \\ (t,z)M(g)\mathrm{\Phi }_t(z)z\hfill \end{array}`$ We say that โthe flow is $`g`$-clean on $`]T,T[\times \mathrm{\Sigma }_E`$โ if zero is a weakly regular value of $`\mathrm{\Psi }`$, i.e. :
* $`\mathrm{\Psi }_g^1(\{0\})=:๐_{E,g}`$ is a finite union of smooth submanifolds of $`\times ^{2d}`$.
* $`(t,z)๐_{E,g}`$, $`T_{(t,z)}๐_{E,g}=\mathrm{ker}d_{(t,z)}\mathrm{\Psi }_g`$.
We say that โthe flow is $`G`$-clean on $`]T,T[\times \mathrm{\Sigma }_E`$โ if it is $`g`$-clean for all $`g`$ in $`G`$.
By proposition 4.3, we see that if $`(t,z)๐_{E,g}`$, then $`\mathrm{ker}d_{(t,z)}\mathrm{\Psi }_g=\mathrm{ker}_{_{}}\text{Hess}\phi _{E,g}(t,z)`$. Thus, if we only know that the support of $`\widehat{f}`$ is in $`]T,T[`$, then the $`g`$-clean flow condition is the minimal hypothesis under which we can apply the stationary phase theorem to $`L_{\eta ,\nu ,\mu ,\gamma ,j}(h)`$. Therefore, we can state the theorem:
###### Theorem 4.4.
Reduced trace formula with $`G`$-clean flow.
Let $`G`$ be a finite subgroup of $`Gl(,d)`$ and $`H:^{2d}`$ a smooth Hamiltonian $`G`$-invariant. Suppose that $`E`$ is such that there exists $`\delta E>0`$ such that $`H^1([E\delta E,E+\delta E])`$ is compact, and $`\mathrm{\Sigma }_E=\{H=E\}`$ has no critical points. Make hypothesis (3.5). Let $`f`$ and $`\psi `$ be real functions in $`๐ฎ()`$ such that $`\text{Supp}(\psi )]E\delta E,E+\delta E[`$ and $`\widehat{f}`$ is compactly supported in $`]T,T[`$, where $`T>0`$. Suppose that the flow is $`G`$-clean on $`]T,T[\times \mathrm{\Sigma }_E`$. Then the spectral density
$$๐ข_\chi (h)=\frac{d_\chi }{|G|}\underset{gG}{}\overline{\chi (g)}I_{g,E}(h)see(\text{1.7}),(\text{3.1})$$
has a complete asymptotic expansion as $`h0^+`$. Moreover, if $`gG`$, and, if $`[๐_{E,g}]`$ denotes the set of connected components of $`๐_{E,g}`$, then the quantity $`_0^tp_s\dot{q_s}๐s`$ is constant on each element $`Y`$ of $`[๐_{E,g}]`$, denoted by $`S_{Y,g}`$, and we have the following expansion:
$$I_{g,E}(h)=\underset{Y[๐_{E,g}]}{}(2\pi h)^{\frac{1dimY}{2}}e^{\frac{i}{h}S_{Y,g}}\frac{1}{2\pi }\left(_Y\widehat{f}(t)\psi (E)d_g(t,z)๐\sigma _Y(t,z)+\underset{j1}{}h^ja_{j,Y}\right)+O(h^+\mathrm{})$$
where $`a_{j,Y}`$ are distributions in $`\widehat{f}(\psi H)`$ with support in $`Y`$, and the density $`d_g(t,z)`$ is defined by:
(4.11)
$$d_g(t,z):=\text{det}_+^{\frac{1}{2}}\left(\frac{\phi _{E,g}^{\prime \prime }(t,z)_{|_{๐ฉ_{(t,z)}Y}}}{i}\right)\text{det}_+^{\frac{1}{2}}\left(\frac{A+iBi(C+iD)}{2}\right).$$
$`\phi _{E,g}`$ is given by (3.16) and $`A,B,C,D`$ are the $`d\times d`$ blocs forming the matrix $`F_z(t):=_z(\mathrm{\Phi }_t(z))`$ (see (5.7)).
Remark: without symmetry, this theorem can be compared to articles of T.Paul and A.Uribe (cf and ) or to the Gutzwiller formula in the PhD. thesis of S.Dozias (), see also . A notion of clean flow is also present in . The density $`d_g(t,z)`$ is difficult to compute in general, even without symmetry. The purpose of next sections is to calculate it in two special cases: when $`\widehat{f}`$ is supported near zero (Weyl part), and under an assumption of non-degenerate periodic orbits of the classical flow in $`\mathrm{\Sigma }_E`$ (oscillating or Gutzwiller part).
Proof : as we have seen before, we can apply the stationary phase theorem on each $`L_{\eta ,\nu ,\mu ,\gamma ,j}(h)`$, which gives an expansion of each $`I_g^{j_0}(h)`$ and each $`I_g(h)`$. The first term is given by:
$$I_g(h)\underset{h0^+}{}\frac{(2\pi h)^d}{2\pi }__t_{_\alpha ^{2d}}\chi _2(\alpha )\widehat{f}(t)\psi (H(\alpha ))\text{det}_+^{\frac{1}{2}}\left(\frac{A+iBi(C+iD)}{2}\right)e^{\frac{i}{h}\phi _{E,g}(t,\alpha )}๐t๐\alpha .$$
By definition of $`๐_{E,g}`$, $`\phi _{E,g}`$ is constant on each connected component of $`๐_{E,g}`$, equal to:
$$\phi _{E,g}(t,\alpha )=S(\alpha ,t)+Et=_0^tp_s\dot{q_s}๐s,\text{ where }(q_s,p_s)=\mathrm{\Phi }_s(\alpha ).$$
This ends the proof of theorem 4.4. $`\mathrm{}`$
### 4.2. The Weyl part
We now deal with one case which leads to an asymptotic expansion at the first order of the counting function of $`\widehat{H}_\chi `$ in an interval of $``$. Fix $`g`$ in $`G`$ and define:
(4.12)
$$_{E,g}:=\{t:z\mathrm{\Sigma }_E:M(g)\mathrm{\Phi }_t(z)=z\}.$$
###### Theorem 4.5.
Let $`G`$ be a finite subgroup of $`Gl(,d)`$ and $`H:^{2d}`$ a smooth $`G`$-invariant Hamiltonian. Let $`E`$ be such that $`H^1([E\delta E,E+\delta E])`$ is compact for some $`\delta E>0`$, and that $`\mathrm{\Sigma }_E=\{H=E\}`$ has no critical points. Make hypothesis (3.5). Let $`f`$ and $`\psi `$ be real functions in $`๐ฎ()`$ with $`\text{Supp}(\psi )]E\delta E,E+\delta E[`$ and $`\widehat{f}`$ is compactly supported. For $`g`$ in $`G`$, we set:
$$\nu _g:=dim\mathrm{ker}(gId_^d),F_g:=\mathrm{ker}(M(g)Id_{^{2d}})\text{ and }\stackrel{~}{F}_g:=\mathrm{ker}(gId_^d).$$
Set $`I_g(h):=Tr\left(\psi (\widehat{H})f\left(\frac{E\widehat{H}}{h}\right)\stackrel{~}{M}(g)\right)`$. Then, under previous assumptions, we have:
โ If $`\text{Supp}\widehat{f}_{E,g}=\mathrm{}`$, then $`I_{g,E}(h)=O(h^+\mathrm{})`$.
โ If $`\text{Supp}\widehat{f}_{E,g}=\{0\}`$ then we have the following expansion modulo $`O(h^+\mathrm{})`$ :
(4.13)
$$I_{g,\lambda }(h)h^{1\nu _g}\underset{k0}{}c_k(\widehat{f},g)h^k,\text{ as }h0^+.$$
uniformly in $`\lambda `$ in a small neighborhood of $`E`$, where $`c_k(\widehat{f},g)`$ are distributions in $`\widehat{f}`$ with support in $`\{0\}`$, and, if $`d(\mathrm{\Sigma }_\lambda F_g)`$ denotes the euclidian measure on $`\mathrm{\Sigma }_\lambda F_g`$, then we have:
(4.14)
$$c_0(\widehat{f},g)=\psi (\lambda )\widehat{f}(0)\frac{(2\pi )^{\nu _g}}{det((Id_^dg)|_{\stackrel{~}{F}_g^{}})}_{\mathrm{\Sigma }_\lambda F_g}\frac{d(\mathrm{\Sigma }_\lambda F_g)(z)}{|H(z)|}.$$
Remark 1: the oscillating term of Theorem 4.4 is now vanishing, since, for $`gG`$, $`S_{Y,g}=0`$ when $`Y=\{0\}\times (\mathrm{\Sigma }_EF_g)`$. Moreover, it is easy to see that, since $`\mathrm{\Sigma }_E`$ is compact and non-critical, zero is isolated in $`_{E,g}`$. Thus the hypothesis $`\text{Supp}\widehat{f}_{E,g}=\{0\}`$ is fulfilled if $`\widehat{f}`$ is supported close enough to zero.
Remark 2: we slightly precised the previous result of Z. El Houakmi given in , by the computation of (4.14). Note that the leading term of $`๐ข_\chi (h)`$ is obtained for $`g=Id`$, and:
$$๐ข_\chi (h)=\frac{d_\chi ^2}{|G|}\psi (E)\widehat{f}(0)(2\pi h)^{1d}\frac{1}{2\pi }_{\mathrm{\Sigma }_E}\frac{d\mathrm{\Sigma }_E}{|H|}+O(h^{2d}),\text{ as }h0^+.$$
Proof : If $`\text{Supp}\widehat{f}_{E,g}=\mathrm{}`$, then $`(\text{Supp}(\widehat{f})\times ^{2d})๐_{E,g}=\mathrm{}`$, and by the non stationary phase theorem, we get the result.
Now suppose that $`\text{Supp}\widehat{f}_{E,g}=\{0\}`$. Then we have:
(4.15)
$$๐_{E,g}(\text{Supp}(\widehat{f})\times ^{2d})=\{0\}\times (\mathrm{\Sigma }_EF_g).$$
We now give some โtrickโ to boil down to the case where $`G`$ is composed of isometries. We recall that, since $`G`$ is compact, there is some $`S_0`$, symmetric $`d\times d`$ positive definite matrix, such that:
(4.16)
$$G_0:=S_0^1GS_0\text{ is a subgroup of the orthogonal group }O(d,).$$
One can indeed classicaly find a scalar product invariant by $`G`$ by averaging with the Haar measure of $`G`$. Thus, we can define a new $`G_0`$-invariant Hamiltonian:
$$H_0(z):=H(M(S_0)z),whereM(S_0):=\left(\begin{array}{cc}S_0& 0\\ 0& {}_{}{}^{t}S_{0}^{1}\end{array}\right).$$
If $`\chi \widehat{G}`$, then one can define $`\chi _0:G_0`$ by:
$$\chi _0(g_0):=\chi (S_0g_0S_0^1).$$
Then it is easy to check that $`\chi _0\widehat{G}_0`$ and that the application $`\chi \chi _0`$ is bijective from $`\widehat{G}`$ to $`\widehat{G}_0`$. Moreover, identity (2.2) implies that:
$$Op_h^w(H_0)=\stackrel{~}{M}(S_0)^1Op_h^w(H)\stackrel{~}{M}(S_0).$$
If $`\chi \widehat{G}`$, then we can define:
$$\stackrel{~}{P}_{\chi _0}:=\frac{d_{\chi _0}}{|G_0|}\underset{g_0G_0}{}\overline{\chi _0(g_0)}\stackrel{~}{M}(g_0).$$
Then we have $`\stackrel{~}{P}_{\chi _0}=\stackrel{~}{M}(S_0)^1P_\chi \stackrel{~}{M}(S_0)`$. Therefore, if $`f(\widehat{H})`$ is trace class, then $`f(\widehat{H}_0)`$ also, and we have:
$$\text{Tr}(f(\widehat{H}_\chi ))=\text{Tr}(f(\widehat{H})P_\chi )=\text{Tr}(f(\widehat{H}_0)\stackrel{~}{P}_{\chi _0})),$$
by cyclicity of trace. This remark apply in particular for the trace (1.7). Moreover, if $`gG`$, if $`g_0:=S_0^1gS_0`$, then $`\text{Tr}(f(\widehat{H})\stackrel{~}{M}(g))=\text{Tr}(f(\widehat{H}_0)\stackrel{~}{M}(g_0))`$. Finally, it is easy to check that hypotheses for ($`H`$, $`G`$) are available for ($`H_0`$, $`G_0`$), and that coefficients of the asymptotic have the same expression in terms of ($`H_0`$, $`G_0`$) as in ($`H`$, $`G`$). $`\mathrm{}`$
From now on, we suppose that $`G`$ is made of isometries, without loss of generality.
First, we remark that $`\mathrm{\Sigma }_E`$ and $`F_g`$ are transverse submanifolds of $`^{2d}`$. Indeed, if $`z\mathrm{\Sigma }_EF_g`$, then, by (4.2), since $`g`$ is an isometry, we have $`H(z)F_g`$, thus $`F_g+[H(z)]^{}=^{2d}`$. Therefore
$$๐ฏ_{(0,z)}๐_{E,g}=\{0\}\times [F_g[H(z)]^{}].$$
If $`(\tau ,\alpha )\mathrm{ker}_{_{}}\text{Hess}\phi _E(0,z)`$ then by Proposition 4.3, $`\tau JH(z)+(M(g)I_{2d})\alpha =0`$. Then one can take the scalar product of this equality with $`JH(z)`$ to obtain $`\tau =0`$ and thus, $`\mathrm{ker}_{}\text{Hess}\phi _E(0,z)=๐ฏ_{(0,z)}๐_{E,g}`$. This means that we have the theorical asymptotic expansion of Theorem 4.5.
Now, we have to compute the leading term of this expansion. Here again, we can suppose that $`g`$ is an isometry, which simplifies the calculus: in particular, $`[M(g),J]=0`$, when $`t=0`$, we have $`\widehat{W_t}=0`$, and $`F_z(0)=Id`$. By Proposition 4.2, we obtain:
$$\text{Hess}\phi _{E,g}(0,z)=\left(\begin{array}{cc}\frac{i}{2}\left|H\left(z\right)\right|^2& ^tH\left(z\right)\\ & \\ H\left(z\right)& \frac{1}{2}J\left(M\left(g\right)M\left(g^1\right)\right)+\frac{i}{2}\left(IM\left(g\right)\right)\left(IM\left(g^1\right)\right)\end{array}\right).$$
We have $`๐ฉ_{(0,z)}๐_{E,g}=\times [F_g^{}+H(z)]`$. Let $`\beta _0`$ be a basis of $`F_g^{}`$. We set:
$$e_0:=\frac{}{t}=(1,0),\epsilon _0:=(0,H(z)).$$
Let $`\beta `$ be the basis of $`๐ฉ_{(0,z)}๐_{E,g}`$ made up of (in this order) $`e_0`$, $`\epsilon _0`$ and $`\beta _0`$. We note that the linear application $`\frac{1}{2}J(M(g)M(g^1))+\frac{i}{2}(IM(g))(IM(g^1))`$ stabilizes the space $`F_g^{}`$. Then by calculating the determinant of the restriction of $`\text{Hess}\phi _E(0,z)`$ to $`๐ฉ_{(0,z)}๐_{E,g}`$ in this basis, we get (noting $`๐ฉ:=๐ฉ_{(0,z)}๐_{E,g}`$):
$$det\left(\frac{\phi _{E,g}^{\prime \prime }(0,z)_{|_๐ฉ}}{i}\right)=|H(z)|^2det\left[\frac{1}{2i}J(M(g)M(g^1))+\frac{1}{2}(IM(g))(IM(g^1))\right]_{|_{F_g^{}}}$$
If $`\mathrm{\Pi }_g`$ is the orthogonal projector on $`\stackrel{~}{F}_g`$, then we have:
$$\frac{1}{|H(z)|^2}det\left(\frac{\phi _{E,g}^{\prime \prime }(0,z)_{|_๐ฉ}}{i}\right)=\left(\begin{array}{cc}\frac{1}{2}(I_dg)(I_dg^1)+\mathrm{\Pi }_g& \frac{1}{2i}(gg^1)\\ & \\ \frac{1}{2i}(gg^1)& \frac{1}{2}(I_dg)(I_dg^1)+\mathrm{\Pi }_g\end{array}\right).$$
Then, since $`g`$ is an isometry, we can suppose that $`g`$ is bloc diagonal with blocs $`I_{p_1}`$, $`I_{p_2}`$, $`R_{\theta _1},\mathrm{},R_{\theta _r}`$, where $`p_1+p_2+2r=d`$, $`\theta _j`$โs are not in $`\pi `$, and $`R_\theta :=\left(\begin{array}{cc}\mathrm{cos}\theta & \mathrm{sin}\theta \\ \mathrm{sin}\theta & \mathrm{cos}\theta \end{array}\right).`$ We then use the fact that $`g`$ commutes with $`\mathrm{\Pi }_g`$, and that when $`[C,D]=0`$, then $`det`$$`\left(\begin{array}{cc}A& B\\ C& D\end{array}\right)`$ $`=det(ADBC)`$, for any blocs $`A`$, $`B`$, $`C`$, $`D`$ of same size. A straightforward calculus then gives (see for details):
$$det\left(\frac{\phi _{E,g}^{\prime \prime }(0,z)_{|_{๐ฉ_{(0,z)}๐_{E,g}}}}{i}\right)=|H(z)|^2det\left[(I_dg)_{|_{\stackrel{~}{F}_g^{}}}\right]^2.$$
Since $`\left[\text{det}_+^{\frac{1}{2}}\right]^2=det`$, we have:
$$\text{det}_+^{\frac{1}{2}}\left(\frac{\phi _{E,g}^{\prime \prime }(0,z)_{|_{๐ฉ_{(0,z)}๐_{E,g}}}}{i}\right)=\pm |H(z)||det(I_dg)_{|_{\stackrel{~}{F}_g^{}}}|.$$
We can proove that the factor $`\pm 1`$ is in fact equal to $`1`$, either by coming back to the calculus of $`\text{det}_+^{\frac{1}{2}}`$ with gaussians, or, classically, by using a weak asymptotic, i.e. by calculating the asymptotic of $`\text{Tr}(\phi (\widehat{H})M(g))`$, when $`\phi :`$ is smooth and $`\phi (\widehat{H})`$ is trace class. See for details.
Using (4.11), the fact that the phase vanishes on $`๐_{E,g}`$, and that $`dim(๐_{E,g})(\text{Supp}(\widehat{f})\times ^{2d})=2\nu _g1`$, we obtain the result we claimed. This ends the proof of Theorem 4.5. $`\mathrm{}`$
As a consequence of Theorem 4.5 near $`t=0`$, using a well known Tauberian argument (see ), we get the following:
###### Corollary 4.6.
Let $`G`$ be a finite group of $`Gl(d,)`$, $`H:^{2d}`$ a $`G`$-invariant smooth Hamiltonian satisfying (3.5). Let $`E_1<E_2`$ in $``$, and $`I:=[E_1,E_2]`$. Suppose that there exists $`\epsilon >0`$ such that $`H^1([E_1\epsilon ,E_2+\epsilon ])`$ is compact. Furthermore suppose that $`E_1`$ and $`E_2`$ are not critical values of $`H`$. If $`\chi \widehat{G}`$, then the spectrum of $`\widehat{H}_\chi `$ is discrete in $`I`$, and we have:
$$N_{I,\chi }(h)=\frac{d_\chi ^2}{|G|}(2\pi h)^d\text{Vol}[H^1(I)]+O(h^{1d}),$$
where $`N_{I,\chi }(h)`$ is the number of eigenvalues of $`\widehat{H}_\chi `$ in $`I`$ counted with multiplicity.
Remark: One can interpret this result by saying that, semi-classically, the proportion of eigenfunctions of $`\widehat{H}`$ having symmetry $`\chi `$ is $`\frac{d_\chi ^2}{|G|}`$. In particular, the same proportion of eigenvalues has multiplicity greater than $`d_\chi `$. The more $`d_\chi `$ is high, the more $`L_\chi ^2(^d)`$ takes part in the spectrum of $`\widehat{H}`$.
### 4.3. The oscillatory part
If $`gG`$ and $`\gamma `$ is a periodic orbit of $`\mathrm{\Sigma }_E`$ globally stable by $`M(g)`$, we set :
$$_{g,\gamma }:=\{t\text{Supp}\widehat{f}:z\gamma :M(g)\mathrm{\Phi }_t(z)=z\}.$$
If $`t_0_{g,\gamma }`$, $`z\gamma `$, then $`P_{\gamma ,g,t_0}`$ denotes the Poincarรฉ map of $`\gamma `$ between $`z`$ and $`M(g^1)z`$ at time $`t_0`$, restricted to $`\mathrm{\Sigma }_E`$. The characteristic polynomial of $`dP_{\gamma ,g,t_0}`$ doesnโt depend on $`z\gamma `$. Note that, by iterating formula (4.3), since $`G`$ is finite, if we have $`M(g)\mathrm{\Phi }_t(z)=z`$, then $`z`$ is a periodic point of the Hamiltonian system (1.2).
###### Theorem 4.7.
Make the same assumptions as in Theorem 4.5, but suppose that $`0\text{Supp}\widehat{f}`$. Make the following hypothesis of non-degeneracy : if $`\gamma \mathrm{\Sigma }_E`$, is such that $`gG`$ and $`t_0_{g,\gamma }`$, $`t_00`$, then $`1`$ is not an eigenvalue of $`M(g)dP_{\gamma ,g,t_0}`$. Then the set of such $`\gamma `$โs is finite and the following expansion holds true modulo $`O(h^+\mathrm{})`$, as $`h0^+`$ :
$$๐ข_\chi (h)\frac{d_\chi }{|G|}\underset{\begin{array}{c}\gamma \text{ periodic }\\ \text{orbit of }\mathrm{\Sigma }_E\end{array}}{}\underset{\begin{array}{c}gG\text{ s.t. }\\ M\left(g\right)\gamma =\gamma \end{array}}{}\overline{\chi (g)}\underset{\begin{array}{c}t_0_{g,\gamma }\\ t_00\end{array}}{}e^{\frac{i}{h}S_\gamma (t_0)}\underset{k0}{}d_k^{\gamma ,g,t_0}(\widehat{f})h^k.$$
Terms $`d_k^{\gamma ,g,t_0}(\widehat{f})`$ are distributions in $`\widehat{f}`$ with support in $`\{t_0\}`$, $`S_\gamma (t_0):=_0^{t_0}p_s\dot{q}_s๐s`$, $`((q_s,p_s):=\mathrm{\Phi }_s(z)\text{ with }z\gamma )`$, and
$$d_0^{\gamma ,g,t_0}(\widehat{f})=\frac{\psi (E)T_\gamma ^{}e^{i\frac{\pi }{2}\sigma _\gamma (g,t_0)}}{2\pi |det(M(g)dP_{\gamma ,g,t_0}Id)|^{\frac{1}{2}}}\widehat{f}(t_0)$$
where $`T_\gamma ^{}`$ is the primitive period of $`\gamma `$ and $`\sigma _\gamma (g,t_0)`$.
Example 1: if $`d=1`$, periodic orbits are always non-degenerate. For example, in the case of a double well Schrรถdinger Hamiltonian, one can illustrate the sum of Theorem 4.7 on figure 1, picturing the classical flow in $`^2`$: some periodic orbits appear only for $`g=Id`$ in the sum, and others arise for both $`g=\pm Id`$. One can also fold the picture to compare with the periodic orbits of the reduced space as in Theorem 1.1.
Example 2: if $`H`$ is a Schrรถdinger operator on $`^d`$ with potential $`V(x)=<Sx,x>`$, where $`S`$ is the diagonal matrix with diagonal non-vanishing $`w_1^2,\mathrm{},w_d^2`$, if one assumes that $`ij`$, $`w_i/w_j`$, then periodic orbits appear as a union of $`d`$ plans, with primitive periods $`T_j^{}=\frac{\pi }{w_j}`$ and are all non-degenerate.
As a particular case of this theorem, we get the Theorem 1.1:
Proof of Theorem 1.1: if we suppose that $`G`$ acts freely on $`\mathrm{\Sigma }_E`$, then $`\mathrm{\Sigma }_E/G`$ inherits a structure of smooth manifold such that the canonical projection $`\pi :\mathrm{\Sigma }_E\mathrm{\Sigma }_E/G`$ is smooth, and the dynamical system restricted to $`\mathrm{\Sigma }_E`$ descends to quotient. If $`t_0^{}`$, $`gG`$ and $`z\mathrm{\Sigma }_E`$, with orbit $`\gamma `$, are such that $`M(g)\mathrm{\Phi }_{t_0}(z)=z`$, then $`\gamma `$ and $`\pi (\gamma )`$ are periodic. If $`P_{\pi (\gamma ),\pi (z)}(t_0)`$ denotes the Poincarรฉ map of $`\pi (\gamma )`$ at time $`t_0`$, then we have:
(4.17)
$$det(M(g)d_zP_{\gamma ,g,t_0}Id)=det(d_{\pi (z)}P_{\pi (\gamma ),\pi (z)}(t_0)Id).$$
Indeed, if $`\stackrel{~}{\mathrm{\Phi }}_t`$ denotes the flow in $`\mathrm{\Sigma }_E/G`$, then one can differentiate the following identity on $`\mathrm{\Sigma }_E`$ with variable $`z`$:
$$\pi (M(g)\mathrm{\Phi }_{t_0}(z))=\stackrel{~}{\mathrm{\Phi }}_{t_0}(\pi (z)),$$
to get the the identity:
$$d_z\pi M(g)F_z(t_0)=\stackrel{~}{F}_{\pi (z)}(t_0)d_z\pi ,$$
where $`\stackrel{~}{F}_{\pi (z)}(t_0)`$ is the differential of $`x\stackrel{~}{\mathrm{\Phi }}_{t_0}(x)`$ at $`\pi (z)`$. Moreover, $`\pi `$ is a submersion, and by a dimensional argument itโs also an immersion. Thus we have (4.17).
Therefore, if we make hypotheses of Theorem 1.1, then hypotheses of Theorem 4.7 are fulfilled. If $`z\mathrm{\Sigma }_E`$ is such that the orbit of $`\pi (z)`$ is periodic with period $`t_00`$, then there is only one $`g=g_\gamma G`$ such that $`M(g)\mathrm{\Phi }_{t_0}(z)=z`$. If $`_{red}`$ denotes the set of periods of $`\mathrm{\Sigma }_E/G`$, then we have:
$$\underset{\begin{array}{c}\gamma \text{ periodic }\\ \text{orbit of }\mathrm{\Sigma }_E\end{array}}{}\underset{\begin{array}{c}gG\text{ s.t. }\\ M\left(g\right)\gamma =\gamma \end{array}}{}\underset{\begin{array}{c}t_0_{g,\gamma }\\ t_00\end{array}}{}\mathrm{}=\underset{t_0_{red}}{}\underset{\begin{array}{c}\gamma \mathrm{\Sigma }_E:\pi (\gamma )\text{ }\text{periodic}\\ \text{ }\text{with }t_0\text{ for period }\end{array}}{}\underset{g=g_\gamma }{}\mathrm{}.$$
If we denote $`Stab(\gamma ):=\{gG:M(g)\gamma =\gamma \}`$, then we have $`Stab(\gamma )=<g_\gamma >`$ and it is easy to see that $`T_{\pi (\gamma )}^{}=\frac{T_\gamma ^{}}{|Stab(\gamma )|}`$. If we denote by $`N_{\pi (\gamma )}`$ the number of orbits of $`\mathrm{\Sigma }_E`$ with image $`\pi (\gamma )`$ by $`\pi `$, then we have $`N_{\pi (\gamma )}=|G|/|Stab(\gamma )|`$. Thus we have:
$$๐ข_\chi (h)=d_\chi \underset{t_0_{red}}{}\widehat{f}(t_0)\underset{\begin{array}{c}\gamma \mathrm{\Sigma }_E:\pi (\gamma )\text{ }\text{periodic}\\ \text{ }\text{with }t_0\text{ for period }\end{array}}{}\overline{\chi (g_{\pi (\gamma )}(t_0))}\frac{T_{\pi (\gamma )}^{}}{N_{\pi (\gamma )}}\frac{e^{\frac{i}{h}S_\gamma (t_0)}e^{i\frac{\pi }{2}\sigma _\gamma (g,t_0)}}{2\pi |det(d_{\pi (z)}P_{\pi (\gamma ),\pi (z)}(t_0)Id)|^{\frac{1}{2}}}+O(h).$$
Then one can show that quantities appearing in the r.h.s. donโt depend on $`\gamma `$ but only on $`\pi (\gamma )`$, and this proves the Theorem 1.1. $`\mathrm{}`$
Proof of the Theorem 4.7: We fix $`g`$ in $`G`$. If $`t_0^{}`$, we set:
$$\mathrm{\Gamma }_{E,g,t_0}:=\{\gamma \text{ orbit of }\mathrm{\Sigma }_E:z\gamma :M(g)\mathrm{\Phi }_{t_0}(z)=z\}.$$
###### Lemma 4.8.
If we make assumptions of non-degeneracy of Theorem 4.7, then $`_{E,g}\text{Supp}(\widehat{f})`$ is finite and we have:
(4.18)
$$๐_{E,g}(\text{Supp}(\widehat{f})\times ^{2d})=\underset{\begin{array}{c}t_0_{E,g}\\ t_00\end{array}}{}\underset{\gamma \mathrm{\Gamma }_{E,g,t_0}}{}\{t_0\}\times \gamma .$$
Proof : one can adapt the proof of the cylinder theorem of . For details, we refer to . $`\mathrm{}`$
Note that periodic orbits appearing in this critical set are the ones stable by $`g`$.
We see that $`๐_{E,g}(\text{Supp}(\widehat{f})\times ^{2d})`$ is a submanifold of $`\times ^{2d}`$ and if $`(t_0,z)๐_{E,g}`$, then we have:
$$T_{(t_0,z)}๐_{E,g}=\{0\}\times JH(z).$$
To apply the stationary phase theorem, we have to show that $`\mathrm{ker}_{_{}}\text{Hess}\phi _{E,g}(t_0,z)T_{(t_0,z)}๐_{E,g}`$. Let $`(\tau ,\alpha )\mathrm{ker}_{}\text{Hess}\phi _{E,g}(t_0,z)`$. By Proposition 4.3, we have $`\alpha H(z)`$ and:
(4.19)
$$\tau JH(z)+(M(g)F_z(t_0)I)\alpha =0.$$
If $`\lambda `$, we denote by $`E_\lambda :=_{k=1}^{2d}\mathrm{ker}(M(g)F_z(t_0)Id)^k`$. Let $`\gamma `$ be the orbit of $`z`$. Since $`1`$ is not an eigenvalue of $`M(g)dP_{\gamma ,g,t_0}`$, $`1`$ is an eigenvalue of $`M(g)F_z(t_0)`$ of multiplicity $`2`$. Thus $`dimE_1=2`$. Using (4.5) and (4.19), we have $`\alpha E_1`$. Let $`u_2^{2d}`$ such that $`(JH(z),u_2)`$ is a basis of $`E_1`$. Note that $`<u_2,H(z)>0`$, otherwise we would have $`u_2(JE_1)^{}`$, which is equal to $`\underset{\lambda 1}{}E_\lambda `$ since $`M(g)F_z(t_0)`$ is symplectic. Since $`\alpha E_1`$ we have $`\lambda _1,\lambda _2`$ in $``$ such that:
$$\alpha =\lambda _1JH(z)+\lambda _2u_2.$$
Then, using the fact that $`<\alpha ,H(z)>=0`$, we get $`\lambda _2=0`$ (since $`<u_2,H(z)>0`$). Thus coming back to (4.19), we get $`\tau =0`$ and $`\alpha JH(z)`$. Thus $`(\tau ,\alpha )T_{(t_0,z)}๐_{E,g}`$.
This shows that we can apply the stationary phase theorem and get a theorical expansion of $`I_g(h)`$ and $`๐ข_\chi (h)`$. We have now to compute the first term of this expansion. We suppose that $`(t_0,z)๐_{E,g}`$. We denote by $`\mathrm{\Pi }`$ the orthogonal projector on $`JH(z)`$. We set $`F:=M(g)F_z(t_0)`$ and $`W:=\widehat{W}_{t_0}`$. Then we have: $`det\left(\frac{\phi _{E,g}^{\prime \prime }(t_0,z)_{|_{๐ฉ_{(t_0,z)}๐_{E,g}}}}{i}\right)=`$
$$det\left(\begin{array}{cc}\frac{1}{2}<(IW)JH(z);JH(z)>& \frac{1}{i}^tH(z)\\ & +\frac{1}{2}^t[(^tFI)(IW)JH(z)]\\ & \\ \frac{1}{i}H(z)& \frac{1}{2i}[JF+^t(JF)]\\ +\frac{1}{2}(^tFI)(IW)JH(z)& +\frac{1}{2}(^tFI)(IW)(FI)+\mathrm{\Pi }\end{array}\right).$$
Since $`F`$ is symplectic, we have $`JF+^t(JF)=(^tF+I)J(FI).`$ Set:
(4.20)
$$K:=\frac{1}{2i}(^tF+I)J+\frac{1}{2}(^tFI)(IW).$$
Then, the forth bloc is equal to $`K(FI)+\mathrm{\Pi }`$.
Using (4.5), we note that the third bloc is equal to $`KJH(z)`$. Let us set:
(4.21)
$$X_1:=\frac{1}{2}(IW)JH(z).$$
We then have:
$$det\left(\frac{\phi _{E,g}^{\prime \prime }(t_0,z)_{|_{๐ฉ_{(t_0,z)}๐_{E,g}}}}{i}\right)=det\left(\begin{array}{cc}{}_{}{}^{t}X_{1}^{}JH(z)& i^tH(z)+^tX_1(FI)\\ & \\ KJH(z)& K(FI)+\mathrm{\Pi }\end{array}\right)$$
The following technical lemma is due to M. Combescure (see in the preprint version or p.87 for the proof):
###### Lemma 4.9.
$`K`$ is invertible and $`K^1=\frac{1}{2}[(FI)+i(F+I)J]`$.
Moreover, if we set $`F=\left(\begin{array}{cc}\stackrel{~}{A}& \stackrel{~}{B}\\ \stackrel{~}{C}& \stackrel{~}{D}\end{array}\right)`$, then $`det(K)=(1)^ddet(\frac{1}{2}(\stackrel{~}{A}+i\stackrel{~}{B}i(\stackrel{~}{C}+i\stackrel{~}{D})))^1.`$
Since
$$det\left(\frac{\phi _{E,g}^{\prime \prime }(t_0,z)_{|_{๐ฉ_{(t_0,z)}Y}}}{i}\right)=det\left(\begin{array}{cc}1& 0\\ & \\ 0& K\end{array}\right)\left(\begin{array}{cc}{}_{}{}^{t}X_{1}^{}JH(z)& i^tH(z)+^tX_1(FI)\\ & \\ JH(z)& (FI)+K^1\mathrm{\Pi }\end{array}\right),$$
using (4.11) and the preceeding lemma, we get:
(4.22)
$$d_g(t,z)^2=(1)^ddet(g^1)det\left(\begin{array}{cc}{}_{}{}^{t}X_{1}^{}JH(z)& i^tH(z)+^tX_1(FI)\\ & \\ JH(z)& (FI)+K^1\mathrm{\Pi }\end{array}\right).$$
We denote by $`\alpha :=<X_1,JH(z)>`$<sup>3</sup><sup>3</sup>3NB : $`\alpha 0`$ since $`IW`$ is invertible and $`JH(z)0`$. and we use the line operation $`L_2L_2\frac{1}{\alpha }JH(z)L_1`$, to get:
(4.23)
$$d_g(t,z)^2=(1)^d\alpha det(D)det(g^1).$$
where
$$D:=(FI)+K^1\mathrm{\Pi }\frac{1}{\alpha }JH(z)[i^tH(z)+^tX_1(FI)].$$
Then, we compute $`det(D)`$ in the basis $`\beta _0:=(v_1,\mathrm{},v_{2d})`$ where $`v_1:=JH(z)`$, $`v_2`$ is such that $`v_2JH(z)`$ and $`(v_1,v_2)`$ is a basis of $`\mathrm{ker}(FI)^2`$. Lastly $`(v_3,\mathrm{}v_{2d})`$ is a basis of $`V_z:=\underset{\lambda 1}{}E_\lambda `$.
Let us set $`w:=\frac{i}{2}(F+I)H(z)`$. We have $`Dv_1=w`$ and, using lemma 4.9:
(4.24)
$$((FI)+K^1\mathrm{\Pi })v_2=(FI)v_2.$$
(4.25)
$$\frac{1}{\alpha }JH(z)[i^tH(z)+^tX_1(FI)]v_2=\frac{1}{\alpha }(i<H(z),v_2>+<X_1,(FI)v_2>)JH(z).$$
Using the fact that $`(FI)v_2E_1`$, one easily gets that there exists $`\lambda _1`$ such that $`(FI)v_2=\lambda _1JH(z)`$. Thus $`<X_1,(FI)v_2>=\lambda _1\alpha .`$ We obtain, using (4.24) and (4.25):
(4.26)
$$Dv_2=\frac{i}{\alpha }<H(z),v_2>JH(z).$$
Note that $`(FI)V_zV_z`$. Moreover $`K^1\mathrm{\Pi }`$ is of rank $`1`$. Hence, since its image is equal to $`K^1\mathrm{\Pi }v_1=w0`$, we can neglect it on others columns than the first column. The same idea holds for $`\frac{1}{\alpha }JH(z)[i^tH(z)+^tX_1(FI)]`$, which we neglect in other columns than the second one (since $`\frac{1}{\alpha }JH(z)[i^tH(z)+^tX_1(FI)]v_20`$). Therefore:
$$det(D)=det\left(\begin{array}{ccc}w_1& \frac{i}{\alpha }<H(z),v_2>& 0\\ w_2& 0& \\ & & \\ w_3& 0& \\ \mathrm{}& \mathrm{}& (FI)_{|_{V_z}}\\ w_{2d}& 0& \end{array}\right)$$
where $`(w_1,\mathrm{},w_{2d})`$ are coordinates of $`w`$ in basis $`\beta _0`$.
Hence $`det(D)=\frac{i}{\alpha }w_2<H(z),v_2>det((FI)_{|_{V_z}})`$.
We write
$$w=\frac{i}{2}(F+I)H(z)=w_1JH(z)+w_2v_2+v$$
where $`vV_z`$, then we take the scalar product with $`H(z)`$. Since $`E_1=(JV_z)^{}`$, we have $`<v,H(z)>=0`$. and $`i|H(z)|^2=w_2<v_2;H(z)>.`$ Thus we get:
$$det(D)=\frac{1}{\alpha }|H(z)|^2det((FI)_{|_{V_z}}).$$
Therefore, according to (4.23)
(4.27)
$$d_g(t,z)^2=(1)^d|H(z)|^2det((FI)_{|_{V_z}})det(g^1).$$
Since $`det(g^1)=\pm 1`$, there exists $`k`$, depending on $`g`$, such that:
$$d_g(t,z)=\frac{e^{ik\frac{\pi }{2}}}{|H(z)||det((FI)_{|_{V_z}})|^{\frac{1}{2}}}.$$
Moreover, $`d_g`$ being continuous, $`k`$ doesnโt depend on $`z\gamma `$. Thus by Theorem 4.4, we have, if $`_{E,g}:=\{t:z\mathrm{\Sigma }_E:M(g)\mathrm{\Phi }_t(z)=z\}`$:
$$I_g(h)=\underset{t_0_{E,g}supp(\widehat{f})}{}\underset{\gamma \mathrm{\Gamma }_{E,g,t_0}}{}e^{\frac{i}{h}S_\gamma (t_0)}\frac{\psi (E)\widehat{f}(t_0)e^{ik\frac{\pi }{2}}}{2\pi |det((FI)_{|_{V_z}})|^{\frac{1}{2}}}_\gamma \frac{d\gamma }{|H|}+O(h).$$
Moreover, if $`z\gamma `$, then:
$$_\gamma \frac{d\gamma }{|H|}=_0^{T_\gamma ^{}}|JH(\varphi _t(z))|\frac{dt}{|H(\varphi _t(z))|}=T_\gamma ^{}.$$
Lastly, we sum on $`gG`$ to get the expansion of $`๐ข_\chi (h)`$. This ends the proof of Theorem 4.7. $`\mathrm{}`$
## 5. Appendix : Coherent states
We recall some basic things on coherent states on $`^{2d}`$ in Schrรถdinger representation. We mainly follow the presentation of M.Combescure and D.Robert (cf , ).
### 5.1. Notations
The $`h`$-scaling unitary operator $`\mathrm{\Lambda }_h:L^2(^d)L^2(^d)`$ is defined by:
$$\mathrm{\Lambda }_h\psi (x)=\frac{1}{h^{\frac{d}{4}}}\psi \left(\frac{x}{h^{\frac{1}{2}}}\right).$$
The phase translation unitary operator associated to $`\alpha =(q,p)=^d\times ^d`$ is given by: $`๐ฏ_h(\alpha ):=\mathrm{exp}[\frac{i}{h}(pxq.hD_x)]`$. We classically have $`๐ฏ_h(\alpha )^{}=๐ฏ_h(\alpha )^1=๐ฏ_h(\alpha )`$ and:
(5.1)
$$๐ฏ_h(\alpha )f(x)=\mathrm{exp}\left(i\frac{p}{h}(x\frac{q}{2})\right).f(xq).$$
The ground state of the harmonic oscillator $`\mathrm{\Delta }+|x|^2`$ is given by $`\stackrel{~}{\psi _0}(x):=\frac{1}{\pi ^{\frac{d}{4}}}\mathrm{exp}(\frac{|x|^2}{2}).`$
We set:
(5.2)
$$\psi _0(x):=\mathrm{\Lambda }_h\stackrel{~}{\psi _0}(x)=\frac{1}{(h\pi )^{\frac{d}{4}}}\mathrm{exp}(\frac{|x|^2}{2h}).$$
Then the coherent state associated to $`\alpha ^{2d}`$ is given by $`\overline{)\phi _\alpha }:=๐ฏ_h(\alpha )\psi _0.`$ By (5.1), we have:
(5.3)
$$\phi _\alpha (x)=\frac{1}{(h\pi )^{\frac{d}{4}}}\mathrm{exp}\left(i\frac{p}{h}(x\frac{q}{2})\right).\mathrm{exp}\left(\frac{|xq|^2}{2h}\right).$$
and we get easily from (5.1) the following formulae:
$$\mathrm{\Lambda }_h^{}๐ฏ_h(\alpha )\mathrm{\Lambda }_h=๐ฏ_1(\frac{\alpha }{\sqrt{h}})\text{ and }\mathrm{\Lambda }_h^{}Op_h^w(a)\mathrm{\Lambda }_h=Op_1^w(a_h),\text{ where }a_h(z):=a(\sqrt{h}z).$$
$$๐ฏ_h(\alpha )๐ฏ_h(\beta )=e^{\frac{i}{2h}<J\alpha ;\beta >}๐ฏ_h(\alpha +\beta )\text{ and }๐ฏ_h(\alpha )^{}Op_h^w(a)๐ฏ_h(\alpha )=Op_h^w[a(\alpha +.)].$$
### 5.2. A trace formula
If $`A(L^2(^d))`$ is trace class, then $`_{_\alpha ^{2d}}|<A\phi _\alpha ;\phi _\alpha >_{L^2(^d)}|d\alpha <+\mathrm{}`$, and we have:
(5.4)
$$\text{Tr}(A)=(2\pi h)^d_{_\alpha ^{2d}}<A\phi _\alpha ;\phi _\alpha >_{L^2(^d)}d\alpha .$$
For a proof, see for example .
### 5.3. Propagation of coherent states
For $`z=(q,p)^d\times ^d`$, let $`z_t=(q_t,p_t):=\mathrm{\Phi }_t(z)`$ be the solution of the Hamiltonian system (1.2) with initial condition $`z`$. We introduce the notations:
(5.5)
$$S(t,z):=_0^t(p_s.\dot{q}_sH(z_s))ds$$
(5.6)
$$\delta (t,z):=S(t,z)\frac{q_tp_tqp}{2}.$$
where $`F_\alpha (t)=_z\mathrm{\Phi }_t(\alpha )Sp(d,)`$. We set:
(5.7)
$$F_\alpha (t)=\left(\begin{array}{cc}A& B\\ C& D\end{array}\right)\text{ where }A,B,C,DM_d().$$
###### Theorem 5.1.
Semi-classical propagation of coherent states (Combescure-Robert) , :
Let $`T>0`$. Let $`H:^{2d}`$ be a smooth Hamiltonian satisfying, for all $`\alpha ^{2d}`$ :
(5.8)
$$|^\alpha H(z)|C_\alpha <x>^{m_\alpha },\text{ where }m_\alpha >0,C_\alpha >0.$$
Let $`\alpha ^{2d}`$ be such that the solution with initial condition $`\alpha `$ of the system $`\dot{z_t}=JH(z_t)`$ is defined for $`t]T,T[`$. We denote by $`U_h(t):=e^{i\frac{t}{h}\widehat{H}}`$ the quantum propagator.
Then, $`M,C_{M,T}(\alpha )>0`$, independant of $`h`$ and of $`t[T,T]`$ such that:
$$U_h(t)\phi _\alpha e^{i\frac{\delta (t,\alpha )}{h}}๐ฏ_h(\alpha _t)\mathrm{\Lambda }_h[\underset{j=0}{\overset{M}{}}h^{\frac{j}{2}}b_j(t,\alpha )(x).e^{\frac{i}{2}<M_0x,x>}]_{L^2(^d)}C_{M,T}(\alpha ).h^{\frac{M+1}{2}}.$$
where $`M_0:=(C+iD)(A+iB)^1`$, for all $`t]T,T[`$, $`b_j(t,\alpha ):^d`$ is a polynomial independant of $`h`$, with degree lower than $`3j`$, with same parity as $`j`$, and smoothly dependant of $`(t,\alpha )`$. In particular, $`b_0(t,\alpha )(x)=\pi ^{\frac{d}{4}}(det(A+iB))_c^{\frac{1}{2}}`$.
Moreover, if solutions of the Hamiltonian classical system are defined on $`[T,T]`$ for initial conditions $`\alpha `$ in a compact $`K`$, then $`\alpha C_{M,T}(\alpha )`$ is upper bounded on $`K`$ by $`\stackrel{~}{C}_{M,T,K}`$ independant of $`\alpha K`$.
|
warning/0506/astro-ph0506320.html
|
ar5iv
|
text
|
# Galaxy clustering from COMBO-17: The halo occupation distribution at โจ๐งโฉ=0.6
## 1 Introduction
In current models of galaxy formation, structure grows hierarchically from small Gaussian density fluctuations. Galaxies are presumed to form within virialized dark matter haloes when the baryonic gas cools and condenses into stars (e.g. Cole et al. 2000). The formation and evolution of galaxies should thus be closely tied to the merging history of dark matter haloes. This paper uses measurements of galaxy clustering at intermediate redshift to test this basic picture.
The complex relation between galaxies and dark matter has become clearer only slowly. Empirically, galaxies display biased clustering in which the amplitude of their correlations varies with galaxy type: older galaxies are generally much more strongly clustered than young, starforming galaxies, and bright galaxies are more strongly clustered than faint galaxies (e.g. Davis & Geller 1976; Norberg et al. 2002; Phleps & Meisenheimer 2003). It is generally believed that such trends can be understood through the tendency for dark-matter haloes to display clustering that is larger for rare massive haloes (e.g. Cole & Kaiser 1989; Mo & White 1996; Sheth & Tormen 1999).
However, a long-standing challenge has been to understand how these ideas could be implemented in the context of Cold Dark Matter (CDM) models. The galaxy correlation function has long been known to be extremely close to a single power law (Totsuji & Kihara 1969, Peebles 1974), and yet this is not the case for the nonlinear mass correlations in a CDM model. Here, the matter correlation function rises above a best-fit power law on scales $`r1h^1\mathrm{Mpc}`$ and falls below it again on scales $`r0.2h^1\mathrm{Mpc}`$ (Jenkins et al. 1998 and references therein). This puzzle was only resolved when it became clear that the correlation function of dark matter haloes (including subhaloes inside large host haloes) differs significantly from the correlation function of the mass. In practice, the predicted correlation function of galaxy-scale haloes follows a power law down to $`100h^1`$ kpc, with an amplitude and slope similar to the data on real galaxies (Kravtsov & Klypin 1999; Neyrinck et al. 2004; Kravtsov et al. 2004; Tasitsiomi et al. 2004). This phenomenon underlies the considerable scale-dependent bias predicted by semianalytic and hydrodynamic simulation models of galaxy formation (Colรญn et al. 1999; Kauffmann et al. 1999; Pearce et al. 1999; Benson et al. 2000; Cen & Ostriker 2000; Somerville et al. 2001; Yoshikawa et al. 2001; Weinberg et al. 2004).
These developments in turn stimulated a simpler and more direct insight into bias and its dependence on scale, through the so-called halo model (e.g. Jing 1998; Seljak 2000; Peacock & Smith 2000; Cooray & Sheth 2002 and references therein). Here, the shape of the correlation function is determined by the linear clustering of the dark matter, and the relation of the galaxies to the dark matter halos in which they reside (the Halo Occupation Distribution; HOD). In particular, a break in slope is expected when the correlation function changes from being dominated by pair counts of galaxies in separate dark matter halos to the small-scale regime, where pairs come from two galaxies that reside in the same halo. Any pure power law correlation function would require coincidental alignment of these two terms, and indeed analyses of the two-point correlation function of galaxies in the local universe have detected small deviations from the power-law form (Hawkins et al. 2003a; Zehavi et al. 2004, 2005; Abazajian et al. 2005).
In this paper, we use the COMBO-17 survey (Wolf et al. 2004) to carry out a similar investigation of the exact shape of the correlation function at higher redshifts. We calculate the projected correlation function $`w(r_p)`$ for red sequence and blue cloud COMBO-17 galaxies (following the definition of Bell et al. 2004), in the redshift bin $`0.4<z<0.8`$. By comparing these results to the predictions of the halo model, we are able to infer the mean number of galaxies per halo of a given mass (the halo occupation number) and also the $`z=0`$ power-spectrum normalization $`\sigma _8`$ (the rms density variation averaged over $`8h^1`$ Mpc spheres).
This paper is structured as follows: The COMBO-17 survey and the data used in this analysis are briefly described in Sect. 2. The halo model is introduced in Sect. 3. The method used to estimate the projected correlation function is explained in Sect. 4. In Sect. 5 we investigate the shape of the correlation function for red sequence and blue cloud galaxies, and in Sect. 6 the results are discussed. We assume a cosmological geometry taken from the WMAP results (Spergel et al. 2003, 2006) and the final 2dFGRS power spectrum results (Cole et al. 2005): a flat model with $`\mathrm{\Omega }_m=0.25`$. All lengths quoted are in comoving units. Normally, we show explicit dependence on $`h`$ (which denotes $`H_0/100\mathrm{km}\mathrm{s}^1\mathrm{Mpc}^1`$); but for absolute magnitudes we suppress this dependence, so that $`M_B`$ denotes $`M_B5\mathrm{log}_{10}h`$.
## 2 Data base: The COMBO-17 Survey
To date, COMBO-17 (Classifying Objects with Medium Band Observations in 17 filters) has surveyed three disjoint $`31^{}\times 30^{}`$ southern equatorial fields (for their coordinates see Wolf et al. 2003) to deep limits in $`5`$ broad and $`12`$ medium passbands, covering wavelengths from $`400`$ to $`930`$ nm. A detailed description of the survey along with filter curves can be found in Wolf et al. (2004). All observations were carried out using the Wide Field Imager at the MPG/ESO 2.2 m-telescope on La Silla, Chile.
In each filter, typically $`10`$ to $`20`$ individual exposures were taken (up to $`50`$ for ultradeep $`R`$-band images totalling $`20`$ ks with seeing $`0\stackrel{}{.}8`$). Galaxies were detected on the deep $`R`$-band images by using SExtractor (Bertin & Arnouts 1996). The spectral energy distributions (SEDs) for $`R`$-band detected objects were measured by performing seeing-adaptive, weighted-aperture photometry in all $`17`$ frames at the position of the $`R`$-band detected object. All magnitudes are quoted with a Vega zero point.
Using the 17-band photometry, objects are classified using a scheme based on template spectral energy distributions (Wolf et al. 2001b, a). The classification algorithm basically compares the observed colours of each object with a colour library of known objects. This colour library is assembled from observed and model spectra by synthetic photometry performed using an accurate representation of the instrumental characteristics of COMBO-17. For galaxy classification, we use PรGASE model spectra (see Fioc & Rocca-Volmerange 1997 for an earlier version of the model). The template spectra are a two-dimensional age/reddening sequence, in which a fixed exponential star formation timescale $`\tau =1`$ Gyr is assumed, ages vary between $`50`$ Myr and $`15`$ Gyr, and the reddening $`E(BV)`$ can be as large as $`0.5`$ mag, adopting a Small Magellanic Cloud Bar extinction curve. Note that we do not apply any morphological star/galaxy separation or use other criteria.
Using a minimum variance estimator, each object is assigned a redshift (if it is not classified as a star). The redshift errors in this process depend on magnitude and type of the object, and for galaxies can be approximated by
$`{\displaystyle \frac{\sigma _z}{(1+z)}}=0.007\left(1+10^{0.8(R21.6)}\right)^{1/2}.`$ (1)
The galaxy redshift estimate quality has been tested by comparison with spectroscopic redshifts for almost 1000 objects (see Wolf et al. 2004). At bright limits $`R<20`$, the redshifts are accurate to $`\sigma _z/(1+z)0.01`$, and the error is dominated by mismatches between template and real galaxy spectra. This error can contain a systematic component that is dictated by the exact filter placement, but these โredshift focusingโ effects are of the order of magnitude of the random redshift errors for $`z<1`$ and are unimportant for the current analysis. At the median apparent magnitude $`R23`$, $`\sigma _z/(1+z)0.02`$. For the faintest galaxies, the redshift accuracy approaches those achievable using traditional broadband photometric surveys, $`\sigma _z/(1+z)0.05`$. We thus restricted our analysis to galaxies with $`I<23`$.
Fig. 1 shows the redshift distribution of the $`\mathrm{22\hspace{0.17em}310}`$ COMBO-17 galaxies between $`z=0.2`$ and $`z=1.2`$ (with $`I<23`$ and $`M_B<18`$). The peak at $`z=0.733`$ in Fig. 1 is due to a real structure in the Chandra Deep Field South, which has been spectroscopically confirmed (Gilli et al. 2003). In order to define a volume limited sample, we restrict our analysis to the redshift range $`0.4<z<0.8`$ and galaxies brighter than $`M_B=18`$, which leaves us with $`\mathrm{10\hspace{0.17em}360}`$ galaxies for the analysis. Note that $`B`$-band luminosities can be determined directly without any $`K`$-correction uncertainty, based on the photometry in our 17 filters between $`400`$ and $`930`$ nm and an interpolation of the corresponding template spectra. We do not apply any evolutionary corrections.
The distribution of the redshift errors for all galaxies in our subsample with $`I<23`$, $`M_B<18`$ and $`0.4<z<0.8`$ is shown in Fig. 2. We use the prescription of Bell et al. (2004) to separate galaxies into the red-sequence component and the remaining blue cloud component:
$$\mathrm{Red}\mathrm{sequence}:(UV)>(UV)_{\mathrm{lim}}$$
(2)
$$\mathrm{Blue}\mathrm{cloud}:(UV)<(UV)_{\mathrm{lim}}$$
(3)
$$(UV)_{\mathrm{lim}}=1.250.4z0.08(M_V5\mathrm{log}_{10}h+20),$$
(4)
where $`z`$ denotes the redshift of each single galaxy.
Note that the cut that separates the red sequence galaxies from the blue ones depends on both redshift and absolute $`V`$ magnitude. This yields 2404 and 7956 galaxies in the red and blue subsamples; the former tend to have more accurate redshifts, as shown in Fig. 2. This is not because the classification scheme works better for the red galaxies, but because they are on average brighter than the blue ones.
Table 1 shows the number of red sequence and blue cloud galaxies per COMBO-17 field.
## 3 The halo model of galaxy clustering
In discussing the results of clustering analyses of the COMBO-17 data, we will make frequent comparisons with theoretical predictions. We therefore now summarise the framework used to carry out this modelling. This goes back to the paradigm introduced by White & Rees (1978): galaxies form through the cooling of baryonic material in virialized haloes of dark matter. The mass function and density profiles of these haloes can be expressed in terms of simple fitting formulae derived from N-body simulations, and the large-scale clustering of haloes can be derived analytically for Gaussian density fields. This concentration on dark-matter haloes gives concrete form to earlier work on the clustering statistics generated by distributions of extended clumps (see e.g. Neyman & Scott 1952 and Scherrer & Bertschinger 1991).
With an accurate description of dark-matter clustering to hand, the stage was set for an extension to galaxies via the โhalo modelโ (Ma & Fry 2000; Seljak 2000; Peacock & Smith 2000; Cooray & Sheth 2002). In this picture, the key remaining uncertainty is the way in which galaxies occupy the dark matter haloes; this can be regarded as an unknown function, to be probed experimentally. The halo model then allows a simple and direct understanding of many features of galaxy clustering, and how the clustering of galaxies differs from that of the mass.
In this approach, the density field is a superposition of dark-matter haloes, with small-scale clustering arising from neighbours in the same halo. The corresponding real-space correlation function can be written as a combination of two parts:
$`\xi _r=\xi _{\mathrm{lin}}+\xi _{\mathrm{halo}},`$ (5)
the first term representing the clustering of the dark matter haloes, and the second correlations from within a single halo. Large-scale halo correlations depend on mass, and the linear bias parameter for a given class of haloes, $`b(M)`$, depends on the rareness of the fluctuation and the rms of the underlying field (Kaiser 1984; Cole & Kaiser 1989; Mo & White 1996; Sheth & Tormen 1999), usually measured in spheres of $`8h^1\mathrm{Mpc}`$ and termed $`\sigma _8`$. The mass profile of the haloes is known from simulations, and may be assumed to follow either an NFW profile (Navarro et al. 1996, 1997, 2004), or a Moore et al. (1999) profile.
The key feature that allows bias to be included is to encode all the complications of galaxy formation via an halo occupation number: the number of galaxies found above some luminosity threshold in a virialized halo of a given mass $`M`$. A simple but instructive model for this halo occupation distribution (HOD) is
$`N(M)=\{\begin{array}{cc}0\hfill & \text{(}M<M_c\text{)}\hfill \\ (M/M_c)^\alpha \hfill & \text{(}M>M_c\text{) .}\hfill \end{array}`$ (6)
This is closely related to the mass-dependent weight introduced by Jing et al. (1998). A model in which light traces mass exactly would have $`M_c0`$ and $`\alpha =1`$. The galaxies are assumed to be split into a central galaxy plus some number of satellite galaxies, which follow the mass distribution in the halo. It is necessary to make an assumption about the statistics of the HOD โ in particular whether $`N`$ is a causal function of $`M`$ or whether it obeys a Poisson distribution. It is known that sensible results require sub-Poisson behaviour, and we assume the extreme limit in which $`N`$ is perfectly determined by $`M`$. Putting all these ingredients together, the galaxy correlation function can be calculated analytically.
Since its initial development, the halo model has been applied successfully to the interpretation of the correlation function of galaxies in the local universe, notably by Zehavi et al. (2004), who detected an inflection in the correlation function of SDSS red galaxies, interpreting it as indicating the transition regime between clustering dominated by 1-halo and 2-halo terms. The occupation model has been elaborated quite significantly (e.g. Abazajian et al. 2005; Zheng et al. 2005; Kravtsov et al. 2004; Tasitsiomi et al. 2004; Zentner et al. 2005), including up to three parameters for the occupation distribution, plus the inclusion of some nonlinear evolution of the power spectrum in the $`\xi _{\mathrm{lin}}`$ term. These sophistications can improve the detailed fit to correlation-function data from simulations, but they do run somewhat counter to the original heuristic spirit of the model. In this work, we shall retain the original method of calculation, as described in detail by Peacock & Smith (2000), together with the simple power-law occupation model. This approach seems justifiable in a first exploration of intermediate-redshift clustering, and the main features of interest are in any case relatively robust.
The form of any transition-regime feature in the correlation function depends mainly on the mean halo mass occupied by the galaxies, โ i.e. the average value of $`M`$, weighted by $`N(M)`$ โ and is insensitive to the details of their distribution within the halo. This typical halo mass is determined by our two-parameter model for $`N(M)`$ plus the halo mass function. We shall use the additional constraint of the observed number density of galaxies under study, so that there remains a single free parameter in the model โ which we take to be the occupation slope $`\alpha `$. For a given value of $`\alpha `$, the number density determines the cutoff mass and hence the average halo mass. Different models of the HOD can of course be used; Zehavi et al. (2004) take $`N=1`$ above $`M_c`$ to represent central galaxies, plus a power-law $`N(M)`$ representing satellites, which commences at a mass of approximately $`20M_c`$ with a slope of $`\alpha 1`$. In practice, this model gives results similar to the single power law with $`\alpha 0.6`$, showing that the detailed shape of the HOD is hard to measure given only correlation-function data. The typical number-weighted halo mass is a more robust quantity, and this is probably the best way to compare different HOD models.
We will normally assume a standard flat cosmology with $`\mathrm{\Omega }_m=0.25`$, $`\mathrm{\Omega }_b=0.045`$ and $`h=0.73`$ where $`H_0=100h`$ km s<sup>-1</sup> Mpc<sup>-1</sup> (Spergel et al. 2006; Cole et al. 2005). The most uncertain cosmological parameter is the normalization of the power spectrum, $`\sigma _8`$. We will often assume a standard value of $`\sigma _8=0.9`$, but it is also of interest to leave both $`\sigma _8`$ and the power-law index $`\alpha `$ of the halo occupation number (equation 6) as free parameters, and determine them from a fit to the measured galaxy correlation functions.
## 4 Redshift space correlations
### 4.1 Projected correlations
With a sample of the present size, only the two-point correlation function, $`\xi (r)`$, can be determined accurately. Even this is not straightforward, because of the need to work in redshift space. While it is possible to measure angular positions on the sky with high precision, peculiar velocities as well as redshift errors distort the galaxy pattern along the line of sight, making $`\xi (๐ซ)`$ appear anisotropic, and tending to reduce its amplitude.
These problems can be overcome by splitting the separation vector $`๐ซ`$ of a pair of objects into components lying on the plane of sky, $`r_p`$, and along the line of sight, $`\pi `$, and compute the correlation function $`\xi (r_p,\pi )`$ as a function of these two components. Projecting $`\xi (r_p,\pi )`$ onto the $`r_p`$ axis gives the projected function $`w(r_p)`$, which is independent of any radial distortions (Davis & Peebles 1983). For small angles $`r^2=r_p^2+\pi ^2`$. Thus the projected correlation function is defined as
$`w(r_p)`$ $`=`$ $`{\displaystyle _{\mathrm{}}^{\mathrm{}}}\xi \left[(r_p^2+\pi ^2)^{1/2}\right]d\pi `$ (7)
$`=`$ $`2{\displaystyle _{r_p}^{\mathrm{}}}\xi (r)(r^2r_p^2)^{1/2}rdr.`$
Note that $`w(r_p)`$ has dimensions of length. Since the correlation function converges rapidly to zero with increasing pair separation, the integration limits do not have to be $`\pm \mathrm{}`$, but they have to be large enough to include all correlated pairs. As will be explained in the following paragraph, this is a crucial point when the redshift errors are large.
### 4.2 The effect of redshift errors on $`\xi (r_p,\pi )`$
We now illustrate the effects of redshift errors, starting from a model for the true $`\xi (r_p,\pi )`$. This is calculated using the halo model, as described above in section 3, together with a prescription for redshift-space distortions. Seljak (2001) showed how to include redshift-space distortions in the halo model, but it is also common to use the following simple model for the ratio of redshift-space power to real-space power:
$$P_s/P_r=\frac{(1+\beta \mu ^2)^2}{(1+k^2\mu ^2\sigma _p^2/2)},$$
(8)
where $`\sigma _p`$ is an effective pairwise velocity dispersion, $`\mu `$ is cosine of the angle between $`๐ค`$ and the line of sight, and $`\beta \mathrm{\Omega }_m^{0.6}/b`$ (e.g. Ballinger, Peacock & Heavens 1996). This approach has the advantage that the redshift-space correlation function can then be found analytically, apart from a radial convolution (Hamilton 1992). We used this prescription, taking $`\sigma _p`$ to be $`\sqrt{2}`$ times the one-dimensional velocity dispersion calculated from the halo model. Fig. 3 shows a model $`\xi (r_p,\pi )`$, which contains two well-known expected anisotropies: the isocorrelation contours of $`\xi (r_p,\pi )`$ are stretched along the $`\pi `$ direction at small separations, because of the effect of virialized velocity dispersions, and compressed at large scales as a consequence of large-scale coherent motions. The former effect is not clearly visible owing to the large scale of the plot.
The effect of redshift errors on this redshift-space correlation function is straightforward: it is a convolution in the radial direction. This reflects the fact that $`1+\xi `$ is a ratio of the observed and expected numbers of pairs of galaxies. The correlation function $`\xi (r_p,\pi )`$ becomes distributed more broadly along the $`\pi `$ axis, but the total correlation signal is conserved โ this is why $`w(r_p)`$ is independent of redshift errors. In order to model this process, we need a model for the convolving function. This is simply deduced, because the classification scheme automatically returns an estimate of the rms redshift error for each galaxy (see Wolf et al. 2004). Thus, given a pair of galaxies $`i`$, $`j`$, with redshift errors $`\sigma _{z_i}`$ and $`\sigma _{z_j}`$ the rms pairwise error is $`\sigma _{\mathrm{pair}_{i,j}}=(\sigma _{z_i}^2+\sigma _{z_j}^2)^{1/2}`$. The signal from this pair is smeared by a Gaussian with this width, so the overall convolving function is a sum of the Gaussians corresponding to all pairs:
$`f(\delta z)={\displaystyle \frac{1}{N}}{\displaystyle \underset{\mathrm{pairs}}{}}(2\pi \sigma _{\mathrm{pair}})^{1/2}\mathrm{exp}\left[(\delta z/\sigma _{\mathrm{pair},n})^2/2\right],`$ (9)
where $`N`$ is the number of pairs $`(i,j)`$. After having transferred the pairwise redshift error distribution into comoving distances we can convolve $`\xi (r_p,\pi )`$ with Eq. (9).
The effect of this convolution is shown in the second panel in Fig. 3. The redshift-space correlations are now heavily elongated in the radial direction, and some care is needed in extracting the projected correlation signal.
### 4.3 Estimation of projected correlations
The simplest strategy for carrying out the projection needed in order to deduce $`w(r_p)`$ would be to integrate $`\xi (r_p,\pi )`$ over a very large radial range. Fig. 3 suggests that a maximum $`\pi `$ value of $`150`$ to $`200h^1\mathrm{Mpc}`$ would be required to capture all the signal. The problem with this strategy is that the random noise in $`\xi (r_p,\pi )`$ is independent of $`\pi `$ at a given $`r_p`$ (because the expected pair counts have a cylindrical dependence $`r_pdr_pd\pi `$). Thus, integration to $`\pi =200h^1\mathrm{Mpc}`$ would yield a random error in $`w(r_p)`$ that is $`\sqrt{2}`$ times larger than integration to $`\pi =100h^1\mathrm{Mpc}`$ โ but the lower limit systematically misses part of the signal.
We have developed a strategy for solving this problem, which depends weakly on some prior knowledge of the likely form of the true clustering signal (after error convolution). A model for $`\xi (r_p,\pi )`$ defines how the real-space signal $`w(r_p)`$ is spread out in $`\pi `$; we are only concerned with the shape of this function, which is dominated by the convolution with the redshift error distribution. Given this probability distribution in the $`\pi `$ direction, the amplitude of $`w(r_p)`$ can be estimated by fitting a scaled version of our model $`\xi (r_p,\pi )`$ to the data at the $`r_p`$ value of interest.
In practice, our model $`\xi (r_p,\pi )`$ will not be exact, and we considered the following compromise procedure for estimating $`w(r_p)`$ so that the result is robust. For each $`r_p`$ value, we fit the amplitude of the (convolved) model $`\xi (r_p=\mathrm{const},\pi )`$ to the data. We then integrate the data for $`\xi (r_p,\pi )`$ out to $`\pi =100h^1\mathrm{Mpc}`$, from which point on we integrate the convolved model out to infinity (see Fig. 4). This combines the exact measurement of $`w(r_p)`$ within $`\pi _{\mathrm{max}}=100h^1\mathrm{Mpc}`$ with an estimate of the missing signal at larger $`\pi `$. Since this correction is typically 20% of the overall signal, we do not need to estimate it very accurately. In practice, the results from the 2-stage procedure were very similar to the direct fitting method. This process is performed separately for the red and blue galaxy samples, using the appropriate pairwise error distributions and the final best-fitting halo-model $`\xi (r_p,\pi )`$ for the unconvolved prediction.
The width of the convolved model in Fig. 4 suggests that the redshift errors yielded by the object classification scheme, which we used for the calculation of the pairwise error distribution, may be slightly overestimated. In order to estimate the effect on the projected correlation function, we tried repeating the analysis with redshift errors scaled to $`80`$% of the values given in the object catalogues. This scaling gives the best fit to the data; however, the resulting changes to $`w(r_p)`$ were small compared to the random errors.
### 4.4 Integral constraint
The mean galaxy density is determined from the observed galaxy counts in each field, which does not necessarily represent the the true density (Groth & Peebles 1977). The estimator will be on average biased low with respect to the true correlation by a constant $``$:
$`w_m(r_p)=w_t(r_p),`$ (10)
where $`w_t(r_p)`$ is the true projected correlation function, $`w_m(r_p)`$ the measurement. The integral constraint $``$ is given by
$`{\displaystyle \frac{1}{S^2}}{\displaystyle w_\mathrm{t}(r_p)\mathrm{d}^2S_1\mathrm{d}^2S_2},`$ (11)
where $`S`$ is the physical area corresponding to the solid angle of the field at the redshift under consideration. For the calculation of the integral constraint, we assume that the three dimensional correlation function $`\xi (r)`$ is to first approximation a power law:
$`\xi (r)=\left({\displaystyle \frac{r}{r_0}}\right)^\gamma .`$ (12)
Then the evaluation of equation 7 yields
$`w(r_p)=Cr_0^\gamma r_p^{1\gamma },`$ (13)
where $`C`$ is a numerical factor, which depends only on the slope $`\gamma `$:
$`C=\sqrt{\pi }{\displaystyle \frac{\mathrm{\Gamma }((\gamma 1)/2)}{\mathrm{\Gamma }(\gamma /2)}}.`$ (14)
If the true correlation function is given by equation (13), the measurement yields
$`w_m(r_p)`$ $`=`$ $`Cr_0^\gamma r_p^{1\gamma }`$ (15)
$`=`$ $`Cr_0^\gamma \left[r_p^{1\gamma }/(Cr_0^\gamma )\right].`$ (16)
The true amplitude $`Cr_0^\gamma `$ is not known, but $`/(Cr_0^\gamma )`$ can be estimated by performing a Monte Carlo integration (where we use the mean of the pair counts $`RR`$ at a projected distance $`r_p`$ of the four fields):
$`{\displaystyle \frac{}{Cr_0^\gamma }}={\displaystyle \frac{\left[RRr_p^{1\gamma }\right]}{RR}}.`$ (17)
The true value of $`Cr_0^\gamma `$ can be estimated by fitting equation (16) to the data, taking the value of $`/(Cr_0^\gamma )`$ from equation (17). This value, multiplied by the fitted amplitude $`Cr_0^\gamma `$, yields the integral constraint $``$. The measurement can then be corrected for the integral constraint by adding $``$ to $`w_m(r_p)`$.
This method yields estimates of $`=0.14h^1\mathrm{Mpc}`$ for the red galaxies and $`=0.33h^1\mathrm{Mpc}`$ for the blue galaxies. These values are negligible in comparison with the observed data for $`r_p20h^1\mathrm{Mpc}`$, demonstrating that the fields are large enough to deliver a fair sample. As a cross-check, note that we expect $`=200\sigma ^2`$ (since we integrated over $`\mathrm{\Delta }\pi =200h^1\mathrm{Mpc}`$), where $`\sigma ^2`$ is the fractional variance in galaxy numbers between different realizations of our survey. With three fields, $`\sigma `$ should be $`\sqrt{3}`$ times smaller than the field-to-field rms variation, so our figures for $``$ suggest 4.5% and 7% expected scatter in the numbers of galaxies per field for red and blue galaxies respectively. This agrees well with the numbers in Table 1.
### 4.5 Error analysis
Finally, there is the crucial issue of setting realistic error bars on our correlation estimates. The three COMBO-17 fields measure $`31^{}\times 30^{}`$ each and are thus large enough to carry out a jack-knife analysis. We divide each field into four quadrants, and then calculate the correlation function $`w(r_p)`$ (including the integral constraint) for twelve realisations of the data, each time omitting one of the quadrants. The variance in $`w`$ is then given approximately by
$`\sigma ^2={\displaystyle \frac{N1}{N}}{\displaystyle \underset{i=1,N}{}}\left[w(r_p)w_i(r_p)\right]^2,`$ (18)
where $`N=12`$ is the number of realisations of the data (e.g. Scranton et al. 2002).
In order to check for cross-correlations between the data points, we can extend the jack-knife method in the obvious way to estimate the covariance between different bins, $`\sigma _{ij}^2`$. The natural way to express this is as a correlation coefficient matrix: $`r_{ij}\sigma _{ij}^2/\sigma _i\sigma _j`$. Results in this form are presented below.
## 5 The clustering of the COMBO-17 galaxies
### 5.1 Results
We calculated $`\xi (r_p,\pi )`$ for all COMBO-17 galaxies in the redshift range $`0.4<z<0.8`$ with $`I`$-band magnitudes $`I<23`$ and absolute restframe $`B`$ band luminosities $`M_B<18`$. We used the estimator invented by Landy & Szalay (1993). An angular mask for the survey was derived by censoring the surroundings of bright stars in the fields. The same mask was applied to a random catalogue consisting of $`\mathrm{30\hspace{0.17em}000}`$ randomly distributed galaxies, each of which was assigned a redshift taken randomly from the real data, where the three fields were put together in order to smooth the redshift distribution. Using a smoothed form of the empirical redshift distribution did not yield a significant change in the results.
The resulting $`\xi (r_p,\pi )`$ is shown in Fig. 5. The field of view of the COMBO-17 fields limits the pair separations accessible for the analysis, so in the transverse direction there is of course no signal at separations larger than the physical distance corresponding to the diagonal diameter of the fields.
For each object, we have an estimate of the redshift and the restframe colours and luminosities; it is therefore possible to divide the sample into two distinct colour classes as described earlier. For both samples we calculated $`w(r_p)`$ as described in section 4, correcting for the integral constraint $``$, and the influence of the redshift errors. These results are shown in Fig. 6.
### 5.2 Fitting the halo model
Fig. 6 also shows predictions from the halo model, varying the single occupation-number parameter $`\alpha `$, and choosing the cutoff $`M_c`$ so as to match the observed comoving densities of $`0.004h^3\mathrm{Mpc}^3`$ (red) and $`0.012h^3\mathrm{Mpc}^3`$ (blue). It is apparent that there is greater sensitivity to $`\alpha `$ at small separations, and that once $`\alpha `$ is fixed from the data there, there is little freedom at large separations, where the data and the model match satisfyingly well. The preferred values are approximately $`\alpha =0.5`$ for the red population and $`\alpha =0.2`$ for blue galaxies. These figures correspond to cutoff masses of respectively $`M_c=10^{12.15}h^1M_{}`$ and $`M_c=10^{11.50}h^1M_{}`$. As discussed earlier, a more meaningful way of casting these numbers may be to apply the HOD model to the halo mass function, to calculate the effective halo mass, weighting by galaxy number. These figures come out as $`M_{\mathrm{eff}}=10^{13.21}h^1M_{}`$ and $`M_{\mathrm{eff}}=10^{12.52}h^1M_{}`$ respectively.
Fig. 6 also shows a magnified view, with the measured correlation functions and the corresponding best-fitting models both divided by a power-law fit (fitted in the range $`\mathrm{log}_{10}r_p<1.1`$), the slope and amplitudes of which are given in Table 2. The data points do not scatter arbitrarily around the power-law fit, but show systematic deviations. For the red galaxies, there is a marked dip around $`r_p1.5h^1\mathrm{Mpc}`$; the blue galaxies are closer to a power law, but with a relatively abrupt step at $`r_p0.2h^1\mathrm{Mpc}`$. Both these features are impressively well accounted for by the halo model predictions, especially when it is considered that there is only one free parameter.
It is interesting to compare our results with those of the VVDS project (Le Fรจvre et al. 2005). They give results to a similar depth for two fields, although not divided by colour, with a total of 7155 redshifts over 0.61 deg<sup>2</sup>. Their redshift bins are not identical, but they quote $`r_0=2.69_{0.59}^{+0.53}h^1\mathrm{Mpc}`$ and $`\gamma =1.71_{0.11}^{+0.18}`$ at $`z=0.6`$ and $`r_0=4.55_{1.56}^{+1.25}h^1\mathrm{Mpc}`$ and $`\gamma =1.48_{0.15}^{+0.28}`$ at $`z=0.7`$. The latter figure is from the CDFS, which is one of our fields, and we have checked that our figure for this field alone agrees well with the VVDS, as it does for our other fields. The VVDS $`2^h`$ field thus gives a somewhat lower clustering strength; this may be because the VVDS sample in that field is about 0.5 mag. deeper than the one studied here, plus the fact that the VVDS analysis has no lower limit in luminosity. The ratio in $`r_0`$ between the VVDS $`2^h`$ field and our overall result is $`1.75\pm 0.21`$, so a factor 1.3 from luminosity-dependent clustering would be required in order to make the results statistically consistent.
We now want to quantify the agreement between model and data in more detail, using the jack-knife error estimates. This would be straightforward if the covariance matrix were diagonal, and if the model was exact. In practice, there is some degree of correlation in the data, and the simple halo model used here may be expected to have some systematic deviations with respect to ideal data. The correlation matrices in Fig. 7 indeed show a strong correlation between the large-scale data points (large matrix indices, lower right corner). This is due to the integral constraint, which is included in the calculation of the jack-knife realisations: all large-scale data points are offset by the same amount and thus become correlated. This issue is not too serious, since the errors in this regime are in any case large. We therefore ignore the large-scale points at $`r_p>10h^1\mathrm{Mpc}`$ and treat the remaining data as independent. Even so, our approximate model is not guaranteed to deliver a perfect fit. In order to achieve a formally acceptable value of $`\chi ^2`$ for the fit, we added in quadrature an error of 5% to the errors on $`w`$ for red galaxies. When fitting the $`z=0`$ data, as discussed below, a covariance matrix was not available, and the formal errors are in any case small. In this case, we therefore took what is effectively a least-square approach, which required an effective error of 10% in both blue and red galaxies.
In performing the fitting, it is interesting to consider variations in both the power-law index $`\alpha `$ of the HOD (equation 6), and in the normalization $`\sigma _8`$. We emphasise that $`\sigma _8`$ is the zero-redshift value, which is connected to the degree of inhomogeneity at $`z=0.6`$ by the growth factor predicted by the cosmological model (a change in linear density contrast by a factor 1.32). Fig. 8 shows the likelihood contours for these two free parameters $`\alpha `$ and $`\sigma _8`$. The preferred values and marginalized rms errors are $`\alpha =0.56\pm 0.03`$ and $`\sigma _8=0.84\pm 0.08`$ for the red sequence galaxies and $`\alpha =0.16\pm 0.03`$ and $`\sigma _8=1.19\pm 0.09`$ for the blue population.
These independent estimates of $`\sigma _8`$ from the red and blue populations are both close to our default value of 0.9. The agreement is not perfect, and the difference in $`\sigma _8`$ is formally $`3\sigma `$, but the mean value of $`\sigma _8=1.02\pm 0.17`$ is certainly plausible. Given the simplicity of the HOD model, this is a satisfying result.
### 5.3 Robustness of the results
In view of the tension between the normalization inferred from the red and blue galaxies, and in the light of the preference for a low normalization of $`\sigma _80.75`$ from the 3-year WMAP data (Spergel et al. 2006), it is important to discuss the extent to which our result is rendered uncertain by simplifying assumptions in the modelling.
The halo model is an idealized approximation in many ways, and one issue in particular has generated considerable discussion recently. It has been shown that the clustering of dark matter haloes is dependent on the halo formation time (Sheth & Tormen 2004a, b; Gao et al. 2005; Wechsler et al. 2005; Harker et al. 2006; Reed et al. 2006), and the question is what impact this has on halo model calculations. To some extent, the effect is already included in the halo-model formalism: when the cosmic density field is smoothed on a given mass scale, the clustering of peaks in the smoothed field is well known to increase with peak height, i.e. with formation redshift (Kaiser 1984). More massive systems are more strongly clustered for the same reason: only the rarest peaks exceed the threshold for collapse when the variance in filtered density is low. When we use the standard expression for bias as a function of mass, this averages over all systems that have collapsed by the present: the dependence of clustering on formation time will thus have no effect on predicted galaxy properties if the occupation numbers are purely a function of halo mass. However, it seems reasonable that the occupation number for a given mass will in fact depend to some extent on collapse redshift (e.g. more red galaxies in a halo that collapses early). At a minimum, the age-clustering effect will then contribute to a stochastic aspect of the occupation number, so that there is some scatter in $`N`$ at a given $`M`$. More seriously, it can also bias the mean clustering compared to all haloes of that mass. The influence of these effects on halo-model predictions remains to be explored, and this task is beyond the scope of the present paper. Some work along these lines has been done by Zentner et al. (2005) and by Croton, Gao & White (2006), where the halo contents in a simulation are scrambled between all haloes of the same mass, thus destroying any correlations with collapse redshift. Zentner et al. (2005) did this for subhaloes, but Croton, Gao & White (2006) considered the case of most direct interest, which is semianalytic galaxy populations. They do detect systematic shifts in correlation amplitude, but for the luminosities of interest here these are no larger than 10%. Such shifts are not important in comparison with the COMBO-17 measuring errors, but this is clearly an issue that should be looked at in more detail, especially as the accuracy in measuring high-redshift clustering improves.
Other degrees of freedom in the halo model are more easily investigated, and we summarise some tests here, the results of which are presented in Table 3. We show the impact on the values of $`\sigma _8`$ inferred from red and blue galaxies by (a) varying cosmological parameters; (b) varying parameters internal to the halo model; (c) varying the occupation number prescription. In the first category, we see that the Hubble parameter has very little effect, but that $`\sigma _8`$ increases with $`\mathrm{\Omega }_m`$ very roughly as $`\mathrm{\Omega }_m^{0.3}`$ to $`\mathrm{\Omega }_m^{0.5}`$, with a larger sensitivity for the blue galaxies. In the second category, we considered altering the assumed density contrast for a virialized halo from the usual figure of 200 to a slightly larger number, which is sometimes assumed in a low-density model. Finally, we modify our simple power-law $`N(M)`$ to something that resembles more closely the prescription used by e.g. Zheng et al. (2005): $`N=1`$ between $`M_c`$ and $`10M_c`$, rising as $`M^\alpha `$ for smaller $`M`$ (we considered $`\alpha 1`$). We also include the formal $`\chi ^2`$ values for some of these alternatives. It appears that the standard model provides the best fit, especially to the red galaxies. However, bearing in mind the simplified nature of the halo model, these differences should not be given high weight; it is more interesting to concentrate on the robustness of the best-fitting parameters.
The overall conclusion of these tests is that plausible variations of some of the degrees of freedom in the modelling can alter $`\sigma _8`$ by 10 to 20%, and in a way that changes the consistency between red and blue results by a similar amount. We therefore conclude that the modelling is working as well as could have been expected, and that there is no need to be concerned by either the internal red-blue tension or by the 1.6$`\sigma `$ discrepancy with the WMAP $`\sigma _8`$. We now turn to the comparison with $`z=0`$; some of the systematics in the analysis should be common to all redshifts, so we should hope for a good level of consistency between measurements based on data from different epochs.
### 5.4 Comparison with local clustering
As a local comparison, we considered $`w(r_p)`$ for a combined set of red and blue galaxies taken from both Sloan Digital Sky Survey (SDSS; York et al. 2000) and 2dFGRS (Colless et al. 2001) data. The sample has been divided into red and blue by either the bimodality of the rest-frame colour distribution, or spectral type, in a way that should compare reasonably well with the COMBO-17 classification. We use the flux-limited 2dFGRS results of Hawkins et al. (2003a), and the $`19>M_r>20`$ results of Zehavi et al. (2005), which have closely comparable amplitudes. Fig. 9 shows a comparison of the local and the high redshift sample. What is apparent here is that the main difference between the COMBO-17 sample at $`z=0.6`$ and the local data is in the shape of the correlation function, with almost identical amplitudes at small scales, but a difference of nearly a factor two for $`r_p>1h^1\mathrm{Mpc}`$.
As shown in Fig. 9, the halo model is capable of accounting for these differences. Fig. 10 repeats the model-fitting exercise for the $`z=0`$ data, and the preferred values and marginalized rms errors are $`\alpha =0.49\pm 0.02`$ and $`\sigma _8=1.03\pm 0.07`$ for the red sequence galaxies and $`\alpha =0.23\pm 0.02`$ and $`\sigma _8=1.00\pm 0.05`$ for the blue population. These values of $`\alpha `$ are slightly smaller than those obtained from COMBO-17; they correspond to cutoff masses of respectively $`M_c=10^{12.20}h^1M_{}`$ and $`M_c=10^{11.50}h^1M_{}`$, or effective halo masses $`M_{\mathrm{eff}}=10^{13.50}h^1M_{}`$ and $`M_{\mathrm{eff}}=10^{12.80}h^1M_{}`$ respectively. Compared to the $`z=0.6`$ results, $`M_c`$ has not changed much, but $`M_{\mathrm{eff}}`$ has increased by about a factor 2. This makes sense in terms of hierarchical growth: the minimum dark-matter mass needed to assemble a galaxy-sized amount of baryons should be invariant, but such haloes inevitably merge into larger systems as time progresses.
The mean value of $`\sigma _8=1.01\pm 0.04`$ from the $`z=0`$ data agrees very well with $`\sigma _8=1.02\pm 0.17`$ from COMBO-17. This agreement assumes hierarchical growth in the halo mass function between $`z=0.6`$ and the present, without which the inferred values of $`\sigma _8`$ would have been expected to differ by a factor 1.3.
## 6 Summary & conclusions
Using a sample of 10 360 galaxies with photometric redshifts from the COMBO-17 survey, we have investigated in some detail the shape of the correlation function at redshift $`z0.6`$. We have shown for the first time that the two-point correlation functions of both red sequence galaxies and blue cloud galaxies at this redshift display deviations from a power law, analogous to the deviations seen at low redshift (Hawkins et al. 2003a; Zehavi et al. 2004).
We have compared these observations to the predictions of a simple halo model, and find a good fit. It appears that the COMBO-17 data allow us to identify the point of transition between 1-halo clustering and 2-halo clustering, as was done at $`z=0`$ by Zehavi et al. (2004). The implication is that the red and blue galaxies at $`z=0.6`$ inhabit haloes of typical effective mass $`M_{\mathrm{eff}}=10^{13.2}h^1M_{}`$ and $`M_{\mathrm{eff}}=10^{12.5}h^1M_{}`$ respectively.
We have also allowed the zero-redshift normalization of the power spectrum, $`\sigma _8`$, to be a free parameter in this analysis. Impressively, both red and blue subsets imply a consistent local normalization of the power spectrum: $`\sigma _81`$. This figure is close to the value inferred by independent means using CMB and gravitational lensing (e.g. Refregier 2003), and certainly within the tolerance expected from the inevitable systematics associated with the simple modelling that we have used in this first analysis of galaxy clustering at intermediate redshifts. This consistency is obtained by assuming that the dark halo mass function grows in a standard hierarchical fashion so that the normalization at $`z=0.6`$ is approximately 30% lower than today. Our results amount to a verification that growth of this order has occurred.
We intend to expand this work to higher redshifts using COMBO-17+4, the NIR extension of COMBO-17, for which observations are currently being carried out using the $`2k\times 2k`$ Omega2000 camera at the 3.5m-telescope on Calar Alto, Spain. Combining the existing optical data base from COMBO-17 with NIR observations in one broad and three medium band filters (covering the wavelength range from 1040 to 1650 nm), we expect to obtain $`4200`$ galaxy redshifts with an accuracy of $`\sigma _z/(1+z)=0.02`$ up to $`z=2`$. This much longer baseline in cosmic time will allow us to observe much larger evolution of halo masses, testing the idea of hierarchical growth back to a time close to the formation of luminous galaxies.
###### Acknowledgements.
S. Phleps acknowledges financial support by the SISCO Network provided through the European Communityโs Human Potential Programme under contract HPRN-CT-2002-00316. JAP was supported by a PPARC Senior Research Fellowship. CW was supported by a PPARC Advanced Fellowship. We thank Luigi Guzzo for helpful comments. We thank the anonymous referee for helpful comments and suggestions.
|
warning/0506/math0506519.html
|
ar5iv
|
text
|
# Geometric Galois Theory, Nonlinear Number Fields and a Galois Group Interpretation of the Idele Class Group Revised Version
## 1. Introduction
Since the introduction of a global understanding of number theory (Gauss, Galois) and geometry (Riemann), the idea that the two subjects exist in parallel duality has exercised a tremendous pull on mathematical thought. An indication of this conjectural relationship can be found in the equivalence between (coverings of) Riemann surfaces and (extensions of) their fields of meromorphic functions, based on which, one can formulate the following meta-principle:
> To every algebraic number field $`K/`$, there exists a โRiemann surfaceโ $`\mathrm{\Sigma }_K`$ for which
>
> $$\mathrm{`}\mathrm{`}\mathrm{๐ฌ๐พ๐}\left(\mathrm{\Sigma }_K\right)K\text{}.$$
>
> If $`L/K`$ is a Galois extension, then $`\mathrm{\Sigma }_L`$ is a Galois covering of $`\mathrm{\Sigma }_K`$ and
>
> $$\mathrm{`}\mathrm{`}\mathrm{๐ฃ๐พ๐ผ๐}\left(\mathrm{\Sigma }_L/\mathrm{\Sigma }_K\right)\mathrm{๐ฆ๐บ๐
}\left(L/K\right)\text{}.$$
If we were interested in an extension $`E`$ over the function field $`๐ฝ_{p^n}[X]`$, then this principle is, suitably interpreted, correct: there exists a curve over $`๐ฝ_{p^n}`$ whose function field is $`E`$. This observation was used by Weil to prove the Riemann hypothesis for function fields over finite fields .
In the case of a number field $`K`$, the principle as stated above must be modified, as any reasonable notion of โfield of meromorphic functionsโ must contain the field of constants $``$, which cannot be a subfield of $`K`$. Thus we might instead ask that $`K`$ generate $`\mathrm{๐ฌ๐พ๐}(\mathrm{\Sigma }_K)`$ over $``$. The additional freedom afforded by the use of $``$-coefficients motivates a second meta-principle:
> If $`K^{\mathrm{ab}}`$ is the maximal abelian extension of $`K`$ and $`๐ข_K`$ is the idele class group of $`K`$, then there is a monomorphism
>
> $$\mathrm{`}\mathrm{`}๐ข_K\mathrm{๐ฆ๐บ๐
}\left(\mathrm{๐ฌ๐พ๐}\left(\mathrm{\Sigma }_{K^{\mathrm{ab}}}\right)/K\right)\text{}.$$
This second meta-principle โ which is true for function fields over finite fields upon using rational function fields in place of meromorphic function fields โ has an important place in Weilโs approach to the classical Riemann hypothesis, as the following often quoted passage reveals :
> โLa recherche dโune interprรฉtation pour $`๐ข_K`$ si $`K`$ est un corps de nombres, analogue en quelque maniรจre ร lโinterprรฉtation par un groupe de Galois quand $`K`$ est un corps de fonctions, me semble constituer lโun des problรจmes fondamentaux de la thรฉorie des nombres ร lโheure actuelle; il se peut quโune telle interprรฉtation renferme la clef de lโhypothรจse Riemannโฆโ
Weilโs speculation regarding a Galois group interpretation of $`๐ข_K`$ has inspired new approaches to the Riemann hypothesis, e.g. especially that of Alain Connes . In this paper, we shall give a certain expression to these principles through a hyperbolized version of the adele class group of $`K`$.
Let us consider first a field $`K`$ of finite degree $`d`$ over $``$. To $`K`$ we may associate the adele class group $`\widehat{๐}_K=๐ธ_K/(K,+)`$, a $`d`$-dimensional solenoid whose leaves are dense and isomorphic to a product of the form
$$K_{\mathrm{}}=^r\times ^s$$
where $`K`$ has $`r`$ real places and $`2s`$ complex places. From $`\widehat{๐}_K`$ we construct a hyperbolization $`\widehat{๐}_K`$ whose leaves are polydisks isomorphic to $`(^2)^d`$ and whose distinguished boundary is $`\widehat{๐}_K`$. See ยงยง2, 3 and 6 for more details.
For a number field $`๐ฆ/`$ of infinite degree, such as the maximal abelian extension $`K^{\mathrm{ab}}`$, the notion of adele class group has not yet, to our knowledge, been defined. In order to redeem the desired properties found in the adele class group of a finite field extension, it is necessary to consider a pair of adele class groups that work in tandem. Representing $`๐ฆ=\underset{}{lim}K_\lambda `$, where the $`K_\lambda `$ are finite degree extensions of $``$, the (ordinary) adele class group $`\widehat{๐}_๐ฆ`$ is formed from $`\underset{}{lim}\widehat{๐}_{K_\lambda }`$ by completing its canonical leaf-wise euclidean metric, see ยง4. Since $`\widehat{๐}_๐ฆ`$ is not locally-compact, we consider in ยง5 a compactification $`\widehat{\widehat{๐}}_๐ฆ`$ called the proto adele class group of $`๐ฆ`$, arising as the inverse limit of the trace maps. We use the hilbertian $`\widehat{๐}_๐ฆ`$ to define the hyperbolization $`\widehat{๐}_๐ฆ`$, and the compact $`\widehat{\widehat{๐}}_๐ฆ`$ to provide Fourier analysis. See ยงยง4, 5.
The origin of the notion of a nonlinear field comes from an enhanced understanding of the character group $`\mathrm{๐ข๐๐บ๐}\left(\widehat{๐}_K\right)`$, described in ยง7. The character group possesses an additional operation making it a field, and there is a canonical isomorphism
(1)
$$\mathrm{๐ข๐๐บ๐}\left(\widehat{๐}_K\right)K.$$
If we denote by $`๐_K=K_{\mathrm{}}/O_K`$ the Minkowski torus, the above isomorphism identifies
$$\mathrm{๐ข๐๐บ๐}\left(๐_K\right)๐ก_K^1O_K$$
where $`๐ก_K^1`$ is the inverse different. Thus $`\mathrm{๐ข๐๐บ๐}\left(๐_K\right)`$ has the enhanced structure of an $`O_K`$-module canonically extending the ring $`O_K`$. The same is true for an infinite field extension $`๐ฆ`$ that is Galois over $``$ if we use the proto adele class group $`\widehat{\widehat{๐}}_๐ฆ`$.
Let $`fL^2(\widehat{๐}_K,)`$ = the Hilbert space of square integrable complex-valued functions with respect to normalized Haar measure. By Fourier theory, $`f`$ has the development
$$f=a_\alpha \varphi _\alpha $$
where $`\alpha K`$, $`a_\alpha `$ and $`\varphi _\alpha \mathrm{๐ข๐๐บ๐}\left(\widehat{๐}_K\right)`$, and so the isomorphism (1) defines an inclusion $`KL^2(\widehat{๐}_K,)`$. Cauchy (point-wise) multiplication of functions, when defined, is denoted $`fg`$ since it restricts to $`+_K`$ in $`K`$. The operation corresponding to $`\times _K`$ is the Dirichlet product $`fg`$, and through it $`L^2(\widehat{๐}_K,)`$ acquires the structure of a partial double group algebra, where partial refers to the fact that the two operations are only defined when square integrability is conserved. The departure from ordinary field theory begins with the observation that Dirichlet multiplication does not distribute over Cauchy multiplication, or to put it differently, the extension of multiplication in $`K`$ to $`L^2(\widehat{๐}_K,)`$ no longer defines a Cauchy bilinear operation. See ยง8.
An interpretation of $`L^2(\widehat{๐}_K,)`$ as boundary values of holomorphic functions on $`\widehat{๐}_K`$ comes through the introduction of graded holomorphicity, treating, in the style of conformal field theory, each notion of holomorphicity on the same footing as the classical one. Suppose first that $`K`$ is totally real. Denote by $`\mathrm{\Theta }_K=\{,+\}^d`$ the sign group. To each $`\theta \mathrm{\Theta }_K`$ we associate the Hardy space $`๐ง_\theta [K]`$ of $`\theta `$-holomorphic functions on $`\widehat{๐}_K`$, and every $`fL^2(\widehat{๐}_K,)`$ determines a $`2^d+1`$ tuple $`(F_\theta ;F_0)`$ consisting of $`\theta `$-holomorphic components and the constant term $`F_0=a_0`$. In particular, there is an isomorphism of Hilbert spaces
$$๐ง_{}[K]:=\underset{\theta }{}๐ง_\theta [K]L^2(\widehat{๐}_K,).$$
The graded Hilbert space $`๐ง_{}[K]`$ inherits the partially-defined operations of $``$ and $``$, where the Dirichlet product has a homogeneous decomposition with respect to the grading of $`๐ง_{}[K]`$. See ยง8.1.
When $`K`$ is totally complex we consider, for each complex place, a pair of order four signings: the singular complex sign group $`\mathrm{\Theta }^{}=\{\sqrt{},,\sqrt{},+\}`$, which signs points on $`(i)0`$ according to the axial component that they belong to, and the complex sign set $`\mathrm{\Omega }=\{\sqrt{}ฯต,ฯต,\sqrt{}ฯต,+ฯต\}`$ which signs points in $`(i)`$ according to the quadrant they belong to. If we denote by $`\mathrm{}=\mathrm{\Theta }^{}\mathrm{\Omega }`$ then there is a subset $`\mathrm{}_K\mathrm{}^s`$ with respect to which we may define, for each $`\vartheta \mathrm{}_K`$, a Hardy space $`๐ง_\vartheta [K]`$ of $`\vartheta `$-holomorphic functions on $`\widehat{๐}_K`$. We obtain in analogy with the real case an isomorphism
$$๐ง_{}[K]:=\underset{\vartheta }{}๐ง_\vartheta [K]L^2(\widehat{๐}_K,).$$
See ยง8.2. In the hybrid case where $`K`$ has both real and complex planes we obtain a direct sum $`๐ง_{}[K]=๐ง_{}^{}[K]๐ง_{}^{}[K]`$ consisting of the $`\theta `$-holomorphic and $`\vartheta `$-holomorphic parts.
Since the vector space structure of $`๐ง_{}[K]`$ does not descend to $`K`$, it is natural to discard it by projectivizing. After removing the functions whose trace is defined and equal to zero, we obtain an infinite-dimensional affine subspace
$$๐ญ_{}[K]๐ง_{}[K]$$
endowed with a partial double semigroup structure induced from $``$ and and $``$, satisfying the following properties:
1. Let $`[K]`$ be the field algebra (double group algebra) associated to $`K`$ and let $`๐ญ_{}^0[K][K]`$ denote the sub double semigroup of elements of non zero trace, graded according to the same scheme described above. Then there is a graded monomorphism $`๐ญ_{}^0[K]๐ญ_{}[K]`$ with dense image.
2. The identity $`\mathrm{๐๐ฝ}_{}`$ is a universal annihilator for the product $``$: for all $`f๐ญ_{}[K]`$ for which $`f\mathrm{๐๐ฝ}_{}`$ is defined, $`f\mathrm{๐๐ฝ}_{}=\mathrm{๐๐ฝ}_{}`$.
We call a topological partial double semi-group satisfying these properties an (abstract) nonlinear number field over $`K`$. The ring of integers $`O_K`$ of $`K`$ generates in turn the nonlinear ring of integers $`๐ญ_{}[O_K]๐ญ_{}[K]`$, a nonlinear extension of $`O_K`$. On a dense subset of $`๐ญ_{}[K]`$, every element may be regarded as a Dirichlet quotient of elements of $`๐ญ_{}[O_K]`$. See ยง9.
An automorphism of the nonlinear number field $`๐ญ_{}[K]`$ is defined to be the restriction of a graded Fubini-Study isometry preserving the operations $``$ and $``$. Given $`K`$ an algebraic number field, denote by
$$\mathrm{๐ฆ๐บ๐
}(๐ญ_{}[K]/K)$$
the group of automorphisms of $`๐ญ_{}[K]`$ fixing $`K`$; if $`L/K`$ is Galois, denote by
$$\mathrm{๐ฆ๐บ๐
}(๐ญ_{}[L]/๐ญ_{}[K])$$
the group of automorphisms of $`๐ญ_{}[L]`$ fixing $`๐ญ_{}[K]`$.
###### Theorem.
For all $`K`$, $`\mathrm{๐ฆ๐บ๐
}(๐ญ_{}[K]/K)`$ is trivial. If $`L/K`$ is Galois, then
$$\mathrm{๐ฆ๐บ๐
}(๐ญ_{}[L]/๐ญ_{}[K])\mathrm{๐ฆ๐บ๐
}(L/K).$$
We consider finally the case $`K=`$. In order to interpret the idele class group $`C_{}`$ as a Galois group within this framework, the operations $``$ and $``$ must be decoupled. Let us consider $`\overline{๐ญ}_{}[]=๐ง_{}[]`$, an abstract nonlinear number field containing $`๐ญ_{}[]`$. We show that there exist flows
$$[\mathrm{\Phi }]:_{\mathrm{}}\mathrm{๐ฆ๐บ๐
}_{}(\overline{๐ญ}_{}[]/),[\mathrm{\Psi }]:_{\mathrm{}}\mathrm{๐ฆ๐บ๐
}_{}(\overline{๐ญ}_{}[]/),$$
where $`\mathrm{๐ฆ๐บ๐
}_{}`$, $`\mathrm{๐ฆ๐บ๐
}_{}`$ denote the groups of automorphisms of $`\overline{๐ญ}_{}[]`$ preserving $``$, $``$ only. Using the above theorem with $`L=^{\mathrm{ab}}`$ leads to the following
###### Theorem.
There are monomorphisms
$$๐ข_{}\mathrm{๐ฆ๐บ๐
}_{}\left(\overline{๐ญ}_{}[^{\mathrm{ab}}]/\right),๐ข_{}\mathrm{๐ฆ๐บ๐
}_{}\left(\overline{๐ญ}_{}[^{\mathrm{ab}}]/\right).$$
These results are proved in ยง10.
Remark: We have recently written some notes which expand on the ideas described in this paper.
Acknowledgements: We thank Michael McQuillen for having pointed out several errors in the original, published version. We would also like to thank the International Centre for Theoretical Physics, for providing generous support and congenial working conditions during the period in which this paper was written. This work was partially supported by proyecto CONACyT Sistemas Dinรกmicos G36357-E.
## 2. Solenoids
We review here a notion fundamental to this paper. References: , , , .
Let $`๐ณ`$ be a $`2^{\mathrm{nd}}`$ countable, Hausdorff space. An n-lamination is a $`2^{\mathrm{nd}}`$ countable, Hausdorff space $`๐ซ`$ equipped with a maximal atlas of homeomorphisms
$$\{\varphi _\alpha :๐ซU_\alpha V_\alpha ^n\times ๐ณ\},$$
in which every overlap $`\varphi _{\alpha \beta }=\varphi _\beta \varphi _\alpha ^1`$ is of the form
$$\varphi _{\alpha \beta }(๐ฑ,๐)=(h_{\alpha \beta }(๐ฑ,๐),f_{\alpha \beta }(๐)),$$
where $`๐h_{\alpha \beta }(,๐)`$ is a continuous family of homeomorphisms and $`f_{\alpha \beta }`$ is a homeomorphism. We call $`๐ซ`$ a foliation if $`๐ณ=^k`$, a solenoid if $`๐ณ`$ is a Cantor set.
Let $`\varphi `$ be a chart, $`D^n`$ an open disk, $`๐ณ^{}๐ณ`$ open. A flowbox is a subset of the form $`\varphi ^1(D\times ๐ณ^{})`$, a flowbox transversal a subset of the form $`\varphi ^1(๐ฑ\times ๐ณ^{})`$ and a plaque a subset of the form $`\varphi ^1(D\times \{๐\})`$. A leaf is a maximal continuation of overlapping plaques: by definition, an $`n`$-manifold. A riemannian metric on a smooth lamination is a family $`\gamma =\{\gamma _{\mathrm{}}\}`$ of smooth riemannian metrics, one on each leaf $`\mathrm{}`$, which when restricted to a flowbox gives a continuous family $`๐\gamma _๐`$ of smooth metrics. If $`๐ซ`$ has the structure of a topological group such that the multiplication and inversion maps take leaves to leaves, $`๐ซ`$ is called a topological group lamination. If in addition $`๐ซ`$ is smooth and the multiplication and inversion maps are smooth along the leaves, $`๐ซ`$ is called a Lie group lamination.
Let $`B`$ be a manifold, $`F`$ a $`2^{\mathrm{nd}}`$ countable Hausdorff space, $`\rho :\pi _1B\mathrm{๐ง๐๐๐พ๐}(F)`$ a representation. The quotient
$$๐ซ_\rho =\left(\stackrel{~}{B}\times F\right)/\pi _1B$$
by the action $`\alpha (\stackrel{~}{x},t)=(\alpha \stackrel{~}{x},\rho _\alpha (t))`$ is a lamination called the suspension of $`\rho `$. The projection $`๐ซ_\rho B`$ displays $`๐ซ_\rho `$ as a fiber bundle with fiber $`F`$, and the restriction of the projection to any leaf is a covering map.
The solenoids considered in this paper are modeled on the following example. Let $`B=๐^d`$ = the $`d`$-torus and let $`F=\widehat{}^d`$ = the profinite completion of $`^d`$. Let
$$\rho :\pi _1๐^d^d\mathrm{๐ง๐๐๐พ๐}(\widehat{}^d),\rho _๐ง(\widehat{๐ฆ})=\widehat{๐ฆ}๐ง$$
for $`๐ง^d`$ and $`\widehat{๐ฆ}\widehat{}^d`$. The associated suspension is called the $`d`$-dimensional torus solenoid $`\widehat{๐}^d`$. Its leaves are its path components, are dense, and may be identified with $`^d`$. Moreover, $`\widehat{๐}^d`$ is a compact, abelian, Lie group solenoid, and the additive group $`^d`$ sits inside as the path-component/leaf containing the identity.
One can extend the definition of a lamination to include infinite dimensional leaves modeled on a locally convex topological vector space $`V`$. If $`V`$ is a Hilbert space, we say we have a Hilbert space lamination. If $`V=^\omega `$ with the Tychonoff topology, we call it an $`^\omega `$-lamination. For Hilbert space laminations, one may use the Frรฉchet derivative to define smoothness, for $`^\omega `$-laminations one uses the Gรขteaux derivative. One may make sense of all other notions discussed above in the infinite-dimensional setting.
## 3. Adele Class Groups I: Finite Field Extensions
The material here is classical and can be found in standard texts on algebraic number theory: , , , . We review it in order to fix notation.
Let $`K`$ be an algebraic number field of degree $`d`$ over $``$, $`O_K`$ its ring of integers. If $`K`$ is Galois over $``$ we denote by $`\mathrm{๐ฆ๐บ๐
}(K/)`$ the Galois group. By a local field, we mean a locally compact field. A local field of characteristic 0 is either $``$, $``$ or a finite extension of $`_p`$ = the field of $`p`$-adic numbers.
Let $`\nu :KK_\nu `$ be an embedding where $`K_\nu `$ is a local field (necessarily of characteristic 0) and $`\nu (K)`$ is dense. Embeddings that are related by a continuous isomorphism of target fields different from complex conjugation are deemed equivalent, and an equivalence class of embedding is called a place. When $`K/`$ is Galois, we view $`\sigma \mathrm{๐ฆ๐บ๐
}(K/)`$ as acting on the left of the places via $`\sigma \nu :=\nu \sigma `$. In practice, we shall use the word place to mean a representative of an embedding equivalence class, and we will write $`q_\nu `$ for $`\nu (q)`$, $`qK`$.
When $`K_\nu `$ is isomorphic to $``$ or $``$, $`\nu `$ is said to be a real or complex infinite place. If $`K`$ has a complex place $`\mu `$, then there is an element $`\sigma _{\mathrm{๐ผ๐๐๐}}\mathrm{๐ ๐๐}(K)`$, called $``$-conjugation, such that $`\sigma _{\mathrm{๐ผ๐๐๐}}\mu =\overline{\mu }`$: so the complex places come in conjugate pairs $`(\mu ,\overline{\mu })`$. We have $`d=r+2s`$ where $`r=`$ the number of real places, $`2s=`$ the number of complex places. Let $`๐ซ_{\mathrm{}}=\{\nu _1,\mathrm{}\nu _r,\mu _1,\overline{\mu }_1,\mathrm{},\mu _s,\overline{\mu }_s\}`$, where $`\nu _j`$, $`j=1,\mathrm{},r`$ are the real places and the $`\mu _k,\overline{\mu }_k`$, $`k=1,\mathrm{},s`$, are the complex places. When $`K/`$ is Galois, the places are either all real or all complex, in which case we will refer to $`K`$ as being either real or complex.
If $`\nu `$ is not infinite, it is said to be finite. The set of finite places will be denoted $`๐ซ_{\mathrm{๐ฟ๐๐}}`$. If $`\nu ๐ซ_{\mathrm{๐ฟ๐๐}}`$, then $`K_\nu `$ has a maximal open subring $`O_\nu `$: its ring of integers. We note that $`K_\nu `$ is locally Cantor (totally-disconnected, perfect and locally compact) and $`O_\nu `$ is Cantor.
Denote
$$K_{\mathrm{}}=\left\{(z_\nu )^๐ซ_{\mathrm{}}=^d\right|\overline{z}_\nu =z_{\overline{\nu }}\}^r\times ^s^d.$$
Note that $`K_{\mathrm{}}`$ has a canonical inner-product induced from the hermitian inner product on $`^d`$, which decomposes as the usual inner product on $`^r`$ and the usual hermitian inner-product on $`^s`$. $`K`$ embeds diagonally into $`K_{\mathrm{}}`$ via $`q(q_\nu )`$ and the image of $`O_K`$ is a lattice in $`K_{\mathrm{}}`$. We shall identify $`K`$ and $`O_K`$ with their images in $`K_{\mathrm{}}`$. The quotient
$$๐_K=K_{\mathrm{}}/O_K$$
is called the Minkowski torus, a torus of (real) dimension $`d`$. When $`K/`$ is totally complex, $`๐_K`$ also has the structure of a complex torus.
The ring of finite adeles is the restricted product
$$๐ธ_K^{\mathrm{๐ฟ๐๐}}=\underset{\nu ๐ซ_{\mathrm{๐ฟ๐๐}}}{^{}}K_\nu $$
with respect to the $`O_\nu `$, $`\nu ๐ซ_{\mathrm{๐ฟ๐๐}}`$. By definition, this is the set of all tuples $`(q_\nu )`$ in which $`q_\nu O_\nu `$ for almost every $`\nu ๐ซ_{\mathrm{๐ฟ๐๐}}`$. $`๐ธ_K^{\mathrm{๐ฟ๐๐}}`$ is a locally Cantor topological ring. The adele ring is defined
$$๐ธ_K=K_{\mathrm{}}\times ๐ธ_K^{\mathrm{๐ฟ๐๐}}.$$
$`๐ธ_K`$ is a locally compact ring, and a solenoid as well since it is locally homeomorphic to (euclidean) $`\times `$ (Cantor). $`K`$ embeds diagonally in $`๐ธ_K`$ as a discrete co-compact subgroup with respect to addition. The quotient
$$\widehat{๐}_K=๐ธ_K/(K,+)$$
is called the adele class group associated to $`K`$.
Given $`๐`$ an ideal in $`O_K`$, denote by $`๐_๐`$ the quotient $`K_{\mathrm{}}/๐`$, a $`d`$-dimensional torus covering $`๐_K`$.
###### Proposition 1.
$`\widehat{๐}_K`$ is a $`d`$-dimensional euclidean Lie group solenoid, isomorphic to
1. The inverse limit of euclidean Lie groups
$$\underset{}{lim}๐_๐,$$
where $`๐`$ ranges over all ideals in $`O_K`$.
2. The suspension
$$\left(K_{\mathrm{}}\times \widehat{O}_K\right)/O_K,$$
where $`\widehat{O}_K=\underset{}{lim}O_K/๐`$.
###### Proof.
For item (1), see , pg. 67. Given $`(๐ณ,\widehat{\gamma })K_{\mathrm{}}\times \widehat{O}_K`$, note that $`๐ณ`$ defines an element $`\widehat{๐ณ}=(๐ณ_๐)`$ of $`\underset{}{lim}๐_๐`$ by projection on to each of the factors. Moreover, $`\widehat{\gamma }`$ is by definition a coherent sequence $`\{\gamma _๐\}`$ of deck transformations of the coverings $`๐_๐๐_K`$. Then the association $`(๐ณ,\widehat{\gamma })\widehat{\gamma }\widehat{๐ณ}=\left(\gamma _๐(๐ณ_๐)\right)`$ defines a homomorphism $`K_{\mathrm{}}\times \widehat{O}_K\widehat{๐}_K`$ which identifies precisely the $`O_K`$-related points, and descends to an isomorphism of $`\left(K_{\mathrm{}}\times \widehat{O}_K\right)/O_K`$ with $`\underset{}{lim}๐_๐`$. โ
By Proposition 1, it follows that the path-component of 0 is a leaf canonically isometric to $`K_{\mathrm{}}`$; the restriction of the projection $`\widehat{๐}_K๐_K`$ to $`K_{\mathrm{}}`$ is the quotient by $`O_K`$. We will endow $`\widehat{๐}_K`$ with the riemannian metric $`\rho `$ along the leaves induced from the inner-product on $`K_{\mathrm{}}`$.
For example, if $`K=`$, then $`๐_{}S^1`$ and $`\widehat{๐}_{}`$ is the classical 1-dimensional solenoid obtained as the inverse limit of circles $`/m`$ under the natural covering homomorphisms.
We now suppose that $`K/`$ is Galois and describe the actions of the Galois groups. If $`๐ณ=(z_\nu )K_{\mathrm{}}`$ and $`\sigma \mathrm{๐ฆ๐บ๐
}(K/)`$ then $`\sigma (๐ณ)=(z_{\sigma \nu })`$. Alternatively, we can write
$$K_{\mathrm{}}_{}K๐ธ_K๐ธ_{}_{}K$$
and the action of $`\sigma `$ on $`K_{\mathrm{}}`$ and on $`๐ธ_K`$ is via $`xqx\sigma (q)`$, where $`x`$ or $`๐ธ_{}`$ as the case may be. In any event it is clear that $`\mathrm{๐ฆ๐บ๐
}(K/)`$ acts orthogonally. The action of $`\sigma `$ on $`๐ธ_K`$ may be understood as the product of the actions on $`K_{\mathrm{}}`$ and $`๐ธ_K^{\mathrm{๐ฟ๐๐}}`$.
Note that the image of $`O_K`$ resp. $`K`$ is preserved by $`\sigma `$ and we induce (leafwise) isometric isomorphisms
$$\sigma :๐_K๐_K\widehat{\sigma }:\widehat{๐}_K\widehat{๐}_K,$$
which are intertwined by the projection $`p:\widehat{๐}_K๐_K`$ in that $`p\widehat{\sigma }=\sigma p`$. This leads to representations
$$\rho :\mathrm{๐ฆ๐บ๐
}(K/)\mathrm{๐จ๐๐๐}(๐_K)\widehat{\rho }:\mathrm{๐ฆ๐บ๐
}(K/)\mathrm{๐จ๐๐๐}(\widehat{๐}_K),$$
where $`\mathrm{๐จ๐๐๐}()`$ means the group of isometric isomorphisms.
Let $`L/K`$ be a finite extension of number fields. Any place of $`L`$, finite or not, defines one on $`K`$ by restriction. We thus obtain injective inclusions of vector spaces resp. rings
$$K_{\mathrm{}}L_{\mathrm{}},๐ธ_K^{\mathrm{๐ฟ๐๐}}๐ธ_L^{\mathrm{๐ฟ๐๐}},๐ธ_K๐ธ_L$$
which scale inner-products/metrics by a factor of $`\mathrm{deg}(L/K)`$. These maps in turn induce injective homomorphisms
(2)
$$๐_K๐_L,\widehat{๐}_K\widehat{๐}_L$$
which scale metrics by $`\mathrm{deg}(L/K)`$ and whose images are, in case $`L/K`$ is Galois, fixed by the action of $`\mathrm{๐ฆ๐บ๐
}(L/K)`$.
We may also define maps in the opposite direction through the trace map $`\mathrm{๐ณ๐}_{L/K}:LK`$, which, when $`L/K`$ is Galois, is given by
$$\mathrm{๐ณ๐}_{L/K}(\alpha )=\underset{\sigma \mathrm{๐ฆ๐บ๐
}(L/K)}{}\sigma (\alpha ),$$
and which has the property that $`\mathrm{๐ณ๐}_{L/K}(O_L)O_K`$. The trace map extends to a linear map $`\mathrm{๐ณ๐}_{L/K}:L_{\mathrm{}}K_{\mathrm{}}`$ as follows: if $`\mathrm{๐ณ๐}_{L/K}(๐ณ)=๐ฐ=(w_\nu )`$ then
$$w_\nu =\underset{\nu ^{}}{}z_\nu ^{}$$
where the sum is over $`\nu ^{}`$ that restrict to $`\nu `$. Note that if $`\nu `$ is a real and $`\nu ^{}`$ is complex, then complex conjugation enters into the above sum, so that the contribution from $`z_\nu ^{}`$ is $`2\mathrm{R}\mathrm{e}(z_\nu ^{})`$. We thus induce epimorphisms
$$\mathrm{๐ณ๐}_{L/K}:๐_L๐_K,\widehat{๐ณ}๐_{L/K}:\widehat{๐}_L\widehat{๐}_K$$
which are, in case $`L/K`$ is Galois, $`\mathrm{๐ฆ๐บ๐
}(L/K)`$-equivariant: $`\mathrm{๐ณ๐}_{L/K}\sigma =\mathrm{๐ณ๐}_{L/K}`$ and $`\widehat{๐ณ}๐_{L/K}\widehat{\sigma }=\widehat{๐ณ}๐_{L/K}`$ for all $`\sigma \mathrm{๐ฆ๐บ๐
}(L/K)`$.
For the extension $`K/`$ we will write the associated trace map $`\mathrm{๐ณ๐}=\mathrm{๐ณ๐}_{K/}`$, referring to it as the absolute trace. The ideal
$$๐ก_K=\{\alpha K|\mathrm{๐ณ๐}(\alpha O_K)\}^1O_K$$
is called the absolute different.
## 4. Adele Class Groups II: Infinite Field Extensions
Let $`๐ฆ`$ be a field occurring as a direct limit $`\underset{}{lim}K_\lambda `$ of fields $`K_\lambda /`$ of finite degree e.g. $`\overline{}`$ = the algebraic closure of $``$ or $`^{\mathrm{ab}}`$ = the maximal abelian extension of $``$. We may associate to $`๐ฆ`$ (non locally-compact) abelian groups by taking the induced direct limits of inclusions (2):
(3)
$$\underset{}{lim}๐_{K_\lambda },\underset{}{lim}\widehat{๐}_{K_\lambda }.$$
The projections $`\widehat{๐}_{K_\lambda }๐_{K_\lambda }`$ induce a projection $`\underset{}{lim}\widehat{๐}_{K_\lambda }\underset{}{lim}๐_{K_\lambda }`$. If the $`K_\lambda `$ are Galois over $``$, then the inverse system of Galois groups $`\left\{\mathrm{๐ฆ๐บ๐
}(K_\lambda /)\right\}`$ acts compatibly on the direct systems of tori and solenoids, inducing an action of the profinite Galois group $`\mathrm{๐ฆ๐บ๐
}(๐ฆ/)=\underset{}{lim}\mathrm{๐ฆ๐บ๐
}(K_\lambda /)`$ on each of the spaces appearing in (3).
Consider also
$$\underset{}{lim}(K_\lambda )_{\mathrm{}},\underset{}{lim}๐ธ_{K_\lambda }^{\mathrm{๐ฟ๐๐}},\underset{}{lim}๐ธ_{K_\lambda },$$
the first a (non locally-compact) topological vector space, the last two (non locally-compact) topological rings. Note that we may identify
$$\underset{}{lim}๐ธ_{K_\lambda }\underset{}{lim}(K_\lambda )_{\mathrm{}}\times \underset{}{lim}๐ธ_{K_\lambda }^{\mathrm{๐ฟ๐๐}},$$
since the direct limit maps preserve the solenoid product structure $`(\text{euclidean})\times (\text{locally Cantor})`$ of the adele spaces. There are natural inclusions $`O_๐ฆ\underset{}{lim}(K_\lambda )_{\mathrm{}}`$ and $`๐ฆ\underset{}{lim}๐ธ_{K_\lambda }`$, and
$$\underset{}{lim}๐_{K_\lambda }\left(\underset{}{lim}(K_\lambda )_{\mathrm{}}\right)/O_๐ฆ\underset{}{lim}\widehat{๐}_{K_\lambda }\left(\underset{}{lim}๐ธ_{K_\lambda }\right)/๐ฆ.$$
It follows that $`\underset{}{lim}\widehat{๐}_{K_\lambda }`$ is a lamination, each leaf of which may be identified with $`\underset{}{lim}(K_\lambda )_{\mathrm{}}`$. Although $`\underset{}{lim}๐ธ_{K_\lambda }^{\mathrm{๐ฟ๐๐}}`$ is totally disconnected and perfect, it is not locally compact since the direct limit maps are not open. Thus $`\underset{}{lim}\widehat{๐}_{K_\lambda }`$ is not a solenoid.
An inner-product on the topological vector space $`\underset{}{lim}(K_\lambda )_{\mathrm{}}`$ is defined by scaling the canonical inner-product on each summand $`(K_\lambda )_{\mathrm{}}`$ by $`(\mathrm{deg}(K_\lambda /))^1`$ and taking the direct limit. The action of $`O_๐ฆ`$ on $`\underset{}{lim}(K_\lambda )_{\mathrm{}}`$ preserves this inner-product and we induce a riemannian metric on $`\underset{}{lim}๐_{K_\lambda }`$. Similarly, the action of $`๐ฆ`$ on $`\underset{}{lim}๐ธ_{K_\lambda }`$ is isometric along the factor $`\underset{}{lim}(K_\lambda )_{\mathrm{}}`$, and we induce a leaf-wise riemannian metric on $`\underset{}{lim}\widehat{๐}_{K_\lambda }`$.
The completion of $`\underset{}{lim}(K_\lambda )_{\mathrm{}}`$ is a Hilbert space denoted $`๐ฆ_{\mathrm{}}`$. Write
$$๐ธ_๐ฆ=๐ฆ_{\mathrm{}}\times \underset{}{lim}๐ธ_{K_\lambda }^{\mathrm{๐ฟ๐๐}}.$$
We define the hilbertian torus resp. the hilbertian adele class group of $`๐ฆ`$ by
$$๐_๐ฆ๐ฆ_{\mathrm{}}/O_๐ฆ\widehat{๐}_๐ฆ๐ธ_๐ฆ/๐ฆ.$$
Thus $`\widehat{๐}_๐ฆ`$ is a Hilbert space lamination whose leaves are isometric to $`๐ฆ_{\mathrm{}}`$. When the system is Galois, the Galois actions on the summands of the direct limits preserve the scaled inner-products, and we obtain representations
$$\mathrm{๐ฆ๐บ๐
}(๐ฆ/)\mathrm{๐จ๐๐๐}(๐_๐ฆ)\mathrm{๐ฆ๐บ๐
}(๐ฆ/)\mathrm{๐จ๐๐๐}(\widehat{๐}_๐ฆ).$$
Any ideal $`๐O_๐ฆ`$ can be realized as the direct limit of ideals $`๐_\lambda =๐O_{K_\lambda }`$. The quotient $`๐_๐=๐ฆ_{\mathrm{}}/๐`$ will be referred to as the hilbertian torus of $`๐`$. Consider the inverse limit
$$\widehat{O}_๐ฆ=\underset{}{lim}O_๐ฆ/๐$$
as $`๐`$ ranges over ideals in $`O_๐ฆ`$. Since ideals in $`O_๐ฆ`$ need not have finite index, $`\widehat{O}_๐ฆ`$ is not compact, or even locally compact. Nevertheless,
###### Proposition 2.
$`\widehat{O}_๐ฆ\underset{}{lim}\widehat{O}_{K_\lambda }`$.
###### Proof.
Observe first that for any ideal $`๐O_๐ฆ`$, we have $`O_๐ฆ/๐\underset{}{lim}O_{K_\lambda }/๐_\lambda `$. If $`๐๐`$, we have a commutative diagram
$$\begin{array}{ccccc}O_๐ฆ/๐& & & & O_๐ฆ/๐\\ & & & & \\ & & & & \\ & & & & \\ O_{K_\lambda }/๐_\lambda & & & & O_{K_\lambda }/๐_\lambda \end{array}.$$
This allows us to interchange direct and inverse limits and write:
$$\widehat{O}_๐ฆ=\underset{}{lim}O_๐ฆ/๐\underset{}{lim}\left(\underset{}{lim}O_{K_\lambda }/๐_\lambda \right)\underset{}{lim}\widehat{O}_{K_\lambda }$$
and the result follows. โ
###### Proposition 3.
$`\widehat{๐}_๐ฆ`$ is isomorphic to
1. The inverse limit of hilbertian tori
$$\underset{}{lim}๐_๐$$
as $`๐`$ ranges over ideals in $`O_๐ฆ`$.
2. The suspension
$$\left(๐ฆ_{\mathrm{}}\times \widehat{O}_๐ฆ\right)/O_๐ฆ.$$
###### Proof.
The spaces appearing in items (1) and (2) are isomorphic by an argument identical to that appearing in Proposition 1. That $`\widehat{๐}_๐ฆ`$ is isomorphic to the inverse limit appearing in (1) follows from the same limit interchange argument used in the proof of Proposition 2. โ
Unfortunately, neither $`๐_๐ฆ`$ nor $`\widehat{๐}_๐ฆ`$ are locally compact. This creates complications, since harmonic analysis plays a fundamental role in the linking of the arithmetic of $`๐ฆ`$ and the algebra of the Hilbert space of $`L^2`$ functions on $`\widehat{๐}_๐ฆ`$. For this reason, we consider in the next section compactifications coming from inverse limits of tori and solenoids.
## 5. Proto Adele Class Groups
Let $`๐ฆ=\underset{}{lim}K_\lambda `$ be a direct limit of finite extensions over $``$. The trace maps induce an inverse limit
$$\widehat{๐ฆ}=\underset{}{lim}K_\lambda ,$$
an abelian group with respect to addition. This system restricts to one of integers, however:
###### Theorem 1.
Let $`๐ฆ`$ be a field containing $`^{\mathrm{ab}}`$. Then
$$\underset{}{lim}O_{K_\lambda }=\{0\}.$$
###### Proof.
Let $`\omega `$ be a primitive $`m`$th root of unity and consider the cyclotomic extension $`K=(\omega )`$. Since $`K`$ is abelian, by assumption it occurs in the direct system defining $`๐ฆ`$. The ring of integers $`O_K`$ is generated by $`1`$ and $`\omega ^j`$, where $`1jd1`$ and $`d`$ is the degree of $`K/`$. If we take say $`m=2^k`$, then $`d=2^{k1}`$ and
$$\mathrm{๐ณ๐}(\omega ^j)=\underset{\sigma \mathrm{๐ฆ๐บ๐
}(K/)}{}\sigma (\omega ^j)=\mathrm{\hspace{0.33em}\hspace{0.33em}0}$$
for each $`j1`$. On the other hand, $`\mathrm{๐ณ๐}(1)=d`$, so it follows that $`\mathrm{๐ณ๐}(O_K)=(d)`$. Since $`d`$ can be taken arbitrarily large, this means that the only coherent sequence that we may form in the inverse limit of rings of integers is $`(0,0,\mathrm{})`$. โ
###### Note 1.
It is worth pointing out that normalizing the trace map by dividing by the degree would not produce a non-trival limit. Indeed, if we let $`\omega `$ be a primitive $`p`$th root of unity, $`p`$ a prime $`>2`$, then $`\mathrm{๐ณ๐}(O_K)=`$. Normalizing would take us out of the integers. Thus, all that survives in the trace inverse limit is a kind of โscaledโ additive number theory.
The trace inverse limits
$$\widehat{๐}_๐ฆ=\underset{}{lim}๐_{K_\lambda },\widehat{\widehat{๐}}_๐ฆ=\underset{}{lim}\widehat{๐}_{K_\lambda }$$
are called, respectively, the proto-torus and the proto-adele class group of $`๐ฆ`$. Each is a compact abelian group, being inverse limits of the same. The trace maps are natural with respect to the epimorphisms $`\widehat{๐}_{K_\lambda }๐_{K_\lambda }`$ and induce in turn an epimorphism $`\widehat{\widehat{๐}}_๐ฆ\widehat{๐}_๐ฆ`$.
Consider as well the inverse limits
$$\widehat{๐ฆ}_{\mathrm{}}=\underset{}{lim}(K_\lambda )_{\mathrm{}},\widehat{๐ธ}_๐ฆ^{\mathrm{๐ฟ๐๐}}=\underset{}{lim}๐ธ_{K_\lambda }^{\mathrm{๐ฟ๐๐}},\widehat{๐ธ}_๐ฆ=\underset{}{lim}๐ธ_{K_\lambda }.$$
Observe that $`\widehat{๐ฆ}_{\mathrm{}}`$ is a locally convex, (non locally compact) topological vector space whose topology is induced from an embedding in $`^\omega `$. Moreover, $`\widehat{๐ธ}_๐ฆ^{\mathrm{๐ฟ๐๐}}`$ and $`\widehat{๐ธ}_๐ฆ`$ are abelian topological groups only. Note that
$$\widehat{๐ธ}_๐ฆ\widehat{๐ฆ}_{\mathrm{}}\times \widehat{๐ธ}_๐ฆ^{\mathrm{๐ฟ๐๐}}.$$
The space $`\widehat{๐ธ}_๐ฆ^{\mathrm{๐ฟ๐๐}}`$ is totally-disconected, perfect but not locally compact. There is a natural inclusion $`\widehat{๐ฆ}\widehat{๐ธ}_๐ฆ`$ and
$$\widehat{\widehat{๐}}_๐ฆ\widehat{๐ธ}_๐ฆ/\widehat{๐ฆ}.$$
Since the action of $`\widehat{๐ฆ}`$ locally preserves the product structure, $`\widehat{\widehat{๐}}_๐ฆ`$ is an $`^\omega `$-lamination. When the system is Galois, the trace system maps are compatible with the Galois actions and we obtain representations of $`\mathrm{๐ฆ๐บ๐
}(๐ฆ/)`$ on $`\widehat{๐}_๐ฆ`$ and on $`\widehat{\widehat{๐}}_๐ฆ`$, acting by (leaf-preserving) topological isomorphisms.
###### Theorem 2.
Let $`๐ฆ=\underset{}{lim}K_\lambda `$ be an infinite field extension containing $`^{\mathrm{ab}}`$. Then the covering homomorphisms $`(K_\lambda )_{\mathrm{}}๐_{K_\lambda }`$ induce a continuous monomorphism
$$\widehat{๐ฆ}_{\mathrm{}}\widehat{๐}_๐ฆ.$$
###### Proof.
The inverse limits giving rise to each of $`\widehat{๐ฆ}_{\mathrm{}}`$ and $`\widehat{๐}_๐ฆ`$ also give rise to a continuous homomorphism $`\widehat{๐ฆ}_{\mathrm{}}\widehat{๐}_๐ฆ`$. An element in the kernel must be a coherent sequence of algebraic integers i.e. an element of $`\underset{}{lim}O_{K_\lambda }`$. By Theorem 1, the latter is trivial and the map is an injective. โ
The map $`\widehat{๐ฆ}_{\mathrm{}}\widehat{๐}_๐ฆ`$ is not surjective since the fibers of the projections $`(K_\lambda )_{\mathrm{}}๐_{K_\lambda }`$ are not compact. Nonetheless, the image of $`\widehat{๐ฆ}_{\mathrm{}}`$ is dense in the compact $`\widehat{๐}_๐ฆ`$. Thus $`\widehat{๐}_๐ฆ`$ is laminated by the cosets of the image of $`\widehat{๐ฆ}_{\mathrm{}}`$. In fact,
###### Theorem 3.
Let $`๐ฆ=\underset{}{lim}K_\lambda `$ be an infinite field extension containing $`^{\mathrm{ab}}`$. Then the trace map inverse systems induce an isomorphism
$$\widehat{\widehat{๐}}_๐ฆ\widehat{๐}_๐ฆ$$
of topological groups.
###### Proof.
For each $`K_\lambda `$, the kernel of $`\widehat{๐}_{K_\lambda }๐_{K_\lambda }`$ is $`\widehat{O}_{K_\lambda }`$, hence the kernel of $`\widehat{\widehat{๐}}_๐ฆ\widehat{๐}_๐ฆ`$ is $`\underset{}{lim}\widehat{O}_{K_\lambda }=0`$. Since the fibers of the projections $`\widehat{๐}_{K_\lambda }๐_{K_\lambda }`$ are compact, the induced map of limits $`\widehat{\widehat{๐}}_๐ฆ\widehat{๐}_๐ฆ`$ is surjective. Since each map $`\widehat{๐}_{K_\lambda }๐_{K_\lambda }`$ is open with compact fibers, the limit map $`\widehat{\widehat{๐}}_๐ฆ\widehat{๐}_๐ฆ`$ is also open. โ
When $`^{\mathrm{ab}}๐ฆ`$, there are no exact analogues of Propositions 1 and 3, due to the lack of a notion of integers in $`\widehat{๐ฆ}`$. On the other hand, given any subfield $`K๐ฆ`$ of finite degree over $``$, one can find โlevel-$`K`$โ suspension structures on $`\widehat{๐}_๐ฆ\widehat{\widehat{๐}}_๐ฆ`$. Specifically, it is not difficult to see that the kernel of the projection $`\widehat{๐}_๐ฆ๐_K`$ is isomorphic to
$$\widehat{O}_K\times \widehat{๐}_{๐ฆ/K},$$
where $`\widehat{๐}_{๐ฆ/K}`$ is the inverse limit of tori occurring as the connected components of $`0`$ of the kernels of the projections $`๐_L๐_K`$. Then
$$\widehat{๐}_๐ฆ\left(K_{\mathrm{}}\times \left(\widehat{O}_K\times \widehat{๐}_{๐ฆ/K}\right)\right)/O_K.$$
These representations are related by homeomorphisms induced by $`\mathrm{๐ณ๐}:LK`$, but they do not survive the trace inverse limit. Similar โlevel-$`K`$โ suspension representations are available for $`\widehat{\widehat{๐}}_๐ฆ`$ as well.
The relationship between the proto constructions of this chapter with the hilbertian constructions of the previous chapter is described by the following
###### Theorem 4.
There are canonical inclusions
$$๐_๐ฆ\widehat{๐}_๐ฆ,๐ฆ_{\mathrm{}}\widehat{๐ฆ}_{\mathrm{}},\widehat{๐}_๐ฆ\widehat{\widehat{๐}}_๐ฆ$$
with dense images, which are $`\mathrm{๐ฆ๐บ๐
}(๐ฆ/)`$-equivariant in case the defining system is Galois.
###### Proof.
We begin by defining the inclusion $`\underset{}{lim}(K_\lambda )_{\mathrm{}}\widehat{๐ฆ}_{\mathrm{}}`$. Let $`๐ณ_{\lambda _0}(K_{\lambda _0})_{\mathrm{}}`$ and define $`๐ณ_{\lambda _0}(๐ณ_\lambda )`$ as follows. If $`\lambda `$ is an index below $`\lambda _0`$, project $`๐ณ_{\lambda _0}`$ by the appropriate trace map to get the $`๐ณ_\lambda `$ coordinate. If $`\lambda `$ is above $`\lambda _0`$, we map $`๐ณ_{\lambda _0}`$ into $`(K_\lambda )_{\mathrm{}}`$ by the inclusion $`(K_{\lambda _0})_{\mathrm{}}(K_\lambda )_{\mathrm{}}`$ and scale by $`1/d`$ where $`d`$ is the degree of $`K_\lambda /K_{\lambda _0}`$. This prescription defines a coherent sequence hence an element of $`\widehat{๐ฆ}`$. This map clearly defines an injective homomorphism of vector spaces $`\underset{}{lim}(K_\lambda )_{\mathrm{}}\widehat{๐ฆ}`$. By virtue of the scaling, this map is also continuous with regard to the inner-product defining $`๐ฆ_{\mathrm{}}`$. Now let $`\{๐ณ_{\lambda _i}\}\underset{}{lim}(K_\lambda )_{\mathrm{}}`$ be a Cauchy sequence defining an element of $`๐ฆ_{\mathrm{}}`$. For every index $`\beta `$, if we let $`\mathrm{๐ณ๐}_{\lambda _i,\beta }`$ denote the trace map from $`(K_{\lambda _i})_{\mathrm{}}`$ to $`(K_\beta )_{\mathrm{}}`$, then $`\left\{\mathrm{๐ณ๐}_{\lambda _i,\beta }(๐ณ_{\lambda _i})\right\}`$ is Cauchy in $`(K_\beta )_{\mathrm{}}`$. But this means precisely that the image of $`\{๐ณ_{\lambda _i}\}`$ in $`\widehat{๐ฆ}_{\mathrm{}}`$ is convergent. Thus the inclusion $`\underset{}{lim}(K_\lambda )_{\mathrm{}}\widehat{๐ฆ}_{\mathrm{}}`$ extends to $`๐ฆ_{\mathrm{}}`$. The other inclusions are induced by this one. That the images are dense follows from the definition of the above map: given any element of $`\widehat{๐ฆ}_{\mathrm{}}`$ and any level $`\lambda `$, there is an element of $`๐ฆ_{\mathrm{}}`$ agreeing up to level $`\lambda `$. โ
## 6. Hyperbolizations
Let $`K/`$ be finite. For each real place $`\nu `$ one pairs $`K_\nu `$ with a single factor of $`(0,\mathrm{})`$ to form the upper half-plane factor $`_\nu =\times i(0,\mathrm{})`$, which we equip with the hyperbolic metric in the usual way.
For a complex place pair $`(\mu ,\overline{\mu })`$, we must alter slightly this prescription in order to take into account the complex algebra of the factor
$$_{(\mu ,\overline{\mu })}=\{(z,\overline{z})|z\}.$$
In order to do this, it is convenient to regard the factor $`(0,\mathrm{})\times (0,\mathrm{})`$ to be attached to $`_{(\mu ,\overline{\mu })}`$ as being complex. Thus let $`๐น`$ be the complex quarter space
$$๐น=\{b=s+it|\mathrm{\hspace{0.33em}\hspace{0.33em}0}<s,t\}$$
and define
$$_{(\mu ,\overline{\mu })}=\{(z,\overline{z})\times (b,\overline{b})|z,b๐น\}(\times ๐น)\times (\overline{}\times (\overline{๐น}))^2\times ^2.$$
Let $`i^2`$ denote the right half-plane. Then we may identify $`_{(\mu ,\overline{\mu })}`$ with $`^2i^2`$ via the map
$$(z,\overline{z})\times (b,\overline{b})\frac{1}{2}(z+\overline{z}+b\overline{b},z\overline{z}+b+\overline{b})=(x+it,s+iy)(u,v)$$
for $`z=x+iy`$. We give $`_{(\mu ,\overline{\mu })}`$ the product hyperbolic metric coming from this identification, and also write
(4)
$$_{(\mu ,\overline{\mu })}=_\mu i_\mu $$
so as to be able to refer to the upper and right half-planes associated to $`(\mu ,\overline{\mu })`$. In the sequel, a function defined in $`_{(\mu ,\overline{\mu })}`$ will be considered holomorphic if it is holomorphic in each of the variables $`(u,v)=(x+it,s+iy)`$ separately.
Notice that the complex conjugation acting in $`_{(\mu ,\overline{\mu })}`$ extends to $`_{(\mu ,\overline{\mu })}`$ by the identity in the $`b`$ variable, and that in the $`(u,v)`$ variables, acts via
$$(u,v)(u,\overline{v}),$$
thus defining an (orientation-reversing) isometry of $`_{(\mu ,\overline{\mu })}`$.
Finally, we define the hyperbolization
$$_K=_K^{}\times _K^{}:=_\nu \times _{(\mu ,\overline{\mu })}=_\nu \times (_\mu i_\mu )K_{\mathrm{}}\times (0,\mathrm{})^d.$$
Thus $`_K`$ has the structure of a $`d`$-dimensional complex polydisk equipped with the product riemannian metric. For $`\nu `$ a real place we write $`\tau _\nu =x_\nu +it_\nu `$ for a point of $`_\nu `$, and for $`\mu `$ complex we write
$$(\kappa _\mu ,\overline{\kappa }_\mu ):=((z_\mu ,b_\mu ),(\overline{z}_\mu ,\overline{b}_\mu ))(u_\mu ,v_\mu ).$$
We write points of $`_K`$ in the form $`๐=๐\times ๐ฟ`$ where $`๐=(\tau _1,\mathrm{},\tau _r)_K^{}`$ are the coordinates of the โreal hyperbolizationโ and $`๐ฟ=((\kappa _1,\overline{\kappa }_1),\mathrm{},(\kappa _s,\overline{\kappa }_s))_K^{}`$ are those of the โcomplex hyperbolizationโ. When $`K/`$ is Galois we have $`_K=_K^{}`$ or $`_K^{}`$ depending on whether $`K`$ is real or complex, and the action of $`\mathrm{๐ฆ๐บ๐
}(K/)`$ extends to an isometric action on $`_K`$ by acting trivially in the extended coordinates.
The subgroups $`O_K`$ and $`K`$ of $`K_{\mathrm{}}`$, viewed as groups of translations, extend to translations of $`_K`$ that are isometries. The quotients
(5)
$$๐_K=_K/O_K(\mathrm{\Delta }^{})^d,\widehat{๐}_K=\left(_K\times ๐ธ_K^{\mathrm{๐ฟ๐๐}}\right)/K\widehat{๐}_K\times (0,\mathrm{})^d$$
are referred to as the hyperbolized torus and the hyperbolized adele class group of $`K`$. (In the above, $`\mathrm{\Delta }^{}`$ denotes the punctured hyperbolic disk.) For $`L/K`$ a finite extension of finite degree extensions of $``$, the canonical inclusions $`๐_K๐_L`$ and $`\widehat{๐}_K\widehat{๐}_L`$ are isometric up to the scaling factor $`\mathrm{deg}(L/K)`$. If $`L/K`$ is Galois, the action of $`\mathrm{๐ฆ๐บ๐
}(L/K)`$ on $`๐_L`$ and $`\widehat{๐}_L`$ extends by isometries to both of $`๐_L`$ and $`\widehat{๐}_L`$ restricting to the identity on the images of $`๐_K`$ and $`\widehat{๐}_K`$.
In the case of an infinite field extension $`๐ฆ`$, one follows the prescription of the preceding paragraphs using the hilbertian torus and solenoid $`๐_๐ฆ`$ and $`\widehat{๐}_๐ฆ`$. Thus $`_๐ฆ`$ is an infinite product of hyperbolic planes. We obtain an infinite-dimensional hyperbolized torus $`๐_๐ฆ`$ and hyperbolized solenoid $`\widehat{๐}_๐ฆ`$ upon quotient by $`O_๐ฆ`$ and $`๐ฆ`$.
## 7. The Character Field
For $`G`$ a locally compact abelian group, by a character we mean a continuous homomorphism $`\chi :GU(1)`$, where $`U(1)`$ is the unit circle. An operation $``$ on characters is defined by multiplying their values, and the set of characters
$$\mathrm{๐ข๐๐บ๐}(G)=\mathrm{๐ง๐๐}_{\mathrm{cont}}(G,U(1))$$
is itself a locally compact abelian group. In the case of interest, the projection $`\widehat{๐}_K๐_K`$ induces an inclusion
$$\mathrm{๐ข๐๐บ๐}(๐_K)\mathrm{๐ข๐๐บ๐}(\widehat{๐}_K)$$
and so we will view $`\mathrm{๐ข๐๐บ๐}(๐_K)`$ as a subgroup of $`\mathrm{๐ข๐๐บ๐}(\widehat{๐}_K)`$. If $`๐ฆ=\underset{}{lim}K_\lambda `$ is an infinite extension over $``$, we have a similar inclusion
$$\mathrm{๐ข๐๐บ๐}(\widehat{๐}_๐ฆ)\mathrm{๐ข๐๐บ๐}(\widehat{\widehat{๐}}_K)$$
and corresponding convention.
The purpose of this section is to give a proof of the well-known identification $`\mathrm{๐ข๐๐บ๐}(\widehat{๐}_K)(K,+)`$, and show that this identification may be used to view $`\mathrm{๐ข๐๐บ๐}(\widehat{๐}_K)`$ as a field, in a way which is natural with respect to the trace maps.
###### Lemma 1.
Let $`K/`$ be a finite extension. If $`๐ฐK_{\mathrm{}}`$ has the property that
$$\mathrm{๐ณ๐}(๐ฐK),$$
then $`๐ฐK`$.
###### Proof.
First suppose that $`K`$ is Galois over $``$. Let $`\alpha _1,\mathrm{},\alpha _d`$ be an integral basis of $`K`$, and let $`A`$ be the $`d\times d`$ invertible matrix whose $`ij`$-element is $`\nu _j(\alpha _i)`$ where the $`\nu _1,\mathrm{},\nu _d`$ are the places of $`K`$. Then the hypothesis $`\mathrm{Tr}(๐ฐK)`$ implies that $`A๐ฐ=๐ช^d`$ or $`๐ฐA^1^d`$. Since $`K`$ is Galois, it is normal, hence all of the entries of $`A`$ belong to $`K`$ (see ), thus the coordinates of $`๐ฐ`$ belong to $`K`$. Suppose that nevertheless $`๐ฐK`$. Then there exists some element $`\sigma _0\mathrm{๐ฆ๐บ๐
}(K/)`$ and a coordinate $`w_{\nu _0}`$ such that $`\sigma _0(w_{\nu _0})w_{\sigma _0\nu _0}`$. Let $`A^\nu `$ denote the column indexed by a place $`\nu `$ (i.e. the vector $`(\nu (\alpha _1),\mathrm{},\nu (\alpha _d))^T`$) so that
(6)
$$w_\nu A^\nu =๐ช^d.$$
Acting by $`\sigma _0`$ on this equation fixes $`๐ช`$, permutes the column vectors $`A^\nu `$, $`\sigma _0(A^\nu )=A^{\sigma _0\nu }`$, but does not similarly permute the entries of $`๐ฐ`$. Thus there exists a vector $`๐ฐ^{}๐ฐ`$, defined $`w_{\sigma _0\nu }^{}=\sigma _0(w_\nu )`$, for which $`A๐ฐ^{}=๐ช`$, implying that the kernel of $`A`$ is nontrivial (since $`A(๐ฐ^{}๐ฐ)=0`$ and $`๐ฐ^{}๐ฐ0`$), contradiction.
Now suppose that $`K/`$ is not Galois. As in the previous paragraph, fix the basis $`\alpha _1,\mathrm{},\alpha _d`$ and note that the columns of the embedding matrix $`A`$ continue to satisfy the vector equation (6). Let $`L/`$ be a finite Galois extension containing the images of $`K`$ by its places: for example, if one writes $`K=(\alpha )`$ one can take $`L`$ to be the splitting field of the minimal polinomial of $`\alpha `$. Note then that $`L/K`$ is Galois. Then $`A`$ has entries belonging to $`L`$, as does its inverse, so that each coordinate of $`๐ฐ`$ belongs to $`L`$. Consider the usual diagonal embedding of vector spaces $`\iota :K_{\mathrm{}}L_{\mathrm{}}`$ defined by
$$\iota (๐ฐ)_\mu =w_\nu $$
for each $`\mu `$ an $`L`$-place with restriction $`\mu |_K=\nu `$. In addition, we have the scaled embedding $`\overline{\iota }:K_{\mathrm{}}L_{\mathrm{}}`$,
$$\overline{\iota }(๐ฐ)_\mu =\frac{w_\nu }{\mathrm{mult}(\nu )},$$
where $`\mu |_K=\nu `$ and $`\mathrm{mult}(\nu )`$ is the number of places having restriction to $`K`$ equal to $`\nu `$. The basis $`\alpha _1,\mathrm{},\alpha _d`$ defines a $`d\times (d[L:K])`$ matrix, denoted $`B`$, whose $`\mu `$th-column is $`B^\mu =(\overline{\iota }(\alpha _1)_\mu ,\mathrm{},(\overline{\iota }(\alpha _d)_\mu )^T`$ (here we are identifying the basis elements with vectors in $`K_{\mathrm{}}`$). Since $`A๐ฐ=๐ช^d`$, we have $`B\iota (๐ฐ)=๐ช^d`$ and we have the analogue of the vector equation (6):
$$\iota (๐ฐ)_\mu B^\mu =๐ช^d.$$
Note that if $`\mu `$ and $`\mu ^{}`$ have the same restriction $`\nu `$ to $`K`$ then $`B^\mu =B^\mu ^{}`$.
Suppose that $`๐ฐ`$ does not belong to $`K`$. Then $`\iota (๐ฐ)`$ does not belong to $`L`$: for if $`\iota (๐ฐ)=\beta L`$, then by definition of $`\iota `$, for all $`\mu `$ extending the identity place on $`K`$, we must have $`\mu (\beta )=\beta `$. But the places extending the identity place on $`K`$ comprise the Galois group $`\mathrm{๐ฆ๐บ๐
}(L/K)`$, hence $`\beta `$ belongs to the fixed field of $`\mathrm{๐ฆ๐บ๐
}(L/K)`$ i.e. $`\iota (๐ฐ)K`$ viewed as a subfield of $`L_{\mathrm{}}`$. But then this would imply that $`๐ฐK`$ contrary to our hypothesis.
Since $`\iota (๐ฐ)`$ does not belong to $`L`$, there exists $`\sigma _0\mathrm{๐ฆ๐บ๐
}(L/)`$ and a coordinate $`\mu _0`$ such that
$$\sigma _0(\iota (๐ฐ)_{\mu _0})\iota (๐ฐ)_{\sigma _0\mu _0}.$$
Then as before we deduce a vector $`๐ฒ`$, defined $`y_{\sigma _0\mu }:=\sigma _0(\iota (๐ฐ)_\mu )`$, and for which $`B๐ฒ=๐ช`$. This vector $`๐ฒ`$ satisfies $`y_{\mu _1}=y_{\mu _2}`$ whenever $`\mu _1`$ and $`\mu _2`$ have the same restriction $`\nu `$ to $`K`$, so that it must belong to the image of the diagonal embedding $`\iota `$. Thus there exists a vector $`๐ฐ^{}K_{\mathrm{}}`$ distinct from $`๐ฐ`$ and which satisfies $`A๐ฐ^{}=๐ช`$, again contradicting the invertibility of $`A`$.
Recall that the inverse different of a finite degree algebraic number field $`K/`$ is the $`O_K`$-module
$$๐ก_K^1=\{\alpha K|\mathrm{Tr}_{K/}(\alpha O_K)\}O_K.$$
Note that if $`L/K`$ is finite and $`\alpha ๐ก_K^1`$ then for all $`\beta O_L`$,
$$\mathrm{Tr}_{L/}(\alpha \beta )=\mathrm{Tr}_{K/}(\alpha \mathrm{Tr}_{L/K}(\beta ))$$
(since $`\mathrm{Tr}_{L/K}(\beta )O_K`$) so that $`๐ก_K^1๐ก_L^1`$. Then if $`๐ฆ/`$ is an infinite degree algebraic extension, we define
$$๐ก_๐ฆ^1=\underset{}{lim}๐ก_K^1$$
where the $`K๐ฆ`$ range over finite subextensions of $``$.
###### Theorem 5.
Suppose that
1. $`K`$ is a finite field extension over $``$. Then $`\mathrm{๐ข๐๐บ๐}(\widehat{๐}_K)`$ possesses a second operation $``$ making it a field and for which their is a canonical isomorphism
$$\mathrm{๐ข๐๐บ๐}(\widehat{๐}_K)K$$
which is natural with respect to the trace maps. This isomorphism identifies $`\mathrm{๐ข๐๐บ๐}(๐_K)`$ with the inverse different $`๐ก_K^1`$ and in particular, there is a canonical embedding of the ring $`O_K`$ in $`\mathrm{๐ข๐๐บ๐}(๐_K)`$.
2. $`๐ฆ=\underset{}{lim}K_\lambda `$ is an infinite extension over $``$. Then $`\mathrm{๐ข๐๐บ๐}(\widehat{\widehat{๐}}_๐ฆ)`$ possesses a second operation $``$ making it a field and for which their is a canonical isomorphism
$$\mathrm{๐ข๐๐บ๐}(\widehat{\widehat{๐}}_๐ฆ)๐ฆ$$
which is natural with respect to the trace maps. This isomorphism identifies $`\mathrm{๐ข๐๐บ๐}(\widehat{๐}_๐ฆ)`$ with the inverse different $`๐ก_๐ฆ^1`$. In particular, there is a canonical embedding of the ring $`O_๐ฆ`$ in $`\mathrm{๐ข๐๐บ๐}(\widehat{๐}_๐ฆ)`$.
###### Note 2.
The statement that $`\mathrm{๐ข๐๐บ๐}(\widehat{๐}_K)(K,+)`$ is classical and has been known since the time of Tateโs thesis .
###### Proof.
We begin with a. Consider the character on $`K_{\mathrm{}}`$ defined
$$\varphi _{\mathrm{}}^K(๐ณ)=\mathrm{exp}(2\pi i\mathrm{๐ณ๐}(๐ณ)).$$
Note that $`\varphi _{\mathrm{}}^K`$ extends continuously to a character $`\varphi ^K`$ of $`\widehat{๐}_K`$. Indeed, $`\varphi _{\mathrm{}}^K`$ is invariant w.r.t. translation by $`O_K`$ and so induces a character $`๐_K๐ด(1)`$, which, when composed with the projection $`\widehat{๐}_K๐_K`$, defines the extension $`\varphi ^K`$. Note that for any finite degree extension $`L/K`$ we have $`\varphi ^L=\varphi ^K\widehat{\mathrm{๐ณ๐}}_{L/K}`$. For each $`\alpha K`$ define $`\varphi _\alpha ^K`$ by
$$\varphi _\alpha ^K(\widehat{๐ณ})=\varphi ^K(\alpha \widehat{๐ณ})$$
(here $`\alpha \widehat{๐ณ}`$ is defined for all $`\widehat{๐ณ}\widehat{๐}_K`$ since $`\widehat{๐}_K`$ is a $`K`$-vector space). Then the map
(7)
$$\alpha \varphi _\alpha ^K$$
defines a monomorphism $`(K,+)\mathrm{Char}(\widehat{๐}_K)`$. We will show that it is an isomorphism. So let $`\varphi :\widehat{๐}_KU(1)`$ be any character. Then the restriction to the leaf through $`0`$, which is just $`K_{\mathrm{}}`$, is of the form
$$๐ณ\mathrm{exp}(2\pi i\mathrm{๐ณ๐}(\mathrm{๐ฐ๐ณ}))$$
for some $`๐ฐK_{\mathrm{}}`$. On the other hand, we may also restrict $`\varphi `$ to the transversal through $`0`$, which is just $`\widehat{O}_K\widehat{๐}_K`$, where it must factor through some finite quotient $`O_K/๐`$. It follows that $`\varphi `$ is the pullback of a character
$$๐_๐=K_{\mathrm{}}/๐U(1)$$
for some ideal $`๐O_K`$. In particular, $`\mathrm{๐ณ๐}(๐ฐ๐)`$ which implies that $`\mathrm{๐ณ๐}(๐ฐK)`$, so by Lemma 1, $`๐ฐ=\alpha K`$ for some $`\alpha `$ and $`\varphi =\varphi _\alpha ^K`$. Thus the map $`\alpha \varphi _\alpha ^K`$ defines an isomorphism between $`(K,+)`$ and $`\mathrm{๐ข๐๐บ๐}(\widehat{๐}_K)`$. One then pushes forward the product operation of $`K`$ to define $``$. Naturality is an expression of the commutative diagram
$$\begin{array}{ccc}L& \stackrel{}{}& \mathrm{Char}(\widehat{๐}_L)\\ & & \widehat{๐ณ}๐_{L/K}^{}& & \\ K& \stackrel{}{}& \mathrm{Char}(\widehat{๐}_K)\end{array}$$
whose commutativity follows from the fact that for all $`\alpha K`$,
$$\widehat{๐ณ}๐_{L/K}^{}(\varphi _\alpha ^K)=\varphi _\alpha ^K\widehat{๐ณ}๐_{L/K}=\varphi _\alpha ^L.$$
The map (7) identifies $`\mathrm{๐ข๐๐บ๐}(๐_K)`$ with $`๐ก_K^1O_K`$ so that the restriction of $``$ to $`\mathrm{๐ข๐๐บ๐}(๐_K)`$ makes the latter an $`O_K`$-module isomorphic to $`๐ก_K^1`$.
To prove b , recall that if $`L/K`$ is a finite extension of number fields finite over $``$, then the trace projection $`\widehat{๐ณ}๐_{L/K}:\widehat{๐}_L\widehat{๐}_K`$ induces an inclusion $`\widehat{๐ณ}๐_{L/K}^{}:\mathrm{๐ข๐๐บ๐}(\widehat{๐}_K)\mathrm{๐ข๐๐บ๐}(\widehat{๐}_L)`$ of fields. Then the direct limit
(8)
$$\underset{}{lim}\mathrm{๐ข๐๐บ๐}(\widehat{๐}_{K_\lambda })$$
is a field isomorphic to $`๐ฆ`$. The direct limit (8) has dense image in $`\mathrm{๐ข๐๐บ๐}(\widehat{\widehat{๐}}_๐ฆ)`$, but the latter is discrete since $`\widehat{\widehat{๐}}_K`$ is compact, so they are equal. Thus every character $`\varphi `$ is the pull-back of one defined on $`\widehat{๐}_K`$ where $`K/`$ is of finite degree. Again, this isomorphism is trace natural. It identifies $`\mathrm{๐ข๐๐บ๐}(\widehat{๐}_๐ฆ)`$ with $`๐ก_๐ฆ^1`$ and so as in the finite extension case, the former has a module structure over the subring corresponding to $`O_๐ฆ`$. โ
Note that for $`K=`$, we have $`\mathrm{๐ข๐๐บ๐}(๐_{})`$ is a ring since $`๐ก_K=`$. On the other hand since $`๐ก_KO_K`$ for $`K`$, we see that $`\mathrm{๐ข๐๐บ๐}(๐_K)`$ is strictly larger than $`O_K`$: it contains $`O_K`$ as a ring, and is to be viewed as an $`O_K`$-module extension of $`O_K`$. If $`๐ฆ^{\mathrm{ab}}`$, then $`๐ก_๐ฆ^1=๐ฆ`$ and $`\mathrm{๐ข๐๐บ๐}(\widehat{๐}_๐ฆ)=\mathrm{๐ข๐๐บ๐}(\widehat{\widehat{๐}}_๐ฆ)`$, which was anticipated by Theorem 3.
## 8. Graded Holomorphic Functions
Let $`\mu `$ be the unit mass Haar measure on $`\widehat{๐}_K`$ and let $`L^2(\widehat{๐}_K,)`$ be the associated space of square integrable complex valued functions on $`\widehat{๐}_K`$. The characters $`\{\varphi _\alpha \}`$ form a complete orthonormal system in $`L^2(\widehat{๐}_K,)`$ and so every element $`fL^2(\widehat{๐}_K,)`$ has the development
$$f=a_\alpha \varphi _\alpha ,$$
where $`\{a_\alpha \}l^2(K)`$, $`\varphi _\alpha \mathrm{๐ข๐๐บ๐}(\widehat{๐}_K)`$ and equality is taken w.r.t. the $`L^2`$ norm. We note that since we are in a Hilbert space, by Parsevalโs Lemma, we have $`f^2=|a_\alpha |^2`$ and so the sum converges with respect to any well-ordering of the indexing set $`K`$. The space $`L^2(๐_K,)`$ may be identified with the subspace of $`L^2(\widehat{๐}_K,)`$ whose Fourier series satisfy $`a_\alpha =0`$ for $`\alpha ๐ก_K^1`$. If we restrict $`f`$ to the dense leaf $`K_{\mathrm{}}`$, then we may identify $`\varphi _\alpha (๐ณ)=\mathrm{exp}(2\pi i\mathrm{Tr}(\alpha ๐ณ))๐ป^\alpha `$ and write $`f`$ in the form of an $`L^2`$ Puiseux series
$$f(๐ป)=a_\alpha ๐ป^\alpha .$$
Let $`f,gL^2(\widehat{๐}_K,)`$ be given by the developments $`f=a_\alpha \varphi _\alpha `$, $`g=b_\alpha \varphi _\alpha `$. The Cauchy product $`fg`$ is defined as the $`L^2`$ extension of the point-wise product of continuous functions, that is,
$$fg=\underset{\alpha }{}c_\alpha \varphi _\alpha =\underset{\alpha }{}\left(\underset{\alpha _1+\alpha _2=\alpha }{}a_{\alpha _1}b_{\alpha _2}\right)\varphi _\alpha ,$$
provided that the sum on the right converges. In this regard, note that $`c_\alpha `$ is equal to the inner-product $`f,\sigma _\alpha \overline{g}`$, where $`\sigma _\alpha \overline{g}=_{\beta K}\overline{b}_{\alpha \beta }\varphi _\beta L^2(\widehat{๐}_K,)`$, hence is an unambiguously defined complex number. Thus the Cauchy product is defined whenever the sequence $`\{c_\alpha \}`$ belongs to $`l^2(K)`$. Given $`fL^2(\widehat{๐}_K,)`$, we denote by $`\mathrm{๐ฃ๐๐}_{}(f)`$ the set of $`gL^2(\widehat{๐}_K,)`$ for which $`fg`$ is defined.
The Dirichlet product $`fg`$ will be defined by the development
(9)
$$fg=\underset{\alpha }{}d_\alpha \varphi _\alpha =\underset{\alpha }{}\left(\underset{\alpha _1\alpha _2=\alpha }{}a_{\alpha _1}b_{\alpha _2}\right)\varphi _\alpha ,$$
provided that it converges. Let us say a few words about what this means. Here, we note that for $`\alpha 0`$, we have that $`d_\alpha =f,\tau _\alpha \overline{g}`$ where $`\tau _\alpha \overline{g}=_{\beta K^{}}b_{\alpha \beta ^1}\varphi _\beta `$, hence is well-defined. Consider also the formal expression
$$d_0=a_0\underset{\alpha K^{}}{}b_\alpha +b_0\underset{\alpha K^{}}{}a_\alpha +a_0b_0.$$
Then we say that the Dirichlet product $`fg`$ converges when $`d_0`$ converges absolutely and the sequence $`\{d_\alpha \}`$ defines an element of $`l^2(K)`$. We denote by $`\mathrm{๐ฃ๐๐}_{}(f)`$ the set of functions having defined Dirichlet product with $`f`$.
In terms of the $`๐ป`$ parameter:
$$(fg)(๐ป)=\underset{\alpha }{}a_\alpha g\left(๐ป^\alpha \right)=\underset{\alpha }{}b_\alpha f\left(๐ป^\alpha \right),$$
showing that Dirichlet multiplication is commutative and distributive over ordinary addition $`+`$ of functions. It follows that $`L^2(\widehat{๐}_K,)`$ is a partial algebra with respect to each of the operations $``$ and $``$ separately: here, partial refers to the fact that $``$ and $``$ are only partially defined, and when they are, the usual axioms of group algebra hold for each operation. When $`๐ฆ`$ is of infinite degree over $``$, the relevant discussion applies to the space of square integrable functions on the proto solenoid $`L^2(\widehat{\widehat{๐}}_๐ฆ,)`$.
###### Note 3.
Suppose that $`K=`$ and $`f,gC_0(๐_{},)C_0(\widehat{๐}_{},)`$ with $`a_n=b_n=0`$ for $`n0`$. Then $`f`$ and $`g`$ define via a Mellin-type transform convergent Dirichlet series
$$๐ฃ_f(y)=a_nn^{2\pi iy}๐ฃ_g(y)=b_nn^{2\pi iy},$$
$`y`$, and
$$D_{fg}=D_fD_g.$$
Thus the algebra of zeta functions, L-functions, Dirichlet series etc. is codified by the Dirichlet product.
Now let $`\widehat{๐}_K`$ be the hyperbolized adele class group defined in ยง6. $`\widehat{๐}_K`$ is a lamination whose leaves are $`d`$-dimensional polydisks, and so we say that a continuous function $`F:\widehat{๐}_K`$ is holomorphic if its restriction $`F|L`$ to each leaf $`L`$ is holomorphic, or equivalently (since all leaves are dense), if its restriction to the canonical leaf $`_K=_K^{}\times _K^{}`$ is holomorphic. We recall that holomorphicity in the factor $`_K^{}`$ is defined in terms of the โupper half-plane $`\times `$ right half-planeโ multi-variable $`๐ฎ\times ๐ฏ`$.
For each $`๐ญ(0,\mathrm{})^d`$ let $`\widehat{๐}_K(๐ญ)\widehat{๐}_K`$ be the subspace of points having extended coordinate $`๐ญ`$ with respect to the decomposition (5). Since $`\widehat{๐}_K(๐ญ)\widehat{๐}_K`$ we may put on $`\widehat{๐}_K(๐ญ)`$ the unit mass Haar measure and define for $`F,G:\widehat{๐}_K`$ the pairing
$$(F,G)_๐ญ=_{\widehat{๐}_K(๐ญ)}F\overline{G}๐\mu .$$
###### Definition 1.
The Hardy space associated to $`K/`$ is the Hilbert space
$$๐ง[K]=\{F:\widehat{๐}_K|F\text{ is holomorphic and }\underset{๐ญ}{sup}(F,F)_๐ญ<\mathrm{}\}.$$
with inner-product $`(,)=sup_๐ญ(,)_๐ญ`$.
Evidently any $`F๐ง[K]`$ has an a.e. defined $`L^2`$ limit for $`๐ญ\mathrm{๐}`$, and such a limit defines an element
$$F:=fL^2(\widehat{๐}_K,).$$
Using the Fourier development available there, we may write the restriction $`F|__K`$ as follows: if $`๐=๐\times ๐ฟ_K=_K^{}\times _K^{}`$ (see ยง6 for the relevant notation) then
(10)
$$F|__K(๐)=a_\alpha \mathrm{exp}(2\pi i\mathrm{๐ณ๐}(\alpha ๐))=a_\alpha \mathrm{exp}(2\pi i\mathrm{๐ณ๐}(\alpha ๐))\mathrm{exp}(2\pi i\mathrm{๐ณ๐}(\alpha ๐ฟ))$$
where
(11)
$$\mathrm{exp}(2\pi i\mathrm{๐ณ๐}(\alpha ๐))=\underset{\nu }{}\mathrm{exp}(2\pi i\alpha _\nu x_\nu )\mathrm{exp}(2\pi \alpha _\nu t_\nu )$$
and
(12)
$$\mathrm{exp}(2\pi i\mathrm{๐ณ๐}(\alpha ๐ฟ))=\underset{(\mu ,\overline{\mu })}{}\mathrm{exp}(4\pi i(\mathrm{Re}(\alpha _\mu z_\mu ))\mathrm{exp}(4\pi \mathrm{Im}(\alpha _\mu b_\mu )).$$
We will sometimes switch to power series notation and shorten this to
$$F|__K(\mathit{\varrho })=\underset{\alpha }{}a_\alpha \mathit{\varrho }^\alpha =\underset{\alpha }{}a_\alpha ๐^\alpha ๐ผ^\alpha ,$$
where $`\mathit{\varrho }\mathrm{exp}(2\pi i\mathrm{๐ณ๐}(๐))`$, $`๐\mathrm{exp}(2\pi i\mathrm{๐ณ๐}(๐))`$ and $`๐ผ\mathrm{exp}(2\pi i\mathrm{๐ณ๐}(๐ฟ))`$.
By the positive cone in $`K_{\mathrm{}}`$ we shall mean the set of $`๐ฑ=(x_\nu ;(z_\mu ,\overline{z}_\mu ))K_{\mathrm{}}`$ for which $`x_\nu >0`$ for all $`\nu `$ real and $`(z_\mu ,\overline{z}_\mu )๐น\times \overline{๐น}`$ for each complex place pair $`(\mu ,\overline{\mu })`$. It is clear then that for the series of (10) to define elements of $`๐ง[K]`$, we must have that $`a_\alpha =0`$ for $`\alpha `$ not contained in the positive cone of $`K_{\mathrm{}}`$. We have that $`F^2=|a_\alpha |^2`$ and hence the correspondence $`FF`$ yields an isometric inclusion of Hilbert spaces
$$๐ง[K]L^2(\widehat{๐}_K,).$$
###### Note 4.
In (12), observe that when $`\alpha `$ is real and positive (i.e. all of its place coordinates are real and positive), then $`๐ผ^\alpha `$ is a holomorphic function purely of the upper half plane variable $`๐ฎ=(u_\mu )_\mu `$ of $`_K`$ i.e. is constant in the right plane variable $`๐ฏ=(v_\mu )(i_\mu )`$. Similarly, when $`\alpha `$ belongs to the positive imaginary axis, $`๐ผ^\alpha `$ only depends on $`๐ฏ`$. In particular, this shows that $`F|__K^{}`$ is holomorphic with respect to the multi-variable $`๐ฎ\times ๐ฏ_K^{}`$.
In order to build up from $`๐ง[K]`$ a kind of holomorphic extension of $`K`$, it will be necessary for us to be able to interpret all elements of $`L^2(\widehat{๐}_K,)`$ as boundaries of holomorphic functions. We shall thus expand upon the $`/2`$-graded (super) convention, which has the virtue of regarding holomorphic and anti-holomorphic functions on equal footing.
### 8.1. The Totally Real Case
We first consider the case $`K=`$. Let $`\mathrm{\Theta }=\{,+\}/2`$ be the sign group and denote by $`^{}=\times (\mathrm{},0)`$ the hyperbolic lower half-plane, $`๐ผ_{}:^{}`$ complex-conjugation. Then every element $`f=a_q\zeta ^qL^2(\widehat{๐}_{},)`$ determines a triple $`(F_{},F_0,F_+)`$
$$F_+(\tau )=\underset{q>0}{}a_q\mathrm{exp}(2\pi iq\tau ),F_0=a_0,F_{}(\tau )=\underset{q<0}{}a_q\mathrm{exp}(2\pi iq๐ผ_{}(\tau )).$$
The functions $`F_+(z)`$ and $`F_{}(z)`$ are viewed as elements of the Hardy spaces $`๐ง_+[]=๐ง[]`$ and $`๐ง_{}[]`$ = Hardy space of anti-holomorphic functions on $`\widehat{๐}_{}`$.
Now let $`K/`$ be a totally real extension of degree $`d`$. Let
$$\mathrm{\Theta }_K=\{,+\}^d(/2)^d$$
and write $`_K=K_{\mathrm{}}_{}^d`$, whose points are written $`๐=(x_{\nu _1}+it_{\nu _1},\mathrm{},x_{\nu _d}+it_{\nu _d})`$. For each $`๐ฝ\mathrm{\Theta }_K`$, define
$$_K^๐ฝ=\left\{๐_K\right|(\text{sign}(t_{\nu _1}),\mathrm{},\text{sign}(t_{\nu _d}))=๐ฝ\}.$$
We associate to $`๐ฝ`$ a conjugation $`๐ผ_๐ฝ:_K_K`$, where the coordinates of $`๐ผ_๐ฝ(๐)=๐^{}`$ satisfy
$$x_{\nu _j}^{}+it_{\nu _j}^{}=x_{\nu _j}+\theta _jit_{\nu _j}$$
for $`j=1,\mathrm{},d`$. The conjugation maps are not holomorphic in $`๐`$ and satisfy for all $`๐ฝ,๐ฝ_1,๐ฝ_2\mathrm{\Theta }_K`$:
$$๐ผ_๐ฝ(_K)=_K^๐ฝ\text{and}๐ผ_{๐ฝ_1}๐ผ_{๐ฝ_2}=๐ผ_{๐ฝ_1๐ฝ_2}.$$
A $`\theta `$-holomorphic function is one of the form $`Fc_\theta `$, where $`F:_K`$ is holomorphic.
Denote by $`K^๐ฝ`$ those elements $`\alpha K`$ whose coordinates with respect to the embedding $`KK_{\mathrm{}}`$ satisfy
$$(\text{sign}(\alpha _{\nu _1}),\mathrm{},\text{sign}(\alpha _{\nu _d}))=๐ฝ$$
Note that
$$K^{๐ฝ_1}K^{๐ฝ_2}K^{๐ฝ_1๐ฝ_2}.$$
Then every element $`f=a_\alpha ๐ป^\alpha L^2L^2(\widehat{๐}_K,)`$ determines a $`(2^d+1)`$-tuple $`(F_๐ฝ;F_0)`$, where for each $`๐ฝ\mathrm{\Theta }_K`$, $`F_๐ฝ:\widehat{๐}_K`$ is defined as the unique extension to $`\widehat{๐}_K`$ of the following $`๐ฝ`$-holomorphic function on $`_K`$:
$$F_๐ฝ(๐)=\underset{\alpha K^๐ฝ}{}a_\alpha \mathrm{exp}\left(2\pi i\mathrm{Tr}(\alpha ๐ผ_๐ฝ(๐))\right)\underset{\alpha K^๐ฝ}{}a_\alpha (๐ผ_๐ฝ(๐))^\alpha $$
where $`๐ผ_๐ฝ(๐)\mathrm{exp}(2\pi i\mathrm{๐ณ๐}(๐ผ_๐ฝ(๐)))`$.The term $`F_0`$ is the constant function $`a_0`$.
The Hardy space of $`\theta `$-holomorphic functions is denoted $`๐ง_๐ฝ[K]`$. The space of $`(2^d+1)`$-tuples is viewed as a graded Hilbert space:
$$๐ง_{}[K]=\left(\underset{๐ฝ}{}๐ง_๐ฝ[K]\right),$$
whose inner-product is the direct sum of the inner-products on each of the summands. We will often write $`๐ง[K]`$ for the summand of $`๐ง_{}[K]`$ corresponding to $`๐ฝ=(+,\mathrm{},+)`$ i.e. the Hardy space of functions holomorphic on $`\widehat{๐}_K`$ in the ordinary sense.
The Cauchy and Dirichlet products are defined on $`๐ง_{}[K]`$ via boundary extensions e.g. $`FG`$ is defined to be the unique element of $`๐ง_{}[K]`$ whose boundary is $`FG`$, provided the latter is defined. The Cauchy product does not generally respect the grading, but the Dirichlet product has the following graded decomposition law:
(13)
$$\left(FG\right)_๐ฝ=\underset{๐ฝ=๐ฝ_1๐ฝ_2}{}F_{๐ฝ_1}G_{๐ฝ_2}\left(FG\right)_0=F(1)G(1)F_0G_0.$$
### 8.2. The Totally Complex Case
Now let us suppose that $`K/`$ is totally complex of degree $`d=2s`$. In order to extend the ideas in the previous paragraphs in a way compatible with the complex places, we develop a complex theory of signs. We consider the singular sign group
$$\mathrm{\Theta }^{}=\{\sqrt{},,\sqrt{},+\}/4$$
and say that $`z(i)0`$ has singular sign $`\sqrt{},,\sqrt{},+`$ according to whether $`z`$ belongs to $`i_+`$, $`_+`$, $`i_+`$ or $`_+`$.
For points which do not belong to $`i`$, we introduce the complex sign set
$$\mathrm{\Omega }=\{\sqrt{}\epsilon ,\epsilon ,\sqrt{}\epsilon ,+\epsilon \},$$
viewed as a $`\mathrm{\Theta }^{}`$-set in the obvious way. We say that $`z(i)`$ has complex sign $`\sqrt{}\epsilon ,\epsilon ,\sqrt{}\epsilon ,+\epsilon `$ according to whether $`z`$ belongs to $`i๐น`$, $`๐น`$, $`i๐น`$ or $`๐น`$, where $`๐น`$ is as before the quarter plane of $`z=x+iy`$ with $`x,y>0`$. Every $`\omega \mathrm{\Omega }`$ can be written uniquely in the form
$$\omega =\theta \epsilon $$
for $`\theta \mathrm{\Theta }^{}`$, and the map
$$e:\mathrm{\Omega }\mathrm{\Theta }^{},e(\theta \epsilon )=\theta $$
is an isomorphism of $`\mathrm{\Theta }^{}`$-sets. The singular sign group $`\mathrm{\Theta }^{}`$ may be viewed intermediate to the โsignlessโ element $`0`$ and the complex sign $`\mathrm{\Theta }^{}`$-set $`\mathrm{\Omega }`$.
If $`z,z^{}`$ have complex signs $`\omega ,\omega ^{}`$ then the product can have sign $`e(\omega )e(\omega ^{})\epsilon `$, or else it can โoverflowโ into the neighboring signs: the singular sign $`\sqrt{}e(\omega )e(\omega ^{})`$ and the complex sign $`\sqrt{}e(\omega )e(\omega ^{})\epsilon `$: explaining why we view $`\mathrm{\Omega }`$ as no more than a $`\mathrm{\Theta }^{}`$-set . We view the union
$$\mathrm{}:=\mathrm{\Theta }^{}\mathrm{\Omega }$$
as an abelian โmulti-valued groupโ i.e. a set equipped with the abelian set-valued product
$$\mathrm{}\times \mathrm{}\mathrm{๐ค}^{\mathrm{}}$$
with identity the sign $`+`$. We write for $`\vartheta ,\vartheta _1,\vartheta _2\mathrm{}`$
$$\vartheta \vartheta _1\vartheta _2$$
to indicate that $`\vartheta `$ is among the possible signs that the product $`z_1z_2`$ can assume when $`\mathrm{๐๐๐๐}(z_1)=\vartheta _1`$, $`\mathrm{๐๐๐๐}(z_2)=\vartheta _2`$. Note that $`|\vartheta _1\vartheta _2|>1`$ only when $`\vartheta _1,\vartheta _2\mathrm{\Omega }`$, in which case $`|\vartheta _1\vartheta _2|=3`$. The function $`e`$ of the previous paragraph extends to $`\mathrm{}`$ by the identity in $`\mathrm{\Theta }^{}`$.
Consider now the set
$$\mathrm{}_K=\left\{\mathit{\vartheta }=(\vartheta _j)\mathrm{}^s\right|\alpha K\text{ such that }\mathrm{sign}(\alpha _{\mu _j})=\vartheta _j\text{ for }j=1,\mathrm{},s\}.$$
Notice that we are only using the first element $`\mu _j`$ of each complex place pair $`(\mu _j,\overline{\mu }_j)`$ to define the sign, and that we have that for all $`K`$ the sign $`(+,\mathrm{},+)\mathrm{}_K`$. The set-valued product of components induces a set-valued product
$$\mathrm{}_K\times \mathrm{}_K\mathrm{๐ค}^\mathrm{}_K.$$
The map
$$e:\mathrm{}_K(\mathrm{\Theta }^{})^s$$
is defined $`e(\mathit{\vartheta })=(e(\vartheta _j))`$.
If we define the type of $`\mathit{\vartheta }\mathrm{}_K`$ to be the vector $`t(\mathit{\vartheta })`$ whose $`j`$th component is $`1`$ or $`\epsilon `$ depending on whether $`\vartheta _j\mathrm{\Theta }^{}`$ or $`\mathrm{\Omega }`$, then
$$\mathit{\vartheta }=e(\mathit{\vartheta })t(\mathit{\vartheta }).$$
We will denote $`๐บ=(\epsilon ,\mathrm{},\epsilon )`$ and $`\mathrm{๐}=(1,\mathrm{},1)`$, and say that $`\mathit{\vartheta }`$ is complex homogeneous (singular homogeneous) if $`t(\mathit{\vartheta })=๐บ`$ ($`t(\mathit{\vartheta })=\mathrm{๐}`$). If $`\mathit{\vartheta }`$ is singular homogeneous, we will sometimes write $`\mathit{\vartheta }=๐ฝ`$. The singular homogeneous elements form a subgroup $`\mathrm{\Theta }_K^{}`$ with respect to which $`\mathrm{}_K`$ is a $`\mathrm{\Theta }_K^{}`$-set.
###### Example 1.
Let $`K`$ be the splitting field for $`X^32`$. Then the sign of any root is (modulo ordering of the places) $`(+,\sqrt{}\epsilon ,\epsilon )`$, so that $`\mathrm{}_K`$ has inhomogeneous triples. On the other hand, if $`K`$ is the splitting field of $`X^31`$ then all elements of $`K`$ are homogeneous. This is also true of the Gaussian numbers $`(i)`$, where $`\mathrm{}_{(i)}=\mathrm{}`$.
For each $`\mathit{\vartheta }\mathrm{}_K`$ we will associate a product of half spaces. For $`\omega \mathrm{\Omega }`$ and each place pair $`(\mu ,\overline{\mu })`$ write
$$_{(\mu ,\overline{\mu })}^\omega =\left\{(\kappa _\mu ,\overline{\kappa }_\mu )^2\times ^2\right|\text{sign}(b_\mu )=\omega \}$$
where we recall that $`\kappa _\mu =(z_\mu ,b_\mu )`$, see ยง6. Notice that when $`\omega =+ฯต`$, $`_{(\mu ,\overline{\mu })}^{+ฯต}=_{(\mu ,\overline{\mu })}`$. We will view these spaces as products of half-spaces in the same way that we viewed $`_{(\mu ,\overline{\mu })}`$ as a product of half-spaces, see (4) in ยง6. For example, when $`\omega =\sqrt{}\epsilon `$, $`b_\mu =s_\mu +it_\mu `$ satisfies $`s_\mu <0`$ and $`t_\mu >0`$ so that we obtain the product of the upper half-plane with the left half-plane:
$$_{(\mu ,\overline{\mu })}^\sqrt{}\epsilon =_\mu i_\mu .$$
The points of the first factor are denoted $`u_\mu =x_\mu +it_\mu `$ and those of the second factor by $`v_\mu =s_\mu +iy_\mu `$. In the same fashion, we have
$$_{(\mu ,\overline{\mu })}^\epsilon =_\mu i_\mu ,_{(\mu ,\overline{\mu })}^\sqrt{}\epsilon =_\mu i_\mu .$$
Also, for each singular sign $`\theta \mathrm{\Theta }^{}`$, we write
$$_{(\mu ,\overline{\mu })}^\theta :=_{(\mu ,\overline{\mu })}^{\theta \epsilon }.$$
Then for $`\mathit{\vartheta }\mathrm{}_K`$ we define the $`\mathit{\vartheta }`$-hyperbolic plane as
$$_K^\mathit{\vartheta }=\underset{j=1}{\overset{s}{}}_{(\mu _j,\overline{\mu }_j)}^{\vartheta _j}.$$
We now define conjugation maps relating these planes. For $`\kappa _\mu =(z_\mu ,b_\mu )^2`$ and $`\theta \mathrm{\Theta }^{}`$ we define
$$๐ผ_\theta (z_\mu ,b_\mu )=(z_\mu ,\theta b_\mu ).$$
This induces a map of $`_{(\mu ,\overline{\mu })}=\{(z,\overline{z})\times (b,\overline{b})|z,b๐น\}`$ such that
$$๐ผ_\theta \left(_{(\mu ,\overline{\mu })}\right)=_{(\mu ,\overline{\mu })}^\theta =_{(\mu ,\overline{\mu })}^{\theta \epsilon }.$$
Notice that the conjugation maps are not (holomorphic) functions of the half-plane variables $`u_\mu =x_\mu +it_\mu ,v_\mu =s_\mu +iy_\mu `$. For example, when $`\theta =\sqrt{}`$,
$$๐ผ_{\sqrt{}}(x_\mu +it_\mu ,s_\mu +iy_\mu )=(x_\mu +is_\mu ,t_\mu +iy_\mu ).$$
We have
$$๐ผ_{\theta _1}๐ผ_{\theta _2}=๐ผ_{\theta _1\theta _2}$$
for all $`\theta _1,\theta _2\mathrm{\Theta }^{}`$.
For each $`\mathit{\vartheta }\mathrm{}_K`$ we associate a conjugation $`๐ผ_\mathit{\vartheta }:_K_K`$ where the coordinates of $`๐ผ_\mathit{\vartheta }(๐ฟ)=๐ฟ^{}`$ are determined by
(14)
$$\kappa _{\mu _j}^{}=(z_{\mu _j}^{},b_{\mu _j}^{})=(z_{\mu _j},e(\vartheta _j)b_{\mu _j})$$
for $`k=1,\mathrm{},s`$. Notice that $`๐ผ_\mathit{\vartheta }`$ restricts to a map $`๐ผ_\mathit{\vartheta }:_K_K^\mathit{\vartheta }`$. A function $`F:_K`$ of the shape $`F=G๐ผ_\mathit{\vartheta }`$ where $`G:_K`$ is holomorphic, will be called $`\mathit{\vartheta }`$-holomorphic.
For each $`\mathit{\vartheta }\mathrm{}_K`$, denote by $`K^\mathit{\vartheta }`$ those elements $`\alpha `$ whose coordinates with respect to the embedding $`KK_{\mathrm{}}`$ satisfy
$$(\text{sign}(\alpha _{\mu _1}),\mathrm{},\text{sign}(\alpha _{\mu _s}))=\mathit{\vartheta }.$$
For $`๐ฝ\mathrm{\Theta }_K^{}`$ and arbitrary $`\mathit{\vartheta }\mathrm{}_K`$ we have
(15)
$$K^๐ฝK^\mathit{\vartheta }K^{๐ฝ\mathit{\vartheta }}.$$
In general, we cannot expect a law of the genre (15) since the complex signs do not form a group. We have instead the multi-valued law:
$$K^{\mathit{\vartheta }_1}K^{\mathit{\vartheta }_2}\underset{\mathit{\vartheta }\mathit{\vartheta }_1\mathit{\vartheta }_2}{}K^\mathit{\vartheta }.$$
Every element $`fL^2(\widehat{๐}_K,)`$ now determines a $`(|\mathrm{}_K|+1)`$-tuple $`(F_\mathit{\vartheta };F_0)`$ as follows. For each $`\mathit{\vartheta }\mathrm{}_K`$, $`F_\mathit{\vartheta }:\widehat{๐}_K`$ is defined as the extension to $`\widehat{๐}_K`$ of the following function on $`_K`$:
$$F_\mathit{\vartheta }(๐ฟ)=\underset{\alpha K^\mathit{\vartheta }}{}a_\alpha \mathrm{exp}\left(2\pi i\mathrm{๐ณ๐}(\alpha ๐ผ_\mathit{\vartheta }(๐ฟ))\right)\underset{\alpha K^\mathit{\vartheta }}{}a_\alpha (๐ผ_\mathit{\vartheta }(๐ผ))^\alpha .$$
where $`๐ผ_\mathit{\vartheta }(๐ผ)\mathrm{exp}(2\pi i\mathrm{๐ณ๐}(๐ผ_\mathit{\vartheta }(๐ฟ)))`$. Observe that the comments found in Note 4 imply that whenever the sign coordinate $`\vartheta _j`$ is singular i.e. $`\vartheta _j\mathrm{\Theta }^{}`$ then $`F_\mathit{\vartheta }`$ is constant in one of the corresponding half-plane coordinates $`u_{\mu _j},v_{\mu _j}`$. This explains why we call the signs in $`\mathrm{\Theta }^{}`$ โsingularโ.
We refer to $`F_\mathit{\vartheta }`$ as the $`\mathit{\vartheta }`$-holomorphic component of $`f`$, and the Hardy space of $`\mathit{\vartheta }`$-holomorphic functions is denoted $`๐ง_\mathit{\vartheta }[K]`$. We obtain a graded Hilbert space:
$$๐ง_{}[K]=\left(\underset{\mathit{\vartheta }}{}๐ง_\mathit{\vartheta }[K]\right),$$
whose inner-product is the direct sum of the inner-products on each of the summands. We will write as in the real case $`๐ง[K]`$ for the summand of $`๐ง_{}[K]`$ corresponding to $`\mathit{\vartheta }=\mathrm{๐}`$.
The Cauchy and Dirichlet products are defined on $`๐ง_{}[K]`$ via boundary extensions just as in the real case. When $`\mathit{\vartheta }\mathit{\vartheta }_1\mathit{\vartheta }_2`$ we write
$$(F_{\mathit{\vartheta }_1}G_{\mathit{\vartheta }_2})|_\mathit{\vartheta }$$
for the projection of $`F_{\mathit{\vartheta }_1}G_{\mathit{\vartheta }_2}`$ onto the sub series indexed by $`\alpha K^\mathit{\vartheta }`$. Then we have the following graded decomposition law generalizing that described in (13) for the totally real case:
(16)
$$\left(FG\right)_\mathit{\vartheta }=\underset{\mathit{\vartheta }\mathit{\vartheta }_1\mathit{\vartheta }_2}{}(F_{\mathit{\vartheta }_1}G_{\mathit{\vartheta }_2})|_\mathit{\vartheta },\left(FG\right)_0=F(1)G(1)F_0G_0.$$
Now consider $`K/`$ a general finite extension. The we obtain a decomposition of the form
(17)
$$๐ง_{}[K]=๐ง_{}^{}[K]๐ง_{}^{}[K]$$
corresponding to the real and complex places (graded accordingly) so that in particular, every $`fL^2(\widehat{๐}_K,)`$ determines a $`(2^r+|\mathrm{}_K|+1)`$-tuple of functions.
###### Note 5.
Let $`K/`$ be Galois. Then the action of the Galois group $`\mathrm{๐ฆ๐บ๐
}(K/)`$ induces a well-defined action on the sign group $`\mathrm{\Theta }_K`$ (when $`K`$ is real) and on the โmulti-valued sign groupโ $`\mathrm{}_K`$ (when $`K`$ is complex) .
For $`K`$ real or complex, we have a sub partial double group algebra $`๐ง_{}[O_K]`$ of $`๐ง_{}[K]`$ defined by Fourier series whose indices belong only to $`O_K`$. The Hilbert space of graded holomorphic functions pulled back from the Minkowski hyperbolized torus $`๐_K`$ are denoted
$$๐ง_{}[๐_K]:=๐ง_{}[๐ก_K^1],$$
where $`๐ก_K^1`$ is the inverse different. This sub Hilbert space is closed with respect to the Cauchy product (where it is defined), and is closed with respect to the action by Dirichlet multiplication with elements of $`๐ง_{}[O_K]`$ (whenever such products are defined). When $`K=`$, then we have $`๐ง_{}[]=๐ง_{}[๐_{}]`$ is a partial double algebra with respect to the Cauchy and Dirichlet products.
We now indicate how to extend this construction to infinite field extensions $`๐ฆ/`$. Here we use the hyperbolized adele class group $`\widehat{๐}_๐ฆ`$ (associated to $`\widehat{๐}_๐ฆ`$) in conjunction with the proto adele class group $`\widehat{\widehat{๐}}_๐ฆ`$. A continuous function $`F:\widehat{๐}_๐ฆ`$ is holomorphic if its restriction to any of the dense leaves is holomorphic. In particular, we note that the restriction $`F|__๐ฆ`$ is holomorphic if $`F`$ is holomorphic separately in each factor of the polydisk decomposition $`_๐ฆ_\nu \times _{(\mu ,\overline{\mu })}`$. As in the case of a finite field extension, we define $`\widehat{๐}_๐ฆ(๐ญ)`$ as the subset of $`\widehat{๐}_๐ฆ`$ having extended coordinate $`๐ญ(0,\mathrm{})^{\mathrm{}}`$. The proto compactification of $`\widehat{๐}_๐ฆ(๐ญ)`$ is a lamination $`\widehat{\widehat{๐}}_๐ฆ(๐ญ)`$ homeomorphic to $`\widehat{\widehat{๐}}_๐ฆ`$. If we let
$$\widehat{\widehat{๐}}_๐ฆ=\widehat{\widehat{๐}}_๐ฆ\times (0,\mathrm{})^{\mathrm{}}=\widehat{\widehat{๐}}_๐ฆ(๐ญ),$$
the Hardy space $`๐ง[๐ฆ]`$ is defined as the space of holomorphic functions $`F:\widehat{๐}_๐ฆ`$ having a continuous extension $`\widehat{F}:\widehat{\widehat{๐}}_๐ฆ`$ and for which the norm
$$\widehat{F}_๐ญ^2=_{\widehat{\widehat{๐}}_๐ฆ(๐ญ)}|\widehat{F}|^2๐\mu $$
is uniformly bounded in $`๐ญ`$ (where $`d\mu `$ is induced from the Haar measure $`\mu `$ on $`\widehat{\widehat{๐}}_๐ฆ\widehat{\widehat{๐}}_๐ฆ(๐ญ)`$). As in the finite-dimensional case, $`๐ง[๐ฆ]`$ is a Hilbert space with respect to the supremum of the integration pairings on each $`\widehat{\widehat{๐}}_๐ฆ(๐ญ)`$. The rest of the development follows that of the finite extension case, where the grading is defined for the real and complex places separately.
We summarize the above remarks in the following
###### Theorem 6.
Let $`๐ฆ`$ be a (possibly infinite degree) algebraic number field over $``$. Then $`๐ง_{}[๐ฆ]`$ is a graded Hilbert space equipped with the structure of a partial double $``$-algebra with respect to the operations of $``$ and $``$.
## 9. Nonlinear Number Fields
Let $`K/`$ be a number field of finite degree over $``$. Let $`[K]`$ denote the vector space of formal, finite $``$-linear combinations of elements in $`K`$ i.e. expressions of the form $`a_\alpha \alpha `$ where $`\alpha K`$ and $`a_\alpha `$, zero for all but finitely many $`\alpha `$. The operations $`+_K`$ and $`\times _K`$ extend linearly to $`[K]`$ yielding two operations, written $``$ and $``$, which define on $`[K]`$ two algebra structures. We refer to $`[K]`$ as the field algebra generated by $`K`$. To avoid confusion, we use the notation $`\mathrm{๐๐ฝ}_{}=0_K`$ and $`\mathrm{๐๐ฝ}_{}=1_K`$; $`0`$ will denote the vector space identity, the element of $`[K]`$ for which $`a_\alpha =0`$ for all $`\alpha `$. There exists a canonical double algebra monomorphism generated by $`\alpha ๐ป^\alpha `$,
$$[K]L^2(\widehat{๐}_K,)๐ง_{}[K]$$
having dense image. The subspace $`[O_K]`$ of $``$-linear combinations of integers is closed with respect to both algebra structures. These algebra structures are not compatible in any familiar sense as the operations $``$ and $``$ do not obey the distributive law.
We define a linear map $`๐ณ:[K]`$ by
$$๐ณ(f)=a_\alpha .$$
Notice that $`f\mathrm{๐๐ฝ}_{}=๐ณ(f)\mathrm{๐๐ฝ}_{}`$. The vector space $`I_K=\mathrm{๐ช๐พ๐}(๐ณ)`$ is an ideal in $`[K]`$ with respect to the operations of $``$ and $``$ and the set-theoretic difference
$$^{}[K]:=[K]I_K$$
is preserved by both of $``$ and $``$. Denote a typical element of $`^{}[K]`$ by $`f^{}`$. Then for any $`f^{}`$ we have
(18)
$$f^{}\mathrm{๐๐ฝ}_{}^{}^{}\mathrm{๐๐ฝ}_{}^{}.$$
Let $`๐ญ^0[K]`$ denote the image of $`^{}[K]`$ in the complex projectivization $`[K]`$ of $`[K]`$. By virtue of (18), the operations $``$ and $``$ descend to $`๐ญ^0[K]`$, making it a double semigroup: a set with two semigroup structures having no a priori compatibility. We denote its elements $`[f]`$. The sub double semigroup $`๐ญ^0[O_K]`$ is defined similarly.
Note that the element $`[\mathrm{๐๐ฝ}_{}]`$ behaves very much like the zero in a field in that it is a universal annihilator with respect to $``$: for all $`[f]๐ญ^0[K]`$,
$$[f][\mathrm{๐๐ฝ}_{}]=[\mathrm{๐๐ฝ}_{}].$$
Furthermore, the natural inclusions $`O_K[O_K]`$ and $`K[K]`$ induce monomorphisms $`O_K๐ญ^0[O_K]`$ and $`K๐ญ^0[K]`$. These echos with field theory make the double semi-group $`๐ญ^0[K]`$ a natural paradigm for the ensuing development of nonlinear fields. We remark that the preceding comments hold without change for an infinite algebraic extension $`๐ฆ/`$.
Motivated by the monomorphism $`K[K]๐ง_{}[K]`$ we set out to create from $`๐ง_{}[K]`$ something akin to a field extension of $`K`$ by graded holomorphic functions. In this connection, we note that the operations $``$ and $``$ restrict to $`+_K`$ and $`\times _K`$ on $`K`$, and on the other hand, the vector space operations of point-wise addition and scalar multiplication do not preserve $`K`$. Accordingly, we discard the vector space structure by projectivizing, retracing the steps in the construction of $`๐ญ^0[K]`$.
Here, the trace operator $`๐ณ`$ is not defined on all of $`๐ง_{}[K]`$. It is unambiguously defined on the subspace of functions $`F`$ having boundary $`F`$ lying in the subspace $`l^1(K)l^2(K)`$ of functions whose Fourier coefficients $`\{a_\alpha \}`$ are absolutely summable. For such elements $`F๐ง_{}[K]`$ we define $`๐ณ(F)=a_\alpha =F(1)`$ and denote $`๐จ_K=\mathrm{๐ช๐พ๐}(๐ณ)`$. We note that $`๐จ_K`$ is not closed in in $`๐ง_{}[K]`$ and is in fact dense there. The set theoretic difference
$$๐ง_{}^{}[K]:=๐ง_{}[K]๐จ_K$$
inherits the grading by restriction. The associated quotient by $`^{}`$, denoted
$$๐ญ_{}[K],$$
is an infinite dimensional subspace of the full projectivization $`๐ง_{}[K]`$ which inherits $``$, $``$ as partially defined operations. The grading gives rise to an arrangement of subspaces
$$\{๐ญ_\mathit{\vartheta }[K]\}$$
where $`๐ญ_\mathit{\vartheta }[K]=๐ง_\mathit{\vartheta }[K]๐ญ_{}[K]`$. The canonical monomorphism
$$๐ญ^0[K]๐ญ_{}[K]$$
induced by $`\alpha \mathit{\varrho }^\alpha `$, has dense image, and the elements of $`๐ญ^0[K]`$ may be Cauchy or Dirichlet multiplied with any element of $`๐ญ_{}[K]`$. We also have a monomorphism $`K๐ญ_{}[K]`$. Elements of $`๐ญ_{}[K]`$ will be denoted by $`[F]`$. The sub partial double semigroup $`๐ญ_{}[O_K]`$ is defined similarly. These remarks are valid without change for an infinite field extension $`๐ฆ/`$.
The grading of $`๐ญ_{}[K]`$ induces one on $`๐ญ^0[K]`$ and so we write $`๐ญ_{}^0[K]`$.
###### Definition 2.
A nonlinear number field is a graded topological abelian partial double semigroup $`๐ฒ_{}`$ with respect to two operations $``$ and $``$ such that
1. There exists a (possibly infinite degree) algebraic number field $`๐ฆ/`$ and a graded double semigroup monomorphism $`ฤฑ:๐ญ_{}^0[๐ฆ]๐ฒ_{}`$ having dense image.
2. The identity $`\mathrm{๐๐ฝ}_{}`$ is a universal annhilator for $``$: for all $`F\mathrm{๐ฃ๐๐}_{}(\mathrm{๐๐ฝ}_{})`$, $`F\mathrm{๐๐ฝ}_{}=\mathrm{๐๐ฝ}_{}`$.
The closure $`๐ฎ`$ of the image $`ฤฑ(๐ญ^0[O_๐ฆ])`$ is called the nonlinear ring of integers.
The qualificative โnonlinearโ refers to the fact that distributivity in an ordinary field is equivalent to the fact that the multiplication map is a bilinear operation. For $`๐ฆ/`$ be a possibly infinite degree extension, notice that both $`๐ญ_{}^0[๐ฆ]`$ and $`๐ญ_{}[๐ฆ]`$ are nonlinear number fields. In addition the following are also nonlinear number fields:
* $`\overline{๐ญ}_{}[๐ฆ]=๐ง_{}[๐ฆ]`$ = the full projectivization of $`๐ง_{}[๐ฆ]`$.
* Let $`๐ถ_{}[๐ฆ]`$ be the projectivization of the subspace of $`๐ง_{}[๐ฆ]`$ consisting of absolutely convergent series with non-zero trace. Then $`๐ถ_{}[๐ฆ]`$ is the Wienerian nonlinear number field associated to $`๐ฆ`$: a full (and not partial) subsemigroup of $`๐ญ_{}[๐ฆ]`$.
We have inclusions $`๐ฆ๐ญ_{}^0[๐ฆ]๐ญ_{}[๐ฆ]๐ถ_{}[๐ฆ]\overline{๐ญ}_{}[๐ฆ]`$, the last three of which are dense.
###### Note 6.
In view of Note 3, all of the arithmetic of classical (single variable) zeta functions, Dirichlet series, $`L`$-functions, etc. is contained in $`๐ญ_{}[]`$.
###### Note 7.
The set of Cauchy units $`๐ด_{}[K]_{}`$ form a dense subset of $`๐ญ_{}[K]`$, since it contains all classes represented by real analytic nonvanishing functions of $`\widehat{๐}_K`$. It is an interesting question as to whether the set of Dirichlet units $`๐ด_{}[K]_{}`$ is also dense, and whether there exists an algorithmic procedure to determine the coefficients of a Dirichlet inverse analogous to the classical Mรถbius inversion formula.
The following theorem says that on a dense subset, $`๐ญ_{}[๐ฆ]`$ is the โnonlinear field of fractionsโ of $`๐ญ_{}[O_๐ฆ]`$:
###### Theorem 7.
Let $`๐ฆ`$ be a (possibly infinite degree) number field. Then there is a dense subset $`๐ฏ๐ญ_{}[๐ฆ]`$ such that for all $`[F]๐ฏ`$, there exists $`[A]๐ญ_{}[O_๐ฆ]`$ with
$$[A][F]๐ญ_{}[O_๐ฆ].$$
###### Proof.
Let $`๐ญ_{}[๐ฆ]_{\mathrm{fin}}`$ be the subspace associated to functions whose nonzero Fourier coefficients are indexed by $`\alpha `$ in some fractional ideal $`O_๐ฆ๐^1๐ฆ`$, where $`๐O_๐ฆ`$. Consider the sub double semigroup $`๐ญ_{}[๐]`$. Then given $`[F]๐ญ_{}[๐ฆ]_{\mathrm{fin}}`$ whose nonzero Fourier coefficients are indexed by $`O_๐ฆ๐^1`$, there exists $`[A]๐ญ_{}[๐]`$ such that $`[A][F]๐ญ_{}[O_๐ฆ]`$. โ
## 10. Galois Groups and the Action of the Idele Class Group of $``$
Let $`๐ฆ/`$ be a possibly infinite degree algebraic number field Galois over $``$. In this case, $`๐ฆ`$ is either totally real or totally complex: in either case we denote by $`\mathrm{}_๐ฆ`$ the sign set. We will also not distinguish $`๐ฆ`$ from its images in $`๐ง_{}[๐ฆ]`$ or $`๐ญ_{}[๐ฆ]`$. We will prefer here to represent elements of $`๐ฆ`$ using the power series notation $`\mathit{\varrho }^\alpha \mathrm{exp}(2\pi i\mathrm{Tr}(๐\alpha ))`$ (as opposed to the character notation $`\varphi _\alpha `$), which has the advantage of allowing us to interpret Dirichlet multiplication with monomials in terms of composition:
$$[F][\mathit{\varrho }^\alpha ]=[F(\mathit{\varrho }^\alpha )].$$
We denote as in the previous section $`\overline{๐ญ}_{}[๐ฆ]=๐ง_{}[๐ฆ]`$.
Equip $`๐ญ_{}[๐ฆ]`$ with the Fubini-Study metric associated to the inner-product on $`๐ง_{}[๐ฆ]`$. A nonlinear automorphism $`\mathrm{{\rm Y}}:๐ญ_{}[๐ฆ]๐ญ_{}[๐ฆ]`$ is the restriction of a graded Fubini-Study isometry of $`\overline{๐ญ}_{}[๐ฆ]`$ respecting the operations $``$, $``$ whenever they are defined: that is,
$$\mathrm{{\rm Y}}([F][G])=\mathrm{{\rm Y}}([F])\mathrm{{\rm Y}}([G]),\mathrm{{\rm Y}}([F][G])=\mathrm{{\rm Y}}([F])\mathrm{{\rm Y}}([G])$$
and for some permutation $`\iota `$ of $`\mathrm{}_๐ฆ`$, $`\mathrm{{\rm Y}}(๐ง_\mathit{\vartheta }[๐ฆ])=๐ง_{\iota (\mathit{\vartheta })}[๐ฆ]`$ for all $`\mathit{\vartheta }\mathrm{}_๐ฆ`$.
For example, let $`/๐ฆ`$ be Galois. Then the Galois group $`\mathrm{๐ฆ๐บ๐
}(/๐ฆ)`$ acts on $`๐ง_{}[]`$ by:
$$a_\alpha \rho ^\alpha a_\alpha \rho ^{\sigma (\alpha )}=a_{\sigma ^1(\alpha )}\rho ^\alpha $$
for $`\sigma \mathrm{๐ฆ๐บ๐
}(/๐ฆ)`$. This action permutes the coefficient set, hence acts by isometries. Viewing the action on $`_{\mathrm{}}`$, where it simply permutes coordinates, we see that there is an induced action on the sign set $`\mathrm{}_{}`$, so that for any $`\mathit{\vartheta }\mathrm{}_{}`$, we have $`\sigma (^\mathit{\vartheta })=^{\sigma (\mathit{\vartheta })}`$. Thus $`\sigma `$ permutes the grading of $`๐ง_{}[]`$. Finally, $`\sigma `$ preserves the elements of trace zero and commutes with multiplication by elements of $`^{}`$, hence induces a nonlinear automorphism
$$\sigma :๐ญ_{}[]๐ญ_{}[]$$
that is trivial on $`๐ญ_{}[๐ฆ]`$.
If $`/๐ฆ`$ is a field extension of number fields of possibly infinite degree over $``$, denote by
$$\mathrm{๐ฆ๐บ๐
}\left(๐ญ_{}[]/๐ญ_{}[๐ฆ]\right)$$
the group of nonlinear automorphisms of $`๐ญ_{}[]`$ fixing the sub nonlinear field $`๐ญ_{}[๐ฆ]`$, and by
$$\mathrm{๐ฆ๐บ๐
}\left(๐ญ_{}[๐ฆ]/๐ฆ\right)$$
the group of nonlinear automorphisms of $`๐ญ_{}[๐ฆ]`$ fixing $`๐ฆ`$.
We recall the following theorem of Wigner .
###### Wignerโs Theorem.
Let $`๐`$ be a complex Hilbert space, $`๐=(๐\{0\})/^{}`$ its projectivization. Let $`[h]:๐๐`$ be a bijection preserving the Fubini-Study metric. Then $`[h]`$ is the projectivization of a unitary or anti-unitary linear map $`h:๐๐`$.
###### Theorem 8.
Let $`๐ฆ`$ be a (possibly infinite degree) number field. Then
$$\mathrm{๐ฆ๐บ๐
}\left(๐ญ_{}[๐ฆ]/๐ฆ\right)\{1\}.$$
###### Proof.
Let $`\sigma \mathrm{๐ฆ๐บ๐
}\left(๐ญ_{}[๐ฆ]/๐ฆ\right)`$. By Wignerโs Theorem, $`\sigma `$ is the projectivization of an (anti) unitary linear map
$$\stackrel{~}{\sigma }:๐ง_{}[๐ฆ]๐ง_{}[๐ฆ].$$
Since $`\sigma `$ fixes $`๐ฆ`$, there exist multipliers $`\lambda _\alpha U(1)`$ with
$$\stackrel{~}{\sigma }(\mathit{\varrho }^\alpha )=\lambda _\alpha \mathit{\varrho }^\alpha .$$
But $`\sigma `$ respects the Cauchy and Dirichlet products, wherein we must have that $`\lambda `$ is simultaneously an additive and multiplicative character:
$$\lambda _{\alpha _1+\alpha _2}=\lambda _{\alpha _1}\lambda _{\alpha _2}=\lambda _{\alpha _1\alpha _2},$$
clearly possible only for $`\lambda `$ trivial. โ
###### Lemma 2.
Let $`๐ฆ`$ be a (possibly infinite degree) algebraic number field Galois over $``$. Given $`\mathbf{\vartheta }\mathrm{}_๐ฆ`$ let $`F๐ง_\mathbf{\vartheta }[๐ฆ]`$ satisfy the functional equation
(19)
$$F(\mathit{\varrho }^r)=(F(\mathit{\varrho }))^r$$
for all $`r_+`$. Then $`F๐ฆ`$ i.e. there exists $`\alpha ๐ฆ`$ with $`F(\mathbf{\varrho })=\mathbf{\varrho }^\alpha `$
###### Proof.
We first consider the case of $`K=`$. In this totally real case we write $`F`$ as a function of the parameter $`\xi =\mathrm{exp}(2\pi i\tau )`$ where $`\tau =x+iy_{}`$. Assume first that $`F`$ is holomorphic and non constant i.e. $`F๐ง[]`$ = Hardy space of holomorphic functions. We will show that $`F(\xi )=\xi ^q`$ for some $`q_+`$. Let us return to viewing $`F`$ as a holomorphic function of the half-plane variable $`\tau `$. Let $`\tau _0_{}\widehat{๐}_{}`$ be such that $`F(\tau _0)0`$. Then there exists a complex number $`\alpha `$ with $`\mathrm{exp}(2\pi i\alpha \tau _0)=F(\tau _0)`$. By (19), the functions $`F(\tau )`$ and $`\mathrm{exp}(2\pi i\alpha \tau )`$ agree on the ray $`_+\tau _0`$ hence by holomorphicity, they coincide. Since $`F`$ is a function of $`\widehat{๐}_{}`$, it follows that $`\alpha =q_+`$ hence $`F(\xi )=\xi ^q`$. For $`F๐ง_{}[]`$ = Hardy space of anti-holomorphic functions, the argument is the same: we use the anti-holomorphic exponential $`\mathrm{exp}(2\pi i\alpha \overline{\tau })`$ and the fact that anti-holomorphic functions agreeing on a co-dimension 1 subspace coincide.
Now let $`K/`$ be totally real of finite degree. Without loss of generality we may assume that $`F๐ง[K]`$ = the Hardy space of holomorphic functions. Let $`\mathrm{\Delta }_K`$ be the diagonal hyperbolic sub-plane, which is the dense leaf of the image of $`\widehat{๐}_{}`$ under the diagonal embedding $`\widehat{๐}_{}\widehat{๐}_K`$. Then by the previous paragraph, the restriction of $`F`$ to $`\mathrm{\Delta }`$ is an exponential
$$\mathrm{exp}(2\pi iq\tau )$$
for $`q_+`$. Let $`\mathrm{\Delta }_1_K`$ be the diagonal
$$\mathrm{\Delta }_1=\{(\tau ,\mathrm{},\tau ,\stackrel{~}{\tau })|\tau ,\stackrel{~}{\tau }\}\mathrm{\Delta }.$$
For each $`\tau `$ fixed, we can (using the same argument employed in the previous paragraph) write the function $`\stackrel{~}{\tau }F(\tau ,\mathrm{},\tau ,\stackrel{~}{\tau })`$ as
$$F(\tau ,\mathrm{},\tau ,\stackrel{~}{\tau })=\mathrm{exp}(2\pi iq\tau )\mathrm{exp}(2\pi i\beta (\tau )\stackrel{~}{\tau })$$
where $`\beta (\tau )`$. Moreover we have
$$F(r(\tau ,\mathrm{},\tau ,\stackrel{~}{\tau }))=\mathrm{exp}(2\pi iqr\tau )\mathrm{exp}(2\pi i\beta (\tau )r\stackrel{~}{\tau })$$
by hypothesis. As we vary $`\tau `$, $`\beta (\tau )`$ varies holomorphically and we obtain that on a real codimension 1 subspace of $`\mathrm{\Delta }_1`$,
$$F(\tau ,\mathrm{},\tau ,\stackrel{~}{\tau })=F_1(\tau ,\mathrm{},\tau ,\stackrel{~}{\tau }):=\mathrm{exp}(2\pi iq\tau )\mathrm{exp}(2\pi i\beta (\tau )\stackrel{~}{\tau })$$
hence they are equal on $`\mathrm{\Delta }_1`$. But this means that $`F_1`$ must also satisfy the functional equation $`F_1(r(\tau ,\mathrm{},\tau ,\stackrel{~}{\tau }))=F_1(\tau ,\mathrm{},\tau ,\stackrel{~}{\tau })^r`$, which implies that $`\beta (\tau )=\beta `$ is a constant. Inductively, we obtain that $`F`$ restricted to $`_K`$ is an exponential function, and since $`F`$ is a function of $`\widehat{๐}_K`$, this restriction must be of the form $`๐^\alpha =\mathrm{exp}(2\pi i\mathrm{Tr}(\alpha ๐))`$ for $`\alpha K_+`$. The case of a totally complex field extension, one of mixed type, or one of infinite degree, is handled in exactly the same manner.
###### Theorem 9.
Let $`/๐ฆ`$ be a Galois extension of (possibly infinite degree) algebraic number fields. Then
$$\mathrm{๐ฆ๐บ๐
}\left(๐ญ_{}[]/๐ญ_{}[๐ฆ]\right)\mathrm{๐ฆ๐บ๐
}(/๐ฆ).$$
###### Proof.
Let $`\sigma \mathrm{๐ฆ๐บ๐
}(๐ญ_{}[]/๐ญ_{}[๐ฆ])`$. We begin by showing that $`\sigma ()=`$, where $``$ is identified with the field of monomials $`[\mathit{\varrho }^\alpha ]`$, $`\alpha `$. Note first that we have already $`\sigma (๐ฆ)=๐ฆ`$. Since $`\sigma ()`$ is a field, all of its elements obey the distributive law. Thus, given any $`[F]\sigma ()`$, since $`\sigma ([\mathit{\varrho }^m])=[\mathit{\varrho }^m]๐ฆ`$, $`m`$, we have
$`[F(\mathit{\varrho }^m)]=[F][\mathit{\varrho }^m]`$ $`=`$ $`[F]\left([\mathit{\varrho }]\mathrm{}[\mathit{\varrho }]\right)`$
$`=`$ $`\left([F][\mathit{\varrho }]\right)\mathrm{}\left([F][\mathit{\varrho }]\right)`$
$`=`$ $`[F(\mathit{\varrho })]\mathrm{}[F(\mathit{\varrho })]`$
$`=`$ $`[F]^m,`$
where $`[F]^m`$ denotes the Cauchy $`m`$th power of $`[F]`$. In fact, the same argument shows that for any $`m/n_+`$
$$([F(\mathit{\varrho }^{m/n})])^n=[F]^m.$$
We may thus find $`F[F]`$ satisfying the functional equation $`F(\mathit{\varrho }^q)=(F(\mathit{\varrho }))^q`$ for all $`q_+`$. By continuity, this extends to $`_+`$. Note that since $`\sigma `$ respects the grading and each element of $``$ is homogeneous (is contained in a fixed projective summand) then $`[F]`$ is also homogeneous. Thus, by Lemma 2, it follows that $`F`$ and $`\sigma ()=`$. We induce in this way a homomorphism
$$\mathrm{\Pi }:\mathrm{๐ฆ๐บ๐
}\left(๐ญ_{}[]/๐ญ_{}[๐ฆ]\right)\mathrm{๐ฆ๐บ๐
}(/๐ฆ),\mathrm{\Pi }(\sigma )=\sigma |_{}.$$
Note that $`\mathrm{\Pi }`$ is clearly onto, as we have already observed that any $`\sigma \mathrm{๐ฆ๐บ๐
}(/๐ฆ)`$ generates an automorphism of $`๐ญ_{}[]`$ fixing $`๐ญ_{}[๐ฆ]`$ via $`\mathit{\varrho }^\alpha \mathit{\varrho }^{\sigma (\alpha )}`$. Suppose now that $`\mathrm{\Pi }(\sigma )=1`$ for $`\sigma `$ in $`\mathrm{๐ฆ๐บ๐
}\left(๐ญ_{}[]/๐ญ_{}[๐ฆ]\right)`$. Then $`\sigma \mathrm{๐ฆ๐บ๐
}\left(๐ญ_{}[]/\right)`$, but by Theorem 8, the latter group is trivial. โ
We now concentrate on the case of a finite Galois extension $`K/`$ and consider each of the operations $``$ and $``$ separately. We will work with the nonlinear number field $`\overline{๐ญ}_{}[K]`$ = $`๐ง_{}[K]`$. Let $`\mathrm{๐ฆ๐บ๐
}_{}\left(\overline{๐ญ}_{}[K]/K\right)`$ denote those isometries fixing $`K`$ and homomorphic with respect to $``$ only. $`\mathrm{๐ฆ๐บ๐
}_{}\left(\overline{๐ญ}_{}[K]/K\right)`$ is defined similarly.
Denote by $`๐ด\left(๐ง_{}[K]\right)`$ the group of unitary operators of $`๐ง_{}[K]`$. The action of $`๐ซK_{\mathrm{}}`$ by translation in $`_K`$, $`๐ณ๐ณ+๐ซ`$, induces an action on $`๐ง_{}[K]`$ by
$$\mathrm{\Phi }_๐ซ(F)=a_\alpha \mathrm{exp}(2\pi i\mathrm{๐ณ๐}(\alpha ๐ซ))\mathit{\varrho }^\alpha $$
for $`F=_\alpha a_\alpha \mathit{\varrho }^\alpha `$, yielding a faithful representation
$$\mathrm{\Phi }:K_{\mathrm{}}๐ด\left(๐ง_{}[K]\right).$$
###### Proposition 4.
The projectivization $`[\mathrm{\Phi }]`$ of $`\mathrm{\Phi }`$ defines a monomorphism
$$[\mathrm{\Phi }]:K_{\mathrm{}}\mathrm{๐ฆ๐บ๐
}_{}\left(\overline{๐ญ}_{}[K]/K\right).$$
###### Proof.
For $`[\mathit{\varrho }^\alpha ]K`$ and $`๐ซK_{\mathrm{}}`$, $`[\mathrm{\Phi }]_๐ซ([\mathit{\varrho }^\alpha ])=[\mathrm{exp}(2\pi i\mathrm{Tr}(\alpha ๐ซ))\mathit{\varrho }^\alpha ]=[\mathit{\varrho }^\alpha ]`$. For any $`[F]`$, $`[G]๐ญ_{}[K]`$, let $`[f]`$, $`[g]`$ be the projective classes of their boundary functions. Then $`[\mathrm{\Phi }]_๐ซ([F][G])`$ is the element of $`๐ญ_{}[K]`$ whose boundary function is
$`[\mathrm{\Phi }]_๐ซ([f][g])`$ $`=`$ $`\left[{\displaystyle \underset{\alpha }{}}\left({\displaystyle \underset{\alpha _1+\alpha _2=\alpha }{}}a_{\alpha _1}b_{\alpha _2}\right)\mathrm{exp}(2\pi i\mathrm{๐ณ๐}(\alpha ๐ซ))๐ป^\alpha \right]`$
$`=`$ $`\left[{\displaystyle \underset{\alpha }{}}\left({\displaystyle \underset{\alpha _1+\alpha _2=\alpha }{}}a_{\alpha _1}\mathrm{exp}(2\pi i\mathrm{Tr}(\alpha _1๐ซ))b_{\alpha _2}\mathrm{exp}(2\pi i\mathrm{๐ณ๐}(\alpha _2๐ซ))\right)๐ป^\alpha \right]`$
$`=`$ $`[\mathrm{\Phi }]_๐ซ([f])[\mathrm{\Phi }]_๐ซ([g]),`$
which is the boundary function of $`[\mathrm{\Phi }]_๐ซ([F])[\mathrm{\Phi }]_๐ซ([G])`$. โ
In like fashion, we may define a flow on $`\overline{๐ญ}_{}[K]`$ respecting $``$ as follows. For a vector $`๐ณK_{\mathrm{}}`$ we denote by $`\mathrm{log}|๐ณ|`$ the vector
$$(\mathrm{log}|z_{\nu _1}|,\mathrm{},\mathrm{log}|z_{\nu _d}|)$$
when $`K`$ is real, or when $`K`$ is complex
$$(\mathrm{log}|z_{\mu _1}|,\mathrm{log}|z_{\mu _1}|\mathrm{},\mathrm{log}|z_{\mu _s}|,\mathrm{log}|z_{\mu _s}|).$$
Then for $`F=_\alpha a_\alpha \mathit{\varrho }^\alpha `$ we define
$$\mathrm{\Psi }_๐ซ(F)=\underset{\alpha K}{}a_\alpha \mathrm{exp}\left(2\pi i\mathrm{๐ณ๐}(๐ซ\mathrm{log}|\alpha |)\right)\mathit{\varrho }^\alpha .$$
This defines a faithful representation
$$\mathrm{\Psi }:K_{\mathrm{}}๐ด\left(๐ง_{}[K]\right).$$
The following is proved exactly as Proposition 4:
###### Proposition 5.
The projectivization $`[\mathrm{\Psi }]`$ of $`\mathrm{\Psi }`$ defines a monomorphism
$$[\mathrm{\Psi }]:K_{\mathrm{}}\mathrm{๐ฆ๐บ๐
}_{}\left(\overline{๐ญ}_{}[K]/K\right).$$
Recall that the idele class group of $``$, $`C_{}`$, may be identified with $`_+^{}\times \mathrm{Gal}(\overline{}^{\mathrm{ab}}/)`$. We have the following Corollary to Propositions 4 and 5:
###### Corollary 1.
There are monomorphisms
$$๐ข_{}\mathrm{๐ฆ๐บ๐
}_{}\left(\overline{๐ญ}_{}[^{\mathrm{ab}}]/\right),๐ข_{}\mathrm{๐ฆ๐บ๐
}_{}\left(\overline{๐ญ}_{}[^{\mathrm{ab}}]/\right).$$
We end by noting that the above structures have an interpretation within the von Neumann depiction of quantum mechanics. The space $`\overline{๐ญ}_{}[K]`$ may be viewed as the space of states of a quantum mechanical system with $`d`$ degrees of freedom. We view each of the flows $`\mathrm{\Phi }`$ and $`\mathrm{\Psi }`$ as generating the coordinate observables of two distinct physical systems and write
$$\mathrm{\Phi }_๐ซ=\mathrm{exp}(2\pi iH_{},๐ซ),\mathrm{\Psi }_๐ซ=\mathrm{exp}(2\pi iH_{},๐ซ),$$
where $`H_{}=\text{diag}[q]_{qK}`$ and $`H_{}=\text{diag}[\mathrm{log}|q|]_{qK}`$ are the associated (vector-valued) Hamiltonian operators. The set of stationary states for each of $`H_{}`$ and $`H_{}`$ is the field $`K`$.
Each $`[F]\overline{๐ญ}_{}[K]`$ defines a Cauchy multiplication operator $`M_{}([F])`$ and a Dirichlet multiplication operator $`M_{}([F])`$. It is easy to see that if $`[F]`$ has a representative $`F`$ whose Fourier coefficients are real, then $`M_{}([F])`$ defines an observable i.e. the projectivization of a self-adjoint operator of $`๐ง_{}[K]`$. This is not true of $`M_{}([F])`$ due to the fact that the Haar measure on $`\widehat{๐}_K`$ โ which we use to define the Hardy inner-product โ is invariant with respect to addition but not multiplication. This suggests that regarding the system defined by $`H_{}`$, it may be more natural to use the Hardy space of functions holomorphic on a hyperbolized idele class group $`๐ข_K`$ (with its multiplicatively invariant Haar measure). A formal computation shows that the operators $`M_{}([F])`$ are self-adjoint for the โidelicโ Hardy inner-product when the Fourier coeficients of some representative $`F`$ are real. We suspect that the eigenvalues of $`M_{}([F])`$ are equal or related in a straightforward manner to the imaginary parts of the zeros of a meromorphic extension of a Dirichlet type series corresponding to $`[F]`$.
## 11. Appendix: Errata to the Published Version \[4p\]
In what follows, the revised version (that is, this version) is denoted \[R\].
1. The definition given of nonlinear number field was erroneous. On page 582, line 15 of \[4p\] the definition of $`^{}[K]`$ should read:
(20)
$$^{}[K]=[K]I_K$$
and not โ$`[K]/I_K`$โ. This correction must be carried out in the more general definition e.g. on page 586 line 30 of \[4p\] it should read $`๐ง_{}^{}[K]=๐ง_{}[K]๐จ_K`$, where $`๐จ_K`$ is the kernel of the trace map on its domain of definition. See ยง9 of \[R\].
2. The discussion of the Dirichlet product structure of $`\mathrm{๐ข๐๐บ๐}(\widehat{๐}_K)`$ (part 1 of Theorem 5, page 581 of \[4p\]) was incomplete. See ยง7 of \[R\].
3. In part 2 of Theorem 5 of \[4p\], the additive group $`(O_K,+)`$ was mistakenly identified with the character group $`\mathrm{๐ข๐๐บ๐}(๐_K)`$ of the Minkowski torus $`๐_K`$. Rather, it is the inverse different $`๐ก_K^1O_K`$ which is identified with $`\mathrm{๐ข๐๐บ๐}(๐_K)`$. The character group $`\mathrm{๐ข๐๐บ๐}(๐_K)`$ is thus an $`O_K`$-module extension of $`O_K`$. See See ยง7 of \[R\].
4. On line 15, page 583 of \[4p\], it was incorrectly asserted that the Cauchy and Dirichlet products are fully defined on the Hilbert space $`L^2(\widehat{๐}_K,)`$. These operations only extend partially so that for each element $`fL^2(\widehat{๐}_K,)`$ one must specify the Cauchy and Dirichlet domains $`\mathrm{๐ฃ๐๐}_{}(f)`$, $`\mathrm{๐ฃ๐๐}_{}(f)`$ of elements $`g`$ for which $`fg`$ resp. $`fg`$ make sense. The definition of nonlinear number field, which appears in Definition 2 on page 587 of \[4p\], must be adjusted accordingly by replacing everywhere the phrase โdouble semigroupโ by โpartial double semigroupโ to take into account this correction. See ยงยง8,9 of \[R\].
5. The proof of Lemma 1 on page 589 of \[4p\] was incorrect. A correct proof can be found in ยง10 of \[R\] (where it is known as โLemma 2โ).
6. The discussion of the idele class group found on pages 591-592 of \[4p\] is only valid for $`K=`$, so that the idele class group $`๐ข_K`$ should be replaced by $`๐ข_{}`$.
7. Apart from implementing the above corrections, ยง6 of the revised version \[R\] contains an enhancement of the hyperbolization $`_K`$ (page 579-580 of \[4p\]). Furthermore, the $`\theta `$-holomorphic grading of functions on $`_K`$ (described on page 585 of \[4p\]) as been expanded by a complex sign set along the complex places. This is described in ยง8.2 of \[R\].
|
warning/0506/cond-mat0506395.html
|
ar5iv
|
text
|
# The effect of disorder on the hierarchical modularity in complex systems
## 1 INTRODUCTION
During the last few years a great number of discoveries have formed our view of complex networks. It turned out that the scale-free property is an ubiquitous feature of most real networks, like the World Wide Web, metabolic networks, and collaboration networks, etc. . The hierarchical topology of most of these networks has also been studied . According to recent studies , a hierarchically modular organization lies beneath this topology: โthere are many highly integrated small modules which group into a few larger modules, which in turn can be integrated into even larger modulesโ . The so-called *hierarchical network model* was introduced , which is a simple illustration of the idea above. However, the modularity of real networks is still an open question. The problem of the unambiguous identification of modules or communities of a system at different hierarchical levels has not been solved ultimately, though important advances have been achieved <sup>1</sup><sup>1</sup>1In the paper by Clauset et al. a quantity called modularity is defined based on already identified modules. We want to avoid this a priori identification (which is often a hard task) and we use the term modularity in the every-day sense for systems consisting of well identifiable and separable composite units.. The *average-linkage hierarchical clustering algorithm* groups the points of a network according to the *topological overlap* between them . Modularity is then attributed to the clustered structure of the overlaps and hierarchical modularity is obtained if the observed clusters can be interpreted such, i.e. if the clusters are hierarchically nested. The advantage of this method is that it avoids the delicate problem of a priori identification of modules.
In Ref. metabolic networks were studied and the interesting conclusion was drawn that complex networks show hierarchical modularity if they are characterized by a clustering coefficient $`C(k)`$ decaying with a power law as a function of the degree $`k`$ of the nodes (often with an exponent close to unity). The clustering coefficient is a measure of the inter-connectedness of the neighbors of a particular node and it is clearly related to the community structure. In Ref. this concept was nicely illustrated by a regular network and used to analyse the experimental data obtained for metabolic networks.
In the present work we address the question of the effect of randomness on the hierarchical modularity. In order to do so we introduce a measure of modularity in scale free (hierarchical) systems without using identification of modules. We apply this concept to randomly rearranged regular fractals and networks.
## 2 MODULARITY OF FRACTALS
### 2.1 The Vicsek Snowflake
Understanding the modularity of networks is rather difficult, because it is hidden in the networkโs topology. Therefore, first we show the effect of modularity in the case of fractals. We investigated the modularity of *randomly rearranged* Vicsek snowflakes embedded into two dimensions.
Our initial object was the well known fractal shown on the left of Fig. 3, which has regular self-similarity . At the highest hierarchical level, it consists of five well-separated blocks, and each of these blocks contains five smaller blocks in the same fashion, and so on. In our case, this hierarchical structure has a finite resolution, so it repeats itself until reaching an elementary block size or lower cutoff. The fractalโs dimension is given by the formula of $`D=\mathrm{ln}5/\mathrm{ln}31.465`$. Obviously, for the regular fractal, the blocks (which have a finer structure not resolved at this level) can be considered as modules at each level of the hierarchy. As these modules are intact until the next, finer level of the hierarchy is reached, there is a clear sequence of separation of scales: We have *hierarchical modularity*. Our goal is to measure quantitatively how this hierarchical modularity changes at different levels of random rearrangement.
### 2.2 Random Rearrangement of the Vicsek Snowflake
We generated the initial regular Vicsek snowflake by taking a $`3^N\times 3^N`$ sized bit matrix evenly filled with *ones*. One bit corresponds to an elementary block of the fractal. Then, starting with the highest hierarchical level, our algorithm โcut outโ the five largest blocks, by turning the corresponding bits to *zero*. After that, taking the next sublevel, the next twenty five sub-blocks were cut out from the five large blocks having been generated in the previous step. After $`N`$ cutting steps one gets the expected Vicsek snowflake with finite resolution.
The random rearrangement algorithm is based on this cutting method above. We got a continuous set of randomly rearranged fractals with a randomness being controlled by the parameter $`p[0,1]`$ in the following way: At each hierarchical level, we shifted all of the five blocks with probability $`p`$ to the remaining free spaces in the larger block. More precisely, we tossed for each of the five blocks whether they would be shifted or not, then we cut out the non-shifted ones, and *after* that we randomly placed the remaining ones to the free spaces with uniform distribution. So, before all of the cutting steps our rearrangement algorithm made that draw described above.
In this model $`p=0`$ corresponds to the original Vicsek snowflake, and $`p=1`$ generates a totally random fractal, in which all the five blocks take place in the nine rooms with uniform random distribution at all hierarchical levels. By visual inspection it is clear that the degree of hierarchical modularity decreases as $`p`$ changes from $`0`$ to $`1`$ (see Figs. 3 \- 3), and in case $`p=1`$, it should be vanished (at least on the average). Note that a $`p=1`$ fractal is also made in a hierarchical manner, but the applied random rearrangement destroys its regular modular structure by mixing the modules together at every hierarchical level.
### 2.3 Measurement of Hierarchical Modularity
In Ref. several methods for measuring the dimension of regular fractals were compared. In order to quantify modularity, we apply the *box counting method*, where we cover growing regions of the fractal with a growing square window started from the centre, and at each step we count the mass (the number of elementary blocks) inside the window. By this means, one gets a mass function: the inside mass of the fractal as a function of the linear size of the window. As pointed out in Ref. , for regular fractals this mass function is not a straight line in a log-log plot: There are periodic deviations due to the above mentioned separation of scales, i.e., due to modularity. Therefore we consider the periodicity of these deviations from the straight lines as the measure of modularity.
From our model, presented in Sec. 2.2, one gets different deviation functions for different values of $`p`$. In order to analyse these functions we applied Fast Fourier Transform (FFT) to them.
### 2.4 Results
In this subsection we discuss the results of the measurement of modularity described above in Sec. 2.3.
The log-log plots of the mass deviation functions as a function of the window size are plotted in Fig. 4 for different values of parameter $`p`$. For non-zero $`p`$, each curve is the average of $`200`$ randomly rearranged fractals (with same $`p`$).
Apparently, the mass deviation functions reflect the hierarchical modularity of the fractals of different $`p`$ values. In the $`p=0`$ case, the function has an inherent structure corresponding to its high degree of modularity. This scale-free structure is more visible on the farther periods of the function, because there we have more points. This inherent structure of the curves vanishes as $`p1`$. The maximum deviation also decreases as we increase $`p`$, and the functions become smoother and smoother (apart from the basic high peaks with values of $`0`$), indicating the vanishing modularity of the fractals. The $`0`$-valued peaks correspond to powers of $`3`$: when the linear size of the window reaches powers of $`3`$, the window contains a whole sub-fractal with the exact dimension. This is the consequence of not mixing the elements in a continuous way. The deviation is non-positive for all values of $`p`$, which can be explained by the geometry of the applied rectangular windowing method in our special case.
The vanishing modularity could be better represented taking the FFT spectrum of the functions in Fig. 4. We analysed the last three periods of the functions with FFT, where they have the finest inherent structure. The results are plotted in Fig. 5.
Because of the triadic organization of the fractals, the fundamental spatial frequency is $`f=1/\mathrm{log}32.096`$. This corresponds to the $`0`$-valued peaks mentioned above, therefore the first peaks in the FFT spectra are direct consequences of the geometry of our method. For the regular fractal ($`p=0`$) one could notice the first, second and third harmonics which indicate the scale-free property of the deviation function. The first harmonics, which indicates the fine structure of the deviation functions, significantly decreases for increasing values of $`p`$, and the second and third harmonics totally disappear, what verifies the decreasing degree of hierarchical modularity of the fractals for increasing rearrangement probability $`p`$.
## 3 MODULARITY OF NETWORKS
### 3.1 The Randomization of the Hierarchical Network Model
Our starting point was the deterministic, modular hierarchical network model of Ravasz and Barabรกsi (see Fig. 6).
Its main features correspond to the real networks : the degree distribution is scale-free, the clustering coefficient is independent of the size, and follows a power-law as a function of degree: $`C(k)k^\alpha `$. In real networks, $`\alpha `$ is usually about 1.
The regular model also has obviously hierarchically modular structure. As real grown networks are to some extent random, the question raises how the modularity is influenced by the randomness. In order study this problem, we used a link randomization procedure, earlier already applied to investigate the influence of randomness on synchronization . In this model two links were chosen randomly, and one node of both links was exchanged between the two links. This process was executed $`M\times p`$ times, where $`M`$ is the number of the links and $`p`$ is the control parameter of the randomization. This method conserves the degree distribution, as it does not change the degree of any node.
As $`p`$ increases from $`0`$ to $`1`$, the average clustering coefficient $`\overline{C}`$ first falls rapidly then becomes constant (see Fig. 7).
For $`p=1`$, the degree dependence of $`C(k)`$ has two regions. It seems that the very low degree ($`k<10`$) nodesโ behavior is significantly different from the rest. Due to this effect and to the crossover it causes, the asymptotics sets in rather late. However, it is clear that the distribution is very broad and it has possibly a power law tail (Fig. 8). Also due to the small $`k`$ anomaly the average $`C`$ decrease with increasing system size for the considered number of nodes.
### 3.2 Topological Overlap Matrix
To investigate the presence or absence of modules and hierarchical structure, we used the topological overlap matrix (TOM) of the network . The $`ij`$ element of this matrix, $`T_{ij}`$ equals the number of mutual neighbours of the nodes $`i`$ and $`j`$ (plus $`1`$ if $`i`$ and $`j`$ are connected), normalized by the minimum degree of $`i`$ and $`j`$, so $`T_{ij}`$ is between $`0`$ and $`1`$. Thus, the $`i`$-th row and column represents the overlap of the node $`i`$ with the other nodes of the network ($`T_{ii}`$ is defined as $`1`$). If there is an isolated module of densely interconnected nodes in the network, and rows/columns representing the nodes of the module are close to each other in the topological overlap matrix, the module appears in the matrix as a square centred on the diagonal with elements close to $`1`$. Of course, to enable this interpretation, the right sequence of the nodes has to be determined, just according to their overlap values, as it is outlined in the next paragraph. The hierarchy of modules is represented by a system of smaller and smaller (and more and more cohesive) squares embedded into the larger squares (in which the overlap of the modules decreases with the module size). These features can be easily observed on the matrix of the deterministic model (Fig. 9).
The sequence of the rows/columns representing the nodes in the matrix is essential to recognize the modules and hierarchy. The rows/columns concerning nodes with big overlap have to be next to each other in the TOM, forming squares centred on the diagonal. In order to get the right sequence, we slightly modified the average linkage clustering method . The original algorithm builds communities joining nodes into โsupernodesโ (see Fig. ??: hierarchical tree, after a contraction the order in the new supernode has to be determined). In each step two nodes are joined, meanwhile the TOM decreases by one row and column.
The basic steps of the algorithm:
* First it finds the highest element in the TOM and contracts the two corresponding nodes into a supernode.
* The corresponding rows and columns of the two original nodes in the matrix are contracted into 1 row and 1 column (matrix elements are averaged), then the next round is started.
* This procedure is repeated until the TOM decreases into a $`1\times 1`$ matrix (every node joined into one supernode).
The previous steps describe the original algorithm. Because it was unable to reconstruct even the simple regular case shown on Fig. 9, some modifications were applied:
* It is possible that the above algorithm finds more than one elements with the same high value in the same step. In this case the contraction resulting the smallest supernode is performed. If this quantity is also degenerated, then the choice is made at random. The goal of this modification is to make the growing of the supernodes
* We are searching for clusters, not just pairs. Therefore the algorithm examines that the selected contraction is good from the view of building a cluster (containing more than $`2`$ nodes). If a better contraction is possible, that will be executed.
This way a sequence of contractions emerges. Parallel with the above algorithm placing the contracted nodes next to each other, the โclusteredโ sequence of the nodes appear. To make the above algorithm more clear, there is a small example:
Applying this algorithm, the TOM of the randomized deterministic network is visually interpretable (Fig. 11):
modular organization is not recognizable with this clustering algorithm after the randomization process for $`p=1`$. This network is therefore scale free, it has a broad, probably power law $`C(k)`$ distribution โ without a modular structure.
We would like to make the vanishing of the regular modularity quantitatively accessible in a similar way as in the case of the fractals: First, the elements of the TOM are raised to the third power in order to weaken the influence of the many small elements (to ยดmake contrastยด). Then the elements are projected (summed) perpendicular to the diagonal (note that it means $`2N1`$ sums for a $`N\times N`$ matrix), and the result is Fourier-analysed (Fig. 12).
A peak in the average of the amplitude-spectra indicates the presence of equal-sized modules. As $`p`$ increases from zero, the peaks indicating the presence of the 5-node and 25-node modules are decreasing to zero rapidly (see Fig. 13).
## 4 DISCUSSION
In this paper we analysed the effect of randomness on the hierarchical modularity of scale free structures. The quantitative analysis was based on an important feature of modularity: the separation of scales. We studied regular structures (Vicsek snowflake, deterministic modular network) where randomness was introduced such that scale freeness and other features (broad distribution of $`C(k)`$) were maintained. Appropriate characteristics of modularity were chosen using Fourier components of the mass deviation function (for fractals) or of projections of the elements of the TOM (networks). In both cases we observed a significant decrease of hierarchical modularity with increasing randomness. For networks, we also observed a rapid fall for small values of $`p`$ in the clustering coefficient and in the Fourier peaks, suggesting a crossover similar to the phenomenon described in Ref. .
It has to be emphasized that the applicability of the Fourier analysis is based on the fact that the regular, hierarchical structures had a regular scale separation. If the separation of scales is less regular, hierarchical modularity can still exist, however, sufficient randomization would destroy it in this case too. In many real networks signatures of hierarchical modularity could be found and the same is true for some model networks like the Holme-Kim network . This indicates that the level of irregularity in these networks is far from that reached by randomization in our models.
## 5 ACKNOWLEDGEMENTS
Thanks are due to T. Vicsek, E. Ravasz and A.-L. Barabรกsi for important discussions.
|
warning/0506/quant-ph0506195.html
|
ar5iv
|
text
|
# Temporal shape manipulation of adiabatons
## I Introduction
Electromagnetically induced transparency (EIT) and coherent population trapping (CPT) can facilitate coherent control of light under propagation through a medium 1 ; 2 . In addition to their fundamental interest, investigations of these processes are stimulated by practical possibilities, such as manipulating a group velocity of light and light storage in atomic medium 3 ; 4 , enhanced nonlinear optical processes 5 , quantum memory 4 and so on.
The CPT is a quantum interference effect and takes place under resonance interaction of two laser fields (probe and coupling) with three-level atomic systems. The essence of this effect is that under certain conditions atoms are trapped into the coherent superposition of two lower states $`|1`$ and $`|2`$, which is called CPT-state 6 ; 7 . Under CPT condition the medium becomes coherent and possesses unusual properties, many of which contradict with the intuitive views. The CPT leads to the maximal coherence at the Raman transition and the medium becomes transparent for the probe and coupling pulses 5 ; 8 . This phenomenon allows recording, storing and reading of information about strong optical pulses 9 ; 10 , to control the degree of excitation of spatially localized regions inside an absorbing three-level medium 11 and to generate matched pulses 12 ; 13 , adiabatons 14 ; 15 and dressed-field pulses 16 .
Recently it was shown how EIT can be used for coherent control of the weak pulse shape 17 . The idea is following. Under EIT the weak probe pulse propagates with a slow group velocity depending on an intensity of the coupling field. If the intensity of the coupling field depends on a time, different points of the probe pulse experience different values of intensity of coupling field and travel with different propagation velocities, giving rise to temporal reshaping of the probe. A proper choice of the temporal shape of the coupling pulse allows control and manipulation of the probe pulse envelope. In this paper we generalize this method for controlling the temporal shape of the intense probe pulse using the peculiarities of CPT propagation dynamics. Temporal compression of adiabatons is demonstrated as special case of pulse tailoring.
## II Principal equations
Consider the propagation of two laser pulses in a medium consisting of three-level atoms (Fig. 1). Pulses propagate along an axis $`z`$ in one direction. A probe pulse (with the slowly varying envelope $`E_p(t)`$ and frequency $`\omega _p`$) is tuned on resonance with $`|3|1`$ transition, and the coupling pulse ($`E_c(t)`$, $`\omega _c)`$ is tuned so that exact two-photon resonance between states $`|1|2`$ is achieved. The coupling pulse is switched on earlier and switched off later than probe. The pulse durations $`T_p`$ and $`T_c`$ are much less than any relaxation times of atoms and $`T_p<T_c`$. Intensities of both pulses are comparable.
A propagation of pulses can be described by Schrรถdinger equation and reduced wave equations for Rabi frequencies (Maxwell-Schrรถdinger equations) which should be solved self-consistently. For the case when the fields are in resonance with their respective transitions, Maxwell-Schrรถdinger equations are
$$\frac{}{\tau }\left(\begin{array}{cccccccccccccccccccc}a_1& & & & & & & & & & & & & & & & & & & \\ a_2& & & & & & & & & & & & & & & & & & & \\ a_3& & & & & & & & & & & & & & & & & & & \end{array}\right)=i\left(\begin{array}{cccccccccccccccccccc}0& 0& G_p^{}& & & & & & & & & & & & & & & & & \\ 0& 0& G_c^{}& & & & & & & & & & & & & & & & & \\ G_p& G_c& 0& & & & & & & & & & & & & & & & & \end{array}\right)\left(\begin{array}{cccccccccccccccccccc}a_1& & & & & & & & & & & & & & & & & & & \\ a_2& & & & & & & & & & & & & & & & & & & \\ a_3& & & & & & & & & & & & & & & & & & & \end{array}\right),$$
(1)
$$\frac{}{\zeta }\left(\begin{array}{cccccccccccccccccccc}G_p& & & & & & & & & & & & & & & & & & & \\ G_c& & & & & & & & & & & & & & & & & & & \end{array}\right)=i\left(\begin{array}{cccccccccccccccccccc}K_pa_1^{}a_3& & & & & & & & & & & & & & & & & & & \\ K_ca_2^{}a_3& & & & & & & & & & & & & & & & & & & \end{array}\right).$$
(2)
Here $`\zeta =z,\tau =tz/c`$ โ space and time coordinates in a frame moving with light velocity $`c`$ in empty space; $`a_{1,2,3}`$ โ the probability amplitudes of atomic states; $`2G_{p,c}=E_{p,c}d_{1,2}/\mathrm{}`$ โ the Rabi frequencies of fields; $`E_{p,c}`$ โ the probe and coupling field strengths; $`d_{13,23}`$ โ the electrical dipole moments of the relevant atomic transitions; $`\mathrm{}`$ โ the Plank constant; $`K_{p,c}=2\pi N\omega _{p,c}\left|d_{13,23}\right|^2/\mathrm{}c`$ โ the field-atomic medium coupling coefficients; $`N`$ โ the atomic concentration. Initially all atoms are in the ground state $`|1`$: $`a_{1,2,3}(\tau =\mathrm{},\zeta )=(1;0;0)`$. The solution of Eqs. (1) and (2) gives the complete evolution of the atom-field system.
The analytical solution of the equation system (1,2) is possible only in adiabatic approximation 8 ; 14 . In this case $`\left|a_3\right|1`$ and $`G_p/G_c=a_2/a_1`$. The condition $`\left|a_3\right|1`$ means, that the population of intermediate state $`|3`$ is close to zero in the interaction of pulses with atoms. The population is trapped in a coherent superposition of states $`|1`$ and $`|2`$ โ the effect of CPT. Under CPT pulses do not interact with medium 2 ; 7 . It means that pulses can propagate practically without absorption.
In the adiabatic approximation Eqs. (1) and (2) lead to photon number conservation law
$`{\displaystyle \frac{G_c^2(\tau ,\zeta )}{K_c}}+{\displaystyle \frac{G_p^2(\tau ,\zeta )}{K_p}}`$ $`=`$
$`{\displaystyle \frac{G_c^2(\tau ,\zeta =0)}{K_c}}+{\displaystyle \frac{G_p^2(\tau ,\zeta =0)}{K_p}}`$ $`=`$ $`V(\tau ,\zeta =0).`$
The conservation law implies that any change in the probe pulse is compensated by a corresponding change in the coupling pulse and so $`V`$ does not depend on the space variable during the propagation. The input fields determine the temporal shape of $`V`$.
In the adiabatic approximation field equations (2) have the form
$$\frac{\stackrel{}{G}}{\zeta }=\widehat{\mathrm{K}}\frac{1}{\stackrel{}{G}^2}\frac{\stackrel{}{G}}{\tau },\stackrel{}{G}=(G_c,G_p).$$
(3)
Let us introduce new variable, mixing angle $`\theta (\tau ,\zeta )`$, which is determined as $`\mathrm{tan}\theta =G_p/G_c`$. The equation for $`\theta (\tau ,\zeta )`$ is 8
$$\frac{\theta }{\zeta }+\frac{K^2\left(\theta \right)}{K_pG_c^2+K_cG_p^2}\frac{\theta }{\tau }=0,$$
(4)
$$K\left(\theta \right)=\left(K_pG_c^2+K_cG_p^2\right)/\stackrel{}{G}^2=K_p\mathrm{cos}^2\theta +K_c\mathrm{sin}^2\theta .$$
The mixing angle appears constant along characteristic curves
$$\zeta (\tau ,\tau _0)=K^2\left(\theta \left(\tau _0,\zeta =0\right)\right)_{\tau _0}^\tau \left(K_pG_c^2+K_cG_p^2\right)๐\tau ^{},$$
$`\tau _0`$ โ the time when given characteristic curve intersects the input of medium. $`\tau _0`$ is to be determined from last equation and the solution of Eq. (4) can be written as:
$$\theta (\tau ,\zeta )=\theta (\tau _0,0).$$
All physical quantities can be expressed through the mixing angle $`\theta (\tau ,\zeta )`$ 8
$$\left(\begin{array}{cccccccccccccccccccc}G_p& & & & & & & & & & & & & & & & & & & \\ G_c& & & & & & & & & & & & & & & & & & & \end{array}\right)=2\sqrt{\frac{\left(K_pG_c^2+K_cG_p^2\right)|_{\zeta =0}}{K\left(\theta \right)}}\left(\begin{array}{cccccccccccccccccccc}\mathrm{sin}\theta & & & & & & & & & & & & & & & & & & & \\ \mathrm{cos}\theta & & & & & & & & & & & & & & & & & & & \end{array}\right),$$
(5)
$$\left(\begin{array}{cccccccccccccccccccc}a_1& & & & & & & & & & & & & & & & & & & \\ a_2& & & & & & & & & & & & & & & & & & & \\ a_3& & & & & & & & & & & & & & & & & & & \end{array}\right)=\left(\begin{array}{cccccccccccccccccccc}\mathrm{cos}\theta & & & & & & & & & & & & & & & & & & & \\ \mathrm{sin}\theta & & & & & & & & & & & & & & & & & & & \\ \left|(\stackrel{}{G}/\stackrel{}{G}^2)/\tau \right|& & & & & & & & & & & & & & & & & & & \end{array}\right).$$
(6)
The solution (5) and (6) can be applied only within the area of adiabaticity which is limited by the relation 14
$$|G_c\frac{G_p}{\tau }G_p\frac{G_c}{\tau }|(G_c^2+G_p^2)^{3/2}.$$
In contrast to usual steady state solution which does not depend on initial conditions, the space-time evolution of the probe and coupling pulses under CPT conditions depends on the pulse forms at the input of medium. Some aspects of this dependance are discussed in 8 .
## III Temporal shape control <br>of the probe pulse by CPT: <br>compression of pulses
Figure 2 demonstrates the evolution of the Rabi frequencies of pulses and the atomic coherence under CPT in the case of the Gaussian pulses at the input of medium, also $`K_p=K_c`$ and $`T_c=10T_p`$. It is visible, that the probe pulse is gradually depleted and the control gets stronger. Note that the pulse shape at the initial stage of propagation shows very little variation with the length, which may considerably exceed the linear absorption length. Complete reemitting of the probe pulse into the control one during propagation is possible. The atomic or Raman coherence is excited only in a part of medium, that is spatially localized. Outside of this area, atoms remain unexcited in the ground state. The spatial distribution of atomic coherence keeps the information about pulses. This can be used for record and storage of information about the probe pulse in the CPT-modified medium 9 ; 10 .
In a case, when $`T_pT_c`$ and the amplitude of the coupling pulse is constant, the probe and coupling pulses have complementary envelopes and propagate without form variation and with equal group velocity. Such pulses are called adiabatons 14 .
Under unequal propagation constants $`K_{p,c}`$ (unequal oscillator strengths of two optical transitions in the atom) the adiabatons are not shape-preserving but undergo a front sharpening (Fig. 3): under $`K_p<K_c`$ a back edge becomes steeper (dash-dot line) , and under $`K_p>K_c`$ \- a leading edge becomes steeper (dashed line).
Since under CPT the space-time evolution of the probe pulse depends on the temporal shape of the coupling pulse, we can manipulate the shape of the probe pulse by proper choice of the coupling pulse envelope at the entry of medium. In this regard CPT can be viewed as a way of the coherent control of temporal pulse shaping. In particular, it is possible to choose such coupling pulse shape, that the trailing edge of the probe pulse travels faster than leading one. This results in the compression of probe pulse. Figure 4 demonstrates an example of the temporal compression of probe pulse using coupling pulse with the envelope shown in Fig. 4a (dashed line). A time evolution of pulses is much similar to adiabatons propagation 14 . Pulse propagation in this case can be treated as adiabatonic pair extended to time shape variation (quasi-adiabatons) since both pulse envelopes vary coherently and travel with equal velocity. The reason of compression is that the leading edge of probe pulse is slowed down more strongly than the trailing one. As a result the pulse is compressed in time under propagation through the medium.
Note that the compression effect is independent on the detailed temporal structure of the coupling pulse. The compression takes place also under a linear growth of amplitude of the coupling pulse. A finite spectral bandwidth of transparency window sets a limit to the temporal duration of the probe pulse that can travel in the medium without absorption.
The pulse compression takes place also in the case of unequal propagation constants $`K_{p,c}`$, which are defined by the oscillator strengths of transitions, as shown on Fig. 5.
Pulse compression is a particular case of temporal shaping. In the general case, the proper choice of the temporal shape of the coupling pulse allows to obtain probe pulse with different temporal shapes at output. For example, we can obtain a flat-top pulse (Fig. 6) or two-peaked pulse (Fig. 7) like in 17 .
For the results presented we have checked that the numerical solution of the Maxwell-Schrรถdinger equations and obtained analytical expression provide exactly the same results.
In conclusion, we have shown that temporal pulse compression can be achieved using CPT schemes. These processes present both fundamental interest and applications in nonlinear optics, because the compressed pulse as a light source can increase the efficiency of nonlinear processes.
This work was supported by Russian Foundation for Basic Research (grant 02 02 16325) and Krasnoyarsk Regional Science Foundation (grant 12F0042c).
|
warning/0506/nucl-ex0506002.html
|
ar5iv
|
text
|
# e+e--pair production in Pb-Au collisions at 158 GeV per nucleon
## 1 Introduction
Strongly interacting matter under extreme conditions of temperature and density is being created by colliding heavy nuclei in fixed target experiments at the Super-Proton Synchrotron (SPS) at CERN and at the Relativistic Heavy-Ion Collider (RHIC) at BNL. The motivation derives from the quest to discover โquark matterโ, the quark gluon plasma (QGP), in which quarks and gluons are deconfined and chiral symmetry is restored quark-matter , two complementary facets of the phase transition predicted by Quantum Chromodynamics (QCD). This transition is expected to occur at a critical energy density of $`ฯต`$ 0.7 GeV/fm<sup>3</sup> and a temperature of 170-180 MeV as finite-temperature lattice calculations have shown karsch02 . The initial energy density at full SPS energy is appreciably larger and reaches about 3 GeV/fm<sup>3</sup> in central Pb-Pb collisions, adopting Bjorkenโs longitudinal expansion scenario for a simple estimate bjorken83 . While the existence of the QGP has not yet been proven, there is circumstantial evidence that its transient formation is imprinted, in one way or another and to varying degrees, on measured distributions of final-state hadrons. Since the fireball terminates in an exploding multi-hadron final state, QGP signatures, if not collective in character, are prone to be masked by hadronic interactions.
Photons and dileptons are potentially more direct probes of the early collision stages since they escape from the impact zone nearly undisturbed by final-state interactions and have their largest emission rates in hot and dense matter. Moreover, according to the vector dominance model sakurai , dilepton production is mediated in the hadronic phase by the light neutral vector mesons $`\rho ,\omega `$, and $`\varphi `$ which mark the low-mass region by distinctive resonance peaks. Among these, especially the short-lived
$`\rho (770)`$ meson ($`\tau =`$ 1.3 fm/$`c`$) has acquired a key role as test particle for โin-medium modificationsโ of hadron properties close to the QCD phase boundary brownrho1991 ; rapp-wambach2000 ; b-r03 . Changes in position and width of the $`\rho `$ have been advocated already 20 years ago as precursor signals of the chiral transition pisarski1982 . Restoration of chiral symmetry in hot and dense matter has become one of the heavily discussed and exciting issues in non-perturbative QCD thermodynamics rapp-wambach2000 ; b-r03 ; hats96 ; b-r96 ; wilcz00 as the melting of the chiral condensate should cause rather drastic changes of the properties of the light vector mesons and thereby on the structure of dilepton spectra. Spectral function calculations on the lattice seem still far from providing model-independent guidelines for the study of thermal modifications of hadron properties karsch02 .
An enhancement of low-mass lepton pair production was first reported by CERES prl1995 ; wurm-qm95 and HELIOS-3 masera1995 ; angelis1998 on the basis of 200 GeV/n S-Au and S-W data, respectively. The CERES $`e^+e^{}`$ mass spectrum is shown in Fig. 1 together with the reference spectra for 450 GeV p-Be and p-Au collisions neutral-meson-pBe <sup>1</sup><sup>1</sup>1The notation of the ordinate in Fig. 1 is no longer in use; it should read as in all other mass spectra shown in this paper. The โcocktailโ has received minor adjustments in the meantime which do not affect any of the conclusions drawn.. While the p-A data are reproduced within errors by final state Dalitz and direct decays of neutral mesons as known from p-p collisions, electron pairs from S-Au collisions reveal a substantial enhancement in the mass range above 250 MeV/$`c^2`$.
At top SPS energy and close to the critical temperature, the prime candidate for โthermal radiationโ from the hadronic phase of the fireball pipi-anni ; mclerran85 is pion annihilation,
$$\pi ^+\pi ^{}\rho e^+e^{}.$$
(1)
This thermal process with a threshold at $`2m_\pi `$ is dynamically enhanced via the electro-magnetic form factor of the pion by the $`\rho `$ resonance cleymans1993 , with a dilepton branching of 1 in $`10^4`$. Yet, the $`\rho `$ serves not only as test particle, but its strong coupling to the $`\pi \pi `$ channel makes it also a major constituent of hot hadronic matter. Numerous theoretical approaches incorporating pion annihilation using vacuum properties of the $`\rho `$ meson, failed without exception to describe the data drees1996 .
This suggested in-medium changes of the $`\rho `$ spectral function that shift dilepton strength down to lower masses. The first calculations that were successful in describing both the CERES and the HELIOS-3 data made use of the scaling conjecture of Brown and Rho brownrho1991 , which postulates that the mass of non-strange vector mesons decreases in dense matter together with the scalar quark condensate, the order parameter of the chiral transition. This โdropping massโ scenario received independent support by work on QCD sum rules hatsuda-lee1992 , and it turned out like a tailor-made concept: linked to a fireball model likob95 ; likobs96 or embedded into transport calculations casseheko95 ; cassehekr96 , it gave excellent fits to the data.
But it was also pointed out dey90 ; rapp-chanfray-w1996 ; harada97 ; mishra02 that chiral symmetry considerations alone would only require that masses of chiral partners, here the vector meson $`\rho `$ and the axial vector meson $`a_1`$, become degenerate, but by no means necessarily massless. Moreover, significant mixing must accompany any mass shift when approaching the phase transition along $`T`$ kim-rapp-brown-rho1999 . The โdropping massโ scaling idea has been very recently revisited by the original authors b-r03 welcoming an alternative scheme of how chiral symmetry might be restored, the Georgi vector limit georgi90 in which the chiral partner of the (longitudinal) $`\rho `$ is the Goldstone pion, both becoming massless in approach of the chiral transition halasz97 ; harada01 . Thermal modifications of hadron properties in general, and of in-medium spectral functions in particular, have not yet come within reach of QCD lattice calculations with finite baryon density karsch02 . Very preliminary results suggesting dropping vector meson masses have been reported muroya2003 .
An alternative approach to explain the low-mass dilepton enhancement in the CERN SPS data focused on the calculation of spectral functions in a hot and strongly interacting hadron resonance gas at finite baryon density by conventional many-body techniques rapp-wambach2000 ; cassing-brat1999 . These confirmed earlier calculations of the two-pion self-energy in nuclear matter which indicated that the $`\rho `$ spectral function suffers significant broadening but only negligible shift in mass herm-frim-noe1993 ; chan-schu1993 . The many-body calculations combined with suitable reaction models describe the observations actually quite well klingl-weise1996 ; rapp-chanfray-w1997 ; cassing1997 ; rw98 : the low-mass wing of the broadened in-medium $`\rho `$ spectral function receives strong thermal Bose enhancement which results in an amplified dilepton strength considerably below the vacuum $`\rho `$ position, while simultaneously the yield at the vacuum position was depleted, consistent with approximate unitarity rapp-gale1999 .
Our discussion so far was limited to aspects of dilepton emission rates with possible modifications by the medium. However, total pair yields derive from space-time integration over a priori unknown density and temperature profiles which are usually modelled by hydro-dynamical sollfrank97 ; hung98 or microscopic transport cassing-brat1999 ; likobs96 ; casseheko95 calculations, or fireball models likob95 ; renk02 . Certainly, the external inputs, e.g. to the hydro-dynamical and fireball calculations, have to conform with whatever knowledge there is on initial conditions, depending on collision geometry, and on the ($`\mu `$, $`T`$) coordinates of the trajectory in the phase diagram. The medium modifications of the $`\rho `$ suggested by the SPS results seem to require a strongly interacting, hot and dense hadronic fireball with sufficient time spent between hadronisation and thermal freeze-out. If this time would be insufficient, the enhanced dilepton production may have stronger links to the hadronisation stage or the plasma phase than hitherto assumed.
The CERES Collaboration measured dilepton production in 158 GeV/n Pb-Au collisions in 1995 and 1996 with a greatly improved setup baur1996 ; ullrich-qm96 compared to the sulfur-beam experiment. The main objective of the Pb runs has been achieved: to corroborate with a large statistics sample the enhanced dielectron production at low masses for the heavy Pb-Au collision system ullrich-qm96 ; ravinovich-qm97 ; plb422 ; lenkeit-qm99 ; lenkeit-paris . Improved background rejection was achieved which was compulsory in an environment of very large rapidity density of hadrons and secondary photons. Among the physics goals considered most important for further insight into the nature of the processes at work was the centrality dependence of the enhancement.
The significance of the baryon chemical potential for in-medium modifications at SPS energies prevailing over that of pion number, or temperature, as found in most calculations cassing-brat1999 ; rapp-wambach2000 , but not in all bleicher00 , was given experimental support by the recent finding of an even somewhat larger enhancement measured in the CERES Pb-Au low energy run at 40 GeV/n adamova-ee03 ; damjanovic-phd , compared to that at 158 GeV/n.
This paper presents the combined results of all data on electron pair production in 158 GeV/n Pb-Au collisions taken by the CERES Collaboration in the years 1995 and 1996. Most of the analyses were performed in the course of Doctoral Dissertations in Heidelberg, Darmstadt and Rehovot dealing with the 1995 voigt-phd ; socol-phd and 1996 socol-phd ; lenkeit-phd ; hering-phd data. Publications of analysis results of the 1995 data ullrich-qm96 ; ravinovich-qm97 ; plb422 and the 1996 data lenkeit-qm99 ; lenkeit-paris are superseded by the combined results presented here. There are no major deviations of the unified results reported in this paper to those published previously for the separate data sets.
The paper starts with a description of the experimental setup and summarises the instrumental means CERES has at its disposal to cope with background. A detailed description of the data analysis is given in sect. 3 which concludes with the centrality determination. The Monte-Carlo simulation method applied for measuring reconstruction efficiency and optimising the rejection of combinatorial background is addressed in sect. 4. The โcocktailโ of hadron decays which serves as an important reference for electron-pair production is discussed in sect. 5. Results of both data sets are presented in sect. 6 which also includes a discussion of statistical and systematic errors. The final mass and transverse momentum spectra are presented and compared to the hadronic cocktail in sect. 7. Section 8 contains a physics discussion, from an experimentalist point of view, on the comparison of data to current theoretical models. The paper concludes by summarising what has been achieved and which issues are still open but might be clarified in the not too distant future.
## 2 The CERES experiment in 1995/96
CERES is dedicated proposal88 to the measurement of electron pairs in the low-mass range from 50 MeV/$`c^2`$ up to about $``$1.5 GeV/$`c^2`$; an upper mass limit is imposed by counting statistics due to the rapid decline in cross section. The spectrometer covers the pseudo-rapidity region close to mid-rapidity, 2.1 $`<`$ $`\eta `$ $`<`$ 2.65. It is axially symmetric around the beam and has 2$`\pi `$ azimuthal coverage. Transverse pair momenta are accepted down to $``$20 MeV/$`c`$ for masses above $``$400 MeV/$`c^2`$. These are great assets when investigating soft processes. CERES maintains transparency for hadrons and photons as strictly as possible.
A schematic view of the spectrometer in the run periods 1995 and 1996 is shown in Fig. 2. At the heart of the spectrometer are the two coaxial ring-imaging Cherenkov detectors rich\_set , one (RICH-1) within the other (RICH-2) along the beam and separated by a compact super-conducting solenoid for momentum analysis. A doublet of silicon-drift chambers (SiDC) chen93 replaces the previously used single SiDC to enable precise charged-particle tracking into RICH-1 for rejection of close tracks, and for off-line measurement of charged multiplicity. A multi-wire proportional counter (Pad Chamber) with pad readout was added behind the mirror of RICH-2 to provide external tracking downstream of RICH-2. The new tracking detectors outside the field were added to cope with the high multiplicities of Pb-Au collisions pb-proposal94 ; cogne95 . These upgrades are described in Ref. baur1996 and Ref. holl96 for RICH and SiDC, respectively. A multiplicity detector (MD) of plastic scintillators behind the Pad Chamber serves as first-level trigger device. Below we introduce the individual detector components in order of their arrangement along the beam.
### 2.1 The target area
The target area, hardly visible in Fig. 2 and enlarged in Fig. 3, comprises the segmented target, the doublet of SiDCโs, the light-collecting parts of the interaction-vetoing beam counter BC3 (see sect. 2.6) and interfaces to the adjacent parts of the spectrometer. It is housed within a hollow cylindrical recess into
RICH-1 which is lined by a cylindrical tungsten mantle of 20 mm thickness which shields the UV-detectors from heavily ionising particles emerging in backward direction from the target. The recess is hermetically closed towards the radiator of RICH-1 by a double-walled aluminised mylar window (2 x 50 $`\mu `$m), the intermediate volume being ventilated with nitrogen. The SiDCโs and a thin mirror for BC3 are mounted within a double-walled tube of 160 mm diameter, made of aluminium and carbon fibre which is water cooled and provides a laminar flow of dry nitrogen for cooling the SiDCโs and the front-end chips mounted on the same mother board. A light guide transfers the Cherenkov light received from a thin mirror.
The beam enters the target area via an evacuated Al tube that reaches until a few millimetres short of the segmented target and is sealed by a thin mylar window .
The segmented Au target consists of 8 discs of 25$`\mu `$m thickness and 600$`\mu `$m diameter spaced uniformly by 3.2 mm along the beam. The total target thickness of 200$`\mu `$m Au corresponds to an interaction length of $`\lambda /\lambda _I=0.83\%`$ for Pb-Au collisions. The segmentation assures that only a fraction of the total radiation length is effectively seen by photons and electrons propagating into the spectrometer acceptance. Under worst conditions of full illumination, there is only a chance of about 20$`\%`$ that the next downstream disc is within the acceptance. The effective radiation length is $`X/X_{}`$ 0.55$`\%`$, compared to $`3\%`$ for an unsegmented target of identical interaction length.
The Au discs are supported by 2.5 $`\mu `$m thick mylar foils, and the entire assembly is contained in a thin-walled (0.5 mm) carbon fibre tube of 60 mm diameter; the Au discs are accurately aligned to the axis of the tube. The Au discs contain 90$`\%`$ of the beam over the full length of the target once collimators and magnet settings of the beam line have been optimised. The beam diameter was measured to have a Gaussian envelope of $`\sigma `$= 220 $`\mu `$m.
### 2.2 The silicon-drift telescope
The SiDCโs fulfil several purposes in the overall concept: (i) to locate the interaction vertex within the segmented target, (ii) to maintain a precise two-point charged-particle tracking free of ambiguities, improving thereby the momentum resolution of the spectrometer, (iii) to assist in the ring pattern recognition of RICH-1, and (iv) to reject conversions and Dalitz pairs which are not resolved in the RICHes, by detecting a double pulse-height signal or two close tracks.
The SiDC telescope made up of two closely spaced cylindrical SiDCโs was the major element in upgrading CERES for the Pb-beam experiments cogne95 . CERES is the first experiment which successfully implemented chen92 this detector concept gatti847 , and it did so from the start, however, with only a single detector chen93 ; fasch93 . This allowed to check that the electron pair originates from a common vertex, but was of course insufficient for full tracking.
Since cylindrical SiDCโs seem not very well known, we insert here a short description. The active area is practically the full area of a 3-inch-diameter wafer 280 $`\mu `$m thick which has a central hole of about 6 mm diameter for the passage of the beam. The signal electrons generated by ionisation of charged particles traversing the wafer drift radially outward to an array of 360 anodes located at the periphery of the detector. The radial coordinate $`r`$ of the point where a charged particle crossed the detector plane is measured by the drift time $`t`$ of the electron cloud. The charge sharing between neighbouring anodes measures the azimuthal coordinate $`\varphi `$. The pair of coordinates $`(r,\varphi )`$ is provided for each crossing charged particle for events with a total charged multiplicity of several hundred. The longest drift distance is about 3 cm. The nominal value of the drift field of 500 V/cm results in a maximum drift time of about 4 $`\mu `$s. The drift field is provided by means of 240 concentric $`p^+`$ electrodes with 130 $`\mu `$m pitch on both sides of the detector, suitably biased by an implanted voltage divider. A major innovation in the design was a โsink anodeโ providing a path for the leakage current generated at the Si-SiO<sub>2</sub> interface away from the signal anode, i.e. without contributing to the anode leakage current.
The ideal circular electrode shape could not be designed in early 1990 due to limitations in the software controlling mask production. Electrodes were shaped instead as regular polygons of 120 sides. We were surprised to see that the distribution of hits displayed peaks at every third anode fasch96 . The โefficientโ anodes were those located in the central part of each 3 triangle forming the polygon.
The telescope implemented in 1995 consisted of a 3-inch detector followed by a 4-inch detector, the latter of the novel AZTEC design holl96 which eliminated the focusing problem. The spacing of the two detectors was 14.3 (15.0) mm and the distance from the target centre to SiDC-1 amounted to 98.5 (110) mm in 1995 (1996).
In production of the 3-inch detector<sup>2</sup><sup>2</sup>2produced by SINTEF, 0134 Oslo 3, Norway, the front and back lithographies were rotated with respect to each other by 1.5 to reduce the non-radial components in the drift field. This trick reduced to negligible levels the focusing effect of the central anodes that had been a severe obstacle for reaching the design azimuthal hit resolution fasch96 .
In 1996, both detectors were of the 4-inch type<sup>3</sup><sup>3</sup>3produced by EURISYS Mรฉsures, F-67380 Lingolsheim, France. The larger anode radius of 42 mm allowed to increase the distance to the target for lower hit occupancy, especially at small radii. The sensitive area of the detectors is increased from 32 cm<sup>2</sup> for the 3-inch devices to 55 cm<sup>2</sup> for the 4-inch detectors. The โfield cageโ polygons consist of 277 concentric $`p^+`$ electrodes each having 360 instead of 120 sides before, and the respective $`p^+`$-implantations on the two wafer sides are rotated by 0.5 with respect to each other for near perfect radial field geometry. Another novel property of the AZTEC detectors is the interlaced anode structure: each one of the 360 anodes is subdivided into 5 segments, two of which (extending over 16$`\%`$ of the anode pitch of 1 degree) are interlaced to the closest of the neighbouring anodes to enforce charge sharing. This provides a more accurate azimuthal position measurement when calculating the centre of gravity of the distribution. Compared to the โ95 runtime, the new design improves the azimuthal resolution from 2.5 to 1 mrad. The resolution in radial direction is 30 $`\mu `$m both for the 95 and 96 set-ups.
Charge signals from 3-inch detectors are amplified by 32-channel front-end OLA chips placed on the detector motherboards which were developed to test ALICE silicon-drift prototype detectors dabrowski93 and were produced in a custom bipolar process<sup>4</sup><sup>4</sup>4Owned by Tektronix at the time of production; since then Maxim Integrated Products. They consist of a charge-sensitive preamplifier, a quasi-Gaussian shaper<sup>5</sup><sup>5</sup>5The time constant $`\tau `$= 38 ns, about twice the design value, deteriorated the potential double-pulse resolution, but avoided a large ballistic deficit over the full drift. and a symmetrical line driver. The rather high gain (30 mV/fC) gave rise to wild collective oscillations. Stable operation was achieved only with the introduction of 20 $`\mathrm{\Omega }`$ damping resistors connected in series into the 50 $`\mathrm{\Omega }`$-terminated output lines.
For the readout of the 4-inch detectors, new 16-channel front-end chips had been designed along a CMOS concept gramegna1997 which incorporated bipolar drivers and were well adjusted in shaping time (37 ns), gain (9 mV/fC), dynamic range (5 $`mips`$<sup>6</sup><sup>6</sup>6minimum ionising particles), and low equivalent noise charge (140 e<sup>-</sup>) to meet our requirements.<sup>7</sup><sup>7</sup>7produced by AMS in 0.8$`\mu `$m biCMOS technology.
An outside buffer stage transmits the bipolar signals over 40 m flat cables to the FADCโs (flash analog to digital converters)<sup>8</sup><sup>8</sup>8Series DL300 of Fa. B. Struck, Tangstedt near Hamburg which sample the data with 50 MHz and store it 256 bytes deep. This corresponds to a drift time range of 5.12 $`\mu `$s. Digitisation is 6 bit with non-linear characteristics.<sup>9</sup><sup>9</sup>9$`\mathrm{Channel}.\mathrm{No}.(0\mathrm{}63)=256\mathrm{U}_{\mathrm{in}}/(0.2+3\mathrm{U}_{\mathrm{in}})`$, input voltage $`\mathrm{U}_{\mathrm{in}}`$ in Volt. Data are continuously sampled until interrupted by an external โstopโ signal so that the channel memory contains always the last 5.12 $`\mu `$s of data.
Since the trigger signal and clock are asynchronous, there is a random phase difference producing a time jitter of 20 ns/$`\sqrt{12}`$ rms. It is measured with a TDC (time-to-digital converter), and correction is done off-line.
Each of the four FADC crates per detector houses a SIM (scanner interface module) which scans the data after the trigger was received for contents above a predefined threshold (readout threshold). Readout is activated whenever the threshold is surpassed in two successive time bins, and stopped, if contents in two successive time bins fall below it. The contents of five preceding channels are also readout (pre-samples) for off-line reconstruction of the baseline.
### 2.3 The RICH detectors
The radiators are filled with methane at atmospheric pressure. The high Cherenkov threshold of $`\gamma _{thr}`$ 32 <sup>10</sup><sup>10</sup>10different in RICH-1 and RICH-2 by about one unit. ensures that more than 95$`\%`$ of all charged particles pass without creating Cherenkov light (โhadron blind trackingโ). The Cherenkov light is reflected backward onto 2-dimensionally position-sensitive gas detectors which are separated from the radiator volume by UV-transparent windows. By their upstream position with respect to the target, the UV-detectors are not exposed to the huge forward flux of charged particles. The price to be payed for this geometry is the limited acceptance in polar angle $`\mathrm{\Theta }`$, indicated by the lines in the upper part of Fig 2.
High-energy electrons produce Cherenkov rings with asymptotic radius, R= 1/$`\gamma _{thr}`$ 30 mrad. The difference in radiator lengths (86 cm and 175 cm for RICH-1 and RICH-2, respectively) is partially compensated by better UV transmission in RICH-1 (CaF<sub>2</sub> window) compared to RICH-2 (quartz window), so that the asymptotic number of photons per ring, 10.8 and 11.5, for RICH-1 and RICH-2, respectively, come out rather similar<sup>11</sup><sup>11</sup>11The numbers of Ref. rich\_set measured with ethane are increased by one photo-electron due to the larger bandwidth in methane.. By the same reason, photon detection in RICH-1 reaches farther into the UV.
In both RICHes spherical mirrors focus the Cherenkov photons radiated from a straight trajectory back onto a ring image in the focal plane of the UV detectors. As the mirror in RICH-1 is traversed by all electrons before the second ring image for momentum measurement is taken in RICH-2, there are stringent physics reasons to keep the radiation length as low as possible: besides reducing the number of external conversions, it is the multiple scattering of low-momentum electrons which reduces the detection efficiency for soft pairs and deteriorates the momentum resolution. The mirror of RICH-1 therefore is made very thin (1.1 mm) so that it adds only 0.4$`\%`$ of a radiation length. It is based on a laminated carbon fibre structure which defines the spherical geometry<sup>12</sup><sup>12</sup>12 manufactured by MAN Technologie AG. An evaporated coating of aluminium protected by magnesium fluoride achieved persistent UV reflectivity of 80$`\%`$ at 300 nm.
The UV-detectors consist of a conversion space followed by two parallel-plate avalanche stages and a multi-wire proportional chamber. The originally planned mode of running only with parallel-plate amplification was abandoned in favour of an added multi-wire stage, following a painful learning process on spark break down problems in pure parallel-plate schemes spark94 .
The operating gas is He + 6$`\%`$ CH<sub>4</sub> at atmospheric pressure + TMAE-saturated<sup>13</sup><sup>13</sup>13Tetrakis-di-Methyl-Amino-Ethylen vapour at 40C as photon converter. The use of TMAE demands that the UV detectors be kept hot to avoid condensation. To avoid temperature gradients across the delicate UV-transparent windows separating the detectors from the radiators, the entire spectrometer is kept hot at about 50 C. The UV detectors operate at a total gain of about 2$``$10<sup>5</sup> for high photon detection efficiency ($`85\%`$). The ion clouds produced in the last wire amplification stage induce signals on the pads. The latter form a grid of pitch 2.74 mm and 7.62 mm, and the resulting total number of pads is 53,800 and 48,400 in RICH-1 and RICH-2, respectively pad\_read .
The UV detectors have been operated without opening since 1991. During the 1995 run, UV-1 degraded in performance. The detector could not be operated at the desired gain of 2$`\times `$10<sup>5</sup> without an excessive spark rate. Early in 1996, the UV-1 detector was opened and all mesh electrodes, in particular the cathodes and the multi-wire plane showed some kind of deposit. Most mesh electrodes were exchanged and the wire anode subjected to ultrasonic cleaning. The refurbished detector performed very well during the 1996 run socol-phd .
### 2.4 Deflection in the magnetic field
The magnetic field for momentum analysis is generated by two super-conducting solenoids carrying currents in opposite sense. Charged particles experience an azimuthal deflection between the two RICHes which is inversely proportional to the momentum,
$$\mathrm{\Delta }\varphi =\frac{\varphi _0}{p}\left(\frac{\mathrm{mrad}}{\mathrm{GeV}/c}\right),$$
(2)
and the sense of which, for fixed polarity of the field, defines the charge sign. The constant is $`\varphi _{}`$= 146 mrad GeV/$`c`$. Between SiDC and the Pad Chamber the deflection is only 66$`\%`$ of this value, $`\varphi _{}`$= 96 mrad GeV/$`c`$.<sup>14</sup><sup>14</sup>14The RICHes measure the change in local angle, while deflection in the Pad Chamber is derived from the displacement relative to the distance from the vertex. To first order, the polar angle is not affected. Particles deflected in azimuth by $`\mathrm{\Delta }\varphi `$ encounter a small second-order deflection towards the beam axis which amounts to
$$\mathrm{\Delta }\theta =78(\mathrm{\Delta }\varphi )^2\left(\frac{\mathrm{mrad}}{\mathrm{rad}^2}\right).$$
(3)
Two sets of warm correction coils are tuned to achieve a field-free radiator in RICH-1 and to align the field lines in the radiator of RICH-2 parallel to the particle trajectories from the target. This way straight trajectories inside both radiators are achieved. Moreover, the absence of deflection in the first RICH detector allows to identify conversion and Dalitz pairs by their small opening angles.
### 2.5 The Pad Chamber
The last tracking detector is a multi-wire proportional chamber located closely behind the mirror of RICH-2, at a distance of about 3.3 m from the target. It has an inner and outer radius of 42 cm and 85 cm, respectively, and is free of radial spokes. The Pad Chamber covers the fiducial pseudo-rapidity interval $`2.05<\eta <2.65`$ of the CERES spectrometer. It was added before the run period in 1995 as an external tracking device behind RICH-2 to assist the ring pattern recognition and reduce the fake-ring background in the high-multiplicity environment of Pb-Au collisions. By providing an absolute reference for the silicon and the RICH detectors, the Pad Chamber, with an angular resolution of about 0.6 mrad (in $`\theta `$), proved a powerful tool in the geometrical inter-calibration of the detectors which helped to improve momentum resolution.
The Pad Chamber is operated with a 90/10$`\%`$ Ar/CO<sub>2</sub> mixture. The multi-wire anode is at equal distance (5 mm) to the upstream mesh cathode and the downstream pad cathode. Only about half the ionisation charges are collected during the $`2\mu `$s charge integration time of the pad readout electronics. The electronic avalanche produced by a charged particle traversing the Pad Chamber induces a signal in some of the $``$ 29000 pads of the pad cathode. The pad size is the same as in RICH-2, and the pad readout electronics was adopted from RICH-2.
### 2.6 The trigger
A system of beam counters (BC) has been specifically developed to meet the requirements of minimal mass exposure in the beam and target region and sufficient radiation hardness volodya97 . The trigger system is based on three small Cherenkov beam counters operating in air, one (BC1) about 60 m upstream, monitoring incident beam particles, and one (BC2) about 6 m downstream of the target detecting ions passing through the spectrometer. The
third Cherenkov counter (BC3) registers each intact Pb ion downstream of the last target in order to veto the interaction trigger. To this purpose, Cherenkov photons (2700 per cm air) emitted in a narrow forward cone are reflected away from the axis onto a UV light guide by means of a tiny (6 mm diameter) aluminised mylar mirror on axis about 6 cm downstream of the target. The Cherenkov light is fed into a photo-multiplier just outside the spectrometer (Fig 3).
A plastic scintillator in front of the spectrometer (VC) is used to veto upstream interactions. The interaction trigger is defined as the logical AND of BC1 and the veto of BC3 and VC, INT= BC1$`\overline{\mathrm{VC}}\overline{\mathrm{BC3}}`$. Centrality is selected with the multiplicity detector (MD), an array of 24 plastic scintillator paddles downstream of the RICH detectors at $`\eta =`$2.9-4.7, the light output of which serves as a measure of the number of ionising particles that have passed. The centrality trigger is defined as INT$``$MD, and a hardware threshold is set at 100 mips. The accuracy of the trigger threshold and its stability over time is limited, mostly due to gain variations in the photo-multiplier tubes. In the off-line analysis, a precise multiplicity measurement is provided by the two SiDCโs. The trigger selection corresponds roughly to the top 30$`\%`$ of the geometrical cross section claudia . A more precise calibration will be presented in sect. 3.8.
### 2.7 Data acquisition
The on-line response of the detectors is shown in Fig. 4 for a semi-central 158 GeV/n Pb-Au event recorded during data taking in 1996.
The large amount of information per event contained in five highly granular detectors is collected by a fast data acquisition system. Data reduction is performed already at the hardware level when individual detectors are readout with zero suppression and pedestal subtraction. The data from the detectors are split into several separate readout chains which are processed in parallel for higher readout speed and transferred into special Memory modules where the data is compressed using Huffmann coding.
A set of two (in 1995, three in 1996) CPU modules in VME technology is used to collect the data from the Memory modules for each event during the burst period (4.8 s) in a round-robin mode and writing it to tape during the intervals between bursts (14 s). As the write speed of the Digital Audio Tape (DAT) drives used is rather low, each CPU has three drives connected to it, allowing for an effective aggregated write speed of about 1.5 MByte/s per CPU (in 1995, about 8 MB/s in 1996). This allowed to record on average 550 events/burst of 40-45 kByte each in 1995; the upgraded DAQ together with optimised detector settings and therefore smaller event sizes (30-35 kB) recorded about 1000 events/burst in 1996.
### 2.8 Data taking in 1995 and 1996
CERES/NA45 had data taking runs of 9 days in fall 1995 and of 27 days in fall 1996 at the CERN SPS with a 158 GeV/n Pb beam on Au targets. The average beam intensity in both years was about 1$`\times `$ 10<sup>6</sup> ions per burst of duration 4.8 s.
Because of the small target and beam dimensions, a readjustment of the beam position was one of the regular shift duties. To keep the spectrometer efficiency at the designed level, the gains in the UV detectors were monitored continuously and held within limits of about 30$`\%`$ at 2$`10^5`$ by adjusting high voltage upon changes of atmospheric pressure.
Collision events were selected with the interaction trigger threshold set on 100 mips equivalent in the multiplicity array. The trigger contained an admixture of downstream interactions on the level of 15% of the target interactions. These were discarded in the off-line analysis on account of the silicon-drift track multiplicity. The latter served for more accurate centrality definition and also revealed some difference in the effective calibration of the multiplicity array between the two runs.
In 1995 8.5 million events were collected with average multiplicity $`dN_{ch}/d\eta `$= 220 corresponding to the top 33$`\%`$ of the geometrical cross section. In 1996, the total sample was 42 million events of average charged particle multiplicity 250, or 26$`\%`$ of the top geometrical cross section. The multiplicity refers to the number of tracks in the SiDCโs and is averaged over the pseudo-rapidity range $`\eta `$= 2-3. In the middle of the runs the polarity of the magnetic field was switched.
### 2.9 Instrumental means of coping with background
We shortly review here the instrumental means by which CERES recovers a weak signal of low-mass electron pairs from high levels of combinatorial background.
Approximate โhadron blindnessโ is achieved by using two Ring Imaging Cherenkov (RICH) detectors with a high threshold $`\gamma _{th}`$ 32. While electrons with momenta above 16 MeV/$`c`$ produce Cherenkov light, pions overcome the threshold only at 4.5 GeV/$`c`$. More than 95% of all charged hadrons pass without producing Cherenkov light.
The radiation length within the spectrometer acceptance has been kept at $`X/X_0`$ $``$ 1 %. This is the level where the number of conversions is about equal to the number of Dalitz pairs. It is the result of persistent efforts in the design of all spectrometer components to reduce the detector materials, among which the thin mirror of RICH-1 and the segmented target are the most important.
Our physics sample are electron pairs with mass above 200 MeV/$`c^2`$. Below 200 MeV/$`c^2`$, photon conversions and $`\pi ^0`$ Dalitz decays shoot up in yield which diminishes the sensitivity to interesting physics. In pursuit of the goal to recognise soft pairs of conversions and $`\pi _0`$-Dalitz pairs with highest possible efficiency, the CERES spectrometer provides two powerful handles: the SiDC doublet detects close tracks by double-dE/dx response in pulse height, or by resolved close hits. The fact that the radiator of RICH-1 is free of magnetic field allows to see electron pairs of small opening angle undeflected, i.e. as close, resolved Cherenkov rings, or rings with a larger number of photons when the two tracks are separated by less than $``$8 mrad. Soft electron tracks from conversions and $`\pi ^{}`$-Dalitz pairs are strongly deflected by the magnetic field between the two RICHes, electrons and positrons in opposite sense. By setting an upper limit to the azimuthal deflection between RICH-1 and RICH-2, or between SiDC and Pad Chamber, a cut on track $`p_t`$ is implemented which is one of the most effective measures to reject background.
To achieve a high efficiency for reconstruction of soft pairs, all detectors have full azimuthal coverage and the vetoing detectors have a slightly larger (minimum 1.9 $`<`$ $`\eta `$ $`<`$ 2.8) fiducial rapidity acceptance than the detectors after the magnetic field (2.15 $`<`$ $`\eta `$ $`<`$ 2.6). The more subtle details of the rejection strategy are discussed in sect. 3.
## 3 Data analysis
### 3.1 Overview
In this overview we sketch the strategy of the data analysis up to the pairing level when identified electron tracks in a given event are selected for combination to pairs. The presentation is mainly based on data and only rarely refers to Monte-Carlo simulations. The latter are, however, implicit in the choice of various quality and rejection cuts and will be treated in sect. 4.
#### 3.1.1 The analysis strategy
The invariant mass squared of the pair is given by the squared sum of the electron 4-momenta,
$$m_{ee}^2c^2=(๐ฉ_{๐^+}+๐ฉ_e^{})^2=2p_{e^+}p_e^{}(1cos\mathrm{\Theta }_{ee}).$$
(4)
For the standard single-electron cut $`p_t`$ 200 MeV/$`c`$ used in the data analysis, the dynamic range of the electron momenta is somewhat restricted so that the dynamic range of the invariant mass is largely determined by that of the laboratory opening angle $`\mathrm{\Theta }_{ee}`$ between the electron tracks.
Signal electron pairs with $`m`$ 200 MeV/$`c^2`$ have opening angles considerably larger than the asymptotic Cherenkov ring radius of 30 mrad; the massive pairs from $`\rho `$, $`\omega `$, and $`\varphi `$ decays are opened about ten times wider (Fig. 5, upper panel).
Our operational definition of signal or open pairs includes an opening angle cut $`\mathrm{\Theta }_{ee}`$ 35 mrad.<sup>15</sup><sup>15</sup>15It reduces the number of Dalitz pairs. As an important side effect, the cut enforces more uniform track distributions in polar angle of pairs from different sources. For $`\eta `$Dalitz decays, the mean opening angle is about 60 mrad. Conversion and $`\pi ^{}`$-Dalitz pairs have small masses<sup>16</sup><sup>16</sup>16Photon conversion pairs are nearly massless but may acquire a small apparent mass by multiple scattering. and average opening angles below 2 mrad and 20 mrad (rms), respectively. The sample of pairs with masses below
200 MeV/$`c^2`$ and $`\mathrm{\Theta }_{ee}`$ 35 mrad will be referred to as the Dalitz sample and is also used for checks on reconstruction efficiency and absolute yields.
The $`p_t`$ distributions of electron tracks from conversions and $`\pi ^{}`$-Dalitz pairs are steeper than those of open signal pairs. This feature provides the only rejection handle at the track level, albeit a very powerful one: the 200 MeV/$`c`$ cut on track $`p_t`$ reduces close-pair tracks much stronger than signal tracks as seen in Fig. 5, lower panel.
The most severe problem of the experiment is the enormous combinatorial background. We do not know which electrons belong to a pair and therefore we accept combinations of all tracks that qualify. However, when we find a pair with $`m<`$ 200 MeV/$`c`$, its tracks are excluded from further pairing. Because the S/B ratio for these pairs is usually very good, we can declare a fully reconstructed Dalitz or conversion pair with good confidence.
The background arises whenever low-mass pairs are only partially reconstructed and the remaining tracks are combined, as visualised in Fig. 6. Clearly, combinatorial pairs cannot be distinguished from genuine signal pairs and contribute
to the entire mass range of interest, exceeding the signal by three orders of magnitude if all low-mass pairs would contribute. The final pair sample at masses above 200 MeV/$`c^2`$ is still only about 10$`\%`$ of the residual level of combinatorial background pairs. This is why already small inefficiencies in reconstruction of soft pairs, acceptance losses, etc., give rise to large relative levels of combinatorial background. Multiple scattering of conversions and $`\pi ^{}`$-Dalitz tracks is a further important source of losses.
A mild $`p_t`$ 50 MeV/$`c`$ cut is applied during the production stage (see below) to limit the search area in azimuth. Many of the conversion and Dalitz pairs with only one leg above 200 MeV/$`c`$ can be reconstructed and later rejected this way. Such pairs are ten times more numerous than pairs with both electrons above 200 MeV/$`c`$, so that many stiff electron tracks are taken out before entering the pairing stage. Applying the strong $`p_t`$ cut before the filter would have kept those tracks in the combinatorics. Tracks attributed to pairs with opening angles $`\mathrm{\Theta }_{ee}`$ 35 mrad - the logical complement of the opening-angle cut for signal pairs - are marked to be excluded from further pairing.
The deflection of electrons and positrons by the magnetic field between the RICHes provides the unique search pattern of V-tracks, i.e. one ring in RICH-1, possibly somewhat blurred as it contains UV photons from two close tracks, and two separated rings in RICH-2 at about the same polar angle (see also Fig. 6).
Rejection of combinatorial background is optimised by tuning various cuts. Since high rejection power and high signal efficiency are competing requirements, an appropriate measure of signal quality is required. The optimisation has to be kept rigorously free of bias. A critical discussion of this important issue is given in sects. 3.4 and 3.5.
#### 3.1.2 The analysis stages
Without a higher-level trigger in the CERES Pb-beam experiments, the actual data volume is much larger than that of interesting events, and a primary data reduction, the โproductionโ stage, is called for. Implementation of an effective production filter requires an accurate geometrical inter-calibration of all detectors. The availability of the Pad Chamber since 1995 allowed precise local tuning of the entire spectrometer using samples of high-$`p_t`$ pions ceretto-phd .
During production, the full analysis chain for electron track reconstruction is at work, albeit under loose quality criteria. Events which contain at least two electron tracks with $`p_t`$ 50 MeV/$`c`$ are stored in a database for further processing. As millions of events have to be processed with sophisticated pattern recognition algorithms, the production is time consuming. The 1995 production on the CS2 parallel computer at the CERN CN division with 32 SUN-SPARC2 processors took about 10 weeks; the 1996 data were preprocessed on a PC farm in several turns with readjusted production filters during 4 months. The final data analysis mainly deals with the optimisation of the pair sample.
### 3.2 Reconstruction of electron tracks
The reconstruction of electron tracks in the present analysis takes full advantage of the external tracking detectors, the doublet of SiDCโs before the RICH spectrometer and the Pad Chamber after it.
#### 3.2.1 Coordinate systems
Raw-data detector coordinates
In RICH detectors and in the Pad Chamber, hits are encoded as pad amplitudes in a two-dimensional mesh of $`(x,y)`$ coordinates of the โpad planeโ. The natural unit is the pad size of 2.74 mm in RICH-1, and 7.62 mm in RICH-2 and the Pad Chamber, which is used up to the reconstruction of rings and ring centres in the RICHes, and of hits in the Pad Chamber.
For the SiDCโs, the symmetry of the radial drift field and the circular ring of anodes is maintained at the raw data level: the intrinsic coordinates are anode numbers (0-359) for the azimuth location and time bins (0-255) of 20 ns for the drift time.
Local detector coordinates
RICH detectors measure angles of particle trajectories that connect centres of Cherenkov rings with the vertex point, the units are $`\mathrm{\Delta }\theta =\mathrm{\Delta }s/f`$, expressed by pad size $`\mathrm{\Delta }s`$ and focal length $`f`$; these are $`\mathrm{\Delta }\theta `$= 2.18 mrad and 1.82 mrad in RICH-1 and RICH-2, respectively. The local coordinates in paraxial approximation are expressed by the tangents of the polar angle $`\theta `$. Because of spherical aberration by the mirrors, the expression using $`\theta `$ instead turns out to be a much better approximation,
$$x=f\theta \mathrm{cos}\varphi +x_{},y=f\theta \mathrm{sin}\varphi +y_{},$$
(5)
where $`x_{},y_{}`$ are the coordinates of the origin of the pad plane. In the off-line analysis, the angles $`\theta `$ and $`\varphi `$ are evaluated by ray tracing using look-up tables to correct for local modulations in focal length.
Hits in the silicon-drift detectors are given by their radial position $`r`$ (referring to the centre of the wafer) and azimuth angle $`\varphi `$ which is defined as in the global laboratory system. The transformation from drift time to radius requires knowledge of the electron drift velocity with possible spatial as well as temporal changes.
Laboratory coordinates
Once track segments of several detectors are to be joined, each detector is put into a three-dimensional global laboratory system with its own geometric calibration parameters, i.e. small $`(x,y)`$-shifts of detector axes away from the optical axis, rotations around $`z`$, tilts, etc. This global laboratory system uses left-handed Cartesian coordinates with the $`z`$-axis along the beam and the $`x`$\- and $`y`$-axes pointing to the right and upward, respectively, when looking with the beam. The origin is in the centre of SiDC-1. As we do not measure space points within the short magnetic field, it is convenient to work with straight trajectories all along from vertex to the Pad Chamber, and store the particleโs momentum and charge sign derived from magnitude and sign of the azimuthal deflection.
Event coordinates
Eventually, in the global event coordinate system, the event vertex $`(x_V,y_V,z_V)`$ is taken as the origin, and tracks are described by polar coordinates $`\theta `$, $`\varphi `$ at the vertex.
#### 3.2.2 Hit finding in the SiDCโs
The 360 readout channels of anodes and the 256 time slices, or time bins, sampled by the FADC, span a matrix each cell of which is assigned a 6-bit non-linear raw-data amplitude. The amplitudes are linearised and pedestals are subtracted.
The data field is then searched for contiguous regions of cells with nonzero amplitude. Each cell is attributed to one such cluster. All cells of fixed anode number within a cluster form a time sequence of amplitudes called a pulse. The signal of a minimum-ionising particle produces pulses of typically 5 time bins and is spread over 2.2 anodes on average. Pulses of less than 3 cells above a hardware threshold are discarded. Examples of such pulses on a few neighbouring anodes are displayed in Fig. 7. Due to complementary signal transmission, the pulses are free of pickup which plagued previous runs.
The centres of gravity of the pulses are calculated by Gaussian regression, or by Gauss fits taking the known, drift time dependent widths from a table. After time $`t`$, the drifting electron cloud arriving at the anodes has developed a time-spread due to diffusion of $`\sigma _t^{in}=\sqrt{2Dt/v_{drift}}`$. Here $`D`$ 35$`\times `$ 10<sup>-4</sup> mm$`{}_{}{}^{2}/\mu `$s is the electron diffusion constant in silicon and $`v_{drift}`$ denotes the drift velocity which varied in the range 6.0 - 8.5 mm/$`\mu `$s depending on the voltage setting. The charge pulse is folded with the quasi-Gaussian response of the shaper, $`\sigma ^{shaper}`$, so that the time spreads add in quadrature. The shaping introduces a ballistic deficit in amplitude $`\delta =\sigma ^{shaper}/\sigma _t^{in}`$ which is corrected for. Pulse heights saturated in the peak cells are approximately reconstructed.
The stop-pulse correction mentioned in sect. 2.2 removes the random phase jitter. The conversion from drift time to drift distance is presented in sect. 3.2.
Pulse trains on neighbouring anodes of the same cluster are merged into a hit if centres of gravity differ by less than one time bin. The hit coordinate in azimuthal direction is calculated as the centre of gravity of the contributing pulses, weighted by their peak amplitudes. The hit amplitude is the sum of the pulse amplitudes.
#### 3.2.3 Hit finding in RICHes and Pad Chamber
Signals of the RICH detectors and the Pad Chamber are read out from the checkerboard-like arrangement of pads which receive the charge amplified in multi-wire proportional chambers. The same hit-reconstruction algorithm is used for these detectors.
Adjacent pads with amplitudes above a readout threshold are connected to *clusters* some of which are caused by background and electronic defects. Typical background clusters, like long and thin stripes from ionising particles on oblique trajectories, or clusters with many pads in saturation, are removed with the help of various cleaning algorithms. Remaining clusters are split into regions containing one local maximum and are identified as UV-photon hits in the RICHes or charged-particle hits in the Pad Chamber. Hit centres are calculated as the centres of gravity of the contributing pads.
#### 3.2.4 Ring candidates in the RICH detectors
RICH detectors require an additional step of pattern recognition to search for rings with asymptotic radius produced by electrons robust96 . A typical RICH-1 ring is shown in Fig. 8. The ring search is done using the Hough transformation hough62 from the data field of the pad plane into the โHough arrayโ (or โparameter spaceโ) which has the same dimension. A given cell is the โimageโ of all data points (hits) on a circle around that cell, and its โHough amplitudeโ is the sum of the data amplitudes on that circle. A ring candidate shows up as a peak in the โHough arrayโ which is the higher the more photon hits, or illuminated pads, lie on the ring. In practice, a โdigitalโ Hough transformation is used: all pads with signal amplitudes above a defined threshold enter with unit weight into the sum amplitudes, irrespective of their amplitudes.
The event display in Fig. 8 demonstrates that rings can be formed also by random arrangement of single photon hits, some of them produced by pions near the Cherenkov threshold. The first step in ring pattern recognition is a linear Hough transformation of the pad plane onto the parameter plane. Besides the real maximum in the Hough array, there are other local maxima connected to fake ring centres. As a counter measure, a second, non-linear Hough transformation is performed which assigns a relative weight to each cell in the Hough array such that each hit counts most for its most favourable ring-centre. This way fake rings are suppressed, as can be seen from Fig. 8, by a larger gap that has developed between the amplitudes of the real and the fake rings. The parameter to select real rings at this stage of the analysis is the amplitude after the second Hough transformation.
All surviving ring candidates are assigned their final centre coordinates by a robust estimation which is based on iterative re-weighted least-squares fits of circles with asymptotic radius to the hits. A second fit with variable ring radius should eliminate charged pions which have a non-asymptotic radius. The function minimised is a modified $`\chi ^2`$ where the fit-potential varies in a Gaussian way (instead of quadratically) with the distance to the minimum. For an extensive description of the fitting algorithm see Refs. ullrich-phd ; ring\_fit . Other parameters besides the amplitude in the Hough array, like the number of hits and the spread of the hits around a perfect circle, can be used for fake-ring suppression.
Ring reconstruction efficiency is determined by Monte-Carlo simulations with a cut on the number of photon hits, assumed to be Poisson-distributed. At the stage described up to here, ring reconstruction efficiencies of about 85$`\%`$ are still confronted with many fake rings, about 10 to 20 for one reconstructed ring that is real. It was in anticipation of such alarming majority of fake rings that the collaboration decided in 1994 to implement full external tracking pb-proposal94 .
#### 3.2.5 Calibration of the SiDC telescope
To first approximation, the relation between the drift distance and drift time is linear. From the known radial extension $`\mathrm{\Delta }R`$ of the active area and the total drift time $`\mathrm{\Delta }t`$, the drift velocity can be calculated as
$$v_{drift}=\mathrm{\Delta }R/\mathrm{\Delta }t,\mathrm{\Delta }R=R_{max}R_{min},$$
(6)
where $`R_{min}`$ is the inner edge of the active area and $`R_{max}`$ corresponds to the anode radius.<sup>17</sup><sup>17</sup>17The anode radius is 32 mm and 42 mm for the 3-inch and 4-inch detectors, respectively, $`R_{min}`$ is typically 10 mm and the total drift time 3 to 5 $`\mu `$s.. The corresponding drift time is
$$\mathrm{\Delta }t=t_{max}t_{min},$$
(7)
where
$`t_{min}`$ is the time corresponding to the shortest drift path (ionisation directly under the anode), and $`t_{max}`$ the drift time of particles starting at $`R_{min}`$. A typical hit distribution as a function of drift time is displayed in Fig. 9. By fitting the edges of the drift-time spectrum, the values of $`t_{min}`$ and $`t_{max}`$ are determined.
Following such preliminary calibration, the geometrical alignment of both detectors and the determination of the interaction vertex is performed. Other corrections, such as the stop-pulse correction, corrections on drift-velocity variations due to temperature changes, etc., are done while maintaining the reconstructed vertex position.
#### 3.2.6 Vertex reconstruction
Once hits are reconstructed, the information of hit positions in the laboratory coordinate system is used to combine hits to tracks and find the vertex position to which almost all tracks of a given event point to. The procedure to minimise the quadratic sum of hit mismatches between the two detectors and its iteration is extremely time consuming. We used therefore a robust vertex fitting approach robust97 : all hits in SiDC-1 and SiDC-2 (typically more than 100) are combined to straight track segments and a weighted sum of their projected distances to the assumed vertex position is calculated. In the next iteration, this centre of gravity becomes a new starting value for the vertex position and each track segment gets a new weight according to its deviation from the mean value in the step before. After the position of the vertex is determined, its $`z`$-position is redefined as the exact position of the closest vertex disc.
Figure 10 displays the density of reconstructed vertex positions along the beam. The peaks reveal the positions of the eight discs of the segmented target assembly with a spacing of 3.2 mm. On average, the reconstruction was done with 160 charged particles per event. The data shown in Fig. 10 was accumulated over $`2.6\times `$10<sup>5</sup> Pb-Au events, or six hours, and demonstrates a certain long-term stability. The resolution<sup>18</sup><sup>18</sup>18i.e. the standard deviation in the mean $`z`$ position of $`\sigma _z=250\mu `$m is sufficient to identify the correct target disc without ambiguity<sup>19</sup><sup>19</sup>19due to the unknown location of the vertex inside the 25 $`\mu `$m thick foil, there is an rms error of 25/$`\sqrt{12}`$, about 7 $`\mu `$m. .
More critical for tracking accuracy is the precision with which the vertex can be localised in the plane transverse to the beam, as there are no fixed points. To measure it we used stiff pion tracks with $`p_t`$ 1.2 GeV/$`c`$ to minimise multiple scattering. The scatter plot in Fig. 10 accumulates over many events the distance in $`x`$ and $`y`$ between the actual event vertex (determined from all tracks) and the point where a given pion track intersects the respective target disc. Projecting on the axes, we obtain the 1-dim distribution shown to the far right; a lateral resolution of 28 $`\mu `$m ($`x`$ and $`y`$) is derived. The transverse vertex resolution is therefore $`\sigma _r=\sqrt{2}\sigma _x40\mu m`$. By choosing the vertex position in the $`x,y`$-plane as the origin of the event coordinate system, we account for event-by-event displacements within the diameter of the target.
#### 3.2.7 Charged particle tracks
Silicon-drift track segments are constructed by connecting the vertex point to hits in SiDC-2 which lie within the fiducial acceptance. A track segment is accepted if there is at least one hit in SiDC-1 within a predefined window around the point of intersection; for more than one hit, the centre of gravity is taken. Once the interaction vertex and the SiDC track segments are reconstructed, the trajectories of charged particles are extrapolated downstream to the Pad Chamber. If a pad hit is found within a certain fiducial window, a track candidate is created.
The sizes of the fiducial windows are expressed as multiples of the rms widths of the corresponding matching distributions; usually these are 5$`\sigma _{match}`$ during production and 3$`\sigma _{match}`$ during the final analysis. The fiducial windows were taken momentum-dependent to reduce efficiency losses at low momentum due to multiple scattering.
The matching of tracks in $`\varphi `$ direction between detectors separated by the magnetic field requires special attention since such tracking window corresponds to a momentum cut. To avoid loss of tracks by multiple scattering for large $`\varphi `$ deflection, the fiducial window in $`\theta `$ direction is opened to assume the shape of a butterfly. During the production stage, we use a tracking window in $`\varphi `$ of $`\pm `$ 0.6 rad. This corresponds to the p<sub>t</sub>-cut of 50 MeV/$`c`$ mentioned already. During off-line analysis, the size of the butterfly is approximated by a sector of 100 mrad in $`\varphi `$, corresponding to a lower momentum cut-off of 1 GeV/$`c`$, the standard $`p_t`$ cut of 200 MeV/$`c`$, and 3 mrad in $`\theta `$ direction. Matching distributions will be discussed in sect. 3.2.
The contribution of background hits was estimated by applying detector rotations. Since the hits in the SiDCโs are highly correlated, the main background contribution comes from random combinations of SiDC track segments with hits in the Pad Chamber. By rotating these hits in a given event by a random angle with respect to the silicon detectors, the true physics signal is destroyed and only background tracks remain.
By requiring that track elements from all five detectors match within three standard deviations of the detector resolutions combined with the rms spread due to multiple scattering, fake tracks are reduced to a negligible level.
#### 3.2.8 Straight pion tracks for calibration
Straight tracks of high-momentum pions are an important tool for fine-tuning the spectrometer calibration ceretto-phd . Pion identification uses external tracking by the SiDCโs and the Pad Chamber to predict ring centre positions in the RICHes. Pions have been selected under tight quality criteria regarding matching, number of photons on Cherenkov rings, and clean environment around rings.
To extend the SiDC track segments into the Pad Chamber, only a narrow window of $`\pm 30`$ mrad in azimuth is searched for a matching hit, variations in deflection being very small. In polar direction, the Pad Chamber is searched merely over the matching window between the two detectors since multiple scattering is negligible. All hits found are candidates. From the coordinates before and after the field, the ring centres in the RICH detectors are predicted. The pointing to RICH-1, without deflection, is unproblematic. With the momentum information derived from the azimuthal deflection between SiDC and Pad Chamber, the tracks are extrapolated into RICH-2, behind the field.
Photon hits that fall into the vicinity of the predictors are collected and used as input for a robust fitting algorithm robust96 . To find centre and radius of the ring, the rms deviation of hit positions from a circle with radius $`R`$ around the ring centre is minimised in a Gaussian fit potential with three free parameters. Once the radius $`R`$ is found, the pion momentum is re-evaluated from the relation
$$\sqrt{p^2+m^2}=\gamma _{thr}\frac{m}{\sqrt{1(R/R_{\mathrm{}})^2)}},$$
(8)
where $`R_{\mathrm{}}`$ stands for the asymptotic ring radius.
#### 3.2.9 Spectrometer calibration
For the analysis of the 1995 and 1996 data every spectrometer component was calibrated. Several successive steps were
necessary to align all detectors and to derive appropriate correction functions for the analysis chain. A first rough calibration was performed to adjust detector offsets and to correct small rotations and tilts.
Since electron mobility in silicon strongly depends on temperature<sup>20</sup><sup>20</sup>20The electron mobility $`\mu =v_{drift}/E`$ depends on temperature as $`\mu T^{2.4}`$., the calibration is strongly affected by temperature variations. To keep the calibration stable over periods of hours, we employed the simple and fast method by which the position of the upper edge of the drift time spectrum is monitored to provide an โonlineโ drift-velocity stabilisation. The feedback procedure successfully stabilises the calibration as can be seen from the resulting stability of vertex positions in a set of test measurements taken while the cooling was switched off, Fig. 11. Under normal running conditions, the temperature variations were typically below 0.5C over 12 hours.
On even longer time scales, drifts in the calibration of the SiDCโs were prevented employing the fixed reference provided by the Pad Chamber using high $`p_t`$ high-statistics pion samples. The method registers the deviations of hits in either one of the SiDCโs from the straight line that connects the event vertex with a selected hit in the Pad Chamber; it works fine in a pre-calibrated system. The effects of fine tuning the parameters $`v_{drift}`$ and $`t_{min}`$ of SiDC-1 can be seen in Fig. 12.
Response from RICH-1 when given close scrutiny by the absolute reference grid of the Pad Chamber revealed slight deformations of the spherical mirrors which caused non-linearities in polar angle. Following the observation that the radius of curvature of mirror-1 decreases towards the rim, application of a small $`\theta `$-dependent variation in the focal length removed this problem.
#### 3.2.10 Matching quality
Starting with an internal calibration of the SiDC vertex telescope, the remaining detectors are aligned to the centre of SiDC-1. Internal consistency and quality of the readjusted calibration can be evaluated from the residual offset in centres of
gravity, the widths of the matching distribution close to the peak, and the amount of background in the tails, for all detector combinations. Examples of matching distributions are shown in Fig. 13.
At low momenta, the matching quality deteriorates because of multiple scattering in the material, and it is no longer determined by the detector resolutions alone. The widths of the matching distributions in $`\theta `$, plotted for SiDC - Pad Chamber and RICH-1 - Pad Chamber shown in Fig. 14 for measured and simulated spectra, exhibit very well the dominance of the multiple-scattering contribution increasing with $`1/p`$ over the momentum-independent detector resolution towards small momenta. The simulations with the actual spectrometer parameters are shown by dashed lines. Final track selection is achieved by appropriate momentum-dependent cuts in the matching distributions.
### 3.3 Momentum analysis and mass resolution
Since particles passing the SiDCโs and RICH-1 have experienced no magnetic field, these detectors are used as zero-deflection reference.
The deflections $`\mathrm{\Delta }\varphi `$ of charged particles of momentum $`p`$ between the RICHes and between the SiDCโs and the Pad Chamber were given in sect. 2.4. The best resolution is obtained from all possible detector combinations, weighted by the respective accuracy. The result of the simulation with the measured detector resolutions as input is shown in Fig. 15. By detailed Monte Carlo simulation which realistically describes the individual detector resolutions and the quality of the matching distributions, the overall resolution is described by the function
$$\mathrm{\Delta }p/p=\sqrt{(2.3\%p)^2+(3.5\%)^2},$$
(9)
with $`p`$ in GeV/$`c`$. The momentum dependent part is due to detector resolutions, the constant part due to multiple scattering. The mass resolution at the $`\rho /\omega `$ is about 6$`\%`$, at the $`\varphi `$ about 7$`\%`$.
### 3.4 Rejection of combinatorial background
With a $`p_t`$ cut of only 50 MeV/$`c`$ at the production level, an important part of the rejection is already achieved: very soft tracks are removed, the tracks of all reconstructed conversion and Dalitz pairs are marked and taken out of the further analysis, including pairs with only one electron of $`p_t>`$ 200 MeV/$`c`$. Only then is the single-track $`p_t`$-cut tightened from 50 to 200 MeV/$`c`$ which reduces the combinatorial pair background by about a factor of 10 while keeping 97 $`\%`$ of signal pairs. Tracking cuts are narrowed down from 5$`\sigma `$ to 3$`\sigma `$. Together with increased requirements in track and ring quality, this results in a drastic suppression of fake tracks as mentioned before.
The most powerful rejection tools on the pair level derive from the fact that target conversions and a large fraction of $`\pi ^{}`$ Dalitz decays produce close tracks. These are either registered as single hits of โdouble-dE/dxโ in the SiDCโs when the separation is less than about 3 mrad, or as a โdouble ringโ in RICH-1 when the two rings coalesce into a single structure for separations below 8 mrad; or as more or less well resolved tracks, still considerably closer than the mean spacing of charged particle tracks in the SiDCโs or of electron tracks in RICH-1.
To maintain high rejection power also in the region between perfect overlap and resolved double hits, we use a โsummationโ window in the SiDCโs of 5 mrad in which hit amplitudes are summed up<sup>21</sup><sup>21</sup>21This allowed to relax demands on double-hit resolution and thereby to avoid excessive (and mysterious) โhit splittingโ.. Figure 16 shows a two-dimensional plot of the hit response in SiDC-1 vs SiDC-2 obtained this way. It demonstrates that with an appropriate two-dimensional rejection-cut, photon conversions in the target are efficiently rejected without cutting much into the Landau tail of single-hit distributions. RICH-1 identifies close tracks of conversions and Dalitz pairs by the number of photon hits, which suffers severely from pile-up losses, however. Alternatively, we use directly the analog sum of the gain-corrected pad amplitudes to identify โdouble ringsโ. Photon conversions occurring in SiDC-1 differ in response from target conversions only by producing on average a single-hit signal in SiDC and can be recognised by such signature.
To be specific, we use the following four rejection steps which evolve in complexity with the information gained along the trajectories:
1. Tracks are rejected if the amplitudes of the 5 mrad summation window in both SiDCโs are larger than the typical single hit response encoded by parameters
$`S1_{high}`$ and $`S2_{high}`$.
2. Conversions in SiDC-1 are rejected by the requirement that the amplitude in SiDC-1 is below $`S1_{low}`$ and the amplitude in SiDC-2 above $`S2_{high}`$, summed over a 7.5-mrad window; and by observation of a double ring in RICH-1.
3. A track with a double ring in RICH-1 is rejected already if $`S_{high}`$ is surpassed in only one of the SiDCโs. Besides improving rejection efficiency in general, this cut rejects conversions in SiDC-2.
4. Tracks are rejected if another ring in RICH-1 is found within a wider window of 35 mrad. To avoid excessive vetoing by accidental structures, the second ring is required to be of high quality and to connect to a track segment in both SiDCโs.
A few comments are in order. The search in the SiDCs for hits near
tracks cannot be extended to larger distances because of the increasing chance to find a pion, vetoing the signal. For the purpose of โDalitz rejectionโ, the veto is better based on RICH-1 where the inspection area can be opened up to 35 mrad without much signal loss because the density of electron tracks is so much lower. The strategy described above is quite successful in rejecting conversions but even with the close-ring cut it is of limited power in rejecting $`\pi ^0`$-Dalitz pairs because of the larger opening angles involved.
In the last step we also reject surviving charged pion tracks. Figure 17 is a scatter plot of track deflection $`\mathrm{\Delta }\varphi `$ as abscissa, and the sum of the reduced ring radii in both RICHes on the ordinate; this quantity only depends on the Lorentz $`\gamma `$ factor. Because of different mass compared to electrons, pions of same momentum (deflection) have a different $`\gamma `$ (radius). It is seen that pion tracks even with asymptotic radius are clearly distinguished from electron tracks by virtue of the much smaller deflection in the field, and they can be rejected by a two-dimensional cut in both RICH detectors on ring radius vs deflection. This cut was used only in the โ96 analysis; it improves the signal-to-background ratio at large masses and has little impact at lower masses.
### 3.5 Optimising signal quality
In tuning the parameters of quality and rejection cuts, we strictly avoided to optimise on the data itself. Such practice is known to increase the risk of statistical fluctuations producing spurious, misleading results. Rather, the signal efficiency was determined by overlaying Monte-Carlo-generated tracks chosen from the hadronic cocktail on real events and measuring the rate of successful reconstructions using the quality and rejection cuts as in the current step of the data analysis. The signal efficiency may also be monitored and optimised using a sample of fully reconstructed Dalitz pairs socol-phd ; damjanovic-phd . The rejection steps were also tuned by Monte-Carlo techniques with generated conversions and $`\pi ^0`$-Dalitz decays. Such a Monte-Carlo simulation of the background is quite demanding and was only achieved in the 1996 data analysis lenkeit-phd . In the 1995 analysis, the background was taken from the data sample itself.
By these procedures, each analysis cut is arbitrated objectively according to how much in background rejection is gained for how little loss in signal efficiency. A clear understanding of what is meant by an โoptimalโ balance is provided by the โequivalent background-free signalโ<sup>22</sup><sup>22</sup>22It is equal to the number of signal pairs of the same statistical significance if there would be no background.
$$S_{eq}=\frac{S^2}{S+2B}.$$
(10)
For a given rejection cut which reduces the signal from $`S`$ to $`S^{}=\epsilon S`$ and rejects a fraction $`(BB^{})/B=r`$ of the background, the ratio
$$\frac{S_{eq}^{}}{S_{eq}}\frac{\epsilon ^2}{1r}$$
(11)
should be maximised.
In summary, the rejection steps together reduce the number of background pairs by factors of 15 and 12 for mass above and below 200 MeV/$`c^2`$, respectively, while signal pairs are reduced by only 20$`\%`$.
### 3.6 Pairing and subtraction of combinatorial background
The combinatorial background $`B`$ of unlike-sign pairs can be accounted for by the like-sign pair sample because by lack of any physics source the latter is purely combinatoric,
$$B=2\sqrt{N_{++}N_{}}.$$
(12)
Here, $`N_{++},N_{}`$ denote the numbers of $`e^+e^+`$ and $`e^{}e^{}`$ pairs, respectively. The relation is exact when charge symmetry is fulfilled. The number of signal pairs $`S`$ is obtained by subtracting the number of background pairs from the number of unlike-sign pairs,
$$S=N_+B.$$
(13)
The variance in $`S`$ is approximately<sup>23</sup><sup>23</sup>23Combinatorial pairs are non-Poissonian, their variance exceeds the mean. Because the density is only of order $`10^4`$ per event, the deviation from eqn. (3.14), however, is below 5$`\%`$.
$$\sigma _S^2=N_++B.$$
(14)
Typically, the open pair signal $`S`$ accounts for only a small fraction of our total unlike-sign pair sample for masses above 200 MeV/$`c^2`$.
The asymmetry induced by Compton electrons and $`K_{e3}`$ decays is below 1$`\%`$ and would alter the factor of 2 in eqn. (3.12) by less than $`1\times 10^3`$. It is worthwhile to note that eqn. (3.12)
remains a very good assumption also in case of a small imbalance in the number of tracks of positive and negative charge, be it due to different acceptance or reconstruction efficiencies, or to a generic imbalance socol-phd . For an asymmetry of 10$`\%`$ in the number of positive and negative tracks, the factor 2 in the above expression would decrease by 2$`\%`$ only.
The asserted symmetry has been tested by generating purely combinatorial pairs of either type, like-sign and unlike-sign, from data by โreversingโ the rejection cuts. The recipe is the following: prepare a clean sample of conversions and Dalitz decays and recombine all tracks such that none meets its original partner. This sample is void of signal pairs and in every respect resembles the experimental pair background. The mass spectra of both components of this artificial background sample are compared with each other in Fig. 18. No significant charge asymmetry is visible between the mass spectra of the like-sign and the unlike-sign pair background, neither in shape nor in absolute yield. Over the full mass range, the ratio of the integral yields is
$$Y_{e^+e^{}}/(Y_{e^+e^+}+Y_{e^{}e^{}})=0.996\pm 0.005,$$
(15)
which excludes an asymmetry larger than 1$`\%`$ at 90$`\%`$ confidence. We shall return to this topic in sect. 6.5.
Let us take an alternative route of estimating combinatorial background. A philosophically correct way is to generate unlike-sign pairs by combining strictly uncorrelated tracks chosen from different events under the same kinematic constraints as applied to the data. This is commonly referred to as the mixed event method. Practically, efficiency losses that occur in the true data for close tracks have not been accounted for. A comparison of combinatorial mass spectra obtained by the two methods from the full 1996 data set
is displayed in Fig. 19. The coarse comparison on a log scale (top panel) does not show deviations. On a finer linear scale (bottom panel), the ratio indicates that the like-sign pairs loose up to 30$`\%`$ compared to the mixed-event generated unlike-sign pairs - which is, however,
due to the mentioned loss in detection efficiency at close track separation; an effect not present in the current realization of the mixed-event method. Yet, the analysis allows to conclude that above $`m`$ 300 MeV/$`c^2`$ the ratio stays at unity within a band of less than $`\pm `$1.5$`\%`$.
Measured mass spectra of like-sign and unlike-sign pairs are shown in Fig. 20. For comparison, a like-sign pair background has also been generated by random pairing of electron tracks with momenta, polar angles and transverse momenta chosen at random from measured distributions. The result of such simulation is shown in the figure and compares very well with the measured like-sign mass spectrum. The simulated opening-angle distribution (not shown) also agrees well with the measured distribution.
Figs. 18, 19 and 20 display the typical shape of the combinatorial mass spectrum for the applied cut on track $`p_t`$. This kinematical condition produces a falloff towards low masses and a broad peak at about twice the minimum track $`p_t`$. The peculiar shape of the combinatorial mass spectrum peaking around mass of 500 MeV/$`c`$<sup>2</sup> $``$ where the enhancement dominates $``$ raises doubts whether the combinatorial background has been correctly assessed and subtracted. We return to this issue in sect. 6.5 to assert an unbiased background handling.
### 3.7 Spectra and statistical errors
The raw signal counts and their relative statistical errors are obtained by subtracting the combinatorial background channel by channel from the spectrum of unlike-sign pairs,
$$s_i=(n_{i,+}b_i),\sigma _i/s_i=\sqrt{n_{i,+}+b_i}/(n_{i,+}b_i).$$
(16)
The counts per channel add up to the total counts known already from eqns. (3.12) and (3.13),
$$\underset{i}{}n_{i,+}=N_+,\underset{i}{}b_i=B.$$
(17)
Pair yields per event, corrected for reconstruction efficiency and normalised to $`N_{ch}`$, are written symbolically
$$y_i=g(1/ฯต)_is_i,$$
(18)
with $`g`$ a scaling factor and $`1/ฯต`$ for efficiency correction, possibly channel dependent. Relative statistical errors are determined by the raw data counts,
$$\sigma _{y,i}/y_i=\sigma _i/s_i.$$
(19)
To reduce staggering in signal counts due to channel fluctuations in the background spectrum, โsmoothedโ combinatorial pair spectra free of bin-to-bin fluctuations have been subtracted. The method allows, moreover, to limit the fluctuations in data points of background-subtracted spectra to truly statistically independent errors; it is discussed in the following.
The Monte-Carlo generation of background spectra is described in the previous section. The subtraction from the unlike-sign pair spectrum is done bin by bin,
$$s_i^{smooth}=n_{i,+}b_i^{MC}=n_{i,+}f_iB.$$
(20)
The fractions $`f_i`$ add up to $`1`$. The bin-to-bin errors
$$\left(\sigma _{y,i}/y_i\right)_{smooth}=\sqrt{n_{i,+}}/(n_{i,+}b_i)$$
(21)
in our case of $`B/S1`$ amount to a fraction quite accurately of $`1/\sqrt{2}`$ of the full error of (3.19). The finite sample error in $`B`$, to which the smooth background spectrum is normalised, was neglected so far. It contributes to each channel a share of the same magnitude as the bin-to-bin error itself<sup>24</sup><sup>24</sup>24because these errors add in quadrature..
Since the โnormalisationโ errors common to all channels differ in size, according to the magnitude of $`f_i`$ in eqn.(3.20), they cannot be represented as an error in scale. However, there is no need to show them separately; we only have to keep in mind that the data points have normalisation errors which are of the same size as the statistical bin-to-bin errors which will be displayed, but are tightly correlated within the entire spectrum.
### 3.8 Centrality determination
The measurement of absolute cross sections is hampered by the fact that the number of particles that actually pass the segmented target is not precisely defined; this is because the diameter of the Au disks is not much larger than the beam diameter. We decided therefore to calibrate the centrality of collisions by the shape of the pseudo-rapidity density of charged particles as measured by the SiDCโs. Once the measured $`N_{ch}`$ distribution is corrected for various instrumental distortions, it is used to calibrate the charge density obtained from minimum-bias UrQMD calculations urqmd . These still differ from the measured distribution at the low-$`N_{ch}`$ tail due to the non-ideal response of the interaction trigger. The final step consists in describing the triggered $`N_{ch}`$ distribution by a linear combination of UrQMD calculations belonging to different impact parameters. These steps are described below.
#### 3.8.1 Charged-particle density
Charged-particle density in the range 2$`\eta `$3 is derived off-line from the number of tracks that emerge from the event vertex and intersect the two SiDCโs sufficiently close to reconstructed hits. The determination of the track reconstruction efficiency is a rather complex problem. To the percent accuracy required, it is influenced not only by hardware imperfections (i.e. dead anodes, electronic noise, pulse shape), but also by the quality of hit and track reconstruction. In particular, โpile-upโ effects due to finite double-hit resolution and artificial hit splitting are the most important effects to be corrected for. To that goal, we simulate charged particle tracks in the SiDCโs generated from UrQMD events. The analysis of the Monte-Carlo events uses the standard data analysis software, besides larger matching windows of 5 $`\sigma _{match}`$. Details are described in sect. 4.
The functional dependence of the inverse of the reconstruction efficiency $`\epsilon `$ on the measured number of charged particles is of the type
$`\epsilon ^1(N_{ch,measured})`$ $`=`$ $`N_{ch,true}/N_{ch,measured}=`$
$`a+bN_{ch,measured}.`$ (22)
The intercept at $`N_{ch,measured}=0`$ corresponds to the inverse โstaticโ efficiency $`a^1`$ and includes not only losses due to dead anodes, but also gains from artificial hit splitting. It amounted to 97$`\%`$ and 93$`\%`$ in 1995 and 1996, respectively. The parameter $`b`$ is about (6-7)$`\times `$10<sup>-4</sup> which corresponds to about 10$`\%`$ relative losses, considerably more than expected from pile-up alone.
The corrected multiplicity distribution of the trigger-selection is shown in Fig. 21. In the middle part and at the high-$`N_{ch}`$ side, the shape of the distribution is supposed to be an undistorted image of partial cross sections, but not so at low $`N_{ch}`$ where the trigger profile of the MD with threshold set at 100 $`\mathrm{๐๐๐๐ }`$ becomes visible (see sect. 2.6). The thresholds of the centrality trigger corresponded roughly to 30% for โ96 and 35$`\%`$ of the total inelastic cross section for โ95.
#### 3.8.2 Calibrating the centrality using UrQMD calculations
A more precise calibration of the centralities is described in the following. Minimum-bias distributions obtained by UrQMD (version 1.3) have been scaled slightly to make their upper corners coincide with the measured trigger distributions. From the UrQMD minimum-bias collisions several centrality classes are sorted out according to certain impact parameter ranges. To reproduce the shape of the trigger profile at the lower edge of the $`N_{ch}`$
distribution, a suitable linear combination of the corresponding $`N_{ch}`$ distributions is used. The resulting distributions are displayed in Fig. 22. The total inelastic cross section of 6.94 barn, and the trigger fractions were calculated with a geometrical overlap model. The resulting $`x\sigma /\sigma _{inel}`$ are:
$$x(^{}95)=0.6\times 35\%+0.2\times 33\%+0.2\times 28\%=\mathrm{\hspace{0.17em}33}\%$$
(23)
$$x(^{}96)=0.16\times 35\%+0.15\times 30\%+0.69\times 23\%=\mathrm{\hspace{0.17em}26}\%.$$
(24)
The systematic error from variation in the linear combinations alone is estimated as $`\pm `$1.5$`\%`$ absolute. For the unified data, an average centrality of 28$`\%`$ of the top geometrical cross section has been adopted with an estimated uncertainty of about $`\pm `$2$`\%`$.
## 4 Monte Carlo simulation
### 4.1 Detector simulation
The spectrometer with all material in proper geometry has been implemented in the GEANT geant detector simulation package version 3.15. The present version of GEANT was not able to describe the number of photons per electron ring correctly, which forced us to write our own function.
The detector response is simulated by taking into account the particular signal generation including all known effects that influence position, width and amplitude. In the SiDCs these include the charge division among adjacent anodes, local variations in drift-velocity over radius, diffusion along the drift path in radial and orthogonal direction, anode-wise gain variations including dead anodes, electronic noise from the readout, and digitisation errors. In the RICH detectors we account for chromatic aberration and mirror quality. In the UV detectors and the Pad Chamber, simulations include the transverse diffusion in the conversion zone and the first amplifying gap, the digitisation effects in the last multi-wire amplification, and the noise of the readout electronics.
Most crucial but difficult to achieve is a realistic simulation of the background in which the simulated signal is to be embedded. The background is caused by charged particles, gamma rays and electro-magnetic radiation produced directly or indirectly by the collision, $`\delta `$-electrons, and particles passing the UV-detectors.
In addition, the entire spectrometer or parts thereof might deviate from response linear with produced signal charges due to transient saturation effects of various kinds. No sufficiently reliable simulation of the background was possible up to now. We use real data events into which the simulated Monte Carlo (MC) signals are embedded. This is done by adding the MC amplitude for each single cell, or pad, on top of the data amplitude. The systematic error introduced by increasing the detector occupancy has been estimated to be negligible.
The MC signal was reconstructed using algorithms identical with those of the data analysis. By comparing the signals of reconstructed MC tracks and reconstructed data tracks we fine-tuned the parameters of the detector simulations to achieve the best description of the data. The characteristics employed for the comparison were the amplitude distributions of hits, the widths, the numbers of pads/time-bins/anodes contributing to a hit, the number of hits belonging to a ring and the local variations of all these parameters over the detectors. An example of the agreement achieved between MC simulation and data analysis in reconstruction of Cherenkov rings is shown in Fig. 23.
It is crucial to model quantitatively the response to compound signal patterns, like accuracy in reconstruction of hit positions or ring centre positions, or the number distributions of reconstructed photons per ring. Quantities of that kind depend also on characteristics of the physics input like opening-angle distributions of pairs in the sample and the polar distribution of electron tracks. They are therefore best checked on the data itself. For that purpose a special sample enriched in $`\pi ^0`$-Dalitz decays was extracted from the data by searching for pairs with opening angles less than 50 mrad plb422 . Such pair sample has a typical signal-to-background ratio of order one.
The properties of this sample compare well with the corresponding properties of simulated $`\pi ^0`$-Dalitz decays in ring-parameters and efficiency. There is but one exception: the measured ring centre resolution of the first RICH detector is worse by a factor of 1.4 compared to the MC simulation, and also deviates from the expected resolution for single photon hits. For lack of understanding, we introduce an ad hoc Gaussian random smearing of 0.9 mrad into the simulation to meet the measured value. The influence on final momentum resolution is very weak, since the azimuthal track position measured in the SiDCโs is anyhow more accurate and outweighs the RICH contribution to tracking.
### 4.2 Determination of reconstruction efficiency
To compare the number of reconstructed electron pairs to any physics-based expectation, an absolute normalisation is required. To correct spectra for reconstruction efficiency<sup>25</sup><sup>25</sup>25For unit pair efficiency, all pairs that have both electrons in the acceptance of the spectrometer are reconstructed., measured yields are multiplied by the inverse pair reconstruction efficiency averaged over the full sample of pair candidates. We may call this a correction โon a statistical basisโ, in contrast to an โevent-by-event correctionโ where the invariant-mass or $`p_t^{ee}`$ spectra are incremented with the inverse efficiency of that event, or of particular track candidates.
One way to determine the reconstruction efficiency is to measure the number of pairs from a well-defined physics sample, and adjusting it to the expected number. We first explain this โDalitz methodโ and then turn to the alternative Overlay Monte-Carlo technique.
#### 4.2.1 The Dalitz method
Our sample of โDalitz pairsโ with $`m<`$200 MeV/$`c^2`$, but opening angles larger than 35 mrad and track $`p_t`$ above 200 MeV/$`c`$ is considered an excellent choice in place. It consists mainly of $`\pi ^0`$-Dalitz pairs and a contribution of $`\eta `$-Dalitz pairs on the 10$`\%`$ level. The opening-angle cut reduces conversions to (5-10) $`\%`$ which can be corrected for. At present, we cannot exclude the (very interesting) possibility that a strong source of thermal electron pairs may compete on the level of a few percent rapp-wambach2000 .
The required average inverse reconstruction efficiency is derived by dividing the measured number ratio of charged-particles to Dalitz pairs by the cocktail expectation,<sup>26</sup><sup>26</sup>26By lower case $`n`$ we denote numbers from physics simulation, by upper case $`N`$ measured counts.
$$\frac{1}{\epsilon }=\frac{N_{ch}}{N_{ee}}/\frac{n_{ch}}{n_{ee}}.$$
(25)
With the event generator GENESIS genesis-old the number of electron pairs of the hadronic cocktail is calculated, but it is normalised to the number of neutral pions; using the $`\pi ^{}`$-to-charged-particle ratio, we obtain the cocktail reference ratio in the required normalisation to charged particles,
$$\frac{n_{ee}}{n_{ch}}=\left(\frac{n_\pi ^{}}{n_{ch}}\right)\frac{n_{ee}}{n_\pi ^{}}.$$
(26)
Note that only the cocktail part factorises into averages of electron pairs and charged particles<sup>27</sup><sup>27</sup>27Since all components of the cocktail scale linearly with $`n_{ch}`$, ratios are constant. while the experimental average does not, since $`N_{ch}/N_{ee}`$ depends, via reconstruction efficiency on $`N_{ch}`$.
The recipe that follows is to average the ratio of the number of charged particles to the number of Dalitz pairs over all events of a given multiplicity class, and divide by the cocktail ratio to obtain the factor 1/$`\epsilon `$ by which the raw data counts have to be multiplied. The discussion emphasises the importance to measure the reconstruction efficiency as a function of $`N_{ch}`$.
#### 4.2.2 The Overlay-Monte-Carlo method
A method of wider application is to reconstruct a MC-simulated sample of electron pairs from the sources under consideration, be it Dalitz decays with masses below 200 MeV/$`c^2`$ as discussed below, or the full hadronic cocktail of electron pairs, always under condition of all acceptance and analysis cuts and using standard analysis software. For realistic background conditions, the simulated pairs are overlaid one by one on data events of given
$`N_{ch}`$, and the desired correction factor is given by the average of the inverse probability of successful reconstruction,
$$\frac{1}{\epsilon }=\frac{1}{p_{rec}}.$$
(27)
The pair efficiencies can be decomposed into the products of the track efficiencies of the detectors for a meaningful check on the reliability of the simulation. Table 1 compares full-track and pair efficiencies for a standard Dalitz sample with those obtained by piecewise multiplication of individual detector efficiencies; all efficiencies were measured by the Overlay-Monte-Carlo technique. The close agreement between the two indicates that the procedures of track reconstruction and pairing work properly.
A further reduction in pair efficiency by about a factor of two occurs in the course of the background rejection discussed in sect. 4.4; also these efficiency losses are understood by the MC simulations.
The Overlay-MC method applied to high-mass electron pairs suffers from the low efficiency combined with acceptance losses which forbid to collect sufficient statistics samples. We have therefore based all corrections on products of track
efficiencies constructed in such a way that on the one hand the correct $`dN_{ch}/d\eta `$ dependence of the Dalitz pair efficiency as shown in Fig.24 is reproduced, and on the other, that the specific differences in pair efficiency are taken care of in an approximate way. Those arise from the $`\theta `$-dependent hit density and are treated separately.
#### 4.2.3 $`N_{ch}`$ dependent efficiency
Reconstruction efficiencies for Dalitz pairs with mass below 200 MeV/$`c^2`$ are displayed in Fig. 24 as a function of charged-particle rapidity density. The pairs are filtered for opening angles $`\mathrm{\Theta }_{ee}`$ 35 mrad to reduce conversions, and they obey the condition $`p_t`$ 200 MeV/$`c`$ on single-electron tracks. For the data points, the dependence on $`N_{ch}`$ is derived by postulating that sources scale with $`N_{ch}`$ while the absolute magnitude is fixed by normalising to the MC simulation as explained above. We observe a loss in MC pair efficiency by a factor of 2.2 when rapidity density is raised from 150 to 350. This degradation factor for the three โ96-data analyses socol-phd ; lenkeit-phd ; hering-phd amounts to 2.30$`\pm `$0.13; the individual results lie remarkably close considering the large spread in absolute efficiencies. The causes of the efficiency loss are twofold. For one, the recognition of hits in the SiDCโs and of Cherenkov rings in the RICHes is deteriorated by an increasing number of background hits. In addition, with increasing track density also the chances increase that rejection of background tracks accidentally vetoes nearby signal tracks.
The strong decrease of pair efficiency with $`N_{ch}`$ agrees very well with the Overlay-MC simulation also displayed in Fig. 24. In the โ96 data analyses, this curve is used for efficiency correction on an event-by-event basis: signal pairs are stored with weights given by the inverse of the efficiency at given event multiplicity. The same procedure was applied in the โ95 data analysis, but the $`N_{ch}`$ dependence is considerably weaker; the ratio quoted above is only 1.4 voigt-phd ), possibly because of a lower event background compared to 1996, but also due to a different treatment of close hits in the SiDCโs.
#### 4.2.4 $`\theta `$-dependent efficiency
With track density increasing towards small polar angles, the track reconstruction efficiency drops by more than a factor of three over the acceptance, as shown in the upper panel of Fig. 25. Since we will deal only with pair yields integrated over the acceptance, one is inclined to disregard this effect.
A $`\theta `$-dependent efficiency, however, combined with the limited $`\theta `$ acceptance, affects pairs of different opening-angle characteristics in different ways; and we have seen already in Fig. 5 that $`\pi ^{}`$-Dalitz pairs and pairs from $`\omega `$ decays have indeed very different $`\mathrm{\Theta }_{ee}`$ behaviour.
The acceptance condition affects wide open pairs considerably. For the given example of $`\omega `$ decays with typical $`\mathrm{\Theta }_{ee}`$ 400 mrad, only pairs with both electrons at large $`\theta `$ are accepted and reconstructed with above-average efficiency. The rise in pair efficiency for very large opening angles is seen in the lower panel of Fig. 25. The shallow minimum at $`\mathrm{\Theta }_{ee}`$ 250 mrad is typical for the majority of the open pairs in our sample. We see also that the efficiency changes only by very little towards lower $`\mathrm{\Theta }_{ee}`$. The flat region even includes our sample of Dalitz pairs; owed to the standard cut $`\mathrm{\Theta }_{ee}`$ 35 mrad, their most-probable polar angle is as large as 190 mrad. So, compared to the โmajority of open pairsโ cited above, the reconstruction efficiency of the sample of Dalitz pairs is almost undegraded. This is very satisfactory for the โDalitz methodโ described above.
Yet, in order to protect characteristic features in spectra of pairs with opening angles comparable to the acceptance of the spectrometer, the efficiency correction should involve the $`\theta `$ distributions of the contributing tracks. We describe this method in the following.
The $`\theta `$ dependence in track efficiency was taken care of in an approximate way under the boundary condition to leave the measured efficiency for Dalitz pairs shown in Fig. 24 unchanged. The pair efficiency is factorised into track efficiencies which are modified according to
$$\epsilon _{track}^{anypair}(N_{ch},\theta )=\epsilon _{track}^{Dalitz}(N_{ch})f(\theta ).$$
(28)
The function $`f(\theta )`$ incorporates the measured $`\theta `$ dependence of the track efficiency shown in Fig. 25, upper part.
The procedure is to increment in an event of multiplicity $`N_{ch}`$ for each pair candidate a weight into the like-sign or unlike-sign spectrum which is
$`w(N_{ch};\theta _1,\theta _2)={\displaystyle \frac{1}{\epsilon _{track}^{anypair}(N_{ch},\theta _1)\epsilon _{track}^{anypair}(N_{ch},\theta _2)}}`$ $``$
$`{\displaystyle \frac{1}{\epsilon _{pair}^{Dalitz}(N_{ch})f(\theta _1)f(\theta _2)}}.`$ (29)
The function $`f(\theta )`$ is suitably normalized to assure that the Dalitz sample is unchanged.
The described event-by-event correction was not performed in the โ95 data analyses while the โ96-analysis of Ref. hering-phd was corrected for a truly two-dimensional efficiency.
### 4.3 Pair acceptance
The acceptance of the CERES spectrometer for electron pairs has been evaluated by MC simulations assuming uniform input distributions in mass $`m`$, pair transverse
momentum $`p_t^{ee}`$, and rapidity $`y`$. How the CERES acceptance depends on $`m`$ and $`p_t^{ee}`$ is shown in Fig. 26 for the standard rapidity range 2.1$`\eta `$2.65; the normalisation is done cutting also the virtual photon distribution to this range in the input. We have applied here the standard analysis cuts $`p_t`$ 200 MeV/$`c`$ and $`\mathrm{\Theta }_{ee}`$ 35 mrad. The results without cuts are shown for comparison. For masses below 100 MeV/$`c^2`$, the acceptance without cuts approaches one, while the angular cut causes a steep reduction. Higher up in mass, the single-electron $`p_t`$ cut reduces the acceptance at low pair $`p_t^{ee}`$, acting like a $`m_t`$ cut, while at $`p_t^{ee}`$ 400 MeV/$`c`$, the acceptance becomes more uniform and even, within a factor of two, independent of mass, due to some โequalisingโ effect of the opening-angle cut. Still higher up, for $`m>`$ 500 MeV/$`c^2`$, the acceptance is essentially uniform both in mass and in $`p_t^{ee}`$.
An acceptance correction for pairs of โnew physicsโ sources is problematic in principle, since the decay dynamics is not precisely known and may depend on additional parameters, such as helicity angle. In comparison of theories to CERES data, proper accounts of the pair acceptance and other cuts are usually taken by the authors themselves who have the necessary insight into the dynamics of the modelled sources to do so. We therefore correct data only for reconstruction efficiency, not for pair acceptance.
### 4.4 Optimising rejection of combinatorial background
To foster confidence in the data analysis and to improve on data quality it was essential to understand the combinatorial background in every detail. We have presented in the preceding section how the measured numbers of background and signal pairs evolve with the rejection steps applied. Here, we use the tool of full MC simulation to arrive at a quantitative understanding of the background sources. It goes without saying that the elaborate simulation of background rejection supplied a very effective handle to fine-tune rejection cuts.
Figure 27 plots the number of electron tracks from various physics sources which survive a given rejection step in the MC simulation. The decrease is
either due to rejection of single tracks that have unwanted properties like low $`p_t`$, bad matching or suspicious environment, or due to removal of recognised conversion or Dalitz pairs which are still intact and fulfil all conditions.
The experimental points are numbers of background tracks from the post-production steps of the 1996 data analysis. They are derived from the measured total number of background pairs accumulated in 42.2$`\mathrm{\hspace{0.17em}10}^6`$ events, assuming Poissonian track statistics. The agreement between absolute numbers from experiment and simulation is quite reassuring; also the size in relative suppression is in good agreement with the simulation. The residual background remaining after all eight rejection steps is dominated by Dalitz pairs.
## 5 Hadronic decay sources
The โconventionalโ sources contributing to the inclusive mass spectra of electron pairs in the mass range below 1.5 GeV/$`c^2`$ are free decays of light neutral mesons up to and including the $`\varphi `$. These contributions have been determined for p-Be and p-Au collisions with considerable precision in an experiment which combined the electron-pair spectrometer of CERES with the photon calorimeter of TAPS neutral-meson-pBe . In addition to the inclusive electron-pair yield, the cross sections and $`p_t`$ distributions of the $`\pi ^{},\eta ,and\omega `$ mesons have been measured via the electro-magnetic decay modes $`\gamma \gamma `$ and $`\pi ^{}\gamma `$, and an exclusive reconstruction of the $`\pi ^{}`$ and $`\eta `$-Dalitz decays $`\pi ^{},\eta e^+e^{}\gamma `$. These data, supplemented by older measurements from NA27 na27 , allowed us to simulate the electron-pair mass spectra originating from decays of neutral mesons. The result was subject to all experimental cuts and folded with mass resolution. Within error limits it was concluded that the inclusive mass spectra of electron-pairs measured in 450 GeV/c p-Be and p-Au collisions are consistently described by the expected $`e^+e^{}`$ pair โcocktailโ of neutral meson decays neutral-meson-pBe .
The hadronic cocktail serves as our reference standard of hadronic sources to expose effects which are specific to nucleus-nucleus collisions, i.e. spectral shapes and yields which cannot be described by a mere superposition of nucleon-nucleon interactions. Supported by evidence from light collision systems becattini96 , particle ratios were originally assumed constant from p-p to heavy systems drees , and particle yields were scaled with charge multiplicity. This cocktail was instrumental in gauging the low-mass excess in 200 GeV/$`c`$ S-Au collisions prl1995 and in 158 GeV/n Pb-Au collisions from the first (1995) data set ravinovich-qm97 ; plb422 .
The hadronic cocktail is calculated with the Monte Carlo event generator GENESIS genesis-old described in considerable detail elsewhere neutral-meson-pBe . It has been improved since then but the changes with respect to previous CERES analyses of Refs. ravinovich-qm97 ; plb422 ; lenkeit-qm99 ; lenkeit-paris are subtle. These publications are superseded by the present paper.
We sketch here only the main points. Mesons are produced with cross sections scaled up from p-p or p-A, taking into account the modified $`p_t`$ and rapidity distributions. Open charm contributions have been neglected on the basis of reliable estimates open-charm98 . Mesons are then allowed to decay with known branching ratios. Decay kinematics are simulated for the Dalitz decays $`PSe^+e^{}\gamma `$ of the pseudo-scalar mesons
$`\pi ^{},\eta and\eta ^{}`$, for the direct decays of the light vector mesons, $`Ve^+e^{}`$, and for the $`\omega `$ Dalitz decay $`\omega e^+e^{}\pi ^{}`$. Electron momenta are Lorentz-transformed into the laboratory system and convoluted with the experimental resolution profile. The simulated events are subject to the same filters concerning acceptance, $`p_t`$ and opening angle.
Production cross sections of the neutral mesons, their rapidity and transverse-momentum distributions are essential ingredients for simulating the decay contributions to the dilepton spectrum. When data on hadron production from Pb-beam experiments at the SPS (NA44, NA49, NA50, WA98) became available in time for the 1996 data analysis, we used this input from Pb-beam data whenever possible lenkeit-qm99 ; lenkeit-paris ; lenkeit-phd . Information not directly available was derived from the statistical model which describes ratios of integrated hadron yields at chemical freeze-out very well with only two fit parameters, the temperature and the baryon chemical potential; the particular values used are $`T`$= 170 MeV and $`\mu _b=`$ 68 MeV chem-eq99 . The collision system is modelled in a state of collective transverse expansion which is based on the observation wessels96 that inverse-slope parameters $`T`$ of transverse momentum spectra, except for pions, systematically increase with mass stachel-paris98 .
Data on $`p_t`$ distributions of pions have been measured by CERES bielcikova ; ceretto-tsukuba , NA49 appels-np98 , NA44 kaneta-99 , and WA98 aggarwal-98-01 . They are exceptional in that they can not be described by a single exponential. The $`\pi ^0`$ spectrum is generated with two slopes: at $`m_t`$ 200 MeV where it is dominated by secondaries, with $`T_{}`$= 100 MeV, and for the higher part with an inverse slope of $`T_{}`$= 230 MeV. The inclusive $`m_t`$ distribution for neutral pions measured by WA98 is extrapolated to small $`m_t`$ using the charged pion distributions from NA44. At small $`m_t`$ the spectra are dominated by secondary decays from heavier mesons. The decay $`\eta 3\pi ^{}`$ is added separately by hand to the $`\pi ^0`$ distribution.
The rapidity distribution sikler99 of negatively charged hadrons, described as a Gaussian centred at $`y_{max}`$= 2.9 with $`\sigma _y`$= 1.5, has been adopted for all mesons. While the widths of the hadron rapidity distributions decrease with particle mass in proton induced collisions, this is not observed in lead-induced collisions hoehne99 .<sup>28</sup><sup>28</sup>28Particle ratios taken at mid-rapidity are therefore the same as those from partially or fully integrated yields.
The parameters from Pb-beam data and the statistical model are given in Table 2. Meson production ratios are implemented relative to the number of $`\pi ^{}`$โs and include feeding from heavier resonances. Compared to the earlier reference to proton induced collisions neutral-meson-pBe , the production ratios of heavier hadrons are enhanced as, e.g. the $`\eta `$/$`\pi ^0`$ peitzmann98 and the $`2\varphi /(\pi ^++\pi ^{})`$ ratio jouan-falco98 ; puehlhofer98 . The meson yields are normalised to the charged-particle density by the ratio
$$N_\pi ^{}/N_{ch}=0.44.$$
(30)
Brackets denote averaging over the CERES acceptance.
The decay branching ratios given in Table 2 are from Ref. pdg2004 . To simulate the Dalitz decays, the Kroll-Wada expression kroll-wada is multiplied by the electro-magnetic transition form factors fitted to the LEPTON-G data lepton-g . The pole approximation $`F(M^2)=(1bM^2)^1`$ is used for the determination of the form factors of the $`\pi ^{}`$ and the $`\eta `$ Dalitz decays lepton-g . For the $`\omega `$ and the $`\eta ^{}`$, the form factors are determined by fitting a Breit-Wigner function
$$|F(M^2)|^2=\frac{m_\rho ^4}{(M^2m_\rho ^2)^2+m_\rho ^2\mathrm{\Gamma }_\rho ^2}$$
(31)
to describe the resonant behaviour according to the vector dominance model. The parameters used are listed in Table 2. The direct decays of the vector-meson were generated following Gunaris and Sakurai gounaris-sakurai .
The 2-body decay of the $`\rho `$ meson has been re-evaluated frimann-knoll00 for the new GENESIS code genesis-new . We defer details of the revised formula as it is implemented in the 2003 version of the New GENESIS to the appendix. The resulting mass distribution, due to a Boltzmann-type phase space factor $`e^{M/T}`$ and a momentum dependent phase space, both omitted in the previous code, receives a shoulder on the low-mass side and a steeper falloff to higher masses.
All decays where assumed isotropic in the rest frame of the decaying meson except for the Dalitz decays to e<sup>+</sup>e<sup>-</sup>$`\gamma `$ which follow a 1 + cos<sup>2</sup>($`\theta `$) distribution, where $`\theta `$ is measured with respect to the virtual photon direction.
Figure 28 depicts the cocktail based on the Pb-beam data with all corrections. It is the mass spectrum of pairs in the CERES acceptance $`2.1\eta 2.65`$ with standard cuts $`\mathrm{\Theta }_{ee}`$ 35 mrad and $`p_t`$ 200 MeV/$`c`$ . The generator output has been folded with the mass resolution function of the 1996 data set, which includes both the momentum resolution as the dominant source of smearing as in eqn. (3.9), as well as the resolution in pair opening angle; the latter is approximately
$$\sigma _{\mathrm{\Theta }_{ee}}^2(\sqrt{2}\sigma _\theta )^2+\overline{sin^2\theta }(\sqrt{2}\sigma _\varphi )^2,$$
(32)
where $`\sigma _\theta `$ 0.6 mrad and $`\sigma _\varphi `$ 3.0 mrad are the angular track resolutions taking information from all detectors together. The effects of bremsstrahlung emission by electrons traversing the detector material have been included in the simulations, but due to the low material budget in the acceptance of $`X/X_{}`$ 1 $`\%`$, they are hardly noticeable. The overall mass resolution is about 6$`\%`$ in the $`\rho /\omega `$ region and 7$`\%`$ in the region of the $`\varphi `$ (see sect. 3.3).
The cocktail for Pb-Au collisions is compared in Fig. 28 to the cocktail of hadronic sources up-scaled from p-p interactions; the latter is also folded with the CERES resolution obtained for the โ96 data analysis.
The systematic errors of the normalised decay cocktail, relevant for the numerical comparison to the experimental pair data in sect. 7, are discussed separately for the mass regions below and above 200 MeV/$`c^2`$.
The low-mass region is dominated by the $`\pi ^{}`$-Dalitz decay, with $`\eta `$-Dalitz contributing about 15 $`\%`$. Errors arise from the relative production cross section of the $`\pi ^{}`$ and from the parametrisations of the input rapidity and transverse momentum distributions. Normalising the decay cocktail to the number of charged particles is a very powerful means to keep the error in the ratio $`\left(n_\pi ^{}/n_{ch}\right)`$ of eqn. (4.26) small; it is estimated to be at most 5$`\%`$. The rapidity distribution is uncritical since it is taken directly from pion data; the errors should be less than 3 $`\%`$. The transverse momentum distribution, due to the single-electron $`p_t`$ cut, is a little more critical. However, quite different assumptions on the shape of the $`p_t`$ spectrum above the cut, e.g., using $`h^{}`$ from CERES rather than $`\pi ^{}`$ from NA49, lead to differences in yield of only very few percent. This error is therefore also estimated to be below 5$`\%`$. Assuming that the systematic errors are uncorrelated, a total systematic error of 8$`\%`$ is obtained for the low-mass region.
In the high-mass region, the errors are dominated by those of the relative production cross sections of the higher-mass mesons and of the detailed properties of the electro-magnetic decays. Since most of the particle yields relevant for the cocktail have not directly been measured, or like the $`\varphi `$, suffer from experimental controversy jouan-falco98 ; puehlhofer98 , the statistical model predictions have been used instead, and it is therefore their uncertainties which enter. Judging the average fit quality to measured particle ratios chem-eq99 , we estimate these errors to be 20$`\%`$. The uncertainties in the branching ratios and in the transition form factors have been discussed in detail in Ref. neutral-meson-pBe . They contribute $``$ 15$`\%`$ for $`m<`$ 450 MeV/$`c^2`$, $``$ 30$`\%`$ in the mass range of 450-750 MeV/$`c^2`$, and 6$`\%`$ for $`m>`$ 750 MeV/$`c^2`$. Taking these supposedly independent sources of uncertainty together, an overall systematic error of 30$`\%`$ is estimated as the weighted average for the high-mass region.
The integral number of electron pairs per charged particle in the CERES acceptance expected from hadronic sources under standard cut conditions $`p_t^e`$ 200 MeV/$`c`$ and $`\mathrm{\Theta }_{ee}`$ 35 mrad is
$`\left[{\displaystyle \frac{dN_{ee}/dm}{N_{ch}}}\right]_{Genesis}`$ $`=`$
$`\{\begin{array}{cc}(9.27\pm 0.74)10^6\hfill & m<200\mathrm{MeV}/c^2\hfill \\ (2.27\pm 0.7)10^6\hfill & m200\mathrm{MeV}/c^2,\hfill \end{array}`$ (35)
where we have quoted the systematic errors of 8$`\%`$ and 30$`\%`$ for the two mass regions, respectively.
## 6 Results
### 6.1 Samples of reconstructed pairs
For an overview, the pair samples reconstructed from the two data sets of 158 GeV/n Pb-Au collisions taken by the CERES Collaboration in 1995 and 1996 are listed in Table 3. All analyses apply standard cuts on track $`p_t`$ and and pair opening angle $`\mathrm{\Theta }_{ee}`$.
The table lists the pair signal (S) obtained by subtracting the combinatorial background (B) from the measured numbers of unlike-sign pairs, for the mass ranges below (Dalitz pairs) and above 200 MeV/$`c^2`$ (open pairs). In this Table, pair yields are not corrected for reconstruction efficiency (other than in the following figures). In analysis no. 5, event mixing was employed to obtain combinatorial background of unlike-sign pairs. Quoted errors are absolute statistical errors in the respective sample numbers amounting to $`\sigma _S=\sqrt{N_++B}`$. The pair reconstruction efficiencies are given in the last column.
We note a considerable spread among the results of different analysis efforts. This originates from the different values of the rejection cuts used along the various analysis chains. However, normalised pair yields, i.e. numbers of pairs per event, per charged particle, and corrected for pair efficiency, as will be shown below, are very stable: the relative spread in the number of Dalitz pairs turns out to be less than 15$`\%`$.
### 6.2 Inclusive mass spectra
Figure 29 shows the mass spectrum of Ref. plb422 ; voigt-phd from 1995 together with that of Ref. lenkeit-qm99 ; lenkeit-paris ; lenkeit-phd from the 1996 data set. The trigger centralities correspond to the most central 33$`\%`$ and 26$`\%`$ of the inelastic cross section for 1995 and 1996, respectively.
The signal is obtained by subtracting the smoothed like-sign pair background from the spectrum of unlike-sign pairs. The differential pair yield $`dN_{ee}/dm_{ee}`$ per event is normalised to the mean number $`N_{ch}`$ of charged particles in the acceptance. The brackets indicate particle yields per event measured within the CERES acceptance. Reconstruction efficiency has been corrected event-by-event by weighting pairs with the value of the inverse pair efficiency at the particular centrality. Shown are the bin-to-bin statistical errors of eqn. (3.18). To include the normalization errors discussed in sect. 3.7, as it may be relevant when individual data points are compared to other data, as in this figure, or to theoretical-model predictions, the errors should be multiplied by 1.4. Note, however, that these larger errors are no longer statistically independent.
It is apparent that the โ95 data points lie systematically higher than the โ96 data points. Apart from the small difference in mean trigger centrality there is no relevant change in setup. The โ95 analysis voigt-phd ; ravinovich-qm97 ; plb422 followed the strategy to optimise the quantity $`S_{eff}\epsilon ^2/B`$. But while the signal efficiency $`\epsilon `$ was determined by Monte-Carlo, the background $`B`$ was taken as the measured like-sign sample; this strategy was discarded when the full MC simulation became available for the โ96 analysis. Although we could find absolutely no indication that statistical fluctuations in $`B`$ might have steered the analysis towards a โbetterโ final sample, such possibility cannot be strictly excluded.
The re-analysis of the โ95 data socol-phd resulted in a mass spectrum closer in absolute yield and shape to the โ96 mass spectrum shown in Fig. 29. In this analysis, any involuntary bias was avoided by sampling the distributions in a random automatic variation of all cut settings simultaneously, and then choosing the centres of gravity. However, seeing no direct evidence for a biased tuning of cuts, we have chosen to keep the original โ95 analysis to expose our actual systematic uncertainties. We return to this issue below.
### 6.3 Centrality dependence
The efficiency-corrected yield of Dalitz pairs is a solid reference for linear $`N_{ch}`$ dependence. We have a first look at the centrality dependence of the open pair yield by comparing in Fig. 30 the centres of gravity of three $`N_{ch}`$ distributions: that of triggered events without any further condition, and two others taken from events which contain a Dalitz or an open pair candidate, respectively. Because of the large open-pair background, the signal is obtained by subtracting the $`N_{ch}`$ distribution of like-sign-pair events from that of unlike-sign-pair events. The $`N_{ch}`$-dependent pair detection efficiency has been corrected for using the curve of Fig. 24.
We see from Fig. 30 that the centres of gravity of Dalitz and open-pair samples are shifted progressively upward; the mean multiplicities are 285 and 310, respectively, compared to $`N_{ch}`$= 250 of the trigger distribution. The averages of the trigger distribution calculated for events depending linearly or quadratically on $`N_{ch}`$ are 287 and 315, respectively. We conclude that the increase of open-pair production is better described by a quadratic dependence on $`N_{ch}`$, than by a linear dependence. We will have a closer look at the differential centrality dependence in sect. 7 but note that the current finding bears no reference to the hadronic cocktail.
### 6.4 Invariant transverse-momentum spectra
Considerable physics potential resides in the spectra of invariant transverse pair momentum, $`p_t^{ee}`$, as displayed in Fig. 31 for the two data samples and three mass bins.
The spectra are normalised and corrected for reconstruction efficiency.
In the Dalitz region, at low $`p_t^{ee}`$ the spectrum is void due to the analysis cut; both samples agree very well. Data points scatter considerably for the two other mass bins. The full physics relevance of these spectra emerges only in comparison to the expectations for hadronic sources.
### 6.5 Stability of results
The optimisation of quality and rejection cuts was done with no feedback from the signal itself not to exploit statistical fluctuations likely in a sample that small. Still, at a signal-to-background level of order 1/10 one is concerned
how reliable and stable a signal can be that results from subtracting two almost equal large numbers. Even by careful inspection of raw mass spectra by shape, or by magnitude, it would be quite hard to distinguish above 200 MeV/$`c^2`$ the signal from background.
Because of cuts and kinematics, it so happens that the combinatorial background spectrum peaks around $`m2p_t^{cut}/c`$= 400 MeV/$`c^2`$ and is surprisingly similar in shape to the low-mass enhancement.
This raises the question, whether combinatorial background has been insufficiently subtracted. Aside from careful studies of analysis cuts, rejection cuts, the determination of the reconstruction efficiency as a function of centrality, we like to address this question in a quantitative way. A most likely cause leading to wrong subtraction of background is an undetected asymmetry in reconstruction efficiencies, i.e. for like-sign as compared to unlike-sign pairs. It is shown in Appendix C that a 5$`\%`$ asymmetry is required to fake an apparent enhancement factor of three over the hadronic sources, assuming the S/B= 1/13 situation of the โ96 analysis. However, this asymmetry was measured to be less than 1$`\%`$, with confidence limit of 90$`\%`$ lenkeit-phd , as reported in sect. 3.6.
If the low-mass enhancement is faked by leakage of some amount of combinatorial background into the spectrum of signal pairs, the enhancement should increase with the amount added. Figure 32, summarising all CERES pair analyses for 158 GeV/n Pb-Au, does not show that. Rather, we see a pair signal which is, in view of the errors, surprisingly stable despite very large variations in background level.
### 6.6 Systematic errors
Estimates of systematic errors in the mass-integrated, normalised yield of โDalitzโ pairs<sup>29</sup><sup>29</sup>29Note that the use of calligraphic $`๐ฉ`$ for normalised pair yields is restricted to this section.,
$$๐ฉ(\mathrm{`}\mathrm{Dalitz}^{})=_0^{0.2}\mathrm{dm}\mathrm{dN}_{\mathrm{ee}}(\mathrm{m})/\mathrm{dm}/\mathrm{N}_{\mathrm{ch}},$$
(36)
is given in Table 4. The definition of the โDalitzโ sample includes the opening angle cut of 35 mrad and the $`p_t`$ cut at 200 MeV/$`c`$. The largest contributions of about 10$`\%`$ each arise from detector efficiencies and the triggered centrality $`N_{ch}`$, while the uncertainties due to matching and rejection cuts account for about 7$`\%`$. The resulting total of 18$`\%`$<sup>30</sup><sup>30</sup>30If errors from both tracks would be correlated, they would add to a total of 24$`\%`$ instead. is larger than the rms deviation of Dalitz pair yields from all our analyses which amounts to 12$`\%`$.
The normalised yields of signal or open pairs
$$๐ฉ(\mathrm{`}\mathrm{Open}^{})=_{0.2}^{1.1}\mathrm{dm}\mathrm{dN}_{\mathrm{ee}}(\mathrm{m})/\mathrm{dm}/\mathrm{N}_{\mathrm{ch}}$$
(37)
plotted in Fig. 32 display a relative spread of 24$`\%`$, twice that of the Dalitz sample and now well above the systematic Dalitz-sample error. We presume that the larger systematic uncertainties in open pair yields are caused by the large combinatorial background. We take the relative sample error of 24$`\%`$ in number of normalised open pairs as a reliable measure of the magnitude of systematic uncertainties, arguing that they were obtained in rather independent efforts and by using diverse strategies.
## 7 Final Results and Comparison to Hadronic Sources
### 7.1 Inclusive mass spectra
The invariant mass spectrum of $`e^+e^{}`$ pairs produced in 158 GeV/n Pb-Au collisions as obtained from a weighted average of the individual data sets, henceforth called โunified mass spectrumโ, is shown in Fig. 33. The squares of the inverse relative statistical errors have been used as weights. Plotted is the differential yield per event corrected for pair efficiency and normalised to the number of charged particles in the acceptance, $`N_{ch}133`$, which corresponds to the rapidity density $`dN_{ch}/d\eta 245`$ and the most central 28$`\%`$ of the reaction cross section.
The mass spectrum is compared to the expectation of hadron decays modelled with the updated GENESIS code. The resonance structure of the light vector mesons is hardly visible in the measured spectrum, and within statistical errors, the data points are compatible with a smooth curve. Note that the resonances were clearly visible in the p-Be and p-Au invariant mass spectra(Fig. 1) despite a comparatively poor resolution of 9$`\%`$ at $`\rho /\omega `$. Note also that the $`\rho /\omega `$ resonance seems to become visible above the continuum in the mass spectrum selected for $`p_t^{ee}>`$ 500 MeV/$`c^2`$ of Fig. 37, to be discussed below.
In the $`\varphi `$ region, the expected mass resolution of 7$`\%`$ being potentially sufficient, statistics hampers further conclusions. We nevertheless quote the ratio of the observed yield (in the mass region 0.95$`<m<`$ 1.1 GeV/$`c^2`$) relative to the cocktail which is dominated by the statistical-model result for the $`\varphi `$ in this region. We obtain a value of 0.68$`\pm `$ 0.28. With one standard deviation below unity, the statistical accuracy is insufficient to settle the pending controversy on the $`\varphi `$ jouan-falco98 ; puehlhofer98 .
### 7.2 Enhancement factors
All CERES results for Pb-Au collisions show pair yields in the $`\pi ^{}`$-Dalitz region which are in good agreement with predictions from known hadron decays. For masses above 200 MeV/$`c^2`$, however, the data overshoot the expectation from hadron decays significantly. The largest enhancement over the hadronic cocktail is in the mass range between 400 and 600 MeV/$`c^2`$ where it reaches a magnitude of six.
Integrating the normalised yields up to 200 MeV/$`c^2`$, for the fraction dominated by $`\pi ^{}`$-Dalitz pairs (A), and from 200 MeV/$`c^2`$ upward for open pairs (B), we obtain
$`N_{ee}/N_{ch}`$ $`=`$
$`\{\begin{array}{cc}(8.52\pm 0.20[stat.]\pm 1.54[syst.])\times 10^6\hfill & (A)\hfill \\ (5.25\pm 0.43[stat.]\pm 1.26[syst.])\times 10^6\hfill & (B)\hfill \end{array}.`$ (40)
We have quoted the statistical and the systematic errors of 18$`\%`$ and 24$`\%`$, respectively.
The ratios of the measured data to the integrated yields of the decay cocktail (given by eqn. (5.33)) for the two mass regions are
$`={\displaystyle \frac{N_{ee}/N_{ch}}{N_{ee}/N_{ch}|_{decays}}}=`$
$`\{\begin{array}{cc}0.92\pm 0.02[stat]\pm 0.17[syst]\pm 0.07[decays]\hfill & (A)\hfill \\ 2.31\pm 0.19[stat]\pm 0.55[syst]\pm 0.69[decays]\hfill & (B).\hfill \end{array},`$ (43)
where our estimate of the systematic error in the decay cocktail is given separately. For the low-mass continuum region 200 $`m`$ 600 MeV/$`c^2`$, the enhancement is even larger,
$$=2.73\pm 0.25[stat]\pm 0.65[syst]\pm 0.82[decays].$$
(44)
### 7.3 Centrality dependence
The enhancement factors $``$ are plotted vs charged-particle density for three mass regions in Fig. 34. The enhancement for the mass region 200$`m`$ 600 MeV/$`c^2`$ reaches
about 4 at the most central collisions, and it is seen to rise about linearly with charged-particle density; a straight-line fit to the five data points gives a slope value deviating from zero by 4 standard deviations. This establishes that the integrated pair yield itself has a stronger-than-linear dependence on charged particle density. (By construction, the cocktail yield per $`N_{ch}`$ does not depend on $`N_{ch}`$). This enhancement extends to the resonance region, although with reduced significance and smaller values of the enhancement factor. Unfortunately, statistical uncertainties, do not allow to confirm or refute interesting details in the growth of the normalized yield with $`N_{ch}`$, like a threshold effect or a saturation behaviour.
To see how the enhancement-typical spectral shape evolves with increasing centrality, unified mass spectra for different ranges in $`N_{ch}`$ are shown in Fig. 35. The series of spectra clearly demonstrates that the excess over the cocktail for all four spectra occurs between the two-pion threshold and the $`\rho /\omega `$-resonance position, increasing with centrality. The spectra, being statistically independent samples, corroborate that the largest enhancement is around 500 MeV/$`c^2`$, well below the $`\rho /\omega `$ position. With respect to the central issue of in-medium modifications, one should be careful not to establish a direct link to the enhancement factor as we defined it. The reason is that pion annihilation per se, i.e. with a vacuum $`\rho `$, proceeding in the hot fireball is already a large, if not the dominant, contribution to the enhancement factor.
### 7.4 Pair transverse momentum
The Lorentz-invariant transverse momentum distributions of produced electron pairs, observed first in the 1995 data analysis voigt-phd , are shown in Fig. 36 for the combined 95/96 data for the $`\pi ^{}`$-Dalitz region and for open
pairs. At low pair $`p_t`$, denoted by $`p_t^{ee}`$, the yield is suppressed due to the single-track $`p_t`$ cut at 200 MeV/c. As to be expected, there is good agreement between data and hadronic cocktail in the Dalitz region. However, the information from the pair $`p_t`$ distribution in the open-pair mass range is striking: the enhancement grows towards small pair $`p_t^{ee}`$ despite the $`p_t`$200 MeV/$`c`$ condition on single electron tracks. The surplus originates from decays of virtual photons within the fireball that are created favourably at rest. The enhancement above 500 MeV/$`c`$ pair transverse momentum is considerably reduced. Whether this observation trivially reflects the annihilation kinematics in a thermal pion gas, or contains information about modified hadron properties, will be discussed in sect. 8.
Conversely, the size of the average transverse pair momentum has a remarkable influence on the shape of the inclusive mass spectra. We observe in Fig. 37 that the measured yield dramatically overshoots the hadron decay contributions when low pair momenta are selected. The enhancement in the mass range 500-700 MeV/$`c^2`$ reaches locally an order of magnitude.
In contrast, for larger transverse pair momenta, the measured yield comes pretty close in shape and magnitude to the cocktail expectations; the resonance region becomes visible as it gains in yield, even on absolute scale, while the previously amplified continuum at 400$`m`$600 MeV/$`c^2`$ is being deflated.
## 8 Physics discussion
The two CERES runs of 1995 and 1996, combined to the largest statistics sample of 158 GeV/n Pb-Au collisions taken so far, corroborate the enhancement in low-mass electron-pair production over that from hadron decays originally observed in 200 GeV/n S-Au collisions by CERES prl1995 and in S-W collisions by Helios-3 masera1995 . Two additional CERES Pb-Au runs have been performed since then with the new TPC and improved mass resolution, one at the reduced energy of 40 GeV/n adamova-ee03 ; appelshauser02 ; wessels02 the results of which are shortly addressed below, the other at 158 GeV/n marin04 with quantitatively consistent results.
In the region of the $`\pi ^{}`$-Dalitz decay, $`m`$ 200 MeV/$`c^2`$, the yield of the unified sample agrees with the cocktail within the estimated systematic uncertainties. In the mass range upward of 200 MeV/$`c^2`$, the normalised pair yield of the unified data analysis significantly exceeds the yield from hadron decays per charged particle, and the enhancement factor is given together with our estimates of statistical and systematic errors in sect. 7.2.
The systematic errors of the decay cocktail were discussed in sect. 4. To address the reliability of the generator in physics terms, the situation has remarkably improved since the practice of scaling yields up from p-p to Pb-Au collisions was abandoned in favour of using particle ratios from statistical model systematics based on Pb-beam data itself. Still, we shortly recall here some particle ratios that have become subject to speculations of being grossly enhanced (i.e. by factors more than two) in nucleus-nucleus compared to p-p reactions, even beyond the statistical-model systematics.
First, the $`\eta /\pi ^{}`$ ratio received interest from the fact that the $`\eta `$-Dalitz decay is the most important single component of the hadronic cocktail being suspect of enhanced production. It was argued drees-eta , however, that an $`\eta /\pi ^{}`$ ratio sufficiently large to explain the electron pair enhancement had not been observed in the photon yield measured by WA80 wa80 .
In the region below the $`\rho /\omega `$, the gap between cocktail and data might be filled by raising the $`\omega `$-Dalitz decay contribution as has been proposed by V. Koch some time ago koch99 . We hold
against that there is no indication otherwise for a strongly enhanced $`\omega `$ production. In addition, a boost in $`\omega `$-Dalitz production is limited to an upper margin set by the data of Fig. 33 to the direct decay $`\omega e^+e^{}`$. Eventually, there is absolutely no reason to believe that open-charm production in nucleus-nucleus collisions should be as enormously enhanced as to explain the low-mass enhancement open-charm98 .
We return to state that the enhanced production of low-mass electron pairs cannot be attributed to decays of produced hadrons. The excess has to originate from processes which are active during the lifetime of the fireball, i.e. between the onset of hadronisation and kinetic freeze-out, if of hadronic origin, and before hadronisation, if created during the plasma phase. Present theoretical studies allocate only a very minor fraction to the partonic part which reflects the supposedly small 4-volumes of deconfined matter at SPS energies.
After having tried to give an overview of theoretical models of dilepton production in the introduction, this discussion will be guided by a few representative theoretical models: the spectral function approach rapp-wambach2000 , the dropping-mass scenario b-r03 , and pion annihilation with a vacuum $`\rho `$. In all calculations to be shown, the same fireball model has been used to describe the space-time evolution rapp-special .
Which are the experimental signatures of pion annihilation, the process most widely ascribed to take place in the dense hadronic fireball ? As for any binary process, the annihilation rate is expected to scale with the squared density of the particles annihilating in the fireball (of course, the argument applies as well to $`q\overline{q}`$ annihilation). Unfortunately, the centrality dependence of the dilepton yield is a topic barely addressed by full transport calculations, except cassing-brat1999 .
We take the observed stronger-than-linear scaling of the pair yield (in the mass region of the strongest enhancement) with $`N_{ch}`$ as strong evidence in support of the two-body annihilation reaction. This deserves some words of justification. An increase in $`N_{ch}`$ signals a larger pion density only to the extent that it is not compensated by an associated increase in volume, maybe even in lifetime, of the fireball. Therefore, pion annihilation does not necessarily go along with quadratic scaling in $`N_{ch}`$. Conversely, however, an observed stronger-than-linear scaling of the pair yield with $`N_{ch}`$ is sufficient reason to infer a binary reaction at work, i.e. strongly suggesting pion annihilation in hadronic matter, or $`q\overline{q}`$ annihilation in the plasma phase.
As other annihilation processes in a thermal medium, $`\pi \pi `$ annihilation takes place favourably at small relative momentum of the constituents. This behaviour was indeed encountered already in the pair transverse momentum spectra of Fig 36. From Fig 38 shown here, we learn in addition that pion annihilation with a vacuum $`\rho `$ is hard to distinguish, by yield and shape of its $`p_t^{ee}`$ spectrum, from the medium-modified spectral function and the dropping-mass approaches.
The invariant mass spectra tell more about the physics processes involved. In Fig. 39, the measured invariant mass spectrum is compared to the model calculations. It is evident at first sight that pion annihilation with a vacuum $`\rho `$ (thick dashed line) does not describe the shape of the spectrum; rather the calculations overshoot the data at the nominal $`\rho `$ position by about a factor of 2 and under-predict the data in the continuum region by about a factor of 3 โ yet, the integral yield comes out about right.
In-medium modifications produce a dramatic change: both the calculations with an in-medium modified $`\rho `$ spectral function (thick solid line), and with a dropping in-medium $`\rho `$ mass (thick dashed-dotted line) describe very well the marked increase of the continuum yield around 500 MeV/$`c^2`$ as well as the depletion at the vacuum $`\rho /\omega `$ position; the differences among the competing approaches again are rather subtle. We remark that calculations adopting the chiral reduction formalism approach give very similar results, except that the depletion at the vacuum $`\rho /\omega `$ position is absent steele-zahed99 .
Before discussing medium modifications in some detail, let us examine the contributions of $`\rho `$ and $`\omega `$ to the cocktail (for $`m>`$400 MeV/$`c^2`$, say), and more speculative, to in-medium pair production. While the $`\omega `$ with its direct decay and part of its Dalitz decay clearly dominates over the $`\rho `$ in the cocktail (see Fig. 33), it is the $`\rho `$ which provides essentially all of the enhancement. One might wonder by which mechanism such drastic change should be accomplished.
Some estimates based on vacuum properties are collected in the appendix. The yield from mesons formed initially by hadronisation is topped by the $`\omega `$ by its larger electro-magnetic branching ratio. In contrast, by
$`e^+e^{}`$ decays from the in-medium meson population, the $`\rho `$ wins by a large margin due to its much larger $`\pi \pi `$ width. Altogether, we find that the $`\rho `$/$`\omega `$ ratio of pair yields from secondary, in-medium generated mesons, is by orders of magnitude larger than for initially produced mesons.
This digression illustrates from another point of view what we know already: the exceptional potential of the $`\rho `$ propagator to dominate electron-pair production quite unrelated to its share in the cocktail โ and that there is no way to describe the surplus of electron pairs other than by direct or thermal radiation out of the fireball, i.e. from mesons which are regeneratively produced in $`\pi \pi `$ annihilation. The chances are feeble to observe such radiation from the $`\omega `$, primarily because of its extremely weak coupling to the hadronic medium. Medium modifications, however, might be observed from primary $`\omega `$ and $`\varphi `$ mesons by the fraction that decays within the lifetime of the fireball.
We return to compare theory to our data. Figure 38 shows that the enhancement at low pair $`p_t^{ee}`$ over the cocktail which is seen in the continuum region and to a lesser extent also in the resonance region is a feature present in all model calculations, with minor differences only between a dropping, broadening, or vacuum $`\rho `$ propagator. We meet here the governing features of pion annihilation, rather than of medium modifications proper.
It is the mass spectrum which uncovers the characteristic of medium modifications as distinct from pion annihilation with vacuum $`\rho `$ propagator as seen from the comparison of the data with the three modell calculations in Fig. 39. The differences between the two models incorporating medium modifications, however, are rather subtle.
Let us inspect the model calculations of $`p_t^{ee}`$-selected mass spectra compared in Fig. 40 to the data. The drastic impact on the shape of the mass spectrum the selection of low $`p_t^{ee}`$ has, is also present in the two model calculations with modified $`\rho `$. In these models, the effect maybe somewhat weaker, yet locally the enhancement over the cocktail reaches 10 (see Fig. 37 for the complete cocktail). The vacuum-$`\rho `$ calculation is only weakly affected by the $`p_t^{ee}`$-selection. For larger $`p_t^{ee}`$, data and model calculations come much closer to the decay cocktail.
The processes causing the in-medium changes of the $`\rho `$ spectral function, or the dropping mass of the $`\rho `$, clearly also favour low pair $`p_t^{ee}`$. Such behaviour would arise most naturally in the dropping-mass scenario from the Boltzmann (or Bose) factor producing the largest gain for small in-medium masses at vanishing 3-momentum (see App. A). For the spectral function approach, the observation had its impact to install the $`s`$-wave N(1520) $`\rho `$-nucleon resonance peters98 ; rapp-wambach2000 as the moving agent in place of the $`p`$-wave N(1720) resonance which had pioneered the importance of $`\rho N`$ resonances for medium modifications friman-pirner97 ; this change also met requirements by photo absorption data to soften the form factor urban98 .
The effective downward shift of strength to lower masses in the melting-$`\rho `$ treatment rapp-chanfray-w1997 is largely due to strong meson-baryon coupling, and most approaches agree to its importance for generating in-medium effects rapp-chanfray-w1996 ; klingl-weise1996 ; steele-zahed99 ; a finite nucleon chemical potential is required also for some meson-meson mixing effects to take place in approaching chiral symmetry restoration chan-del-erik98 ; theo01 . Only very small effects of baryon density have been reported for UrQMD transport calculations bleicher00 .<sup>31</sup><sup>31</sup>31This conclusion rests on the (false) premise that the data have been satisfactorily described without medium modifications using boosted $`\omega `$ and $`\eta `$ Dalitz decays. We have remarked that the CERES run at reduced SPS energy of 40 GeV/n observed an even larger enhancement as in 158 GeV/n collisions adamova-ee03 reaffirming the conclusion rapp-nassau that the increase in baryon density
seems to have won over the reduced pion density, or lower temperature. As it is the total baryon density that matters - vector mesons interact symmetrically with baryons and anti-baryons rapp-prc01 \- the situation at RHIC energies will not be greatly different from top SPS energy (despite vanishing net baryon density).
In concluding this review of selected theory descriptions of the CERES Pb-Au dilepton data, we like to add that the spectral function approach had also other successes. Within the same framework, the intermediate-mass enhancement observed by NA50 abreu00 has been successfully described as thermal radiation with a 30$`\%`$ share of the quark gluon plasma rapp-shuryak00 ; there was no need to invoke open-charm enhancement. This finding may be seen as the first glimpse of light in the long search for thermal $`\overline{q}q`$ radiation from the quark-gluon plasmashu80 ; mclerran85 .
Still within the same framework, the $`p_t`$ spectra of photons in 158 GeV/n Pb-Pb collisions measured by
WA98 aggarwal02 have been reproduced; up to transverse momenta of about 1.5 GeV/$`c`$, thermal emission from the expanding hadronic fireball has been found to dominate with photons mainly of baryonic origin turbide04 , fully consistent with the closely related calculations that describe the low-mass dilepton excess observed by CERES.
## 9 Conclusion and Outlook
The analysis of the large unified data sample has substantiated the earlier finding of a strong source of continuum electron pairs which contributes to the invariant mass spectrum beyond the decays of produced mesons, most strongly around 500 MeV/$`c^2`$. There is ample evidence that we observe dilepton radiation from the interior of the hadronic fireball in which pion annihilation mediated by the $`\rho `$ propagator plays a major role. The excess yield rises significantly steeper than linearly with charged-particle density, consistent with the binary annihilation process. Another piece of circumstantial evidence for $`\pi \pi `$ annihilation is delivered by the invariant pair transverse-momentum spectra for the continuum pairs of masses between 200 and 600 MeV/$`c^2`$: the dramatic enhancement over the cocktail occurs at very low $`p_t^{ee}`$.
Full calculations with vacuum $`\rho `$ describe the measured yield about correctly, but fail to account for the characteristic shape of the spectrum of excess dileptons. The $`\rho `$ propagator is manifestly modified in the medium which is well described by two theories incorporating medium modifications of the $`\varrho `$, which are, however, very different in concept: while the Brown-Rho scaling hypothesis explicitly refers to restoration of chiral symmetry, the many-body spectral-function approach of Rapp and Wambach, although tracing some of the induced mixings, does not have chiral symmetry restoration as a ruling concept. That it may very well be implicitly included is inferred from the fact that the hadronic dilepton rates extrapolated up to $`T_c`$ (โbottom upโ) come out very similar to the (โtop downโ) extrapolated perturbative QGP rates rw98 . Both theories also give a good description of the mass spectra for selected ranges of pair transverse momentum $`p_t^{ee}`$, where an additional preference of the mechanism generating the medium modifications has shown up.
Three unsolved issues remain. To the present data accuracy, it is not possible to decide between one or the other of the two competing theories so that the role of chiral symmetry restoration for in-medium modifications remains unclear.
Patience seems also advised in localising the source of medium-modified electron-pair production within the phase diagram. On one hand, preformation of vector and axial-vector correlator strengths in the non-perturbative plasma might influence dilepton production across the
phase boundary lee98 ; jaikumar02 . On the late end of the time scale, the impact of a growing pion chemical potential for dilepton production might not be entirely settled yet.
An issue widely overlooked is whether the hadronic fireball is โboilingโ long enough to radiate sufficient amounts of dileptons. The time spent by the system between chemical and kinetic freeze-out came under scrutiny on the basis of recent pion interferometric data adamova03-hbt which indicated only a 30$`\%`$ change in volume. If, in addition, chemical freeze-out should occur essentially at the phase boundary between hadronic and quark matter, as suggested by the asymptotic statistical-model result of $`T_{chem}`$ 170 MeV at higher energies pbm04 , the purely hadronic origin of the low-mass dilepton enhancement might have to be negotiated again.
We look out to further progress that can only be expected from radically better data, with greatly improved statistics and less combinatorial background, but also with improved mass resolution. This will not be easy. But a first step in this direction is the CERES 2000 run with the new Time Projection Chamber; preliminary data have been presented very recently marin04 .
###### Acknowledgements.
Acknowledgement
We acknowledge the good performance of the CERN PS and SPS accelerators and the excellent support for the central data recording from the IT division. We are grateful to D.A. Pinelli at BNL and O. Runolfsson at CERN for their delicate work in the assembly of motherboards for SIDCโs. We acknowledge the support by Deutsches Bundesministerium fรผr Bildung, Wissenschaft, Forschung und Technologie (BMBF), the U.S. Department of Energy, the Minerva Foundation, the Israeli Science Foundation, and the German Israeli Foundation for Scientific Research and Development.
## Appendix A Appendix $`\rho e^+e^{}`$ decay rate
The thermal emission rate of dielectrons from two-body decay of rho mesons has been worked out from Ref. song by B. Friman and J. Knoll frimann-knoll00 as follows:
$`{\displaystyle \frac{dR}{dMd^{\mathrm{\hspace{0.17em}3}}\stackrel{}{q}}}={\displaystyle \frac{\alpha ^2m_\rho ^4}{3(2\pi )^4}}{\displaystyle \frac{(14m_\pi ^2/M^2)^{3/2}}{(M^2m_\rho ^2)^2+M^2\mathrm{\Gamma }_{tot}^{0^2}}}`$ $`\times `$
$`\left\{e^{\sqrt{M^2+\stackrel{}{q}^2}/T}{\displaystyle \frac{M}{\sqrt{M^2+\stackrel{}{q}^2}}}\right\}.`$ (45)
Here, q is the 3-momentum of the rho meson. The new expression differs from the one used in the previous GENESIS code by the Boltzmann-type phase space factor in the curled parentheses being explicitly momentum dependent. To obtain the final invariant mass distribution, the differential rate $`dR/dMd^3\text{q}`$ has to be integrated over 3-momentum q. The final result used in the 2003 version of New GENESIS genesis-new for comparison to the data reads
$$\frac{dR}{dM}=\frac{\alpha ^2m_\rho ^4}{3(2\pi )^4}\frac{(14m_\pi ^2/M^2)^{3/2}}{(M^2m_\rho ^2)^2+M^2\mathrm{\Gamma }_{tot}^{0^2}}(2\pi MT)^{3/2}e^{M/T}.$$
(46)
## Appendix B Estimate of $`\rho `$/$`\omega `$ ratios
We estimate the $`\rho /\omega `$ ratio of electron-pair yields from initially produced mesons and from those produced in a hadronic fireball by using simplified order-of magnitude estimates expressed solely by the (vacuum) particle properties pdg2004 .
Primary mesons are produced during hadronisation. They decay into electron pairs with a fraction given by the electro-magnetic (e.m.) branching ratio $`B_{ee}=\mathrm{\Gamma }_{ee}/\mathrm{\Gamma }`$. As in p-p collisions neutral-meson-pBe , we assume equal primary populations of $`\rho `$ and $`\omega `$. To the mass range $`m`$ 400 MeV/$`c^2`$, the $`\omega `$ contributes by its direct decay and by about 15$`\%`$ of its Dalitz decay. In this mass range, the ratio of the number of electron pairs emitted by the initial population of $`\rho `$โs and $`\omega `$โs is
$$\left(\frac{Y_\rho }{Y_\omega }\right)_{prim}=\frac{B_{ee}(\rho )}{B_{ee}(\omega )}0.3.$$
(47)
The $`\rho `$ looses in this comparison of e.m. branching ratios due its so much larger total width of $`\mathrm{\Gamma }(\rho )`$= 150 MeV, compared to $`\mathrm{\Gamma }(\omega )`$= 8.4 MeV.
We are mostly interested in mesons that decay into electron pairs within the lifetime of the fireball since the others cannot probe medium modifications. Besides primary produced $`\rho `$ mesons, an additional source of in-medium decays are from
Secondary mesons. These are continuously produced in the fireball by two-pion annihilation, and most of the time disintegrate back into two pions as expressed by eqn. (1.1). We like to present here a simple estimate of the relative in medium-pair production yields from $`\rho `$ and $`\omega `$ which neglects all finer details of the reaction dynamics by assuming that the ratio of rates is given as the ratio of the 2-pion annihilation cross sections, which are estimated from detailed balance, times the ratio of the e.m. decay widths. The two vector mesons differ in an essential manner in strength of coupling to the hadronic medium: while the $`\pi \pi `$ channel exhausts the large width of the $`\rho `$, $`B_{\pi \pi }(\rho )`$ 1, the $`\omega `$ (with larger branching into $`3\pi `$) is extremely weakly coupled to the $`\pi \pi `$ channel, $`B_{\pi \pi }(\omega )`$ 2 $`\%`$. With $`Y_{med}\mathrm{\Gamma }_{\pi \pi }\mathrm{\Gamma }_{ee}`$,
$$\left(\frac{Y_\rho }{Y_\omega }\right)_{med}4\times 10^3.$$
(48)
The estimate demonstrates that by in-medium produced electron pairs, the $`\omega `$ meson is completely outnumbered by the $`\rho `$, unless its in-medium properties should drastically be changed.
## Appendix C Instrumental asymmetries as source of enhancement?
Let us assume there is an asymmetry in reconstruction efficiencies for like-sign pairs (L) as compared to unlike-sign pairs (U) which would affect the signal in a most direct way. Asymmetries of this kind might be expected since rings tend to end up in a more or less dense environment in RICH-2 depending on whether they are deflected away from each other, or not.
Writing the pair efficiencies as $`\epsilon _L=\epsilon _{},\epsilon _U=\epsilon _{}(1+\delta )`$, the signal is given by
$$S(U(1\delta )L)/\epsilon =S_{}\delta U/\epsilon .$$
(49)
To be definite, let us assume a signal $`S_{}`$ three times the hadronic background, i.e. an enhancement factor of three. The enhancement will be gone if the asymmetry term reduces the signal to one third of $`S_{}`$. The corresponding asymmetry parameter is
$$\delta =2(UL)/3U\frac{2}{3}(UL)/L.$$
(50)
For a signal-to-background ratio $`(UL)/L=S/B=`$ 1/13, we find that an asymmetry of $`\delta `$ 5.1$`\%`$ in efficiencies for like-sign and unlike-sign pairs is sufficient to fake an enhancement factor of 3.
|
warning/0506/astro-ph0506652.html
|
ar5iv
|
text
|
# The search for the host galaxy of the gamma-ray burst GRB 000214 Based on observations obtained at the ESO 3.6 m telescope under ESO programme 165.H-0464(I).
## 1 Introduction
GRB 000214 was detected by both the GRB monitor (GRBM) and the Wide Field Cameras (WFC) on board BeppoSAX on 14 February 2000, 01:01:01 UT (Piro (2000)). In the GRB monitor it exhibited a duration of $``$9 s, and a 40โ700 keV fluence of $``$1.4 $`\times `$ 10<sup>-5</sup> erg cm<sup>-2</sup>. In the WFC (2โ30 keV) the duration was $``$115 s and the fluence $``$1.0 $`\times `$ 10<sup>-6</sup> erg cm<sup>-2</sup> (Paolino et al. (2000)). Follow-up observations with the BeppoSAX Narrow-Field Instrument (NFI) began about 12 hr after the burst. A previously unknown X-ray fading point source 1SAX J1854.4-6627, was detected in the MECS and LECS field of view at a position of R.A. (J2000)=18<sup>h</sup>54<sup>m</sup>27.0<sup>s</sup>, Dec (J2000)=-662730<sup>โฒโฒ</sup> (error radius 50<sup>โฒโฒ</sup>) with a 2โ10 keV flux of 5 $`\times `$ 10<sup>-13</sup> erg cm<sup>-2</sup> s<sup>-1</sup> (Antonelli et al. 2000a ). Within the 50<sup>โฒโฒ</sup> radius of the NFI error circle, radio (Subrahmanyan et al. (2000)) and IR (Rhoads et al. Rhoa00 (2000)) observations did not find any variable source. An estimation of the redshift based on the Fe K$`\alpha `$ X-ray emission line yielded 0.37โ0.47 (Antonelli et al. 2000a , 2000b ; Kotake & Nagataki Kota01 (2001)).
Here we present optical observations of the GRB 000214 NFI error box in the UBVRI-bands in order to search for objects with photometric redshifts in the range 0.37โ0.47, which could be potential candidates for the GRB 000214 host galaxy. Throughout, we assume a cosmology where $`H_0=65`$ km s<sup>-1</sup> Mpc<sup>-1</sup>, $`\mathrm{\Omega }_\mathrm{\Lambda }=0`$ and $`\mathrm{\Omega }_M=1`$.
## 2 Observations
All observations were obtained with the 3.6 m ESO telescope at La Silla (Chile). The CCD used was a Loral 2048 $`\times `$ 2048 detector giving a 5.4 $`\times `$ 5.4 field of view. The observations were carried out in 2$`\times `$2 binning mode, yielding a pixel scale of 0.31<sup>โฒโฒ</sup>/pixel. Table 1 displays the observing log. The photometry performed to study the content of the NFI error box, is based on aperture photometry carried out using SExtractor (Bertin & Arnouts (1996)) to study the content of the NFI error box. The field was calibrated observing the Landolt star LTT 4816 (Landolt (1992)), at an airmass similar to that of the GRB. Table 2 shows the positions and magnitudes of the selected secondary standards present in the NFI field (see Fig. 1).
## 3 Results and discussion
48 objects located closer than $`1^{}`$ from the NFI position were detected in at least three optical bands (out of the five UBVRI filters). The magnitudes (and upper limits in the bands where no detection was possible) of these objects were used to feed the HyperZ code, yielding the photometric redshift, extinction ($`A_v`$), galaxy type and dominant stellar population age for each object (see Bolzonella et al. Bolz00 (2000) for more details on the HyperZ outputs). The photometric redshifts derived by HyperZ for GRB host galaxies have been tested in the past using a sample of 10 hosts with known spectroscopic redshifts, yielding excellent results (specially for GRB host galaxies classified as starbursts; see Table 2 of Christensen et al. 2004a ). For the construction of the HyperZ synthetic templates, we assumed a Miller & Scalo (Mill79 (1979)) initial mass function, and a small Magellanic cloud (SMC) extinction law (Prevot et al. Prev84 (1984)), typical of GRB hosts galaxies.
Table 3 provides the coordinates, magnitudes and photometric redshifts for our four best candidates. The photometric fluxes corresponding to our measurements have been obtained convolving the 3.6 m ESO filter transmittances with the Loral CCD, yielding the AB offsets (AB<sub>off</sub><sup>1</sup><sup>1</sup>1The AB offset is defined as AB$`{}_{off}{}^{}=2.5\mathrm{log}(F_\nu )48.60m_{vega}`$, where $`F_\nu `$ is the flux density measured in erg s<sup>-1</sup> cm<sup>-2</sup> Hz<sup>-1</sup>, and $`m_{vega}`$ is the magnitude in the Vega system.) given in Table 1.
Only object #1 shows a photometric redshift fully consistent, within $`1\sigma `$, with the 0.37โ0.47 redshift range, being the photometric redshift of candidate #2 just at $`1\sigma `$ from the X-ray redshift range lower limit. The two remaining objects (candidates #3 and #4) have photometric redshifts separated by $`2\sigma `$ from the X-ray redshift range upper limit.
However, we note that candidate #1 is formally outside of the 50<sup>โฒโฒ</sup> radius NFI error circle and object #3 is just on its edge (see Fig. 1). Both candidates are fully consistent with the IPN annulus so we decided not to discard them. Candidate #2 is well centered in the NFI error circle, but its photometric redshift is only marginally consistent (at $`1\sigma `$) with the X-ray redshift. Thus, inside the 90% confidence level NFI error box, no object has a photometric redshift fully consistent (within 1$`\sigma `$) with the 0.37โ0.47 X-ray redshift range.
An alternative possibility is that the host galaxy of GRB 000214 is indeed placed within the 50<sup>โฒโฒ</sup> radius NFI error circle, but it is fainter in three or more filters than the limits reported in Table 1. In this case no computation of photometric redshift is possible and the object would be automatically discarded in our analysis. A second alternative scenario is possible if the GRB 000214 host galaxy is detected in three o more filters, but it is located in the outskirts of the NFI error circle (i.e. on the tail of the probability distribution). This might still be the case for object #1, which is located only $`4.5^{\prime \prime }`$ out of the NFI error circle 90% boundary. The same conclusion stands for object #3 which is just on the border of the NFI error circle.
In Fig. 2 we display the R-band magnitudes compiled for 32 host galaxies (rhomboids) and our four candidates (squares), once they are corrected for foreground Galactic extinction (E(B$``$V)=0.061, Schlegel et al. Schl98 (1998)). The curves display the apparent R-band magnitude of a reference $`M_R^{}`$ galaxy when it is redshifted from $`z=0`$ to $`z=4`$. $`M_R^{}`$ represents the R-band absolute magnitude and determines the knee of the luminosity function, separating the intrinsically bright from the subluminous galaxies. We assumed a value of $`M_R^{}=20.29+5\mathrm{log}(H_0/100)`$ (Lin et al. Lin96 (1996)) estimated adopting an Einstein-de Sitter Universe (as in the present study). In order to perform the K-correction (Oke & Sandage Oke68 (1968)) the spectrum of the $`M_R^{}`$ galaxy has been assumed to be a power law ($`F_\nu \nu ^\beta `$) with the spectral index ranging from $`\beta =0`$ (lower dotted line) to $`\beta =2`$ (upper dashed line). This spectral index range generates a broad set of colours, ($`0.3<`$B-R$`<1.2`$) accounting for most of the galaxies seen in the Hubble Deep Field (Williams et al. Will96 (1996)). As it is shown, the four candidates seem to be subluminous galaxies, tending to be above the dotted curves. Thus, our candidates show apparent and absolute magnitudes similar to GRB host galaxies at similar redshifts ($`z0.5`$).
The four objects were classified as starbursts by HyperZ, consistent with the hostsโ photometric spectral energy distributions (SEDs) studied to date (Gorosabel et al. 2003a , 2003b , Goro05 (2005); Christensen et al. 2004a , 2004b ). The intrinsic extinction values of the host candidates range from $`A_\mathrm{v}=0.0`$ (objects #3 and #4) to $`A_\mathrm{v}=2.85`$ (object #2), while object #1 has an intermediate $`A_\mathrm{v}`$ value of $`1.41`$ (see Fig. 3).
Three of our four candidates (#1, #2, and #4) show compact appearance, at least under our seeing conditions (see Table 1), displaying full width half maxima (FWHM) similar to other stellar objects present in the GRB field. Object #3 is slightly extended in the images having the best seeing, so it corresponds very likely to a galaxy. The potential stellar nature of objects #1, #2, and #4 has been checked using the CLASS\_STAR keyword given by SExtractor. Objects #1, #2, and #4 shows CLASS\_STAR values below the mode of the CLASS\_STAR distribution, specially in the $`B`$-band filter displaying CLASS\_STAR $`<0.8`$. Systematically the object with the largest CLASS\_STAR value is #2. Therefore the four objects correspond very likely to galaxies, may be with the exception of object #2 which stellar nature can not be completely excluded.
One potential problem might be the presence of Active Galactic Nuclei (AGN) in our NFI error box, for which HyperZ (not accounting for emission due to a nebular component or/and a central massive compact source) would not be an appropriate tool to fit our SEDs. The expected number of AGNs brighter than z=22.5 (comparable to R=23.4, the $`3\sigma `$ limit of our R-band image) closer than $`1^{}`$ from the NFI position is $`1`$ (Treister et al. Trei04 (2004)). Thus, for the sample of 48 objects the AGN contamination is expected to be only $`2\%`$.
Even for objects #1 and #2, which show the highest extinction among our four candidates ($`A_v=1.41`$ and $`A_v=2.85`$ magnitudes), their low photometric redshifts ($`z=0.49`$ and $`z=0.32`$) do not imply a high near-infrared restframe extinction. In particular the K-band limits imposed by Rhoads et al. (Rhoa00 (2000)) would be only affected by intrinsic host extinctions of $`A_{14400\mathrm{\AA }}=0.3`$ and $`A_{16300\mathrm{\AA }}=0.4`$, for objects #1 and #2 respectively (assuming a SMC extinction law Prevot et al. Prev84 (1984)). For objects #3 and #4, the K-band limit is even less extincted ($`A_v0`$). Therefore, if the host were one of our four objects, then it would be difficult to explain the K-band non detection as an effect of the global intrinsic host extinction.
In fact, neither De Pasquale et al. (DePa03 (2003)) nor Jakobsson et al. (Jako04 (2004)) classified GRB 000214 as an intrinsically dark GRB. According to these authors the K-band and X-ray observations reported for this GRB are not fast/deep enough to constrain the physical parameters determining the SED.
## 4 Conclusions
We presented here the result of UBVRI photometry for all objects down to R=23.4 inside the GRB 000214 error box. After photometric reduction of the images and modeling of synthetic SEDs, we have found no object within the 50<sup>โฒโฒ</sup> radius NFI error circle fully consistent with the redshift inferred from the X-ray spectrum.
However, we report four host galaxy candidates with photometric redshifts consistent within $`2\sigma `$ with the 0.37โ0.47 X-ray redshift range, so they are still statistically acceptable. Three of them are located inside (or just on the border of) the NFI error box, although they do not show photometric redshifts consistent (within $`1\sigma `$) with the X-ray spectroscopic redshift range. A fourth R=21.1 mag object, shows a photometric redshift of $`z=0.49_{0.07}^{+0.05}`$, fully consistent within $`1\sigma `$. We note that this candidate, although consistent with the IPN annulus, is slightly ($`4.5^{\prime \prime }`$) outside of the 90% NFI error circle.
We can not discard that an object fainter (in three or more bands) than our UBVRI-band detection limits might be the actual GRB 000214 host galaxy. Further spectrophotometric observations of our four objects would definitively shed light on the reliability of the proposed candidates.
###### Acknowledgements.
The data reported in the present paper were taken under the ESO programme 165.H-0464(I). We are grateful to the ESO staff at La Silla for performing the observations in the context of GRACEโs host galaxy programme. S. Guziy acknowledges the receipt of a fellowship grant from Spainโs Ministerio de Ciencia y Tecnologรญa (ref. SB 2003-0236), and the hospitality at IAA-CSIC, where this research was carried out. J. Gorosabel acknowledges the receipt of a Ramรณn y Cajal Fellowship from Spainโs Ministerio de Ciencia y Tecnologรญa. This research was partially funded by the Spanish ESP2002-04124-C03-01 and AYA2004-01515 programmes (including FEDER Funds). We thank N. Masetti, E. Pian, and C. Kouveliotou for useful conversations. We acknowledge our anonymous referee for fruitful and constructive comments.
|
warning/0506/cond-mat0506551.html
|
ar5iv
|
text
|
# Quantized Transport in Two-Dimensional Spin-Ordered Structures
## 1 Introduction
A fascinating and relatively common problem in solid-state physics is the motion of electrons in a lattice structure where localized spins are present, possibly in an organised form. The scattering of the electrons by the spins represents a complex physics problem for the theoretical investigator, especially when (like it seems to happen for the high-$`T_c`$ cuprate superconductors) the spin and charge degrees of freedom are attached to the same particles โ a situation which we do not consider. In a simplified picture we assume first of all that the spins are localised and interacting, but that the mobile electrons do not interact with each other (or interact weakly and give rise to independent quasiparticles) and are separate entities from the electrons that produce the magnetic ions. In this case the individual electrons (or quasiparticles) experience the localized spins as an effective local magnetic flux (Mรผller-Hartmann and Dagotto, 1996). This flux produces a Berry (or Peierls) phase in the hopping terms of the electronsโ Hamiltonian. In this paper we shall study a situation with a locally staggered flux, created by plaquettes of ordered spins, where the global flux is zero. This case is realized, for instance, in a Kagome lattice (Ohgushi, Murakami and Nagaosa, 2000), a two dimensional (2D) lattice consisting of triangles and hexagons of the same interatomic distance and that can be viewed as a triangular Bravais lattice with a three-point atomic basis forming an equilateral triangle of size half that of the triangular lattice parameterโs. In Fig.1 we remind the reader of the Kagome lattice structure.
This type of lattice may have experimental relevance in the planes of pyrochlore compounds (Ohgushi, Murakami and Nagaosa, 2000, Ramirez, 1994, Harris and Zinkin, 1996) and the transport properties we describe may be appropriate for such materials. Ferromagnetic pyrochlore crystals of the type R<sub>2</sub>Mo<sub>2</sub>O<sub>7</sub> (R=Nd, Sm or Gd) have revealed interesting transport properties like an anomalous Hall effect increasing as the temperature $`T`$ is lowered, (Taguchi and Tokura, 1999) a feature that seems to be connected with the geometrical frustration of pyrochlore lattices that is partly embodied by the Kagome lattice itself, viewed now as the (1,1,1) cross-section of the 3D pyrochloreโs lattice. Motivated by the transport properties of pyrochlore compounds, as well as by those of the manganite ones, we study in this paper some quantum transport properties of the Kagome lattice with a canted localised spin texture in which independent electrons can move.
The behavior of the Berry phase as a function of the spin canting for the model at hand, and its consequences in terms of the macroscopic transport properties of the model are studied in this paper. We discuss in detail the energy spectrum as it depends strongly on the canting of the localized spins. In particular, the nodes in the spectrum and the opening of energy gaps are investigated, including their consequences for the transport properties. Moreover, we evaluate explicitely the longitudinal conductivity $`\sigma _{xx}`$ and the Hall conductivity $`\sigma _{xy}`$ as a function of the canting angle $`\theta `$ or flux $`\varphi `$. Our results for the transport properties are then compared with those of another famous model of this class, where a staggered magnetic field is applied to electrons within a honeycomb lattice (Haldane, 1988). The latter has very similar spectral properties as the model, first proposed by Ohgushi, Murakami and Nagaosa, (2000) defined on the Kagome lattice and studied here in greater detail. The model on the honeycomb lattice was proposed by Haldane as the condensed-matter (solid-state) equivalent of the quantum Hall effect, in that a quantization of the Hall $`\sigma _{xy}`$ conductivity can be achieved by varying the local flux per plaquette $`\varphi `$, but without the need to introduce an external, homogeneous magnetic field. In the case of the Kagome lattice model at hand, the same result will be shown to be attained through the introduction of a localised spin texture. Our results confirm and complete the work by Ohgushi, Murakami and Nagaosa for the Hall $`\sigma _{xy}`$ conductivity, with explicit calculations in terms of expansions around the gapโs nodes shown in detail, and moreover the quantized values of the longitudinal $`\sigma _{xx}`$ conductivity are obtained indicating the existence of some sort of metal-insulator transition as the canting angle moves away from some special values.
The paper is organized as follows: In Section 2 the tight-binding model of Ohgushi, Murakami and Nagaosa for localized spins is described. The relationship between the localized spinโs wavefunction and the Berry phase of the hopping elctron is discussed in Section 2.1, and the application to the Kagome lattice (Section 2.2) is presented. In Section 2.3 the tigh-binding model for the honeycomb lattice with a staggered magnetic field is also presented and compared to the model on the Kagome lattice. Transport properties are then studied in Section 3 by making explicit use of Kuboโs formula and an expansion of the energy spectrum near the nodes next to the Fermi energy, and the results obtained are discussed in Section 4.
## 2 A Model of Hopping Electrons in a Spin Texture
### 2.1 The Model
We consider the electronic hopping between nearest neighbours on a Kagome lattice as described by the tight-binding Hamiltonian. The electronic degrees of freedom are coupled to a set of localized spin-$`S`$ degrees of freedom on the same lattice via a local Hund coupling $`J_H`$. In this work $`S=\frac{1}{2}`$, but generalization to the physically and theoretically interesting case of larger $`S`$ is possible. When $`J_H`$ is strong enough the spin of the hopping electron is forced to allign parallel to the localised spin $`๐_i`$ at each site and through double-exchange mechanism (Zener, 1951, Anderson and Hasegawa, 1955, de Gennes, 1960 ) the tight-binding hopping parameter $`t_{ij}`$ between two neighbouring sites $`i,j`$ becomes proportional to the projection of the localised-spin wave function at site $`j`$ onto that at site $`i`$. The effective Hamiltonian representing the hopping is then
$$H=\underset{i,j}{}t_{ij}^{eff}c_i^{}c_j+\mathrm{h}.\mathrm{c}.$$
(1)
where $`t_{ij}^{eff}=t๐ง_i|๐ง_j`$, $`t`$ being the bare hopping parameter and $`|๐ง`$ the spin wave function for a spin-$`\frac{1}{2}`$ quantized along the direction defined by the unit vector $`๐ง=(\mathrm{sin}\theta \mathrm{cos}\varphi ,\mathrm{sin}\theta \mathrm{sin}\varphi ,\mathrm{cos}\theta )`$. This spinor wave function clearly satisfies (with $`\stackrel{}{\sigma }=(\sigma _x,\sigma _y,\sigma _z)`$ the vector of Pauli matrices) $`๐ง\stackrel{}{\sigma }|๐ง=+|๐ง`$ and is given by
$$|๐ง=e^{ib}\left(\begin{array}{c}\mathrm{cos}\frac{\theta }{2}\\ e^{i\varphi }\mathrm{sin}\frac{\theta }{2}\end{array}\right),$$
(2)
where $`b`$ is an undetermined overall gauge degree of freedom. The effective hopping parameter is then
$$t_{ij}^{eff}=te^{i(b_ib_j)}\left\{\mathrm{cos}\frac{\theta _i}{2}\mathrm{cos}\frac{\theta _j}{2}+e^{i(\varphi _i\varphi _j)}\mathrm{sin}\frac{\theta _i}{2}\mathrm{sin}\frac{\theta _j}{2}\right\}$$
(3)
and since $`|๐ง_i|๐ง_j|^2=\mathrm{cos}^2\frac{\theta _{ij}}{2}`$, with $`\mathrm{cos}\theta _{ij}=๐ง_i๐ง_j`$ or $`\theta _{ij}`$ being the angle between the two localized spinsโ directions of quantization so that $`\mathrm{cos}^2\frac{\theta _{ij}}{2}=\frac{1}{2}(1+๐ง_i๐ง_j)`$, we see that we can put
$$t_{ij}^{eff}=te^{ia_{ij}}\mathrm{cos}\frac{\theta _{ij}}{2}$$
(4)
where the Berry phase $`a_{ij}`$ is obtained (ignoring the gauge parameters) through
$$e^{ia_{ij}}=\frac{\mathrm{cos}\frac{\theta _i}{2}\mathrm{cos}\frac{\theta _j}{2}+e^{i(\varphi _i\varphi _j)}\mathrm{sin}\frac{\theta _i}{2}\mathrm{sin}\frac{\theta _j}{2}}{\mathrm{cos}\frac{\theta _{ij}}{2}}$$
(5)
and can be evaluated, e.g., by means of
$$\mathrm{sin}a_{ij}=\frac{\mathrm{sin}\frac{\theta _i}{2}\mathrm{sin}\frac{\theta _j}{2}\mathrm{sin}(\varphi _i\varphi _j)}{\mathrm{cos}\frac{\theta _{ij}}{2}}.$$
(6)
To see what the phase $`a_{ij}`$ is, geometrically, we introduce the unit vector $`\widehat{z}`$ and evaluate the triple product
$$๐ง_i\times ๐ง_j\widehat{z}=\mathrm{sin}\theta _i\mathrm{sin}\theta _j\mathrm{sin}(\varphi _i\varphi _j)=4\mathrm{sin}\frac{\theta _i}{2}\mathrm{sin}\frac{\theta _j}{2}\mathrm{cos}\frac{\theta _i}{2}\mathrm{cos}\frac{\theta _j}{2}\mathrm{sin}(\varphi _i\varphi _j)$$
(7)
which shows that
$$\mathrm{sin}a_{ij}=\frac{๐ง_i\times ๐ง_j\widehat{z}}{4\mathrm{cos}\frac{\theta _i}{2}\mathrm{cos}\frac{\theta _j}{2}\mathrm{cos}\frac{\theta _{ij}}{2}}.$$
(8)
This expression is a special case of the formula giving the solid angle $`\mathrm{\Omega }(๐ง_1,๐ง_2,๐ง_3)`$ between three unit vectors $`๐ง_1`$, $`๐ง_2`$ and $`๐ง_3`$:
$$\mathrm{sin}\frac{\mathrm{\Omega }(๐ง_1,๐ง_2,๐ง_3)}{2}=\frac{๐ง_1\times ๐ง_2๐ง_3}{4\mathrm{cos}\frac{\theta _{12}}{2}\mathrm{cos}\frac{\theta _{13}}{2}\mathrm{cos}\frac{\theta _{23}}{2}}$$
(9)
(as can be verified by taking, e.g., $`๐ง_1=\widehat{x}`$, $`๐ง_2=\widehat{y}`$ and $`๐ง_3=\widehat{z}`$; $`\mathrm{\Omega }(๐ง_1,๐ง_2,๐ง_3)`$ can also be seen as the area of the portion of unit sphere enclosed by the maximum circles passing through the unit vectorsโ tips). In the last formula, of course, $`\mathrm{cos}\theta _{kk^{}}=๐ง_k๐ง_k^{}`$ and therefore $`\mathrm{cos}^2\frac{\theta _{kk^{}}}{2}=\frac{1}{2}(1+๐ง_k๐ง_k^{})`$. We remark that this formula for three spins is completely analogous to that for the chirality gauge field in the formulation of Lee and Nagaosa (1992) for the chiral spin liquid theory of high-temperature superconductivity (Wen, Wilczek and Zee, 1989). In this formulation the instantaneous gauge field flux through the triangular plaquette made up by the three spins is $`\mathrm{\Phi }(๐ง_1,๐ง_2,๐ง_3)=\frac{1}{2}\mathrm{\Omega }(๐ง_1,๐ง_2,๐ง_3)`$. Back to our two-spins hopping problem, we then conclude that $`\mathrm{sin}a_{ij}=\mathrm{sin}\frac{1}{2}\mathrm{\Omega }(๐ง_i,๐ง_j,\widehat{z})`$, or
$$a_{ij}=\pi +\frac{1}{2}\mathrm{\Omega }(๐ง_i,๐ง_j,\widehat{z})$$
(10)
(the factor $`\frac{1}{2}`$ being probably due to our specially chosen localised spin value, which leads to the conjecture that for a generic spin-$`S`$ situation the Berry phase would be $`a_{ij}=\pi +S\mathrm{\Omega }(๐ง_i,๐ง_j,\widehat{z})`$). Since the solid angle $`\mathrm{\Omega }(๐ง_i,๐ง_j,\widehat{z})`$ is also the unit sphereโs surface area between the tips of the three vectors $`๐ง_i`$, $`๐ง_j`$ and $`\widehat{z}`$, the phase $`a_{ij}=_i^j๐๐ซ๐`$ can also be seen as the flux of a magnetic monopoleโs field of modulus $`|๐|=\frac{1}{2}`$ with the monopole placed in the sphereโs center, or, alternatively, as the flux of the related gauge field $`๐`$ through the triangle bearing on the segment $`(i,j)`$ of a triangular Kagome latticeโs unit cell (Ohgushi, Murakami and Nagaosa, 2000). In this way, the Berry phase $`a_{ij}`$ acquires some physical meaning too.
We now consider this tight-binding model on the Kagome lattice with a fixed localised-spin configuration (or spin texture) as suggested by Ohgushi, Murakami and Nagaosa (2000) , in which the unit vectors $`๐ง_i`$ at each site of a triangular unit cell are tilted outwards at a fixed angle $`\theta `$ over the unit vector $`\widehat{z}`$ orthogonal to the lattice plane. This means (labelling the spins clockwise A, B and C in the unit cell)
$$\mathrm{sin}a_{AB}=\mathrm{sin}a_{BC}=\mathrm{sin}a_{CA}=\frac{\sqrt{3}\mathrm{sin}^2\theta }{4(1+\mathrm{cos}\theta )\sqrt{1\frac{3}{4}\mathrm{sin}^2\theta }}.$$
(11)
The flux generated by the spins in every triangular unit is set equal to $`\varphi `$ with the condition
$$e^{i\varphi }=e^{i(a_{AB}+a_{BC}+a_{CA})}=e^{3ia_{AB}}$$
(12)
with $`\varphi =3a_{AB}`$ (mod $`2\pi `$) and thus
$$\mathrm{sin}\frac{\varphi }{3}=\frac{\sqrt{3}(1\mathrm{cos}\theta )}{2\sqrt{1+3\mathrm{cos}^2\theta }}.$$
(13)
This is equivalent to the expression proposed by Ohgushi, Murakami and Nagaosa (preprint of Ohgushi, Murakami and Nagaosa (2000)) $`\varphi =\pi +3\mathrm{arg}(1i\sqrt{3}\mathrm{cos}\theta )`$. The graph for this expression of $`\varphi =\varphi (\theta )`$ is shown in Fig.2 for convenience.
As pointed out by Ohgushi, Murakami and Nagaosa (2000), the flux per triangular unit cell $`\varphi `$ is cancelled out for the chosen spin texture by the flux $`2\varphi `$ generated by each of the remaining hexagonal hopping plaquettes on the Kagome lattice. There are indeed twice as many triangular units as hexagonal plaquettes, so that the overall gauge field flux through the lattice is zero. This situation is reminiscent of the analogous tight-binding model in a staggered magnetic field as was proposed by Haldane (1988) to mimick the quantized Hall effect in a condensed-matter situation. In the present model, the chosen spin-texture, with all localised spins tilted by the same angle $`\theta `$ in each unit cell, is presumably the one corresponding to the mean-field solution for some magnetic spin-spin Heisenberg Hamiltonian which should be added to our tight-binding Hamiltonian, Eq. (1), to give a total Hamiltonian of the type
$$H_{tot}=\underset{i,j}{}t_{ij}^{eff}(\{๐_i^{(0)}\})c_i^{}c_j+\mathrm{h}.\mathrm{c}.+\underset{i,j}{}J_{ij}๐_i๐_j.$$
(14)
The role of the spin fluctuations around this ordered spin texture, $`\{๐_i^{(0)}\}`$, as well as the effects of different spin textures, (e.g. AFM ones) could serve as an interesting further research problem for future studies.
### 2.2 Band structure for the Kagome lattice
The Kagome lattice is made up of triangular and hexagonal plaquettes and can be seen as a triangular lattice with a 3-point basis where every triangular unit cell contains three sites, $`A`$, $`B`$, $`C`$ (Fig 1). The displacement vectors between these sites are $`\stackrel{}{a}_1=(1/2,\sqrt{3}/2)`$, $`\stackrel{}{a}_2=(1,0)`$ and $`\stackrel{}{a}_3=(1/2,\sqrt{3}/2)`$, with $`_i\stackrel{}{a}_i=0`$. The effective hopping parameter of (1) can be written as
$$t_{ij}^{eff}=te^{ia_{ij}}\mathrm{cos}(\frac{\theta _{ij}}{2})$$
(15)
and since $`\mathrm{cos}(\theta _{ij}/2)=\sqrt{1\frac{3}{4}\mathrm{sin}^2\theta }`$ is fixed for the chosen spin texture, $`\theta _{ij}`$ being the angle between the n.n. pair of localized spins, we can choose the convention where $`t\mathrm{cos}(\frac{\theta _{ij}}{2})1`$. Then, in momentum space, the Hamiltonian can be rewritten as
$$H=\underset{\stackrel{}{k}}{}\psi ^{}(\stackrel{}{k})h(\stackrel{}{k})\psi (\stackrel{}{k})$$
(16)
where $`\psi (\stackrel{}{k})=(c_A(\stackrel{}{k}),c_B(\stackrel{}{k}),c_C(\stackrel{}{k}))`$ and $`h(\stackrel{}{k})`$ is the (suitably symmetrized) matrix
$`h(\stackrel{}{k})=\left(\begin{array}{ccc}0& 2\mathrm{cos}(\stackrel{}{k}\stackrel{}{a}_1)e^{i\varphi /3}& 2\mathrm{cos}(\stackrel{}{k}\stackrel{}{a}_3)e^{i\varphi /3}\\ 2\mathrm{cos}(\stackrel{}{k}\stackrel{}{a}_1)e^{i\varphi /3}& 0& 2\mathrm{cos}(\stackrel{}{k}\stackrel{}{a}_2)e^{i\varphi /3}\\ 2\mathrm{cos}(\stackrel{}{k}\stackrel{}{a}_3)e^{i\varphi /3}& 2\mathrm{cos}(\stackrel{}{k}\stackrel{}{a}_2)e^{i\varphi /3}& 0\end{array}\right).`$ (20)
The three eigenvalues of this Hamiltonian are
$`E_{up}(\stackrel{}{k})`$ $`=`$ $`4\sqrt{{\displaystyle \frac{1+f(\stackrel{}{k})}{3}}}\mathrm{cos}({\displaystyle \frac{\theta (\stackrel{}{k})}{3}})`$
$`E_{mid}(\stackrel{}{k})`$ $`=`$ $`4\sqrt{{\displaystyle \frac{1+f(\stackrel{}{k})}{3}}}\mathrm{cos}({\displaystyle \frac{\theta (\stackrel{}{k})2\pi }{3}})`$ (21)
$`E_{down}(\stackrel{}{k})`$ $`=`$ $`4\sqrt{{\displaystyle \frac{1+f(\stackrel{}{k})}{3}}}\mathrm{cos}({\displaystyle \frac{\theta (\stackrel{}{k})+2\pi }{3}})`$
with
$$\theta (\stackrel{}{k})=arg\left[f(\stackrel{}{k})\mathrm{cos}(\varphi )+i\sqrt{4\left(\frac{1+f(\stackrel{}{k})}{3}\right)^3(f(\stackrel{}{k})\mathrm{cos}(\varphi ))^2}\right]$$
(22)
and
$$f(\stackrel{}{k})=2\mathrm{cos}(\stackrel{}{k}\stackrel{}{a}_1)\mathrm{cos}(\stackrel{}{k}\stackrel{}{a}_2)\mathrm{cos}(\stackrel{}{k}\stackrel{}{a}_3).$$
(23)
The three bands touch in six points only for $`\varphi =0`$ and $`\varphi =\pm \pi `$, while for $`\varphi `$ different from these values there is a gap between the bands, as shown in Fig. 3 ($`\varphi =0`$) and Fig. 4 ($`\varphi =\pi /3`$).
These nodes are: $`(\pm \frac{2\pi }{3},0)`$ and $`(\pm \frac{\pi }{3},\pm \frac{\sqrt{3}}{3}\pi )`$, on the vertices of a hexagon, as shown in Fig. 5.
The problem of calculating transport coefficients with this $`3\times 3`$ matrix is not exactly solvable, so we reduce this matrix to a $`2\times 2`$ one by expanding $`h(\stackrel{}{k})`$ around the nodes. This can be done with a unitary transformation which allows us to neglect the terms related to the lower band; in fact this band is far from the other two and gives no relevant contribution to the Green function present in the Kubo formula. To find this unitary transformation we consider the Hamiltonian evaluated at a node $`๐ค_0`$. If we apply the unitary matrix that diagonalizes $`h(k_{x0},k_{y0})`$, where $`(k_{x0},k_{y0})`$ are the nodeโs coordinates, to the Hamiltonian evaluated at the general point $`(k_x,k_y)`$, we find a matrix $`H^{}`$ with the structure
$`H^{}=\left(\begin{array}{ccc}\lambda _1& \alpha [\stackrel{~}{k}_xi\stackrel{~}{k}_y]& \beta [\stackrel{~}{k}_xi\stackrel{~}{k}_y]\\ \overline{\alpha }[\stackrel{~}{k}_x+i\stackrel{~}{k}_y]& \lambda _2& \gamma [\stackrel{~}{k}_x+i\stackrel{~}{k}_y]\\ \overline{\beta }[\stackrel{~}{k}_x+i\stackrel{~}{k}_y]& \overline{\gamma }[\stackrel{~}{k}_xi\stackrel{~}{k}_y]& \lambda _3\end{array}\right)`$ (27)
with $`\stackrel{~}{k}_x=(k_xk_{x0})`$ and $`\stackrel{~}{k}_y=(k_yk_{y0})`$. The elements on the diagonal are the eigenvalues of the Hamiltonian $`h(k_{x0},k_{y0})`$, while the off-diagonal elements are complex combinations of $`\stackrel{~}{k}_{x,y}`$. Near the nodes the distance between the upper and the middle band is small, while the lower band is distant and gives no relevant contribution. To justify this we can consider a projection of the Green function around the node.
We want to find a projection only on the first two eigenvalues, so we choose a projector such that
$$PH^{^{}}P=\left(\begin{array}{cc}\lambda _1& H_{12}^{^{}}\\ H_{21}^{^{}}& \lambda _2\end{array}\right).$$
(28)
We define the Green function as $`G=(zH^{^{}})^1`$ and the projection operator $`P`$ with the convention that $`(A)_P^1=(PAP)^1`$ is the inverse operation on the projected space, $`(1P)`$ being the projection operator complementary to $`P`$. The Green function can now be written as
$$G=PGP+(1P)GP+PG(1P)+(1P)G(1P)$$
(29)
and the projected Green function is
$`PGP=P(zH^{^{}})^1P=(zPH^{^{}}PPH^{^{}}(1P)(zH^{^{}})_{1P}^1(1P)H^{^{}}P)_P^1.`$
The last terms are of higher order and can be neglected. If we consider the terms related to the lower eigenvalue, we can see that $`1/(z\lambda _3)1/2`$, because we are considering $`\stackrel{}{k}`$ near the nodes. This eigenvalue does not give an important contribution, so it can be neglected and we can use the approximation
$$PGP(zPH^{^{}}P)_P^1.$$
(30)
Now we can write the projection of the Hamiltonian $`H^{}`$ as
$`PH^{}P`$ $`=`$ $`\left(\begin{array}{cc}\lambda _1& \alpha [\stackrel{~}{k}_xi\stackrel{~}{k}_y]\\ \overline{\alpha }[\stackrel{~}{k}_x+i\stackrel{~}{k}_y]& \lambda _2\end{array}\right)`$ (33)
$`=`$ $`\left(\begin{array}{cc}1+\frac{\sqrt{3}\varphi }{3}& h_1ih_2\\ h_1+ih_2& 1\frac{\sqrt{3}\varphi }{3}\end{array}\right)`$ (36)
$`=`$ $`I+\left(\begin{array}{cc}m& h_1ih_2\\ h_1+ih_2& m\end{array}\right)=I+h`$ (39)
where $`I`$ is the identity matrix, with
$`h_1`$ $`=`$ $`\alpha _1\stackrel{~}{k}_x+\alpha _2\stackrel{~}{k}_y`$
$`h_2`$ $`=`$ $`\alpha _2\stackrel{~}{k}_x+\alpha _1\stackrel{~}{k}_y`$ (40)
$`m`$ $`=`$ $`{\displaystyle \frac{\sqrt{3}}{3}}\varphi ,`$
with $`\alpha _1`$ and $`\alpha _2`$ are the components of a complex parameter, depending on the node we are considering. The new matrix representing the Hamiltonian has eigenvalues $`\pm \lambda `$ and eigenvectors $`\mathrm{\Psi }_\pm `$ with
$`\lambda `$ $`=`$ $`\sqrt{m^2+h_1^2+h_2^2}`$
$`\mathrm{\Psi }_\pm `$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{1+(\pm \lambda m)^2/|k|^2}}}\left(\begin{array}{c}1\\ \frac{\pm \lambda m}{k}\end{array}\right).`$ (43)
As was said, the bands touch only when $`\varphi =0`$ and $`\varphi =\pm \pi `$. Here we consider $`\varphi `$ different from zero, but small enough so that we create a small gap between the two bands.
### 2.3 Band structure of the honeycomb lattice
Here we also consider the case of a honeycomb lattice, as was first envisaged by Haldane (1988); this is made up by two sublattices that we call A and B (Fig. 6), or by a triangular lattice with a 2-point basis. The displacement vectors from a B site to the three nearest neighbours are: $`\stackrel{}{a}_1=(\sqrt{3}/2,1/2)`$, $`\stackrel{}{a}_2=(0,1)`$ and $`\stackrel{}{a}_3=(\sqrt{3}/2,1/2)`$, while the displacement vectors from the site B and the nearest neighbours on the same sublattice are $`\stackrel{}{b}_1=(\sqrt{3}/2,3/2)`$, $`\stackrel{}{b}_2=(\sqrt{3},0)`$ and $`\stackrel{}{b}_3=(\sqrt{3}/2,3/2)`$ (again, $`\stackrel{}{b}_i=0`$).
Here too, we consider a tight-binding model in the presence of a staggered magnetic flux (Haldane, 1988); the Hamiltonian for this Haldane model is $`H=_\stackrel{}{k}\psi ^{}(\stackrel{}{k})h(\stackrel{}{k})\psi (\stackrel{}{k})`$ with
$`h(\stackrel{}{k})=`$ $`2t_2\mathrm{cos}\varphi {\displaystyle \underset{i}{}}\mathrm{cos}(๐ค๐_๐ข)๐+t_1{\displaystyle \underset{i}{}}[\mathrm{cos}(๐ค๐_๐ข)\sigma _\mathrm{๐}+\mathrm{sin}(๐ค๐_๐ข)\sigma _\mathrm{๐}]`$
$`+`$ $`[M2t_2\mathrm{sin}\varphi {\displaystyle \underset{i}{}}\mathrm{sin}(๐ค๐_๐ข)]\sigma _\mathrm{๐},`$ (44)
where $`t_1`$ is a hopping parameter between nearest neighbours on different sublattices, $`t_2`$ is a hopping parameter between nearest neighbour sites on the same sublattice, and $`\sigma _i`$ are the three Pauli matrices. If we rewrite the Hamiltonian as $`H=a\sigma _1+b\sigma _2+c\sigma _3`$, the Hamiltonian matrix reads
$`H=\left(\begin{array}{cc}c& aib\\ a+ib& c\end{array}\right).`$ (47)
The eigenvalues of this matrix are $`\pm \lambda `$ with $`\lambda =\sqrt{a^2+b^2+c^2}`$ (Fig. 7 ($`\varphi =0`$) and Fig. 8 ($`\varphi =\frac{\pi }{3}`$)), while the eigenvectors are
$$\psi _\pm =\frac{1}{\sqrt{1+\frac{(\pm \lambda )^2}{|k|^2}}}\left(\begin{array}{c}1\\ \frac{\pm \lambda c}{k}\end{array}\right)$$
(48)
where $`k=aib`$. Formally this case is similar to that of the Kagome lattice, after the reduction of the original matrix to a $`2\times 2`$ one. The two bands meet when the condition $`a^2+b^2+c^2=0`$ is satisfied. This becomes a condition on the parameter $`M`$: there are nodes when $`M=3\sqrt{3}t_2\alpha \mathrm{sin}\varphi `$, with $`\alpha =\pm 1`$. When $`\varphi =0`$ and $`M=0`$ there are six nodes: $`(\pm \frac{4\pi }{3\sqrt{3}},0)`$ and $`(\pm \frac{2\pi }{3\sqrt{3}},\pm \frac{2\pi }{3})`$ (Fig. 9), while when $`M=3\sqrt{3}t_2\alpha \mathrm{sin}\varphi `$ there are only three of these nodes.
## 3 Transport Properties
Based on the linear-response theory, a suitable Kubo formula and the corresponding conductivity tensor can be studied for the Hamiltonians considered in Sect. 2. Some details are given in Appendix A. From this result we can derive the Hall conductivity $`\sigma _{xy}`$ and the longitudinal conductivity $`\sigma _{xx}`$ of our two-dimensional tight-binding model. We verify explicitely the quantization of $`\sigma _{xy}`$ as a function of $`\varphi `$ (or $`\theta `$) and calculate explicitely the longitudinal conductivity $`\sigma _{xx}`$ which also appears to be quantized in the absence of disorder or other symmetry-breaking conditions. For the two models considered, these are our main new results.
### 3.1 Hall conductivity
For $`\mu \nu `$ the third term in Eq. (95) (Appendix A) vanishes and after the integration with respect to $`E`$ we find that the Hall conductivity, in the limit of $`\omega =0`$ and $`T=0`$, is
$$\sigma _{xy}=\frac{1}{\mathrm{}\eta }Re\underset{k}{}\lambda _k[U^{}j_\mu (h\lambda _k+2i\eta )^1j_\nu U]_{kk},$$
(49)
where $`U`$ is the unitary matrix that diagonalizes the Hamiltonian matrix $`h(\stackrel{}{k})`$, while $`j_x`$ and $`j_y`$ are the current matrices. From this we find that the Hall conductivity for every node $`n`$ for the case of the Kagome lattice is
$$\sigma _{xy}^n=\frac{e^2}{\mathrm{}\eta }_{\mathrm{}}^+\mathrm{}\frac{12m\eta }{8(m^2+h_1^2+h_2^2)^{\frac{3}{2}}}\frac{d^2k}{(2\pi )^2}=\frac{e^2}{2h}sgn(\varphi ).$$
(50)
The integration being over the hexagonal Brillouin zone, we considered only one third of the integral and then we have to multiply for the number of nodes. We can conclude that the Hall conductivity is different from zero and it is quantized in the presence of a gap between the bands (that is for $`\varphi `$ different from 0, $`\pm \pi `$) and is equal to
$$\sigma _{xy}=\frac{e^2}{h}sgn(\varphi ).$$
(51)
So, we have another model of transverse conductivity quantization in the absence of an external uniform magnetic field.
Now we consider the case of the honeycomb lattice, in Haldaneโs model. For $`M=0`$ and $`\varphi =0`$ the bands touch in six points: $`(\pm \frac{4\pi }{3\sqrt{3}},0)`$, $`(\pm \frac{2\pi }{3\sqrt{3}},\pm \frac{2\pi }{3})`$ and the Hamiltonian is simply of the form
$`\left(\begin{array}{cc}0& aib\\ a+ib& 0\end{array}\right).`$ (54)
Expanding the terms around the nodes, we find that the function to integrate in order to find the Hall conductivity is
$$\pm \frac{9ab}{8(a^2+b^2)^{\frac{3}{2}}}$$
(55)
but the integral of this term gives zero contribution. To generate a gap we have to move from the situation in which $`M=0`$ and $`\varphi =0`$. We add a small mass contribution $`M1`$, but we mantain $`\varphi =0`$; now the Hamiltonian is
$`\left(\begin{array}{cc}M& aib\\ a+ib& M\end{array}\right).`$ (58)
In this case the function to integrate is
$$\pm \frac{9t_1^2M\eta }{8(a^2+b^2+M^2)^{\frac{3}{2}}};$$
(59)
three nodes give a positive contribution and three a negative one, to give
$$\pm \frac{e^2}{2h}sgn(M)$$
(60)
Summing up all the contributions we find that the Hall conductivity is zero. Different is the situation in which we consider $`M=0`$, but we add a small flux $`\varphi `$. We rewrite the Hamiltonian as
$`\left(\begin{array}{cc}c& aib\\ a+ib& c\end{array}\right),`$ (63)
where
$$c=2t_2\mathrm{sin}\varphi \underset{i}{}\mathrm{sin}(๐ค๐_๐ข).$$
(64)
Now the function to integrate is ($`k=aib`$)
$$P(k)=\pm \frac{9c\eta t_1^2}{8(a^2+b^2+c^2)^{\frac{3}{2}}}.$$
(65)
Near three of the six nodes $`P(k)`$ is negative, but approximating $`c`$ around these points we find $`c3\sqrt{3}t_2\mathrm{sin}\varphi `$, so the function to integrate is negative. After the integration we find that every one of these three points gives a conductivity equal to
$$\frac{e^2}{2h}sgn(\mathrm{sin}\varphi ).$$
(66)
For the other three points $`P(k)`$ is positive, but the expansion of $`c`$ is $`c3\sqrt{3}t_2\mathrm{sin}(\varphi )`$. So, now the function to integrate is negative too and it gives the same result as before. Summing over all the points and remembering that the integration is over the exagon we find that the Hall conductivity for the honeycomb lattice case is
$$\sigma _{xy}=\frac{e^2}{h}sgn(\mathrm{sin}\varphi ).$$
(67)
We can conclude that the Hall conductivity can be rewritten as
$$\sigma _{xy}=\nu \frac{e^2}{h}$$
(68)
with $`\nu =\pm 1`$, depending on the sign of $`\varphi `$.
### 3.2 Longitudinal conductivity
Here too we use the general expression derived from the Kubo formula. In this case ($`\mu =\nu =x`$) the longitudinal conductivity $`\sigma _{xx}`$ derived by expression (95), after the energy integration, is
$`{\displaystyle \frac{1}{\mathrm{}}}{\displaystyle \underset{k,m}{}}{\displaystyle \frac{\rho _0(\lambda _k)(U^{}j_xU)_{km}(U^{}j_xU)_{mk}}{\omega +2i\eta }}\left[{\displaystyle \frac{1}{\lambda _m\lambda _k\omega +2i\eta }}+{\displaystyle \frac{1}{\lambda _m\lambda _k+\omega 2i\eta }}\right].`$
After summing over the eigenvalues, substituting the values of $`\lambda _k`$ and $`ฯต=\eta +\frac{i\omega }{2}`$ we find
$$\sigma _{xx}=\frac{1}{2\mathrm{}ฯต}_{\mathrm{}}^+\mathrm{}(U^{}j_xU)_{21}(U^{}j_xU)_{12}\frac{\sqrt{m^2+h_1^2+h_2^2}}{m^2+h_1^2+h_2^2+ฯต^2}\frac{d^2k}{(2\pi )^2}.$$
(69)
\>From this expression, we have still to subtract the diamagnetic term and so we evaluate (Ludwig, Fisher, Shankar and Grinstein, 1994)
$$\stackrel{~}{\sigma }_{xx}=\sigma _{xx}\frac{1}{ฯต}\underset{ฯต0}{lim}ฯต\sigma _{xx}.$$
(70)
For the Kagome lattice the product of the matrix elements of the currents is
$$(U^{}j_xU)_{21}(U^{}j_xU)_{12}=\frac{3(4m^2+3h_1^2+2\sqrt{3}h_1h_2+h_2^2)}{4(m^2+h_1^2+h_2^2)},$$
(71)
and remembering that $`h_1`$ and $`h_2`$ are symmetric variables and using polar coordinates we can rewrite $`\sigma _{xx}`$ as
$`\sigma _{xx}={\displaystyle \frac{e^2}{4hฯต}}\left[{\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{m^2r}{\sqrt{m^2+r^2}(m^2+r^2+ฯต^2)}}๐r+{\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{r\sqrt{m^2+r^2}}{m^2+r^2+ฯต^2}}๐r\right],`$ (72)
so that carrying out the integrals we find
$`{\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{rm^2}{ฯต\sqrt{m^2+r^2}(m^2+r^2+ฯต^2)}}๐r=\{\begin{array}{cc}\frac{m^2}{ฯต^2}\mathrm{arccos}(\frac{m}{\sqrt{m^2+ฯต^2}})\hfill & \hfill \text{if }m>0\\ 0\hfill & \hfill \text{if }m=0\\ \frac{m^2}{ฯต^2}(\pi \mathrm{arccos}(\frac{m}{\sqrt{m^2+ฯต^2}})\hfill & \hfill \text{if }m<0\end{array}`$ (76)
For the second integral we introduce a cut-off $`\lambda `$ and evaluate
$`{\displaystyle _0^\lambda }{\displaystyle \frac{r\sqrt{m^2+r^2}}{ฯต(m^2+r^2+ฯต^2)}}๐r=\{\begin{array}{cc}\frac{\lambda }{ฯต}(m+\sqrt{m^2+\lambda ^2})+\mathrm{arctan}(\frac{m}{ฯต})\mathrm{arctan}(\frac{\sqrt{m^2+\lambda ^2}}{ฯต})\hfill & \hfill \text{if }m>0\\ & \\ \frac{\lambda }{ฯต}\mathrm{arctan}(\frac{\lambda }{ฯต})\hfill & \hfill \text{if }m=0\\ & \\ \frac{\lambda }{ฯต}(m+\sqrt{m^2+\lambda ^2})\mathrm{arctan}(\frac{m}{ฯต})\mathrm{arctan}(\frac{\sqrt{m^2+\lambda ^2}}{ฯต})\hfill & \hfill \text{if }m<0.\end{array}`$ (82)
We consider two cases: $`m`$ equal to zero (that is, the flux $`\varphi `$ is zero) and $`m`$ different from zero. In the first case, after having removed the diamagnetic term, the conductivity is
$$\stackrel{~}{\sigma }_{xx}=\frac{e^2}{4h}\mathrm{arctan}(\frac{\lambda }{ฯต}).$$
(83)
Now we can take the limit for $`ฯต0`$ (thus making the cutoff irrelevant). We find that the conductivity for every node is different from zero and is equal to
$$\stackrel{~}{\sigma }_{xx}=\frac{1}{3}\frac{e^2}{4h}\frac{\pi }{2}.$$
(84)
After the sum over all six nodes is done we can conclude that the longitudinal conductivity for $`m=0`$ is
$$\stackrel{~}{\sigma }_{xx}=\frac{e^2\pi }{4h}.$$
(85)
The case where $`m`$ is not zero is quite different. Now, after having done the diamagnetic subtraction (70), the conductivity is
$$\stackrel{~}{\sigma }_{xx}=\frac{e^2}{4h}[\frac{m^2}{ฯต}\mathrm{arccos}(\frac{m}{\sqrt{m^2+ฯต^2}})\frac{m}{ฯต}+\mathrm{arctan}(\frac{m}{ฯต})\mathrm{arctan}(\frac{\sqrt{m^2+\lambda ^2}}{ฯต})],$$
(86)
but the limit for $`ฯต0`$ gives a vanishing result. We conclude that the longitudinal conductivity is different from zero only when there is no gap between the two bands, that is in our case for $`m=0`$.
Next we consider the longitudinal conductivity for the honeycomb lattice. We expect that it is different from zero when the bands touch. This happens in six points, when $`M=0`$ and $`\varphi =0`$, and in three points when $`M=3\sqrt{3}\alpha t_2\mathrm{sin}(\varphi )`$. To calculate $`\sigma _{xx}`$, we use the expression (69). In the first case every node gives a contribution different from zero and equal to each other, so with the same observations made for the Kagome lattice we find that the longitudinal conductivity is
$$\sigma _{xx}=\frac{e^2\pi }{4h}(M=\varphi =0).$$
(87)
When $`M>0`$ and $`\varphi =0`$ we find that the conductivity is
$`\sigma _{xx}=`$ $``$ $`{\displaystyle \frac{1}{2\mathrm{}ฯต}}{\displaystyle \frac{9}{8}}{\displaystyle \frac{1}{(2\pi )^2}}{\displaystyle _{\mathrm{}}^{\mathrm{}}}{\displaystyle \frac{\sqrt{h_1^2+h_2^2+M^2}}{h_1^2+h_2^2+M^2+ฯต^2}}d^2k`$
$``$ $`{\displaystyle \frac{1}{2\mathrm{}ฯต}}{\displaystyle \frac{9}{8}}{\displaystyle \frac{1}{(2\pi )^2}}{\displaystyle _{\mathrm{}}^{\mathrm{}}}{\displaystyle \frac{M^2}{\sqrt{h_1^2+h_2^2+M^2}(h_1^2+h_2^2+M^2+ฯต^2)}}d^2k`$
but, as in the case of the Kagome lattice, this integral vanishes; so in this case the longitudinal conductivity is zero.
The last case we consider is for $`M=\pm 3\sqrt{3}\alpha t_2\mathrm{sin}(\varphi )0`$. Here the bands touch only in three points and these give a contribution to the conductivity since now there is a gap where before there were three nodes. Around the three nodes the term $`c`$ is zero, so formally the problem is the same as that of the case $`M=0`$ and $`\varphi =0`$ and only in these three points the longitudinal conductivity is different from zero. The result is equal to half of what was found in the case in which $`M`$ and $`\varphi `$ are zero, that is
$$\sigma _{xx}=\frac{e^2\pi }{8h}(M=\pm 3\sqrt{3}\alpha t_2\mathrm{sin}(\varphi )0).$$
(88)
We can conclude that the longitudinal conductivity can be rewritten as
$$\sigma _{xx}=\mu \frac{e^2\pi }{8h}$$
(89)
with $`\mu `$ = 0, 1, 2 for the honeycomb lattice and $`\mu `$ = 0, 2 for the Kagome lattice.
## 4 Discussion and Conclusions
We have considered both Haldaneโs model for electrons in a staggered flux on the honeycomb lattice and the model by Ohgushi, Murakami and Nagaosa for electrons in the presence of a canted spin-1/2 texture on the Kagome lattice. We have shown how similar these two models are in that the transverse Hall conductivity $`\sigma _{xy}`$ is quantized as $`\pm e^2/h`$ as a function of the tuning parameter (e.g. the magnetic flux per plaquette, $`\varphi `$).
Whilst for the Hall conductivity $`\sigma _{xy}`$ we have obtained the same results both for the Kagome lattice in the presence of a spin texture (as found by Oshgushi et al. (2000)) and for the honeycomb lattice with staggered magnetic field (as found by Haldane (1988)), we stress that we have used a different method of calculation based on implementing the band structure of each model in the Kubo formula. Furthermore, we have explicitely evaluated in this way, and for the first time, also the longitudinal conductivity $`\sigma _{xx}`$ starting from the Kubo formula. This quantity is also quantized in the absence of symmetry-breaking, non-ideal features of the system, but not in terms of integer multiples of $`e^2/h`$.
For the Kagome lattice model, we find metallic behavior for a ferromagnetic state of localized spins perpendicular to the plane of the lattice. This state has a vanishing flux ($`\varphi =0`$) in each plaquette of the Kagome lattice. Metallic behavior exists also for a canted state where the spins are inside the plane ($`\theta =\pi /2`$, in this case the local flux is $`\varphi =\pi `$). The longitudinal conductivity is for both cases $`\sigma _{xx}=e^2\pi /4h`$ and the Hall conductivity vanishes. In Fig. 10 we show the schematic behaviour of the Hall conductivity for the model defined on the Kagome lattice and as a function of the parameter $`\varphi `$. This is to be compared with the richer phase diagram for the Haldane model on the honeycomb lattice, reported in Fig. 11 also as a function of $`\varphi `$. The longitudinal conductivity $`\sigma _{xx}`$ as evaluated in this work is shown schematically in Fig. 12.
Removing some of the nodes in the DOS by breaking symmetries (like for the case of a square lattice with next-nearest neighbor terms) alters the Hall conductivity substantially. Also the introduction of disorder (e.g. slow fluctuations of the localized spins, fluctuations around the perfect canted spin texture) may remove some of the nodes and yield non-universal features in the transport properties. There is also another interesting effect due to disorder in our two-dimensional lattices. The longitudinal conductivity $`\sigma _{xx}`$ is usually based on diffusion of charge carriers; however, the diffusion coefficient $`D`$ is infinite in our model, since there is no scattering in the absence of imperfections. Nevertheless, the longitudinal conductivity, expressed through the Einstein relation
$$\sigma _{xx}=\frac{e^2}{\mathrm{}}D\rho ,$$
is finite thanks to a vanishing density of states at the nodes. The cancellation of the divergent diffusion coefficient and the vanishing density of states is subtle. Since there is scattering by impurities in a realistic system, a finite diffusion coefficient is more natural. On the other hand, impurities create additional states near the nodes such that a non-vanishing density of states exists. This effect was studied in the case of 2D Dirac fermions with random scatterers (Ziegler, 1997, 1998, Ziegler and Jug, 1997). In particular, it was found that random scattering broadens the metallic state (Ziegler and Jug, 1997), and the maximal conductivity value is lowered by a factor $`1/(1+g/2\pi )`$, where $`g`$ is the strength of the random fluctuations.
Acknowledgement:
We are grateful to MIUR (Ministero dellโIstruzione, Universitaโ e della Ricerca) for support through PRIN-2003 grant and to the Deutsche Forschungsgemeinschaft for support through Sonderforschungsbereich 484.
## Appendix A: Linear Response and Kubo Formula
From the Kubo formula we know that the conductivity tensor can be written as (Madelung, 1978)
$$\sigma _{\mu \nu }=\frac{e}{i\mathrm{}}\underset{\alpha 0}{lim}_{\mathrm{}}^0e^{(i\omega +\alpha )t}Tr([\rho _0,r_\mu ]e^{iHt/\mathrm{}}j_\nu e^{iHt/\mathrm{}})๐t$$
(90)
where $`\rho _0`$ is the Fermi function. Using the Green functions defined as
$$G_\pm (E)=(H/\mathrm{}+E\pm i\eta )^1$$
(91)
we can use the substitution
$$e^{\pm iHt/\mathrm{}}=\pm \underset{\eta 0}{lim}_{\mathrm{}}^{\mathrm{}}e^{iEt}G_{}(E)\frac{dE}{2\pi i}(t0)$$
(92)
the conductivity can be rewritten as
$$\sigma _{\mu \nu }=\frac{e}{\mathrm{}}_{\mathrm{}}^{\mathrm{}}Tr[\rho _0,r_\mu ]G_+(E)j_\nu G_{}(E+\omega )\frac{dE}{2\pi i}$$
(93)
The current operator is
$$j_\nu =\frac{e}{i}[H,r_\nu ]$$
(94)
Using this expression iteratively, we find that the conductivity can be rewritten as a sum of three terms
$`\sigma _{\mu \nu }={\displaystyle \frac{1}{\mathrm{}}}{\displaystyle _{\mathrm{}}^{\mathrm{}}}Tr\{\rho _0G_+(E)j_\mu G_+(E)j_\nu G_{}(E+\omega )\}{\displaystyle \frac{dE}{2\pi }}`$
$`{\displaystyle \frac{1}{\mathrm{}}}{\displaystyle _{\mathrm{}}^{\mathrm{}}}Tr\{\rho _0G_+(E)j_\nu G_{}(E+\omega )j_\mu G_{}(E+\omega )\}{\displaystyle \frac{dE}{2\pi }}`$ (95)
$`+{\displaystyle \frac{e}{\mathrm{}}}{\displaystyle _{\mathrm{}}^{\mathrm{}}}Tr\{\rho _0G_+(E)[j_\nu ,r_\mu ]G_{}(E+\omega )\}{\displaystyle \frac{dE}{2\pi }}.`$
|
warning/0506/math0506200.html
|
ar5iv
|
text
|
# On packing spheres into containers (about Keplerโs finite sphere packing problem)
## 1. Introduction
How many equally sized spheres can be packed into a given container? In 1611, Kepler discussed this question in his booklet \[Kep11\] and came to the following conclusion:
> โCoaptatio fiet arctissima, ut nullo praeterea ordine plures globuli in idem vas compingi queant.โ
>
> โThe (cubic or hexagonal close) packing will be the tightest possible, so that in no other arrangement more spheres could be packed into the same container.โ
In this note we want to show that Keplerโs assertion is false for many containers (see Section 5, Corollary 2). Even more general we show, roughly speaking, that the set of solutions to the finite container problem (see below) in an Euclidean space of dimension $`d2`$ has no โsimple structureโ (see Definition 1).
To make this precise, we consider the Euclidean $`d`$-space $`^d`$ endowed with inner product $`,`$ and norm $`||`$. Let $`B^d=\{๐^d:|๐|1\}`$ denote the (solid) unit sphere and $`S^{d1}=\{๐^d:|๐|=1\}`$ its boundary. Then a discrete set $`X^d`$ is a packing set and defines a sphere packing $`X+\frac{1}{2}B^d=\{๐+\frac{1}{2}๐:๐X,๐B^d\}`$, if distinct elements $`๐,๐^{}X`$ have distance $`|๐๐^{}|1`$. The sphere packing is called finite if $`X`$ is of finite cardinality $`|X|`$. Here we consider finite sphere packings contained in a convex body (container) $`C`$, that is, a compact, convex subset of $`^d`$ with nonempty interior. The finite container problem may be stated as follows.
###### Problem.
Given $`d2`$, $`n`$ and a convex body $`C^d`$, determine
$$\lambda (C,n)=\mathrm{min}\{\lambda >0:\lambda CX+\frac{1}{2}B^d\text{ a packing, }X^d\text{ with }|X|=n\text{ }\}$$
and packing sets $`X`$ attaining the minimum.
Many specific instances of this container problem have been considered (see for example \[Bez87\], \[BW04\], \[Fod99\], \[Mel97\], \[Nร97\],\[Spe04\], \[SMC+06\]). Independent of the particular choice of the container $`C`$, solutions tend to densest infinite packing arrangements for growing $`n`$ (see Section 5, cf. \[CS95\]). In dimension $`2`$ these packings are known to be arranged hexagonally. Nevertheless, although close, solutions to the container problem are not hexagonally arranged for all sufficiently large $`n`$ and various convex disks $`C`$, as shown by the author in \[Sch02\], Theorem 9 (cf. \[LG97\] for corresponding computer experiments). Here we show that a similar phenomenon is true in arbitrary Euclidean spaces of dimension $`d2`$.
We restrict ourselves to smooth convex bodies $`C`$ as containers. That is, we assume the support function $`h_C(๐)=sup\{๐,๐:๐C\}`$ of $`C`$ is differentiable at all $`๐^d\{\mathrm{๐}\}`$, or equivalently, we require that $`C`$ has a unique supporting hyperplane through each boundary point (see \[Sch93\], Chapter 1.7).
Our main result shows that families of packing sets with a โsimple structureโ can not be solutions to the container problem if $`C`$ is smooth and $`n`$ sufficiently large. This applies for example to the family of solutions to the lattice restricted container problem. In it, we only consider packing sets which are isometric to a subset of some lattice (a discrete subgroup of $`^d`$).
###### Theorem 1.
Let $`d2`$ and $`C^d`$ a smooth convex body. Then there exists an $`n_0`$, depending on $`C`$, such that $`\lambda (C,n)`$ is not attained by any lattice packing set for $`nn_0`$.
## 2. Packing families of limited complexity
The result of Theorem 1 can be extended to a more general class of packing sets.
###### Definition 1.
A family $``$ of packing sets in $`^d`$ is of limited complexity (an lc-family), if
1. there exist isometries $`_X`$, for each $`X`$, such that
(1)
$$\{๐๐:๐,๐_X(X)\text{ and }X\}$$
has only finitely many accumulation points in any bounded region.
2. there exists a $`\varrho >0`$, such that for all $`๐X`$ with $`X`$, every affine subspace spanned by some elements of
$$\{๐X:|๐๐|=1\}$$
either contains $`๐`$ or its distance to $`๐`$ is larger than $`\varrho `$.
Condition (i) shows that point configurations within an arbitrarily large radius around a point are (up to isometries of $`X`$ and up to finitely many exceptions) arbitrarily close to one out of finitely many possibilities. Condition (ii) limits the possibilities for points at minimum distance further. Note that the existence of a $`\varrho >0`$ in (ii) follows if (1) in (i) is finite within $`S^{d1}`$.
An example of an lc-family in which isometries can be chosen so that (1) is finite in any bounded region, is the family of hexagonal packing sets. These are isometric copies of subsets of a hexagonal lattice, in which every point in the plane is at minimum distance $`1`$ to six others. For the hexagonal packing sets, condition (ii) is satisfied for all $`\varrho <\frac{1}{2}`$. More general, isometric copies of subsets of a fixed lattice give finite sets (1) in any bounded region and satisfy (ii) for suitable small $`\varrho >0`$. Similar is true for more general families of packing sets, as for example for the hexagonal close configurations in dimension $`3`$ (see Section 5).
An example of an lc-family, in which the sets (1) are not necessarily finite in any bounded region, are the solutions to the lattice restricted container problem. As shown at the end of Section 3, condition (ii) in Definition 1 is nevertheless satisfied. Thus we derive Theorem 1 from the following, more general result.
###### Theorem 2.
Let $`d2`$, $`C^d`$ a smooth convex body and $``$ an lc-family of packing sets in $`^d`$. Then there exists an $`n_0`$, depending on $``$ and $`C`$, such that $`\lambda (C,n)`$ is not attained by any packing set in $``$ for $`nn_0`$.
Proofs are given in the next section. In Section 4 we briefly mention some possible extensions of Theorem 2. In Section 5 we discuss consequences for the quoted assertion of Kepler, if interpreted as a container problem (see Corollary 2).
## 3. Proofs
Idea. The proof of Theorem 2 is subdivided into four preparatory steps and corresponding propositions. These technical ingredients are brought together at the end of this section. Given an lc-family $``$ of packing sets, the idea is the following: We show that packing sets $`X`$, with $`|X|`$ sufficiently large, allow the construction of packing sets $`X^{}`$ with $`|X^{}|=|X|`$ and with $`X^{}+\frac{1}{2}B^d`$ fitting into a smaller dilate of $`C`$. Roughly speaking, this is accomplished in two steps. First we show that โrearrangementsโ of spheres near the boundary of $`C`$ are possible for sufficiently large $`n`$. This allows us to obtain arbitrarily large regions in which spheres have no contact, respectively in which points of $`X^{}`$ have distance greater than $`1`$ to all other points (Proposition 2, depending on property (i) of Definition 1). Such an initial modification then allows rearrangements of all spheres (Proposition 3 and 4, depending on property (ii) of Definition 1), so that the resulting packing fits into a smaller dilate of $`C`$. For example, consider a hexagonal packing in the plane: It is sufficient to initially rearrange (or remove) two disks in order to subsequently rearrange all other disks, so that no disk is in contact with others afterwards (see Figure 1, cf. \[Sch02\]).
How do we know that the new sphere packings $`X^{}+\frac{1}{2}B^d`$ fit into a smaller dilate of $`C`$? Consider
$$\lambda (C,X)=\mathrm{min}\{\lambda >0:\lambda C๐+X+\frac{1}{2}B^d\text{ for some }๐^d\}$$
for a fixed finite packing set $`X`$. Here and in the sequel we use $`๐+X`$ to abbreviate $`\{๐\}+X`$. Clearly
$$\lambda (C,n)=\mathrm{min}\{\lambda (C,X):\text{ }X\text{ is a packing set with }|X|=n\text{ }\},$$
and $`\lambda (C,X^{})<\lambda (C,X)`$ whenever the convex hull $`\mathrm{conv}X^{}`$ of $`X^{}`$ (and hence $`X^{}`$ itself) is contained in the interior $`\mathrm{int}\mathrm{conv}X`$ of the convex hull of $`X`$. Thus in order to prove that $`X`$ does not attain $`\lambda (C,|X|)`$ for any convex container $`C`$, it is sufficient to describe a way of attaining a packing set $`X^{}`$ with $`|X^{}|=|X|`$ and
(2)
$$X^{}\mathrm{int}\mathrm{conv}X.$$
I. Let us first consider the โshapesโ of packing sets $`X_n`$ attaining $`\lambda (C,n)`$. Here and in what follows, $`X_n`$ denotes a packing set with $`|X_n|=n`$.
In order to define the โshapeโ, let
$$R(M)=\mathrm{min}\{R0:M๐+RB^d\text{ for some }๐^d\}$$
denote the circumradius of a compact set $`M^d`$ and let $`๐(M)`$ denote the center of its circumsphere. Hence $`M๐(M)+R(M)B^d`$. Then the shape of $`M`$ is defined by
$$๐ฎ(M)=\left(\mathrm{conv}(M)๐(M)\right)/R(M)B^d.$$
The family of nonempty compact subsets in $`^d`$ can be turned into a metric space, for example with the Hausdorff metric (cf. \[Sch93\]). Shapes of packing sets $`X_n`$ attaining $`\lambda (C,n)`$ converge to the shape of $`C`$, that is,
(3)
$$\underset{n\mathrm{}}{lim}๐ฎ\left(X_n\right)=๐ฎ(C).$$
This is seen by โreorganizing elementsโ in a hypothetical convergent subsequence of $`\{X_n\}_n`$ not satisfying (3).
The convergence of shapes leads for growing $`n`$ to shrinking sets of outer (unit) normals
(4)
$$\{๐S^{d1}:๐,๐๐,๐\text{ for all }๐\mathrm{conv}X_n\}$$
at boundary points $`๐`$ of the center polytope $`\mathrm{conv}X_n`$. For general terminology and results on convex polytopes used here and in the sequel we refer to \[Zie97\].
Since $`C`$ is smooth, the sets of outer normals (4) at boundary points of $`\mathrm{conv}X_n`$ become uniformly small for large $`n`$. Also, within a fixed radius around a boundary point, the boundary of $`\mathrm{conv}X_n`$ becomes โnearly flatโ for growing $`n`$.
###### Proposition 1.
Let $`d2`$ and $`C^d`$ a smooth convex body. Let $`\{X_n\}`$ be a sequence of packing sets in $`^d`$ attaining $`\lambda (C,n)`$. Then
1. for $`\epsilon >0`$ there exists an $`n_1`$, depending on $`C`$ and $`\epsilon `$, such that for all $`nn_1`$, outer normals $`๐,๐^{}S^{d1}`$ of $`\mathrm{conv}X_n`$ at $`๐X_n`$ satisfy
$$|๐๐^{}|<\epsilon ;$$
2. for $`\epsilon >0`$ and $`r>0`$ there exists an $`n_1`$, depending on $`C`$, $`\epsilon `$ and $`r`$, such that for all $`nn_1`$, and for $`๐,๐^{}\mathrm{bd}\mathrm{conv}X_n`$ with $`|๐๐^{}|r`$, outer normals $`๐S^{d1}`$ of $`\mathrm{conv}X_n`$ at $`๐`$ satisfy
$$๐,๐๐^{}>\epsilon .$$
II. In what follows we use some additional terminology. Given a packing set $`X`$, we say $`๐X`$ is in a free position, if the set
$$๐ฉ_X(๐)=\{๐X:|๐๐|=1\}$$
is empty. If some $`๐X`$ is not contained in $`\mathrm{int}\mathrm{conv}๐ฉ_X(๐)`$, then it is possible to obtain a packing set $`X^{}=X\{๐\}\{๐^{}\}`$ in which $`๐^{}`$ is in a free position. We say $`๐`$ is moved to a free position in this case (allowing $`๐^{}=๐`$). We say $`๐`$ is moved into or within a set $`M`$ (to a free position), if $`๐^{}M`$. Note, in the resulting packing set $`X^{}`$ less elements may have minimum distance $`1`$ to others, and therefore possibly further elements can be moved to free positions.
Assuming $`X`$ attains $`\lambda (C,|X|)`$ with $`|X|`$ sufficiently large, the following proposition shows that it is possible to move elements of $`X`$ into free positions within an arbitrarily large region, without changing the center polytope $`\mathrm{conv}X`$.
###### Proposition 2.
Let $`d2`$ and $`R>0`$. Let $`C^d`$ a smooth convex body and $``$ a family of packing sets in $`^d`$ satisfying (i) of Definition 1. Then there exists an $`n_2`$, depending on $`R`$, $``$ and $`C`$, such that for all $`X`$ attaining $`\lambda (C,|X|)`$ with $`|X|n_2`$, there exists a $`๐ญ_X^d`$ with
1. $`(๐_X+RB^d)\mathrm{conv}X`$, and
2. all elements of $`X\mathrm{int}(๐_X+RB^d)`$ can be moved to free positions by subsequently moving elements of $`X\mathrm{int}\mathrm{conv}X`$ to free positions within $`\mathrm{int}\mathrm{conv}X`$.
###### Proof.
Preparations. By applying suitable isometries to the packing sets in $``$ we may assume that
(5)
$$\{๐:\text{ }๐X๐\text{ with }|๐|<r\text{ for }๐X\text{ and }X\}$$
has only finitely many accumulation points for every $`r>1`$. For each $`X`$, the container $`C`$ is transformed to possibly different isometric copies. This is not a problem though, since the container is not used aside of Proposition 1, which is independent of the chosen isometries. Note that the smoothness of $`C`$ is implicitly used here.
We say $`๐X`$ is moved in direction $`๐ฏS^{d1}`$, if it is replaced by an $`๐^{}`$ on the ray $`\{๐+\lambda ๐:\lambda _{>0}\}`$. Note that it is possible to move $`๐`$ in direction $`๐S^{d1}`$ to a free position, if
(6)
$$๐ฉ_X(๐,๐)=\{๐๐ฉ_X(๐)๐:๐,๐>0\}$$
is empty. If we want a fixed $`๐X`$ to be moved to a free position, in direction $`๐S^{d1}`$ say, we have to move the elements $`๐๐+๐ฉ_X(๐,๐)`$ first. In order to do so, we move the elements of $`๐+๐ฉ_X(๐,๐)`$ to free positions, and so on. By this we are lead to the definition of the access cone
(7)
$$\text{acc}_{,n}(๐)=\text{pos}\{๐ฉ_X(๐,๐):๐X\text{ for }X\text{ with }|X|n\}$$
of $``$ and $`n`$ in direction $`๐S^{d1}`$. Here,
$$\text{pos}(M)=\{\underset{i=1}{\overset{m}{}}\lambda _i๐_i:m,\lambda _i0\text{ and }๐_iM\text{ for }i=1,\mathrm{},m\text{ }\}$$
denotes the positive hull of a set $`M^d`$, which is by definition a convex cone. Note that $`\text{acc}_{,n}(๐)`$ is contained in the halfspace $`\{๐^d:๐,๐0\}`$ and that $`\text{acc}_{,n}(๐)\text{acc}_{,n^{}}(๐)`$ whenever $`nn^{}`$.
By the assumption that (5) has only finitely many accumulation points for $`r>1`$, there exist only finitely many limits $`lim_n\mathrm{}\left(\text{acc}_{,n}(๐)B^d\right)`$. Here, limits are defined using the Hausdorff metric on the set of nonempty compact subsets of $`^d`$ again.
Strategy. We choose a $`๐S^{d1}`$ such that there exists an $`\epsilon >0`$ with
$$\underset{n\mathrm{}}{lim}\left(\text{acc}_{,n}(๐)B^d\right)=\underset{n\mathrm{}}{lim}\left(\text{acc}_{,n}(๐^{})B^d\right),$$
for all $`๐^{}`$ in the $`\epsilon `$-neighborhood $`S_\epsilon (๐)=S^{d1}(๐+\epsilon B^d)`$ of $`๐S^{d1}`$.
In order to prove the proposition, we show the following for every $`X`$, attaining $`\lambda (C,|X|)`$ with $`|X|`$ sufficiently large: There exists a $`๐_X^d`$ such that
1. $`(๐_X+RB^d)+\text{acc}_{,n}(๐)`$ does not intersect $`X\mathrm{bd}\mathrm{conv}X`$, while
2. $`(๐_X+RB^d)\mathrm{conv}X`$.
It follows that $`\mathrm{bd}\mathrm{conv}X`$ has to intersect the unbounded set
(8)
$$(๐_X+RB^d)+\text{acc}_{,n}(๐)$$
and by the definition of the access cone it is possible to move the elements in $`X\mathrm{int}(๐_X+RB^d)`$ to free positions as asserted. For example, after choosing a direction $`๐^{}S_\epsilon (๐)`$, we may subsequently pick non-free elements $`๐`$ in (8) with maximal $`๐,๐^{}`$. These elements can be moved to a free position within $`\mathrm{int}\mathrm{conv}X`$, since $`๐ฉ_X(๐,๐^{})`$ is empty by the definition of the access cone.
Bounding the boundary intersection. We first estimate the size of the intersection of (8) with $`\mathrm{bd}\mathrm{conv}X`$. For $`๐^{}S_\epsilon (๐)`$ and $`n`$, we consider the sets
$$M(๐^{},n)=\{๐RB^d+\text{acc}_{,n}(๐):๐,๐^{}=R\}.$$
By the definition of the access cones (7), $`M(๐^{},n)M(๐^{},n^{})`$ for $`nn^{}`$. We choose
$$r>sup\{|๐๐|:๐,๐M(๐^{},n)\text{ with }๐^{}S_\epsilon (๐)\},$$
as a common upper bound on the diameter of the sets $`M(๐^{},n)`$ with $`n`$ sufficiently large, say $`nn^{}`$. Note that $`R`$ as well as $``$, $`๐`$ and $`\epsilon `$ have an influence on the size of $`r`$ and $`n^{}`$.
By Proposition 1 (ii) we can choose $`n^{}`$ possibly larger to ensure the following for all $`X`$ attaining $`\lambda (C,|X|)`$ with $`|X|n^{}`$: The intersection of (8) with $`\mathrm{bd}\mathrm{conv}X`$ has a diameter less than $`r`$, no matter which $`๐ญ_X\mathrm{conv}X`$ at distance $`R`$ to $`\mathrm{bd}\mathrm{conv}X`$ we choose. Moreover, $`(๐ญ_X+RB^d)\mathrm{conv}X`$.
Ensuring an empty intersection. It remains to show that for $`X`$, attaining $`\lambda (C,|X|)`$ with $`|X|`$ sufficiently large, $`๐_X`$ can be chosen such that (8) does not intersect $`X\mathrm{bd}\mathrm{conv}X`$. For this we prove the following claim: There exists an $`n^{\prime \prime }`$, depending on $`r`$, $`๐ฏ`$ and $`\epsilon `$, such that for all $`X`$ with $`|X|n^{\prime \prime }`$, there exists a vertex $`๐ฑ`$ of $`\mathrm{conv}X`$ with outer normal $`๐ฏ^{}S_\epsilon (๐ฏ)`$ and
(9)
$$\{๐\}=X(\mathrm{bd}\mathrm{conv}X)(๐+rB^d).$$
Thus these vertices $`๐`$ have a distance larger than $`r`$ to any other element of $`X\mathrm{bd}\mathrm{conv}X`$. Therefore, by choosing $`n_2\mathrm{max}\{n^{},n^{\prime \prime }\}`$, we can ensure that there exists a $`๐_X^d`$ at distance $`R`$ to $`\mathrm{bd}\mathrm{conv}X`$ such that (iโ) and (iiโ) are satisfied for all $`X`$ attaining $`\lambda (C,|X|)`$ with $`|X|n_2`$. Note that $`n^{}`$, $`n^{\prime \prime }`$, and hence $`n_2`$, depend on the choice of $`๐`$ and $`\epsilon `$. But we may choose $`๐`$ and $`\epsilon `$, depending on $``$, so that $`n_2`$ can be chosen as small as possible. In this way we get an $`n_2`$ which solely depends on $`R`$, $``$ and $`C`$.
It remains to prove the claim. Since (5) has only finitely many accumulation points, the set of normals $`๐^{}S^{d1}`$ with hyperplane $`\{๐^d:๐^{},๐=0\}`$ running through $`\mathrm{๐}`$ and an accumulation point $`๐`$ of (5) all lie in the union $`๐ฐ_r`$ of finitely many linear subspaces of dimension $`d1`$. Thus for any $`\delta >0`$ the normals of these hyperplanes all lie in $`๐ฐ_{r,\delta }=๐ฐ_r+\delta B^d`$ if we choose $`|X|`$ sufficiently large, depending on $`\delta `$. By choosing $`\delta `$ small enough, we find a $`๐^{}S_\epsilon (๐)`$ with $`๐^{}๐ฐ_{r,\delta }`$. Moreover, there exists an $`\epsilon ^{}>0`$ such that $`S_\epsilon ^{}(๐^{})๐ฐ_{r,\delta }=\mathrm{}`$. Since every center polytope $`\mathrm{conv}X`$ has a vertex $`๐`$ with outer normal $`๐^{}`$, we may choose $`|X|`$ sufficiently large by Proposition 1 (i) (applied to $`2\epsilon ^{}`$), such that $`\mathrm{conv}X`$ has no outer normal in $`๐ฐ_{r,\delta }`$ at $`๐`$.
Moreover, for sufficiently large $`|X|`$, faces of $`\mathrm{conv}X`$ intersecting $`๐+rB^d`$ can not contain any vertex in $`X(๐+rB^d)`$ aside of $`๐`$. Thus by construction, there exists an $`n^{\prime \prime }`$ such that (9) holds for all $`X`$ with $`|X|n^{\prime \prime }`$. This proves the claim and therefore the proposition. โ
Note that the proof offers the possibility to loosen the requirement on $``$ a bit, for the prices of introducing another parameter: For suitable large $`r`$, depending on $``$, the proposition holds, if
1. there exist isometries $`_X`$ for each $`X`$, such that
$$\{๐๐:๐,๐_X(X)\text{ and }X\text{ }\}$$
has only finitely many accumulation points within $`rB^d`$.
III. For all $`X`$ attaining $`\lambda (C,|X|)`$, with $`|X|`$ sufficiently large, we are able to obtain contact free regions $`(๐_X+RB^d)\mathrm{conv}X`$, with $`R`$ as large as we want, by Proposition 2. That is, we can modify these packing sets $`X`$ by moving elements to free positions within $`\mathrm{int}(๐_X+RB^d)`$. By choosing $`R`$ large enough, such an initial contact free region allows to move further elements to free positions. The following proposition takes care of interior points.
###### Proposition 3.
Let $`d2`$ and $``$ a family of packing sets in $`^d`$ satisfying (ii) in Definition 1 with $`\varrho >0`$. Let $`R\frac{1}{\varrho }`$, $`X`$ and $`๐ฑX\mathrm{int}\mathrm{conv}X`$. Let $`๐ญ^d`$ with $`|๐ญ๐ฑ|R+\frac{\varrho }{2}`$ and with all elements of $`X(๐ญ+RB^d)`$ in a free position. Then $`๐ฑ`$ can be moved to a free position within $`\mathrm{int}\mathrm{conv}X`$.
###### Proof.
Assume $`๐\mathrm{int}\mathrm{conv}๐ฉ_X(๐)`$. By the assumption on $``$,
$$๐+\varrho B^d\mathrm{int}\mathrm{conv}๐ฉ_X(๐).$$
Thus there exists a $`๐๐ฉ_X(๐)`$, such that the orthogonal projection $`๐^{}`$ of $`๐`$ onto the line through $`๐`$ and $`๐`$ satisfies $`|๐^{}๐|\varrho `$ and $`|๐^{}๐|R\frac{\varrho }{2}`$. Then
$$|๐๐|^2=|๐^{}๐|^2+|๐๐^{}|^2\left(R\frac{\varrho }{2}\right)^2+\left(1\varrho ^2\right)<R^2.$$
Thus $`๐`$ is in a free position by the assumptions of the proposition, which contradicts $`๐๐ฉ_X(๐)`$. โ
IV. After Propositions 2 and 3 it remains to take care of points in $`X\mathrm{bd}\mathrm{conv}X`$, for $`X`$ attaining $`\lambda (C,|X|)`$, and with $`|X|`$ sufficiently large. It turns out that these points can all be moved to free positions within $`\mathrm{int}\mathrm{conv}X`$. As a consequence we obtain the following.
###### Proposition 4.
Let $`d2`$, $`C^d`$ a smooth convex body and $``$ a family of packing sets in $`^d`$ satisfying (ii) of Definition 1. Then there exists an $`n_4`$, depending on $`C`$ and $``$, such that $`X`$ with $`|X|n_4`$ does not attain $`\lambda (C,|X|)`$, if all elements of $`X\mathrm{int}\mathrm{conv}X`$ are in a free position.
###### Proof.
Let $`\varrho >0`$ as in (ii) of Definition 1. We choose $`n_4`$ by Proposition 1 (ii), applied to $`\epsilon =\varrho `$ and $`r=1`$. Assume $`X`$ with $`|X|n_4`$ attains $`\lambda (C,|X|)`$ and all elements of $`X\mathrm{int}\mathrm{conv}X`$ are in a free position. We show that every element $`๐X\mathrm{bd}\mathrm{conv}X`$ can be moved to a free position into $`\mathrm{int}\mathrm{conv}X`$. This gives the desired contradiction, because after moving (in an arbitrary order) all $`X\mathrm{bd}\mathrm{conv}X`$ to free positions into $`\mathrm{int}\mathrm{conv}X`$, we obtain a packing set $`X^{}`$ with $`|X^{}|=|X|`$ and $`X^{}\mathrm{int}\mathrm{conv}X`$.
It is possible to move a given $`๐X\mathrm{bd}\mathrm{conv}X`$ to a free position $`๐^{}=๐+\delta ๐`$ for a (sufficiently small) $`\delta >0`$, if $`๐S^{d1}`$ is contained in the non-empty polyhedral cone
$$C_๐=\{๐^d:๐,๐๐0\text{ for all }๐๐ฉ_X(๐)\}.$$
If $`๐C_๐`$ can be chosen, so that $`๐^{}\mathrm{int}\mathrm{conv}X`$, the assertion follows. Otherwise, because $`C_๐`$ and $`\mathrm{conv}X`$ are convex, there exists a hyperplane through $`๐`$, with normal $`๐S^{d1}`$, which separates $`\mathrm{conv}X`$ and $`๐+C_๐`$. That is, we may assume that
$$๐\mathrm{pos}\left\{๐๐:๐๐ฉ_X(๐)\right\}$$
and $`๐`$ is an outer normal of $`\mathrm{conv}X`$ at $`๐`$.
Then for some $`\delta >0`$, there exists a point $`๐=๐+\delta ๐\mathrm{bd}\mathrm{conv}๐ฉ_X(๐)`$, which is a convex combination of some $`๐_1,\mathrm{},๐_k๐ฉ_X(๐)`$. That is, there exist $`\alpha _i0`$ with $`_{i=1}^k\alpha _i=1`$ and $`๐=_{i=1}^k\alpha _i๐_i`$ . Therefore
$$\delta =๐๐,๐=\underset{i=1}{\overset{k}{}}\alpha _i๐_i๐,๐<\varrho ,$$
because $`๐_i๐,๐<\varrho `$ due to $`|X|n_4`$ and $`๐_i\mathrm{bd}\mathrm{conv}X`$. This contradicts the assumption on $``$ with respect to $`\varrho `$ though. โ
Finish. The proof of Theorem 2 reduces to the application of Propositions 1, 2, 3 and 4. Let $``$ be an lc-family of packing sets in $`^d`$, with a $`\varrho >0`$ as in (ii) of Definition 1. We choose $`R1/\varrho `$ and $`n_2`$ and $`n_4`$ according to Propositions 2 and 4. By Proposition 1 (ii), we choose $`n_1`$ such that packing sets $`X`$ attaining $`\lambda (C,|X|)`$ with $`|X|n_1`$ satisfy the following: For each $`๐X`$, there exists a $`๐^d`$ with $`|๐๐|=R+\frac{\varrho }{2}`$ and $`๐+RB^d\mathrm{conv}X`$.
We choose $`n_0\mathrm{max}\{n_1,n_2,n_4\}`$ and assume that $`X`$ with $`|X|n_0`$ attains $`\lambda (C,|X|)`$. By Proposition 2 we can modify the packing set $`X`$ to obtain a new packing set $`X^{}`$ with a contact free region $`(๐_X+RB^d)\mathrm{int}\mathrm{conv}X`$, and with the same points $`X^{}\mathrm{bd}\mathrm{conv}X^{}=X\mathrm{bd}\mathrm{conv}X`$ on the boundary of the center polytope $`\mathrm{conv}X^{}=\mathrm{conv}X`$.
The following gives a possible order, in which we may subsequently move non-free elements $`๐X\mathrm{int}\mathrm{conv}X`$ to free positions: By the choice of $`n_0`$ we can guarantee that for each $`๐X\mathrm{int}\mathrm{conv}X`$, there exists a $`๐`$ with $`|๐๐|R+\frac{\varrho }{2}`$ and $`๐+RB^d\mathrm{conv}X`$. Let $`๐_๐`$ be the $`๐`$ at minimal distance to $`๐_X`$. Then among the non-free $`๐\mathrm{int}\mathrm{conv}X`$, the one with minimal distance $`|๐_๐๐_X|`$ satisfies the assumptions of Proposition 3, because a non-free element $`๐X(๐_๐+B^d)`$ would satisfy $`|๐_๐๐_X|<|๐_๐๐_X|`$ due to $`\mathrm{conv}\{๐_๐,๐_X\}+B^d\mathrm{conv}X`$.
Thus by Proposition 3 we can subsequently move the non-free elements within $`X\mathrm{int}\mathrm{conv}X`$ to free positions. By this we obtain a contradiction to Proposition 4, which proves the theorem.
The lattice packing case. We end this section with the proof of Theorem 1. We may apply Theorem 2 after showing that the family of solutions to the lattice restricted container problem is of limited complexity. The space of lattices can be turned into a topological space (see \[GL87\]). The convergence of a sequence $`\{\mathrm{\Lambda }_n\}`$ of lattices to a lattice $`\mathrm{\Lambda }`$ in particular involves that sets of lattice points within radius $`r`$ around a lattice point tend to translates of $`\mathrm{\Lambda }rB^d`$ for growing $`n`$. As a consequence, a convergent sequence of packing lattices, as well as subsets of them, form an lc-family. Solutions to the lattice restricted container problem tend for growing $`n`$ towards subsets of translates of densest packing lattices (see \[Zon99\]). These lattices are the solutions of the lattice (sphere) packing problem. Up to isometries, there exist only finitely many of these lattices in each dimension (see \[Zon99\]). Thus the assertion follows, since a finite union of lc-families is an lc-family.
## 4. Extensions
Let us briefly mention some possible extensions of Theorem 2. These have been treated in \[Sch02\] for the $`2`$-dimensional case and could be directions for further research.
Packings of other convex bodies. Instead of sphere packings, we may consider packings $`X+K`$ for other convex bodies $`K`$. If the difference body $`DK=KK`$ is strictly convex, then the proofs can be applied after some modifications: Instead of measuring distances with the norm $`||`$ given by $`B^d`$, we use the norm $`|๐|_{DK}=\mathrm{min}\{\lambda >0:\lambda ๐DK\}`$ given by $`DK`$. The strict convexity of $`DK`$ is then used for the key fact, that elements $`๐`$ of a packing set $`X`$ can be moved to a free position, whenever they are not contained in $`\mathrm{int}\mathrm{conv}๐ฉ_X(๐)`$ (see II in Section 3). Note though that the sets in (6) and depending definitions have to be adapted for general convex bodies.
Packings in other containers. The restriction to smooth convex containers simplifies the proof, but we strongly believe that Theorem 2 is valid for other containers as well, e.g. certain polytopes. On the other hand there might exist containers for which Theorem 2 is not true. In particular in dimension $`3`$ it seems very likely that Theorem 2 is not true for polytopal containers $`C`$ with all their facets lying in planes containing hexagonal sublattices of the fcc lattice (see Section 5). That is, for these polytopal containers $`C`$ we conjecture the existence of infinitely many $`n`$, for which subsets of the fcc lattice attain $`\lambda (C,n)`$. An example for at least โlocal optimalityโ of sphere packings (with respect to differential perturbations) in suitable sized tetrahedra was given by Dauenhauer and Zassenhaus \[DZ87\]. A proof of โglobal optimalityโ seems extremely difficult though, as it would provide a new proof of the sphere packing problem (โKepler conjectureโ, see Section 5).
Other finite packing problems. Similar โphenomenaโ occur for other packing problems. For example, if we consider finite packing sets $`X`$ with minimum diameter or surface area of $`\mathrm{conv}X`$, or maximum parametric density with large parameter (cf. \[FCG91\], \[BHW94\], \[Bรถr04\], \[BP05\]). This is due to the fact that the shapes of solutions tend to certain convex bodies, e.g. a sphere.
## 5. Keplerโs assertion
Keplerโs statement, quoted in the introduction, was later referred to as the origin of the famous sphere packing problem known as the Kepler conjecture (cf. e.g. \[Hal02\] p.5, \[Hsi01\] p.4). In contrast to the original statement, this problem asks for the maximum sphere packing density (see (10) below) of an infinite arrangement of spheres, where the โcontainerโ is the whole Euclidean space. As a part of Hilbertโs famous problems \[Hil01\], it attracted many researchers in the past. Its proof by Hales with contributions of Ferguson (see \[Hal02\], \[Hal05\], \[Hal06\]), although widely accepted, had been a matter of discussion (cf. \[Lag02\], \[Szp03\], \[FL06\]).
Following Kepler \[Kep11\], the cubic or hexagonal close packings in $`^3`$ can be described via two dimensional layers of spheres, in which every sphere center belongs to a planar square grid, say with minimum distance $`1`$. These layers are stacked (in a unique way) such that each sphere in a layer touches exactly four spheres of the layer above and four of the layer below.
The packing attained in this way is the well known face centered cubic (fcc) lattice packing. We can build up the fcc lattice by planar hexagonal layers as well, but then there are two choices for each new layer to be placed, and only one of them yields an fcc lattice packing. All of them, including the uncountably many non-lattice packings, are referred to as hexagonal close packings (hc-packings). Note that the family of hc-packings is of limited complexity, because up to isometries they can be built from a fixed hexagonal layer.
Let
$$n(C)=\mathrm{max}\{|X|:CX+\frac{1}{2}B^d\text{ is a packing }\}.$$
Then in our terminology Kepler asserts that, in $`^3`$, $`n(C)`$ is attained by hc-packings. His assertion, if true, would imply an โanswerโ to the sphere packing problem (Kepler conjecture), namely that the density of the densest infinite sphere packing
(10)
$$\delta _d=\underset{\lambda \mathrm{}}{lim\; sup}\frac{n(\lambda C)\mathrm{vol}(\frac{1}{2}B^d)}{\mathrm{vol}(\lambda C)}$$
is attained by hc-packings for $`d=3`$; hence $`\delta _3=\pi /\sqrt{18}`$. Note that this definition of density is independent of the chosen convex container $`C`$ (see \[Hla49\] or \[GL87\]).
As a consequence of Theorem 2, Keplerโs assertion turns out to be false, even if we think of arbitrarily large containers. Consider for example the containers $`\lambda (C,n)C`$ for $`nn_0`$.
###### Corollary 1.
Let $`d2`$, $`C^d`$ a smooth convex body and $``$ an lc-family of packing sets in $`^d`$. Then there exist arbitrarily large $`\lambda `$ such that $`n(\lambda C)`$ is not attained by packing sets in $``$.
We may as well think of arbitrarily small spheres packed into a fixed container $`C`$. For $`r>0`$, we call $`X+rB^d`$ a sphere packing if distinct elements $`๐`$ and $`๐^{}`$ of $`X`$ have distance $`|๐๐^{}|2r`$. Specializing to $`^3`$, the following corollary of Theorem 2 refers directly to Keplerโs assertion.
###### Corollary 2.
Let $`C^3`$ a smooth convex body. Then there exist arbitrarily small $`r>0`$, such that
$$\mathrm{max}\{|X|:CX+rB^d\text{ is a packing }\}$$
is not attained by fcc or hexagonal close packing sets.
## Acknowledgments
I like to thank Thomas C. Hales, Tyrrell B. McAllister, Frank Vallentin, Jรถrg M. Wills, Gรผnter M. Ziegler and the two anonymous referees for many helpful suggestions.
|
warning/0506/cond-mat0506481.html
|
ar5iv
|
text
|
# Cell Dynamics Simulation of Kolmogorov-Johnson-Mehl-Avrami Kinetics of Phase Transformation
## 1 Introduction
Phase transformation occurs by the nucleation and subsequent growth of a nucleus in a system where the first-order phase transformation takes place. It has attracted much attention for more than a half century from a fundamental point of view as well as from technological interests. These include the mechanical properties of metallic materials , the recrystallization of deformed metals , and the manufacturing of basic thin-film transistor devices, such as solar cells and active matrix-addressed flat-panel displays .
The nucleation and growth processes are often described in terms of the old standard theory called the KJMA theory developed by Kolmogorov , Johnson and Mehl , and Avrami . According to this theory, the time evolution of the volume fraction of a new transformed phase follows the linear KJMA plot with the integral Avrami exponent $`n`$ that is given by the slope. However, it is recognized that this theory often fails to explain experimental results ; neither the KJMA plot becomes linear nor the Avrami exponent becomes an integer. There is also some debate about the validity of the assumption used in the theory . To resolve the discrepancy, a realistic yet efficient simulation method that could take various factors into account is indispensable.
The direct atomic-scale computer simulation of the kinetics of phase transformation using the molecular dynamics or Monte Carlo method is still a difficult task. Even the most fundamental phenomenon, such as nucleation, is not easy to study using these methods. Instead, the problem of phase transformation has been tackled using a coarse-grained Ginzburg-Landau-type model called the Cahn-Hilliard , Ginzburg-Landau or phase-field model, which requires the solution of highly nonlinear partial differential equations. Since this model requires the time integration of the partial differential equations, it is not easy to simulate the long-time behavior of the dynamics of phase transformation except for special traveling wave solutions .
A model based on a cellular automaton instead of a partial differential equation can improve the efficiency of numerical integration . It has been used to test the KJMA theory in the recrystallization of metals. Although the cellular automaton is sufficiently flexible to implement the various local reactions of recrystallization, it does not have a direct connection to the equilibrium phase diagram. Therefore, the connection between the phase diagram and the phase transformation is not so clear compared to the Ginzburg-Landau-type model mentioned above.
In this study, instead, we use the cell dynamics method to study the validity of the KJMA theory. This method is attractive because it has the merit of cellular automata and is computationally efficient, and yet it keeps the connection to the phase diagram through the Landau-type free energy. The format of this paper is as follows: in ยง2, we review the cell dynamics method and present the necessary modification for studying the nucleation and growth. In ยง3, we follow closely the work by Jou and Lusk and test the validity of the KJMA theory using the cell dynamics method. We conclude in section 4.
## 2 Cell Dynamics Method for Nucleation and Growth
To study the phase transformation, it is customary to study the partial differential equation called the phase-field model which is equivalent to the time-dependent Ginzburg-Landau (TDGL) or Cahn-Hilliard model :
$$\frac{\psi }{t}=\frac{\delta }{\delta \psi },$$
(1)
where $`\delta `$ denotes the functional differentiation, $`\psi `$ is the nonconserved order parameter, and $``$ is the free energy functional. This free energy is usually written as the square-gradient form
$$[\psi ]=\frac{1}{2}\left[D(\psi )^2+h(\psi )\right]d๐ซ.$$
(2)
The local part $`h(\psi )`$ of the free energy $``$ determines the bulk phase diagram and the value of the order parameter in equilibrium phases. The double-well form was frequently used for $`h(\psi )`$ to express the two-phase coexistence and study the phase transformation between these two phases.
This TDGL equation (1) for the nonconserved order parameter $`\psi `$ was loosely transformed into a space-time discrete cell dynamics equation by Puri and Oono following a similar transformation of the kinetic equation for the conserved order parameter called the Cahn-Hilliard-Cook equation . In their cell dynamics method, the partial differential equation (1) is replaced by a finite difference equation in space and time in the form
$$\psi (t+1,n)=F[\psi (t,n)],$$
(3)
where the time $`t`$ is discrete and an integer, and the space is also discrete and is expressed by the integral site index $`n`$. The mapping $`F`$ is given by
$$F[\psi (t,n)]=f(\psi (t,n))+[\psi (t,n)\psi (t,n)],$$
(4)
where $`f(\psi )=dh(\psi )/d\psi `$, and the definition of $`\mathrm{}`$ for a two-dimensional square grid is given by
$$\psi (t,n)=\frac{1}{6}\underset{i=\text{nn}}{}\psi (t,i)+\frac{1}{12}\underset{i=\text{nnn}}{}\psi (t,i),$$
(5)
where โnnโ denotes nearest neighbors and โnnnโ next-nearest neighbors. An improved form of this mapping for a three-dimensional case was also obtained .
Oono and Puri further approximated the derivative of the local free energy $`f(\psi )`$ called the โmap functionโ in the $`\mathrm{tanh}`$ form
$$f(\psi )=\frac{dh}{d\psi }\psi A\mathrm{tanh}\psi ,$$
(6)
with $`A=1.3`$ that corresponds to the free energy
$$h(\psi )=A\mathrm{ln}\left(\mathrm{cosh}\psi \right)+\frac{1}{2}\psi ^2$$
(7)
and is the lowest order ($`O(\psi ^2)`$) approximation to the double well form of the free energy
$$h(\psi )=\frac{1}{2}\psi ^2+\frac{1}{4}\psi ^4$$
(8)
when $`A=1.5`$. They used this simplification since this cell dynamics system is invented not to simulate the mathematical TDGL partial differential equation but to simulate and describe the behavior of nature directly. Later, Chakrabarti and Brown discussed that this simplification is justified since the detailed form of the double-well potential $`h(\psi )`$ is irrelevant to the long-time dynamics and the scaling exponent.
Subsequently, however, several authors used the map function $`f(\psi )`$ directly obtained from the free energy $`h(\psi )`$ as it is and found that the cell dynamics equation (3) is still tractable numerically. Ren and Hamley argued that one can easily include the effect of the asymmetry of the free energy and the asymmetric characteristic of two phases using the original form of the free energy function $`f(\psi )`$.
Despite the popularity of this cell dynamics method in the soft-condensed matter community as a simulator of pattern formation due to various factors, it has not yet been used to study the most fundamental problem of phase transformation by nucleation and growth. In the next section, we use a parameterized free energy function, study the kinetics of phase transformation, and test the validity of the KJMA theory.
## 3 Numerical Results
### 3.1 Two-dimensional growth of single domain
To simulate the growing stable phase after the nucleation, we have to prepare the system in a state where one phase is metastable and has the higher free energy than the other stable phase. The free energy difference between the stable and metastable phases is determined from the supersaturation in liquid condensation and from the undercooling in crystal nucleation. Microscopically, this free energy difference is necessary for the nucleus to overcome the additional curvature effect caused by the interfacial tension and to continue to grow .
To study the growth of the stable phase using the cell dynamics method, we consider the time-dependent Ginzburg-Landau (TDGL) equation (1) with the square gradient free energy functional (2). The local part of the free energy $`h(\psi )`$ we used is
$$h(\psi )=\frac{1}{4}\eta \psi ^2(1\psi )^2+\frac{3}{2}ฯต\left(\frac{\psi ^3}{3}\frac{\psi ^2}{2}\right).$$
(9)
This free energy is shown in Fig. 1, where one phase at $`\psi =0`$ is metastable while another phase at $`\psi =1`$ is stable. The free energy difference $`\mathrm{\Delta }h`$ between the stable phase at $`\psi =1`$ and the metastable phase at $`\psi =0`$ is solely determined from the parameter $`ฯต`$:
$$\mathrm{\Delta }h=h(\psi =0)h(\psi =1)=\frac{ฯต}{4}.$$
(10)
Therefore, $`ฯต`$ represents the supersaturation or the undercooling.
The metastable phase at $`\psi =0`$ becomes unstable when $`\eta =3ฯต`$, which defines the spinodal. The height $`\mathrm{\Delta }E`$ of the free energy barrier at $`\psi =(\eta 3ฯต)/2\eta `$ can be tuned by the parameters $`\eta `$ and $`ฯต`$:
$$\mathrm{\Delta }E=h\left(\psi =\frac{\eta 3ฯต}{2\eta }\right)h(\psi =0)=\frac{\eta ^44\eta ^3ฯต+27ฯต^4}{32\eta ^3},$$
(11)
which vanishes when $`\eta =3ฯต`$ at the spinodal.
The steady-state analytical solution of the TDGL with a constant interfacial velocity has been obtained in one dimension by Chan when the free energy $`h(\psi )`$ is written in the quadratic form such as in eq. (9). Using Chanโs formula, the interfacial velocity $`v`$ of our TDGL model (1), (2) with the free energy (9) is given by
$$v=\sqrt{\frac{D}{2\eta }}3ฯต=\sqrt{\frac{D}{2\eta }}12\mathrm{\Delta }h.$$
(12)
Chan further suggested that if the interfacial width is small, the interfacial velocity of a two-dimensional circular or three-dimensional spherical growing nucleus is asymptotically given by eq. (12) of the one-dimensional model. The larger the free energy difference $`ฯต`$ and the lower the free energy barrier $`\eta `$ are, the higher the front velocity $`v`$ becomes.
The critical radius $`R_c`$ of a two-dimensional circular nucleus is also given analytically :
$$R_c=\frac{D}{v}=\frac{\sqrt{2\eta D}}{3ฯต}.$$
(13)
In the metastable phase, the circular nucleus of the stable phase with a radius ($`R`$) smaller than $`R_c`$ shrinks, while that with a radius larger than $`R_c`$ grows and its front velocity approaches eq. (12). Again, the larger the free energy difference $`ฯต`$ and the lower the free energy barrier $`\eta `$ are, the smaller the critical radius $`R_c`$ is.
We implemented the above free energy (9) into the cell dynamics code written by Mathematica TM for the animation of spinodal decomposition developed by Gaylord and Nishidate , and simulated the growth of a single circular nucleus of a stable phase.
Initially, we prepared a small circular nucleus of a stable phase within a metastable phase and simulated the growth of that nucleus. The system size is 100$`\times `$100 cells, $`D`$ in eq. (2) is $`D=0.5`$, and the periodic boundary condition is imposed. The initial order parameter $`\psi `$ is randomly chosen from $`0.9\psi 1.1`$ for the circular nucleus of the stable phase and from $`0.1\psi 0.1`$ for the metastable environment. The diameter of the initial nucleus is fixed at $`d=11`$. Therefore, the initial nucleus occupies a part of 11$`\times `$11 cells. The effective area of the stable phase is computed by counting the number of cells with the order parameter $`\psi >1/2`$.
Figure 2 shows the effective radius of the circular nucleus of the stable phase calculated from the effective area of the nucleus as a function of time step $`t`$. The nearly linear growth of the nucleus of the stable phase is clearly visible, which indicates a constant front velocity for the stable-metastable interface predicted from the analytical solution for the TDGL . The velocities $`v`$ estimated from Fig. 2 and predicted from eq. (12) are compared in Table 1 together with the critical radius $`R_c`$ calculated from eq. (13). It can be seen that eq. (12) correctly predicts the general trend of the front velocity $`v`$ when the two parameters $`\eta `$ and $`ฯต`$ are altered. Since the cell dynamics method does not solve the TDGL directly, the discrepancy between cell dynamics simulation and theoretical prediction (12) by a factor of roughly 2 is not very significant.
From the above comparison of the values estimated by cell dynamics simulation and theoretical prediction using the steady-state solution of the TDGL for a two-phase system, we consider that this cell dynamics method is effective for studying the phase transformation by the growth of the multiple nucleus of the stable phase.
### 3.2 KJMA kinetics by cell dynamics simulation
#### 3.2.1 Site-saturation nucleation
In site-saturation nucleation, a fixed number of nuclei are prepared initially, and subsequent growth is monitored. The KJMA theory gives an analytical expression for the volume fraction $`f`$ of the stable phase as a function of time $`t`$. In two dimensions, the formula leads to
$$f=1\mathrm{exp}\left(\pi n_0v^2\left(t+t_0\right)^2\right),$$
(14)
where $`n_0`$ is the number density (number per unit area) of the randomly distributed initial nuclei. $`v`$ is the growth rate of the radius of each nucleus discussed in the previous section. $`t_0`$ is the origin of time which can hopefully take the incubation time of nucleation into account .
From eq. (14), we have
$$\mathrm{log}\left(\mathrm{ln}(1f)\right)=2\mathrm{log}\left(t+t_0\right)+\text{constant}.$$
(15)
Therefore, the KJMA theory predicts that a double logarithms $`\mathrm{log}\left(\mathrm{ln}(1f)\right)`$ versus $`\mathrm{log}\left(t+t_0\right)`$ is a straight line that is known as the KJMA plot with the integral tangent $`n=2`$, which is called the โAvrami exponentโ.
We have simulated the site-saturation nucleation using the cell dynamics method. Now a finite number of nuclei of the stable phase is distributed over the area we considered. The initial nuclei are circular and have the diameter $`d=8`$, which is larger than the critical radius $`R_c`$ in Table 1. Then, the evolution of the transformed volume is monitored as a function of time. Again, we have considered the 100$`\times `$100 system and introduced a finite number ($`N_0=20`$) of nuclei as the initial condition. Therefore, the number density of the initial nucleus is $`n_0=20/10000=0.002`$.
The time evolution of the transformed volume $`f`$ is plotted as a function of time $`t`$ in Fig. 3. When the effect of the incubation time with $`t_0=10`$ is included, a better agreement between the simulation and theoretical results is attained. This incubation time $`t_0=10`$ is the time necessary for a infinitely small nucleus to become a larger nucleus with the diameter $`d=8`$ in our simulation, and is estimated by fitting the theoretical curve (14) to the simulation data. Since an infinitely small nucleus cannot grow because it is smaller than the critical nucleus, we use the terminology โincubation timeโ to indicate both the time necessary for a critial nucleus to appear and the time necessary for it to grow to be a larger nucleus.
The KJMA plot of the double logarithm of the volume fraction $`f`$ is shown as a function of $`\mathrm{log}t`$ in Fig. 4, where we ignore the effect of the incubation time and set $`t_0=0`$. The time evolutions for several combinations of the potential parameters $`\eta `$ and $`ฯต`$ are shown. They do not fit the expected straight lines.
Figure 5 shows the KJMA plot when the incubation time $`t_0`$ is considered. Again, the incubation time $`t_0`$ is estimated by fitting the theoretical curve (14) to the simulation data. Now, the time evolutions for several combinations of the potential parameters $`\eta `$ and $`ฯต`$ all fit the straight lines with almost the same Avrami exponent $`n2`$, which is very close to the theoretical prediction, as shown in Table 2. The results in Figs. 4 and 5 clearly suggest that the incubation time $`t_0`$ should be carefully taken into account to deduce the Avrami exponent $`n`$ when we analyze experimental as well as simulation data.
Figure 6 shows the evolution of the morphology of the two-dimensional system for the site-saturation nucleation when $`\eta =0.4`$ and $`ฯต=0.1`$. We observe the almost isotropic growth of every nucleus of the stable phase. At the time step $``$100, almost all cells are transformed into the stable phase.
#### 3.2.2 Continuous nucleation
In the continuous nucleation, a new nuclear embryo is continuously introduced. The KJMA theory of continuous nucleation gives the analytical expression for the volume fraction $`f`$ of the growing stable phase. In two dimensions, it leads to
$$f=1\mathrm{exp}\left(\frac{\pi \dot{n}v^2}{3}\left(t+t_0\right)^3\right),$$
(16)
where $`\dot{n}`$ is the steady nucleation rate per unit area and $`v`$ is the growth rate of the single nucleus discussed in the previous section.
From eq. (16), we have
$$\mathrm{log}\left(\mathrm{ln}(1f)\right)=3\mathrm{log}\left(t+t_0\right)+\text{constant}.$$
(17)
Therefore, a double logarithmic KJMA plot should give the โAvrami exponentโ $`n=3`$ instead of $`n=2`$ of the site-saturation nucleation.
We have also simulated the continuous nucleation using the cell dynamics method. In our simulation, a constant nucleation rate $`\dot{n}`$ is achieved by introducing a new nucleus every $`1/\dot{n}`$ time step (nucleation time). At each nucleation time step, a position within the two-dimensional area is randomly selected. If the position is already occupied by the stable phase, no new nucleus is placed. If the position is not occupied by the stable phase, a new nucleus is placed and allowed to grow there. In this simulation, we have used a larger 200$`\times `$200 system to avoid the finite-size effect as much as possible. The steady nucleation rate $`\dot{n}=0.1/40000`$ is used. Therefore, a single nucleus is produced at every 10 time steps in the area 200$`\times `$200.
The time evolution of the transformed volume $`f`$ is plotted as a function of time $`t`$ in Fig. 7 as the double logarithmic KJMA plot. The time evolutions for several combinations of the potential parameters $`\eta `$ and $`ฯต`$ show almost the same straight line with the Avrami exponent very close to the theoretically predicted $`n=3`$, as shown in Table 3.
Figure 8 shows the evolution of the morphology of the two-dimensional systems for the continuous nucleation when $`\eta =0.4`$ and $`ฯต=0.1`$. Because a nucleus is continuously produced, almost all cells are occupied at the later stage, and then the production of a nucleus stops. The situation becomes closer to the site-saturation nucleation. The Avrami exponent $`n`$ smaller than the theoretically predicted $`n=3`$ for the continuous nucleation but closer to the site saturation nucleation $`n=2`$ is expected. This finite-size effect is one of the reasons why the Avrami exponent estimated from the simulation data in Table 3 is always smaller than the theoretically predicted $`n=3`$.
There is also a problem of incubation time in continuous nucleation. In our simulation, we have introduced a fairly large nucleus, which is sufficiently large to grow continuously. Thus, the same problem of time origin or incubation time $`t_0`$ as that for the site-saturation nucleation could occur. Since a nucleus is continuously produced, we could not incorporate the effect of incubation time in a reasonable manner in our analysis.
In continuous nucleation, we found that the KJMA theory correctly describes the overall behavior of the time dependence of the transformed volume fraction $`f`$. However, there is a small inflection in the slope of the KJMA plot at around $`\mathrm{log}(\mathrm{ln}(1f))0.0`$, which can be explained by the impingement in which the growing circular grains collide with each other. The volume fraction $`f`$ when the impingement starts to occur is roughly estimated to be $`f=\pi r^2/4r^20.79`$ using the ratio of the area of a circular grain with the radius $`r`$ to that of a square with the side length $`2r`$. Thus, the inflection of the KJMA straight line is expected at
$$\mathrm{log}(\mathrm{ln}(1f))=\mathrm{log}(\mathrm{ln}(0.21))0.19,$$
(18)
which is very close to the point where the inflection is actually observed in Fig. 7. The same effect was discussed by Jou and Lusk . The other factors affecting the Avrami exponent are discussed in the next subsection.
### 3.3 Discussion
Experimentally, a reasonably linear KJMA behavior was observed in the recrystallization of some metals and in the crystallization of metallic glasses . However, a considerable variation in Avrami exponent $`n`$ defined by
$$\mathrm{log}\left(\mathrm{ln}(1f)\right)=n\mathrm{log}\left(t+t_0\right)+\text{constant}$$
(19)
was observed from the electrical resistivity and differential scanning calorimetry (DSC) data.
Following the argument of Christian , Price proposed a formula for the Avrami exponent
$$n=a+b(1q),$$
(20)
where $`a`$ is the nucleation component; $`a=0`$ for the site saturation and $`a=1`$ for the continuous nucleation. $`b`$ defines the dimensionality of the growth ($`b=3`$ for a three-dimensional problem and $`b=2`$ for our two-dimensional problem). The exponent $`q`$ includes contributions from various types of power-law reaction.
For example, in our cell dynamics model based on the Landau-type free energy (9), the driving force of phase transformation comes from the undercooling defined by eq. (10) that leads to the linear time-dependent growth of a circular nucleus with the constant front velocity $`v`$ given by eq. (12). However, as more materials are transformed into the stable phase, the driving force decreases somehow and the front velocity $`v`$ is expected to decelerate from eq. (12). Thus, it is reasonable to assume a power-law decay of the front velocity
$$vt^q,$$
(21)
which exactly gives eq. (20). Hence, the Avrami exponent becomes smaller than the KJMA predicted $`n=a+b`$, as observed in our numerical simulations for the continuous nucleation.
There are many other factors affecting the exponent $`n`$. Possible reasons for the nonideal exponent and even the nonlinear growth in the KJMA plot include the nonrandomness of the nucleation site and the preferential nucleation, for example, at the grain boundary , the effect of the time dependence of the nucleation rate and so forth. The net result of these effects is a negative deviation from the KJMA linear plot , which leads to again the smaller exponent $`n`$ in accordance to the many experimental results and our simulation. In our cell dynamics method, these effects can be easily included by changing the probability of selecting the nucleation rate from cell to cell. Hesselbarth and Gรถbel have included such effects in their cellular automaton and could successfully explain the deviation of experimental data from KJMA predicted data.
There is also a problem of two-stage crystallization . In some alloys, the KJMA plot shows an inflection in which the exponent $`n`$ changes markedly from a large value at an early stage to a small value at a later stage. However, our simulation data shown in Fig. 7 shows the opposite trend; the exponent $`n`$ is large at the later stage. This phenomena is explained by assuming that the early stage corresponds to the continuous nucleation and that the later stage corresponds to the site saturation because of the exhaustion of the nucleation site . Recent theoretical model calculation supports this two-stage nucleation model .
Our cell dynamics method could easily incorporate such a two-stage transformation by assuming that the continuous nucleation terminates at a certain stage. Then, the growth process is continuous nucleation with the exponent $`n=3`$ in the early stage, but it becomes site saturation with the exponent $`n=2`$ in the later stage. In our cell dynamics method, it is not necessary to assume a discrete lattice and is easier to incorporate various modifications to KJMA kinetics. Using our cell dynamics method, a more quantitative study is feasible in the future.
## 4 Conclusion
In this study, we used a cell dynamics method to test the validity of the Kolmogorov-Johnson-Mehl-Avrami (KJMA) kinetic theory of phase transformation. First, we used this method to study the growth of a single circular nucleus and found that the nucleus grows with a constant front velocity in accordance to the analytical solution . Next, we used the cell dynamics method to simulate the growth of an ensemble of nuclei under the conditions of both the site saturation and continuous nucleation. We found a nearly linear behavior of the KJMA plot with the Avrami exponent close to the KJMA predicted value. Finally, we suggested several extensions of the cell dynamics method to study various contributions that may lead to the nonlinear KJMA behavior or nonideal Avrami exponent.
The results obtained in this study are summarized as follows:
* The cell dynamics method with a realistic free energy can succesfully simulate the steady growth of a single nucleus and confirm the prediction of Chan based on the time-dependent Ginzburg-Landau (TDGL) equation.
* It can also simulate the growth of multiple nuclei and confirm the time evolution of the volume fraction of the transformed material predicted from the KJMA kinetic theory and numerical simulation using TDGL .
* Therefore, the cell dynamics method can be used to simulate more complex scenarios of nucleation and growth.
* Our simulation indicates that the incubation time should be carefully taken into account when we deduce the Avrami exponent from the experimental and simulation data.
The cell dynamics method is similar to the time-dependent Ginzbug-Landau or Cahn-Hilliard model based on the free energy functional. In contrast to the conventional cellular automaton approach to the phase transformation , no phenomenological energy that induces phase transformation is necessary. Therefore, the cell dynamics method is numerically efficient as a cellular automaton, yet it keeps the direct connection to the equilibrium phase diagram. This cell dynamics method can be used to test various scenarios of nucleation and growth in a unified manner.
|
warning/0506/astro-ph0506587.html
|
ar5iv
|
text
|
# Fundamental Parameters of Low Mass X-ray Binaries II: X-Ray Persistent Systems
## 1 Introduction
Sco X-1 was the first Galactic X-ray source discovered by X-ray satellites and its powerful X-ray luminosity explained through mass accretion into a degenerate star . Sco X-1 became the prototype of a new class of objects, the low mass X-ray binaries (hereafter LMXBs), where late-type, Roche lobe-filling stars transfer matter onto compact objects . Their orbital periods are very compact, clustering at 3-6hr, and they mostly contain accreting neutron stars . This is demonstrated by the exhibition of Type I X-ray bursts () and, in a few exceptional cases (X1822-371 , X1626-67 ) X-ray pulses. There are about 150 X-ray luminous or โpersistentโ LMXBs in the Galaxy, 30 of which with confirmed optical counterparts . For comparison, it is estimated that the Galaxy contains a population of $`10^3`$ โtransientโ LMXBs which only show X-ray activity sporadically and mostly harbour black holes (see accompanying review by Charles & Casares).
The reason for the different luminosity behaviour stems from the interplay between the mass transfer rate from the donor star $`\dot{M_2}`$ and irradiation effects. $`\dot{M_2}`$ determines the temperature structure of the accretion flow and, if it is somewhere below a critical temperature, $`T_{crit}`$ (that of H ionisation), then instability cycles (outbursts) can be triggered (see ). However, in persistent LMXBs, irradiation can keep the outer disc hotter than $`T_{crit}`$ (even for low $`\dot{M_2}10^9`$ M yr<sup>-1</sup>), so that outburst cycles are suppressed and discs appear persistently bright (, ). They display a large $`L_X/L_{opt}`$ and the optical and X-ray fluxes are correlated, a strong indication that optical emission is caused by reprocessing of higher energy photons by material in the vicinity of the X-ray source. Further evidence for X-ray reprocessing includes:
* Statistical properties: dereddened optical colours follow the distribution $`(BV)_0=0.09\pm 0.17`$ , $`(UB)_0=0.97\pm 0.17`$ . This is consistent with F$`{}_{\nu }{}^{}`$ constant as expected for the reprocessing of high energy photons.
* Correlated X-ray/optical Type I bursts: the optical lags by a few seconds, consistent with light travel times within the binary. Optical profiles are smeared versions of the X-ray profiles, indicating an extended reprocessing site (e.g. ).
* Presence of high excitation lines: e.g. HeII $`\lambda `$4686, CIII/NIII $`\lambda \lambda `$4640-50, and their flux is well correlated with $`L_X`$. Bowen fluorescence was proposed to explain the enhanced NIII emission , a mechanism subsequently confirmed by the detection of OIII cascades in Her X-1 and Sco X-1 (Fig. 1).
However, X-ray irradiation has systematically plagued attempts to determine system parameters in persistent LMXBs. These rely on dynamical information from the companion star, which is typically $`>10^3`$ times fainter than the X-ray heated accretion disc at optical-IR wavelengths. Fortunately, there are methods which can exploit the effects of irradiation and X-ray variability. Here we provide an overview of these recent advances in the determination of fundamental parameters of persistent LMXBS.
## 2 Optical light curves
Soon after the discovery of the first optical counterparts it was clear that constraining binary parameters in X-ray bright LMXBs was a difficult challenge . Even the determination of the most fundamental binary parameter, $`P_{orb}`$, proved elusive due to the lack of obvious photometric variability. This led Milgrom to propose a scenario where the companion star was effectively shadowed from the central X-ray source by a flared accretion disc. The absence of eclipsing systems was hence a pure selection effect. Since then, the gain in X-ray sensitivity and deeper surveys has presented several examples of eclipsing LMXBs and others with regular optical/X-ray modulations and dips which indicate the binary periods .
Optical lightcurves are quasi-sinusoidal with superposed erratic variations and flickering, probably in response to variable X-ray illumination. It is widely assumed that optical maxima are associated with the irradiated inner face of the companion i.e. orbital phase $`\varphi =0.5`$. This is supported by the relative phasing of X-ray eclipses ($`\varphi =0`$) and dips ($`\varphi =0.8`$) (e.g. ). Interestingly, the amplitude A seems to be correlated with inclination , from which LMXBs can be placed into three broad categories: (i) eclipsing, with A $``$0.5-1.5 mag and $`i80^{}`$ (e.g. X1822-371, X0748-636, X2129+47) (ii) dippers, with A $``$0.5 mag and $`i7080^{}`$ (e.g. X1254-690, X1755-338, X1916-05) and (iii) low i, with A $``$0.5 mag and $`i70^{}`$ (e.g. X1636-536, X1735-444, Sco X-1).<sup>1</sup><sup>1</sup>1However, lightcurves of the transient LMXB XTE J2123-058, obtained throughout the outburst cycle, show a factor of 2 increase in optical amplitude when $`L_X`$ drops by one order of magnitude (see , ). Therefore, a straight correlation between A and i should be treated with caution.
Tighter constraints on $`i`$ can be set by detailed modelling of optical/X-ray eclipses and light curves of accretion disc coronae (ADC) sources . In particular, simultaneous fits to EXOSAT X-ray/optical light curves in X1822-371 have led to the most accurate determination of the disc geometry and $`i`$ ($`83^{}\pm 2^{}`$) in an LMXB , . The model includes irradiation, shadowing and obscuration by the disc, donor star, bulge and ADC structures (see Fig. 2).
## 3 Dynamical Information
With the exception of a few long period ($`P>1`$ d) systems with evolved donors (e.g. Cyg X-2 , X0921-630 ), spectroscopic features of companion stars in persistent LMXBs are totally veiled by the accretion disc continuum. Attempts to derive dynamical information of the compact star using emission lines (mainly Balmer or HeII $`\lambda `$4686) have proven unreliable because the emission lines are very broad and show a complex, variable, multi-component structure. The bulk of the emission tends to be dominated by the disc bulge with superior conjunction at orbital phase $``$0.75<sup>2</sup><sup>2</sup>2As determined by photometric ephemerides from eclipses or lightcurve minima. e.g. X1636-536, X1735-444 .
### 3.1 Fluorescent emision from the companion
New prospects for dynamical studies have been opened by the discovery of high-excitation emission lines arising from the donor star in Sco X-1 . The most prominent are in the core of the Bowen blend, namely the triplets NIII $`\lambda `$4634-40 and CIII $`\lambda `$4647-50. In particular, the NIII lines are powered by fluorescence resonance which requires seed photons of HeII Ly-$`\alpha `$. These narrow components move in phase with each other and are not resolved (i.e. their FWHM is the instrumental resolution, 50 km s<sup>-1</sup>), an indication that the reprocessing region is very localized (Fig. 3). The extreme narrowness rules out the accretion flow or the hot spot and points to the companion star as the reprocessing site.
The radial velocity curve of the Bowen lines (Fig.3) is in antiphase with the HeII $`\lambda `$4686 wings, which approximately trace the motion of the compact star. Furthermore, they are also in phase with the maximum of the photometric light curve, ascribed to the irradiated face of the donor star . This work represents the first detection of the companion star in Sco X-1 and has opened a new avenue for dynamical studies of luminous LMXBs.
We currently know that fluorescence is not peculiar to Sco X-1, but is a general signature of active LMXBs. This is exemplified by recent work on X1822-371, the archetypal eclipsing ADC which also contains a 0.59s pulsar . Both $`i`$ and the pulsar orbit are extremely well constrained and hence only the radial velocity curve of the companion star is needed for a full determination of the system parameters. The faintness of X1822-371, coupled with the high spectral resolution ($`70`$ km s<sup>-1</sup>) required to resolve the Bowen lines, prevents detection in individual spectra. However, exploiting Doppler Tomography we can combine all the information contained in individually phase-resolved spectra to reconstruct the emissivity distribution. Fig. 4 shows the Doppler maps of HeII $`\lambda `$4686 and NIII $`\lambda `$4640 in velocity space. Unlike HeII $`\lambda `$4686 (which shows the classic accretion disc ring) the NIII map displays a compact spot located at the expected velocity and phasing of the donor star, as predicted from the pulsar ephemeris. The centroid of the spot lies at 300 km s<sup>-1</sup> and establishes a lower limit to the velocity of the companion because it arises from the irradiated inner face. In order to derive the true velocity (and then accurate masses) one needs to model the displacement between the reprocessing site and the donorโs center of mass as a function of the mass ratio q. The so-called K-correction depends on details of reprocessing physics and irradiation geometry, such as shielding effects by the disc (see ).
Follow-up campaigns, using 8m-class telescopes (e.g. VLT), has enabled us to extend this analysis to fainter LMXBs. This has led to the first detection of the donors in X1636-536, X1735-444 and GX 9+9 and the determination of their orbital velocities, which lie in the range 200-300 km s<sup>-1</sup> (, ). In addition, this technique has been applied to transient LMXBs in outburst, such as the millisecond pulsar XTE J1814-338, Aql X-1 and the black hole candidate GX339-4 . In the latter case, the observations provided the first dynamical proof that it is a black hole.
### 3.2 Burst Oscillations
Despite LMXBs having long been considered the progenitors of millisecond pulsars, such pulsations have escaped detection until recently. This changed thanks to RXTE with the discovery of: (i) persistent pulses in 5 transient LMXBs with $`P_{spin}`$ in the range 185-435 Hz, and (ii) nearly coherent oscillations during X-ray bursts in 13 luminous LMXBs. In SAX J1808-3658 and XTE J1814-338 both burst oscillations and persistent pulses were detected and with identical frequencies, confirming that burst oscillations are indeed modulated with the neutron star spin. Furthermore, a smooth frequency drift in the oscillation could be observed during a superburst in X1636-536, caused by the doppler motion of the neutron star . Burst oscillations can therefore be used to trace the neutron star orbit in persistent LMXBs and, in combination with Bowen fluorescence, these luminous LMXBs can become double-lined spectroscopic binaries.
## 4 Echo-tomography
Echo Tomography is a powerful technique which employs time delay between X-ray and UV/optical variability to map the reprocessing sites in a binary . The optical lightcurve results from the convolution of the X-ray lightcurve with a transfer function representing the binary response to the illuminating flux. The transfer function contains a phase-dependent component, associated with the donor, which encodes information on the most fundamental parameters, namely $`i`$, $`a`$ and $`q`$. Succesful Echo-tomography experiments have been performed in several X-ray active LMXBs using X-ray and broad-band UV/optical lightcurves. However, the results indicate that the reprocessing flux is mostly dominated by the accretion disc with no orbital phase dependency (e.g. , ). Exploiting emission-line reprocessing rather than broad-band photometry has two potential benefits: a) it amplifies the response of the donorโs contribution by suppressing most of the background continuum light (dominated by the disc); b) since the emission line reprocessing time is instantaneous, the response is sharper (i.e. only smeared by geometry). Recent results on Sco X-1, using narrow-band filters centered on the Bowen blend + HeII $`\lambda `$4686 region, simultaneously with RXTE, have shown evidence for delayed echoes associated with the donor (see Muรฑoz-Darias et al., these proceedings).
## 5 Conclusions
The study of Bowen fluorescence in X-ray active LMXBs is producing significant progress in the determination of fundamental system parameters. This is possible because of: i) dynamical information obtained by detecting the donors through high-resolution spectroscopy of the Bowen blend; ii) echo-mapping reprocessing sites through simultaneous Bowen-line/X-ray lightcurves. These techniques, together with results from burst oscillations, will likely provide the first accurate neutron star masses in luminous LMXBs in the near future. High-speed and high-resolution instruments in new generation large telescopes (such as OSIRIS at GTC and PFIS at SALT) will play a crucial role in this goal.
|
warning/0506/gr-qc0506032.html
|
ar5iv
|
text
|
# Intrinsic definitions of โrelative velocityโ in general relativity
## 1 Introduction
The need for a strict definition of โradial velocityโ was treated at the General Assembly of the International Astronomical Union (IAU), held in 2000 (see , ), due to the ambiguity of the classic concepts in general relativity. As result, they obtained three different concepts of radial velocity: kinematic (which corresponds most closely to the line-of-sight component of space velocity), astrometric (which can be derived from astrometric observations) and spectroscopic (also called barycentric, which can be derived from spectroscopic measurements). The kinematic and astrometric radial velocities were defined using a particular reference system, called Barycentric Celestial Reference System (BCRS). The BCRS is suitable for accurate modelling of motions and events within the solar system, but it has not into account the effects produced by gravitational fields outside the solar system, since it describes an asymptotically flat metric at large distances from the Sun. Moreover, from a more theoretical point of view, these concepts can not be defined in an arbitrary spacetime since they are not intrinsic, i.e. they only have sense in the framework of the BCRS. So, in this work we are going to define them intrinsically. In fact, we obtain in a natural way four intrinsic definitions of relative velocity (and consequently, radial velocity) of one observer $`\beta ^{}`$ with respect to another observer $`\beta `$, following the original ideas of the IAU.
This paper has two big parts:
* The first one is formed by Sections 3 and 4, where all the concepts are defined, trying to make the paper as self-contained as possible. In Section 3, we define the kinematic and Fermi relative velocities in the framework of spacelike simultaneity (also called Fermi simultaneity), obtaining some general properties and interpretations. The kinematic relative velocity generalizes the usual concept of relative velocity when the two observers $`\beta `$, $`\beta ^{}`$ are at the same event. On the other hand, the Fermi relative velocity does not generalize this concept, but it is physically interpreted as the variation of the relative position of $`\beta ^{}`$ with respect to $`\beta `$ along the world line of $`\beta `$. Analogously, in Section 4, we define and study the spectroscopic and astrometric relative velocities in the framework of observed (lightlike) simultaneity.
* In the second one (Sections 5 and 6) we give some relations between these concepts in special and general relativity. In Section 5 we find general expressions, in special relativity, for the relation between kinematic and Fermi relative velocities, and between spectroscopic and astrometric relative velocities. Finally, in Section 6 we show some fundamental examples in Schwarzschild and Robertson-Walker spacetimes.
## 2 Preliminaries
We work in a 4-dimensional lorentzian spacetime manifold $`(,g)`$, with $`c=1`$ and $``$ the Levi-Civita connection, using the Landau-Lifshitz Spacelike Convention (LLSC). We suppose that $``$ is a convex normal neighborhood . Thus, given two events $`p`$ and $`q`$ in $``$, there exists a unique geodesic joining $`p`$ and $`q`$ and there are not caustics. The parallel transport from $`p`$ to $`q`$ along this geodesic will be denoted by $`\tau _{pq}`$. If $`\beta :I`$ is a curve with $`I`$ a real interval, we will identify $`\beta `$ with the image $`\beta I`$ (that is a subset in $``$), in order to simplify the notation. If $`u`$ is a vector, then $`u^{}`$ denotes the orthogonal space of $`u`$. The projection of a vector $`v`$ onto $`u^{}`$ is the projection parallel to $`u`$. Moreover, if $`x`$ is a spacelike vector, then $`x`$ denotes the modulus of $`x`$. Given a pair of vectors $`u,v`$, we use $`g(u,v)`$ instead of $`u^\alpha v_\alpha `$. If $`X`$ is a vector field (typically, vector fields will be denoted by uppercase letters), $`X_p`$ denotes the unique vector of $`X`$ in $`T_p`$.
In general, we will say that a timelike world line $`\beta `$ is an observer (or a test particle). Nevertheless, we will say that a future-pointing timelike unit vector $`u`$ in $`T_p`$ is an observer at $`p`$, identifying it with its 4-velocity.
The relative velocity of an observer (or a test particle) with respect to another observer is completely well defined only when these observers are at the same event: given two observers $`u`$ and $`u^{}`$ at the same event $`p`$, there exists a unique vector $`vu^{}`$ and a unique positive real number $`\gamma `$ such that
$$u^{}=\gamma \left(u+v\right).$$
(1)
As consequences, we have $`0v<1`$ and $`\gamma :=g(u^{},u)=\frac{1}{\sqrt{1v^2}}`$. We will say that $`v`$ is the relative velocity of $`u^{}`$ observed by $`u`$, and $`\gamma `$ is the gamma factor corresponding to the velocity $`v`$. From (1), we have
$$v=\frac{1}{g(u^{},u)}u^{}u.$$
(2)
We will extend this definition of relative velocity in two different ways (kinematic and spectroscopic) for observers at different events. Moreover, we will define another two concepts of relative velocity (Fermi and astrometric) that do not extend (2) in general, but they have clear physical sense as the variation of the relative position.
A light ray is given by a lightlike geodesic $`\lambda `$ and a future-pointing lightlike vector field $`F`$ defined in $`\lambda `$, tangent to $`\lambda `$ and parallelly transported along $`\lambda `$ (i.e. $`_FF=0`$), called frequency (or wave) vector field of $`\lambda `$. Given $`p\lambda `$ and $`u`$ an observer at $`p`$, there exists a unique vector $`wu^{}`$ and a unique positive real number $`\nu `$ such that
$$F_p=\nu \left(u+w\right).$$
(3)
As consequences, we have $`w=1`$ and $`\nu =g(F_p,u)`$. We will say that $`w`$ is the relative velocity of $`\lambda `$ observed by $`u`$, and $`\nu `$ is the frequency of $`\lambda `$ observed by $`u`$. In other words, $`\nu `$ is the modulus of the projection of $`F_p`$ onto $`u^{}`$. A light ray from $`q`$ to $`p`$ is a light ray $`\lambda `$ such that $`q`$, $`p\lambda `$ and $`\mathrm{exp}_q^1p`$ is future-pointing.
## 3 Relative velocity in the framework of spacelike simultaneity
The spacelike simultaneity was introduced by E. Fermi (see ), and it was used to define the Fermi coordinates. So, some concepts given in this section are very related to the work of Fermi, as the Fermi surfaces, the Fermi derivative or the Fermi distance. The original Fermi paper and most of the modern discussions of this notion (see , ) use a coordinate language (Fermi coordinates). On the other hand, in the present work we use a coordinate-free notation that allows us to get a better understanding of the basic concepts of the Fermi work, studying them from an intrinsic point of view and, in the next section, extending them to the framework of lightlike simultaneity.
Let $`u`$ be an observer at $`p`$ and $`\mathrm{\Phi }:`$ defined by $`\mathrm{\Phi }\left(q\right):=g(\mathrm{exp}_p^1q,u)`$. Then, it is a submersion and the set $`L_{p,u}:=\mathrm{\Phi }^1\left(0\right)`$ is a regular 3-dimensional submanifold, called Landau submanifold of $`(p,u)`$ (see , ), also known as Fermi surface. In other words, $`L_{p,u}=\mathrm{exp}_pu^{}`$. An event $`q`$ is in $`L_{p,u}`$ if and only if $`q`$ is simultaneous with $`p`$ in the local inertial proper system of $`u`$.
###### Definition 3.1
Given $`u`$ an observer at $`p`$, and a simultaneous event $`qL_{p,u}`$, the relative position of $`q`$ with respect to $`u`$ is $`s:=\mathrm{exp}_p^1q`$ (see Fig. 1).
We can generalize this definition for two observers $`\beta `$ and $`\beta ^{}`$.
###### Definition 3.2
Let $`\beta `$, $`\beta ^{}`$ be two observers and let $`U`$ be the 4-velocity of $`\beta `$. The relative position of $`\beta ^{}`$ with respect to $`\beta `$ is the vector field $`S`$ defined on $`\beta `$ such that $`S_p`$ is the relative position of $`q`$ with respect to $`U_p`$, where $`p\beta `$ and $`q`$ is the unique event of $`\beta ^{}L_{p,U_p}`$.
### 3.1 Kinematic relative velocity
We are going to introduce the concept of โkinematic relative velocityโ of one observer $`u^{}`$ with respect to another observer $`u`$ generalizing the concept of relative velocity given by (2), when the two observers are at different events.
###### Definition 3.3
Let $`u`$, $`u^{}`$ be two observers at $`p`$, $`q`$ respectively such that $`qL_{p,u}`$. The kinematic relative velocity of $`u^{}`$ with respect to $`u`$ is the unique vector $`v_{\mathrm{kin}}u^{}`$ such that $`\tau _{qp}u^{}=\gamma \left(u+v_{\mathrm{kin}}\right)`$, where $`\gamma `$ is the gamma factor corresponding to the velocity $`v_{\mathrm{kin}}`$ (see Fig. 2). So, it is given by
$$v_{\mathrm{kin}}:=\frac{1}{g(\tau _{qp}u^{},u)}\tau _{qp}u^{}u.$$
(4)
Let $`s`$ be the relative position of $`q`$ with respect to $`p`$, the kinematic radial velocity of $`u^{}`$ with respect to $`u`$ is the component of $`v_{\mathrm{kin}}`$ parallel to $`s`$, i.e. $`v_{\mathrm{kin}}^{\mathrm{rad}}:=g(v_{\mathrm{kin}},\frac{s}{s})\frac{s}{s}`$. If $`s=0`$ (i.e. $`p=q`$) then $`v_{\mathrm{kin}}^{\mathrm{rad}}:=v_{\mathrm{kin}}`$. On the other hand, the kinematic tangential velocity of $`u^{}`$ with respect to $`u`$ is the component of $`v_{\mathrm{kin}}`$ orthogonal to $`s`$, i.e. $`v_{\mathrm{kin}}^{\mathrm{tng}}:=v_{\mathrm{kin}}v_{\mathrm{kin}}^{\mathrm{rad}}`$.
So, the kinematic relative velocity of $`u^{}`$ with respect to $`u`$ is the relative velocity of $`\tau _{qp}u^{}`$ observed by $`u`$, in the sense of expression (2). Note that $`v_{\mathrm{kin}}<1`$, since the parallel transported observer $`\tau _{qp}u^{}`$ defines an observer at $`p`$.
We can generalize these definitions for two observers $`\beta `$ and $`\beta ^{}`$.
###### Definition 3.4
Let $`\beta `$, $`\beta ^{}`$ be two observers, and let $`U`$, $`U^{}`$ be the $`4`$-velocities of $`\beta `$, $`\beta ^{}`$ respectively. The kinematic relative velocity of $`\beta ^{}`$ with respect to $`\beta `$ is the vector field $`V_{\mathrm{kin}}`$ defined on $`\beta `$ such that $`V_{\mathrm{kin}p}`$ is the kinematic relative velocity of $`U_q^{}`$ observed by $`U_p`$ (in the sense of Definition 3.3), where $`p\beta `$ and $`q`$ is the unique event of $`\beta ^{}L_{p,U_p}`$. In the same way, we define the kinematic radial velocity of $`\beta ^{}`$ with respect to $`\beta `$, denoted by $`V_{\mathrm{kin}}^{\mathrm{rad}}`$, and the kinematic tangential velocity of $`\beta ^{}`$ with respect to $`\beta `$, denoted by $`V_{\mathrm{kin}}^{\mathrm{tng}}`$.
We will say that $`\beta `$ is kinematically comoving with $`\beta ^{}`$ if $`V_{\mathrm{kin}}=0`$.
Let $`V_{\mathrm{kin}}^{}`$ be the kinematic relative velocity of $`\beta `$ with respect to $`\beta ^{}`$. Then, $`V_{\mathrm{kin}}=0`$ if and only if $`V_{\mathrm{kin}}^{}=0`$, i.e. the relation โto be kinematically comoving withโ is symmetric and so, we can say that $`\beta `$ and $`\beta ^{}`$ are kinematically comoving (each one with respect to the other). Note that it is not transitive in general.
### 3.2 Fermi relative velocity
We are going to define the โFermi relative velocityโ as the variation of the relative position.
###### Definition 3.5
Let $`\beta `$, $`\beta ^{}`$ be two observers, let $`U`$ be the 4-velocity of $`\beta `$, and let $`S`$ be the relative position of $`\beta ^{}`$ with respect to $`\beta `$. The Fermi relative velocity of $`\beta ^{}`$ with respect to $`\beta `$ is the projection of $`_US`$ onto $`U^{}`$, i.e. it is the vector field
$$V_{\mathrm{Fermi}}:=_US+g(_US,U)U$$
(5)
defined on $`\beta `$. The right-hand side of (5) is known as the Fermi derivative. The Fermi radial velocity of $`\beta ^{}`$ with respect to $`\beta `$ is the component of $`V_{\mathrm{Fermi}}`$ parallel to $`S`$, i.e. $`V_{\mathrm{Fermi}}^{\mathrm{rad}}:=g(V_{\mathrm{Fermi}},\frac{S}{S})\frac{S}{S}`$ if $`S0`$; if $`S_p=0`$ (i.e. $`\beta `$ and $`\beta ^{}`$ intersect at $`p`$) then $`V_{\mathrm{Fermi}p}^{\mathrm{rad}}:=V_{\mathrm{Fermi}p}`$. On the other hand, the Fermi tangential velocity of $`\beta ^{}`$ with respect to $`\beta `$ is the component of $`V_{\mathrm{Fermi}}`$ orthogonal to $`S`$, i.e. $`V_{\mathrm{Fermi}}^{\mathrm{tng}}:=V_{\mathrm{Fermi}}V_{\mathrm{Fermi}}^{\mathrm{rad}}`$.
We will say that $`\beta `$ is Fermi-comoving with $`\beta ^{}`$ if $`V_{\mathrm{Fermi}}=0`$.
It is important to remark that the modulus of the vectors of $`V_{\mathrm{Fermi}}`$ is not necessarily smaller than one.
Since $`g(V_{\mathrm{Fermi}},S)=g(_US,S)`$, if $`S0`$ we have
$$V_{\mathrm{Fermi}}^{\mathrm{rad}}=g(_US,\frac{S}{S})\frac{S}{S}.$$
(6)
The relation โto be Fermi-comoving withโ is not symmetric in general.
An expression similar to (5) is given by the next proposition, that can be proved easily.
###### Proposition 3.1
Let $`\beta `$, $`\beta ^{}`$ be two observers, let $`U`$ be the 4-velocity of $`\beta `$, let $`S`$ be the relative position of $`\beta ^{}`$ with respect to $`\beta `$, and let $`V_{\mathrm{Fermi}}`$ be the Fermi relative velocity of $`\beta ^{}`$ with respect to $`\beta `$. Then $`V_{\mathrm{Fermi}}=_USg(S,_UU)U`$. Note that if $`\beta `$ is geodesic, then $`_UU=0`$, and hence $`V_{\mathrm{Fermi}}=_US`$ .
If $`S_p=0`$, i.e. $`\beta `$ and $`\beta ^{}`$ intersect at $`p`$, then $`V_{\mathrm{Fermi}p}=\left(_US\right)_p`$. So, it does not coincide in general with the concept of relative velocity given in expression (2).
We are going to introduce a concept of distance from the concept of relative position given in Definition 3.2. This concept of distance was previously introduced by Fermi.
###### Definition 3.6
Let $`u`$ be an observer at an event $`p`$. Given $`q`$, $`q^{}L_{p,u}`$, and $`s`$, $`s^{}`$ the relative positions of $`q`$, $`q^{}`$ with respect to $`u`$ respectively, the Fermi distance from $`q`$ to $`q^{}`$ with respect to $`u`$ is the modulus of $`ss^{}`$, i.e. $`d_u^{\mathrm{Fermi}}(q,q^{}):=ss^{}`$.
We have that $`d_u^{\mathrm{Fermi}}`$ is symmetric, positive-definite and satisfies the triangular inequality. So, it has all the properties that must verify a topological distance defined on $`L_{p,u}`$. As a particular case, if $`q^{}=p`$ we have
$$d_u^{\mathrm{Fermi}}(q,p)=s=\left(g(\mathrm{exp}_p^1q,\mathrm{exp}_p^1q)\right)^{1/2}.$$
(7)
The next proposition shows that the concept of Fermi distance is the arclength parameter of a spacelike geodesic, and it can be proved taking into account the properties of the exponential map (see ).
###### Proposition 3.2
Let $`u`$ be an observer at an event $`p`$. Given $`qL_{p,u}`$ and $`\alpha `$ the unique geodesic from $`p`$ to $`q`$, if we parameterize $`\alpha `$ by its arclength such that $`\alpha \left(0\right)=p`$, then $`\alpha \left(d_u^{\mathrm{Fermi}}(q,p)\right)=q`$.
###### Definition 3.7
Let $`\beta `$, $`\beta ^{}`$ be two observers and let $`S`$ be the relative position of $`\beta ^{}`$ with respect to $`\beta `$. The Fermi distance from $`\beta ^{}`$ to $`\beta `$ with respect to $`\beta `$ is the scalar field $`S`$ defined in $`\beta `$.
We are going to characterize the Fermi radial velocity in terms of the Fermi distance.
###### Proposition 3.3
Let $`\beta `$, $`\beta ^{}`$ be two observers, let $`S`$ be the relative position of $`\beta ^{}`$ with respect to $`\beta `$, and let $`U`$ be the 4-velocity of $`\beta `$. If $`S0`$, the Fermi radial velocity of $`\beta ^{}`$ with respect to $`\beta `$ reads $`V_{\mathrm{Fermi}}^{\mathrm{rad}}=U\left(S\right)\frac{S}{S}`$.
By Definition 3.7 and Proposition 3.3, the Fermi radial velocity of $`\beta ^{}`$ with respect to $`\beta `$ is the rate of change of the Fermi distance from $`\beta ^{}`$ to $`\beta `$ with respect to $`\beta `$. So, if we parameterize $`\beta `$ by its proper time $`\tau `$, the Fermi radial velocity of $`\beta ^{}`$ with respect to $`\beta `$ at $`p=\beta \left(\tau _0\right)`$ is given by $`V_{\mathrm{Fermi}p}^{\mathrm{rad}}=\frac{\mathrm{d}\left(S\beta \right)}{\mathrm{d}\tau }\left(\tau _0\right)\frac{S_p}{S_p}`$, where $`S\beta `$ is the Fermi distance as a function of $`\tau `$.
## 4 Relative velocity in the framework of lightlike simultaneity
The lightlike (or observed) simultaneity is based on โwhat an observer is really observingโ and it provides an appropriate framework to study optical phenomena and observational cosmology (see ).
Let $`p`$ and $`\phi :`$ defined by $`\phi \left(q\right):=g(\mathrm{exp}_p^1q,\mathrm{exp}_p^1q)`$. Then, it is a submersion and the set
$$E_p:=\phi ^1\left(0\right)\left\{p\right\}$$
(8)
is a regular 3-dimensional submanifold, called horismos submanifold of $`p`$ (see , ). An event $`q`$ is in $`E_p`$ if and only if $`qp`$ and there exists a lightlike geodesic joining $`p`$ and $`q`$. $`E_p`$ has two connected components, $`E_p^{}`$ and $`E_p^+`$ ; $`E_p^{}`$ (respectively $`E_p^+`$) is the past-pointing (respectively future-pointing) horismos submanifold of $`p`$, and it is the connected component of (8) in which, for each event $`qE_p^{}`$ (respectively $`qE_p^+`$), the preimage $`\mathrm{exp}_p^1q`$ is a past-pointing (respectively future-pointing) lightlike vector. In other words, $`E_p^{}=\mathrm{exp}_pC_p^{}`$, and $`E_p^+=\mathrm{exp}_pC_p^+`$, where $`C_p^{}`$ and $`C_p^+`$ are the past-pointing and the future-pointing light cones of $`T_p`$ respectively.
This section is analogous to Section 3, but using $`E_p^{}`$ instead of $`L_{p,u}`$.
###### Definition 4.1
Given $`u`$ an observer at $`p`$, and an observed event $`qE_p^{}\left\{p\right\}`$, the relative position of $`q`$ observed by $`u`$ (or the observed relative position of $`q`$ with respect to $`u`$) is the projection of $`\mathrm{exp}_p^1q`$ onto $`u^{}`$ (see Fig. 3), i.e. $`s_{\mathrm{obs}}:=\mathrm{exp}_p^1q+g(\mathrm{exp}_p^1q,u)u`$.
We can generalize this definition for two observers $`\beta `$ and $`\beta ^{}`$.
###### Definition 4.2
Let $`\beta `$, $`\beta ^{}`$ be two observers and let $`U`$ be the 4-velocity of $`\beta `$. The relative position of $`\beta ^{}`$ observed by $`\beta `$ is the vector field $`S_{\mathrm{obs}}`$ defined in $`\beta `$ such that $`S_{\mathrm{obs}p}`$ is the relative position of $`q`$ observed by $`U_p`$, where $`p\beta `$ and $`q`$ is the unique event of $`\beta ^{}E_p^{}`$.
### 4.1 Spectroscopic relative velocity
In a previous work (see ), we defined a concept of relative velocity of an observer observed by another observer in the framework of lightlike simultaneity. We are going to rename this concept as โspectroscopic relative velocityโ, and to review its properties in the context of this work.
###### Definition 4.3
Let $`u`$, $`u^{}`$ be two observers at $`p`$, $`q`$ respectively such that $`qE_p^{}`$ and let $`\lambda `$ be a light ray from $`q`$ to $`p`$. The spectroscopic relative velocity of $`u^{}`$ observed by $`u`$ is the unique vector $`v_{\mathrm{spec}}u^{}`$ such that $`\tau _{qp}u^{}=\gamma \left(u+v_{\mathrm{spec}}\right)`$, where $`\gamma `$ is the gamma factor corresponding to the velocity $`v_{\mathrm{spec}}`$ (see Fig. 4). So, it is given by
$$v_{\mathrm{spec}}:=\frac{1}{g(\tau _{qp}u^{},u)}\tau _{qp}u^{}u.$$
(9)
We define the spectroscopic radial and tangential velocity of $`u^{}`$ observed by $`u`$ analogously to Definition 3.3, using $`s_{\mathrm{obs}}`$ (see Definition 4.1) instead of $`s`$.
So, the spectroscopic relative velocity of $`u^{}`$ observed by $`u`$ is the relative velocity of $`\tau _{qp}u^{}`$ observed by $`u`$, in the sense of expression (2), and $`v_{\mathrm{spec}}<1`$.
Note that if $`w`$ is the relative velocity of $`\lambda `$ observed by $`u`$ (see (3)), then $`w=\frac{s_{\mathrm{obs}}}{s_{\mathrm{obs}}}`$, and so
$$v_{\mathrm{spec}}^{\mathrm{rad}}=g(v_{\mathrm{spec}},w)w.$$
(10)
We can generalize these definitions for two observers $`\beta `$ and $`\beta ^{}`$.
###### Definition 4.4
Let $`\beta `$, $`\beta ^{}`$ be two observers, we define $`V_{\mathrm{spec}}`$ (the spectroscopic relative velocity of $`\beta ^{}`$ observed by $`\beta `$) and its radial and tangential components analogously to Definition 3.4, using $`E_p^{}`$ instead of $`L_{p,U_p}`$.
We will say that $`\beta `$ is spectroscopically comoving with $`\beta ^{}`$ if $`V_{\mathrm{spec}}=0`$.
The relation โto be spectroscopically comoving withโ is not symmetric in general.
The following result can be found in .
###### Proposition 4.1
Let $`\lambda `$ be a light ray from $`q`$ to $`p`$ and let $`u`$, $`u^{}`$ be two observers at $`p`$, $`q`$ respectively. Then
$$\nu ^{}=\gamma \left(1g(v_{\mathrm{spec}},w)\right)\nu ,$$
(11)
where $`\nu `$, $`\nu ^{}`$ are the frequencies of $`\lambda `$ observed by $`u`$, $`u^{}`$ respectively, $`v_{\mathrm{spec}}`$ is the spectroscopic relative velocity of $`u^{}`$ observed by $`u`$, $`w`$ is the relative velocity of $`\lambda `$ observed by $`u`$, and $`\gamma `$ is the gamma factor corresponding to the velocity $`v_{\mathrm{spec}}`$.
Expression (11) is the general expression for Doppler effect (that includes gravitational redshift, see ). Therefore, if $`\beta `$ is spectroscopically comoving with $`\beta ^{}`$, and $`\lambda `$ is a light ray from $`\beta ^{}`$ to $`\beta `$, then, by (11), we have that $`\beta `$ and $`\beta ^{}`$ observe $`\lambda `$ with the same frequency. So, if $`\beta ^{}`$ emits $`n`$ light rays in a unit of its proper time, then $`\beta `$ observes also $`n`$ light rays in a unit of its proper time. Hence, $`\beta `$ observes that $`\beta ^{}`$ uses the โsame clockโ as its.
Taking into account (10), expression (11) can be written in the form
$$\nu ^{}=\frac{1\pm v_{\mathrm{spec}}^{\mathrm{rad}}}{\sqrt{1v_{\mathrm{spec}}^2}}\nu ,$$
(12)
where we choose โ$`+`$โ if $`g(v_{\mathrm{spec}},w)<0`$ (i.e. if $`u^{}`$ is moving away from $`u`$), and we choose โ$``$โ if $`g(v_{\mathrm{spec}},w)>0`$ (i.e. if $`u^{}`$ is getting closer to $`u`$).
###### Remark 4.1
We can not deduce $`v_{\mathrm{spec}}`$ from the shift, $`\nu ^{}/\nu `$, unless we make some assumptions (like considering negligible the tangential component of $`v_{\mathrm{spec}}`$, as we will see in Remark 4.2). For instance, if $`\nu ^{}/\nu =1`$ then $`v_{\mathrm{spec}}`$ is not necessarily zero. Let us study this particular case: by (11) we have
$$1=\frac{\nu ^{}}{\nu }=\frac{1g(v_{\mathrm{spec}},w)}{\sqrt{1v_{\mathrm{spec}}^2}}g(v_{\mathrm{spec}},w)=1\sqrt{1v_{\mathrm{spec}}^2}.$$
Since $`\left(1\sqrt{1v_{\mathrm{spec}}^2}\right)0`$, it is necessary that $`g(v_{\mathrm{spec}},w)0`$, i.e. the observed object has to be getting closer to the observer. In this case, by (12) we have $`v_{\mathrm{spec}}^{\mathrm{rad}}=1\sqrt{1v_{\mathrm{spec}}^2}`$. So, it is possible that $`\nu ^{}/\nu =1`$ and $`v_{\mathrm{spec}}0`$ if the observed object is getting closer to the observer. On the other hand, if the observed object is moving away from the observer then $`\nu ^{}/\nu =1`$ if and only if $`v_{\mathrm{spec}}=0`$. That is, for objects moving away, the shift is always redshift; and for objects getting closer, the shift can be blueshift, 1, or redshift.
###### Remark 4.2
If we suppose that $`v_{\mathrm{spec}}^{\mathrm{tng}}=0`$, i.e. $`v_{\mathrm{spec}}=v_{\mathrm{spec}}^{\mathrm{rad}}=kw`$ with $`k]1,1[`$, then we can deduce $`v_{\mathrm{spec}}`$ from the shift $`\nu ^{}/\nu `$:
$$\frac{\nu ^{}}{\nu }=\frac{1g(v_{\mathrm{spec}},w)}{\sqrt{1v_{\mathrm{spec}}^2}}=\frac{1k}{\sqrt{1k^2}}=\frac{\sqrt{1k}}{\sqrt{1+k}}k=\frac{1\left(\frac{\nu ^{}}{\nu }\right)^2}{1+\left(\frac{\nu ^{}}{\nu }\right)^2},$$
and hence
$$v_{\mathrm{spec}}=\left(\frac{1\left(\frac{\nu ^{}}{\nu }\right)^2}{1+\left(\frac{\nu ^{}}{\nu }\right)^2}\right)w=\left(\frac{1\left(\frac{\nu ^{}}{\nu }\right)^2}{1+\left(\frac{\nu ^{}}{\nu }\right)^2}\right)\frac{s_{\mathrm{obs}}}{s_{\mathrm{obs}}}.$$
(13)
### 4.2 Astrometric relative velocity
We are going to define the โastrometric relative velocityโ as the variation of the observed relative position.
###### Definition 4.5
Let $`\beta `$, $`\beta ^{}`$ be two observers, we define $`V_{\mathrm{ast}}`$ (the astrometric relative velocity of $`\beta ^{}`$ observed by $`\beta `$) and its radial and tangential components analogously to Definition 3.5, using $`S_{\mathrm{obs}}`$ (see Definition 4.2) instead of $`S`$. So,
$$V_{\mathrm{ast}}:=_US_{\mathrm{obs}}+g(_US_{\mathrm{obs}},U)U,$$
(14)
where $`U`$ is the 4-velocity of $`\beta `$.
We will say that $`\beta `$ is astrometrically comoving with $`\beta ^{}`$ if $`V_{\mathrm{ast}}=0`$.
It is important to remark that the modulus of the vectors of $`V_{\mathrm{ast}}`$ is not necessarily smaller than one.
Analogously to (6), since $`g(V_{\mathrm{ast}},S_{\mathrm{obs}})=g(_US_{\mathrm{obs}},S_{\mathrm{obs}})`$, if $`S_{\mathrm{obs}}0`$ we have
$$V_{\mathrm{ast}}^{\mathrm{rad}}=g(_US_{\mathrm{obs}},\frac{S_{\mathrm{obs}}}{S_{\mathrm{obs}}})\frac{S_{\mathrm{obs}}}{S_{\mathrm{obs}}}.$$
(15)
The relation โto be astrometrically comoving withโ is not symmetric in general.
An expression similar to (14) is given by the next proposition, which proof is analogous to the proof of Proposition 3.1.
###### Proposition 4.2
Let $`\beta `$, $`\beta ^{}`$ be two observers, let $`U`$ be the 4-velocity of $`\beta `$, let $`S_{\mathrm{obs}}`$ be the relative position of $`\beta ^{}`$ observed by $`\beta `$, and let $`V_{\mathrm{ast}}`$ be the astrometric relative velocity of $`\beta ^{}`$ observed by $`\beta `$. Then $`V_{\mathrm{ast}}=_US_{\mathrm{obs}}g(S_{\mathrm{obs}},_UU)U`$. Note that if $`\beta `$ is geodesic, then $`_UU=0`$, and hence $`V_{\mathrm{ast}}=_US_{\mathrm{obs}}`$.
If $`S_{\mathrm{obs}p}=0`$, i.e. $`\beta `$ and $`\beta ^{}`$ intersect at $`p`$, then $`V_{\mathrm{ast}p}=\left(_US_{\mathrm{obs}}\right)_p`$. So, it does not coincide in general with the concept of relative velocity given in (2).
We are going to introduce another concept of distance from the concept of observed relative position given in Definition 4.1. This distance was previously introduced in and studied in , and it plays a basic role for the construction of optical coordinates whose relevance for cosmology was stressed in many articles by G. Ellis and his school (see ).
###### Definition 4.6
Let $`u`$ be an observer at an event $`p`$. Given $`q`$, $`q^{}E_p^{}\left\{p\right\}`$, and $`s_{\mathrm{obs}}`$, $`s_{\mathrm{obs}}^{}`$ the relative positions of $`q`$, $`q^{}`$ observed by $`u`$ respectively, the affine distance from $`q`$ to $`q^{}`$ observed by $`u`$ is the modulus of $`s_{\mathrm{obs}}s_{\mathrm{obs}}^{}`$, i.e. $`d_u^{\mathrm{aff}}(q,q^{}):=s_{\mathrm{obs}}s_{\mathrm{obs}}^{}`$.
We have that $`d_u^{\mathrm{aff}}`$ is symmetric, positive-definite and satisfies the triangular inequality. So, it has all the properties that must verify a topological distance defined on $`E_p^{}\left\{p\right\}`$. As a particular case, if $`q^{}=p`$ we have
$$d_u^{\mathrm{aff}}(q,p)=s_{\mathrm{obs}}=g(\mathrm{exp}_p^1q,u).$$
(16)
The next proposition shows that the concept of affine distance is according to the concept of โlengthโ (or โtimeโ) parameter of a lightlike geodesic for an observer, and it is proved in .
###### Proposition 4.3
Let $`\lambda `$ be a light ray from $`q`$ to $`p`$, let $`u`$ be an observer at $`p`$, and let $`w`$ be the relative velocity of $`\lambda `$ observed by $`u`$. If we parameterize $`\lambda `$ affinely (i.e. the vector field tangent to $`\lambda `$ is parallelly transported along $`\lambda `$) such that $`\lambda \left(0\right)=p`$ and $`\stackrel{.}{\lambda }\left(0\right)=\left(u+w\right)`$, then $`\lambda \left(d_u^{\mathrm{aff}}(q,p)\right)=q`$.
###### Definition 4.7
Let $`\beta `$, $`\beta ^{}`$ be two observers and let $`S_{\mathrm{obs}}`$ be the relative position of $`\beta ^{}`$ observed by $`\beta `$. The affine distance from $`\beta ^{}`$ to $`\beta `$ observed by $`\beta `$ is the scalar field $`S_{\mathrm{obs}}`$ defined in $`\beta `$.
We are going to characterize the astrometric radial velocity in terms of the affine distance. The proof of the next proposition is analogous to the proof of Proposition 3.3, taking into account expression (15).
###### Proposition 4.4
Let $`\beta `$, $`\beta ^{}`$ be two observers, let $`S_{\mathrm{obs}}`$ be the relative position of $`\beta ^{}`$ observed by $`\beta `$, and let $`U`$ be the 4-velocity of $`\beta `$. If $`S_{\mathrm{obs}}0`$, the astrometric radial velocity of $`\beta ^{}`$ observed by $`\beta `$ reads $`V_{\mathrm{ast}}^{\mathrm{rad}}=U\left(S_{\mathrm{obs}}\right)\frac{S_{\mathrm{obs}}}{S_{\mathrm{obs}}}`$.
By Definition 4.7 and Proposition 4.4, the astrometric radial velocity of $`\beta ^{}`$ observed by $`\beta `$ is the rate of change of the affine distance from $`\beta ^{}`$ to $`\beta `$ observed by $`\beta `$. So, if we parameterize $`\beta `$ by its proper time $`\tau `$, the astrometric radial velocity of $`\beta ^{}`$ observed by $`\beta `$ at $`p=\beta \left(\tau _0\right)`$ is given by $`V_{\mathrm{ast}p}^{\mathrm{rad}}=\frac{\mathrm{d}\left(S_{\mathrm{obs}}\beta \right)}{\mathrm{d}\tau }\left(\tau _0\right)\frac{S_{\mathrm{obs}p}}{S_{\mathrm{obs}p}}`$, where $`S_{\mathrm{obs}}\beta `$ is the affine distance as a function of $`\tau `$.
## 5 Special Relativity
In this section, we are going to work in the Minkowski spacetime, considering $`\beta `$, $`\beta ^{}`$ two observers with 4-velocities $`U`$, $`U^{}`$ respectively. The goal is to find expressions for $`V_{\mathrm{Fermi}}`$ and $`V_{\mathrm{ast}}`$ in terms of $`U`$, $`_UU`$, $`U^{}`$, $`S`$ and $`S_{\mathrm{obs}}`$, i.e. without $`_US`$, $`_US_{\mathrm{obs}}`$, or any term involving the evolution of $`S`$, $`S_{\mathrm{obs}}`$.
###### Proposition 5.1
Let $`S`$ be the relative position of $`\beta ^{}`$ with respect to $`\beta `$, and let $`V_{\mathrm{Fermi}}`$ be the Fermi relative velocity of $`\beta ^{}`$ with respect to $`\beta `$. Then
$$V_{\mathrm{Fermi}}=\left(1+g(S,_UU)\right)\left(\frac{1}{g(U^{},U)}U^{}U\right),$$
(17)
where $`V_{\mathrm{Fermi}}`$, $`U`$, $`S`$, $`_UU`$ are evaluated at an event $`p`$ of $`\beta `$, and $`U^{}`$ is evaluated at the event $`q=\beta ^{}L_{p,U_p}`$.
###### Proof.
We are going to consider the observers parameterized by their proper times. Let $`p=\beta \left(\tau \right)`$ be an event of $`\beta `$, let $`u\left(\tau \right)`$ be the 4-velocity of $`\beta `$ at $`p`$, and let $`q=\beta ^{}\left(\tau ^{}\left(\tau \right)\right)`$ be the event of $`\beta ^{}`$ such that $`g(u\left(\tau \right),qp)=0`$ (note that the Minkowski spacetime has an affine structure, and $`qp`$ denotes the vector which joins $`p`$ and $`q`$). So, $`\tau ^{}\left(\tau \right)`$ is the proper time of $`q=\beta ^{}L_{p,u}`$, and the relative position of $`q`$ with respect to $`u`$, denoted by $`s`$, is $`qp`$. If $`u^{}\left(\tau ^{}\right)`$ is the 4-velocity of $`\beta ^{}`$ at $`q`$, then
$$s\left(\tau \right)=\beta ^{}\left(\tau ^{}\left(\tau \right)\right)\beta \left(\tau \right)\dot{s}=u^{}\left(\tau ^{}\right)\dot{\tau }^{}u,$$
(18)
where the dot denotes $`\frac{d}{d\tau }`$. On the other hand
$$g(s,u)=0g(\dot{s},u)+g(s,\dot{u})=0.$$
(19)
Applying (18) in (19) we have
$$g(u^{}\left(\tau ^{}\right)\dot{\tau }^{}u,u)+g(s,\dot{u})=0\dot{\tau }^{}=\frac{1+g(s,\dot{u})}{g(u^{}\left(\tau ^{}\right),u)}.$$
(20)
Combining (18) and (20), we obtain
$$\dot{s}=\frac{1+g(s,\dot{u})}{g(u^{}\left(\tau ^{}\right),u)}u^{}\left(\tau ^{}\right)u.$$
(21)
Let $`U`$, $`U^{}`$ be the 4-velocities of $`\beta `$ and $`\beta ^{}`$ respectively, and let $`S`$ be the relative position of $`\beta ^{}`$ with respect to $`\beta `$. Then, from (21) we have
$$_US=\frac{1+g(S,_UU)}{g(U^{},U)}U^{}U,$$
(22)
where $`U`$, $`S`$, $`_UU`$, $`_US`$ are evaluated at $`p`$, and $`U^{}`$ is evaluated at $`q`$. So, by Proposition 3.1 and expression (22), the Fermi relative velocity $`V_{\mathrm{Fermi}}`$ of $`\beta ^{}`$ with respect to $`\beta `$ is given by
$`V_{\mathrm{Fermi}}`$ $`=`$ $`_USg(S,_UU)U`$
$`=`$ $`\left(1+g(S,_UU)\right)\left({\displaystyle \frac{1}{g(U^{},U)}}U^{}U\right),`$
where $`V_{\mathrm{Fermi}}`$, $`U`$, $`S`$, $`_UU`$ are evaluated at $`p`$, and $`U^{}`$ is evaluated at $`q`$. โ
Taking into account the expression of the kinematic relative velocity given in (4), we obtain the next corollary:
###### Corollary 5.1
The Fermi relative velocity of $`\beta ^{}`$ with respect to $`\beta `$ reads
$$V_{\mathrm{Fermi}}=\left(1+g(S,_UU)\right)V_{\mathrm{kin}}.$$
(23)
So, $`V_{\mathrm{Fermi}}`$ and $`V_{\mathrm{kin}}`$ are proportional. Moreover, if $`\beta `$ is geodesic, then $`V_{\mathrm{Fermi}}=V_{\mathrm{kin}}`$.
###### Proposition 5.2
Let $`S_{\mathrm{obs}}`$ be the relative position of $`\beta ^{}`$ observed by $`\beta `$, and let $`V_{\mathrm{ast}}`$ be the astrometric relative velocity of $`\beta ^{}`$ with respect to $`\beta `$. Then
$$V_{\mathrm{ast}}=\frac{1}{g(U^{},\frac{S_{\mathrm{obs}}}{S_{\mathrm{obs}}}U)}\left(U^{}+g(U^{},U)U\right)+S_{\mathrm{obs}}_UU,$$
(24)
where $`V_{\mathrm{ast}}`$, $`U`$, $`S_{\mathrm{obs}}`$, $`_UU`$ are evaluated at an event $`p`$ of $`\beta `$, and $`U^{}`$ is evaluated at the event $`q=\beta ^{}E_p^{}`$.
###### Proof.
We are going to consider the observers parameterized by their proper times. Let $`p=\beta \left(\tau \right)`$ be an event of $`\beta `$, let $`u\left(\tau \right)`$ be the 4-velocity of $`\beta `$ at $`p`$, and let $`q=\beta ^{}\left(\tau ^{}\left(\tau \right)\right)`$ be the event of $`\beta ^{}`$ such that $`g(qp,qp)=0`$ (note that the Minkowski spacetime has an affine structure, and $`qp`$ denotes the vector which joins $`p`$ and $`q`$). So, $`\tau ^{}\left(\tau \right)`$ is the proper time of $`q=\beta ^{}E_p^{}`$, and the relative position of $`q`$ observed by $`u`$, denoted by $`s_{\text{obs}}`$, is the projection of $`qp`$ onto $`u^{}`$. Let us denote $`s_{\mathrm{obs}}`$ by $`s`$ for the shake of readability. Hence
$$s\left(\tau \right)=\beta ^{}\left(\tau ^{}\left(\tau \right)\right)\beta \left(\tau \right)+s\left(\tau \right)u,$$
(25)
where $`s=\sqrt{g(s,s)}`$ is the affine distance from $`p`$ to $`q`$. If $`u^{}\left(\tau ^{}\right)`$ is the 4-velocity of $`\beta ^{}`$ at $`q`$, deriving (25) with respect to $`\tau `$ we obtain
$$\dot{s}=u^{}\left(\tau ^{}\right)\dot{\tau }^{}u+g(\dot{s},\frac{s}{s})u+s\dot{u},$$
(26)
where the dot denotes $`\frac{d}{d\tau }`$. Taking into account that $`g(s,u)=0`$ and (26), we have
$$g(\dot{s},\frac{s}{s})=g(u^{}\left(\tau ^{}\right)\dot{\tau }^{}+s\dot{u},\frac{s}{s})=\dot{\tau }^{}g(u^{}\left(\tau ^{}\right),\frac{s}{s})+g(\dot{u},s),$$
(27)
and hence, by (26) and (27) we obtain
$$\dot{s}=u^{}\left(\tau ^{}\right)\dot{\tau }^{}+\left(\dot{\tau }^{}g(u^{}\left(\tau ^{}\right),\frac{s}{s})+g(\dot{u},s)1\right)u+s\dot{u}.$$
(28)
On the other hand
$$g(s,u)=0g(\dot{s},u)+g(s,\dot{u})=0.$$
(29)
Applying (28) in (29) and taking into account that $`g(\dot{u},u)=0`$, we find
$$\dot{\tau }^{}=\frac{1}{g(u^{}\left(\tau ^{}\right),\frac{s}{s}u)}.$$
(30)
Combining (28) and (30), we obtain
$$\dot{s}=\frac{1}{g(u^{}\left(\tau ^{}\right),\frac{s}{s}u)}\left(u^{}\left(\tau ^{}\right)+g(u^{}\left(\tau ^{}\right),u)u\right)+g(s,\dot{u})u+s\dot{u}.$$
(31)
Let $`U`$, $`U^{}`$ be the 4-velocities of $`\beta `$ and $`\beta ^{}`$ respectively, and let $`S=S_{\mathrm{obs}}`$ (for the shake of readability) be the relative position of $`\beta ^{}`$ observed by $`\beta `$. Then, from (31) we have
$$_US=\frac{1}{g(U^{},\frac{S}{S}U)}\left(U^{}+g(U^{},U)U\right)+g(S,_UU)U+S_UU,$$
(32)
where $`U`$, $`S`$, $`_UU`$, $`_US`$ are evaluated at $`p`$, and $`U^{}`$ is evaluated at $`q`$. So, by Proposition 4.2 and expression (32), the astrometric relative velocity $`V_{\mathrm{ast}}`$ of $`\beta ^{}`$ with respect to $`\beta `$ is given by
$`V_{\mathrm{ast}}`$ $`=`$ $`_USg(S,_UU)U`$
$`=`$ $`{\displaystyle \frac{1}{g(U^{},\frac{S}{S}U)}}\left(U^{}+g(U^{},U)U\right)+S_UU,`$
where $`V_{\mathrm{ast}}`$, $`U`$, $`S`$, $`_UU`$ are evaluated at $`p`$, and $`U^{}`$ is evaluated at $`q`$. โ
Taking into account the expression of the spectroscopic relative velocity given in (9), we obtain the next corollary:
###### Corollary 5.2
The astrometric relative velocity of $`\beta ^{}`$ with respect to $`\beta `$ reads
$$V_{\mathrm{ast}}=S_{\mathrm{obs}}_UU+\frac{1}{1+g(V_{\mathrm{spec}},\frac{S_{\mathrm{obs}}}{S_{\mathrm{obs}}})}V_{\mathrm{spec}}.$$
(33)
So, $`V_{\mathrm{spec}}`$ and $`V_{\mathrm{ast}}`$ are not proportional unless $`\beta `$ is geodesic.
If $`\beta ^{}`$ is geodesic then it is clear that $`V_{\mathrm{spec}}=V_{\mathrm{kin}}`$. Moreover, if $`\beta `$ is also geodesic then $`V_{\mathrm{spec}}=V_{\mathrm{kin}}=V_{\mathrm{Fermi}}`$.
###### Remark 5.1
Let us suppose that $`\beta `$ and $`\beta ^{}`$ intersect at $`p`$, let $`u`$, $`u^{}`$ be the 4-velocities of $`\beta `$, $`\beta ^{}`$ at $`p`$ respectively, and let $`v`$ be the relative velocity of $`u^{}`$ observed by $`u`$, in the sense of expression (2). Let us study the relations between $`v`$, $`V_{\mathrm{kin}p}`$, $`V_{\mathrm{Fermi}p}`$, $`V_{\mathrm{spec}p}`$ and $`V_{\mathrm{ast}p}`$.
It is clear that $`V_{\mathrm{kin}p}=V_{\mathrm{spec}p}=v`$, even in general relativity. Moreover, since $`S_p=0`$, by (17) we have $`V_{\mathrm{Fermi}p}=v`$. On the other hand, since $`S_{\mathrm{obs}p}=0`$, it is easy to prove that $`V_{\mathrm{ast}p}=\frac{1}{1\pm v}v`$, where we choose โ$`+`$โ if we consider that $`\beta ^{}`$ is leaving from $`\beta `$, and we choose โ$``$โ if we consider that $`\beta ^{}`$ is arriving at $`\beta `$. Therefore, if $`\beta `$ and $`\beta ^{}`$ intersect at $`p`$, then it is not possible to write $`V_{\mathrm{ast}p}`$ in a unique way in terms of $`v`$.
###### Example 5.1
Using rectangular coordinates $`(t,x,y,z)`$, let us consider the following observers: $`\beta \left(\tau \right):=(\tau ,0,0,0)`$, and $`\beta ^{}\left(\tau ^{}\right):=\{\begin{array}{c}(\gamma \tau ^{},v\gamma \tau ^{},0,0)\mathrm{if}\tau ^{}[0,\frac{1}{\gamma v}]\hfill \\ \\ (\gamma \tau ^{},2v\gamma \tau ^{},0,0)\mathrm{if}\tau ^{}]\frac{1}{\gamma v},\frac{2}{\gamma v}]\hfill \end{array}`$ where $`v]0,1[`$ and $`\gamma :=\frac{1}{\sqrt{1v^2}}`$, parameterized by their proper times. That is, $`\beta `$ is a stationary observer with $`x=0`$, $`y=0`$, $`z=0`$ and $`\beta ^{}`$ is an observer moving from $`x=0`$, $`y=0`$, $`z=0`$ to $`x=1`$, $`y=0`$, $`z=0`$ with velocity of modulus $`v`$ and returning (see Fig. 5). It is satisfied that
$$V_{\mathrm{kin}\beta \left(\tau \right)}=\{\begin{array}{c}v\frac{}{x}|_{\beta \left(\tau \right)}\mathrm{if}\tau [0,\frac{1}{v}]\hfill \\ v\frac{}{x}|_{\beta \left(\tau \right)}\mathrm{if}\tau ]\frac{1}{v},\frac{2}{v}]\hfill \end{array},$$
$$V_{\mathrm{spec}\beta \left(\tau \right)}=\{\begin{array}{c}v\frac{}{x}|_{\beta \left(\tau \right)}\mathrm{if}\tau [0,\frac{1+v}{v}]\hfill \\ v\frac{}{x}|_{\beta \left(\tau \right)}\mathrm{if}\tau ]\frac{1+v}{v},\frac{2}{v}]\hfill \end{array}.$$
Applying (17), we obtain $`V_{\mathrm{Fermi}\beta \left(\tau \right)}=V_{\mathrm{kin}\beta \left(\tau \right)}`$. Moreover
$$S_{\mathrm{obs}\beta \left(\tau \right)}=\{\begin{array}{c}\frac{v\tau }{1+v}\frac{}{x}|_{\beta \left(\tau \right)}\mathrm{if}\tau [0,\frac{1+v}{v}]\hfill \\ \frac{2v\tau }{1v}\frac{}{x}|_{\beta \left(\tau \right)}\mathrm{if}\tau ]\frac{1+v}{v},\frac{2}{v}]\hfill \end{array}.$$
Hence, by (24) we have
$$V_{\mathrm{ast}\beta \left(\tau \right)}=\{\begin{array}{c}\frac{v}{1+v}\frac{}{x}|_{\beta \left(\tau \right)}\mathrm{if}\tau [0,\frac{1+v}{v}]\hfill \\ \frac{v}{1v}\frac{}{x}|_{\beta \left(\tau \right)}\mathrm{if}\tau ]\frac{1+v}{v},\frac{2}{v}]\hfill \end{array}.$$
Consequently, $`V_{\mathrm{ast}\beta \left(\tau \right)}]0,1/2[`$ if $`\tau [0,\frac{1+v}{v}]`$, i.e. if $`\beta ^{}`$ is moving away radially. On the other hand, $`V_{\mathrm{ast}\beta \left(\tau \right)}]0,+\mathrm{}[`$ if $`\tau ]\frac{1+v}{v},\frac{2}{v}]`$, i.e. if $`\beta ^{}`$ is getting closer radially (see fig. 6). This corresponds to what $`\beta `$ observes.
###### Example 5.2
Let us suppose that the spacetime is flat and we see an alien spaceship coming to Earth from a planet at 9 lightyears (this distance can be measured by parallax, since this method estimates the affine distance from the planet to Earth observed by someone on Earth). Let us suppose that the spaceship is coming radially, and so, we can measure the modulus of its spectroscopic relative velocity (see 4.2). Supposing that this modulus is $`v=0.9`$, the spaceship will take 10 years to arrive at Earth from its planet. However, since light takes 9 years to arrive at us, there is only 1 year left for the arrival of the spaceship. This result can also be obtained by using expression (24): in our case, the modulus of the astrometric relative velocity is $`\frac{0.9}{10.9}=9`$, and we will therefore observe that it takes 1 year to arrive.
###### Remark 5.2
There is an open problem in general relativity, that consists on finding expressions for $`V_{\mathrm{Fermi}}`$ and $`V_{\mathrm{ast}}`$ in terms of $`U`$, $`_UU`$, $`U^{}`$, $`S`$ and $`S_{\mathrm{obs}}`$, analogously to Propositions 5.1 and 5.2, avoiding $`_US`$, $`_US_{\mathrm{obs}}`$, or any term involving the evolution of $`S`$, $`S_{\mathrm{obs}}`$. It would be very useful in the calculations of the relative velocities.
## 6 Examples in General Relativity
In this section, we are going to study some fundamental examples in Schwarzschild and Robertson-Walker spacetimes. See for an interesting and complete study of the relative velocities of a radially receding test particle with respect to / observed by a central observer in a Schwarzschild-de Sitter spacetime.
### 6.1 Stationary observers in Schwarzschild spacetime
In the Schwarzschild metric with spherical coordinates
$$\mathrm{d}s^2=a^2\left(r\right)\mathrm{d}t^2+\frac{1}{a^2\left(r\right)}\mathrm{d}r^2+r^2\left(\mathrm{d}\theta ^2+\mathrm{sin}^2\theta \mathrm{d}\phi ^2\right),$$
where $`a\left(r\right)=\sqrt{1\frac{2m}{r}}`$ and $`r>2m`$, let us consider two equatorial stationary observers, $`\beta _1\left(\tau \right)=(\frac{1}{a_1}\tau ,r_1,\pi /2,0)`$ and $`\beta _2\left(\tau \right)=(\frac{1}{a_2}\tau ,r_2,\pi /2,0)`$ with $`\tau `$, $`r_2>r_1>2m`$, $`a_1:=a\left(r_1\right)`$ and $`a_2:=a\left(r_2\right)`$, and let $`U`$ be the 4-velocity of $`\beta _2`$, i.e. $`U:=\frac{1}{a_2}\frac{}{t}`$. We are going to study the relative velocities of $`\beta _1`$ with respect to / observed by $`\beta _2`$.
#### 6.1.1 Kinematic and Fermi relative velocities. Fermi distance
Let us consider the vector field $`X:=a\left(r\right)\frac{}{r}`$; it is spacelike, unit, geodesic, and orthogonal to $`U`$. Since $`_X\left(\frac{1}{a\left(r\right)}\frac{}{t}\right)=0`$, we have that the kinematic relative velocity $`V_{\mathrm{kin}}`$ of $`\beta _1`$ with respect to $`\beta _2`$ is given by $`V_{\mathrm{kin}}=0`$.
It is clear (a priori) that the relative position $`S`$ of $`\beta _1`$ with respect to $`\beta _2`$ is proportional to $`\frac{}{r}`$ and the proportionality factor is constant. So, it is easy to prove that $`_US`$ is proportional to $`U`$ and therefore, the Fermi relative velocity $`V_{\mathrm{Fermi}}`$ of $`\beta _1`$ with respect to $`\beta _2`$ reads $`V_{\mathrm{Fermi}}=0`$.
Nevertheless, we are going to calculate the Fermi distance and $`S`$:
Let $`\alpha \left(\sigma \right)=(t_0,\alpha ^r\left(\sigma \right),\pi /2,0)`$ be an integral curve of $`X`$ such that $`q:=\alpha \left(\sigma _1\right)\beta _1`$ and $`p:=\alpha \left(\sigma _2\right)\beta _2`$, with $`\sigma _2>\sigma _1`$ (i.e. $`\alpha \left(\sigma \right)`$ is a spacelike geodesic from $`q`$ to $`p`$, parameterized by its arclength, and its tangent vector at $`p`$ is $`X_p`$). Then, by Proposition 3.2, the Fermi distance $`d_{U_p}^{\mathrm{Fermi}}(q,p)`$ from $`q`$ to $`p`$ with respect to $`U_p`$ is $`\sigma _2\sigma _1`$. Since $`\alpha `$ is an integral curve of $`X`$, we have $`\stackrel{.}{\alpha }^r\left(\sigma \right)=\sqrt{1\frac{2m}{\alpha ^r\left(\sigma \right)}}`$. So, $`_{r_1}^{r_2}\left(1\frac{2m}{\alpha ^r\left(\sigma \right)}\right)^{1/2}\stackrel{.}{\alpha }^r\left(\sigma \right)d\sigma =\sigma _2\sigma _1`$, and then
$$d_{U_p}^{\mathrm{Fermi}}(q,p)=2m\mathrm{ln}\left(\frac{\left(1a_1\right)\sqrt{r_1}}{\left(1a_2\right)\sqrt{r_2}}\right)+r_2a_2r_1a_1.$$
(34)
Since (34) does not depends on $`t_0`$, the Fermi distance from $`\beta _1`$ to $`\beta _2`$ with respect to $`\beta _2`$ is also given by expression (34). Hence, by (7), the relative position $`S`$ of $`\beta _1`$ with respect to $`\beta _2`$ is given by
$$S=\left(2m\mathrm{ln}\left(\frac{\left(1a_2\right)\sqrt{r_2}}{\left(1a_1\right)\sqrt{r_1}}\right)+r_1a_1r_2a_2\right)a_2\frac{}{r}.$$
#### 6.1.2 Spectroscopic and astrometric relative velocities. Affine distance
It is easy to prove that the spectroscopic relative velocity $`V_{\mathrm{spec}}`$ of $`\beta _1`$ observed by $`\beta _2`$ is radial. Since the gravitational redshift is given by $`\frac{a_2}{a_1}`$ (see ), by (13) we obtain
$$V_{\mathrm{spec}}=a_2\frac{a_2^2a_1^2}{a_2^2+a_1^2}\frac{}{r}.$$
(35)
Expression (35) is also obtained in . Since $`V_{\mathrm{spec}}=\frac{a_2^2a_1^2}{a_2^2+a_1^2}`$, we have $`lim_{r_12m}V_{\mathrm{spec}}=1`$ (see Fig. 7).
On the other hand, it is clear (a priori) that the relative position $`S_{\mathrm{obs}}`$ of $`\beta _1`$ observed by $`\beta _2`$ is proportional to $`\frac{}{r}`$ and the proportionality factor is constant. So, it is easy to prove that $`_US_{\mathrm{obs}}`$ is proportional to $`U`$ and therefore, the astrometric relative velocity $`V_{\mathrm{ast}}`$ of $`\beta _1`$ observed by $`\beta _2`$ reads $`V_{\mathrm{ast}}=0`$.
Nevertheless, we are going to calculate the affine distance and $`S_{\mathrm{obs}}`$:
In it is proved (by using Proposition 4.3) that the affine distance from $`\beta _1`$ to $`\beta _2`$ observed by $`\beta _2`$ is $`\frac{r_2r_1}{a_2}`$. Hence, by (16), the relative position $`S_{\mathrm{obs}}`$ of $`\beta _1`$ observed by $`\beta _2`$ is given by
$$S_{\mathrm{obs}}=\left(r_1r_2\right)\frac{}{r}.$$
(36)
### 6.2 Free-falling observers in Schwarzschild spacetime
Let us consider a radial free-falling observer $`\beta _1`$ parameterized by the coordinate time $`t`$, $`\beta _1\left(t\right)=(t,\beta _1^r\left(t\right),\pi /2,0)`$. Given an event $`q=(t_1,r_1,\pi /2,0)\beta _1`$, the 4-velocity of $`\beta _1`$ at $`q`$ is given by
$$u_1=\frac{E}{a_1^2}\frac{}{t}|_q\sqrt{E^2a_1^2}\frac{}{r}|_q,$$
(37)
where $`E`$ is a constant of motion given by $`E:=\left(\frac{12m/r_0}{1v_0^2}\right)^{1/2}`$, $`r_0`$ is the radial coordinate at which the fall begins, $`v_0`$ is the initial velocity (see ), and $`a_1:=a\left(r_1\right)`$. Moreover, let us consider an equatorial stationary observer $`\beta _2\left(\tau \right)=(\frac{1}{a_2}\tau ,r_2,\pi /2,0)`$ with $`\tau `$, $`r_2r_1>2m`$, $`a_2:=a\left(r_2\right)`$, and $`U:=\frac{1}{a_2}\frac{}{t}`$ its 4-velocity. We are going to study the relative velocities of $`\beta _1`$ with respect to / observed by $`\beta _2`$ at $`p`$, where $`p`$ will be a determined event of $`\beta _2`$.
#### 6.2.1 Kinematic and Fermi relative velocities
Let $`p=(t_1,r_2,\pi /2,0)`$. This is the unique event of $`\beta _2`$ such that $`qL_{p,U_p}`$, i.e. there exists a spacelike geodesic $`\alpha \left(\sigma \right)`$ from $`q=\alpha \left(\sigma _1\right)`$ to $`p=\alpha \left(\sigma _2\right)`$ such that the tangent vector $`\stackrel{.}{\alpha }\left(\sigma _2\right)`$ is orthogonal to $`U_p`$. We can consider $`\alpha \left(\sigma \right)`$ parameterized by its arclength and $`\sigma _2>\sigma _1`$. So, $`\alpha \left(\sigma \right)`$ is an integral curve of the vector field $`X=a\left(r\right)\frac{}{r}`$. If we parallelly transport $`u_1`$ from $`q`$ to $`p`$ along $`\alpha `$ we obtain $`\tau _{qp}u_1=\frac{E}{a_1a_2}\frac{}{t}|_p\frac{a_2}{a_1}\sqrt{E^2a_1^2}\frac{}{r}|_p`$. By (4), the kinematic relative velocity $`V_{\mathrm{kin}p}`$ of $`\beta _1`$ with respect to $`\beta _2`$ at $`p`$ reads
$$V_{\mathrm{kin}p}=a_2\sqrt{1\frac{a_1^2}{E^2}}\frac{}{r}|_p.$$
Since $`V_{\mathrm{kin}p}=\sqrt{1\frac{a_1^2}{E^2}}`$, it is satisfied that $`lim_{r_12m}V_{\mathrm{kin}p}=1`$. See Appendix for a deeper analysis of this function.
On the other hand, by (34), the relative position $`S`$ of $`\beta _1`$ with respect to $`\beta _2`$ is given by
$$S=\left(2m\mathrm{ln}\left(\frac{\left(1a_2\right)\sqrt{r_2}}{\left(1a\left(\beta _1^r\left(t\right)\right)\right)\sqrt{\beta _1^r\left(t\right)}}\right)+\beta _1^r\left(t\right)a\left(\beta _1^r\left(t\right)\right)r_2a_2\right)a_2\frac{}{r}.$$
By (5), the Fermi relative velocity $`V_{\mathrm{Fermi}}`$ of $`\beta _1`$ with respect to $`\beta _2`$ reads
$$V_{\mathrm{Fermi}}=\left(_US\right)^r\frac{}{r}=\frac{1}{a_2}\frac{S^r}{t}\frac{}{r}=\frac{1}{a_2}\frac{\stackrel{.}{\beta }_1^r\left(t\right)}{a\left(\beta _1^r\left(t\right)\right)}\frac{}{r}$$
Taking into account (37), we have $`\stackrel{.}{\beta }_1^r\left(t_1\right)=a_1^2\sqrt{1\frac{a_1^2}{E^2}}`$. Hence
$$V_{\mathrm{Fermi}p}=\frac{a_1}{a_2}\sqrt{1\frac{a_1^2}{E^2}}\frac{}{r}|_p.$$
Since $`V_{\mathrm{Fermi}p}=\frac{a_1}{a_2^2}\sqrt{1\frac{a_1^2}{E^2}}`$, it is satisfied that $`lim_{r_12m}V_{\mathrm{Fermi}p}=0`$. See Appendix for a deeper analysis of this function.
#### 6.2.2 Spectroscopic and astrometric relative velocities
Let $`p`$ be the unique event of $`\beta _2`$ such that there exists a light ray $`\lambda `$ from $`q`$ to $`p`$, and let us suppose that $`p=(t_2,r_2,\pi /2,0)`$. In it is shown that the spectroscopic relative velocity $`V_{\mathrm{spec}p}`$ of $`\beta _1`$ observed by $`\beta _2`$ at $`p`$ is given by
$$V_{\mathrm{spec}p}=a_2\frac{\left(a_2^2+a_1^2\right)\sqrt{1\frac{a_1^2}{E^2}}+\left(a_2^2a_1^2\right)}{\left(a_2^2a_1^2\right)\sqrt{1\frac{a_1^2}{E^2}}+\left(a_2^2+a_1^2\right)}\frac{}{r}|_p.$$
(38)
Since $`V_{\mathrm{spec}p}=\frac{\left(a_2^2+a_1^2\right)\sqrt{1\frac{a_1^2}{E^2}}+\left(a_2^2a_1^2\right)}{\left(a_2^2a_1^2\right)\sqrt{1\frac{a_1^2}{E^2}}+\left(a_2^2+a_1^2\right)}`$, it follows that $`lim_{r_12m}V_{\mathrm{spec}p}=1`$. See Appendix for a deeper analysis of this function.
On the other hand, it can be checked that
$$\lambda \left(r\right):=(t_1+rr_1+2m\mathrm{ln}\left(\frac{r2m}{r_12m}\right),r,\pi /2,0),r[r_1,r_2]$$
is a light ray from $`q=\lambda \left(r_1\right)`$ to $`p=\lambda \left(r_2\right)`$. So,
$$t_2=\lambda ^t\left(r_2\right)=t_1+r_2r_1+2m\mathrm{ln}\left(\frac{r_22m}{r_12m}\right).$$
(39)
Let us define implicitly the function $`f\left(t\right)`$ by the expression
$$f\left(t\right):=t\left(r_2\beta _1^r\left(f\left(t\right)\right)+2m\mathrm{ln}\left(\frac{r_22m}{\beta _1^r\left(f\left(t\right)\right)2m}\right)\right).$$
(40)
Taking into account (39), $`f\left(t\right)`$ is the coordinate time at which $`\beta _1`$ emits a light ray that arrives at $`\beta _2`$ at coordinate time $`t`$. Applying (36), the relative position $`S_{\mathrm{obs}}`$ of $`\beta _1`$ observed by $`\beta _2`$ reads
$$S_{\mathrm{obs}}=\left(\beta _1^r\left(f\left(t\right)\right)r_2\right)\frac{}{r}.$$
By (14), the astrometric relative velocity $`V_{\mathrm{ast}}`$ of $`\beta _1`$ observed by $`\beta _2`$ is given by
$$V_{\mathrm{ast}}=\left(_US_{\mathrm{obs}}\right)^r\frac{}{r}=\frac{1}{a_2}\frac{S_{\mathrm{obs}}^r}{t}\frac{}{r}=\frac{1}{a_2}\stackrel{.}{\beta }_1^r\left(f\left(t\right)\right)\stackrel{.}{f}\left(t\right)\frac{}{r}.$$
From (40), we have $`\stackrel{.}{f}\left(t_2\right)=\frac{a_1^2}{a_1^2\left(a_1^21\right)\stackrel{.}{\beta }_1^r\left(t_1\right)}`$. Moreover, taking into account (37), we have $`\stackrel{.}{\beta }_1^r\left(t_1\right)=a_1^2\sqrt{1\frac{a_1^2}{E^2}}`$. Hence
$$V_{\mathrm{ast}p}=\frac{a_1^2}{a_2}\frac{\sqrt{1\frac{a_1^2}{E^2}}}{1+\left(a_1^21\right)\sqrt{1\frac{a_1^2}{E^2}}}\frac{}{r}|_p,$$
(41)
and, in consequence, $`V_{\mathrm{ast}p}=\frac{a_1^2}{a_2^2}\frac{\sqrt{1\frac{a_1^2}{E^2}}}{1+\left(a_1^21\right)\sqrt{1\frac{a_1^2}{E^2}}}`$, concluding that $`lim_{r_12m}V_{\mathrm{ast}p}=\frac{1}{a_2^2}\frac{2E^2}{1+2E^2}]0,+\mathrm{}[`$. See Appendix for a deeper analysis of this function.
### 6.3 Comoving observers in Robertson-Walker spacetime
See for an interesting and complete study of the Fermi relative velocity of a comoving test particle with respect to / observed by a comoving observer in an expanding Robertson-Walker spacetime. Currently, there is a work in process that studies the other relative velocities in this case, with examples in the Milne, de-Sitter, radiation-dominated an matter-dominated universes.
In a Robertson-Walker metric with cartesian coordinates
$$\mathrm{d}s^2=\mathrm{d}t^2+\frac{a^2\left(t\right)}{\left(1+\frac{1}{4}kr^2\right)^2}\left(\mathrm{d}x^2+\mathrm{d}y^2+\mathrm{d}z^2\right),$$
where $`a\left(t\right)`$ is the scale factor, $`k=1,0,1`$ and $`r:=\sqrt{x^2+y^2+z^2}`$, we consider two comoving (in the classical sense, see ) observers $`\beta _0\left(\tau \right)=(\tau ,0,0,0)`$ and $`\beta _1\left(\tau \right)=(\tau ,x_1,0,0)`$ with $`\tau `$ and $`x_1>0`$. Let $`t_0`$, $`p:=\beta _0\left(t_0\right)`$ and $`u:=\stackrel{.}{\beta _0}\left(t_0\right)=\frac{}{t}|_p`$ (i.e. the 4-velocity of $`\beta _0`$ at $`p`$). We are going to study the relative velocities of $`\beta _1`$ with respect to / observed by $`\beta _0`$ at $`p`$.
#### 6.3.1 Kinematic and Fermi relative velocities
The vector field
$$X:=\sqrt{\frac{a_0^2}{a^2\left(t\right)}1}\frac{}{t}+\frac{a_0}{a^2\left(t\right)}\left(1+\frac{1}{4}kx^2\right)\frac{}{x}$$
is geodesic, spacelike, unit, and $`X_p`$ is orthogonal to $`u`$, i.e. it is tangent to the Landau submanifold $`L_{p,u}`$. Let $`\beta _1\left(t_1\right)=:q`$ be the unique event of $`\beta _1L_{p,u}`$. We can find $`t_1`$ for a given scale factor $`a\left(t\right)`$ taking into account the expression of $`X`$, but we can not find an explicit expression in the general case. If $`u^{}:=\stackrel{.}{\beta _1}\left(t_1\right)=\frac{}{t}|_q`$, then $`\tau _{qp}u^{}=\frac{a_0}{a_1}\frac{}{t}|_p+\sqrt{\frac{1}{a_1^2}\frac{1}{a_0^2}}\frac{}{x}|_p`$, where $`a_1:=a\left(t_1\right)`$ (it is well defined because $`a_0a_1>0`$). So, by (4), the kinematic relative velocity $`V_{\mathrm{kin}p}`$ of $`\beta _1`$ with respect to $`\beta _0`$ at $`p`$ is given by
$$V_{\mathrm{kin}p}=\frac{1}{a_0^2}\sqrt{a_0^2a_1^2}\frac{}{x}|_p.$$
Given a scale factor $`a\left(t\right)`$, the Fermi distance $`d^{\mathrm{Fermi}}`$ from $`\beta _1`$ to $`\beta _0`$ with respect to $`\beta _0`$ can be also found, taking into account the expression of $`X`$. So, the relative position $`S`$ of $`\beta _1`$ with respect to $`\beta _0`$ reads
$$S=d^{\mathrm{Fermi}}\frac{\left(1+\frac{1}{4}kr^2\right)}{a\left(t\right)}\frac{}{x},$$
because $`d^{\mathrm{Fermi}}=S`$. Hence, the Fermi relative velocity $`V_{\mathrm{Fermi}p}`$ of $`\beta _1`$ with respect to $`\beta _0`$ at $`p`$ is given by
$$V_{\mathrm{Fermi}p}=\left(\frac{\mathrm{d}}{\mathrm{d}t}\left(\frac{d^{\mathrm{Fermi}}}{a\left(t\right)}\right)|_{t=t_0}+d_p^{\mathrm{Fermi}}\frac{\stackrel{.}{a}\left(t_0\right)}{a_0^2}\right)\frac{}{x}|_p.$$
#### 6.3.2 Spectroscopic and astrometric relative velocities
Let $`\lambda `$ be a light ray received by $`\beta _0`$ at $`p`$ and emitted from $`\beta _1`$ at $`\beta _1\left(t_1\right)`$. Note that $`t_1`$ can be found from $`x_1`$ and $`t_0`$ taking into account that $`_0^{x_1}\frac{\mathrm{d}x}{1+\frac{1}{4}kx^2}=_{t_1}^{t_0}\frac{\mathrm{d}t}{a\left(t\right)}`$. It can be easily proved that the spectroscopic relative velocity $`V_{\mathrm{spec}p}`$ of $`\beta _1`$ observed by $`\beta _0`$ at $`p`$ is radial (by isotropy). So, by (13) taking into account that the cosmological shift is given by $`\frac{a_0}{a_1}`$ (see ), where $`a_0:=a\left(t_0\right)`$ and $`a_1:=a\left(t_1\right)`$, we have
$$V_{\mathrm{spec}p}=\frac{1}{a_0}\frac{a_0^2a_1^2}{a_0^2+a_1^2}\frac{}{x}|_p.$$
(42)
Given a scale factor $`a\left(t\right)`$, the affine distance $`d^{\mathrm{aff}}`$ from $`\beta _1`$ to $`\beta _0`$ observed by $`\beta _0`$ can be found. So, the relative position $`S_{\mathrm{obs}}`$ of $`\beta _1`$ observed by $`\beta _0`$ is given by
$$S_{\mathrm{obs}}=d^{\mathrm{aff}}\frac{\left(1+\frac{1}{4}kr^2\right)}{a\left(t\right)}\frac{}{x},$$
because $`d^{\mathrm{aff}}=S_{\mathrm{obs}}`$. Hence, the astrometric relative velocity $`V_{\mathrm{ast}p}`$ of $`\beta _1`$ observed by $`\beta _0`$ at $`p`$ reads
$$V_{\mathrm{ast}p}=\left(\frac{\mathrm{d}}{\mathrm{d}t}\left(\frac{d^{\mathrm{aff}}}{a\left(t\right)}\right)|_{t=t_0}+d_p^{\mathrm{aff}}\frac{\stackrel{.}{a}\left(t_0\right)}{a_0^2}\right)\frac{}{x}|_p.$$
(43)
Let us study these relative velocities in more detail. In cosmology it is usual to consider the scale factor in the form
$$a\left(t\right)=a_0\left(1+H_0\left(tt_0\right)\frac{1}{2}q_0H_0^2\left(tt_0\right)^2\right)+๐ช\left(H_0^3\left(tt_0\right)^3\right),$$
where $`t_0`$, $`a_0=a\left(t_0\right)>0`$, $`H\left(t\right)=\stackrel{.}{a}\left(t\right)/a\left(t\right)`$ is the Hubble โconstantโ, $`H_0=H\left(t_0\right)>0`$, $`q\left(t\right)=a\left(t\right)\stackrel{..}{a}\left(t\right)/\stackrel{.}{a}\left(t\right)^2`$ is the deceleration coefficient, and $`q_0=q\left(t_0\right)`$, with $`\left|H_0\left(tt_0\right)\right|1`$ (see ). This corresponds to a universe in decelerated expansion and the time scales that we are going to use are relatively small. Let us define $`p:=\beta _0\left(t_0\right)`$ and $`u:=\stackrel{.}{\beta _0}\left(t_0\right)=\frac{}{t}|_p`$.
We are going to express the spectroscopic and the astrometric relative velocity of $`\beta _1`$ observed by $`\beta _0`$ at $`p`$ in terms of the redshift parameter at $`t=t_0`$, defined as $`z_0:=\frac{a_0}{a_1}1`$, where $`a_1:=a\left(t_1\right)`$. This parameter is very usual in cosmology since it can be measured by spectroscopic observations. By (42), the spectroscopic relative velocity $`V_{\mathrm{spec}p}`$ of $`\beta _1`$ observed by $`\beta _0`$ at $`p`$ is given by
$$V_{\mathrm{spec}p}=\frac{1}{a_0}\frac{a_0^4\left(z_0+1\right)^2}{a_0^4+\left(z_0+1\right)^2}\frac{}{x}|_p.$$
(44)
In it is shown that the affine distance $`d^{\mathrm{aff}}`$ from $`\beta _1`$ to $`\beta _0`$ observed by $`\beta _0`$ reads
$$d^{\mathrm{aff}}\left(t\right)=\frac{z\left(t\right)}{H\left(t\right)}\left(1\frac{1}{2}\left(3+q\left(t\right)\right)z\left(t\right)\right)+๐ช\left(z^3\left(t\right)\right),$$
where $`z\left(t\right)`$ is the redshift function. So, by (43), the astrometric relative velocity $`V_{\mathrm{ast}p}`$ of $`\beta _1`$ observed by $`\beta _0`$ at $`p`$ is given by
$$V_{\mathrm{ast}p}=\left(\frac{\stackrel{.}{z}\left(t_0\right)}{a_0H_0}+\frac{z_0}{a_0}\left(q_0+1\frac{\stackrel{.}{z}\left(t_0\right)}{H_0}\left(3+q_0\right)\right)+๐ช\left(z_0^2\right)\right)\frac{}{x}|_p.$$
Hence, if we suppose that $`\stackrel{.}{z}\left(t_0\right)0`$ (i.e., the redshift is constant in our time scale), then
$$V_{\mathrm{ast}p}\left(\frac{z_0}{a_0}\left(q_0+1\right)+๐ช\left(z_0^2\right)\right)\frac{}{x}|_p.$$
(45)
## 7 Discussion and comments
It is usual to consider the spectroscopic relative velocity as a non-acceptable โphysical velocityโ. However, in this paper we have defined it in a geometric way, showing that it is, in fact, a very plausible physical velocity.
* Firstly, in other works (see , ), we have discussed pros and cons of spacelike and lightlike simultaneities, coming to the conclusion that lightlike simultaneity is physically and mathematically more suitable. Since the spectroscopic relative velocity is the natural generalization (in the framework of lightlike simultaneity) of the usual concept of relative velocity (given by (2)), it might have a lot of importance.
* Secondly, there are some good properties suggesting that the spectroscopic relative velocity has a lot of physical sense. For instance, if we work with the spectroscopic relative velocity, it is shown in that gravitational redshift is just a particular case of a generalized Doppler effect.
Nevertheless, all four concepts of relative velocity have full physical sense and they must be studied equally.
Finally, one can wonder whether the discussed concepts of relative velocity can be actually determined experimentally. A priori, only the spectroscopic and astrometric relative velocities can be measured by direct observation. The shift allows us to find relations between the modulus of the spectroscopic relative velocity and its tangential component, as we show in (12). But, in general, it is not enough information to determine it completely (as we discuss in Remark 4.1), unless we make some assumptions (see Remark 4.2) or we use a model for the spacetime and apply some expressions like (35), (38), or (44). Finding the astrometric relative velocity is basically the same problem as finding the optical coordinates. It is non-trivial and it has been widely treated, for instance, in . Nevertheless, expressions like (41) or (45) could be very useful in particular situations. Since the measure of these velocities is rather difficult, any expression relating them can be very helpful in order to determine them, as, for example, expression (24) in special relativity.
## Appendix
### A.1 Free-falling observers in Schwarzschild spacetime
We are going to study the modulus of the relative velocities of a radially inward free-falling observer (or test particle) at $`r_1>2m`$ with respect to / observed by a stationary observer at $`r_2r_1`$, according to the results of Section 6.2. The radial coordinate that we are going to use is $`a=\sqrt{1\frac{2m}{r}}`$, taking values from $`0`$ (when $`r2m`$) to $`1`$ (when $`r+\mathrm{}`$); so, the radial parameters are $`a_1=a\left(r_1\right)`$ and $`a_2=a\left(r_2\right)`$. Another parameter is given by the energy $`E>0`$ of the free falling test particle. In our study, we are going to consider the modulus of the relative velocities as functions of $`a_1`$, taking $`a_2`$ and $`E`$ as parameters. So, taking into account the definition of $`E`$, it is clear that $`0<a_1a_{1\mathrm{m}\mathrm{a}\mathrm{x}}:=\mathrm{min}\{E,a_2\}`$.
#### A.1.1 Kinematic relative velocity
The modulus of the kinematic relative velocity is given by
$$v_{\mathrm{kin}}=\sqrt{1\frac{a_1^2}{E^2}}.$$
Note that $`v_{\mathrm{kin}}`$ does not depend on $`a_2`$. It satisfies $`0v_{\mathrm{kin}}<1`$, it is decreasing with $`a_1`$ (i.e. increasing with time), and $`lim_{a_10}v_{\mathrm{kin}}=1`$. Moreover:
* If $`Ea_2`$, then $`v_{\mathrm{kin}}`$ takes its minimum at $`a_1=a_{1\mathrm{m}\mathrm{a}\mathrm{x}}=E`$ and it is $`0`$.
* If $`E>a_2`$, then $`v_{\mathrm{kin}}`$ takes its minimum at $`a_1=a_{1\mathrm{m}\mathrm{a}\mathrm{x}}=a_2`$ and it is given by
$$v_{\mathrm{kin}}_{\mathrm{min}}:=\sqrt{1\frac{a_2^2}{E^2}}.$$
(46)
We have that $`lim_{E+\mathrm{}}v_{\mathrm{kin}}_{\mathrm{min}}=1`$.
#### A.1.2 Fermi relative velocity
The modulus of the Fermi relative velocity is given by
$$v_{\mathrm{Fermi}}=\frac{a_1}{a_2^2}\sqrt{1\frac{a_1^2}{E^2}}.$$
It satisfies $`lim_{a_10}v_{\mathrm{Fermi}}=0`$. Moreover:
* If $`E<\sqrt{2}a_2`$, then $`v_{\mathrm{Fermi}}`$ takes its maximum at $`a_1=\frac{E}{\sqrt{2}}`$ and it is given by
$$v_{\mathrm{Fermi}}_{\mathrm{max}}:=\frac{E}{2a_2^2}<\frac{1}{\sqrt{2}a_2}.$$
It is increasing with $`E`$, becoming superluminal (i.e. $`>1`$) if, in addition, $`E>2a_2^2`$. Note that it is only possible if $`a_2<\frac{1}{\sqrt{2}}`$ (i.e. $`r_2<4m`$). In this case, $`v_{\mathrm{Fermi}}`$ is superluminal if
$$\frac{E^2}{2}\left(1\sqrt{14\frac{a_2^4}{E^2}}\right)<a_1^2<\frac{E^2}{2}\left(1+\sqrt{14\frac{a_2^4}{E^2}}\right).$$
* If $`E\sqrt{2}a_2`$, then $`v_{\mathrm{Fermi}}`$ is increasing with $`a_1`$ (i.e. decreasing with time) and takes its maximum at $`a_1=a_{1\mathrm{m}\mathrm{a}\mathrm{x}}=a_2`$, given by
$$v_{\mathrm{Fermi}}_{\mathrm{max}}:=\frac{1}{a_2}\sqrt{1\frac{a_2^2}{E^2}}.$$
(47)
It is increasing with $`E`$, becoming superluminal if $`E>\frac{a_2}{\sqrt{1a_2^2}}`$; nevertheless, it is bounded by $`lim_{E+\mathrm{}}v_{\mathrm{Fermi}}_{\mathrm{max}}=\frac{1}{a_2}>1`$. In this case, $`v_{\mathrm{Fermi}}`$ is superluminal if
$$a_1^2>\frac{E^2}{2}\left(1\sqrt{14\frac{a_2^4}{E^2}}\right).$$
On the other hand,
* If $`Ea_2`$, then $`v_{\mathrm{Fermi}}`$ takes its minimum at $`a_1=a_{1\mathrm{m}\mathrm{a}\mathrm{x}}=E`$ and it is $`0`$.
* If $`a_2<E<\sqrt{2}a_2`$, then $`v_{\mathrm{Fermi}}`$ has a relative minimum at $`a_1=a_{1\mathrm{m}\mathrm{a}\mathrm{x}}=a_2`$ and it is given by (47). Note that it is superluminal if, in addition, $`E>\frac{a_2}{\sqrt{1a_2^2}}`$.
#### A.1.3 Spectroscopic relative velocity
The modulus of the spectroscopic relative velocity is given by
$$v_{\mathrm{spec}}=\frac{\left(a_2^2+a_1^2\right)\sqrt{1\frac{a_1^2}{E^2}}+\left(a_2^2a_1^2\right)}{\left(a_2^2a_1^2\right)\sqrt{1\frac{a_1^2}{E^2}}+\left(a_2^2+a_1^2\right)}.$$
It satisfies $`0v_{\mathrm{spec}}<1`$, it is decreasing with $`a_1`$ (i.e. increasing with time), and $`lim_{a_10}v_{\mathrm{spec}}=1`$. Moreover:
* If $`Ea_2`$, then $`v_{\mathrm{spec}}`$ takes its minimum at $`a_1=a_{1\mathrm{m}\mathrm{a}\mathrm{x}}=E`$ and it is given by
$$v_{\mathrm{spec}}_{\mathrm{min}}:=\frac{a_2^2E^2}{a_2^2+E^2}.$$
We have that $`v_{\mathrm{spec}}_{\mathrm{min}}`$ is decreasing with $`E`$, and it only vanishes at $`E=a_2`$.
* If $`E>a_2`$, then $`v_{\mathrm{spec}}`$ takes its minimum at $`a_1=a_{1\mathrm{m}\mathrm{a}\mathrm{x}}=a_2`$ and it is given by
$$v_{\mathrm{spec}}_{\mathrm{min}}:=\sqrt{1\frac{a_2^2}{E^2}}.$$
Note that this is the same minimum as in the kinematic case (see (46)).
#### A.1.4 Astrometric relative velocity
The modulus of the astrometric relative velocity is given by
$$v_{\mathrm{ast}}=\frac{a_1^2}{a_2^2}\frac{\sqrt{1\frac{a_1^2}{E^2}}}{1+\left(a_1^21\right)\sqrt{1\frac{a_1^2}{E^2}}}.$$
It is important to note that $`lim_{E+\mathrm{}}v_{\mathrm{ast}}=\frac{1}{a_2^2}>1`$ for all $`a_1`$. So, given $`a_2`$, there exists always a big enough energy (see (48) below) such that $`v_{\mathrm{ast}}`$ is superluminal for all $`a_1`$.
It is decreasing with $`a_1`$ (i.e. increasing with time), and it has a supremum
$$v_{\mathrm{ast}}_{\mathrm{sup}}:=\underset{a_10}{lim}v_{\mathrm{ast}}=\frac{1}{a_2^2}\frac{2E^2}{1+2E^2}.$$
We have that $`v_{\mathrm{ast}}_{\mathrm{sup}}`$ is increasing with $`E`$, becoming superluminal if $`E>\frac{1}{\sqrt{2}}\frac{a_2}{\sqrt{1a_2^2}}`$ (but it is bounded by $`\frac{1}{a_2^2}`$). In this case, $`v_{\mathrm{ast}}`$ is superluminal if
$$a_1^2<\frac{E^2}{2}\left(1+\sqrt{1+\frac{4}{E^2}\frac{a_2^2}{1a_2^2}}\right)\frac{a_2^2}{1a_2^2}.$$
Moreover:
* If $`Ea_2`$, then $`v_{\mathrm{ast}}`$ takes its minimum at $`a_1=a_{1\mathrm{m}\mathrm{a}\mathrm{x}}=E`$ and it is $`0`$.
* If $`E>a_2`$, then $`v_{\mathrm{ast}}`$ takes its minimum at $`a_1=a_{1\mathrm{m}\mathrm{a}\mathrm{x}}=a_2`$ and it is given by
$$v_{\mathrm{ast}}_{\mathrm{min}}:=\frac{\sqrt{1\frac{a_2^2}{E^2}}}{1+\left(a_2^21\right)\sqrt{1\frac{a_2^2}{E^2}}}.$$
It is increasing with $`E`$, becoming superluminal if
$$E>\frac{a_2\left(2a_2^2\right)}{\sqrt{\left(2a_2^2\right)^21}}.$$
(48)
See Figures 8 ($`a_2=0.2`$), 9 ($`a_2=0.5`$), 10, ($`a_2=0.70711`$, i.e. $`r_2=4m`$), 11 ($`a_2=0.9`$), and 12 (exterior limit $`a_2=1`$). In all figures at low energies (top left) there is not any superluminal velocity and all the velocities vanishes at $`a_1=a_{1\mathrm{m}\mathrm{a}\mathrm{x}}=E`$ except for $`v_{\mathrm{spec}}`$. At $`E=a_2`$, all the velocities vanish at $`a_1=a_{1\mathrm{m}\mathrm{a}\mathrm{x}}=E=a_2`$, and these minima begin to increase for higher energies; moreover, $`v_{\mathrm{kin}}`$ and $`v_{\mathrm{spec}}`$ have the same minimum. At high energies (bottom right), $`v_{\mathrm{kin}}`$ and $`v_{\mathrm{spec}}`$ tends to $`1`$, $`v_{\mathrm{Fermi}}`$ tends to $`\frac{a_1}{a_2^2}`$, and $`v_{\mathrm{ast}}`$ tends to $`\frac{1}{a_2^2}`$.
## Acknowledgments
I would like to thank Ettore Minguzzi, Pedro Sancho, Vicente Miquel, David Klein and the referees of the journal Communications in Mathematical Physics for their valuable help and comments.
|
warning/0506/math0506212.html
|
ar5iv
|
text
|
# Energy of Twisted Harmonic Maps of Riemann Surfaces
## Introduction
Let $`S`$ a closed orientable smooth surface with $`\chi (S)<0`$ and $`G`$ a Lie group. This paper discusses an analytic invariant of a representation $`\pi _(S)\stackrel{๐}{}G`$, and applies to the action of the mapping class group $`\pi _0\left(\mathrm{๐ฃ๐๐ฟ๐ฟ}(S)\right)`$ of $`S`$ on the space of representations $`Hom(\pi _1(S),G)/G`$.
We assume $`G`$ is a reductive real algebraic group with maximal compact subgroup $`K`$ and symmetric space $`X=G/K`$. Suppose that $`\rho `$ is reductive, that is, a representation whose image is Zariski dense in a reductive subgroup of $`G`$. Then according to Corlette , for every conformal structure $`\sigma `$ on $`S`$, there is a $`\rho `$-equivariant harmonic map
$$\stackrel{~}{S}\stackrel{f_{\rho ,\sigma }}{}X,$$
which is unique up to isometries of $`X`$. (Such an equivariant harmonic map is called a twisted harmonic map.) In particular its energy
$$E_\rho (\sigma )$$
is well-defined. Letting $`\sigma `$ vary over Teichmรผller space $`๐ฏ_S`$ defines a function
$$๐ฏ_S\stackrel{E_\rho }{}.$$
The starting point of our paper is the following result:
###### Theorem A.
Suppose that $`\rho `$ is convex cocompact. Then $`E_\rho `$ is a proper function on $`๐ฏ_S`$.
Recall that a discrete subgroup $`\mathrm{\Gamma }G`$ is convex cocompact if there exists a $`\mathrm{\Gamma }`$-invariant closed geodesically convex subset $`NX`$ such that $`N/\mathrm{\Gamma }`$ is compact. A representation $`\rho `$ is convex cocompact if $`\rho `$ is an isomorphism of $`\pi _1(S)`$ onto a convex cocompact discrete subgroup of $`G`$.
ยฟFrom this theorem follows the example which motivated this study. Let $``$ be the subset of $`Hom(\pi _1(S),G)/G`$ consisting of equivalence classes of convex cocompact representations.
###### Corollary B.
$`\pi _0\left(\mathrm{๐ฃ๐๐ฟ๐ฟ}(S)\right)`$ acts properly on $``$.
When $`G=\mathrm{๐ฏ๐ฒ๐ซ}(2,)`$, a convex cocompact representation is quasi-Fuchsian, that is a discrete embedding whose action on $`S2=๐ง^3`$ is topologically conjugate to the action of a discrete subgroup of $`\mathrm{๐ฏ๐ฒ๐ซ}(2,)`$. The corollary is just the known fact that $`\pi _0\left(\mathrm{๐ฃ๐๐ฟ๐ฟ}(S)\right)`$ acts properly on the space $`๐ฌ_S`$ of quasi-Fuchsian embeddings. Bersโs simultaneous uniformization theorem provides a $`\pi _0\left(\mathrm{๐ฃ๐๐ฟ๐ฟ}(S)\right)`$-equivariant homeomorphism
$$๐ฌ_S๐ฏ_S\times \overline{๐ฏ}_S.$$
Properness of the action of $`\pi _0\left(\mathrm{๐ฃ๐๐ฟ๐ฟ}(S)\right)`$ on $`๐ฏ_S`$ implies properness on on $`๐ฌ_S`$.
The basic idea goes back to work of Sacks-Uhlenbeck and Schoen-Yau . When $`\rho `$ is a Fuchsian representation (corresponding to a hyperbolic structure on $`S`$), Tromba proved that $`E_\rho `$ is proper and has a unique critical point (necessarily a minimum). When $`\rho `$ is a quasi-Fuchsian $`\mathrm{๐ฏ๐ฒ๐ซ}(2,)`$-representation, $`E_\rho `$ is proper. Uhlenbeck gave an explicit criterion for when $`E_\rho `$ has a unique minimum. Generally $`E_\rho `$ admits more than one critical point, for quasi-Fuchsian $`\rho `$. This follows from the existence of quasi-Fuchsian hyperbolic 3-manifolds containing arbitrarily many minimal surfaces, as constructed by Joel Hass and Bill Thurston (unpublished).
However, as first shown by Kleiner and Leeb (see also Quint ), convex cocompactness is highly restrictive, only interesting when $`G`$ has $``$-rank one. A more general condition guaranteeing properness of the action of $`\pi _0\left(\mathrm{๐ฃ๐๐ฟ๐ฟ}(S)\right)`$ is given by the notion of Anosov representations introduced by Labourie .
We generalize these results in two directions. First, we extend the results on isometric actions of surface groups to isometric actions on non-positively curved metric spaces as developed by Korevaar-Schoen . Second, following a suggestion of Bruce Kleiner, we consider embeddings of surface groups onto normal subgroups of a convex cocompact group $`\mathrm{\Gamma }`$ of isometries of an NPC space. The quotient group $`Q=\mathrm{\Gamma }/\rho (\pi _1(S))`$ acts on $`๐ฏ_S`$. Since $`E_\rho `$ is $`Q`$-invariant, it induces a function $`E_\rho ^{}`$ on $`๐ฏ_S/Q`$ and we show:
###### Theorem C.
The mapping
$$๐ฏ_S/Q\stackrel{E_\rho ^{}}{}$$
is proper.
This generalization was motivated by hyperbolic 3-manifolds fibering over the circle. The hyperbolic 3-manifold determines a representation $`\rho `$ of the fundamental group of the fiber surface $`S`$. Furthermore the monodromy of the fibration determines an automorphism $`\varphi `$ of $`\pi _1(S)`$ such that $`\rho `$ is conjugate to $`\rho \varphi `$. According to Thurston (see also Otal ), $`\varphi `$ is a pseudo-Anosov or hyperbolic mapping class, and generates a proper $``$-action on $`๐ฏ_S`$. In particular every orbit is an infinite discrete subset of $`๐ฏ_S`$. Since $`E_\rho `$ is $`\varphi `$-invariant and constant on each infinite discrete orbit, $`E_\rho `$ is not proper. Kleiner observed that $`E_\rho `$ induces a proper map on the cyclic quotient $`๐ฏ_S/\varphi `$.
Properness of the energy function fails for surface group representations containing โaccidental parabolicsโ. Such representations are discrete embeddings mapping some nontrivial simple loop $`c`$ to a parabolic isometry. One can find a sequence of bounded energy mappings for which the conformal structures $`\sigma `$ degenerate as to shorten $`c`$, contradicting properness. Using the theory of geometric tameness developed by Thurston and Bonahon (see the recent papers of Agol , Calegari-Gabai and Choi ), we obtain a sharp converse to Theorem A for discrete embeddings into $`\mathrm{๐ฏ๐ฒ๐ซ}(2,)`$:
###### Theorem D.
Let $`\rho :\pi _1(S)\mathrm{๐ฏ๐ฒ๐ซ}(2,)`$ be a discrete embedding. Then $`E_\rho `$ is proper if and only if $`\rho `$ is convex cocompact (that is, quasi-Fuchsian).
## Acknowledgments
This paper grew out of conversations with many mathematicians over a period of several years. In particular we thank Ian Agol, Francis Bonahon, Dick Canary, David Dumas, Cliff Earle, Joel Hass, Misha Kapovich, Bruce Kleiner, Franรงois Labourie, John Loftin, Yair Minsky, Rick Schoen, Bill Thurston, Domingo Toledo, Karen Uhlenbeck, Mike Wolf, and Scott Wolpert for helpful discussions. We also thank the referee for several helpful suggestions.
## Notation and Terminology
If $`X`$ is a metric space, we denote the distance function by $`d_X`$. If $`c`$ is a curve in a length space $`X`$, we denote its length by $`L_X(c)`$. We denote by $`[a]`$ the equivalence class of $`a`$, in various contexts. We denote the identity transformation by $`๐`$. We shall sometimes implicitly assume a fixed basepoint $`s_0S`$ in discussing the fundamental group $`\pi _1(S)`$ and the corresponding universal covering space $`\stackrel{~}{S}S`$.
Although it is more customary to define the mapping class group by orientation-preserving diffeomorphisms, for our purposes it seems more natural to consider all diffeomorphisms. Orientation-reversing mapping classes induce anti-holomorphic isometries of $`๐ฏ_S`$, which are nonetheless appropriate in our setting.
## 1. Flat bundles and harmonic maps
Let $`S`$ be a closed oriented surface with $`\chi (S)<0`$, and let $`\pi _1(S)`$ be its fundamental group. Let $`(X,d)`$ be a complete nonpositively curved length space (an *NPC space*) with isometry group $`G`$.
Choose a universal covering space $`\stackrel{~}{S}S`$ with group of deck transformations $`\pi _1(S)`$. An isometric action of $`\pi _1(S)`$ on $`X`$ is a homomorphism $`\pi _1(S)\stackrel{๐}{}G`$, where $`G`$ is the isometry group of $`X`$. Such a homomorphism defines a flat $`(G,X)`$-bundle $`X_\rho `$ over $`S`$, whose total space is the quotient $`\stackrel{~}{S}\times X`$ by the (diagonal) $`\pi `$-action by deck transformations on $`\stackrel{~}{S}`$ and by $`\rho `$ on $`X`$. Since every flat bundle over a simply connected space is trivial, a section over the universal covering space $`\stackrel{~}{S}`$ is the graph of a mapping $`\stackrel{~}{S}X`$. Sections of $`X_\rho `$ correspond to $`\rho `$-equivariant mappings
$$\stackrel{~}{S}\stackrel{๐ข}{}X.$$
Since $`X`$ is contractible (see, for example, Bridson-Haefliger ), sections always exist.
An important case (and the only one treated in this paper) occurs when $`\rho `$ is a discrete embedding (otherwise known as a discrete faithful representation). Then $`\rho `$ maps $`\pi _1(S)`$ isomorphically onto a discrete subgroup $`\mathrm{\Gamma }G`$ and determines a properly discontinouous free isometric action of $`\pi _1(S)`$ on $`X`$. The quotient $`X/\mathrm{\Gamma }`$ is a NPC space locally isometric to $`X`$ with fundamental group $`\mathrm{\Gamma }\pi _1(S)`$. Indeed, the representation $`\rho `$ defines a preferred isomorphism of $`\pi _1(S)`$ with $`\pi _1(X_\rho )`$, that is, a preferred homotopy class of homotopy equivalences $`SX/\mathrm{\Gamma }`$. Sections of the flat $`(G,X)`$-bundle $`X_\rho `$ correspond to maps in this homotopy class.
For us, a conformal structure on $`S`$ will be an almost complex structure $`\sigma `$ on $`S`$, that is, an automorphism of the tangent bundle $`TS`$ satisfying $`\sigma 2=๐`$. A Riemannian metric $`g`$ is in the conformal class of $`\sigma `$ if and only if
$$g(\sigma v_1,\sigma v_2)=g(v_1,v_2)$$
for tangent vectors $`v_1,v_2`$.
Choose a conformal structure $`\sigma `$ on $`S`$. Let $`\stackrel{~}{S}\stackrel{๐}{}X`$ be a continuously differentiable $`\rho `$-equivariant mapping. Its differential defines a continuous section $`df`$ of the vector bundle $`T^{}\stackrel{~}{S}f^{}TX`$, Choose a Riemannian metric $`g`$ on $`S`$ in the conformal class of $`\sigma `$. Denote by $`\stackrel{~}{g}`$ its pullback to $`\stackrel{~}{S}`$ and $`\stackrel{~}{dA}`$ the corresponding area form on $`\stackrel{~}{S}`$. Let $`_{g,X}`$ denotes the Hilbert-Schmidt norm with respect to the metric on $`\stackrel{~}{S}`$ induced by $`g`$ and the metric on $`X`$. Define energy density of $`f`$ with respect to $`g`$ on $`\stackrel{~}{S}`$ as
$$\stackrel{~}{e}(f)=df_{g,X}^2dA,$$
The energy density $`\stackrel{~}{e}(f)`$ is a $`\pi _1(S)`$-invariant exterior 2-form on $`\stackrel{~}{S}`$ and hence defines an exterior 2-form, the energy density $`e(f)`$ on $`S`$. The energy $`E_{\rho ,g}(f)`$ is the integral
$$E_{\rho ,g}(f)=_Se(f).$$
Alternatively, $`E_{\rho ,g}(f)`$ is the integral of $`\stackrel{~}{e}(f)`$ on $`\stackrel{~}{S}`$ over a fundamental domain for the $`\pi _1(S)`$-action on $`\stackrel{~}{S}`$. Since $`S`$ is two-dimensional, $`E_{\rho ,g}(f)`$ depends only on the conformal structure $`\sigma `$, and we denote it $`E_{\rho ,\sigma }(f)`$.
When the target $`X`$ is only a metric space, define the energy density via
$$\stackrel{~}{e}(f)=\underset{\epsilon 0}{lim}_{d_{\stackrel{~}{S}}(x,y)=\epsilon }\frac{d_X^2(f(x),f(y))}{\epsilon ^2}\frac{ds(y)}{2\pi \epsilon }.$$
(See Korevaar-Schoen or Jost .) For finite energy maps the energy density $`e(f)`$ is a well-defined measure which is absolutely continuous with respect to Lebesgue measure. The Radon-Nikodym derivative plays the role of $`df^2`$. For more details, see Korevaar-Schoen . Finite energy maps always exist. Furthermore, energy minimizing sequences of uniformly Lipschitz equivariant mappings exist (, Theorem 2.6.4). In addition to providing a definition of energy minimizing maps to metric spaces, their construction defines a Sobolev completion of the continuously differentiable maps to Riemannian targets which does not appeal to an isometric embedding of $`X`$ into euclidean space.
In many cases the infimum of the energy is realized. In the context of NPC targets, recall that a map $`f`$ is called harmonic if it minimizes $`E_{\rho ,\sigma }`$ among all $`\rho `$-equivariant maps of finite energy.
The fundamental existence theorem for harmonic maps to nonpositively curved Riemannian manifolds is due to Eells-Sampson . In the twisted (that is, equivariant) setting there are various conditions on $`\rho `$ which guarantee existence. When $`X`$ is a symmetric space of noncompact type, $`\rho `$ is said to be *reductive* if its Zariski closure has trivial unipotent radical. Existence of a twisted harmonic map for reductive $`\rho `$ was proven by Corlette , Donaldson , Labourie and Jost-Yau . A geometric notion of reductivity involving stabilizers of flat totally geodesic subspaces was used in (see also Jost ). Korevaar and Schoen introduced the notion of a *proper action* (not to be confused with the more standard use of the term *proper* below) which is the condition that the sublevel sets of the displacement function associated to a generating set of $`\pi _1(S)`$ are bounded. This condition guarantees the existence of an energy minimizer when $`X`$ is a general NPC space (see also ).
## 2. Bounded geometry
Let $`\gamma G`$. Its translation length $`|\gamma |`$ is defined by:
(2.1)
$$|\gamma |:=\underset{xX}{inf}d(x,\gamma x).$$
###### Lemma 2.1.
Let $`\mathrm{\Gamma }G`$ be a convex cocompact discrete subgroup. Then $`\epsilon _0>0`$ such that $`|\gamma |\epsilon _0`$ for all $`\gamma \mathrm{\Gamma }\{๐\}`$.
###### Proof.
Suppose not. Then $`\gamma _i\mathrm{\Gamma }`$ such that $`|\gamma _i|0`$ for all $`i`$, and $`|\gamma _i|0`$. Let $`N`$ be a closed convex $`\mathrm{\Gamma }`$-invariant subset such that $`N/\mathrm{\Gamma }`$ is compact. Since $`N`$ is convex and $`\mathrm{\Gamma }`$-invariant, $`x_iN`$ such that
$$d(x_i,\gamma _ix_i)0.$$
Since $`N/\mathrm{\Gamma }`$ is compact, $`\lambda _i\mathrm{\Gamma }`$ and $`xN`$ such that, after passing to a subsequence, $`\lambda _ix_ix`$. Set $`\stackrel{~}{\gamma }_i=\lambda _i\gamma _i\lambda _i^1`$. Then
$$d(\lambda _ix_i,\stackrel{~}{\gamma }_i\lambda x_i)0.$$
Properness of the action of $`\mathrm{\Gamma }`$ near $`xN`$ implies that for only finitely many $`i`$ does $`|\stackrel{~}{\gamma }_i|=|\gamma _i|`$ This contradicts the assumption that
$$0|\gamma _i|0.$$
###### Lemma 2.2.
Let $`\epsilon _0`$ satisfy Lemma 2.1. Let $`\gamma _1,\gamma _2\mathrm{\Gamma }`$ and $`x,yX`$. If
* $`d(x,y)<\epsilon _0/2`$;
* $`d(\gamma _1x,\gamma _2y)<\epsilon _0/2`$,
then $`\gamma _1=\gamma _2`$.
###### Proof.
$`|\gamma _2^1\gamma _1|`$ $`d(\gamma _2^1\gamma _1x,x)`$
$`=d(\gamma _1x,\gamma _2x)`$
$`d(\gamma _1x,\gamma _2y)+d(\gamma _2y,\gamma _2x)`$
$`=d(\gamma _1x,\gamma _2y)+d(x,y)<\epsilon _0.`$
Now apply Lemma 2.1. โ
## 3. Existence of harmonic maps
###### Proposition 3.1.
Suppose that $`\pi _1(S)\stackrel{๐}{}G`$ is convex cocompact. Then there exists a $`\rho `$-equivariant harmonic map $`\stackrel{~}{S}\stackrel{๐ข}{}X`$.
We deduce this proposition as an immediate corollary of the following more general proposition, which we state here for later applications.
###### Proposition 3.2.
Suppose that $`\pi _1(S)\stackrel{๐}{}G`$ is an embedding onto a normal subgroup of a convex cocompact subgroup $`\mathrm{\Gamma }Iso(X)`$ such that $`\rho (\pi _1(S))`$ has trivial centralizer in $`\mathrm{\Gamma }`$. Then there exists a $`\rho `$-equivariant harmonic map $`\stackrel{~}{S}\stackrel{๐ข}{}X`$.
###### Proof.
For any NPC space $`X`$ and compact surface $`S`$, there exists an energy minimizing sequence $`u_i`$ of uniformly Lipschitz $`\rho `$-equivariant mappings $`\stackrel{~}{S}X`$ (Korevaar-Schoen \[27, Theorem 2.6.4\]). Let $`NX`$ be a $`\rho `$-invariant convex set such that $`N/\mathrm{\Gamma }`$ is compact. Projection $`XN`$ decreases distances, and therefore decreases energy. Thus we may assume that the image of $`u_i`$ lies in $`N`$.
Fix any point $`\stackrel{~}{s_0}\stackrel{~}{S}`$ with image $`\stackrel{~}{s}S`$. Since $`N/\mathrm{\Gamma }`$ is compact, after passing to a subsequence, $`\gamma _i\mathrm{\Gamma }`$ such that $`v_i(\stackrel{~}{s_0})`$ converges to a point in $`N`$, where
$$v_i:=\rho (\gamma _i)u_i.$$
The $`v_i`$ are uniformly Lipschitz and $`v_i(\stackrel{~}{s_0})`$ converges. The Arzรฉla-Ascoli theorem implies that a subsequence of $`v_i`$ converges uniformly on compact subsets of $`\stackrel{~}{S}`$. Choose $`\epsilon _0>0`$ satisfying Lemma 2.1. For each compact $`K\stackrel{~}{S}`$, there exists $`I>0`$ so that
(3.1)
$$d(v_i(w),v_j(w))<\epsilon _0/2$$
whenever $`i,jI`$ and $`wK`$.
Each $`v_i`$ is equivariant with respect to $`\rho _i=\rho \mathrm{๐จ๐๐}_{\gamma _i}`$, where $`\mathrm{๐จ๐๐}_{\gamma _i}`$ denotes the inner automorphism of $`\pi _1(S)`$ defined by $`\gamma _i`$. Fix $`i,jI`$, and set $`x=v_i(\stackrel{~}{s_0})`$ and $`y=v_j(\stackrel{~}{s_0})`$.
Choose a finite generating set $`\mathrm{\Pi }\pi _1(S)`$. Applying (3.1) to the finite set $`K=\mathrm{\Pi }\stackrel{~}{s_0}`$,
$$d(\rho _i(c)x,\rho _j(c)y)=d(v_i(c\stackrel{~}{s_0}),v_j(c\stackrel{~}{s_0}))<\epsilon _0/2$$
whenever $`c\mathrm{\Pi }`$. Since
$$d(x,y)=d(v_i(\stackrel{~}{s_0}),v_j(\stackrel{~}{s_0}))<\epsilon _0/2,$$
Lemma 2.2 implies $`\rho _i(c)=\rho _j(c)`$ for all $`c\mathrm{\Pi }`$. As $`\mathrm{\Pi }`$ generates $`\pi _1(S)`$ it follows $`\rho _i=\rho _j`$ if $`i,jI`$. Since $`\rho `$ is injective and the centralizer of $`\pi _1(S)`$ in $`\mathrm{\Gamma }`$ is trivial, $`c_i=c_j`$ for all $`i,jI`$. Therefore $`u_i`$ itself converges locally uniformly to the desired minimizer. โ
## 4. The action of $`\mathrm{๐ฃ๐๐ฟ๐ฟ}(S)`$ on $`๐ฏ_S`$
For later use, as well as a perspective on the theme of this paper, we summarize in this section general facts on the action of the diffeomorphism group on the space of metrics. A good general reference for this material is Trombaโs book .
Denote by $`\mathrm{๐ฃ๐๐ฟ๐ฟ}(S)`$ the group of smooth diffeomorphisms of $`S`$ with the $`C^{\mathrm{}}`$ topology. Let $`\mathrm{๐ฃ๐๐ฟ๐ฟ}^0(S)`$ denote the identity component of $`\mathrm{๐ฃ๐๐ฟ๐ฟ}(S)`$, that is, the group of all diffeomorphisms isotopic to the identity. The mapping class group of $`S`$ is the quotient
$$\pi _0\left(\mathrm{๐ฃ๐๐ฟ๐ฟ}(S)\right)=\mathrm{๐ฃ๐๐ฟ๐ฟ}(S)/\mathrm{๐ฃ๐๐ฟ๐ฟ}^0(S).$$
The mapping class group relates to $`\pi _1(S)`$ as follows. Let $`s_0S`$ be a fixed basepoint. A diffeomorphism $`\varphi `$ determines an automorphism of the fundamental group $`\pi _1(S,s_0)`$ if $`\varphi (s_0)=s_0`$. Let $`\varphi \mathrm{๐ฃ๐๐ฟ๐ฟ}(S)`$. Although $`\varphi `$ may not fix $`s_0`$, it is isotopic to one which fixes $`s_0`$, which we call $`\varphi _1`$. This isotopy describes a path $`q_1`$ from $`\varphi (s_0)`$ to $`s_0`$.. Suppose $`\varphi _2`$ is another diffeomorphism isotopic to $`\varphi `$ which fixes $`s_0`$, with corresponding path $`q_2`$ from $`\varphi (s_0)`$ to $`s_0`$. Then the automorphisms of $`\pi _1(S,s_0)`$ induced by $`\varphi _1`$ and $`\varphi _2`$ differ by the inner automorphism $`\mathrm{๐จ๐๐}_\gamma `$ where $`\gamma \pi _1(S,s_0)`$ is the homotopy class of the based loop $`q_1(q_2)^1`$ in $`S`$. There results a homomorphism
$$\pi _0(\mathrm{๐ฃ๐๐ฟ๐ฟ}(S))\mathrm{๐ฎ๐๐}\left(\pi _1(S)\right)$$
where
$$\mathrm{๐ฎ๐๐}\left(\pi _1(S)\right):=\mathrm{๐ ๐๐}\left(\pi _1(S)\right)/\mathrm{๐จ๐๐}\left(\pi _1(S)\right)$$
is the quotient of $`\mathrm{๐ ๐๐}\left(\pi _1(S)\right)`$ by its normal subgroup of inner automorphisms.
###### Theorem 4.1 (Dehn-Nielsen).
The homomorphism
$$\pi _0(\mathrm{๐ฃ๐๐ฟ๐ฟ}(S))\mathrm{๐ฎ๐๐}\left(\pi _1(S)\right)$$
is an isomorphism.
We shall henceforth pass freely between these two approaches of the mapping class group. This was first proved by Nielsen and Dehn (unpublished). For proof and discussion, see Stillwell ) and Farb-Margalit ).
Denote by $`\mathrm{๐ฌ๐พ๐}(S)`$ the space of smooth Riemannian metrics on $`S`$ with the $`C^{\mathrm{}}`$ topology. For any smooth manifold $`S`$, the natural action of $`\mathrm{๐ฃ๐๐ฟ๐ฟ}(S)`$ on $`\mathrm{๐ฌ๐พ๐}(S)`$ is proper (Ebin and Palais (unpublished) in general, and Earle-Eels in dimension $`2`$). In particular its restriction to the subspace $`\mathrm{๐ฌ๐พ๐}_1(S)`$ of metrics of curvature $`1`$ is also proper.
Then $`\mathrm{๐ฃ๐๐ฟ๐ฟ}^0(S)`$ acts properly on $`\mathrm{๐ฌ๐พ๐}_1(S)`$. The quotient, comprising isotopy classes of hyperbolic structures on $`S`$, identifies with the Teichmรผller space of $`S`$
$$\mathrm{๐ฌ๐พ๐}_1(S)/\mathrm{๐ฃ๐๐ฟ๐ฟ}^0(S)๐ฏ_S$$
and inherits an action of the mapping class group The properness of the action of $`\mathrm{๐ฃ๐๐ฟ๐ฟ}(S)`$ on $`\mathrm{๐ฌ๐พ๐}_1(S)`$ implies the following basic fact:
###### Theorem 4.2.
$`\pi _0\left(\mathrm{๐ฃ๐๐ฟ๐ฟ}(S)\right)`$ acts properly on $`๐ฏ_S`$.
Closely related is the existence of a $`\mathrm{๐ฃ๐๐ฟ๐ฟ}(S)`$-invariant Riemannian metric (in the Frรฉchet sense) on $`\mathrm{๐ฌ๐พ๐}(S)`$. This induces the $`\pi _0\left(\mathrm{๐ฃ๐๐ฟ๐ฟ}(S)\right)`$-invariant Weil-Petersson metric on $`๐ฏ_S`$. This metric is incomplete, but complete metrics (for example the Finslerian Teichmรผller metric) exist which are $`\pi _0\left(\mathrm{๐ฃ๐๐ฟ๐ฟ}(S)\right)`$-invariant. For later applications, all we need is some $`\pi _0\left(\mathrm{๐ฃ๐๐ฟ๐ฟ}(S)\right)`$-invariant metric $`d_๐ฏ`$ on $`๐ฏ_S`$. (For a survey of invariant metrics on $`๐ฏ_S`$ see Wolpertโs paper in this volume.)
Theorem 4.2 is commonly attributed to Fricke. The customary proof uses a different set of ideas, more directly related to representations of the fundamental group. We briefly digress to sketch these ideas.
The uniformization theorem identifies $`๐ฏ_S`$ with a component of the space of conjugacy classes of discrete embeddings $`\pi _1(S)\mathrm{๐ฒ๐ซ}(2,)`$. Such a representation is determined up to conjugacy by its character
$`\pi _1(S)`$ $`\stackrel{\chi _\rho }{}`$
$`c`$ $`Tr\rho (c).`$
Geometrically $`\chi _\rho `$ corresponds to the marked length spectrum $`\mathrm{}_\rho `$ which associates to a free homotopy class of oriented loops in $`S`$ the length of the closed geodesic on $`๐ง^2/\rho (\pi _1(S))`$ in that homotopy class. Homotopy classes of oriented loops in $`S`$ correspond to conjugacy classes in $`\pi _1(S)`$). Denote this set of conjugacy classes by $`\widehat{\pi _1(S)}`$. The key point is that the marked length spectrum
$$\widehat{\pi _1(S)}\stackrel{\mathrm{}_\rho }{}_+$$
is finite-to-one (a proper map, where $`\widehat{\pi _1(S)}`$ is discretely topologized).
Choose a $`\pi _0\left(\mathrm{๐ฃ๐๐ฟ๐ฟ}(S)\right)`$-invariant metric $`d_๐ฏ`$ on $`๐ฏ_S`$. An isometric action on a locally compact metric space is proper if and only if some (and hence every) orbit is discrete. Therefore it suffices to prove that every $`\pi _0\left(\mathrm{๐ฃ๐๐ฟ๐ฟ}(S)\right)`$-orbit is discrete. Suppose that $`\varphi _n\mathrm{๐ ๐๐}\left(\pi _1(S)\right)`$ is a sequence of automorphisms and $`\rho `$ is a representation such that its images $`\rho \varphi _n`$ converge to a representation $`\rho _{\mathrm{}}`$. Let $`\mathrm{\Pi }\pi _1(S)`$ be a finite generating set and choose $`C`$ sufficiently large so that
$$\mathrm{}_\rho _{\mathrm{}}(\gamma )C$$
for $`\gamma \mathrm{\Pi }`$. Then
$$A:=\{\gamma \pi _1(S)\mathrm{}_\rho _{\mathrm{}}(\gamma )C\}.$$
is a finite union of conjugacy classes in $`\pi _1(S)`$ containing $`\mathrm{\Pi }`$. Let $`ฯต>0`$. Then $`I`$ such that
$$\mathrm{}_{\rho \varphi _i}(\gamma )C+ฯต$$
for $`iI`$ and $`\gamma A`$. Since
$$\mathrm{}_{\rho \varphi _i}(\gamma )=\mathrm{}_\rho (\varphi _i(\gamma )),$$
the set $`A`$ is invariant under all $`\varphi _i(\varphi _j)^1`$ for $`i,jI`$. ยฟFrom this one can prove that the set of equivalence classes $`[\varphi _i]\mathrm{๐ฎ๐๐}\left(\pi _1(S)\right)`$ for $`iI`$ is finite, so that the sequence $`[\rho \varphi _i]`$ is finite, as desired.
For further details, see Abikoff , ยง2.2, Farb-Margalit , Harvey , ยง2.4.1, Buser , ยง6.5.6 (p.156), Imayoshi-Tanigawa ,ยง6.3, Nag ,ยง2.7, and Bers-Gardiner , Theorem II.
## 5. Properness of the energy function
We now prove that for $`\rho `$ convex cocompact, the function $`E_\rho `$ on $`๐ฏ_S`$ is proper. With little extra effort, we prove a more general theorem (suggested by Bruce Kleiner), concerning homomorphisms
$$\pi _1(S)\stackrel{๐}{}\mathrm{\Gamma }G$$
where $`\mathrm{\Gamma }`$ is convex cocompact and $`\rho (\pi _1(S))`$ is a normal subgroup $`\mathrm{\Gamma }_1\mathrm{\Gamma }`$. Furthermore we assume that the centralizer of $`\mathrm{\Gamma }_1`$ in $`\mathrm{\Gamma }`$ is trivial. Let
$$\mathrm{\Gamma }\stackrel{๐}{}\mathrm{๐ ๐๐}\left(\pi _1(S)\right)$$
be the homomorphism induced by the inclusion $`\mathrm{\Gamma }_1\mathrm{\Gamma }`$ and the isomorphism $`\pi _1(S)\stackrel{๐}{}\mathrm{\Gamma }_1`$. As $`\mathrm{\Gamma }_1`$ has trivial centralizer, $`\psi `$ is injective. Thus $`\psi `$ induces a monomorphism
$$Q\mathrm{๐ฎ๐๐}\left(\pi _1(S)\right)$$
where $`Q:=\mathrm{\Gamma }/\mathrm{\Gamma }_1`$. Hence $`Q`$ acts on $`๐ฏ_S`$ via (4.1). Furthermore $`E_\rho `$ is $`Q`$-invariant and hence induces a map $`๐ฏ_S/Q\stackrel{E_\rho ^{}}{}`$
###### Proposition 5.1.
The map $`๐ฏ_S/Q\stackrel{E_\rho ^{}}{}`$ is proper.
Suppose that $`[\sigma _i]๐ฏ_S`$ is a sequence whose image in $`๐ฏ_S/Q`$ diverges. Suppose further that
$$E_\rho ([\sigma _i])B$$
for some constant $`B>0`$, and all $`i=1,2,\mathrm{}`$.
Our assumption that the images of $`[\sigma _i]`$ diverge in $`๐ฏ_S/Q`$ means the following. Choose any invariant $`\pi _0\left(\mathrm{๐ฃ๐๐ฟ๐ฟ}(S)\right)`$-invariant metric $`d_๐ฏ`$ on $`๐ฏ_S`$. We may assume, for each $`\eta Q`$, that
(5.1)
$$d_๐ฏ(\psi (\eta )[\sigma _i],[\sigma _j])1$$
for $`ij`$.
By the $`\rho `$-equivariant harmonic maps
$$(\stackrel{~}{S},\stackrel{~}{g_i})\stackrel{u_i}{}X$$
have a uniform Lipschitz constant $`K`$ (depending on $`B`$), where $`\stackrel{~}{g_i}`$ denotes the hyperbolic metric on $`\stackrel{~}{S}`$ associated to $`\sigma _i`$. In particular, given a closed curve $`c`$ in $`S`$, choose a lift $`\stackrel{~}{c}\stackrel{~}{S}`$ running from $`\stackrel{~}{s_0}`$ to $`[c]\stackrel{~}{s_0}`$, where $`[c]\pi _1(S;s_0)`$ denotes the deck transformation corresponding to $`c`$. Denote the length of $`c`$ with respect to the metric $`g_i`$ on $`S`$ by $`L_i(c)`$. Then
$`|\rho ([c])|`$ $`d(u_i(\stackrel{~}{s_0}),\rho ([c])u_i(\stackrel{~}{s_0}))`$
$`=d(u_i(\stackrel{~}{s_0}),u_i(\rho ([c])\stackrel{~}{s_0}))`$
$`L_X(u_i(\stackrel{~}{c}))`$
(5.2) $`KL_i(c).`$
Suppose that $`c\mathrm{\Sigma }`$ is any closed essential curve. Since $`\rho `$ is injective, the isometry $`\rho (c)`$ is nontrivial. Let $`\epsilon _0>0`$ satisfy Lemma 2.1. Then (5) implies
$$\mathrm{}_c(\sigma _i)\epsilon _0/K$$
where $`\mathrm{}_c(\sigma )`$ denotes the geodesic length function of $`c`$ with respect to $`\sigma `$, that is, the length of the unique closed geodesic freely homotopic to $`c`$ in the hyperbolic metric corresponding to $`\sigma `$.
Mumfordโs compactness theorem implies that the conformal structures $`[\sigma _i]`$ project to a compact subset of the Riemann moduli space $`๐ฏ_S/\pi _0\left(\mathrm{๐ฃ๐๐ฟ๐ฟ}(S)\right)`$. Thus $`[\phi _i]\pi _0\left(\mathrm{๐ฃ๐๐ฟ๐ฟ}(S)\right)`$ and $`[\sigma _{\mathrm{}}]๐ฏ_S`$ exist such that, after passing to a subsequence,
(5.3)
$$[\phi _i][\sigma _i][\sigma _{\mathrm{}}].$$
As $`\mathrm{๐ฃ๐๐ฟ๐ฟ}(S)`$ acts properly on the set of Riemannian metrics (ยง4), representatives $`g_i\mathrm{๐ฌ๐พ๐}_1(S)`$ and $`\phi _i\mathrm{๐ฃ๐๐ฟ๐ฟ}(S)`$ exist with $`\phi _i(g_i)g_{\mathrm{}}`$, where $`g_{\mathrm{}}`$ denotes the hyperbolic metric associated to $`\sigma _{\mathrm{}}`$. Choose a base point $`\stackrel{~}{s_0}\stackrel{~}{S}`$ with image $`s_0S`$. We may assume that $`\phi _i(s_0)=s_0`$. Let $`\stackrel{~}{\phi _i}\mathrm{๐ฃ๐๐ฟ๐ฟ}(\stackrel{~}{S})`$ be the unique lift of $`\phi _i`$ such that $`\stackrel{~}{\phi _i}(\stackrel{~}{s_0})=\stackrel{~}{s_0}`$.
The map
$$v_i:=u_i\stackrel{~}{\phi _i}^1:\stackrel{~}{S}X$$
is harmonic with respect to the metric $`\varphi _i(g_i)`$ and equivariant with respect to the homomorphism
$$\rho (\phi _i^1):\pi _1(S)G.$$
The maps $`v_i`$ are uniformly Lipschitz with respect to the metric $`\stackrel{~}{g_{\mathrm{}}}`$ on $`\stackrel{~}{S}`$ induced from the metric $`g_{\mathrm{}}`$ on $`S`$. In particular the family $`\{v_i\}`$ is equicontinuous.
Since $`N/\mathrm{\Gamma }`$ is compact, $`\gamma _i\mathrm{\Gamma }`$ such that all $`\gamma _iv_i(\stackrel{~}{s_0})`$ lie in a compact subset of $`X`$.
By the Arzรฉla-Ascoli theorem, a subsequence of
$$w_i:=\gamma _iv_i$$
converges uniformly on compact sets. For $`I`$ sufficiently large,
(5.4)
$$\underset{z\stackrel{~}{S}}{sup}d_X(w_i(z),w_j(z))<\epsilon _0/2$$
for $`i,jI`$.
Each $`v_i=u_i\stackrel{~}{\phi _i}^1`$ is equivariant with respect to $`\rho (\phi _i)_{}^1`$ and is harmonic with respect to $`\phi _i(g_i)`$. Thus each $`w_i=\gamma _iv_i`$ is equivariant with respect to
$$\rho _i:=\rho (\phi _i)_{}^1\psi (\gamma _i)$$
and also harmonic with respect to $`\phi _i(g_i)`$ (since $`\gamma _i`$ is an isometry).
Fix $`i,jI`$, and let $`x=w_i(\stackrel{~}{s_0})`$ and $`y=w_j(\stackrel{~}{s_0})`$. For every $`c\pi _1(S)`$, (5.4) implies
$`d_X(\rho _i(c)x,\rho _j(c)y)`$ $`=d_X(w_i(c\stackrel{~}{s_0}),w_j(c\stackrel{~}{s_0}))`$
$`<\epsilon _0/2.`$
Since $`d_X(x,y)<\epsilon _0/2`$, Lemma 2.2 implies $`\rho _i(c)=\rho _j(c)`$ for all $`c\pi _1(S)`$. Since $`\rho `$ is injective,
$$\psi (\gamma _i)(\phi _i)_{}=\psi (\gamma _j)(\phi _j)_{}$$
Theorem 4.1, implies the natural homomorphism
$$\pi _0\left(\mathrm{๐ฃ๐๐ฟ๐ฟ}(S)\right)\mathrm{๐ฎ๐๐}\left(\pi _1(S)\right)$$
is injective; thus $`\psi (\gamma _i)\phi _i`$ is isotopic to $`\psi (\gamma _j)\phi _j`$ for all $`i,jI`$. Call this common mapping class $`[\phi ]`$. Thus, for $`iI`$,
(5.5)
$$\psi (\gamma _i)(\phi _i)_{}=[\phi ]$$
If $`i,jI`$, then
$`d_๐ฏ(\psi (\gamma _i)^1[\sigma _i],`$ $`\psi (\gamma _j)^1[\sigma _j])`$
$`=d_๐ฏ([\phi ]^1(\phi _i^1)_{}[\sigma _i],[\phi ]^1(\phi _j^1)_{}[\sigma _j])`$
$`=d_๐ฏ((\phi _i^1)_{}[\sigma _i],(\phi _j^1)_{}[\sigma _j])`$
$`0`$
by (5.3), contradicting (5.1). Thus $`E_\rho `$ is proper, as claimed.
## 6. Action of the mapping class group
Corollary B follows from the properness of the action of $`\pi _0\left(\mathrm{๐ฃ๐๐ฟ๐ฟ}(S)\right)`$ on $`๐ฏ_S`$ and a general fact on proper actions on metric spaces. Let $`X`$ be a metric space and let $`๐ฆ(X)`$ denote the space of compact subsets of $`X`$, with the Hausdorff metric.
###### Lemma 6.1.
A group $`\mathrm{\Gamma }`$ of homeomorphisms of $`X`$ acts properly on $`X`$ if and only if $`\mathrm{\Gamma }`$ acts properly on $`๐ฆ(X)`$.
###### Proof.
The mapping
$`X`$ $`\stackrel{๐}{}๐ฆ(X)`$
$`x`$ $`\{x\}`$
is a proper isometric $`\mathrm{\Gamma }`$-equivariant embedding. If $`\mathrm{\Gamma }`$ acts properly on $`๐ฆ(X)`$, then equivariance implies that $`\mathrm{\Gamma }`$ acts properly on $`X`$.
Conversely, suppose that $`\mathrm{\Gamma }`$ acts properly on $`X`$. For any compact subset $`K๐ฆ(X)`$ of $`๐ฆ(X)`$, its union
$$UK:=\underset{AK}{}A$$
is a compact subset of $`X`$. For $`\gamma \mathrm{\Gamma }`$ the condition
(6.1)
$$\gamma (K)K\mathrm{}$$
implies the condition
(6.2)
$$\gamma (UK)UK\mathrm{}.$$
To show that $`\mathrm{\Gamma }`$ acts properly on $`๐ฆ(X)`$, let $`K๐ฆ(X)`$ be a compact subset. Since $`\mathrm{\Gamma }`$ acts properly on $`X`$, only finitely many $`\gamma \mathrm{\Gamma }`$ satisfy (6.2), and hence only finitely many $`\gamma \mathrm{\Gamma }`$ satisfy (6.1). Thus $`\mathrm{\Gamma }`$ acts properly on $`๐ฆ(X)`$. โ
We now prove Corollary B. Let $`[\rho ]`$. By Theorem A, $`E_\rho `$ is a proper function on $`๐ฏ_S`$, and assumes a minimum $`m_0(E_\rho )`$ Furthermore
$$\mathrm{๐ฌ๐๐}(\rho ):=\{[\sigma ]๐ฏ_SE_\rho (\sigma )=m_0(E_\rho )\}$$
is a compact subset of $`๐ฏ_S`$, and
$$\stackrel{๐ฌ๐๐}{}๐ฆ(๐ฏ_S)$$
is a $`\pi _0\left(\mathrm{๐ฃ๐๐ฟ๐ฟ}(S)\right)`$-equivariant continuous mapping.
###### Conclusion of Proof of Corollary B .
Lemmas 4.2 and 6.1 together imply $`\pi _0\left(\mathrm{๐ฃ๐๐ฟ๐ฟ}(S)\right)`$ acts properly on $`๐ฆ(๐ฏ_S)`$. By equivariance, $`\pi _0\left(\mathrm{๐ฃ๐๐ฟ๐ฟ}(S)\right)`$ acts properly on $``$.โ
## 7. Accidental parabolics
Now we illustrate with a well-known construction how properness of the energy functional can fail if the action contains non-semisimple isometries. For simplicity, assume in this section that $`X`$ is a simply connected nonpositively curved complete Riemannian manifold (a Cartan-Hadamard manifold) and $`G`$ its group of isometries.
###### Theorem 7.1.
Let $`\pi _1(S)\stackrel{๐}{}G`$ be a homomorphism. Assume that for some simple closed curve $`c`$ in $`S`$, there is a complete geodesic
$$\stackrel{๐พ}{}X$$
and constants $`C,\delta >0`$ such that
(7.1)
$$d_X(\gamma (t),\rho [c]\gamma (t))Ce^{\delta t},$$
for all $`t0`$. Then the energy functional $`E_\rho `$ is not proper.
###### Proof.
It suffices to construct a family $`\sigma _t`$, $`0<t1`$, of conformal structures on $`S`$ such that the corresponding points in $`๐ฏ_S`$ diverge as $`t0`$, and a family $`u_t`$ of $`\rho `$-equivariant maps $`\stackrel{~}{S}X`$ such that $`E_{\rho ,\sigma _t}(u_t)`$ is uniformly bounded in $`t`$.
Fix an initial conformal structure $`\sigma _1`$ on $`S`$. Let $`A_\epsilon `$ denote a tubular neighborhood of the geodesic representative of $`c`$ with respect to the hyperbolic metric $`g_1`$ associated to $`\sigma _1`$. We denote this geodesic also by $`c`$. Let $`A_\epsilon ^\pm `$ be the connected components of $`A_\epsilon c`$.
We furthermore choose $`A_\epsilon `$ such that in the uniformization of $`(S,g_1)`$, $`\stackrel{~}{A}_\epsilon ^\pm `$ are isometric to the strip
$$\times [\epsilon ,\frac{1}{\mathrm{}_c(\sigma _1)}),$$
where $`\epsilon _1`$ is some positive number, and $`\mathrm{}_c(g_1)`$ denotes the length of the geodesic. This realizes the isometry $`[c]\pi _1(S)`$ as the isometry
$$(x,y)(x+1,y).$$
Define the family $`\sigma _t`$ of conformal structures by the plumbing construction discussed by Wolpert . The conformal structure on the complement $`S_\epsilon =SA_\epsilon `$ remains fixed whereas the conformal structure on $`A_\epsilon ^\pm `$ is equivalent to the annulus
$$A_t^\pm :=/\times [\epsilon ,1/\mathrm{}_c(\sigma _t))$$
where $`\mathrm{}_c(\sigma _t)0`$ as $`t0`$.
Next, let $`\gamma `$ be the geodesic satisfying (7.1). and let $`W(t)`$ be the quantity on the left-hand-side of (7.1):
$$W(t):=d_X(\gamma (t),\rho [c]\gamma (t)).$$
Geodesically connect points on the geodesic $`\gamma `$ to the points on its image $`\rho [c]\gamma `$ along geodesics as follows. Define
$$\times [0,\mathrm{})\stackrel{๐ผ}{}X$$
so that $`s\alpha (s,t)`$ is the complete unit speed geodesic satisfying
$`\alpha (0,t)`$ $`=\gamma (t)`$
$`\alpha (W(t),t)`$ $`=\rho [c]\gamma (t).`$
Writing
$$L(t)=(1/\delta )\mathrm{log}(t/\epsilon ),$$
notice that $`W(L(t))C\epsilon /t`$. Define
$`[0,1]\times [\epsilon ,\mathrm{})`$ $`\stackrel{๐ฝ}{}X`$
$`(s,t)`$ $`\alpha (W(L(t))s,L(t)).`$
Since
$$_t\beta (0,t)=\frac{\alpha ^{}(L)t}{\delta }=\frac{t}{\delta }.$$
the nonpositive curvature of $`X`$ implies
(7.2)
$$_t\beta (s,t)t/\delta $$
for all $`0s1`$. Also,
$$(_s\alpha )(W(L(t))s,L(t))=1$$
so
$`_s\beta (s,t)`$ $`=W(L(t))(_s\alpha )(W(L(t))s,L(t))`$
(7.3) $`C\epsilon /t`$
Extend $`\beta `$ to $`\times [\epsilon ,\mathrm{})`$ equivariantly with respect to the $``$-action on $`\times [\epsilon ,\mathrm{})`$ and $`\rho (c)`$ on $`X`$. The derivative bounds (7) and (7.2) imply that $`\beta `$ has finite energy as an equivariant map.
Choose a finite energy $`\rho `$-equivariant map $`(\stackrel{~}{S},\sigma _1)\stackrel{๐ข}{}X`$. The energy of its restriction $`u_1^c`$ to $`\stackrel{~}{S}_\epsilon `$ is finite as well. By interpolating near the boundary $`S_\epsilon `$ we may assume that $`u_1`$ restricted to the connected components of $`\stackrel{~}{S}_\epsilon `$ coincides with the geodesic $`s\alpha (s,0)`$. Then for each $`t`$, $`u_1^c`$ extends to a map $`u_t:\stackrel{~}{S}X`$ by requiring
$$u_t|_{\stackrel{~}{A}_t^\pm }=\beta |_{\times [\epsilon ,1/\mathrm{}_c(\sigma _t)]}.$$
Then $`u_t`$ is equivariant and has finite energy with respect to $`\sigma _t`$, uniformly in $`t`$. This completes the proof. โ
## 8. When $`G=\mathrm{๐ฏ๐ฒ๐ซ}(2,)`$
For discrete embeddings in $`G=\mathrm{๐ฏ๐ฒ๐ซ}(2,)`$, R. Canary and Y. Minsky have explained a partial converse to Theorem A. Namely suppose that $`\rho `$ is a discrete embedding of a closed surface group $`\pi _1(S)`$ into $`G`$. Let $`M:=๐ง^3/\rho (\pi _1(S))`$ be the corresponding hyperbolic 3-manifold. We show (Theorem C) that unless $`\rho `$ is quasi-Fuchsian, then $`E_\rho `$ is not proper. Assume that $`\rho `$ is not quasi-Fuchsian. Further assume that $`\rho (\pi )`$ contains no parabolics; otherwise by Theorem 7.1, $`E_\rho `$ is not proper.
Under these assumptions, the work of Thurston and Bonahon guarantees a sequence of pleated surfaces
$$S\stackrel{f_n}{}M$$
which exhaust the ends of the hyperbolic 3-manifold $`M3`$.
The intrinsic geometry of each $`f_n`$ is that of a totally geodesic surface in $`๐ง^3`$ and therefore its energy (computed with respect to the intrinsic hyperbolic metric) equals
$$2\pi \chi (S)=\mathrm{area}(S).$$
Let $`\sigma _n`$ be the conformal structure underlying this intrinsic metric; then
$$E_\rho (\sigma _n)2\pi \chi (S)$$
is bounded.
However, the corresponding sequence $`[\sigma _n]๐ฏ_S`$ tends to $`\mathrm{}`$. It suffices to show that for some $`c\pi _1(S)`$, the geodesic length $`\mathrm{}_c(\sigma _n)`$ is unbounded. Choose a nontrivial element $`c\pi _1(S)`$. Since each pleated surface $`f_n`$ is an isometric map, it suffices to show that the closed geodesics $`c_n`$ on $`f_n`$ become arbitrarily long. Otherwise, $`C`$ such that
$$\mathrm{}_{f_n}(c_n)C.$$
Let $`c`$ denote geodesic in $`M`$ corresponding to $`\rho (c)`$. Since the pleated surfaces $`f_n`$ tend to $`\mathrm{}`$,
$$d(c,f_n)\mathrm{}$$
and in particular the curves $`c_n`$ (each homotopic to $`c`$) become arbitrarily long, as claimed.
Thus, the energy function $`E_\rho `$ for a discrete embedding $`\pi _1(S)\stackrel{๐}{}\mathrm{๐ฒ๐ซ}(2,)`$ is proper if and only if $`\rho `$ is quasi-Fuchsian.
## 9. Speculation
Deformation spaces of flat bundles over a surface $`S`$ are natural geometric objects upon which the mapping class group of $`S`$ acts. When $`G`$ is a compact group, then the action is ergodic (Goldman and Pickrell-Xia ). At the other extreme, uniformization identifies the Teichmรผller space $`๐ฏ_S`$ of $`S`$ with a connected component in the deformation space of flat $`\mathrm{๐ฏ๐ฒ๐ซ}(2,)`$-bundles over $`S`$, and $`\pi _0\left(\mathrm{๐ฃ๐๐ฟ๐ฟ}(S)\right)`$ acts properly on $`๐ฏ_S`$. In general one expects the dynamics of $`\pi _0\left(\mathrm{๐ฃ๐๐ฟ๐ฟ}(S)\right)`$ to intermediate between these two extremes.
As mentioned earlier, convex cocompactness excludes all higher rank examples which do not come from rank one. However it may be possible to replace geodesic convexity of the Riemannian structure by another notion. All that is needed is a compact core $`N/\mathrm{\Gamma }`$ of the locally symmetric space $`X/\mathrm{\Gamma }`$ in which all all harmonic mappings $`SX/\mathrm{\Gamma }`$ take values.
For example, when $`G=\mathrm{๐ฒ๐ซ}(3,)`$, the mapping class group $`\pi _0\left(\mathrm{๐ฃ๐๐ฟ๐ฟ}(S)\right)`$ acts properly on the component of $`Hom(\pi ,G)/G`$ corresponding to convex $`๐ฏ^2`$-structures (Goldman ). Recently using his notion of Anosov representations, Labourie has proved that for any split real form $`G`$, the action of $`\pi _0\left(\mathrm{๐ฃ๐๐ฟ๐ฟ}(S)\right)`$ on the Hitchin-Teichmรผller component of $`Hom(\pi ,G)/G`$ (see Hitchin ) is proper.
Labourieโs definition is as follows. The unit tangent bundle
$$US\stackrel{\Pi }{}S$$
induces a central extension of fundamental groups
$$\pi _1(US)\stackrel{\mathrm{\Pi }_{}}{}\pi $$
where the center $``$ of $`\pi _1(US)`$ corresponds to the fundamental group of the fibers of $`\mathrm{\Pi }`$. A representation $`\rho :\pi G`$ and a linear representation of $`G`$ on a vector space $`V`$ defines a flat vector bundle
$$V_\rho US$$
with holonomy representation $`\rho \mathrm{\Pi }_{}`$. Let $`\stackrel{~}{\xi }_t`$ denote the lift of the vector field on $`US`$ defining the geodesic flow to the total space $`๐_\rho `$. Labourie defines an Anosov structure to be a continuous splitting of the vector bundle
$$V_\rho =V_+V_0V_{}$$
so that vectors in $`V_+`$ (respectively in $`V_{}`$) are exponentially expanded (respectively contracted) under $`\stackrel{~}{\xi }_t`$.
Labourie proves that the mapping class group acts properly on all such representations. All known examples of open sets of representations upon which the mapping class group acts properly satisify Labourieโs condition. The key point is reminiscent of the proof of properness in ยง4: from the representation he constructs a class function $`\pi _1(S)\stackrel{\mathrm{}_\rho }{}_+`$ which is bounded with respect to length function for (any) hyperbolic structure on $`S`$ (or the word metric on $`\pi _1(S)`$.
In another direction, using ideas generalizing those of Bowditch Tan, Wong and Zhang have shown that the action of $`\pi _0\left(\mathrm{๐ฃ๐๐ฟ๐ฟ}(S)\right)`$ on representations satisfying the analogue of Bowditchโs Q-conditions is proper. This also generalizes the properness of the action on the space of quasi-Fuchsian representations.
|
warning/0506/hep-th0506094.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
The dynamical degree of freedom of a probe $`AdS_d`$ brane embedded in an $`AdS_{d+1}`$ target space is the Nambu-Goldstone world volume field $`\varphi (x)`$, with $`x^\mu `$ the $`AdS_d`$ world volume coordinates. $`\varphi (x)`$ describes the co-volume oscillatons of the brane in $`AdS_{d+1}`$ space. Its long wavelength dynamics is given by the reparametrization invariant volume of the brane times the constant brane tension $`\sigma `$ and is encoded in an $`AdS`$ form of the Nambu-Goto action as
$`\mathrm{\Gamma }`$ $`=`$ $`\sigma {\displaystyle d^dxdete}=\sigma {\displaystyle d^dxdet\overline{e}detN}`$ (1.1)
$`=`$ $`\sigma {\displaystyle d^dxdet\overline{e}\mathrm{cosh}^d\sqrt{m^2\varphi ^2}\sqrt{1\frac{๐_m\varphi \eta ^{mn}๐_n\varphi }{\mathrm{cosh}^2\sqrt{m^2\varphi ^2}}}}.`$ (1.2)
The induced vielbein $`e_\mu ^m`$ has a product form, $`e_\mu ^m=\overline{e}_\mu ^nN_n^m`$ where $`\overline{e}_\mu ^m`$ is the static background vielbein for the $`AdS_d`$ world volume of the brane which yields a background world volume Ricci tensor $`\overline{R}_{\mu \nu }=m^2(d1)\overline{g}_{\mu \nu }`$ and hence a background Ricci scalar $`\overline{R}=m^2d(d1)`$ with $`m^2>0`$, while the $`\varphi `$ dependent $`N_n^m`$ is given by
$`N_m^n(x)`$ $`=`$ $`\delta _m^n\mathrm{cosh}(\sqrt{m^2\varphi ^2(x)})`$ (1.5)
$`+[\sqrt{\left(\mathrm{cosh}^2(\sqrt{m^2\varphi ^2(x)})๐_r\varphi (x)\eta ^{rs}๐_s\varphi (x)\right)}`$
$`\mathrm{cosh}(\sqrt{m^2\varphi ^2(x)})]{\displaystyle \frac{๐_m\varphi (x)๐^n\varphi (x)}{(๐\varphi )^2(x)}},`$
where the derivative $`๐_m`$ is defined as $`๐_m=\overline{e}_m^{1\mu }_\mu `$. Expanding the action through terms bilinear in $`\varphi `$ gives
$$\mathrm{\Gamma }=\sigma d^dxdet\overline{e}\left\{1+\frac{1}{2}(m^2d)\varphi ^2\frac{1}{2}_\mu \varphi \overline{g}^{\mu \nu }_\nu \varphi +\mathrm{}\right\}.$$
(1.6)
It is seen that the Nambu-Goldstone boson carries the $`(E,s)=(d,0)`$ representation of $`SO(2,d1)`$. That is, it has mass squared equal to $`m^2d`$ and hence energy $`d`$ in units of $`m^2`$ while being spin zero -. It should be noted that the $`m^2=0`$ case reproduces the massless bosonic brane Nambu-Goto action.
The brane action is invariant under a nonlinear realization of the $`AdS_{d+1}`$ target space global isometry group of transformations $`SO(2,d)`$. In order to have invariance under general coordinate transformations, additional gravitational fields must be introduced. The purpose of this paper is to construct the action of the world volume localized gravitational fields when the brane is embedded in curved space -. In short, the dynamics of the oscillating brane in curved space is described by a world volume localized massless graviton represented by a dynamical veilbein $`e_\mu ^m(x)`$ and a world volume localized vector field represented by a dynamical field $`A_\mu (x)`$. As a consequence of the Higgs mechanism -, the vector field is massive. The action for these fields is derived in a model independent manner using coset methods in which the $`AdS_{d+1}`$ local symmetry group $`SO(2,d)`$ is nonlinearly realized. In section 2, the nonlinear local transformations of the Nambu-Goldstone fields are introduced via the coset method . The locally covariant Maurer-Cartan one-form building blocks for the invariant action are obtained along with the introduction of the dynamical veilbein and vector fields. Derivatives of these Maurer-Cartan world volume vectors that are covariant with respect to local Lorentz and Einstein transformations are defined using the spin and related affine connections. In section 3, these covariant derivatives are used to construct the low energy locally $`SO(2,d)`$ invariant action. Exploiting the spontaneously broken local (pseudo-) translation and Lorentz transformations, the action is transformed to and analyzed in the unitary gauge. The physical degrees of freedom so obtained are the dynamical world volume veilbein and massive vector field.
## 2 The Coset Construction
The embedding of $`AdS_d`$ space spontaneously breaks the symmetry group of the $`AdS_{d+1}`$ space from $`SO(2,d)`$ to $`SO(2,d1)`$. The low energy action governing the dynamics of the Nambu-Goldstone modes associated with the symmetry breakdown can be constructed using coset methods. This technique begins by introducing the coset element $`\mathrm{\Omega }SO(2,d)/SO(1,d1)`$ where $`SO(1,d1)`$ corresponds to the Lorentz structure (stability) group of transformations in $`AdS_d`$. The $`AdS_d=SO(2,d1)/SO(1,d1)`$ coordinates, $`x^\mu `$, act as parameters for pseudo-translations in the world volume and are part of the coset so that
$$\mathrm{\Omega }(x)=e^{ix^\mu P_\mu }e^{i\varphi (x)Z}e^{iv^\mu (x)K_\mu },$$
(2.1)
The $`SO(2,d)`$ generators can be expressed in terms of the unbroken $`SO(1,d1)`$ Lorentz subgroup representation content of the $`SO(2,d1)`$ symmetry group of the brane. The unbroken $`SO(2,d1)`$ symmetry group is generated by the subgroup Lorentz transformation generators $`M_{\mu \nu }`$, where $`\mu ,\nu =0,1,2,\mathrm{},d1`$ and the pseudo-translations in $`AdS_d`$ space with charges $`P_\mu `$. The remaining charges are the generating elements of the $`SO(2,d)/SO(2,d1)`$ coset. They are the broken $`SO(2,d)`$ symmetry transformation charges. $`Z`$ generates the broken $`SO(2,d)`$ pseudo-translations in the co-volume direction normal to the brane, while $`K_\mu `$ generates the broken $`AdS_{d+1}`$ Lorentz transformations. Thus the $`SO(2,d)`$ algebra can be written in terms of the $`P_\mu `$, $`M_{\mu \nu }`$, $`Z`$ and $`K_\mu `$ charges as
$`[M_{\mu \nu },M_{\rho \sigma }]`$ $`=`$ $`i\left(\eta _{\mu \rho }M_{\nu \sigma }\eta _{\mu \sigma }M_{\nu \rho }+\eta _{\nu \sigma }M_{\mu \rho }\eta _{\nu \rho }M_{\mu \sigma }\right)`$ (2.2)
$`[M_{\mu \nu },P_\lambda ]`$ $`=`$ $`i\left(P_\mu \eta _{\nu \lambda }P_\nu \eta _{\mu \lambda }\right)`$ (2.3)
$`[M_{\mu \nu },K_\lambda ]`$ $`=`$ $`i\left(K_\mu \eta _{\nu \lambda }K_\nu \eta _{\mu \lambda }\right)`$ (2.4)
$`[M_{\mu \nu },Z]`$ $`=`$ $`0`$ (2.5)
$`[P_\mu ,P_\nu ]`$ $`=`$ $`im^2M_{\mu \nu }`$ (2.6)
$`[K_\mu ,K_\nu ]`$ $`=`$ $`iM_{\mu \nu }`$ (2.7)
$`[P_\mu ,K_\nu ]`$ $`=`$ $`i\eta _{\mu \nu }Z`$ (2.8)
$`[P_\mu ,Z]`$ $`=`$ $`im^2K_\mu `$ (2.9)
$`[Z,K_\mu ]`$ $`=`$ $`iP_\mu .`$ (2.10)
The coset so defined corresponds to a particular choice of coordinates, specifically denoted as $`x^\mu `$, for the $`AdS_d`$ world volume. The fields are also defined as functions of $`x^\mu `$. The Nambu-Goldstone field $`\varphi (x)`$ along with $`v^\mu (x)`$ act as the remaining coordinates needed to parametrize the coset manifold $`SO(2,d)/SO(2,d1)`$.
Left multiplication of the coset elements $`\mathrm{\Omega }`$ by an $`SO(2,d)`$ group element $`g`$ which is specified by local infinitesimal parameters $`ฯต^\mu (x),z(x),b^\mu (x),\lambda ^{\mu \nu }(x)`$ so that
$$g(x)=e^{iฯต^\mu (x)P_\mu }e^{iz(x)Z}e^{ib^\mu (x)K_\mu }e^{\frac{i}{2}\lambda ^{\mu \nu }(x)M_{\mu \nu }},$$
(2.11)
results in transformations of the space-time coordinates and the Nambu-Goldstone fields according to the general form
$$g(x)\mathrm{\Omega }(x)=\mathrm{\Omega }^{}(x^{})h(x).$$
(2.12)
The transformed coset element, $`\mathrm{\Omega }^{}`$, is a function of the transformed world volume coordinates and the total variations of the fields so that
$$\mathrm{\Omega }^{}=e^{ix^\mu P_\mu }e^{i\varphi ^{}(x^{})Z}e^{iv^\mu (x^{})K_\mu },$$
(2.13)
while $`h`$ is a field dependent element of the stability group $`SO(1,d1)`$:
$$h=e^{\frac{i}{2}\theta ^{\mu \nu }(x)M_{\mu \nu }}.$$
(2.14)
Exploiting the algebra of the $`SO(2,d)`$ charges displayed in equation (2.10), along with extensive use of the Baker-Campbell-Hausdorf formulae, the local $`SO(2,d)`$ transformations are obtained as
$`x^\mu `$ $`=`$ $`\left[1z(x)\sqrt{m^2}\mathrm{tanh}\sqrt{m^2\varphi ^2}{\displaystyle \frac{\mathrm{sin}\sqrt{4m^4x^2}}{\sqrt{m^2x^2}}}\right]x^\mu \lambda ^{\mu \nu }(x)x_\nu `$ (2.17)
$`+\left[P_L^{\mu \nu }(x)+\sqrt{m^2x^2}\mathrm{cot}\sqrt{m^2x^2}P_T^{\mu \nu }(x)\right]ฯต_\nu (x)`$
$`+{\displaystyle \frac{\mathrm{tanh}\sqrt{m^2\varphi ^2}}{\sqrt{m^2}}}\left[\mathrm{cos}\sqrt{m^2x^2}P_L^{\mu \nu }(x)+{\displaystyle \frac{\sqrt{m^2x^2}}{\mathrm{sin}\sqrt{m^2x^2}}}P_T^{\mu \nu }(x)\right]b_\nu (x)`$
$`\varphi ^{}(x^{})`$ $`=`$ $`\varphi (x)+z(x)\mathrm{cos}\sqrt{m^2x^2}+b_\mu (x)x^\mu {\displaystyle \frac{\mathrm{sin}\sqrt{m^2x^2}}{\sqrt{m^2x^2}}}`$ (2.19)
$`v^\mu (x^{})`$ $`=`$ $`v^\mu (x)\lambda ^{\mu \nu }(x)v_\nu {\displaystyle \frac{m^2}{2}}{\displaystyle \frac{\mathrm{tan}\sqrt{m^2x^2/4}}{\sqrt{m^2x^2/4}}}(ฯต^\mu (x)x^\nu ฯต^\nu (x)x^\mu )v_\nu `$ (2.25)
$`z(x){\displaystyle \frac{m^2}{\mathrm{cosh}\sqrt{m^2\varphi ^2}}}{\displaystyle \frac{\mathrm{sin}\sqrt{m^2x^2}}{\sqrt{m^2x^2}}}\left[P_L^{\mu \nu }(v)+\sqrt{v^2}\mathrm{coth}\sqrt{v^2}P_T^{\mu \nu }(v)\right]x_\nu `$
$`+\sqrt{m^2}{\displaystyle \frac{\mathrm{tan}\sqrt{m^2x^2/4}}{\sqrt{m^2x^2}}}\mathrm{tanh}\sqrt{m^2\varphi ^2}(b^\mu (x)x^\nu b^\nu (x)x^\mu )v_\nu `$
$`+{\displaystyle \frac{1}{\mathrm{cosh}\sqrt{m^2\varphi ^2}}}\left[P_L^{\mu \nu }(v)+\sqrt{v^2}\mathrm{coth}\sqrt{v^2}P_T^{\mu \nu }(v)\right]`$
$`\times \left[\mathrm{cos}\sqrt{m^2x^2}P_{L\nu \rho }(x)+P_{T\nu \rho }(x)\right]b^\rho (x)`$
$`\theta ^{\mu \nu }(x)`$ $`=`$ $`\lambda ^{\mu \nu }(x)+{\displaystyle \frac{m^2}{2}}{\displaystyle \frac{\mathrm{tan}\sqrt{m^2x^2/4}}{\sqrt{m^2x^2/4}}}(ฯต^\mu (x)x^\nu ฯต^\nu (x)x^\mu )`$ (2.31)
$`z(x){\displaystyle \frac{m^2}{\mathrm{cosh}\sqrt{m^2\varphi ^2}}}{\displaystyle \frac{\mathrm{sin}\sqrt{m^2x^2}}{\sqrt{m^2x^2}}}(v^\mu x^\nu v^\nu x^\mu ){\displaystyle \frac{\mathrm{tanh}\sqrt{v^2/2}}{\sqrt{v^2}}}`$
$`\sqrt{m^2}{\displaystyle \frac{\mathrm{tan}\sqrt{m^2x^2/4}}{\sqrt{m^2x^2}}}\mathrm{tanh}\sqrt{m^2\varphi ^2}(b^\mu (x)x^\nu b^\nu (x)x^\mu )`$
$`{\displaystyle \frac{1}{\mathrm{cosh}\sqrt{m^2\varphi ^2}}}{\displaystyle \frac{\mathrm{tanh}\sqrt{v^2/2}}{\sqrt{v^2}}}`$
$`\times [\mathrm{cos}\sqrt{m^2x^2}P_L^{\mu \rho }(x)b_\rho (x)v^\nu +P_T^{\mu \rho }(x)b_\rho (x)v^\nu (\mu \nu )].`$
Here the transverse and longitudinal projectors for $`x^\mu `$ are defined as
$`P_{T\mu \nu }(x)`$ $`=`$ $`\eta _{\mu \nu }{\displaystyle \frac{x_\mu x_\nu }{x^2}}`$ (2.33)
$`P_{L\mu \nu }(x)`$ $`=`$ $`{\displaystyle \frac{x_\mu x_\nu }{x^2}}`$ (2.34)
and $`\eta _{\mu \nu }`$ is the metric tensor for $`d`$โdimensional Minkowski space having signature $`(+1,1,\mathrm{},1)`$. In the above, the indices are raised, lowered and contracted using $`\eta _{\mu \nu }`$. Both Nambu-Goldstone fields $`\varphi `$ and $`v^\mu `$ transform inhomogeneously under the broken local translations $`Z`$ and broken local Lorentz transformations $`K_\mu `$. Thus these broken transformations can be used to transform to the unitary gauge in which both $`\varphi `$ and $`v^\mu `$ vanish. This will be done in section 3 in order to exhibit the physical degrees of freedom in a more transparent fashion.
The nonlinearly realized $`SO(2,d)`$ transformations induce a coordinate and field dependent general coordinate transformation of the world volume space-time coordinates. From the $`x^\mu `$coordinate transformation given above, the $`AdS_{d+1}`$ general coordinate Einstein transformation for the world volume space-time coordinate differentials is given by
$$dx^\mu =dx^\nu G_\nu ^\mu (x),$$
(2.35)
with $`G_\nu ^\mu (x)=x^\mu /x^\nu `$. The $`SO(2,d)`$ invariant interval can be formed using the metric tensor $`g_{\mu \nu }(x)`$ so that $`ds^2=dx^\mu g_{\mu \nu }(x)dx^\nu =ds^2=dx^\mu g_{\mu \nu }^{}(x^{})dx^\nu `$ where the metric tensor transforms as
$$g_{\mu \nu }^{}(x^{})=G_\mu ^{1\rho }(x)g_{\rho \sigma }(x)G_\nu ^{1\sigma }(x).$$
(2.36)
The form of the vielbein (and hence the metric tensor) as well as the locally $`SO(2,d)`$ covariant derivatives of the Nambu-Goldstone boson fields and the spin connection can be extracted from the locally covariant Maurer-Cartan one-form, $`\mathrm{\Omega }^1D\mathrm{\Omega }`$, which can be expanded in terms of the generators as
$`\mathrm{\Omega }^1D\mathrm{\Omega }`$ $``$ $`\mathrm{\Omega }^1(d+i\widehat{E})\mathrm{\Omega }`$ (2.37)
$`=`$ $`i\left[\omega ^mP_m+\omega _ZZ+\omega _K^mK_m+{\displaystyle \frac{1}{2}}\omega _M^{mn}M_{mn}\right].`$ (2.38)
Here Latin indices $`m,n=0,1,\mathrm{},d1`$, are used to distinguish tangent space local Lorentz transformation properties from world volume Einstein transformation properties which are denoted using Greek indices. In what follows Latin indices are raised and lowered with use of the Minkowski metric tensors, $`\eta ^{mn}`$ and $`\eta _{mn}`$, while Greek indices are raised and lowered with use of the curved $`AdS_d`$ metric tensors, $`g^{\mu \nu }`$ and $`g_{\mu \nu }`$. Since the Nambu-Goldstone fields vanish in the unitary gauge it is useful to exhibit the one-form gravitational fields in terms of their pseudo-translated form
$$\widehat{E}=e^{+ix^\mu P_\mu }Ee^{ix^\mu P_\mu }.$$
(2.39)
The world volume one-form gravitational fields $`E`$ have the expansion in terms of the charges as
$$E=E^mP_m+AZ+B^mK_m+\frac{1}{2}\gamma ^{mn}M_{mn}.$$
(2.40)
Similarly expanding $`\widehat{E}`$ as
$$\widehat{E}=\widehat{E}^mP_m+\widehat{A}Z+\widehat{B}^mK_m+\frac{1}{2}\widehat{\gamma }^{mn}M_{mn},$$
(2.41)
one finds the various fields are related according to
$`\widehat{E}`$ $`=`$ $`\left[\mathrm{cos}\sqrt{m^2x^2}P_{Tn}^m(x)+P_{Ln}^m(x)\right]E^n{\displaystyle \frac{\mathrm{sin}\sqrt{m^2x^2}}{\sqrt{m^2x^2}}}\gamma ^{mn}x_n`$ (2.42)
$`\widehat{A}`$ $`=`$ $`A\mathrm{cos}\sqrt{m^2x^2}{\displaystyle \frac{\mathrm{sin}\sqrt{m^2x^2}}{\sqrt{m^2x^2}}}B^mx_m`$ (2.43)
$`\widehat{B^m}`$ $`=`$ $`\left[P_{Tn}^m(x)+\mathrm{cos}\sqrt{m^2x^2}P_{Ln}^m(x)\right]B^n+m^2{\displaystyle \frac{\mathrm{sin}\sqrt{m^2x^2}}{\sqrt{m^2x^2}}}Ax^m`$ (2.44)
$`\widehat{\gamma }^{mn}`$ $`=`$ $`\gamma ^{mn}m^2{\displaystyle \frac{\mathrm{sin}\sqrt{m^2x^2}}{\sqrt{m^2x^2}}}\left(E^mx^nE^nx^m\right)`$ (2.46)
$`\left(\mathrm{cos}\sqrt{m^2x^2}1\right)\left[\gamma ^{ms}P_{Ls}^n(x)\gamma ^{ns}P_{Ls}^m(x)\right].`$
Defining the one-form gravitational fields to transform as a gauge field so that
$$\widehat{E}^{}(x^{})=g(x)\widehat{E}(x)g^1(x)ig(x)dg^1(x),$$
(2.47)
the covariant Maurer-Cartan one-form transforms analogously to the way it varied for global transformations:
$$\omega ^{}(x^{})=h(x)\omega (x)h^1(x)+h(x)dh^1(x),$$
(2.48)
with $`h=e^{\frac{i}{2}\theta ^{mn}(x)M_{mn}}`$. Expanding in terms of the $`SO(2,d)`$ charges, the individual one-forms transform according to their local Lorentz nature
$`\omega ^m(x^{})`$ $`=`$ $`\omega ^n(x)\mathrm{\Lambda }_n^m(\theta (x))`$ (2.49)
$`\omega _Z^{}(x^{})`$ $`=`$ $`\omega _Z(x)`$ (2.50)
$`\omega _K^m(x^{})`$ $`=`$ $`\omega _K^n(x)\mathrm{\Lambda }_n^m(\theta (x))`$ (2.51)
$`\omega _M^{mn}(x^{})`$ $`=`$ $`\omega _M^{rs}(x)\mathrm{\Lambda }_r^m(\theta (x))\mathrm{\Lambda }_s^n(\theta (x))d\theta ^{mn}(x).`$ (2.52)
For infinitesimal transformations, the local Lorentz transformations are $`\mathrm{\Lambda }_n^m(\theta (x))=\delta _n^m+\theta _n^m(x)`$, while the infinitesimal local $`SO(2,d)`$ transformations of the gravitational one-forms take the form
$`\widehat{E}^m`$ $`=`$ $`\widehat{E}^m+\widehat{\gamma }^{mn}ฯต_nz\widehat{B}^m+b^m\widehat{A}\lambda ^{mn}\widehat{E}_ndฯต^m`$ (2.53)
$`\widehat{A}^{}`$ $`=`$ $`\widehat{A}ฯต_m\widehat{B}^m+b_m\widehat{E}^mdz`$ (2.54)
$`\widehat{B}^m`$ $`=`$ $`\widehat{B}^m+ฯต^mm^2\widehat{A}zm^2\widehat{E}^m+\widehat{\gamma }^{mn}b_n\lambda ^{mn}\widehat{B}_ndb^m`$ (2.55)
$`\widehat{\gamma }^{mn}`$ $`=`$ $`\widehat{\gamma }^{mn}+m^2(ฯต^m\widehat{E}^nฯต^n\widehat{E}^m)(b^m\widehat{B}^nb^n\widehat{B}^m)`$ (2.57)
$`+(\lambda ^{mr}\widehat{\gamma }_r^n\lambda ^{nr}\widehat{\gamma }_r^m)d\lambda ^{mn}.`$
Using the Feynman formula for the variation of an exponential operator in conjunction with the Baker-Campell-Hausdorff formulae, the individual world volume one-forms appearing in the above decomposition of the covariant Maurer-Cartan one-form are secured as
$`\omega ^m`$ $`=`$ $`dx^\mu e_\mu ^m`$ (2.58)
$`=`$ $`dx^\mu _\mu ^nN_n^m`$ (2.59)
$`\omega _Z`$ $`=`$ $`dx^\mu \omega _{Z\mu }`$ (2.60)
$`=`$ $`dx^\mu \mathrm{cosh}\sqrt{v^2}_\mu ^m\left[v_m{\displaystyle \frac{\mathrm{tanh}\sqrt{v^2}}{\sqrt{v^2}}}\mathrm{cosh}\sqrt{m^2\varphi ^2}+_m^{1\nu }\left(_\nu \varphi +A_\nu \right)\right]`$ (2.61)
$`\omega _K^m`$ $`=`$ $`dx^\mu \omega _{K\mu }^m`$ (2.62)
$`=`$ $`\left[P_L^{mn}(v)+{\displaystyle \frac{\mathrm{sinh}\sqrt{v^2}}{\sqrt{v^2}}}P_T^{mn}(v)\right]\left(dv_n(\overline{\omega }_{Mnr}+\gamma _{nr})v^r\right)`$ (2.65)
$`+[P_L^{mn}(v)+\mathrm{cosh}\sqrt{v^2}P_T^{mn}(v)][(\overline{\omega }_n+E_n)\sqrt{m^2}\mathrm{sinh}\sqrt{m^2\varphi ^2}`$
$`+B_n\mathrm{cosh}\sqrt{m^2\varphi ^2}]`$
$`\omega _M^{mn}`$ $`=`$ $`dx^\mu \omega _{M\mu }^{mn}`$ (2.66)
$`=`$ $`(\overline{\omega }_M^{mn}+\gamma ^{mn})\mathrm{cosh}\sqrt{m^2\varphi ^2}{\displaystyle \frac{\mathrm{sinh}\sqrt{v^2}}{\sqrt{v^2}}}\left[B^mv^nB^nv^m\right]`$ (2.70)
$`+\sqrt{m^2}{\displaystyle \frac{\mathrm{sinh}\sqrt{v^2}}{\sqrt{v^2}}}\mathrm{sinh}\sqrt{m^2\varphi ^2}\left[v^mP_{Ts}^n(v)v^nP_{Ts}^m(v)\right](\overline{\omega }^s+E^s)`$
$`+[\mathrm{cosh}\sqrt{v^2}1][{\displaystyle \frac{v^mdv^nv^ndv^m}{v^2}}`$
$`(P_{Lr}^m(v)(\overline{\omega }_M^{nr}+\gamma ^{nr})P_{Lr}^n(v)(\overline{\omega }_M^{mr}+\gamma ^{mr}))].`$
In these expressions, the background $`AdS_d`$ covariant coordinate differential, $`\overline{\omega }^m`$, and spin connection, $`\overline{\omega }_M^{mn}`$, are obtained from the $`AdS_d`$ coordinate one-form $`(e^{ix^mP_m})d(e^{ix^nP_n})=i[\overline{\omega }^mP_m+\frac{1}{2}\overline{\omega }_M^{mn}M_{mn}]`$ as
$`\overline{\omega }^m`$ $`=`$ $`{\displaystyle \frac{\mathrm{sin}\sqrt{m^2x^2}}{\sqrt{m^2x^2}}}P_T^{mn}(x)dx_n+P_L^{mn}(x)dx_n`$ (2.72)
$`\overline{\omega }_M^{mn}`$ $`=`$ $`\left[\mathrm{cos}\sqrt{m^2x^2}1\right]{\displaystyle \frac{(x^mdx^nx^ndx^m)}{x^2}}.`$ (2.73)
The background differential $`\overline{\omega }^m`$ is related to the $`x^\mu `$ world volume coordinate differential by the background veilbein $`\overline{e}_\mu ^m(x)`$ as
$$\overline{\omega }^m=dx^\mu \overline{e}_\mu ^m(x).$$
(2.74)
Using equation (2.73) along with $`d=dx^\mu _\mu ^x`$, the background veilbein is obtained as
$$\overline{e}_\mu ^m(x)=\frac{\mathrm{sin}\sqrt{m^2x^2}}{\sqrt{m^2x^2}}P_{T\mu }^m(x)+P_{L\mu }^m(x).$$
(2.75)
The covariant coordinate differential $`\omega ^m`$ is related to the world volume coordinate differential $`dx^\mu `$ by the vielbein $`e_\mu ^m`$ which in turn can be written in a factorized form as the product of the dynamic vielbein $`_\mu ^m`$ and the Nambu-Goto vielbein $`N_n^m`$
$`_\mu ^m`$ $`=`$ $`\overline{e}_\mu ^m+E_\mu ^m+B_\mu ^m{\displaystyle \frac{\mathrm{tanh}\sqrt{m^2\varphi ^2}}{\sqrt{m^2}}}`$ (2.76)
$`N_n^m`$ $`=`$ $`\mathrm{cosh}\sqrt{m^2\varphi ^2}\left[P_{Tn}^m(v)+\mathrm{cosh}\sqrt{v^2}P_{Ln}^m(v)\right]`$ (2.78)
$`\mathrm{cosh}\sqrt{v^2}_n^{1\nu }(_\nu \varphi +A_\nu )v^m{\displaystyle \frac{\mathrm{tanh}\sqrt{v^2}}{\sqrt{v^2}}}.`$
The one-forms and their covariant derivatives are the building blocks of the locally $`SO(2,d)`$ invariant action. Indeed a $`m^{\mathrm{th}}`$-rank contravariant local Lorentz and $`n^{\mathrm{th}}`$-rank covariant Einstein tensor, $`T_{\mu _1\mathrm{}\mu _n}^{m_1\mathrm{}m_m}`$ is defined to transform as
$$T_{\mu _1^{}\mathrm{}\mu _n^{}}^{m_1^{}\mathrm{}m_m^{}}(x^{})=G_{\mu _1^{}}^{1\mu _1}(x)\mathrm{}G_{\mu _n^{}}^{1\mu _n}(x)T_{\mu _1\mathrm{}\mu _n}^{m_1\mathrm{}m_m}(x)\mathrm{\Lambda }_{m_1}^{m_1^{}}(\theta (x))\mathrm{}\mathrm{\Lambda }_{m_m}^{m_m^{}}(\theta (x)).$$
(2.79)
For example, the veilbein transforms as $`e_\mu ^m(x^{})=G_\mu ^{1\nu }(x)e_\nu ^n(x)\mathrm{\Lambda }_n^m(\theta (x))`$. Hence, the veilbein and its inverse can be used to convert local Lorentz indices into world volume indices and vice versa. Since the Minkowski metric, $`\eta _{mn}`$, is invariant under local Lorentz transformations the metric tensor $`g_{\mu \nu }`$
$$g_{\mu \nu }=e_\mu ^m\eta _{mn}e_\nu ^n,$$
(2.80)
is a rank 2 Einstein tensor. It can be used to define covariant Einstein tensors given contravariant ones. Likewise, the Minkowski metric can be used to define covariant local Lorentz tensors given contravariant ones.
Since the Jacobian of the $`x^\mu x^\mu `$ transformation is simply
$`d^dx^{}`$ $`=`$ $`d^dxdetG,`$ (2.81)
it follows that $`d^dx^{}dete^{}(x^{})=d^dxdete(x)`$ since $`det\mathrm{\Lambda }=1`$. Thus an $`SO(2,d)`$ invariant action is constructed as
$$\mathrm{\Gamma }=d^dxdete(x)(x),$$
(2.82)
with the Lagrangian an invariant $`^{}(x^{})=(x)`$. The invariants that make up the Lagrangian can be found by contracting the indices of tensors with the appropriate vielbein, its inverse and the Minkowski metric. For example $`\omega _{Z\mu }g^{\mu \nu }\omega _{Z\nu }`$ is an invariant term used to construct the action.
Besides products of the covariant Maurer-Cartan one-forms, their covariant derivatives can also be used to construct invariant terms of the Lagrangian. The covariant derivative of a general tensor can be defined using the affine and related spin connections. Consider the covariant derivative of the Lorentz tensor $`T^{mn}`$
$$_\rho T^{mn}=_\rho T^{mn}\omega _{M\rho r}^mT^{rn}\omega _{M\rho r}^nT^{mr}.$$
(2.83)
Since the spin connection transforms inhomogeneously according to equation (2.52), the covariant derivative of $`T^{mn}`$ transforms homogeneously again
$$(_\rho T^{mn})^{}=G_\rho ^{1\sigma }(_\sigma T^{rs})\mathrm{\Lambda }_r^m\mathrm{\Lambda }_s^n.$$
(2.84)
Converting the Lorentz index $`n`$ to a world index $`\nu `$ using the vielbein, the covariant derivative for mixed tensors is obtained
$`_\rho T^{m\nu }`$ $``$ $`e_n^{1\nu }_\rho T^{mn}=_\rho T^{m\nu }\omega _{M\rho }^{mr}T_r^\nu +\mathrm{\Gamma }_{\sigma \rho }^\nu T^{m\sigma },`$ (2.85)
where the spin connection $`\omega _{M\rho }^{mn}`$ and $`\mathrm{\Gamma }_{\sigma \rho }^\nu `$ are related according to
$$\mathrm{\Gamma }_{\sigma \rho }^\nu =e_n^{1\nu }_\rho e_\sigma ^ne_n^{1\nu }\omega _{M\rho }^{nr}e_\sigma ^s\eta _{rs}.$$
(2.86)
(Note that this relation as well follows from the requirement that the covariant derivative of the vielbein vanishes, $`_\rho e_\mu ^m=0`$.) Applying the above to the Minkowski metric Lorentz 2-tensor yields the formula relating the affine connection $`\mathrm{\Gamma }_{\mu \nu }^\rho `$ to derivatives of the metric
$`_\rho \eta ^{mn}`$ $`=`$ $`_\rho \eta ^{mn}\omega _{M\rho r}^m\eta ^{rn}\omega _{M\rho r}^n\eta ^{mr}=\omega _{M\rho }^{mn}\omega _{M\rho }^{nm}`$ (2.87)
$`=`$ $`0`$ (2.88)
$`=`$ $`e_\mu ^me_\nu ^n_\rho g^{\mu \nu }`$ (2.89)
$`=`$ $`e_\mu ^me_\nu ^n\left(_\rho g^{\mu \nu }+\mathrm{\Gamma }_{\sigma \rho }^\mu g^{\sigma \nu }+\mathrm{\Gamma }_{\sigma \rho }^\nu g^{\mu \sigma }\right).`$ (2.90)
The solution to this equation yields the affine connection in terms of the derivative of the metric (the space is torsionless, hence the connection is symmetric $`\mathrm{\Gamma }_{\mu \nu }^\rho =\mathrm{\Gamma }_{\nu \mu }^\rho `$)
$$\mathrm{\Gamma }_{\mu \nu }^\rho =\frac{1}{2}g^{\rho \sigma }\left[_\mu g_{\sigma \nu }+_\nu g_{\mu \sigma }_\sigma g_{\mu \nu }\right].$$
(2.91)
Finally a covariant field strength two-form can be constructed out of the inhomogeneously transforming spin connection $`\omega _{M\mu }^{mn}`$
$`F^{mn}`$ $`=`$ $`d\omega _M^{mn}+\eta _{rs}\omega _M^{mr}\omega _M^{ns}.`$ (2.92)
Expanding the forms yields the field strength tensor
$$F_{\mu \nu }^{mn}=_\mu \omega _{M\nu }^{mn}_\nu \omega _{M\mu }^{mn}+\eta _{rs}\omega _{M\mu }^{mr}\omega _{M\nu }^{ns}\eta _{rs}\omega _{M\nu }^{mr}\omega _{M\mu }^{ns}.$$
(2.93)
It can be shown that $`F_{\mu \nu }^{mn}=e^{1n\sigma }e_\rho ^mR_{\sigma \mu \nu }^\rho `$ where $`R_{\sigma \mu \nu }^\rho `$ is the Riemann curvature tensor
$$R_{\sigma \mu \nu }^\rho =_\nu \mathrm{\Gamma }_{\sigma \mu }^\rho _\mu \mathrm{\Gamma }_{\sigma \nu }^\rho +\mathrm{\Gamma }_{\sigma \mu }^\lambda \mathrm{\Gamma }_{\lambda \nu }^\rho \mathrm{\Gamma }_{\sigma \nu }^\lambda \mathrm{\Gamma }_{\lambda \mu }^\rho .$$
(2.94)
The Ricci tensor is given by $`R_{\mu \nu }=R_{\mu \nu \rho }^\rho `$ and hence the scalar curvature is an invariant
$$R=g^{\mu \nu }R_{\mu \nu }=e_m^{1\mu }e_n^{1\nu }F_{\mu \nu }^{mn}.$$
(2.95)
## 3 The Invariant Action
The covariant derivatives of the Maurer-Cartan one-forms provide additional building blocks out of which the invariant action is to be constructed. For example the covariant derivatives of $`\omega _{Z\nu }`$ and $`\omega _{K\nu }^n`$ yield the mixed tensors
$`_\mu \omega _{Z\nu }`$ $`=`$ $`_\mu \omega _{Z\nu }\mathrm{\Gamma }_{\mu \nu }^\rho \omega _{Z\rho }`$ (3.1)
$`_\mu \omega _{K\nu }^n`$ $`=`$ $`_\mu \omega _{K\nu }^n\mathrm{\Gamma }_{\mu \nu }^\rho \omega _{K\rho }^n\omega _{M\mu }^{nr}\omega _{K\nu }^s\eta _{rs}.`$ (3.2)
So proceeding, the invariant action describing the curved $`AdS_d`$ brane embedded in curved $`AdS_{d+1}`$ space has the general low energy form
$`\mathrm{\Gamma }`$ $`=`$ $`{\displaystyle }d^dxdete\{\mathrm{\Lambda }+\kappa ^2R+{\displaystyle \frac{1}{2}}\omega _{Z\mu }[(M^2+\xi R)g^{\mu \nu }+\zeta R^{\mu \nu }]\omega _{Z\nu }`$ (3.9)
$`{\displaystyle \frac{1}{2}}_\mu \omega _{Z\nu }_\rho \omega _{Z\sigma }\left[Z_1(g^{\mu \rho }g^{\nu \sigma }g^{\mu \sigma }g^{\nu \rho })+Z_2g^{\mu \nu }g^{\rho \sigma }\right]`$
$`+{\displaystyle \frac{1}{2}}\omega _{K\mu }^m\omega _{K\nu }^n\left[ae_m^{1\mu }e_n^{1\nu }+be_n^{1\mu }e_m^{1\nu }+cg^{\mu \nu }\eta _{mn}\right]`$
$`\omega _{K\mu }^m_\nu \omega _{Z\rho }[\alpha e_m^{1\mu }g^{\nu \rho }+\beta e_m^{1\nu }g^{\mu \rho }+\gamma e_m^{1\rho }g^{\mu \nu }]\}.`$
Many invariant terms are possible. The above includes a reduced set of terms which leads to a consistent effective theory. The model can be further simplified by setting the parameters $`\xi `$ and $`\zeta `$ to zero. On the other hand, due to the Higgs mechanism, the parameter $`M`$ cannot be zero and is an independent scale in the theory. Since the massive vector $`A_\mu `$ is a Proca field, it can be consistently quantized by further setting $`Z_2`$ to zero. Moreover, exploiting the identity $`_\mu (deteT^\mu )=dete_\mu T^\mu `$ along with the chain rule for covariant differentiation, integration by parts has been used to eliminate redundant terms. A term of the form $`detee_m^{1\mu }\omega _{K\mu }^m`$ has also been excluded from the action since, when $`\omega _{K\mu }^m`$ is eliminated as below, it will not result in any new terms.
The action is independent of any terms containing derivatives acting on $`\omega _{K\mu }^m`$. Hence varying the action with respect to $`\omega _{K\mu }^m`$ yields an algebraic identity relating it to $`_\mu \omega _{Z\nu }`$ as
$`\omega _{K\mu }^m\left[ae_m^{1\mu }e_\sigma ^r+b\delta _\sigma ^\mu \delta _m^r+ce^{1\mu r}e_{\sigma m}\right]`$ (3.10)
$`=`$ $`_\nu \omega _{Z\rho }\left[\alpha g^{\nu \rho }e_\sigma ^r+\beta e^{1\rho r}\delta _\sigma ^\nu +\gamma e^{1\nu r}\delta _\sigma ^\rho \right].`$ (3.11)
Introducing $`\omega _{K\mu }^m=e^{1\nu m}B_{\mu \nu }`$ allows the solution
$`B_{\sigma \rho }`$ $`=`$ $`g_{\rho \sigma }{\displaystyle \frac{1}{(b+c)}}\left[\alpha a{\displaystyle \frac{(\alpha d+\beta +\gamma )}{(ad+b+c)}}\right]g^{\mu \nu }_\mu \omega _{Z\nu }`$ (3.13)
$`+{\displaystyle \frac{(\beta +\gamma )}{2(b+c)}}\left[_\sigma \omega _{Z\rho }+_\rho \omega _{Z\sigma }\right]+{\displaystyle \frac{(\beta \gamma )}{2(bc)}}\left[_\sigma \omega _{Z\rho }_\rho \omega _{Z\sigma }\right].`$
Substituting this back into the action allows $`\omega _{K\mu }^m`$ to be eliminated in favor of terms involving the vielbein and $`\omega _{Z\mu }`$ yielding
$`\mathrm{\Gamma }`$ $`=`$ $`{\displaystyle }d^dxdete\{\mathrm{\Lambda }+\kappa ^2R+{\displaystyle \frac{1}{2}}\omega _{Z\mu }[(M^2+\xi R)g^{\mu \nu }+\zeta R^{\mu \nu }]\omega _{Z\nu }`$ (3.16)
$`{\displaystyle \frac{1}{2}}Z_1_\mu \omega _{Z\nu }_\rho \omega _{Z\sigma }(g^{\mu \rho }g^{\nu \sigma }g^{\mu \sigma }g^{\nu \rho })\},`$
where the form of the contractions of the product of two $`_\mu \omega _{Z\nu }`$ terms are similar to those of the initial action and hence the constants have just been redefined and the effective $`Z_2`$ has been set to zero. Exploiting the form of the covariant derivative of the $`Z`$ one-form in order to define the anti-symmetric field strength tensor $`F_{\mu \nu }`$,
$$F_{\mu \nu }=(_\mu \omega _{Z\nu }_\nu \omega _{Z\mu })=(_\mu \omega _{Z\nu }_\nu \omega _{Z\mu }),$$
(3.17)
the action becomes
$`\mathrm{\Gamma }`$ $`=`$ $`{\displaystyle }d^dxdete\{\mathrm{\Lambda }+\kappa ^2R+{\displaystyle \frac{1}{2}}\omega _{Z\mu }[(M^2+\xi R)g^{\mu \nu }+\zeta R^{\mu \nu }]\omega _{Z\nu }`$ (3.20)
$`{\displaystyle \frac{Z_1}{4}}F_{\mu \nu }g^{\mu \rho }g^{\nu \sigma }F_{\rho \sigma }\}.`$
According to equation (LABEL:variations), $`\varphi `$ and $`v^m`$ transform inhomogeneously under the broken translation and Lorentz transformation local transformations. Hence we now fix the unitary gauge defined by $`\varphi =0=v^m`$. So doing, the covariant one-forms take a simplified form
$`\omega ^m`$ $`=`$ $`dx^\mu e_\mu ^m=dx^\mu _\mu ^m=dx^\mu (\overline{e}_\mu ^m+E_\mu ^m)`$ (3.21)
$`\omega _Z`$ $`=`$ $`dx^\mu A_\mu `$ (3.22)
$`\omega _K^m`$ $`=`$ $`dx^\mu B_\mu ^m`$ (3.23)
$`\omega _M^{mn}`$ $`=`$ $`(\overline{\omega }_M^{mn}+\gamma ^{mn})=dx^\mu (\overline{\omega }_{M\mu }^{mn}+\gamma _\mu ^{mn}).`$ (3.24)
Note that, in this gauge, equation (2.78) reduces to $`_\mu ^m=\overline{e}_\mu ^m+E_\mu ^m`$ and $`N_b^a=\delta _b^a`$. Consequently the vielbein $`e_\mu ^m=_\mu ^bN_b^a=\overline{e}_\mu ^m+E_\mu ^m`$ and thus depends only on the gravitational fluctuation field, $`E_\mu ^m`$, about the $`AdS_d`$ background vielbein $`\overline{e}_\mu ^m`$ and is independent of the vector field. As such, the $`dete`$ gives no contribution to the vector mass even though it is the source of Nambu-Goldstone boson kinetic term in the model with spontaneously broken global isometry. Instead, the mass of the vector, $`M`$, is a completely new scale arising from an independent monomial. This realization of the Higgs mechanism is strikingly different from what occurs when gauging internal symmetries. In that case, when the symmetry is made local, the Nambu-Goldstone boson kinetic term gets replaced by the square of the covariant derivative containing the vector connection. In unitary gauge, the Nambu-Goldstone field vanishes leaving the residual vector mass term whose scale is set by the Nambu-Goldstone decay constant, a scale already present in the global model.
The action, equation (3.20), reduces to that of a massive vector field coupled to a gravitational field with cosmological constant
$$\mathrm{\Gamma }=d^dxdete\left\{\mathrm{\Lambda }+\kappa ^2R\frac{Z_1}{4}F_{\mu \nu }g^{\mu \rho }g^{\nu \sigma }F_{\rho \sigma }+\frac{1}{2}A_\mu \left[(M^2+\xi R)g^{\mu \nu }+\zeta R^{\mu \nu }\right]A_\nu \right\},$$
(3.25)
with the field strength tensor $`F_{\mu \nu }`$ for the vector field
$$F_{\mu \nu }=_\mu A_\nu _\nu A_\mu .$$
(3.26)
The action was constructed by considering gravitational fluctuations about a background static $`AdS`$ space and describes the embedding of a brane in that curved space. The world volume action of the brane is equivalent to that of a world volume gravitational field Einstein-Hilbert action, with corresponding cosmological constant as dictated by the field equations evaluated on the $`AdS`$ background, and the action for a massive vector field in that gravitating space. Furthermore, the action can equally well be used to describe a bosonic brane embedded in a space gravitating about a background Minkowski space by taking the limit $`m^20`$. The vector field remains massive with the mass $`M`$ still being an independent scale. Setting the parameters $`\xi `$ and $`\zeta `$ to zero, the world volume action for a brane embedded in curved space has the form of a massive Abelian gauge theory coupled to gravity.
The work of TEC and STL was supported in part by the U.S. Department of Energy under grant DE-FG02-91ER40681 (Task B) while MN was supported by the Japan Society for the Promotion of Science under the Post-Doctoral Research Program. STL thanks the hospitality of the Fermilab theory group during his sabbatical leave while this project was undertaken. TtV would like to thank the theoretical physics groups at Purdue University and the Tokyo Institute of Technology for their hospitality during visits while this work was being completed.
|
warning/0506/hep-th0506035.html
|
ar5iv
|
text
|
# On braneworld cosmologies from six dimensions, and absence thereof
## Abstract
We consider (thin) braneworlds with conical singularities in six-dimensional Einstein-Gauss-Bonnet gravity with a bulk cosmological constant. The Gauss-Bonnet term is necessary in six dimensions for including non-trivial brane matter. We show that this model for axially symmetric bulks does not possess isotropic braneworld cosmological solutions.
Much work on braneworlds in six-dimensional spacetimes has been done, especially during the last two years. In classical six-dimensional gravity gra or supergravity sugra theories a codimension-two object induces a conical singularity coni , and a cancelation occurring between the brane tension and the bulk gravitational degrees of freedom gives rise to a vanishing effective cosmological constant. Couplings of six-dimensional gravity to sigma models have been discussed in sigma . Other works have focused on static/time-dependent solutions and issues of stability static . It is known that six-dimensional Einstein gravity cannot support a (thin gravitating) braneworld with a non-trivial matter content different than a brane tension cline . Proposals for generalizing the brane equation of state, or deriving cosmologies have been made propo . The situation can be improved if a Gauss-Bonnet term is added to the bulk action, in which case the generic matching conditions of a 3-brane with conical singularities were derived in ruth (see also charmousis ; others ; efforts ). The conservation equation of the braneworld was derived in kofinas . In the present paper we consider the isotropic braneworld cosmology of this theory for axially symmetric bulks around the defect, and with a bulk cosmological constant. We show that the model is incompatible with such braneworld configurations.
We consider the total gravitational brane-bulk action
$`S_{gr}={\displaystyle \frac{1}{2\kappa _6^2}}{\displaystyle }d^6x\sqrt{|๐|}\{2\mathrm{\Lambda }_6+\alpha (^24_{AB}^{AB}`$
$`+_{ABCD}^{ABCD})\}+{\displaystyle \frac{r_c^2}{2\kappa _6^2}}{\displaystyle }d^4x\sqrt{|g|}(R2\mathrm{\Lambda }_4),`$ (1)
where calligraphic quantities refer to the bulk metric tensor $`๐`$, while the regular ones to the brane metric tensor $`g`$. The Gauss-Bonnet coupling $`\alpha `$ has dimensions $`(length)^2`$ and is defined as
$`\alpha ={\displaystyle \frac{1}{8g_s^2}},`$ (2)
with $`g_s`$ the string energy scale, while from the induced-gravity crossover lenght scale $`r_c`$ we can define
$`r_c={\displaystyle \frac{\kappa _6}{\kappa _4}}={\displaystyle \frac{M_4}{M_6^2}}.`$ (3)
Here, $`M_6`$ is the fundamental six-dimensional Planck mass $`M_6^4=\kappa _6^2=8\pi G_6`$, while $`M_4`$ is given by $`M_4^2=\kappa _4^2=8\pi G_4`$. The brane tension is
$`\lambda ={\displaystyle \frac{\mathrm{\Lambda }_4}{\kappa _4^2}}.`$ (4)
The field equations arising from the action (1) are
$`๐ข_{AB}{\displaystyle \frac{\alpha }{2}}(^24_{CD}^{CD}+_{CDEF}^{CDEF})๐_{AB}+2\alpha `$
$`\times (_{AB}2_{AC}_B^C2_{ACBD}^{CD}+_{ACDE}_B^{CDE})`$
$`=\kappa _6^2๐ฏ_{AB}\mathrm{\Lambda }_6๐_{AB}+\kappa _6^2{}_{}{}^{(loc)}T_{AB}^{}\delta ^{(2)},`$ (5)
where $`๐ฏ_{AB}`$ is a regular bulk energy-momentum tensor, $`T_{AB}`$ is the brane energy-momentum tensor, $`{}_{}{}^{(loc)}T_{AB}^{}=T_{AB}\lambda g_{AB}(r_c^2/\kappa _6^2)G_{AB}`$, and $`\delta ^{(2)}`$ is the two-dimensional delta function. Capital indices $`A,B,\mathrm{}`$ are six-dimensional. Assuming that the bulk metric in the brane-adapted coordinate system takes the axially symmetric form
$`ds_6^2=dr^2+L^2(x,r)d\phi ^2+g_{\mu \nu }(x,r)dx^\mu dx^\nu ,`$ (6)
with $`g_{\mu \nu }(x,0)`$ being the braneworld metric and $`\phi `$ having the standard periodicity $`2\pi `$, under the usual assumptions for conical singularities $`L(x,r)=\beta (x)r+๐ช(r^2)`$ for $`r0`$, $`_rL(x,0)=1`$, $`_rg_{\mu \nu }(x,0)=0`$, the general matching conditions for imbedding the 3-brane in the six-dimensional theory (1) were found in ruth (see also charmousis ) as follows
$`K_\lambda ^{\alpha \lambda }K_{\alpha \mu \nu }K_\mu ^{\alpha \lambda }K_{\alpha \nu \lambda }+{\displaystyle \frac{1}{2}}(K^{\alpha \lambda \sigma }K_{\alpha \lambda \sigma }K_\lambda ^{\alpha \lambda }K_{\alpha \sigma }^\sigma )g_{\mu \nu }`$
$`+(\beta ^11+{\displaystyle \frac{r_c^2}{8\pi \alpha \beta }})G_{\mu \nu }+{\displaystyle \frac{\kappa _6^2\lambda 2\pi (1\beta )}{8\pi \alpha \beta }}g_{\mu \nu }={\displaystyle \frac{\kappa _6^2}{8\pi \alpha \beta }}T_{\mu \nu }.`$
(7)
Here, $`K_{\alpha \mu \nu }=๐(_\mu n_\alpha ,_\nu )=n_{\alpha \mu ;\nu }`$ (at $`r=0^+`$) denote the extrinsic curvatures of the brane (symmetric in $`\mu ,\nu `$), where $`n_\alpha `$ ($`\alpha =1,2`$) are arbitrary unit normals to the brane (indices $`\alpha ,\beta ,\mathrm{}`$ are lowered/raised with the matrix $`๐_{\alpha \beta }=๐(n_\alpha ,n_\beta )`$ and its inverse $`๐^{\alpha \beta }`$), while $``$ (also denoted by ;) refers to the Christoffel connection of $`๐`$. For extracting this singular part of equations (5), one has to focus on the worst behaving pieces with the structure $`\delta (r)/L\delta (r)/r`$. Note that with respect to local rotations $`n_\alpha O_\alpha ^\beta (x^A)n_\beta `$, $`K_{\alpha \mu \nu }O_\alpha ^\beta K_{\beta \mu \nu }`$ transforming as a vector, thus Eq.(7) is invariant under changes of the normal frame.
Focusing on the $`๐ช(1/r)`$ terms in the $`r\mu `$ components of equations (5) (which cannot be canceled by any regular $`๐ฏ_{AB}`$ in (5)) we obtain the equation
$`_{\nu \sigma }^{\alpha \sigma }K_{\alpha \lambda }^\lambda _{\lambda \sigma }^{\alpha \sigma }K_{\alpha \nu }^\lambda _{\nu \sigma }^{\alpha \lambda }K_{\alpha \lambda }^\sigma ={\displaystyle \frac{\beta _{,\mu }}{\beta }}[G_\nu ^\mu {\displaystyle \frac{1}{4\alpha }}\delta _\nu ^\mu `$
$`+K_\nu ^{\alpha \sigma }K_{\alpha \sigma }^\mu K_\sigma ^{\alpha \sigma }K_{\alpha \nu }^\mu +{\displaystyle \frac{1}{2}}(K_\sigma ^{\alpha \sigma }K_{\alpha \lambda }^\lambda K^{\alpha \sigma \lambda }K_{\alpha \sigma \lambda })\delta _\nu ^\mu ].`$
(8)
In kofinas it was shown that equation (8) is equivalent to the standard conservation equation on the brane
$`T_{\nu |\mu }^\mu =0,`$ (9)
where $`|`$ refers to the Christoffel connection $`\gamma _{\mu \nu \lambda }=๐(_\lambda _\nu ,_\mu )`$ of the induced brane metric $`g_{\mu \nu }`$. Thus, we do not consider equation (8) further, but only equation (9).
From the $`๐ช(1/r)`$ part of the $`rr`$ component of equations (5) we obtain the following equation, valid at the position of the brane
$`g^{\mu \nu }g_{\mu \nu }^{}[\mathrm{\hspace{0.17em}4}R(g^{\kappa \lambda }g_{\kappa \lambda }^{})^23g^{\kappa \lambda }g_{\kappa \lambda }^{}+2\alpha ^1]8R^{\mu \nu }g_{\mu \nu }^{}`$
$`2g_{\mu \nu }^{}g^{\mu \kappa }g^{\nu \lambda }g_{\kappa \lambda }=0,`$ (10)
where a prime denotes differentiation with respect to $`r`$. Note that in the coordinates (6) it is $`K_{r\mu \nu }=g_{\mu \nu }^{}/2`$, $`K_{\phi \mu \nu }=0`$. We will transform equation (10) to an equivalent and simpler form. To do so, we contract the matching conditions (7) with $`g^{\mu \nu }`$ and replace from this equation the last term of equation (10). Making also use of the trace of equations (7), equation (10) gets the form
$`(\sigma _1G^{\mu \nu }+\sigma _2g^{\mu \nu }+\sigma _3T^{\mu \nu })g_{\mu \nu }^{}=0,`$ (11)
where
$`\sigma _1=1+{\displaystyle \frac{r_c^2}{8\pi \alpha }},\sigma _2={\displaystyle \frac{\kappa _6^2\lambda 2\pi }{8\pi \alpha }},\sigma _3={\displaystyle \frac{\kappa _6^2}{8\pi \alpha }}.`$ (12)
This equation is linear and homogeneous in the components of the extrinsic curvature, does not contain the deficit angle $`\beta `$, and will facilitate our analysis.
The only nontrivial remaining components of equations (5) with a $`๐ช(1/r)`$ part are the $`\mu \nu `$ ones, which give the equation
$`4{\displaystyle \frac{\beta _{,\kappa }}{\beta }}g^{\kappa \lambda }[_{r(\mu |\lambda |\nu )}_{r\sigma \tau (\mu }g_{\nu )\lambda }g^{\sigma \tau }_{r\sigma \lambda \tau }g^{\sigma \tau }g_{\mu \nu }]=\text{c}G_{\mu \nu }`$
$`+{\displaystyle \frac{5}{4}}g_{\mu \kappa }^{}g_{\nu \lambda }^{}g^{\kappa \lambda }+(4R5\text{b}3\text{c}^2+{\displaystyle \frac{2}{\alpha }}){\displaystyle \frac{g_{\mu \nu }^{}}{8}}+\text{c}(5\text{b}+\text{c}^2{\displaystyle \frac{2}{\alpha }}){\displaystyle \frac{g_{\mu \nu }}{8}}`$
$`2R_{(\mu }^\lambda g_{\nu )\lambda }^{}+R^{\kappa \lambda }g_{\kappa \lambda }^{}g_{\mu \nu }+R_{\mu \kappa \nu \lambda }g^{\kappa \lambda }{\displaystyle \frac{1}{2}}g_{\kappa \sigma }^{}g_{\lambda \rho }^{}g^{\kappa \lambda }g^{\sigma \rho }g_{\mu \nu }`$
$`+\text{c}g_{\mu \kappa }^{}g_{\nu \lambda }^{}g^{\kappa \lambda }+{\displaystyle \frac{1}{2}}[2\widehat{g^{\prime \prime }}_{\kappa (\mu }g_{\nu )\lambda }^{}g^{\kappa \lambda }+(\widehat{\text{f}}+\text{c}\widehat{\text{f}})g_{\mu \nu }\text{c}\widehat{g^{\prime \prime }}_{\mu \nu }\widehat{\text{f}}g_{\mu \nu }^{}]`$
$`+{\displaystyle \frac{\widehat{L^{\prime \prime }}}{2\beta }}[4G_{\mu \nu }\text{c}g_{\mu \nu }^{}+g_{\mu \kappa }^{}g_{\nu \lambda }^{}g^{\kappa \lambda }+({\displaystyle \frac{\text{b}+\text{c}^2}{2}}{\displaystyle \frac{1}{\alpha }})g_{\mu \nu }],`$ (13)
where for abbreviating the expression we have defined
$`\text{b}=g^{\mu \nu }g_{\mu \nu }^{},\text{c}=g^{\mu \nu }g_{\mu \nu }^{},\text{f}=g^{\mu \nu }g_{\mu \nu }^{\prime \prime },\text{f}=g^{\mu \nu }g_{\mu \nu }^{\prime \prime },`$ (14)
and an overhat means the regular part of the corresponding quantity.
The only equations remaining to be valid on the brane come from the regular part of the system (5).
There are two cases concerning the form of the possible braneworld solutions: (a) $`K_{\alpha \mu \nu }`$ is not identically zero, and (b) $`K_{\alpha \mu \nu }=0`$. In the case (a) one has to consider all the previous equations together. In the case (b) the matching condition (7) takes the form of purely 4-dimensional Einstein gravity, equation (8) implies $`\beta =`$ constant, equations (10), (11) are identically satisfied, while equation (13) implies $`\widehat{L^{\prime \prime }}=0`$. Considering the six-dimensional Ricci scalar, this contains singular $`\delta (r)/r`$ terms, and, in general, also terms of the form $`1/r`$ (multiplied by $`g_{\mu \nu }^{}`$). Thus, in the case (b) these last $`1/r`$ terms vanish, while in case (a) tidal forces appear in the vicinity of the braneworld. Our aim is to find any 4-dimensional isotropic cosmology compatible with the model or to show that no such cosmology exists. We are interested here in a bulk with a pure cosmological constant $`\mathrm{\Lambda }_6`$; however, for possible use of the present formulation elsewhere we let $`๐ฏ_{AB}`$ non-vanishing.
We consider the bulk cosmological metric of the form (6)
$`ds_6^2=dr^2+L^2(t,r)d\phi ^2n^2(t,r)dt^2+a^2(t,r)\gamma _{ij}(x)dx^idx^j,`$
(15)
where $`\gamma _{ij}`$ is a maximally symmetric 3-dimensional metric characterized by its spatial curvature $`k=1,0,1`$. For the metric (15) the matching conditions (7) are written equivalently as
$`A^2=(1{\displaystyle \frac{1}{\beta }})(X+{\displaystyle \frac{1}{12\alpha }})+{\displaystyle \frac{\sigma _3}{3\beta }}^{(loc)}T_t^t`$ (16)
$`AN=(1{\displaystyle \frac{1}{\beta }})(Y+{\displaystyle \frac{1}{12\alpha }})+{\displaystyle \frac{\sigma _3}{6\beta }}(^{(loc)}T_\mu ^\mu 2^{(loc)}T_t^t),`$ (17)
where
$`A={\displaystyle \frac{a^{}}{a}},N={\displaystyle \frac{n^{}}{n}}`$ (18)
$`X=H^2+{\displaystyle \frac{k}{a^2}},Y={\displaystyle \frac{\dot{H}}{n}}+H^2,`$ (19)
with $`H=\dot{a}/na`$ being the Hubble parameter of the brane and a dot denotes differentiation with respect to $`t`$. Throughout, the lapse function $`n`$ is left undetermined and does not affect the analysis since it corresponds to the temporal gauge choice on the brane. The matter on the brane is taken to be a perfect fluid with energy density $`\rho `$ and pressure $`p=w\rho `$. Equation (11) takes the simple form
$`N=fA,`$ (20)
where
$`f=3{\displaystyle \frac{\sigma _3p+\sigma _2\sigma _1(X+2Y)}{\sigma _3\rho \sigma _2+3\sigma _1X}}.`$ (21)
The $`tt`$ component of equation (13) is
$`A(A^2X{\displaystyle \frac{1}{4\alpha }}+{\displaystyle \frac{2\widehat{a^{\prime \prime }}}{a}})+{\displaystyle \frac{\widehat{L^{\prime \prime }}}{\beta }}(A^2X{\displaystyle \frac{1}{12\alpha }})=0,`$ (22)
while the $`ij`$ components of the same equation give
$`{\displaystyle \frac{4\dot{\beta }}{n\beta }}[{\displaystyle \frac{\dot{A}}{n}}+H(AN)]=NX+2AY3NA^2+{\displaystyle \frac{N+2A}{4\alpha }}`$
$`2(A+N){\displaystyle \frac{\widehat{a^{\prime \prime }}}{a}}2A{\displaystyle \frac{\widehat{n^{\prime \prime }}}{n}}+{\displaystyle \frac{\widehat{L^{\prime \prime }}}{\beta }}[X+2YA(A+2N)+{\displaystyle \frac{1}{4\alpha }}].`$ (23)
From equations (16), (17), (20), we can find the extrinsic curvature and the deficit angle
$`A^2={\displaystyle \frac{2(\sigma _3\rho \sigma _2\frac{\sigma _1}{4\alpha })(YX)+3\sigma _3(\rho +p)(X+\frac{1}{12\alpha })}{\sigma _3(\rho +9p)+8\sigma _26\sigma _1(X+3Y)}},`$ (24)
$`\beta ={\displaystyle \frac{\sigma _3\rho \sigma _2+3\sigma _1X}{3(XA^2+\frac{1}{12\alpha })}}.`$ (25)
The regular part of the $`r\mu `$ components of equations (5) gives on the brane
$`(XA^2+{\displaystyle \frac{1}{4\alpha }}+2H{\displaystyle \frac{\dot{\beta }}{n\beta }}){\displaystyle \frac{\dot{A}}{nA}}+H(1{\displaystyle \frac{N}{A}})(XA^2+{\displaystyle \frac{1}{4\alpha }})`$
$`+[2H^2(1{\displaystyle \frac{N}{A}}){\displaystyle \frac{N}{A}}(XA^2+{\displaystyle \frac{1}{12\alpha }})]{\displaystyle \frac{\dot{\beta }}{n\beta }}={\displaystyle \frac{n\kappa _6^2๐ฏ_r^t}{12\alpha A}}.`$ (26)
(Note that for the case (b) equation (26) is trivially satisfied with $`๐ฏ_r^t=0`$). Similarly, the regular part of the $`rr`$ component of equations (5) gives
$`(XA^2+{\displaystyle \frac{1}{4\alpha }}+2Y2AN){\displaystyle \frac{H\dot{\beta }}{n\beta }}+(XA^2+{\displaystyle \frac{1}{4\alpha }})(YAN+{\displaystyle \frac{1}{4\alpha }})`$
$`+(XA^2+{\displaystyle \frac{1}{12\alpha }})\left[{\displaystyle \frac{1}{n}}({\displaystyle \frac{\dot{\beta }}{n\beta }})^^.+({\displaystyle \frac{\dot{\beta }}{n\beta }})^2\right]={\displaystyle \frac{\mathrm{\Lambda }_6\kappa _6^2๐ฏ_r^r}{12\alpha }}+{\displaystyle \frac{1}{16\alpha ^2}}.`$ (27)
The other regular parts of the system (5) (namely, equations $`\phi \phi `$, $`\mu \nu `$) contain the quantities $`\widehat{a^{\prime \prime }}`$, $`\widehat{n^{\prime \prime }}`$. Considering, now, the bulk system (5), it is expected, due to the Bianchi-Bach-Lanczos identities, that one of these equations, say the $`ij`$ one, is redundant and it is derived from the other equations of the system. Thus, both equations (23), and the $`ij`$ regular part of (5) are redundant. The remaining two regular equations $`\phi \phi `$, $`tt`$ determine $`\widehat{a^{\prime \prime }}`$, $`\widehat{n^{\prime \prime }}`$, while equation (22) gives the value of $`\widehat{L^{\prime \prime }}`$. Equation (26), when $`A,\beta `$ are substituted from (24), (25) becomes an equation for $`\dot{Y}`$ (i.e. $`\ddot{H}`$, or more precisely an autonomous equation for $`\stackrel{\dot{}\dot{}\dot{}}{a}`$) which is the candidate cosmological equation of the model. This equation remains to be compatible with equation (27), which means that the compatibility has to be checked at the order $`\ddot{Y}`$. For the case (b), equation (27) becomes
$`(X+{\displaystyle \frac{1}{4\alpha }})(Y+{\displaystyle \frac{1}{4\alpha }})={\displaystyle \frac{\mathrm{\Lambda }_6\kappa _6^2๐ฏ_r^r}{12\alpha }}+{\displaystyle \frac{1}{16\alpha ^2}},`$ (28)
which is seen to be inconsistent with the solution $`X=(\beta \sigma _1)^1(\sigma _3\rho \sigma _2\beta /4\alpha )/3`$ of the matching conditions (7).
Continuing with the general case (a), we define the variables
$`x=X+{\displaystyle \frac{1}{12\alpha }},P=\sigma _3\rho \sigma _2+3\sigma _1X,ร={\displaystyle \frac{1}{\beta }},`$ (29)
and replacing $`\dot{A}`$ from equation (16), we write the system of equations (26), (27) equivalently as
$`(x{\displaystyle \frac{1}{12\alpha }}{\displaystyle \frac{k}{a^2}})\left({\displaystyle \frac{d\mathrm{ln}ร}{d\mathrm{ln}a}}\right)^2{\displaystyle \frac{1}{6}}(รP6fA^2+{\displaystyle \frac{1}{2\alpha }}){\displaystyle \frac{d\mathrm{ln}ร}{d\mathrm{ln}a}}`$
$`={\displaystyle \frac{n\kappa _6^2A๐ฏ_r^t}{4\alpha HรP}}`$ (30)
$`รP(x{\displaystyle \frac{1}{12\alpha }}{\displaystyle \frac{k}{a^2}}){\displaystyle \frac{d^2\mathrm{ln}ร}{d(\mathrm{ln}a)^2}}รP[{\displaystyle \frac{รP}{6}}(2+5f)(1+f){\displaystyle \frac{k}{a^2}}`$
$`+(x+{\displaystyle \frac{1}{12\alpha }})(1f)]{\displaystyle \frac{d\mathrm{ln}ร}{d\mathrm{ln}a}}{\displaystyle \frac{1}{6}}(รP+{\displaystyle \frac{1}{2\alpha }})[{\displaystyle \frac{1}{\alpha }}รP(1+f)]`$
$`={\displaystyle \frac{n\kappa _6^2A๐ฏ_r^t}{4\alpha H}}+{\displaystyle \frac{\kappa _6^2๐ฏ_r^r\mathrm{\Lambda }_6}{4\alpha }}{\displaystyle \frac{3}{16\alpha ^2}}.`$ (31)
For $`๐ฏ_r^t=0`$, equation (30) is solved for $`d\mathrm{ln}ร/d\mathrm{ln}a`$ as
$`{\displaystyle \frac{d\mathrm{ln}ร}{d\mathrm{ln}a}}={\displaystyle \frac{รP6fA^2+1/2\alpha }{6(xka^21/12\alpha )}}.`$ (32)
Differentiating equation (32) and replacing in equation (31), we obtain the following algebraic equation
$`\chi _5๐^5+\chi _4๐^4+\chi _3๐^3+\chi _2๐^2+\chi _1๐+\chi _0=0,`$ (33)
where $`๐=A^2`$, and $`\chi `$โs are functions of $`x`$, $`\varrho =\sigma _3\rho `$ given in the appendix. Now, the system of equations (26), (27) has been substituted equivalently by the system of equations (32), (33). Dropping from now on $`๐ฏ_r^r`$ completely from the notation, differentiating equation (33) once more, and comparing with equation (32), we finally substitute the system of equations (26), (27) by the algebraic system (33), (34):
$`\psi _7๐^7+\psi _6๐^6+\psi _5๐^5+\psi _4๐^4+\psi _3๐^3+\psi _2๐^2+\psi _1๐+\psi _0=0,`$
(34)
where $`\psi `$โs are functions of $`x,\varrho `$, given in the appendix. After some algebraic manipulation, the system of equations (33), (34) is written equivalently as the following system
$`\text{H}_2๐^2+\text{H}_1๐+1=0`$ (35)
$`\text{H}_1๐+\text{H}_0=0,`$ (36)
where Hโs, Hโs are functions of $`x,\varrho `$ given in the appendix. From equations (35), (36) one obtains
$`(x,\varrho )\text{H}_2\text{H}_0^2\text{H}_1\text{H}_0\text{H}_1+\text{H}_1^2=0.`$ (37)
This equation could still be the (first order) Hubble equation of the model even without containing any integration constants. However, this is not the case, since the consistency of equation (37) with equation (36) gives
$`๐ฅ(x,\varrho )\{3(1+w)x\varrho \text{H}_1+[(1+9w)\varrho 8(\sigma 3\sigma _1x)]\text{H}_0\}_{,x}`$
$`+3(1+w)\varrho [(\varrho +\sigma )\text{H}_1+9\sigma _1\text{H}_0]_{,\varrho }=0,`$ (38)
where $`\sigma =\sigma _2+\sigma _1/4\alpha `$. It can now be checked (e.g. numerically) that on the two-dimensional plane $`(x,\varrho )`$ the two curves $`(x,\varrho )=0`$, $`๐ฅ(x,\varrho )=0`$ do not coincide, which completes our statement of non-existence of isotropic braneworld cosmologies <sup>2</sup><sup>2</sup>2Attempting to generalize metric (15) to an off-diagonal ansatz by adding the term $`2J^i(t,r)\gamma _{ij}(x)dtdx^j`$, the matching conditions (7) provide $`J^i{}_{}{}^{^{}}(t,0^+)=0`$. Additionally, in order to assure an untilded isotropic 4-cosmology, one has to impose $`J^i(t,0)=0`$. Therefore, all the first order brane equations implying the inconsistency remain unaffected..
If we are interested in looking at the compatibility of embedding a maximally symmetric 3-brane (with $`R=4\mathrm{}`$) carrying only a tension in a static bulk, we have to put in the line-element (15) $`L(t,r)=\stackrel{~}{L}(r)`$ (thus $`\beta `$=constant), $`n(t,r)=\stackrel{~}{n}(r)`$, and $`a(t,r)=\stackrel{~}{n}(r)\stackrel{~}{a}(t)`$, where $`\dot{\stackrel{~}{a}}^2+k=\mathrm{}\stackrel{~}{n}(0)^2\stackrel{~}{a}^2/3`$. For the regular case (b), equations (16), (17) coincide giving
$`\sigma _2+\beta /4\alpha =\mathrm{}(\sigma _1\beta ),`$ (39)
equations (20), (26) are trivially satisfied, and equation (27) gives the value of the bulk cosmological constant
$`\mathrm{\Lambda }_6=2\mathrm{}(1+2\alpha \mathrm{}/3),`$ (40)
making the embedding of maximally symmetric branes permissible. This solution generalizes known results from cosmic strings. For the case (a), equation (20) gives
$`\sigma _2=\mathrm{}\sigma _1,`$ (41)
the matching conditions (16), (17) coincide giving
$`A^2=N^2=(\mathrm{}+1/4\alpha )/3,`$ (42)
equation (26) is trivially satisfied, and equation (27) gives again a value for the bulk cosmological constant
$`\mathrm{\Lambda }_6=5/12\alpha ,`$ (43)
with the deficit angle $`\beta `$ remaining undetermined. This is a new solution with a maximally symmetric 3-brane embedded in a six-dimensional bulk with negative cosmological constant (non-$`AdS_6`$), where divergences of the bulk scalar curvature of the form $`1/r`$ appear as approaching the brane.
In conclusion, we have considered a codimension two (thin) braneworld with conical singularities in Einstein-Gauss-Bonnet (-induced gravity) theory with a bulk cosmological constant, where the addition of the Gauss-Bonnet term is known to make meaningful the situation when non-trivial braneworld matter content is included. Considering all the field equations at the position of the brane, we have shown that for axially symmetric bulks an isotropic braneworld cosmological ansatz is incompatible with the model. Technically, this is because there is (excluding the gauge arbitrariness) one equation more than the unknowns, which is finally inconsistent with the other equations. Having developed to some degree our formulation on a general basis, makes it also applicable to other braneworld configurations. It is easily seen that the case of a maximally symmetric 3-brane is compatible with the formulation.
Acknowlegements We wish to thank C. Carvalho, C. Charmousis, T. Christodoulakis, R. Emparan, K. Koyama, R. Maartens, E. Papantonopoulos, E. Verdaguer and in particular J. Garriga for useful discussions. This work is supported by a European Commission Marie Curie Fellowship, under contract MEIF-CT-2004-501432.
Appendix We provide here the quantities $`\chi (x,\varrho )`$ appearing in equation (33)
$`\chi _5=9\sigma _1^2\{(1+9w)\varrho 4[2\sigma 9\sigma _1(x\stackrel{~}{\alpha }\stackrel{~}{k})]\}`$
$`{\displaystyle \frac{\chi _4}{3\sigma _1}}=2(1+12w+27w^2)\varrho ^2+\{4\sigma (5+21w)3\sigma _1[12\stackrel{~}{\alpha }(27w^2`$
$`+30w+2)+18\stackrel{~}{k}(1+17w+18w^2)(13+261w+324w^2)`$
$`x]\}\varrho 4[8\sigma ^23\sigma \sigma _1(19x+3\stackrel{~}{\alpha }9\stackrel{~}{k})54\sigma _1^2x(x\stackrel{~}{\alpha }\stackrel{~}{k})]`$
$`\chi _3=(1+15w+54w^2)\varrho ^36\{(2+13w9w^2)\sigma +2\sigma _1[(2+30w`$
$`+27w^2)x3\stackrel{~}{a}(2+23w+18w^2)\stackrel{~}{k}(4+51w+54w^2)]\}\varrho ^2+3`$
$`\{(931w)\sigma ^24\sigma \sigma _1[3\stackrel{~}{\alpha }(123w18w^2)+2(27w^2+42w+14)`$
$`x\stackrel{~}{k}(11+51w+54w^2)]+3\sigma _1^2x[36\stackrel{~}{\alpha }(2+13w+9w^2)(324w^2`$
$`+297w+53)x+12\stackrel{~}{k}(5+30w+27w^2)]\}\varrho +2[20\sigma ^3+243\sigma _1^3(3\omega `$
$`2x^2)(x\stackrel{~}{\alpha }\stackrel{~}{k})+6\sigma ^2\sigma _1(x9\stackrel{~}{\alpha }+7\stackrel{~}{k})36\sigma \sigma _1^2x(10x+9\stackrel{~}{\alpha }3\stackrel{~}{k})]`$
$`\chi _2=27\sigma _1^2x^3[(13+9w)\varrho +4\sigma ]2\stackrel{~}{\alpha }(\varrho +\sigma )[2\sigma (9w^2+45w+2)\varrho `$
$`34\sigma ^2243\sigma _1^2\omega +2(1+9w9w^2)\varrho ^2]+6\sigma _1x^2\{(81w^2+72w`$
$`+11)\varrho ^2+10\sigma (2\sigma +9\stackrel{~}{\alpha }\sigma _1)+2\varrho [(29+63w+54w^2)\sigma 9\stackrel{~}{\alpha }\sigma _1`$
$`(4+9w)]\}x(\varrho +\sigma )\{(1+21w+144w^2)\varrho ^2+124\sigma ^2+486\sigma _1^2\omega `$
$`360\stackrel{~}{\alpha }\sigma \sigma _1[(55+339w+36w^2)\sigma 72\stackrel{~}{\alpha }\sigma _1(2+16w+9w^2)]`$
$`\varrho \}2\stackrel{~}{k}\{(1+9w18w^2)\varrho ^3+6[(2+15w)\sigma +2\sigma _1(27w^2+30w`$
$`+4)x]\varrho ^2\sigma [26\sigma ^212\sigma \sigma _1x+27\sigma _1^2(2x^2+9\omega )]\}6\stackrel{~}{k}\varrho \{4\sigma \sigma _1`$
$`(5+30w+27w^2)x+9\sigma _1^2[(7+9w)x^29\omega ]\sigma ^2(527w6w^2)\}`$
$`{\displaystyle \frac{\chi _1}{\varrho +\sigma }}=2\stackrel{~}{k}\{2x[(1+6w9w^2)\varrho ^2+(5+42w+9w^2)\sigma \varrho 14\sigma ^2]`$
$`+6\sigma _1x^2[(4+9w)\varrho 5\sigma ]27\sigma _1\omega (\varrho +\sigma )\}18\sigma _1(\varrho +\sigma )(3\stackrel{~}{\alpha }\omega `$
$`+2x^3)+2x\{2\stackrel{~}{\alpha }(2+15w9w^2)\varrho ^2+[27\sigma _1\omega 2\stackrel{~}{\alpha }\sigma (251w`$
$`9w^2)]\varrho +\sigma (27\sigma _1\omega 44\stackrel{~}{\alpha }\sigma )\}x^2\{(19w90w^2)\varrho ^24\sigma `$
$`(20\sigma 63\stackrel{~}{\alpha }\sigma _1)[36\stackrel{~}{\alpha }\sigma _1(2+9w)(47+243w+36w^2)\sigma ]\varrho \}`$
$`{\displaystyle \frac{\chi _0}{(\varrho +\sigma )^2}}=x^2\{[(13w)\varrho +4\sigma ]x4\stackrel{~}{\alpha }[(1+6w)\varrho 5\sigma ]`$
$`2\stackrel{~}{k}[(1+3w)\varrho 2\sigma ]\}2\omega (\varrho +\sigma )(x\stackrel{~}{\alpha }\stackrel{~}{k}),`$
where $`\stackrel{~}{\alpha }1/12\alpha `$, $`\omega (\mathrm{\Lambda }_6\kappa _6^2๐ฏ_r^r+5/12\alpha )/6\alpha `$, and $`\stackrel{~}{k}k/a^2=k(\varrho /\varrho _o)^{2/3(1+w)}`$, with $`\varrho _o>0`$ integration constant.
We give here the quantities $`\psi (x,\varrho )`$ appearing in equation (34)
$`\psi _j=\stackrel{~}{\psi }_j+c_j,\mathrm{\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}0}j7,`$
where
$`\stackrel{~}{\psi }_j==2(x\stackrel{~}{\alpha }\stackrel{~}{k})\{3(1+w)\varrho [x\chi _{j,x}+(\varrho +\sigma )\chi _{j,\varrho }][(1+9w)\varrho `$
$`8(\sigma 3\sigma _1x)]\chi _{j1,x}27(1+w)\sigma _1\varrho \chi _{j1,\varrho }\}`$
$`(c_7,c_6,c_5,c_4,c_3,c_2,c_1,c_0)=(15\sigma _1\chi _5,12\sigma _1\chi _4+5\widehat{\zeta }\chi _5,`$
$`9\sigma _1\chi _3+4\widehat{\zeta }\chi _45\stackrel{ห}{\zeta }\chi _5,6\sigma _1\chi _2+3\widehat{\zeta }\chi _34\stackrel{ห}{\zeta }\chi _4+5\zeta \chi _5,`$
$`3\sigma _1\chi _1+2\widehat{\zeta }\chi _23\stackrel{ห}{\zeta }\chi _3+4\zeta \chi _4,\widehat{\zeta }\chi _12\stackrel{ห}{\zeta }\chi _2+3\zeta \chi _3,`$
$`\stackrel{ห}{\zeta }\chi _1+2\zeta \chi _2,\zeta \chi _1)`$
$`\zeta =x(x+2\stackrel{~}{\alpha })(\varrho +\sigma )`$
$`\widehat{\zeta }=(1+6w)\varrho 5\sigma 6\sigma _1(6x11\stackrel{~}{\alpha }8\stackrel{~}{k})`$
$`\stackrel{ห}{\zeta }=6\stackrel{~}{\alpha }[(12w)\varrho +3\sigma ]+4\stackrel{~}{k}[(13w)\varrho +4\sigma ]+9\sigma _1x^2`$
$`2x[(19w)\varrho +10\sigma 9\stackrel{~}{\alpha }\sigma _1].`$
We provide now the quantities $`\text{H}(x,\varrho )`$, $`\text{H}(x,\varrho )`$ appearing in equations (35), (36)
$`(\text{H}_2,\text{H}_1)=(F_3[(C_2F_1C_1)(B_1p_1)+F_1(B_2p_2)+p_3B_3],`$
$`(F_1F_2F_3)[B_2p_2C_1(B_1p_1)](C_4F_2C_2)(B_1p_1)`$
$`F_2(B_3p_3)+B_5p_5)/[(F_1^2F_2)[B_2p_2C_1(B_1p_1)]`$
$`(C_3F_1C_2)(B_1p_1)F_1(B_3p_3)+B_4p_4]`$
$`(\text{H}_1,\text{H}_0)=((F_3\text{H}_2F_1)[B_2p_2C_1(B_1p_1)]+(C_4\text{H}_2C_2)`$
$`(B_1p_1)+\text{H}_2(B_3p_3)+p_5B_5,(F_2\text{H}_1F_1)[B_2p_2C_1`$
$`(B_1p_1)]+(C_3\text{H}_1C_2)(B_1p_1)+\text{H}_1(B_3p_3)+p_4B_4)`$
where
$`(F_3,F_2,F_1)=(C_4[C_1(B_1p_1)+p_2B_2],(C_1C_3C_4)(B_1p_1)`$
$`C_3(B_2p_2)+B_5p_5,(C_1C_2C_3)(B_1p_1)C_2(B_2p_2)`$
$`+B_4p_4)/[(C_1^2C_2)(B_1p_1)C_1(B_2p_2)+B_3p_3]`$
$`(C_4,C_3,C_2,C_1)=(B_5(B_1p_1),B_4(B_1p_1)+p_5B_5,B_3(B_1`$
$`p_1)+p_4B_4,B_2(B_1p_1)+p_3B_3)/(B_1^2B_2B_1p_1+p_2)`$
$`(B_5,B_4,B_3,B_2,B_1)=(q_7,p_5(p_1q_1)+q_6,p_4(p_1q_1)+q_5p_5,`$
$`p_3(p_1q_1)+q_4p_4,p_2(p_1q_1)+q_3p_3)/(p_1^2p_2p_1q_1+q_2)`$
and $`p_i=\chi _i/\chi _0`$ ($`1i5`$), $`q_j=\psi _j/\psi _0`$ ($`1j7`$).
|
warning/0506/math0506057.html
|
ar5iv
|
text
|
# Nonโvanishing for Koszul cohomology of curves
## 1 Introduction
Let $`X`$ be a smooth complex projective variety. The geometry of $`X`$ is reflected in the behaviour of the Koszul cohomology groups $`K_{p,q}(X,L)`$ introduced by Green , more specifically the vanishing/nonvanishing of certain Koszul cohomology groups. The fundamental result in this direction is the nonvanishing theorem of GreenโLazarsfeld . This theorem states that if a line bundle $`L`$ admits a decomposition $`L=L_1L_2`$ with $`r_i=h^0(X,L_i)11`$ ($`i=1,2`$) then $`K_{r_1+r_21,1}(X,L)0`$. Voisin \[9, (1.1)\] has given a different proof of this result under the hypothesis that $`L_1`$ and $`L_2`$ are globally generated.
The aim of this note is to give a more geometric approach to this type of problems. The starting point is the following construction due to Voisin. Given a rank two vector bundle $`E`$ on $`X`$ with determinant $`L`$, Voisin \[11, (2.22)\] defined a homomorphism
$$\phi :S^pH^0(X,E)^{p+2}H^0(X,E)^pH^0(X,L)H^0(X,L).$$
By \[11, Lemma 5\], this homomorphism produces elements of $`K_{p,1}(X,L)`$. If we take $`E=L_1L_2`$, we get back the classes constructed by Green and Lazarsfeld. As one of the referees pointed out to us, Koh and Stillman had generalised the GreenโLazarsfeld construction before from a different point of view.
Recall that the rank of a Koszul class $`\gamma K_{p,1}(X,L)`$ is the minimal dimension of a linear subspace $`WH^0(X,L)`$ such that $`\gamma `$ is represented by an element in $`^pWH^0(X,L)`$; cf. \[6, Definition 2.2\]. (Note that the subspace $`W`$ is uniquely determined if $`p2`$.) By definition, the Koszul classes constructed in this paper are of rank $`p+2`$ if the vector bundle $`E`$ is indecomposable.
Section 3 contains the main results of this paper. We first give a necessary and sufficient condition for nonvanishing of Koszul classes on smooth curves obtained from rank 2 vector bundles (Theorem 3.1). This result generalises the nonvanishing theorem of GreenโLazarsfeld in the case of curves. Our second main result, Theorem 3.4, states that every rank $`p+2`$ Koszul class on a smooth curve comes from a rank two vector bundle. This theorem is a generalisation of \[6, Theorem 6.7\].
## 2 Preliminaries
### 2.1 The method of Voisin
Let $`E`$ be a rank two vector bundle on a smooth projective variety $`X`$ defined over an algebraically closed field $`k`$ of characteristic zero. Write $`L=detE`$ and $`V=H^0(X,L)`$, and let
$$d:^2H^0(X,E)V$$
be the determinant map. Given $`tH^0(X,E)`$, define a linear map
$$d_t:H^0(X,E)V$$
by $`d_t(u)=d(tu)`$, and choose a subspace $`UH^0(X,E)`$ with $`Uker(d_t)=0`$. Suppose that $`dim(U)=p+2`$ with $`p1`$, and put $`W=d_t(U)U`$. The restriction of $`d`$ to $`^2U`$ defines a map $`^2UV`$, which we can view as an element of
$$^2U^{}V^pUV.$$
Let
$$\gamma ^pWV^pVV$$
be the image of this element under the map $`d_t`$.
Following Voisin \[11, (2.22)\], we prove that $`\gamma `$ defines a Koszul class in $`K_{p,1}(X,L)`$. To this end, we make the previous construction explicit using coordinates. If we choose a basis $`\{e_1,\mathrm{},e_{p+3}\}`$ of $`tUH^0(X,E)`$ such that $`e_1=t`$, we have
$`\gamma `$ $`=`$ $`{\displaystyle \underset{i<j}{}}(1)^{i+j}d(te_2)\mathrm{}\widehat{d(te_i)}\mathrm{}`$
$`\mathrm{}\widehat{d(te_j)}\mathrm{}d(te_{p+3})d(e_ie_j).`$
As in one shows that the image of the $`\gamma `$ by the Koszul differential
$$\delta :^pVH^0(X,L)^{p1}VS^2H^0(X,L)$$
equals
$`{\displaystyle \underset{i<j<k}{}}(1)^{i+j+k}d(te_2)\mathrm{}\widehat{d(te_i)}\mathrm{}\widehat{d(te_j)}\mathrm{}\widehat{d(te_k)}\mathrm{}d(te_{p+3})`$ (2)
$`\{d(te_i)d(e_je_k)d(te_j)d(e_ie_k)+d(te_k)d(e_ie_j)\}.`$
###### Lemma 2.1 (Voisin)
Given four elements $`w_1`$, $`w_2`$, $`w_3`$, $`wH^0(X,E)`$ we have the relation
$$d(ww_1)d(w_2w_3)d(ww_2)d(w_1w_3)+d(ww_3)d(w_1w_2)=0$$
in $`H^0(X,L^2)`$.
Proof: See \[11, Lemma 5\]. $`\mathrm{}`$
The previous lemma shows that $`\gamma `$ belongs to the kernel of the Koszul differential
$$\delta _X:^pVH^0(X,L)^{p1}VH^0(X,L^2).$$
Hence $`\gamma `$ defines a Koszul class $`[\gamma ]K_{p,1}(X,L,W)K_{p,1}(X,L)`$. Clearly the given class only depends on $`t`$ and $`W`$; we write $`[\gamma ]=\gamma (W,t)`$.
### 2.2 The method of GreenโLazarsfeld
Let $`L_1`$, $`L_2`$ be two line bundles on a smooth projective variety $`X`$ such that $`r_i=h^0(X,L_i)11`$ ($`i=1`$, 2). Write $`L_i=M_i+F_i`$ with $`M_i`$ the mobile part and $`F_i`$ the fixed part. Let $`B`$ be the divisorial part of $`F_1F_2`$. It is possible to choose $`s_iH^0(X,L_i)`$ such that $`V(s_1,s_2)=BZ`$ with $`codim(Z)2`$. Set $`L=L_1L_2`$, and put $`t=(s_1,s_2)H^0(X,L_1L_2)`$, $`W=im(d_t)H^0(X,L(B))`$. By construction $`h^0(X,๐ช_X(B))=1`$, hence $`dimW=r_1+r_2+1`$. By the previous discussion, we obtain a Koszul class $`\gamma (W,t)K_{r_1+r_21,1}(X,L)`$. We call such classes GreenโLazarsfeld classes.
Note that the rank of a GreenโLazarsfeld class is either $`p+1`$ or $`p+2`$. Classes of rank $`p+1`$ are of scrollar type; see e.g. or \[6, Corollary 5.2\].
###### Definition 2.2
Given a nonnegative integer $`k0`$, let $`K_{k,1}(X,L)_{\mathrm{GL}}K_{k,1}(X,L)`$ be the subspace generated by GreenโLazarsfeld classes for all decompositions $`L=L_1L_2`$ with $`k=r_1+r_21`$, ($`r_11`$, $`r_21`$).
### 2.3 The method of KohโStillman
Voisinโs method produces syzygies of rank $`p+2`$. As we have seen in the previous subsection, rank $`p+1`$ syzygies are GreenโLazarsfeld syzygies of scrollar type. Rank $`p+2`$ syzygies can be obtained in the following way. Suppose that $`L`$ is a globally generated line bundle on a projective variety $`X`$, and let $`[\gamma ]K_{p,1}(X,L)`$ be a nonzero class represented by an element $`\gamma ^pWV`$ with $`dimW=p+2`$. We view $`\gamma `$ as an element in $`^2W^{}VHom(^2W,V)`$. Following \[6, Proof of Theorem 6.1\] we consider the map
$$\gamma ^{}:^2(W)=W^2WV$$
defined by taking the direct sum of $`\gamma `$ and the inclusion $`WV`$. If we choose a generator $`e_1`$ for the first summand and a basis $`\{e_2,\mathrm{},e_{p+3}\}`$ for $`W`$, we obtain a skewโsymmetric $`(p+3)\times (p+3)`$ matrix $`A`$ by setting
$$a_{ij}=\gamma ^{}(e_ie_j).$$
By construction, the inclusion $`WV`$ corresponds to the map $`\gamma ^{}(e_1)`$. This allows us to identify $`a_{1j}`$ and $`e_j`$, $`2jp+3`$. Let $`\alpha `$ be the image of $`\gamma `$ under the Koszul differential
$$\delta :^pVV^{p1}VS^2V.$$
Writing this out, we obtain
$$\alpha =\underset{i<j<k}{}(1)^{i+j+k}a_{12}\mathrm{}\widehat{a_{1,i}}\mathrm{}\widehat{a_{1,j}}\mathrm{}\widehat{a_{1,k}}\mathrm{}a_{1,p+3}Pf_{1ijk}(A).$$
(3)
As the elements $`\{a_{12},\mathrm{}a_{1,p+3}\}=\{e_2,\mathrm{},e_{p+3}\}`$ are linearly independent, this expression is nonzero if and only if at least one of the Pfaffians $`Pf_{1ijk}(A)`$ is nonzero. Furthermore, since $`\alpha `$ maps to zero in $`^{p1}VH^0(X,L^2)`$ the Pfaffians $`Pf_{1ijk}(A)`$ have to vanish on the image of $`X`$.
The preceding discussion shows that every rank $`p+2`$ syzygy arises from a skewโsymmetric $`(p+3)\times (p+3)`$ matrix $`A`$ such that
1. the elements $`\{a_{12},\mathrm{}a_{1,p+3}\}`$ are linearly independent;
2. there exists a nonzero Pfaffian $`Pf_{1ijk}(A)`$;
3. the Pfaffians $`Pf_{1ijk}(A)`$ vanish on the image of $`X`$ in $`(V^{})`$.
This is exactly the method used by Koh and Stillman to produce syzygies; see \[7, Lemma 1.3\].
###### Remark 2.3
In the geometric setting of subsection 2.1, let $`Y`$ be the image of $`X`$ in $`(V^{})`$. The expression (2) shows that the canonical isomorphism
$$K_{p,1}(X,L)K_{p1,2}(^r,_Y,๐ช_{}(1))$$
maps the class $`\gamma (W,t)`$ to the element $`\alpha `$ defined in (3). Moreover, if $`d`$ does not vanish on decomposable elements then $`\gamma (W,t)0`$. Indeed, this condition is satisfied if and only if the matrix $`A`$ has no generalised zero; cf. \[7, Definition (1.1)\]. One then applies \[loc. cit., Remark p. 122\].
## 3 Main results
###### Theorem 3.1
Let $`X`$ be a smooth curve, let $`L`$ be a baseโpoint free line bundle on $`X`$ and let $`WH^0(X,L)`$ be a linear subspace. Put $`B=Bs(W)`$, and let $`t`$ be a section of $`H^0(X,๐ช_X(B))`$ vanishing on $`B`$. Consider an extension
$$0๐ช_X(B)EL(B)0$$
(4)
such that
$$W(kerH^0(X,L(B))\begin{array}{c}\delta \hfill \end{array}H^1(X,๐ช_X(B))).$$
Then the Koszul class $`\gamma (W,t)`$ defined in section 2.1 is nonzero is and only if the extension (4) is non-split.
Proof: The proof proceeds in several steps. We use the notation of section 2.1.
Step 1. Suppose that the extension (4) splits. In this case, one readily verifies that $`d`$ vanishes identically on $`^2U`$. The formula (2.1) then shows that $`\gamma (W,t)=0`$.
Step 2. If $`\gamma (W,t)=0`$ there exists a linear map $`h:U`$ such that
$$d(u_1u_2)=h(u_2)d_t(u_1)h(u_1)d_t(u_2)$$
(5)
for all $`u_1,u_2U`$.
Indeed, suppose that there exists a nonzero element $`\stackrel{~}{\gamma }^{p+1}WW^{}`$ such that $`\gamma `$ is the image of $`\stackrel{~}{\gamma }`$ under the Koszul differential. Then $`\gamma `$ coincides with the composition of maps
$$^2W\begin{array}{c}\delta \hfill \end{array}WW\begin{array}{c}\stackrel{~}{\gamma }id\hfill \end{array}WV.$$
Since
$`d(u_1u_2)`$ $`=`$ $`\gamma (d_t(u_1)d_t(u_2))`$
$`=`$ $`\stackrel{~}{\gamma }(d_t(u_2))d_t(u_1)\stackrel{~}{\gamma }(d_t(u_1))d_t(u_2)),`$
condition (5) is satisfied with $`h=\stackrel{~}{\gamma }d_t:U`$.
Step 3. Let $`u_1`$, $`u_2U`$ be two sections such that $`d_t(u_1)`$ and $`d_t(u_2)`$ generate $`L(B)`$. If $`d(u_1u_2)=0`$, the extension (4) splits.
To prove this assertion, put $`s_i=d_t(u_i)`$ ($`i=1,2`$) and consider the commutative diagram
$$\begin{array}{ccccccccc}0& & ๐ช_X(B)& \begin{array}{c}\hfill \end{array}& E& \begin{array}{c}\hfill \end{array}& L(B)& & 0\\ & & & & ev_1& & ev_2& & \\ & & 0& & u_1,u_2๐ช_X& \stackrel{}{}& s_1,s_2๐ช_X& & 0.\end{array}$$
Put $`M=ker(ev_1)`$, and note that $`ker(ev_2)L^1(B)`$ since $`ev_2`$ is surjective. By the Snake Lemma we obtain an exact sequence
$$0ML^1(B)๐ช_X(B)coker(ev_1)0.$$
Note that
$$d(u_1u_2)=0rankim(u_1,u_2๐ช_XE)=1rankM=1$$
where the first equivalence follows from \[10, p. 380\]. If $`d(u_1u_2)=0`$ the above exact sequence shows that $`ML^1(B)`$, hence the isomorphism $`u_1,u_2๐ช_X\stackrel{}{}s_1,s_2๐ช_X`$ induces an isomorphism $`im(ev_1)L(B)`$. The inverse of this isomorphism provides a splitting of the extension (4).
Step 4. Suppose that $`\gamma (W,t)=0`$. Then there exists a linear map $`h:U`$ as in Step 2. Consider the morphism
$$\pi :X(W^{})$$
defined by the baseโpoint free linear system $`WH^0(X,L(B))`$, and choose a linear subspace $`\mathrm{\Lambda }(W^{})`$ of codimension two such that $`\mathrm{\Lambda }\pi (X)=\mathrm{}`$. The hyperplane $`ker(h)W`$ corresponds to a point $`p(W^{})`$. Put $`H_1=\mathrm{\Lambda },p`$ and choose a hyperplane $`H_2(W^{})`$ containing $`\mathrm{\Lambda }`$ such that $`pH_2`$. Let $`u_1`$, $`u_2`$ be the sections corresponding to $`H_1`$, $`H_2`$. Then $`d_t(u_1)`$ and $`d_t(u_2)`$ generate $`L(B)`$ and $`u_1ker(h)`$, $`u_2ker(h)`$. Equation (5) yields the identity
$$d(u_1u_2)=h(u_2)d_t(u_1).$$
Rewriting this identity, we obtain $`d(u_1(u_2+h(u_2)t))=0`$. Since the pair $`\{d_t(u_1),d_t(u_2+h(u_2)t)\}=\{d_t(u_1),d_t(u_2)\}`$ generates $`L(B)`$, Step 3 implies that the extension (4) splits. $`\mathrm{}`$
###### Remark 3.2
In the statement of Theorem 3.1 it is not necessary to suppose that $`L`$ is globally generated, since $`K_{p,1}(X,L(Bs(L)))K_{p,1}(X,L)`$.
Theorem 3.1 yields a short, geometric proof of the GreenโLazarsfeld nonvanishing theorem for curves.
###### Theorem 3.3
(GreenโLazarsfeld) Let $`X`$ be a smooth curve, and let $`L`$ be a line bundle on $`X`$ that admits a decomposition $`L=L_1L_2`$ with $`r_i=dim|L_i|1`$ for $`i=1,2`$. Then $`K_{r_1+r_21,1}(X,L)0`$.
Proof: We define $`s_1`$, $`s_2`$, $`t`$, $`W`$, $`B`$ and $`\gamma (W,t)`$ as in section 2.2. Let $`C`$ be the base locus of $`W`$, seen as a subspace of $`H^0(X,L(B))`$. We prove that $`\gamma (W,t)0`$. Suppose that $`\gamma (W,t)=0`$. Consider the extension
$$0๐ช_X(B)L_1L_2L(B)0.$$
Pulling back this extension along the injective homomorphism $`L(BC)L(B)`$, we obtain an induced extension
$$0๐ช_X(B)EL(BC)0.$$
Applying Theorem 3.1 to the line bundle $`L(C)`$, we find that this extension splits. Hence there exists an injective homomorphism
$$๐ช_X(B)L(BC)L_1L_2.$$
In particular there exists $`i\{1,2\}`$ such that $`Hom(L(BC),L_i)0`$. This implies that
$$r_i+1=h^0(X,L_i)h^0(X,L(BC))dimW=r_1+r_2+1,$$
and this is impossible since $`r_11`$ and $`r_21`$. $`\mathrm{}`$
###### Theorem 3.4
Let $`X`$ be a smooth curve, and let $`\alpha 0K_{p,1}(X,L)`$ be a Koszul class of rank $`p+2`$ represented by an element of $`^pWH^0(X,L)`$ with $`dimW=p+2`$. There exist a rank 2 vector bundle $`E`$ on $`X`$ and a section $`tH^0(X,E)`$ such that $`\alpha =\gamma (W,t)`$.
Proof: Put $`T=W`$, and choose a basis $`\{e_1,\mathrm{},e_{p+3}\}`$ of $`T`$ such that $`t=e_1`$ is the generator of the first summand. Writing $`z_{ij}=e_ie_j`$, we obtain a skewโsymmetric matrix $`Z=(z_{ij})`$ and coordinates $`(z_{ij})_{1i<jp+3}`$ on $`(^2T^{})`$. Consider the Grassmannian $`G=G(2,T)`$ of 2โdimensional quotients of $`T`$. The ideal of $`G`$ under the Plรผcker embedding $`G(^2T^{})`$ is generated by the $`4\times 4`$ Pfaffians $`Pf_{ijkl}(Z)`$ of the matrix $`Z`$. Taking exterior powers in the exact sequence
$$0tTW0$$
we obtain an exact sequence
$$0tW^2T^2W0.$$
The linear subspace $`(^2W^{})\left(^2T^{}\right)`$ is defined by the vanishing of the linear forms $`z_{1j}`$, $`j=2,\mathrm{},p+3`$. A straightforward computation then shows that the ideal of the union
$$G(2,T)(^2W^{})(^2T^{})$$
is generated by the Pfaffians $`Pf_{1ijk}(Z)`$. The tautological exact sequence
$$0ST๐ช_GQ0$$
induces an isomorphism $`TH^0(G,Q)`$. Under this isomorphism, we have $`G(2,W)=V(t)`$. As in section 2.3 we associate to the Koszul class $`\alpha `$ a matrix $`A=(a_{ij})`$ of linear forms $`A=(a_{ij})`$ such that
1. The linear forms in the first row of $`A`$ span $`W`$;
2. There exists a nonzero $`4\times 4`$ Pfaffian of $`A`$ involving the first row and column;
3. The $`4\times 4`$ Pfaffians involving the first row and column of $`A`$ vanish on the image of $`X`$ in $`H^0(X,L)^{}`$.
Let $`C`$ be the base locus of the image of $`A`$. Replacing $`L`$ by $`L(C)`$ if necessary ($`W`$ is obviously contained in the image of $`A`$) we can suppose that $`C`$ is empty, hence the matrix $`A`$ defines a morphism
$$\psi :X(^2T^{}).$$
Condition (c) implies that the image $`Y=\psi (X)`$ is contained in the union $`G(2,T)(^2W^{})`$, and condition (a) shows that $`Y`$ is not contained in $`(^2W^{})`$. As $`Y`$ is irreducible, this implies that $`Y`$ is contained in $`G(2,T)`$.
Put $`E=\psi ^{}Q`$. Twisting the exact sequence
$$0_Y๐ช_G\psi _{}๐ช_X0$$
by the universal quotient bundle $`Q`$ and taking global sections, we obtain an exact sequence
$$0H^0(G,Q_Y)H^0(G,Q)\begin{array}{c}\psi ^{}\hfill \end{array}H^0(G,\psi _{}๐ช_XQ)H^0(X,E).$$
Condition (a) implies that $`Y`$ is not contained in $`G(2,W)=G(2,T)(^2W^{})`$, hence $`t`$ does not vanish identically on $`X`$ and defines a global section of $`E`$. The zero locus of this section is given by the equations $`a_{12}=\mathrm{}=a_{1,p+3}=0`$, hence it coincides with $`B`$. Consequently the line bundle $`E`$ is given by an extension
$$0๐ช_X(B)EL(B)0.$$
(6)
Consider the commutative diagram
$$\begin{array}{ccc}0& & 0\\ & & \\ H^0(G,๐ช_G)& \begin{array}{c}.t\hfill \end{array}& H^0(X,๐ช_X(B))\\ t& & t\\ H^0(G,Q)& \begin{array}{c}\psi ^{}\hfill \end{array}& H^0(X,E)\\ & & d_t\\ W& \begin{array}{c}i\hfill \end{array}& H^0(X,L(B)).\end{array}$$
Note that $`keri=WH^0(G,๐ช_G(1)_Y)=0`$ by condition (a). As the map $`H^0(G,Q)W`$ is surjective, we find that $`W`$ is contained in the image of the map $`d_t:H^0(X,E)H^0(X,L(B))`$. Hence the condition of Theorem 3.1 is satisfied. By condition (b) we have $`\gamma (W,t)0`$. Hence the extension (6) does not split by Theorem 3.1. $`\mathrm{}`$
###### Remark 3.5
The union $`G(2,T)(^2W^{})`$ is a generic syzygy scheme; see \[6, Theorem 6.1\]. In \[loc. cit., Theorem 6.7\] it was shown that a rank $`p+2`$ syzygy gives rise to a rank 2 vector bundle if $`L`$ is very ample and the ideal of $`X`$ is generated by quadrics.
The condition of Theorem 3.1 can be reinterpreted in terms of surjectivity of a natural multiplication map.
###### Proposition 3.6
Let $`X`$ be a smooth curve, and let $`WH^0(X,L)`$ be a linear subspace. We put $`B=Bs(W)`$ and view $`W`$ as a baseโpoint free linear subspace of $`H^0(X,L(B))`$. Let
$$\mu :WH^0(X,K_X(B))H^0(K_XL(2B))$$
be the multiplication map. The following conditions are equivalent.
1. The map $`\mu `$ is not surjective;
2. There exists a non-split extension
$$0๐ช_X(B)EL(B)0$$
such that $`W`$ is contained in the kernel of the map $`\delta :H^0(X,L(B))H^1(X,๐ช_X(B))`$.
Proof: We first show that (i) implies (ii). Since $`\mu `$ is not surjective, there exists a hyperplane $`HH^0(X,K_XL(B))`$ that contains $`im(\mu )`$. Let $`\eta `$ be a linear functional defining $`H`$. Put $`0\xi =\eta ^{}H^1(X,L^1(B))`$, and let
$$0๐ช_X(B)EL(B)0$$
be the corresponding non-split extension. Given $`wW`$ and $`vH^0(X,K_X(B))`$, the formula
$$\delta (w)(v)=(\eta \mu )(wv)$$
(7)
shows that $`W`$ is contained in the kernel of $`\delta `$.
For the converse, note that formula (7) implies that $`\eta |_{im\mu }0`$. $`\mathrm{}`$
###### Remark 3.7
If $`B`$ is a fixed divisor, the result of the previous Proposition follows from Greenโs duality theorem \[4, Corollary (2.c.10)\]. Indeed,
$$coker\mu K_{0,1}(X,K_X(B),L(B),W)K_{p,1}(X,B,L(B),W)^{}$$
(8)
and since $`h^0(X,๐ช_X(B))=1`$ we have an injection
$$K_{p,1}(X,B,L(B),W)K_{p,1}(X,L).$$
Theorem 3.4 shows that Voisinโs method may produce nontrivial Koszul classes that are not contained in the space $`K_{p,1}(X,L)_{\mathrm{GL}}`$ spanned by GreenโLazarsfeld classes.
###### Example 3.8
By \[2, Theorem 3.6 and Theorem 4.3\] there exists a smooth curve of genus 14 and Clifford index 5 whose Clifford index is computed by a unique line bundle $`L`$ such that $`L^2=K_X`$. The line bundle $`L`$ embeds $`X`$ in $`^4`$ as a projectively normal curve of degree 13 which is not contained in any quadric of rank $`4`$, and the ideal of $`X`$ is generated by the $`4\times 4`$ Pfaffians of a skewโsymmetric matrix $`(a_{ij})_{1i,j5}`$ with
$$deg(a_{ij})=\{\begin{array}{c}2\text{ if }i=1\text{ or }j=1\hfill \\ 1\text{ if }i2\text{ and }j2\hfill \end{array}$$
such that the quadric $`Q=a_{23}a_{45}a_{24}a_{35}+a_{25}a_{34}`$ has rank 5.
By \[loc.cit.\] the group $`K_{1,1}(X,L)`$ is generated by $`[Q]`$, hence $`I_X`$ contains no quadrics of rank $`4`$. If $`K_{1,1}(X,L)`$ contains a GreenโLazarsfeld class this class would be of scrollar type, since it necessarily comes from two pencils $`|L_1|`$, $`|L_2|`$. This is impossible, since classes of scrollar type give rise to quadrics of rank $`4`$.
The Koszul class $`[Q]K_{1,1}(X,L)`$ has rank 3, since it is represented by the linear subspace $`W=a_{23},a_{24},a_{25}`$. Hence $`[Q]`$ comes from Voisinโs method by Theorem 3.4.
###### Remark 3.9
A more geometric description of a subspace $`W`$ representing $`[Q]`$ is the following. A smooth intersection of the quadric $`V(Q)H^0(X,L)^{}`$ with one of the cubic Pfaffians is a $`K3`$ surface in $`H^0(X,L)^{}`$ containing a line $`\mathrm{}`$ which is disjoint from $`X`$ by \[2, Prop. 4.1\]. The line $`\mathrm{}`$ corresponds to a $`3`$-dimensional linear subspace $`WH^0(X,L)`$, which is base-point-free since $`\mathrm{}`$ does not meet $`X`$.
One could ask whether the syzygies constructed in section 2.1 span $`K_{p,1}(X,L)`$. In principle it may be possible to obtain higher rank syzygies as linear combinations of rank $`p+2`$ syzygies. However, if $`K_{p,1}(X,L)`$ is spanned by a single syzygy of rank $`p+3`$ this is not possible.
###### Example 3.10 (EusenโSchreyer)
Eusen and Schreyer \[3, Theorem 1.7 (a)\] have constructed a smooth curve $`X^5`$ of genus 7 and Clifford index 3 embedded by the linear system $`|K_X(x)|`$ such that $`K_{2,1}(X,K_X(x))`$ is spanned by a syzygy $`s_0`$. The explicit expression for $`s_0`$ given on p.8 of \[loc. cit.\] shows that $`s_0`$ is a rank 5 syzygy. Hence $`s_0`$ cannot be obtained by the GreenโLazarsfeld construction or the method of section 2.1.
Acknowledgements. The first named author was partially supported by a Humboldt Research Fellowship, and by the ANCS contract 2-CEx 06-11-20/2006. We would like to thank Universitรฉ Grenoble 1, I.H.E.S., Universitรคt Bayreuth and Universitรฉ Lille 1 for hospitality during the first stage of this work.
We thank the referees for several comments that helped to improve the presentation of the paper, and for pointing out an error in the previous version of this paper.
M. Aprodu, Romanian Academy, Institute of Mathematics โSimion Stoilowโ, P.O.Box 1-764, RO-014700, Bucharest, ROMANIA, email: Marian.Aprodu@imar.ro
ลcoala Normalฤ SuperioarฤโBucureลti 21, Calea Grivitei Str. 010702-Bucharest, Sector 1 ROMANIA
J. Nagel, Universitรฉ Lille 1, Mathรฉmatiques โ Bรขt. M2, F-59655 Villeneuve dโAscq Cedex, FRANCE, email: nagel@math.univ-lille1.fr
|
warning/0506/hep-ph0506028.html
|
ar5iv
|
text
|
# MINT - A SIMPLE MODEL FOR LOW ENERGY HADRONIC INTERACTIONS
## 1 Introduction
As illustrated in Fig.(1), proton-nucleus ($`pA`$) collisions can proceed with or without colour exchange. The latter case applies for example in diffractive scattering, where beam or target can emerge in a high mass excited state. Reactions with colour exchange, referred to as โnormal inelastic interactionsโ in the following, constitute the bulk of the total cross section. Here a single high mass system is created which decays into the observable final state particles.
Interactions without colour exchange are characterized by an exponential spectrum $`dn/dte^{bt}`$ for the momentum transfer $`t`$ and mass spectra $`dn/dM1/M`$ for diffractively excited states, essentially independent of the target nucleus $`A`$. This is consistent with the picture that reactions without colour exchange are peripheral, involving momentum transfer only between the beam proton and a single nucleon of the target nucleus. In contrast, normal inelastic interactions show a significant $`A`$-dependence, indicating a more central character of such collisions.
A Monte Carlo model for the generation of exclusive final states in $`pA`$ collisions can be constructed in a straightforward way, using as external input the measured cross sections for the different sub-processes and a prescription for the transition of an initial high mass system into final particles. The latter is provided by MINT, based on the assumption that this transition is universal, i.e. it is the same for normal inelastic interactions or diffractive masses.
## 2 The MINT Model
The phenomenological basis of the MINT model is the observation that in hadronic interactions with center-of-mass energy $`E_{cm}`$ large against the nucleon mass, the typical transverse momenta of final state particles are negligible compared to their longitudinal momenta. Assuming further that dynamical effects such as hard parton-parton scattering can be ignored, which is a reasonable approximation at low energies, the final state is governed by longitudinal phase space. Then, for zero transverse momentum, $`p_T=0`$, the Lorentz invariant phase space element of a free particle, $`d^3p/2E=dp_T^2d\varphi dy/4`$, implies a uniform particle density in rapidity $`y`$. In MINT, for normal inelastic interactions the longitudinal direction in the center-of-mass system is along the axis defined by the momenta of the colliding particles. For diffractive scattering it is chosen along the direction of the outgoing systems.
To reconcile longitudinal phase space and finite transverse momenta of the final state particles, MINT employs a two step procedure. The basic idea is to start by generating primary clusters with zero transverse momentum, which then perform two-body decays into the actual final state particles. The mass spectrum of the primary clusters
$$\frac{dn}{dm}=a^2me^{am}\text{ }\text{easily generated by}\text{ }m=\frac{1}{a}\mathrm{ln}(r_1r_2)$$
(1)
from two uniform random deviates $`r_1,r_2[0,1]`$ and the still to be defined parameter $`a`$, thus determines the $`p_T`$-spectrum of the final state particles. The case of massless secondaries can be solved analytically to yield $`dn/dp_T=4a^2p_TK_0(2ap_T)`$ where $`K_0`$ is the modified Bessel function of order zero. For large $`p_T`$ it shows an approximately exponential behaviour, in agreement with observations. The mean transverse momentum is given by $`p_T=\pi /4a`$ which motivates to use the rescaled quantity $`\alpha =4a/\pi `$ as the free parameter of the model. For decays into massive secondaries the spectrum will be slightly different, but will not change qualitatively.
Given the mass $`m`$ of a cluster, its rapidity is generated uniformly over the kinematically allowed range $`\pm \mathrm{ln}(E_{cm}/m)`$. The generation of mass and rapidity is iterated until the total invariant mass $`M`$ of the system exceeds $`E_{cm}`$. Then the generation stops and the primary clusters are shifted in rapidity such that the longitudinal momentum balances, using $`p_L=m_T\mathrm{sinh}y`$ and $`m_T=m`$ for $`p_T=0`$. Finally all masses are scaled such that $`M=E_{cm}`$, i.e. MINT satisfies exact energy-momentum conservation. Note also that in this scheme the particle multiplicity distribution is an absolute prediction by the model.
Since flavour physics is not the objective of MINT, only decays into pions and photons are considered. Clusters with a mass $`m<2m_\pi `$, where $`m_\pi `$ is the charged-pion mass, are neutral and decay into two photons. Heavier clusters are assumed to behave like $`\rho `$-mesons. They are randomly assigned charges $`Q=\{1,0,+1\}`$ with equal probability but subject to the constraint of global charge conservation. Charged primaries decay according to $`X^\pm \pi ^\pm \pi ^0\pi ^\pm \gamma \gamma `$ and neutral ones via the mode $`X^0\pi ^+\pi ^{}`$.
The above discussion defines the generation of final state particles from a single high mass system. It remains to specify how normal inelastic $`pA`$ interactions shall be modeled. In MINT this is done as the incoherent sum of $`n`$ subsystems. Here $`1nA`$ is the number of participating nucleons from the target nucleus, drawn from a Poisson distribution with mean value given by the number of nucleons intercepted by the beam proton. A uniform distribution of the impact point of the beam proton on the target nucleus is assumed. Every subsystem has the same invariant mass, which is a natural assumption for the case where the beam proton hits the nucleus at rest, its charge is randomly chosen from $`Q=\{0,+1\}`$, with a probability $`Z/A`$ for $`Q=+1`$. To account for the charge of the beam proton, the charge of the first subsystem is increased by one unit. A more detailed discussion of the implementation of the model can be found in a HERA-B note describing the MINT model $`^\mathrm{?}`$.
## 3 Model Tuning and Comparison with Real Data
Since MINT is based on the assumption that the transition of a primary high mass system into final state particles is universal, the adjustment of the only free parameter of the model is done to the transverse momentum spectrum observed in target single diffraction processes $`^\mathrm{?}`$. Figure (2) shows for three values $`\alpha `$ how, as a function of the mass $`M_x`$ of the diffractive system, MINT compares with data for the average $`p_T`$ in the range $`0.25\text{GeV}/c<p_T<2\text{GeV}/c`$. The value $`\alpha =0.28`$ roughly matches the measurements and was used throughout later on. Given the relative simplicity of the model, no attempt was made to perform an actual fit to the data.
As a first test, in Fig.(3) the mean charged particle multiplicities for inelastic proton-proton collisions predicted by MINT as a function of the center-of-mass energy is compared to a real data. Up to the energies reached at the CERN Intersecting Storage Rings (ISR) the agreement is surprisingly good, with a discrepancy below one track per event. Approximate KNO scaling is found for charged and total multiplicities in the energy range from $`E_{cm}=10`$ to $`60`$ GeV.
In addition, a comparison was done with data from $`pA`$ collisions recorded by the HERA-B $`^\mathrm{?}`$ fixed-target experiment at the HERA storage ring of DESY/Hamburg. The nucleon-nucleon center-of-mass energy was $`\sqrt{s_{NN}}=41.5`$ GeV. The detector is a forward magnetic spectrometer with an angular acceptance of $`15220`$ mrad in the bending plane. The tracking systems consists of a vertex detector (VDS) before and Outer Tracker chambers behind the magnet; particle identification is performed by a RICH detector, an electromagnetic calorimeter (ECAL) and a muon system. Target materials used in 2002/03 were Carbon, Titanium and Tungsten.
Figure (4) shows how MINT and the FRITIOF $`^\mathrm{?}`$ model after the full detector simulation compare to real data. The comparison covers sub-detector specific quantities, such as track segments reconstructed in the VDS, the number of hits seen in the RICH and the number of clusters per event from the ECAL, as well as physics quantities like the number of charged tracks per event passing through the entire tracking system, the transverse momentum distribution and the pseudo-rapidity distribution of those tracks. In general, MINT describes the data as well as FRITIOF. Given the simplicity and minimal amount of tuning that went into the model, it is surprisingly accurate. Interestingly, the number of hits in the RICH and the transverse momentum spectrum with its high-$`p_T`$ tail is better reproduced by MINT than by FRITIOF. The same qualitative findings apply for the lighter Carbon and the heavier Tungsten target without retuning the model.
## 4 Summary
A simple and surprisingly accurate model has been presented for the description of $`pA`$-collisions with nucleon-nucleon center-of-mass energies up to $`E_{cm}60`$ GeV. The model has only a single adjustable parameter, which makes it very convenient for systematic studies exploring the sensitivity of a physics analysis to details of a Monte Carlo model. MINT, which incorporates also elastic and diffractive scattering in its implementation satisfies exact energy-momentum and charge conservation, but features only charged pions and photons in the final state.
## Acknowledgments
Sincere thanks go to Mikhail Zavertyaev and Marco Bruschi for the detailed comparison between MINT and HERA-B data.
## References
|
warning/0506/physics0506059.html
|
ar5iv
|
text
|
# Energy-conserving Finite-๐ฝ Electromagnetic Drift-fluid Equations
## I Introduction
The derivation of dissipationless nonlinear reduced-fluid equations that explicitly conserve energy represents an important goal of the numerical simulation communities in fusion Red\_Fus\_1 -Red\_Fus\_8 and space redfluid\_space\_1 -redfluid\_space\_2 plasma physics. The existence of an exact energy conservation law plays an important role in benchmarking numerical investigations of energy-transfer processes in a nonlinear plasma turbulent state between the fluid kinetic energy, its internal energy, and the electromagnetic field energy. Hence, if and when dissipative effects and/or advanced fluid-closure schemes are introduced in a nonlinear dissipationless reduced-fluid model, one can develop a deeper understanding of the roles played by dissipationless and dissipative effects on the long-time evolution of the turbulent plasma state.
The variational formulation of dissipationless fluid dynamics ensures that the nonlinear fluid equations satisfy an exact energy conservation law derived by application of the Noether method Noether\_1 -Noether\_3 . Previous variational formulations of reduced fluid models include the drift-fluid work of Pfirsch and Correa-Restrepo PCR\_1 ; PCR\_2 and the recent gyrofluid works of Strintzi and Scott SS and Strintzi, Scott, and Brizard SSB . In contrast to Refs. SS ; SSB , however, we ignore finite-Larmor-radius (FLR) effects to focus our attention on the drift-fluid polarization and magnetization effects introduced by the low-frequency magnetic field fluctuations.
In the energy-conserving drift-Alfvรฉn model presented here, the perturbed electric and magnetic fields
$$๐\varphi \frac{\widehat{๐ป}_0}{c}\frac{A_{}}{t}\mathrm{and}๐_{}A_{}\mathbf{\times }\widehat{๐ป}_0,$$
(1)
are expressed in terms of the fluctuation scalar potentials $`\varphi `$ and $`A_{}`$, and the time-independent background magnetic field $`๐_0=B_0\widehat{๐ป}_0`$ is assumed to be spatially nonuniform. From Eq. (1), we define the following perturbed $`E\times B`$ velocities:
$`๐ฎ_E`$ $``$ $`{\displaystyle \frac{c\widehat{๐ป}_0}{B_0}}\mathbf{\times }_{}\varphi =๐_{}\mathbf{\times }{\displaystyle \frac{c\widehat{๐ป}_0}{B_0}},`$ (2)
$`V_{}`$ $``$ $`{\displaystyle \frac{c_{}\varphi }{B_0}}\mathbf{}{\displaystyle \frac{_{}A_{}}{B_0}}={\displaystyle \frac{c\widehat{๐ป}_0}{B_0^2}}\mathbf{}\left(๐_{}\mathbf{\times }๐_{}\right),`$ (3)
where $`๐ฎ_E`$ denotes the linear perturbed $`E\times B`$ velocity and $`V_{}`$ denotes the parallel component of the nonlinear perturbed $`E\times B`$ velocity. Lastly, according to the finite-$`\beta `$ ordering finite\_beta\_1 -finite\_beta\_3 ($`m_e/m_i<\beta 1`$), we neglect the perpendicular component of the fluctuating vector potential ($`๐_{}0`$), so that the parallel component of the fluctuating magnetic field is assumed to vanish in what follows.
The remainder of this paper is organized as follows. In Sec. II, we present the heuristic derivation of the nonlinear drift-fluid Lagrangian density to be used, in Sec. III, in the variational derivation of our self-consistent, energy-conserving drift-Alfvรฉn equations. In Sec. IV, we use the Noether method to derive the local form of the energy conservation law for our drift-Alfvรฉn model. Next, we integrate the local form of the energy conservation law to obtain its global form, in which the energy-transfer processes between the parallel kinetic energy, internal energy, and electromagnetic field energy are manifestly expressed. In Sec. V, we summarize our variational derivation of the energy-conserving drift-Alfvรฉn model. Lastly, in Appendix A, we present a procedure developed for including heat fluxes within a fluid Lagrangian variational principle, while in Appendix B, we briefly outline the diamagnetic-cancellation procedure involving the addition of energy-conserving gyroviscous terms in the drift-fluid equations for $`(u_{},p_{},p_{})`$.
## II Drift-fluid Lagrangian Density
The four-moment drift-fluid Lagrangian for the present anisotropic-temperature finite-$`\beta `$ electromagnetic model is based on the previous electrostatic gyrofluid work of Strintzi, Scott, and Brizard SSB . Here, the drift-fluid Lagrangian density (for a fluid species of mass $`m`$ and charge $`e`$) is defined as
$$_{\mathrm{df}}\frac{mn}{2}u_{}^2\left(p_{}+\frac{p_{}}{2}\right)+en\left(๐_0+A_{}\widehat{๐ป}_0\right)\mathbf{}\frac{๐ฎ}{c}en\varphi _{\mathrm{ZLR}}\left[H_{2\mathrm{g}\mathrm{y}}\right],$$
(4)
where $`u_{}๐ฎ\mathbf{}\widehat{๐ป}_0`$ denotes the drift-fluid velocity parallel to the background magnetic field and all terms (except the last one) are standard fluid Lagrangian terms. The last term represents the zero-Larmor-radius (ZLR) limit of the electromagnetic nonlinearities contained in the second-order finite-$`\beta `$ gyrocenter Hamiltonian HLB\_88 :
$`H_{2\mathrm{g}\mathrm{y}}`$ $`=`$ $`{\displaystyle \frac{mc^2}{2B_0^2}}|_{}\varphi |^2+mv_{}{\displaystyle \frac{c_{}\varphi }{B_0}}\mathbf{}{\displaystyle \frac{_{}A_{}}{B_0}}{\displaystyle \frac{mv_{}^2}{2}}{\displaystyle \frac{|_{}A_{}|^2}{B_0^2}}`$
$``$ $`{\displaystyle \frac{m}{2}}\left(|๐ฎ_E|^2+v_{}^2{\displaystyle \frac{|๐_{}|^2}{B_0^2}}\right)+mv_{}V_{},`$
where $`v_{}`$ denotes the parallel gyrocenter particle velocity, the $`E\times B`$ velocities $`๐ฎ_E`$ and $`V_{}`$ are defined in Eqs. (2) and (3), respectively, and the background magnetic-field nonuniformity was neglected. By heuristically transforming this second-order gyrocenter Hamiltonian to its drift-fluid version, following a procedure outlined in Refs. PCR\_1 ; PCR\_2 , the drift-fluid Lagrangian density (5) for each fluid species becomes
$`_{\mathrm{df}}`$ $`=`$ $`{\displaystyle \frac{mn}{2}}\left|u_{}(\widehat{๐ป}_0+๐_{}/B_0)+๐ฎ_E\right|^2\left(p_{}+{\displaystyle \frac{p_{}}{2}}\right)+en\left(๐_0+A_{}\widehat{๐ป}_0\right)\mathbf{}{\displaystyle \frac{๐ฎ}{c}}en\varphi `$ (5)
$``$ $`{\displaystyle \frac{mn}{2}}U_{}^2+en๐_0\mathbf{}{\displaystyle \frac{๐ฎ}{c}}\left(p_{}+{\displaystyle \frac{p_{}}{2}}\right)en\left(\mathrm{\Phi }^{}{\displaystyle \frac{u_{}}{c}}A_{}^{}\right),`$
where we introduced the definitions
$$\begin{array}{ccc}\hfill U_{}& & u_{}bu_{}\left(1+|๐_{}|^2/B_0^2\right)^{1/2}\hfill \\ & & \\ \hfill e\mathrm{\Phi }^{}& & e\varphi m|๐ฎ_E|^2/2\hfill \\ & & \\ \hfill (e/c)A_{}^{}& & (e/c)A_{}mV_{}\hfill \end{array}\},$$
(6)
with the effective potentials $`\mathrm{\Phi }^{}`$ and $`A_{}^{}`$ both containing linear and nonlinear field terms. We point out that, while $`u_{}`$ denotes the parallel drift-fluid velocity along the unperturbed (background) magnetic field $`๐_0`$, $`U_{}`$ denotes the parallel drift-fluid velocity along the total magnetic field $`๐_0+๐_{}`$. Moreover, although the term $`b^21`$ is considered small, it is kept in the drift-fluid Lagrangian (5) to allow a covariant treatment of drift-fluid polarization and magnetization effects in our model \[see Eqs. (10)-(11) below\].
We note that the fluid velocity $`๐ฎ`$ appearing in the drift-fluid Lagrangian density (5) is defined as the particle-fluid velocity expressed in terms of moments of the gyroangle-independent gyrocenter distribution function $`F(๐,v_{},\mu ,t)`$ defined in Ref. Brizard\_89 :
$`n(๐ซ,t)๐ฎ(๐ซ,t)`$ $`=`$ $`{\displaystyle f(๐ฑ,๐ฏ,t)๐ฏ\delta ^3(๐ฑ๐ซ)d^3xd^3v}`$ (7)
$``$ $`{\displaystyle F(๐,v_{},\mu ,t)(๐ฏ_{}+๐ฏ_{\mathrm{gc}}+\mathrm{})\delta ^3(๐+๐๐ซ)d^6Z}.`$ (8)
Here, $`\mathrm{}`$ denotes a gyroangle average and we used the transformation from a particle-fluid moment (7) to a gyrofluid moment (8), which was introduced in Ref. brizard , where $`๐ฏ_{\mathrm{gc}}`$ denotes the guiding-center (gc) velocity and higher-order corrections (e.g., electromagnetic fluctuations) are omitted for simplicity. Furthermore, because of the presence of $`\delta ^3(๐+๐๐ซ)`$ in Eq. (8), the gyrofluid moment of the perpendicular particle velocity $`๐ฏ_{}`$ does not vanish but, instead, yields the divergenceless term
$$F(๐,v_{},\mu ,t)๐ฏ_{}\delta ^3(๐+๐๐ซ)d^6Z=\mathbf{\times }\left(p_{}\frac{c\widehat{๐ป}_0}{eB_0}\right)$$
which introduces an important diamagnetic contribution to the particle flux (8):
$$n๐ฎ_\mathrm{P}\frac{c\widehat{๐ป}_0}{eB_0}\mathbf{\times }\left(p_{}\mathrm{ln}B_0+p_{}\widehat{๐ป}_0\mathbf{}\widehat{๐ป}_0\right)\mathbf{\times }\left(p_{}\frac{c\widehat{๐ป}_0}{eB_0}\right)=\frac{c\widehat{๐ป}_0}{eB_0}\mathbf{\times }\left(\mathbf{}๐ฏ\right),$$
(9)
where $`๐ฏ`$ denotes the Chew-Goldberger-Low (CGL) pressure tensor CGL \[see Eq. (21) below\].
The appearance of the perturbed electric and magnetic fields in the drift-fluid Lagrangian (5) yields the following expressions for the drift-fluid polarization and magnetization vectors:
$`๐_{}`$ $``$ $`{\displaystyle \frac{_{\mathrm{df}}}{๐_{}}}={\displaystyle \frac{mnc^2}{B_0^2}\left(๐_{}+\frac{u_{}}{c}\widehat{๐ป}_0\mathbf{\times }๐_{}\right)},`$ (10)
$`๐_{}`$ $``$ $`{\displaystyle \frac{_{\mathrm{df}}}{๐_{}}}={\displaystyle mn\frac{u_{}}{B_0}\left(u_{}\frac{๐_{}}{B_0}+๐_{}\mathbf{\times }\frac{c\widehat{๐ป}_0}{B_0}\right)}.`$ (11)
Because of the mass dependence in the drift-fluid polarization and magnetization vectors (10)-(11), the drift-fluid polarization and magnetization effects are especially important for ion fluid species (in a quasineutral plasma).
The total Lagrangian density for our four-moment electromagnetic drift-fluid model is, therefore, defined as the sum of the Lagrangian density of the electromagnetic field and the drift-fluid Lagrangian density (5) for each fluid species (summation over species is implied):
$``$ $`=`$ $`{\displaystyle \frac{1}{8\pi }}\left(\left|๐_{}\right|^2|๐_{}|^2\right)+en\left(๐_0\mathbf{}{\displaystyle \frac{๐ฎ}{c}}+{\displaystyle \frac{u_{}}{c}}A_{}^{}\mathrm{\Phi }^{}\right)`$ (12)
$`+{\displaystyle \frac{mn}{2}}U_{}^2\left(p_{}+{\displaystyle \frac{p_{}}{2}}\right),`$
where we have omitted the contribution from the parallel electric field in the electric field energy (i.e., $`|๐|^2|๐_{}|^2`$), which removes the parallel displacement current $`(_tE_{})`$ in the parallel drift-fluid Ampรจre equation \[see Eq. (36)\]. The variational fields are the four drift-fluid moments $`(n,๐ฎ,p_{},p_{})`$ for each fluid species and the electromagnetic potentials $`(\varphi ,A_{})`$. Here, we note that only the first four drift-fluid moments $`(n,๐ฎ,p_{},p_{})`$ appear in the drift-fluid Lagrangian density (5) and higher-order drift-fluid moments (e.g., heat fluxes) are omitted as variational dynamical fields; future work will consider the introduction of higher-order drift-fluid moments in a Lagrangian variational setting thermal\_momentum (see Appendix A for further comments). Lastly, we note that, because the background magnetic field $`๐_0`$ is independent of the variational fields, the magnetic-energy term $`|๐_0|^2/8\pi `$ has been removed from the Lagrangian density (12).
## III Variational Derivation of Drift-fluid Dynamical Equations
The drift-Alfvรฉn variational principle
$$\delta (n,๐ฎ,p_{},p_{};\varphi ,A_{};๐ฑ)d^3x๐t=\mathrm{\hspace{0.33em}0}$$
(13)
associated with the drift-fluid Lagrangian density (12) must be expressed in terms of the Eulerian variations $`(\delta n,\delta ๐ฎ,\delta p_{},\delta p_{};\delta \varphi ,\delta A_{})`$:
$`\delta `$ $``$ $`\left(\delta n{\displaystyle \frac{}{n}}+\delta ๐ฎ\mathbf{}{\displaystyle \frac{}{๐ฎ}}+\delta p_{}{\displaystyle \frac{}{p_{}}}+\delta p_{}{\displaystyle \frac{}{p_{}}}+\delta \varphi {\displaystyle \frac{}{\varphi }}+\delta A_{}{\displaystyle \frac{}{A_{}}}\right)`$ (14)
$`+\delta \varphi \mathbf{}{\displaystyle \frac{}{(\varphi )}}+\delta A_{}\mathbf{}{\displaystyle \frac{}{(A_{})}},`$
where the additional $`๐ฑ`$-dependence in $`(\mathrm{};๐ฑ)`$ arises from the nonuniform background magnetic field $`๐_0`$. Here, the Eulerian variations for the drift-fluid moments $`(\delta n,\delta ๐ฎ,\delta p_{},\delta p_{})`$ are not independent of each other but are instead expressed in terms of a virtual fluid displacement $`๐`$ (for each fluid species).
The derivation of the Eulerian variations $`(\delta n,\delta p_{},\delta p_{})`$ are based on dynamical constraints for the drift-fluid moments $`\eta ^a=(n,p_{},p_{})`$. The first dynamical constraint is associated with mass conservation, expressed in terms of the continuity equation
$$\frac{dn}{dt}+n\mathbf{}๐ฎ=\mathrm{\hspace{0.33em}0},$$
(15)
where $`d/dt=/t+๐ฎ\mathbf{}`$ denotes the total time derivative. The next two dynamical constraints are associated with the conservation of the first two single-particle adiabatic invariants, expressed in terms of the Chew-Goldberger-Low equations for the perpendicular and parallel pressures CGL :
$`{\displaystyle \frac{dp_{}}{dt}}+p_{}\mathbf{}๐ฎ+p_{}(๐\widehat{๐ป}_0\widehat{๐ป}_0):๐ฎ`$ $`=`$ $`0,`$ (16)
$`{\displaystyle \frac{dp_{}}{dt}}+p_{}\mathbf{}๐ฎ+\mathrm{\hspace{0.33em}2}p_{}\widehat{๐ป}_0\widehat{๐ป}_0:๐ฎ`$ $`=`$ $`0,`$ (17)
where the higher-order heat-flux moments $`๐ช_{}^{()}`$ and $`๐ช_{}^{()}`$ are omitted here but are considered later in Appendix B.
We begin our derivation of expressions for the Eulerian variations to be used in Eq. (14) by transforming the dynamical constraints (15) and (16)-(17) into Lagrangian variations $`\mathrm{\Delta }\eta ^a=(\mathrm{\Delta }n,\mathrm{\Delta }p_{},\mathrm{\Delta }p_{})`$ by introducing the following limiting process
$$\underset{\mathrm{\Delta }t0}{lim}\left(\mathrm{\Delta }t\frac{d\eta ^a}{dt}\right)\mathrm{\Delta }\eta ^a\mathrm{and}\underset{\mathrm{\Delta }t0}{lim}\left(๐ฎ\mathrm{\Delta }t\right)๐$$
(18)
where $`๐`$ denotes the virtual fluid-displacement (for each fluid species), with the Lagrangian variation of the fluid velocity is defined as $`\mathrm{\Delta }๐ฎd๐/dt`$. Hence, using the procedure (18) on the constraint equations (15)-(17), we find the Lagrangian variations
$`\mathrm{\Delta }n`$ $`=`$ $`n\mathbf{}๐,`$
$`\mathrm{\Delta }p_{}`$ $`=`$ $`p_{}\mathbf{}๐p_{}(๐\widehat{๐ป}_0\widehat{๐ป}_0):๐,`$
$`\mathrm{\Delta }p_{}`$ $`=`$ $`p_{}\mathbf{}๐2p_{}\widehat{๐ป}_0\widehat{๐ป}_0:๐.`$
Next, using the relation $`\delta \eta ^a\mathrm{\Delta }\eta ^a๐\mathbf{}\eta ^a`$ between Lagrangian and Eulerian variations Newcomb , the Eulerian variations $`(\delta n,\delta ๐ฎ,\delta p_{},\delta p_{})`$ to be used in the variational principle (14) are now defined as
$$\begin{array}{ccc}\hfill \delta n& =& \mathbf{}(n๐)\hfill \\ & & \\ \hfill \delta ๐ฎ& =& _t๐+(๐ฎ\mathbf{})๐(๐\mathbf{})๐ฎ\hfill \\ & & \\ \hfill \delta p_{}& =& \mathbf{}(p_{}๐)\mathrm{\hspace{0.33em}2}p_{}\widehat{๐ป}_0\widehat{๐ป}_0:๐\hfill \\ & & \\ \hfill \delta p_{}& =& \mathbf{}(p_{}๐)p_{}(๐\widehat{๐ป}_0\widehat{๐ป}_0):๐\hfill \end{array}\}.$$
(19)
The remaining drift-fluid equations include the evolution equation for the parallel drift-fluid velocity $`u_{}`$ and an explicit expression for the perpendicular component of the drift-fluid velocity $`๐ฎ`$. Lastly, a self-consistent treatment involves the appropriate Poisson and parallel Ampรจre equations for the potentials $`(\varphi ,A_{})`$ expressed in terms of the drift-fluid moments $`(n,u_{})`$.
### III.1 Variation of the Lagrangian density
We now insert the Eulerian variations (19) into the drift-fluid Lagrangian variation (14), and rearranging several terms to isolate $`๐`$ and $`\delta \psi ^i=(\delta \varphi ,\delta A_{})`$, to obtain
$`\delta `$ $``$ $`๐\mathbf{}\left[{\displaystyle \frac{}{t}}\left({\displaystyle \frac{}{๐ฎ}}\right)+\mathbf{}\left(๐ฎ{\displaystyle \frac{}{๐ฎ}}\right)+๐ฎ\mathbf{}{\displaystyle \frac{}{๐ฎ}}+\mathbf{}๐ฏ\left(\eta ^a{\displaystyle \frac{}{\eta ^a}}\right)\right]`$ (20)
$`+\delta \psi ^i\left[{\displaystyle \frac{}{\psi ^i}}\mathbf{}{\displaystyle \frac{}{(\psi ^i)}}\right]+{\displaystyle \frac{\mathrm{\Lambda }}{t}}+\mathbf{}๐ช,`$
where
$$\eta ^a\frac{}{\eta ^a}n\frac{}{n}+p_{}\frac{}{p_{}}+p_{}\frac{}{p_{}},$$
and the tensor $`๐ฏ`$ denotes the CGL pressure tensor CGL
$$๐ฏp_{}\frac{}{p_{}}(๐\widehat{๐ป}_0\widehat{๐ป}_0)\mathrm{\hspace{0.33em}2}p_{}\frac{}{p_{}}\widehat{๐ป}_0\widehat{๐ป}_0=p_{}(๐\widehat{๐ป}_0\widehat{๐ป}_0)+p_{}\widehat{๐ป}_0\widehat{๐ป}_0.$$
(21)
While the space-time divergence terms $`_t\mathrm{\Lambda }+\mathbf{}๐ช`$ in Eq. (20), expressed in terms of the Noether fields
$$\mathrm{\Lambda }๐\mathbf{}\frac{}{๐ฎ},$$
(22)
and
$$๐ช๐ฎ\left(๐\mathbf{}\frac{}{๐ฎ}\right)+\delta \psi ^i\frac{}{(\psi ^i)}+\left[๐ฏ\left(\eta ^a\frac{}{\eta ^a}\right)๐\right]\mathbf{}๐.$$
(23)
do not play a role in the variational principle $`\delta d^3x๐t=0`$, they play a crucial role in the derivation of exact conservation laws based on the Noether method Noether\_1 -Noether\_3 (see Sec. IV).
### III.2 Drift-fluid velocity equations
The stationarity of the action functional $`d^3x๐t`$ with respect to a arbitrary virtual fluid displacement $`๐`$ yields the Euler-Poincarรฉ equation EP\_eq (for each fluid species)
$$0=\frac{}{t}\left(\frac{}{๐ฎ}\right)+\mathbf{}\left(๐ฎ\frac{}{๐ฎ}\right)+๐ฎ\mathbf{}\frac{}{๐ฎ}+\mathbf{}๐ฏ\left(\eta ^a\frac{}{\eta ^a}\right),$$
(24)
which describes the time evolution of the drift-fluid velocity $`๐ฎ`$. Upon substituting derivatives of the drift-fluid Lagrangian density (12) into the Euler-Poincarรฉ equation (24), and using the fact that the background magnetic field $`๐_0`$ is assumed to be a time-independent nonuniform vector field, Eq. (24) can be written as
$$0=n\widehat{๐ป}_0\frac{}{t}\left(mu_{}b^2+\frac{e}{c}A_{}^{}\right)\frac{en}{c}๐ฎ\mathbf{\times }๐^{}+\mathbf{}๐ฏ+n\left[e\mathrm{\Phi }^{}+\frac{m}{2}(u_{}b)^2\right],$$
(25)
where we have introduced the following divergenceless field
$`๐^{}`$ $``$ $`๐_0+\mathbf{\times }\left[\left(A_{}^{}+{\displaystyle \frac{mc}{e}}u_{}b^2\right)\widehat{๐ป}_0\right],`$ (26)
from which we define
$`B_{}^{}`$ $``$ $`\widehat{๐ป}_0\mathbf{}๐^{}=B_0+\left[A_{}^{}+\left({\displaystyle \frac{mc}{e}}\right)u_{}b^2\right]\widehat{๐ป}_0\mathbf{}\mathbf{\times }\widehat{๐ป}_0,`$ (27)
$`๐ป^{}`$ $``$ $`{\displaystyle \frac{๐^{}}{B_{}^{}}}=\widehat{๐ป}_0+{\displaystyle \frac{c\widehat{๐ป}_0}{eB_{}^{}}}\mathbf{\times }\left[\left(mu_{}b^2+{\displaystyle \frac{e}{c}}A_{}^{}\right)๐ฟ_0\left(mu_{}b^2+{\displaystyle \frac{e}{c}}A_{}^{}\right)\right],`$ (28)
where $`๐ฟ_0=\widehat{๐ป}_0\mathbf{}\widehat{๐ป}_0`$ denotes the background magnetic curvature. We note that the term $`\widehat{๐ป}_0\mathbf{}\mathbf{\times }\widehat{๐ป}_0`$ in Eq. (27), which is related to the unperturbed parallel current flowing along the background magnetic field, may vanish for some magnetic geometries (e.g., magnetic-dipole geometry).
The Euler-Poincarรฉ equation (25) can be divided into two equations: one equation that expresses the drift-fluid velocity $`๐ฎ`$ in terms of the drift-fluid moments $`(n,u_{},p_{},p_{})`$ and the scalar potentials $`(\varphi ,A_{})`$, and one equation that describes the time evolution of the parallel drift-fluid velocity $`u_{}`$.
#### III.2.1 Drift-fluid velocity
The first equation is obtained by taking the cross-product of Eq. (25) with $`\widehat{๐ป}_0`$, which yields the following first-order expression for the drift-fluid velocity:
$$๐ฎu_{}๐ป^{}+\frac{c\widehat{๐ป}_0}{enB_{}^{}}\mathbf{\times }\left[\mathbf{}๐ฏ+n\left(e\mathrm{\Phi }^{}+\frac{m}{2}(u_{}b)^2\right)\right],$$
(29)
where $`๐ป^{}`$ is defined in Eq. (28). Here, the divergence of the CGL pressure tensor (21) is
$$\mathbf{}๐ฏ=p_{}+p_\mathrm{\Delta }๐ฟ_0+\left[\mathbf{}\left(p_\mathrm{\Delta }\widehat{๐ป}_0\right)\right]\widehat{๐ป}_0,$$
(30)
(with $`p_\mathrm{\Delta }p_{}p_{}`$ denotes the pressure anisotropy) so that $`\widehat{๐ป}_0\mathbf{\times }(\mathbf{}๐ฏ)=\widehat{๐ป}_0\mathbf{\times }(p_{}+p_\mathrm{\Delta }๐ฟ_0)`$. According to Eq. (29), the drift-fluid velocity $`๐ฎ`$, therefore, consists of the parallel velocity $`u_{}๐ฎ\mathbf{}\widehat{๐ป}_0`$ (note that $`\widehat{๐ป}_0\mathbf{}๐ป^{}1`$) and the following perpendicular drift-fluid velocities
$$\begin{array}{ccc}\hfill ๐ฎ_D& & \left(c\widehat{๐ป}_0/enB_{}^{}\right)\mathbf{\times }p_{}\hfill \\ & & \\ \hfill ๐ฎ_\mathrm{\Phi }& & \left(c\widehat{๐ป}_0/eB_{}^{}\right)\mathbf{\times }\left(e\mathrm{\Phi }^{}+mU_{}^2/2\right)\hfill \\ & & \\ \hfill ๐ฎ_C& & \left(c\widehat{๐ป}_0/enB_{}^{}\right)\mathbf{\times }p_\mathrm{\Delta }๐ฟ_0\hfill \end{array}\},$$
(31)
corresponding to the diamagnetic velocity, the generalized $`E\times B`$ velocity (which includes the gradients of $`|๐ฎ_E|^2`$ and $`U_{}^2u_{}^2b^2`$), and the curvature-drift velocity, respectively. We note that, although the higher-order polarization drift velocity
$$๐ฎ_{\mathrm{pol}}\frac{\widehat{๐ป}_0}{\mathrm{\Omega }_0}\mathbf{\times }\frac{d๐ฎ_E}{dt}+\mathrm{}$$
(32)
is not included in the drift-fluid velocity (29), it is self-consistently introduced in the net current density $`๐`$ (through the polarization current density $`_t๐_{}`$) in the form of the quasineutrality condition $`\mathbf{}๐=0`$ (see Sec. III.4 for more details).
#### III.2.2 Evolution equation for $`u_{}`$
The second equation to be extracted from the Euler-Poincarรฉ equation (25), which describes the time evolution equation for parallel drift-fluid velocity $`u_{}`$, can be obtained by taking the dot-product of Eq. (25) with $`๐ป^{}`$:
$$n\frac{}{t}\left(mu_{}b^2+\frac{e}{c}A_{}^{}\right)=๐ป^{}\mathbf{}\left[\mathbf{}๐ฏ+n\left(e\mathrm{\Phi }^{}+\frac{m}{2}U_{}^2\right)\right].$$
(33)
Here, the terms on the left side are
$$\frac{}{t}\left(u_{}b^2\right)=b\frac{U_{}}{t}+u_{}\left(\frac{_{}A_{}}{B_0^2}\mathbf{}_{}\frac{A_{}}{t}\right)$$
and
$$\frac{e}{c}\frac{A_{}^{}}{t}\frac{e}{c}\frac{A_{}}{t}m\frac{V_{}}{t},$$
which contains both the parallel inductive electric field and the partial time derivative of the parallel component of the nonlinear $`E\times B`$ velocity (3); note that the parallel component of the partial time derivative of the linear $`E\times B`$ velocity vanishes $`(\widehat{๐ป}_0\mathbf{}_t๐ฎ_E0)`$. Note also that, although the convective derivative $`๐ฎ\mathbf{}(mu_{}b^2)`$ along the drift-fluid velocity $`๐ฎ`$ appears to be absent in Eq. (33), it is actually hidden in the last term $`mnU_{}๐ป^{}\mathbf{}U_{}`$; this term may appear explictly in Eq. (25) by redefining the divergenceless vector (26) into a non-divergenceless vector as was done in Ref. SSB .
The set of drift-fluid equations of motion for the four drift-fluid moments $`(n,u_{},p_{},p_{})`$ are, thus, given by Eqs. (15),(33), and (16)-(17), respectively, while the drift-fluid velocity $`๐ฎ`$ is defined by Eq. (29). Note that each of the drift-fluid equations for $`(u_{},p_{},p_{})`$ involves the diamagnetic advection operator $`๐ฎ_D\mathbf{}`$, which must be cancelled by the addition of so-called gyroviscous cancellations (see Appendix B for further details). Furthermore, the drift-fluid moment equations now need to be complemented by suitable evolution equations for the electromagnetic scalar potentials $`(\varphi ,A_{})`$ through the low-frequency version of Maxwellโs equations, in which the net charge and current densities are expressed in terms of drift-fluid charge and current densities as well as polarization and magnetization terms.
### III.3 Poisson and parallel Ampรจre equations
We now derive the drift-fluid Poisson and parallel Ampรจre equations from the variational principle (13) as Euler-Lagrange equations for $`\psi ^i=(\varphi ,A_{})`$:
$$0=\frac{}{\psi ^i}\mathbf{}\frac{}{(\psi ^i)}.$$
(34)
where the drift-fluid polarization and magnetization effects, represented by the vectors (10)-(11), are introduced naturally in these respective equations.
The drift-fluid Poisson equation for $`\varphi `$ can be found from the Euler-Lagrange equation (34) for $`\psi ^i=\varphi `$, which can be expressed as
$`{\displaystyle \frac{\mathbf{}๐_{}}{4\pi }}`$ $`=`$ $`{\displaystyle en}+_{}\mathbf{}\left[{\displaystyle \frac{mnc^2}{B_0^2}\left(_{}\varphi \frac{u_{}}{c}_{}A_{}\right)}\right]`$ (35)
$``$ $`{\displaystyle en}\mathbf{}๐_{},`$
where summation over fluid species is shown explicitly. This equation shows the effects of the finite-$`\beta `$ electromagnetic generalization of the polarization density $`\rho _{\mathrm{pol}}=\mathbf{}๐_{}`$, where the drift-fluid polarization vector is defined in Eq. (10).
Next, the parallel drift-fluid Ampรจre equation can be found from the Euler-Lagrange equation (34) for $`\psi ^i=A_{}`$, which can be expressed as
$`{\displaystyle \frac{_{}^2A_{}}{4\pi }}`$ $`=`$ $`{\displaystyle \frac{en}{c}u_{}}+\mathbf{}\left[{\displaystyle \frac{mnu_{}c}{B_0^2}\left(_{}\varphi \frac{u_{}}{c}_{}A_{}\right)}\right]`$ (36)
$``$ $`{\displaystyle \frac{en}{c}u_{}}+\mathbf{}\left(๐_{}\mathbf{\times }\widehat{๐ป}_0\right),`$
where the absence of the parallel displacement current $`_tE_{}`$ results from the omission of $`E_{}^2`$ in the drift-fluid Lagrangian density (12). We also see that the second term in the divergence on the right side of the drift-fluid Ampรจre equation represents the effects of the parallel component of the magnetization current $`\widehat{๐ป}_0\mathbf{}๐_{\mathrm{mag}}c\widehat{๐ป}_0\mathbf{}\mathbf{\times }๐_{}`$, where the magnetization vector $`๐_{}`$ is defined in Eq. (11) and background magnetic nonuniformity is ignored. Since the polarization vector (10) is perpendicular to the background magnetic field, however, the drift-fluid polarization current $`๐_{\mathrm{pol}}๐_{}/t`$ has no parallel component ($`\widehat{๐ป}_0\mathbf{}๐_{\mathrm{pol}}0`$) and, thus, it does not appear in the parallel drift-fluid Ampรจre equation (36).
Lastly, we note that the covariant electromagnetic polarization and magnetization effects in the drift-fluid Poisson-Ampรจre equations are due to (a) the $`\varphi A_{}`$ symmetry of the nonlinear Lagrangian term $`mnu_{}V_{}`$ in the drift-fluid Lagrangian (5), and (b) the nonlinear kinetic energy terms $`mn(|๐ฎ_E|^2+U_{}^2)/2`$. While such low-frequency polarization and magnetization effects appear naturally in nonlinear finite-$`\beta `$ gyrokinetic theory HLB\_88 , they have not appeared explicitly in previous derivations of gyrofluid equations (e.g., Ref. brizard ).
### III.4 Quasineutral drift-fluid dynamics
A fundamental aspect of drift-fluid dynamics in a strongly magnetized plasma involves the fact that the time evolution of the turbulent plasma state maintains a condition of quasineutrality BDS\_2003 , where $`\rho =\mathbf{}๐_{}/4\pi 0`$ and, thus, according to the drift-fluid Poisson equation (35), the net drift-fluid charge density PCR\_1
$$en\mathbf{}๐_{}$$
(37)
is expressed in terms of the drift-fluid polarization vector (10). This drift-fluid quasineutrality condition, therefore, can be used with Eq. (35) to obtain the drift-fluid charge conservation law:
$$0=\mathbf{}\left(en๐ฎ+\frac{๐_{}}{t}\right)\mathbf{}๐,$$
(38)
where the net current density
$$๐en๐ฎ+\frac{๐_{}}{t}+\mathbf{\times }๐$$
(39)
is defined as the sum of the net drift-fluid current density ($`en๐ฎ`$), the drift-fluid polarization current ($`_t๐_{}`$), and a divergenceless term ($`\mathbf{\times }๐`$).
The vector field $`๐`$ in Eq. (39) is determined by considering the parallel drift-fluid Ampรจre equation (36) and introducing the definition for the net parallel current density:
$$J_{}\frac{c}{4\pi }_{}^2A_{}=enu_{}+c\widehat{๐ป}_0\mathbf{}\mathbf{\times }๐_{},$$
(40)
where the drift-fluid magnetization vector $`๐_{}`$ is defined in Eq. (11). Hence, we find $`๐c๐_{}`$ and the net current density (39) is now expressed as PCR\_1
$`๐`$ $`=`$ $`{\displaystyle en๐ฎ}+{\displaystyle \frac{๐_{}}{t}}+c\mathbf{\times }๐_{}`$ (41)
$`=`$ $`J_{}\widehat{๐ป}_0+{\displaystyle en๐ฎ_{}}+{\displaystyle \frac{๐_{}}{t}}+c(\mathbf{\times }๐_{})_{},`$
where we have introduced the net parallel current density (40) and $`(\mathbf{\times }๐_{})_{}`$ denotes perpendicular components of the drift-fluid magnetization current.
Lastly, we note that, while the standard polarization drift velocity (32) is absent from the drift-fluid velocity (29), the associated drift-fluid polarization current density $`_t๐_{}`$ is introduced naturally into our formalism through the drift-fluid quasineutrality condition (37) and the drift-fluid charge conservation law (38). Furthermore, we note that the drift-fluid polarization vector (10) can be expressed as $`๐_{}en๐ฐ_{}`$ so that the drift-fluid polarization current density
$$\frac{๐_{}}{t}\left[en\frac{d๐ฐ_{}}{dt}\mathbf{}\left(en๐ฎ๐ฐ_{}\right)\right]$$
can also be expressed in terms of the total time derivative $`d/dt/t+๐ฎ\mathbf{}`$ and the polarization velocity (32) now appears explicitly in the term $`d๐ฐ_{}/dt`$.
## IV Local and Global Drift-fluid Energy Conservation Laws
In this Section, we present the local and global forms of the energy conservation law, as they arise from an application of the Noether method Noether\_1 -Noether\_3 . For this purpose, we point out that, as a result of the variational principle $`\delta d^3x๐t=0`$, the only remaining terms in the Eulerian variation of the Lagrangian density (20) yield the Noether equation
$$\delta =\frac{\mathrm{\Lambda }}{t}+\mathbf{}๐ช,$$
(42)
where the Noether fields $`\mathrm{\Lambda }`$ and $`๐ช`$ are defined in Eqs. (22)-(23). The energy and momentum conservation laws are derived from the Noether equation (42) by considering infinitesimal time and space translations PCR\_1 , respectively. In the present work, we focus our attention on the local and global energy conservation laws associated with our drift-Alfvรฉn model.
### IV.1 Local energy conservation law
We derive the local form of the energy conservation law from the Noether equation (42) by considering infinitesimal time translations $`tt+\delta t`$, from which we obtain the following expressions for the virtual fluid displacement $`๐`$ and the Eulerian variations $`\delta \varphi `$ and $`\delta `$:
$$๐=๐ฎ\delta t\mathrm{and}(\delta \varphi ,\delta A_{},\delta )=\delta t(\frac{\varphi }{t},\frac{A_{}}{t},\frac{}{t}).$$
(43)
Inserting these expressions into Eqs. (22)-(23), the Noether fields become
$`\mathrm{\Lambda }`$ $``$ $`\delta t\left(๐ฎ\mathbf{}{\displaystyle \frac{}{๐ฎ}}\right),`$ (44)
$`๐ช`$ $``$ $`\delta t\left[๐ฎ\left(๐ฎ\mathbf{}{\displaystyle \frac{}{๐ฎ}}\eta ^a{\displaystyle \frac{}{\eta ^a}}\right)+๐ฏ\mathbf{}๐ฎ+{\displaystyle \frac{\psi ^i}{t}}{\displaystyle \frac{}{(\psi ^i)}}\right],`$ (45)
where summation over fluid species is, henceforth, implied wherever appropriate. By combining Eqs. (43)-(45), we obtain the primitive form of the local energy conservation law:
$$\frac{\epsilon ^{}}{t}+\mathbf{}๐^{}=\mathrm{\hspace{0.33em}0},$$
(46)
where the primitive energy density is
$`\epsilon ^{}`$ $``$ $`๐ฎ\mathbf{}{\displaystyle \frac{}{๐ฎ}}`$ (47)
$`=`$ $`{\displaystyle \frac{1}{2}}nmU_{}^2+\left(p_{}+{\displaystyle \frac{p_{}}{2}}\right)+en\mathrm{\Phi }^{}+{\displaystyle \frac{1}{8\pi }}\left(|_{}A_{}|^2|_{}\varphi |^2\right),`$
and the primitive energy-density flux is
$`๐^{}`$ $``$ $`๐ฎ\left(๐ฎ\mathbf{}{\displaystyle \frac{}{๐ฎ}}\eta ^a{\displaystyle \frac{}{\eta ^a}}\right)+๐ฏ\mathbf{}๐ฎ+{\displaystyle \frac{\psi ^i}{t}}{\displaystyle \frac{}{(\psi ^i)}}`$ (48)
$`=`$ $`\left[n\left({\displaystyle \frac{m}{2}}U_{}^2+e\mathrm{\Phi }^{}\right)+\left(p_{}+{\displaystyle \frac{p_{}}{2}}\right)\right]๐ฎ+๐ฏ\mathbf{}๐ฎ+{\displaystyle \frac{\psi ^i}{t}}{\displaystyle \frac{}{(\psi ^i)}}.`$
Note that the energy conservation law (46) has the following gauge-invariance property: the energy conservation law is unaffected by the transformation
$$\epsilon =\epsilon ^{}+\mathbf{}๐\mathrm{and}๐=๐^{}\frac{๐}{t},$$
(49)
where $`๐`$ is an arbitrary vector field.
In order to arrive at the final form of the local energy conservation law, we need to rearrange some terms in Eq. (47). First, by substituting the drift-fluid Poisson equation (35), we find
$$en\mathrm{\Phi }^{}\frac{|\varphi |^2}{8\pi }=\frac{|\varphi |^2}{8\pi }+\frac{mn}{2}|๐ฎ_E|^2mnu_{}V_{}\mathbf{}๐,$$
(50)
where the gauge vector field $`๐`$ is defined as
$$๐\varphi \frac{}{(\varphi )}=\varphi \left(\frac{๐_{}}{4\pi }+๐_{}\right).$$
(51)
Next, we express the primitive energy density (47) as $`\epsilon ^{}\epsilon \mathbf{}๐`$, where the final form of the energy density is defined as
$$\epsilon \frac{1}{2}mn\left|u_{}(\widehat{๐ป}_0+๐_{}/B_0)+๐ฎ_E\right|^2+\left(p_{}+\frac{p_{}}{2}\right)+\frac{1}{8\pi }\left(|๐_{}|^2+|๐_{}|^2\right).$$
(52)
The final form of the local energy conservation law is, therefore, expressed as
$$\frac{\epsilon }{t}+\mathbf{}๐=\mathrm{\hspace{0.33em}0},$$
(53)
where the final form of the energy density flux, defined as
$$๐๐^{}\frac{๐}{t},$$
(54)
is expressed, after some partial cancelations, as
$$๐=\left[n\left(\frac{m}{2}U_{}^2+e\mathrm{\Phi }^{}\right)+\left(p_{}+\frac{p_{}}{2}\right)\right]๐ฎ+๐ฏ\mathbf{}๐ฎ+๐_\phi ,$$
(55)
where the field energy density flux $`๐_\phi `$ is defined as
$$๐_\phi \varphi \frac{}{t}\left(\frac{}{(\varphi )}\right)+\frac{A_{}}{t}\frac{}{(A_{})}.$$
(56)
In Appendix B, after gyroviscous diamagnetic cancellations are inserted into the drift-fluid equation for $`(u_{},p_{},p_{})`$, the local energy conservation law (53) is converted into a new energy equation
$$\frac{\epsilon }{t}+\mathbf{}๐^{}=\mathrm{\hspace{0.33em}0},$$
(57)
in which diamagnetic-cancellation terms result in a heat-modified energy density flux $`๐^{}`$ \[see Eq. (91)\]. This new form ensures that the total energy $`=\epsilon d^3x`$ still satisfies the global energy conservation law $`d/dt=0`$.
### IV.2 Global energy conservation law
The global energy conservation law
$$0=\frac{d}{dt}\frac{d}{dt}(_u+_p+_\varphi +_A),$$
(58)
can be derived from the local energy conservation law (53) by integrating it over space (and neglecting surface terms), where the component energies
$`_u`$ $`=`$ $`{\displaystyle d^3x\left(\frac{mn}{2}U_{}^2\right)},`$ (59)
$`_p`$ $`=`$ $`{\displaystyle d^3x\left(p_{}+\frac{p_{}}{2}\right)},`$ (60)
$`_\varphi `$ $`=`$ $`{\displaystyle d^3x\left(\frac{|\varphi |^2}{8\pi }+\frac{mn}{2}|๐ฎ_E|^2mnu_{}V_{}\right)},`$ (61)
$`_A`$ $`=`$ $`{\displaystyle d^3x\left(\frac{|_{}A_{}|^2}{8\pi }\right)},`$ (62)
represent the parallel kinetic energy, the internal energy, and the electromagnetic field energy (defined as the sum of the $`\varphi `$-field energy $`_\varphi `$ and the $`A_{}`$-field energy $`_A`$), respectively. We now present the time evolution of each energy component (in the absence of diamagnetic cancellations and heat fluxes) in order to identify the energy-exchange processes that allow the transfer of energy between the three types of drift-fluid (parallel kinetic, internal, electromagnetic field) energies.
First, we begin with the time derivative of the parallel kinetic energy density
$`{\displaystyle \frac{}{t}}\left({\displaystyle \frac{mn}{2}}U_{}^2\right)`$ $`=`$ $`\mathbf{}\left[\left({\displaystyle \frac{mn}{2}}U_{}^2\right)๐ฎ\right]๐ฎ\mathbf{}\left(\mathbf{}๐ฏ+en\mathrm{\Phi }^{}\right)`$ (63)
$`nu_{}\left(mu_{}b{\displaystyle \frac{b}{t}}+{\displaystyle \frac{e}{c}}{\displaystyle \frac{A_{}^{}}{t}}\right),`$
derived from Eq. (33) after rearranging some terms and using the continuity equation (15). Hence, the time evolution of the parallel-kinetic energy is
$$\frac{d_u}{dt}=d^3x\left[๐ฎ\mathbf{}\left(\mathbf{}๐ฏ+en\mathrm{\Phi }^{}\right)+nu_{}\left(mu_{}b\frac{b}{t}+\frac{e}{c}\frac{A_{}^{}}{t}\right)\right].$$
(64)
Next, we evaluate the time derivative of the internal energy density $`๐ซp_{}+p_{}/2`$ using the parallel and perpepdicular CGL pressure equations (16)-(17):
$$\frac{d๐ซ}{dt}+๐ซ\mathbf{}๐ฎ+๐ฏ:๐ฎ=\mathrm{\hspace{0.33em}0},$$
so that, by rearranging terms, we obtain
$$\frac{๐ซ}{t}=\mathbf{}\left(๐ซ๐ฎ+๐ฏ\mathbf{}๐ฎ\right)+๐ฎ\mathbf{}\left(\mathbf{}๐ฏ\right),$$
(65)
and the time evolution of the internal energy is
$$\frac{d_p}{dt}=d^3x๐ฎ\mathbf{}\left(\mathbf{}๐ฏ\right).$$
(66)
Thirdly, we evaluate the time derivative of the $`A_{}`$-field energy density
$`{\displaystyle \frac{}{t}}\left({\displaystyle \frac{|_{}A_{}|^2}{8\pi }}\right)`$ $`=`$ $`{\displaystyle \frac{1}{4\pi }}{\displaystyle \frac{A_{}}{t}}_{}^2A_{}+\mathbf{}\left({\displaystyle \frac{A_{}}{t}}{\displaystyle \frac{_{}A_{}}{4\pi }}\right)`$ (67)
$`=`$ $`\mathbf{}\left({\displaystyle \frac{A_{}}{t}}{\displaystyle \frac{}{A_{}}}\right)+{\displaystyle \frac{en}{c}}u_{}{\displaystyle \frac{A_{}}{t}}`$
$`{\displaystyle \frac{mnu_{}c}{B_0^2}}\left(_{}\varphi {\displaystyle \frac{u_{}}{c}}_{}A_{}\right)\mathbf{}\left(_{}{\displaystyle \frac{A_{}}{t}}\right),`$
where the second equality is obtained by substituting the drift-fluid Ampรจre equation (36), so that the time evolution of the $`A_{}`$-field energy is
$$\frac{d_A}{dt}=d^3x\left[nu_{}\left(mu_{}b\frac{b}{t}+\frac{e}{c}\frac{A_{}}{t}\right)\frac{mnu_{}c}{B_0^2}_{}\varphi \mathbf{}\left(_{}\frac{A_{}}{t}\right)\right].$$
(68)
Lastly, we evaluate time derivative of the $`\varphi `$-field energy density
$`{\displaystyle \frac{}{t}}\left({\displaystyle \frac{|_{}\varphi |^2}{8\pi }}+{\displaystyle \frac{mn}{2}}|๐ฎ_E|^2mnu_{}V_{}\right)`$ $`=`$ $`{\displaystyle \frac{}{t}}\left(en\mathrm{\Phi }^{}{\displaystyle \frac{|_{}\varphi |^2}{8\pi }}+\mathbf{}๐\right)`$ (69)
$`=`$ $`\mathbf{}\left({\displaystyle \frac{๐}{t}}๐ฎen\mathrm{\Phi }^{}\right)+๐ฎ\mathbf{}\left(en\mathrm{\Phi }^{}\right)`$
$`+\left(en{\displaystyle \frac{\mathrm{\Phi }^{}}{t}}{\displaystyle \frac{_{}\varphi }{4\pi }}\mathbf{}_{}{\displaystyle \frac{\varphi }{t}}\right),`$
where the first equality is obtained from the definition (50) while the second equality arises from the use of the continuity equation (15). By using the definition
$`en{\displaystyle \frac{\mathrm{\Phi }^{}}{t}}{\displaystyle \frac{_{}\varphi }{4\pi }}\mathbf{}_{}{\displaystyle \frac{\varphi }{t}}`$ $`=`$ $`{\displaystyle \frac{\varphi }{t}}\left[en+\mathbf{}\left({\displaystyle \frac{\varphi }{4\pi }}+{\displaystyle \frac{mnc^2}{B_0^2}}_{}\varphi \right)\right]`$
$`\mathbf{}\left[{\displaystyle \frac{\varphi }{t}}\left({\displaystyle \frac{_{}\varphi }{4\pi }}+{\displaystyle \frac{mnc^2}{B_0^2}}_{}\varphi \right)\right]`$
$`=`$ $`\mathbf{}\left({\displaystyle \frac{\varphi }{t}}{\displaystyle \frac{}{\varphi }}\right)mnu_{}\left({\displaystyle \frac{c}{B_0^2}}_{}{\displaystyle \frac{\varphi }{t}}\right)\mathbf{}_{}A_{},`$
so that Eq. (69) becomes
$`{\displaystyle \frac{}{t}}\left({\displaystyle \frac{|_{}\varphi |^2}{8\pi }}+{\displaystyle \frac{mn}{2}}|๐ฎ_E|^2mnu_{}V_{}\right)=\mathbf{}\left[\varphi {\displaystyle \frac{}{t}}\left({\displaystyle \frac{}{\varphi }}\right)en\mathrm{\Phi }^{}๐ฎ\right]`$
$`+๐ฎ\mathbf{}\left(en\mathrm{\Phi }^{}\right)mnu_{}\left({\displaystyle \frac{c}{B_0^2}}_{}{\displaystyle \frac{\varphi }{t}}\right)\mathbf{}_{}A_{},`$ (70)
and the time evolution of the $`\varphi `$-field energy is
$$\frac{d_\varphi }{dt}=d^3x\left[๐ฎ\mathbf{}\left(en\mathrm{\Phi }^{}\right)mnu_{}\left(\frac{c}{B_0^2}_{}\frac{\varphi }{t}\right)\mathbf{}_{}A_{}\right].$$
(71)
By combining Eqs. (68) and (71), and using the expression
$$\frac{V_{}}{t}=\frac{c}{B_0^2}\left(_{}\frac{\varphi }{t}\mathbf{}_{}A_{}+_{}\varphi \mathbf{}_{}\frac{A_{}}{t}\right)$$
with the definition (6) for $`A_{}^{}`$, we obtain the time evolution of the electromagnetic energy $`_\psi =_\varphi +_A`$:
$$\frac{d_\psi }{dt}=d^3x\left[en๐ฎ\mathbf{}\mathrm{\Phi }^{}+nu_{}\left(mu_{}b\frac{b}{t}+\frac{e}{c}\frac{A_{}^{}}{t}\right)\right].$$
(72)
We now see that by combining Eqs. (64), (66), and (72), the global energy conservation law (58) is satisfied exactly. Hence, the energy-transfer processes (64), (66), and (72) between the parallel kinetic energy $`_u`$, internal energy $`_p`$, and electromagnetic field energy $`_\psi `$ involve the following pathways:
$$(p_{},p_{})(n,u_{})(\varphi ,A_{}),$$
(73)
where Eqs. (64) and (66), for example, show that internal energy $`_p`$ is transferred to and from parallel kinetic energy $`_u`$.
## V Summary
The nonlinear drift-Alfvรฉn equations (15), (33), and (16)-(17) for the anisotropic drift-fluid moments $`(n,u_{},p_{},p_{})`$, respectively, and the drift-fluid Poisson-Ampรจre equations (35) and (36) were derived in this paper from a Lagrangian variational principle based on the drift-fluid Lagrangian density (12). The local and global forms of the exact energy conservation law for these drift-Alfvรฉn equations, given by Eqs. (53) and (58), respectively, were derived by the Noether method as a natural consequence of the variational formulation presented here.
The inclusion of the linear and nonlinear $`E\times B`$ velocities (2)-(3) as well as the parallel flow along the total magnetic field lines into the drift-fluid Lagrangian density (12) enabled a derivation of generalized (electromagnetic) drift-fluid polarization and magnetization effects. Under the drift-fluid quasineutrality condition, the drift-fluid Poisson-Ampรจre equations (35) and (36) were transformed into the drift-fluid charge conservation law (38), with the net current density $`๐`$ defined by Eq. (41), which represents a generalized (electromagnetic) form of the so-called vorticity equation.
Although higher-order drift-fluid moments have been omitted from the present variational formulation of our four-moment nonlinear anisotropic drift-Alfvรฉn model, Appendix A shows how higher-order heat fluxes might be introduced into a variational formulation, while Appendix B shows how diamagnetic cancellations in the drift-fluid equations for $`(u_{},p_{},p_{})`$ can be incorporated without affected the global energy conservation law. Future work will investigate the extension of variational principles by including higher-order heat fluxes as dynamical variational fields as well as the extension of Generalized Lagrangian Mean (GLM) methods GLM to plasma fluid models.
###### Acknowledgements.
The Author wishes to acknowledge Drs. Bruce Scott, T. S. Hahm, and R. E. Denton, for their useful comments and continued interest in this work. The present work was supported by the National Science Foundation under grant No. DMS-0317339.
## Appendix A Notes on Macroscopic Plasma Lagrangians
In this Appendix, we present an adapted version of Y.-K. M. Pengโs Ph. D. dissertation Peng\_PhD (with selected parts also published in Ref. Peng\_Crawford ). The macroscopic Lagrangian presented by Peng is based on constrained variations of the fluid density, velocity, and pressure tensor, and includes the effects of finite heat fluxes self-consistently.
### A.1 Lagrangian Density in Eulerian Coordinates
Following Pengโs work, we begin by combining the pressure constraint equations (16)-(17) to form the scalar pressure equation for $`๐ซp_{}+p_{}/2`$:
$$\frac{d๐ซ}{dt}+๐ซ\mathbf{}๐ฎ+๐ฏ:๐ฎ=\mathbf{}๐ช,$$
(74)
where the heat flux $`๐ช`$ now appears on the right side of Eq. (74). Next, the heat flux vector $`๐ช`$ is represented as
$$๐ช\frac{d๐}{dt}=\frac{๐}{t}+๐ฎ\mathbf{}๐,$$
(75)
where $`๐`$ denotes the total density of heat energy transported across the boundary of the macroscopic (coarse-graining) cell as it moves along its Lagrangian trajectory Peng\_PhD ; Peng\_Crawford . Hence, the right side of Eq. (74) can be written as
$$\mathbf{}๐ช\mathbf{}\frac{d๐}{dt}=\frac{d\mathbf{}๐}{dt}๐ฎ:๐,$$
so that Eq. (74) may be rewritten as
$$\frac{d}{dt}\left(๐ซ+\mathbf{}๐\right)=๐ซ\mathbf{}๐ฎ\left(๐ฏ+๐\right):๐ฎ.$$
(76)
By using the Lagrangian-variation limitting procedure (18), we, thus, obtain the Lagrangian variation
$$\mathrm{\Delta }\left(๐ซ+\mathbf{}๐\right)=๐ซ\mathbf{}๐\left(๐ฏ+๐\right):๐,$$
(77)
from which we obtain the Eulerian variation
$$\delta \left(๐ซ+\mathbf{}๐\right)=\mathbf{}\left(๐ซ๐+๐\mathbf{}๐\right)๐ฏ:๐.$$
(78)
Lastly, the heat flux $`๐ช`$ is introduced in the Lagrangian density $``$, defined by Eq. (12), by adding the divergence term $`\mathbf{}๐`$:
$$^{}\mathbf{}๐_{cold}\left(๐ซ+\mathbf{}๐\right),$$
(79)
where $`_{cold}`$ denotes the cold-fluid part of the drift-fluid Lagrangian density (12).
### A.2 Eulerian Variational Principle
We now write the expression for the Eulerian variation of the heat-modified Lagrangian density (79):
$$\delta ^{}=\delta _{cold}+\mathbf{}\left(๐ซ๐+๐\mathbf{}๐\right)+๐ฏ:๐\delta +\mathbf{}\left(๐\mathbf{}๐\right),$$
(80)
where the Eulerian variation $`\delta `$ is given by the Noether equation (42). We note that the quantity $`๐`$ does not play a dynamic role since it appears inside a spatial divergence. We shall see, however, that it plays an important role in terms of the energy conservation law derived from the heat-modified Noether equation
$$\delta ^{}=\frac{\mathrm{\Lambda }}{t}+\mathbf{}\left(๐ช+๐\mathbf{}๐\right).$$
(81)
### A.3 Energy Conservation Law by Noether Method
The invariance of the Lagrangian density (79) under time translations $`tt+\delta t`$ yields an energy equation derived from the Noether equation (81). By applying the Noether method (as outlined in Sec. IV), we obtain the heat-modified energy equation:
$$0=\frac{}{t}\left(\epsilon +\mathbf{}๐\right)+\mathbf{}\left(๐+๐ฎ\mathbf{}๐\right)\frac{\epsilon }{t}+\mathbf{}\left(๐+\frac{d๐}{dt}\right),$$
(82)
where the energy density $`\epsilon `$ and energy-density flux $`๐`$ are defined in Eqs. (52) and (55), respectively. Lastly, using the definition (75) for the heat flux $`๐ช`$, we obtain the energy equation
$$\frac{\epsilon }{t}+\mathbf{}๐=\mathbf{}๐ช.$$
(83)
Although the heat flux $`๐ช`$ appears self-consistently within Pengโs variational principle, we note that the evolution equation for $`๐ช`$ is outside of its scope. It is a topic of future research to determine whether the evolution equation for the heat flux can be derived within a variational formulation.
## Appendix B Diamagnetic cancelations and Energy conservation
The drift-fluid equations (33) and (16)-(17) contain diamagnetic advective derivatives $`๐ฎ_D\mathbf{}\eta ^a`$ of the drift-fluid moments $`\eta ^a=(u_{},p_{},p_{})`$ that must be cancelled by the addition of so-called higher-order gyroviscous cancellations. Since these higher-order moments cannot be derived from the drift-fluid Lagrangian, the diamagnetic cancelations must be inserted in drift-fluid equations in the post-variational phase. However, there exists a constraint in the addition of higher-order moment terms, namely, that the global energy conservation law should not be altered.
### B.1 Parallel drift-fluid dynamics
The diamagnetic cancelation needed for the parallel momentum equation (33) involves the addition of the term $`\mathbf{}๐ท_{}^{}`$ on the right side of Eq. (33), associated with the non-diagonal part of the pressure tensor. From Vlasov theory, the diamagnetic-cancelation term is found to be Braginskii ; HM ; Chang\_Callen :
$$\mathbf{}๐ท_{}^{}=\mathbf{}\left(p_{}\frac{mc\widehat{๐ป}_0}{eB_{}^{}}\mathbf{\times }u_{}\right)=mn๐ฎ_D\mathbf{}u_{}+p_{}๐ฆ(mu_{}),$$
(84)
where the term $`B_{}^{}`$ replaces the denominator $`B_0`$ to ensure an exact diamagnetic cancellation and the magnetic differential operator $`๐ฆ(\mathrm{})`$ is defined by the identity
$$\mathbf{}\left(g\frac{c\widehat{๐ป}_0}{eB_{}^{}}\mathbf{\times }f\right)=\frac{c\widehat{๐ป}_0}{eB_0}\mathbf{}f\mathbf{\times }gg๐ฆ(f),$$
(85)
valid for arbitrary functions $`f`$ and $`g`$.
Note that the diamagnetic-cancellation term (84) is energy conserving since
$$u_{}\mathbf{}๐ท_{}^{}\mathbf{}\left(u_{}๐ท_{}^{}\right)$$
and, thus, the time evolution of the parallel kinetic energy density (63) is now expressed as
$`{\displaystyle \frac{}{t}}\left({\displaystyle \frac{mn}{2}}U_{}^2\right)`$ $`=`$ $`\mathbf{}\left({\displaystyle \frac{mn}{2}}U_{}^2๐ฎ+u_{}๐ท_{}^{}\right)๐ฎ\mathbf{}\left(\mathbf{}๐ฏ+en\mathrm{\Phi }^{}\right)`$ (86)
$`nu_{}\left(mu_{}b{\displaystyle \frac{b}{t}}+{\displaystyle \frac{e}{c}}{\displaystyle \frac{A_{}^{}}{t}}\right),`$
and the global time evolution equation (64) for $`_u`$ is still valid.
### B.2 Internal energy
To consider the diamagnetic cancellations in the anisotropic pressure equations (16)-(17), we write the modified CGL pressure equations weiland :
$`{\displaystyle \frac{dp_{}}{dt}}+p_{}\mathbf{}๐ฎ+p_{}(๐\widehat{๐ป}_0\widehat{๐ป}_0):๐ฎ`$ $`=`$ $`\mathbf{}๐ช_{}^{()}\mathrm{\hspace{0.33em}2}๐ช_{}^{()}\mathbf{}(\widehat{๐ป}_0\mathbf{}\widehat{๐ป}_0),`$ (87)
$`{\displaystyle \frac{dp_{}}{dt}}+p_{}\mathbf{}๐ฎ+\mathrm{\hspace{0.33em}2}p_{}\widehat{๐ป}_0\widehat{๐ป}_0:๐ฎ`$ $`=`$ $`\mathrm{\hspace{0.33em}2}\mathbf{}๐ช_{}^{()}+\mathrm{\hspace{0.33em}4}๐ช_{}^{()}\mathbf{}(\widehat{๐ป}_0\mathbf{}\widehat{๐ป}_0),`$ (88)
which have been modified by the addition of the parallel and perpendicular gyroviscous heat fluxes $`๐ช_{}^{()}`$ and $`๐ช_{}^{()}`$, respectively, derived from the Vlasov equation directly and found to be
$$๐ช_{}^{()}=\frac{1}{2}\frac{cp_{}\widehat{๐ป}_0}{eB_{}^{}}\mathbf{\times }T_{}+p_{}๐ฎ_C\mathrm{and}๐ช_{}^{()}=\mathrm{\hspace{0.33em}2}\frac{cp_{}\widehat{๐ป}_0}{eB_{}^{}}\mathbf{\times }T_{}.$$
(89)
Inserting these diamagnetic heat fluxes into Eqs. (87)-(88), the time evolution of the internal energy can, therefore, be expressed as
$$\frac{๐ซ}{t}=\mathbf{}\left(๐ซ๐ฎ+๐ฏ\mathbf{}๐ฎ+๐ช\right)+๐ฎ\mathbf{}\left(\mathbf{}๐ฏ\right),$$
(90)
where the net heat flux is $`๐ช๐ช_{}^{()}+๐ช_{}^{()}`$. Hence, we see that the insertion of diamagnetic cancellations into our finite-$`\beta `$ electromagnetic drift-fluid model does not jeopardize the local and global energy conservation laws, where the energy density flux (55) is now replaced by the flux
$$๐^{}=๐+u_{}๐ท_{}^{}+๐ช_{}^{()}+๐ช_{}^{()}.$$
(91)
|
warning/0506/math0506311.html
|
ar5iv
|
text
|
# Renormalization analysis of catalytic Wright-Fisher diffusions
## 1 Introduction and main result
Several authors \[BCGdH95, BCGdH97, dHS98, Sch98, CDG04\] have studied maps where a function, describing the local diffusion matrix of a diffusion process, is mapped into the average of that function with respect to the unique invariant measure of the diffusion process itself. Such mappings arise in the analysis of infinite systems of diffusion processes indexed by the hierarchical group, with a linear attractive interaction between the components \[DG93a, DG96, DGV95\]. In this context, the mappings are called renormalization transformations. We follow this terminology. For more on the relation between hierarchically interacting diffusions and renormalization transformations, see Appendix A.1.
Formally, such renormalization transformations can be defined as follows.
###### Definition 1.1
(Renormalization class and transformation) Let $`D^d`$ be nonempty, convex, and open. Let $`๐ฒ`$ be a collection of continuous functions $`w`$ from the closure $`\overline{D}`$ into the space $`M_+^d`$ of symmetric non-negative definite $`d\times d`$ real matrices, such that $`\lambda w๐ฒ`$ for every $`\lambda >0`$, $`w๐ฒ`$. We call $`๐ฒ`$ a prerenormalization class on $`\overline{D}`$ if the following three conditions are satisfied:
1. For each constant $`c>0`$, $`w๐ฒ`$, and $`x\overline{D}`$, the martingale problem for the operator $`A_x^{c,w}`$ is well-posed, where
$$A_x^{c,w}f(y):=\underset{i=1}{\overset{d}{}}c(x_iy_i)\frac{}{y_i}f(y)+\underset{i,j=1}{\overset{d}{}}w_{ij}(y)\frac{^2}{y_iy_j}f(y)(y\overline{D}),$$
(1.1)
and the domain of $`A_x^{c,w}`$ is the space of real functions on $`\overline{D}`$ that can be extended to a twice continuously differentiable function on $`^d`$ with compact support.
2. For each $`c>0`$, $`w๐ฒ`$, and $`x\overline{D}`$, the martingale problem for $`A_x^{c,w}`$ has a unique stationary solution with invariant law denoted by $`\nu _x^{c,w}`$.
3. For each $`c>0`$, $`w๐ฒ`$, $`x\overline{D}`$, and $`i,j=1,\mathrm{},d`$, one has $`{\displaystyle _{\overline{D}}}\nu _x^{c,w}(\mathrm{d}y)|w_{ij}(y)|<\mathrm{}`$.
If $`๐ฒ`$ is a prerenormalization class, then we define for each $`c>0`$ and $`w๐ฒ`$ a matrix-valued function $`F_cw`$ on $`\overline{D}`$ by
$$F_cw(x):=_{\overline{D}}\nu _x^{c,w}(dy)w(y)(x\overline{D}).$$
(1.2)
We say that $`๐ฒ`$ is a renormalization class on $`\overline{D}`$ if in addition:
1. For each $`c>0`$ and $`w๐ฒ`$, the function $`F_cw`$ is an element of $`๐ฒ`$.
If $`๐ฒ`$ is a renormalization class and $`c>0`$, then the map $`F_c:๐ฒ๐ฒ`$ defined by (1.2) is called the renormalization transformation on $`๐ฒ`$ with migration constant $`c`$. In (1.1), $`w`$ is called the diffusion matrix and $`x`$ the attraction point. $`\mathrm{}`$
###### Remark 1.2
(Associated SDE) It is well-known that $`\overline{D}`$-valued (weak) solutions $`๐ฒ=(๐ฒ^1,\mathrm{},๐ฒ^d)`$ to the stochastic differential equation (SDE)
$$\mathrm{d}๐ฒ_t^i=c(x_i๐ฒ_t^i)\mathrm{d}t+\sqrt{2}\underset{j=1}{\overset{n}{}}\sigma _{ij}(๐ฒ_t)\mathrm{d}B_t^j(t0,i=1,\mathrm{},d),$$
(1.3)
where $`B=(B^1,\mathrm{},B^n)`$ is $`n`$-dimensional (standard) Brownian motion ($`n1`$), solve the martingale problem for $`A_x^{c,w}`$ if the $`d\times n`$ matrix-valued function $`\sigma `$ is continuous and satisfies $`_k\sigma _{ik}\sigma _{jk}=w_{ij}`$. Conversely \[EK86, Theorem 5.3.3\], every solution to the martingale problem for $`A_x^{c,w}`$ can be represented as a solution to the SDE (1.3), where there is some freedom in the choice of the root $`\sigma `$ of the diffusion matrix $`w`$. $`\mathrm{}`$
In the present paper, we concern ourselves with the following renormalization class on $`[0,1]^2`$.
###### Definition 1.3
(Renormalization class of catalytic Wright-Fisher diffusions) We set $`๐ฒ_{\mathrm{cat}}:=\{w^{\alpha ,p}:\alpha >0,p\}`$, where
$$w^{\alpha ,p}(x):=\left(\begin{array}{cc}\alpha x_1(1x_1)& 0\\ 0& p(x_1)x_2(1x_2)\end{array}\right)(x=(x_1,x_2)[0,1]^2),$$
(1.4)
and
$$:=\{p:p\text{ a real function on }[0,1],p0,p\text{ Lipschitz continuous}\}.$$
(1.5)
Moreover, we put
$$_{l,r}:=\{p:\mathrm{\hspace{0.33em}1}_{\{p(0)>0\}}=l,1_{\{p(1)>0\}}=r\}(l,r=0,1),$$
(1.6)
and set $`๐ฒ_{\mathrm{cat}}^{l,r}:=\{w^{\alpha ,p}:\alpha >0,p_{l,r}\}(l,r=0,1)`$. $`\mathrm{}`$
By Remark 1.2, solutions $`๐ฒ=(๐ฒ^1,๐ฒ^2)`$ to the martingale problem for $`A_x^{c,w^{\alpha ,p}}`$ can be represented as solutions to the SDE
$$\begin{array}{cccc}\hfill (\mathrm{i})& \hfill \mathrm{d}๐ฒ_t^1& =& c(x_1๐ฒ_t^1)\mathrm{d}t+\sqrt{2\alpha ๐ฒ_t^1(1๐ฒ_t^1)}\mathrm{d}B_t^1,\hfill \\ \hfill (\mathrm{ii})& \hfill \mathrm{d}๐ฒ_t^2& =& c(x_2๐ฒ_t^2)\mathrm{d}t+\sqrt{2p(๐ฒ_t^1)๐ฒ_t^2(1๐ฒ_t^2)}\mathrm{d}B_t^2.\hfill \end{array}$$
(1.7)
We call $`๐ฒ^1`$ the Wright-Fisher catalyst with resampling rate $`\alpha `$ and $`๐ฒ^2`$ the Wright-Fisher reactant with catalyzing function $`p`$.
For any renormalization class $`๐ฒ`$ and any sequence of (strictly) positive migration constants $`(c_k)_{k0}`$, we define iterated renormalization transformations $`F^{(n)}:๐ฒ๐ฒ`$, as follows:
$$F^{(n+1)}w:=F_{c_n}(F^{(n)}w)(n0)\text{with}F^{(0)}w:=w(w๐ฒ_{\mathrm{cat}}).$$
(1.8)
We set $`s_0:=0`$ and
$$s_n:=\underset{k=0}{\overset{n1}{}}\frac{1}{c_k}(1n\mathrm{}).$$
(1.9)
Here is our main result:
###### Theorem 1.4
(Main result)
(a) The set $`๐ฒ_{\mathrm{cat}}`$ is a renormalization class on $`[0,1]^2`$ and $`F_c(๐ฒ_{\mathrm{cat}}^{l,r})๐ฒ_{\mathrm{cat}}^{l,r}`$ $`(c>0,l,r=0,1)`$.
(b) Fix (positive) migration constants $`(c_k)_{k0}`$ such that
$$(\mathrm{i})s_n\underset{n\mathrm{}}{}\mathrm{}\text{and}(\mathrm{ii})\frac{s_{n+1}}{s_n}\underset{n\mathrm{}}{}1+\gamma ^{}$$
(1.10)
for some $`\gamma ^{}0`$. If $`w๐ฒ_{\mathrm{cat}}^{l,r}`$ $`(l,r=0,1)`$, then uniformly on $`[0,1]^2`$,
$$s_nF^{(n)}w\underset{n\mathrm{}}{}w^{},$$
(1.11)
where the limit $`w^{}`$ is the unique solution in $`๐ฒ_{\mathrm{cat}}^{l,r}`$ to the equation
$$\begin{array}{ccccc}\hfill (\mathrm{i})& \hfill (1+\gamma ^{})F_{1/\gamma ^{}}w^{}& =& w^{}\hfill & \text{if}\gamma ^{}>0,\hfill \\ \hfill (\mathrm{ii})& \hfill \frac{1}{2}\underset{i,j=1}{\overset{2}{}}w_{ij}^{}(x)\frac{^2}{x_ix_j}w^{}(x)+w^{}(x)& =& 0(x[0,1]^2)\hfill & \text{if}\gamma ^{}=0.\hfill \end{array}$$
(1.12)
(c) The matrix $`w^{}`$ is of the form $`w^{}=w^{1,p^{}}`$, where $`p^{}=p_{l,r,\gamma ^{}}^{}_{l,r}`$ depends on $`l,r,`$ and $`\gamma ^{}`$. One has
$$p_{0,0,\gamma ^{}}^{}0\text{and}p_{1,1,\gamma ^{}}^{}1\text{ for all }\gamma ^{}0.$$
(1.13)
For each $`\gamma ^{}0`$, the function $`p_{0,1,\gamma ^{}}^{}`$ is concave, nondecreasing, and satisfies $`p_{0,1,\gamma ^{}}^{}(0)=0`$, $`p_{0,1,\gamma ^{}}^{}(1)=1`$. By symmetry, analoguous statements hold for $`p_{1,0,\gamma ^{}}^{}`$.
Conditions (1.10) (i) and (ii) are satisfied, for example, for $`c_k=(1+\gamma ^{})^k`$. Note that the functions $`p_{0,0,\gamma ^{}}^{}`$ and $`p_{1,1,\gamma ^{}}^{}`$ are independent of $`\gamma ^{}0`$. We believe that on the other hand, $`p_{0,1,\gamma ^{}}^{}`$ is not constant as a function of $`\gamma ^{}`$, but we have not proved this.
The function $`p_{0,1,0}^{}`$ is the unique nonnegative solution to the equation
$$\frac{1}{2}x(1x)\frac{^2}{x^2}p(x)+p(x)(1p(x))=0(x[0,1])$$
(1.14)
with boundary conditions $`p(0)=0`$ and $`p(1)>0`$. This function occurred before in the work of Greven, Klenke, and Wakolbinger \[GKW01, formulas (1.10)โ(1.11)\]. In Section 4.1 we discuss the relation between their work and ours.
Outline In Part I of the paper (Sections 14) we present our results and our main techniques for proving them. Part II (Sections 59) contains detailed proofs. Since the motivation for studying renormalization classes comes from the study of linearly interacting diffusions on the hierarchical group, we explain this connection in Appendix A.
Outline of Part I In the next section, we place our main result in a broader context. We give a more thorough introduction to the theory of renormalization classes on compact sets and discuss earlier results on this topic. In Section 3, we discuss special properties of the renormalization class $`๐ฒ_{\mathrm{cat}}`$ from Definition 1.3. In particular, we show how techniques from the theory of spatial branching processes can be used to prove Theorem 1.4. In Section 4 we discuss the relation of our work with that in \[GKW01\] and mention some open problems.
Notation If $`E`$ is a separable, locally compact, metrizable space, then $`๐(E)`$ denotes the space of continuous real functions on $`E`$. If $`E`$ is compact then we equip $`๐(E)`$ with the supremumnorm $`_{\mathrm{}}`$. We let $`B(E)`$ denote the space of all bounded Borel measurable real functions on $`E`$. We write $`๐_+(E)`$ and $`๐_{[0,1]}(E)`$ for the spaces of all $`f๐(E)`$ with $`f0`$ and $`0f1`$, respectively, and define $`B_+(E)`$ and $`B_{[0,1]}(E)`$ analogously. We let $`(E)`$ denote the space of all finite measures on $`E`$, equipped with the topology of weak convergence. The subspaces of probability measures is denoted by $`_1(E)`$. We write $`๐ฉ(E)`$ for the space of finite counting measures, i.e., measures of the form $`\nu =_{i=1}^m\delta _{x_i}`$ with $`x_1,\mathrm{},x_mE`$ ($`m0`$). We interpret $`\nu `$ as a collection of particles, situated at positions $`x_1,\mathrm{},x_m`$. For $`\mu (E)`$ and $`fB(E)`$ we use the notation $`\mu ,f:=_Efd\mu `$ and $`|\mu |:=\mu (E)`$. By definition, $`๐_E[0,\mathrm{})`$ is the space of cadlag functions $`w:[0,\mathrm{})E`$, equipped with the Skorohod topology. We denote the law of a random variable $`y`$ by $`(y)`$. If $`๐ฒ=(๐ฒ_t)_{t0}`$ is a Markov process in $`E`$ and $`xE`$, then $`P^x`$ denotes the law of $`๐ฒ`$ started in $`๐ฒ_0=x`$. If $`\mu `$ is a probability law on $`E`$ then $`P^\mu `$ denotes the law of $`๐ฒ`$ started with initial law $`(๐ฒ_0)=\mu `$. For time-inhomogeneous processes, we use the notation $`P^{t,x}`$ or $`P^{t,\mu }`$ to denote the law of the process started at time $`t`$ with initial state $`๐ฒ_t=x`$ or initial law $`(๐ฒ_t)=\mu `$, respectively. We let $`E^x,E^\mu ,\mathrm{}`$ etc. denote expectation with respect to $`P^x,P^\mu ,\mathrm{}`$, respectively.
## 2 Renormalization classes on compact sets
### 2.1 Some general facts and heuristics
In this section, we explain that our main result is a special case of a type of theorem that we believe holds for many more renormalization classes on compact sets in $`^d`$. Moreover, we describe some elementary properties that hold generally for such renormalization classes. The proofs of Lemmas 2.12.8 can be found in Section 5.1 below.
Fix a prerenormalization class $`๐ฒ`$ on a set $`\overline{D}`$ where $`D^d`$ is open, bounded, and convex. Then $`๐ฒ`$ is a subset of the cone $`๐(\overline{D},M_+^d)`$ of continuous $`M_+^d`$-valued functions on $`\overline{D}`$. We equip $`๐(\overline{D},M_+^d)`$ with the topology of uniform convergence. Our first lemma says that the equilibrium measures $`\nu _x^{c,w}`$ and the renormalized diffusion matrices $`F_cw(x)`$ are continuous in their parameters.
###### Lemma 2.1
(Continuity in parameters)
* The map $`(x,c,w)\nu _x^{c,w}`$ from $`\overline{D}\times (0,\mathrm{})\times ๐ฒ`$ into $`_1(\overline{D})`$ is continuous.
* The map $`(x,c,w)F_cw(x)`$ from $`\overline{D}\times (0,\mathrm{})\times ๐ฒ`$ into $`M_+^d`$ is continuous.
In particular, $`x\nu _x^{c,w}`$ is a continuous probability kernel on $`\overline{D}`$, and $`F_cw๐(\overline{D},M_+^d)`$ for all $`c>0`$ and $`w๐ฒ`$. Recall from Definition 1.1 that $`\lambda w๐ฒ`$ for all $`w๐ฒ`$ and $`\lambda >0`$. The reason why we have included this assumption is that it is convenient to have the next scaling lemma around, which is a consequence of time scaling.
###### Lemma 2.2
(Scaling property of renormalization transformations) One has
$$\begin{array}{cccc}\hfill (\mathrm{i})& \hfill \nu _x^{\lambda c,\lambda w}& =& \nu _x^{c,w}\hfill \\ \hfill (\mathrm{ii})& \hfill F_{\lambda c}(\lambda w)& =& \lambda F_cw\hfill \end{array}\}(\lambda ,c>0,w๐ฒ,x\overline{D}).$$
(2.1)
The following simple lemma will play a crucial role in what follows.
###### Lemma 2.3
(Mean and covariance matrix) For all $`x\overline{D}`$ and $`i,j=1,\mathrm{},d`$, the mean and covariances of $`\nu _x^{c,w}`$ are given by
$$\begin{array}{cccc}\hfill (\mathrm{i})& \hfill _{\overline{D}}\nu _x^{c,w}(\mathrm{d}y)(y_ix_i)& =& 0,\hfill \\ \hfill (\mathrm{ii})& \hfill _{\overline{D}}\nu _x^{c,w}(\mathrm{d}y)(y_ix_i)(y_jx_j)& =& \frac{1}{c}F_cw_{ij}(x).\hfill \end{array}$$
(2.2)
For any $`w๐(\overline{D},M_+^d)`$, we call
$$_wD:=\{x\overline{D}:w_{ij}(x)=0i,j=1,\mathrm{},d\}$$
(2.3)
the effective boundary of $`D`$ (associated with $`w`$). If $`๐ฒ`$ is a solution to the martingale problem for the operator $`_{i,j=1}^dw_{ij}(y)\frac{^2}{y_iy_j}`$ (i.e., the operator in (1.1) without the drift), then, by martingale convergence, $`๐ฒ_t`$ converges a.s. to a limit $`๐ฒ_{\mathrm{}}`$; it is not hard to see that $`๐ฒ_{\mathrm{}}_wD`$ a.s. The next lemma says that the effective boundary is invariant under renormalization.
###### Lemma 2.4 (Invariance of effective boundary)
One has $`_{F_cw}D=_wD`$ for all $`w๐ฒ`$, $`c>0`$.
For example, for diffusion matrices $`w`$ from the renormalization class $`๐ฒ=๐ฒ_{\mathrm{cat}}`$, there occur four different effective boundaries, depending on whether $`w๐ฒ_{\mathrm{cat}}^{1,1}`$, $`๐ฒ_{\mathrm{cat}}^{0,1}`$, $`๐ฒ_{\mathrm{cat}}^{1,0}`$, or $`๐ฒ_{\mathrm{cat}}^{0,0}`$. These effective boundaries are depicted in Figure 1. The statement from Theorem 1.4 (a) that $`F_c(๐ฒ_{\mathrm{cat}}^{l,r})๐ฒ_{\mathrm{cat}}^{l,r}`$ is just the translation of Lemma 2.4 to the special set-up there.
From now on, let $`๐ฒ`$ be a renormalization class, i.e., $`๐ฒ`$ satisfies also condition (iv) from Definition 1.1. Fix a sequence of (positive) migration constants $`(c_k)_{k0}`$. By definition, the iterated probability kernels $`K^{w,(n)}`$ associated with a diffusion matrix $`w๐ฒ`$ (and the constants $`(c_k)_{k0}`$) are the probability kernels on $`\overline{D}`$ defined inductively by
$$K_x^{w,(n+1)}(\mathrm{d}z):=_{\overline{D}}\nu _x^{c_n,F^{(n)}w}(\mathrm{d}y)K_y^{w,(n)}(\mathrm{d}z)(n0)\text{with}K_x^{w,(0)}(\mathrm{d}y):=\delta _x(\mathrm{d}y),$$
(2.4)
with $`F^{(n)}`$ as in (1.8). Note that
$$F^{(n)}w(x)=_{\overline{D}}K_x^{w,(n)}(\mathrm{d}y)w(y)(x\overline{D},n0).$$
(2.5)
The next lemma follows by iteration from Lemmas 2.1 and 2.3. It their essence, this lemma and Lemma 2.6 below go back to \[BCGdH95\].
###### Lemma 2.5 (Basic properties of iterated probability kernels)
For each $`w๐ฒ`$, the $`K^{w,(n)}`$ are continuous probability kernels on $`\overline{D}`$. Moreover, for all $`x\overline{D}`$, $`i,j=1,\mathrm{},d`$, and $`n0`$, the mean and covariance matrix of $`K_x^{w,(n)}`$ are given by
$$\begin{array}{cccc}\hfill (\mathrm{i})& \hfill _{\overline{D}}K_x^{w,(n)}(\mathrm{d}y)(y_ix_i)& =& 0,\hfill \\ \hfill (\mathrm{ii})& \hfill _{\overline{D}}K_x^{w,(n)}(\mathrm{d}y)(y_ix_i)(y_jx_j)& =& s_nF^{(n)}w_{ij}(x).\hfill \end{array}$$
(2.6)
We equip the space $`๐(\overline{D},_1(\overline{D}))`$ of continuous probability kernels on $`\overline{D}`$ with the topology of uniform convergence (since $`_1(\overline{D})`$ is compact, there is a unique uniform structure on $`_1(\overline{D})`$ generating the topology). For โniceโ renormalization classes, it seems reasonable to conjecture that the kernels $`K^{w,(n)}`$ converge as $`n\mathrm{}`$ to some limit $`K^{w,}`$ in $`๐(\overline{D},_1(\overline{D}))`$. If this happens, then formula (2.6) (ii) tells us that the rescaled renormalized diffusion matrices $`s_nF^{(n)}w`$ converge uniformly on $`\overline{D}`$ to the covariance matrix of $`K^{w,}`$. This gives a heuristic explanation why we need to rescale the iterates $`F^{(n)}w`$ with the scaling constants $`s_n`$ from (1.9) to get a nontrivial limit in (1.11).
We now explain the relevance of the conditions (1.10) (i) and (ii) in the present more general context. If the iterated kernels converge to a limit $`K^{w,}`$, then condition (1.10) (i) guarantees that this limit is concentrated on the effective boundary:
###### Lemma 2.6 (Concentration on the effective boundary)
If $`s_n\underset{n\mathrm{}}{}\mathrm{}`$, then for any $`f๐(\overline{D})`$ such that $`f=0`$ on $`_wD`$:
$$\underset{n\mathrm{}}{lim}\underset{x\overline{D}}{sup}\left|_{\overline{D}}K_x^{w,(n)}(\mathrm{d}y)f(y)\right|=0.$$
(2.7)
In fact, the condition $`s_n\mathrm{}`$ guarantees that the corresponding system of hierarchically interacting diffusions with migration constants $`(c_k)_{k0}`$ clusters in the local mean field limit, see \[DG93a, Theorem 3\] or Appendix A.1 below.
To explain also the relevance of condition (1.10) (ii), we observe that using Lemma 2.2, we can convert the rescaled iterates $`s_nF^{(n)}`$ into (usual, not rescaled) iterates of another transformation. For this purpose, it will be convenient to modify the definition of our scaling constants $`s_n`$ a little bit. Fix some $`\beta >0`$ and put
$$\overline{s}_n:=\beta +s_n(n0).$$
(2.8)
Define rescaled renormalization transformations $`\overline{F}_\gamma :๐ฒ๐ฒ`$ by
$$\overline{F}_\gamma w:=(1+\gamma )F_{1/\gamma }w(\gamma >0,w๐ฒ).$$
(2.9)
Using (2.1) (ii), one easily deduces that
$$\overline{s}_nF^{(n)}w=\overline{F}_{\gamma _{n1}}\mathrm{}\overline{F}_{\gamma _0}(\beta w)(w๐ฒ,n1),$$
(2.10)
where
$$\gamma _n:=\frac{1}{\overline{s}_nc_n}(n0).$$
(2.11)
We can reformulate the conditions (1.10) (i) and (ii) in terms of the constants $`(\gamma _n)_{n0}`$. Indeed, it is not hard to check<sup>1</sup><sup>1</sup>1To see this, let $`\overline{s}_{\mathrm{}}(0,\mathrm{}]`$ denote the limit of the $`\overline{s}_n`$ and note that on the one hand, $`_n1/(\overline{s}_nc_n)_n\mathrm{log}(1+1/(\overline{s}_nc_n))=\mathrm{log}(_n\overline{s}_{n+1}/\overline{s}_n)=\mathrm{log}(\overline{s}_{\mathrm{}}/\overline{s}_1)`$, while on the other hand $`_n1/(\overline{s}_nc_n)_n(1+1/(\overline{s}_nc_n))=_n\overline{s}_{n+1}/\overline{s}_n=\overline{s}_{\mathrm{}}/\overline{s}_1`$. that equivalent formulations of condition (1.10) (i) are:
$$(\mathrm{i})s_n\underset{n\mathrm{}}{}\mathrm{},(\mathrm{ii})\overline{s}_n\underset{n\mathrm{}}{}\mathrm{},(\mathrm{iii})\underset{n}{}\gamma _n=\mathrm{}.$$
(2.12)
Since $`\overline{s}_{n+1}/\overline{s}_n=1+\gamma _n`$ we see moreover that, for any $`\gamma ^{}[0,\mathrm{}]`$, equivalent formulations of condition (1.10) (ii) are:
$$(\mathrm{i})\frac{s_{n+1}}{s_n}\underset{n\mathrm{}}{}1+\gamma ^{},(\mathrm{ii})\frac{\overline{s}_{n+1}}{\overline{s}_n}\underset{n\mathrm{}}{}1+\gamma ^{},(\mathrm{iii})\gamma _n\underset{n\mathrm{}}{}\gamma ^{}.$$
(2.13)
If $`0<\gamma ^{}<\mathrm{}`$, then, in the light of (2.10), we expect $`\overline{s}_nF^{(n)}w`$ to converge to a fixed point of the transformation $`\overline{F}_\gamma ^{}`$. If $`\gamma ^{}=0`$, the situation is more complex. In this case, we expect the orbit $`\overline{s}_nF^{(n)}w\overline{s}_{n+1}F^{(n+1)}w\mathrm{}`$, for large $`n`$, to approximate a continuous flow, the generator of which is
$$\underset{\gamma 0}{lim}\gamma ^1\left(\overline{F}_\gamma ww\right)(x)=\frac{1}{2}\underset{i,j=1}{\overset{d}{}}w_{ij}(x)\frac{^2}{x_ix_j}w(x)+w(x)(x\overline{D}).$$
(2.14)
To see that the right-hand side of this equation equals the left-hand side if $`w`$ is twice continuously differentiable, one needs a Taylor expansion of $`w`$ together with the moment formulas (2.2) for $`\nu _x^{1/\gamma ,w}`$. Under condition condition (2.12) (iii), we expect this continuous flow to reach equilibrium.
In the light if these considerations, we are led to at the following general conjecture.
###### Conjecture 2.7
(Limits of rescaled renormalized diffusion matrices) Assume that $`s_n\mathrm{}`$ and $`s_{n+1}/s_n1+\gamma ^{}`$ for some $`\gamma ^{}[0,\mathrm{}]`$. Then, for any $`w๐ฒ`$,
$$s_nF^{(n)}w\underset{n\mathrm{}}{}w^{},$$
(2.15)
where $`w^{}`$ satisfies
$$\begin{array}{ccccc}\hfill (\mathrm{i})& \hfill \overline{F}_\gamma ^{}w^{}& =& w^{}\hfill & \text{if}0<\gamma ^{}<\mathrm{},\hfill \\ \hfill (\mathrm{ii})& \hfill \frac{1}{2}\underset{i,j=1}{\overset{d}{}}w_{ij}^{}(x)\frac{^2}{x_ix_j}w^{}(x)+w^{}(x)& =& 0(x\overline{D})\hfill & \text{if}\gamma ^{}=0,\hfill \\ \hfill (\mathrm{iii})& \hfill \underset{\gamma \mathrm{}}{lim}\overline{F}_\gamma w^{}& =& w^{}\hfill & \text{if}\gamma ^{}=\mathrm{}.\hfill \end{array}$$
(2.16)
We call (2.16) (ii), which is in some sense the $`\gamma ^{}0`$ limit of the fixed point equation (2.16) (i), the asymptotic fixed point equation. A version of formula (2.16) (ii) occured in \[Swa99, formula (1.3.5)\] (a minus sign is missing there).
In particular, one may hope that for a given effective boundary, the equations in (2.16) have a unique solution. Our main result (Theorem 1.4) confirms this conjecture for the renormalization class $`๐ฒ_{\mathrm{cat}}`$ and for $`\gamma ^{}<\mathrm{}`$. In the next section, we discuss numerical evidence that supports Conjecture 2.7 in the case $`\gamma ^{}=0`$ for other renormalization classes on compacta as well.
In previous work on renormalization classes, fixed shapes have played an important role. By definition, for any prerenormalization class $`๐ฒ`$, a fixed shape is a subclass $`\widehat{๐ฒ}๐ฒ`$ of the form $`\widehat{๐ฒ}=\{\lambda w:\lambda >0\}`$ with $`0w๐ฒ`$, such that $`F_c(\widehat{๐ฒ})\widehat{๐ฒ}`$ for all $`c>0`$. The next lemma describes how fixed shapes for renormalization classes on compact sets typically arise.
###### Lemma 2.8
(Fixed shapes) Assume that for each $`0<\gamma ^{}<\mathrm{}`$, there is a $`0w^{}=w_\gamma ^{}^{}๐ฒ`$ such that $`s_nF^{(n)}w\underset{n\mathrm{}}{}w_\gamma ^{}^{}`$ whenever $`w๐ฒ`$, $`s_n\mathrm{}`$, and $`s_{n+1}/s_n1+\gamma ^{}`$. Then:
(a) $`w_\gamma ^{}^{}`$ is the unique solution in $`๐ฒ`$ of equation (2.16) (i).
(b) If $`w^{}=w_\gamma ^{}^{}`$ does not depend on $`\gamma ^{}`$, then
$$F_c(\lambda w^{})=(\frac{1}{\lambda }+\frac{1}{c})^1w^{}(\lambda ,c>0).$$
(2.17)
Moreover, $`\{\lambda w^{}:\lambda >0\}`$ is the unique fixed shape in $`๐ฒ`$.
(c) If the $`w_\gamma ^{}^{}`$ for different values of $`\gamma ^{}`$ are not constant multiples of each other, then $`๐ฒ`$ contains no fixed shapes.
Note that by Theorem 1.4, $`๐ฒ_{\mathrm{cat}}^{0,1}`$ is a renormalization class satisfying the general assumptions of Lemma 2.8. The unique solution of (2.16) (i) in $`๐ฒ_{\mathrm{cat}}^{0,1}`$ is of the form $`w^{}=w^{1,p^{}}`$ where $`p^{}=p_{0,1,\gamma ^{}}^{}`$. We conjecture that the $`p_{0,1,\gamma ^{}}^{}`$ for different values of $`\gamma ^{}`$ are not constant multiples of each other, and, as a consequence, that $`๐ฒ_{\mathrm{cat}}^{0,1}`$ contains no fixed shapes.
Many facts and conjectures that we have discussed can be generalized to renormalization classes on unbounded $`D`$, but in this case, the second moments of the iterated kernels $`K^{w,(n)}`$ may diverge as $`n\mathrm{}`$. As a result, because of formula (2.6) (ii), the $`s_n`$ may no longer be the right scaling factors to find a nontrivial limit of the renormalized diffusion matrices; see, for example, \[BCGdH97\].
### 2.2 Numerical solutions to the asymptotic fixed point equation
Let $`tw(t,)`$ be a solution to the continuous flow with the generator in (2.14), i.e., $`w`$ is an $`M_+^d`$-valued solution to the nonlinear partial differential equation
$$\frac{}{t}w(t,x)=\frac{1}{2}\underset{i,j=1}{\overset{d}{}}w_{ij}(t,x)\frac{^2}{x_ix_j}w(t,x)+w(t,x)(t0,x\overline{D}).$$
(2.18)
Solutions to (2.18) are quite easy to simulate on a computer. We have simulated solutions for all kind of diffusion matrices (including nondiagonal ones) on the unit square $`[0,1]^2`$, with the effective boundaries 1โ6 depicted in Figure 2. For all initial diffusion matrices $`w(0,)`$ we tried, the solution converged as $`t\mathrm{}`$ to a fixed point $`w^{}`$. In all cases except case 6, the fixed point was unique. The fixed points are listed in Figure 2. The functions $`p_{0,1,0}^{}`$ and $`q^{}`$ from Figure 2 are plotted in Figure 3. Here $`p_{0,1,0}^{}`$ is the function from Theorem 1.4 (c).
The fixed points for the effective boundaries in cases 1,2, and 4 are the unique solutions of equation (1.12) (ii) from Theorem 1.4 in the classes $`๐ฒ_{\mathrm{cat}}^{1,1}`$, $`๐ฒ_{\mathrm{cat}}^{0,1}`$, and $`๐ฒ_{\mathrm{cat}}^{0,0}`$, respectively. The simulations suggest that the domain of attraction of these fixed points (within the class of โallโ diffusion matrices on $`[0,1]^2`$) is actually a lot larger than the classes $`๐ฒ_{\mathrm{cat}}^{1,1}`$, $`๐ฒ_{\mathrm{cat}}^{0,1}`$, and $`๐ฒ_{\mathrm{cat}}^{0,0}`$.
The function $`q^{}`$ from case 3 satisfies $`q^{}(x_1,1)=x_1(1x_1)`$ and is zero on the other parts of the boundary. In contrast to what one might perhaps guess in view of case 2, $`q^{}`$ is not of the form $`q^{}(x_1,x_2)=f(x_2)x_1(1x_1)`$ for some function $`f`$.
Case 5 is somewhat degenerate since in this case the fixed point is not continuous.
The only case where the fixed point is not unique is case 6. Here, $`m`$ can be any positive definite matrix, while $`g^{}`$, depending on $`m`$, is the unique solution on $`(0,1)^2`$ of the equation $`1+\frac{1}{2}_{i,j=1}^2m_{ij}\frac{^2}{x_ix_i}g^{}(x)=0`$, with zero boundary conditions.
### 2.3 Previous rigorous results
In this section we discuss some results that have been derived previously for renormalization classes on compact sets.
###### Theorem 2.9
\[BCGdH95, DGV95\] (Universality class of Wright-Fisher models) Let $`D:=\{x^d:x_i>0i,_{i=1}^dx_i<1\}`$, and let $`\{e_0,\mathrm{},e_d\}`$, with $`e_0:=(0,\mathrm{},0)`$ and $`e_1:=(1,0,\mathrm{},0),\mathrm{},e_d:=(0,\mathrm{},0,1)`$ be the extremal points of $`\overline{D}`$. Let $`w_{ij}^{}(x):=x_i(\delta _{ij}x_j)`$ $`(x\overline{D}`$, $`i,j=1,\mathrm{},d)`$ denote the standard Wright-Fisher diffusion matrix, and assume that $`๐ฒ`$ is a renormalization class on $`\overline{D}`$ such that $`w^{}๐ฒ`$ and $`_w\overline{D}=\{e_0,\mathrm{},e_d\}`$ for all $`w๐ฒ`$. Let $`(c_k)_{k0}`$ be migration constants such that $`s_n\mathrm{}`$ as $`n\mathrm{}`$. Then, for all $`w๐ฒ`$, uniformly on $`\overline{D}`$,
$$s_nF^{(n)}w\underset{n\mathrm{}}{}w^{}.$$
(2.19)
The convergence in (2.19) is a consequence of Lemmas 2.5 and 2.6: The first moment formula (2.6) (i) and (2.7) show that $`K_x^{w,(n)}`$ converges to the unique distribution on $`\{e_0,\mathrm{},e_d\}`$ with mean $`x`$, and by the second moment formula (2.6) (ii) this implies the convergence of $`s_nF^{(n)}w`$.
In order for the iterates in (2.19) to be well-defined, Theorem 2.9 assumes that a renormalization class $`๐ฒ`$ of diffusion matrices $`w`$ on $`\overline{D}`$ with effective boundary $`\{e_0,\mathrm{},e_d\}`$ is given. The problem of finding a nontrivial example of such a renormalization class is open in dimensions greater than one. In the one-dimensional case, however, the following result is known.
###### Lemma 2.10
\[DG93b\] (Renormalization class on the unit interval) The set
$$๐ฒ_{\mathrm{DG}}:=\{w๐[0,1]:w=0\text{ on }\{0,1\},w>0\text{ on }(0,1),w\text{ Lipschitz}\}$$
(2.20)
is a renormalization class on $`[0,1]`$.
About renormalization of isotropic diffusions, the following result is known. Below, $`D:=\overline{D}\backslash D`$ denotes the topological boundary of $`D`$.
###### Theorem 2.11
\[dHS98\] (Universality class of isotropic models) Let $`D^d`$ be open, bounded, and convex and let $`mM_+^d`$ be fixed and (strictly) positive definite. Set $`w_{ij}^{}(x):=m_{ij}g^{}(x)`$, where $`g^{}`$ is the unique solution of $`1+\frac{1}{2}_{ij}m_{ij}\frac{^2}{x_ix_j}g^{}(x)=0`$ for $`xD`$ and $`g^{}(x)=0`$ for $`xD`$. Assume that $`๐ฒ`$ is a renormalization class on $`\overline{D}`$ such that $`w^{}๐ฒ`$ and such that each $`w๐ฒ`$ is of the form
$$w_{ij}(x)=m_{ij}g(x)(x\overline{D},i,j=1,\mathrm{},d),$$
(2.21)
for some $`g๐(\overline{D})`$ satisfying $`g>0`$ on $`D`$ and $`g=0`$ on $`D`$. Let $`(c_k)_{k0}`$ be migration constants such that $`s_n\mathrm{}`$ as $`n\mathrm{}`$. Then, for all $`w๐ฒ`$, uniformly on $`\overline{D}`$,
$$s_nF^{(n)}w\underset{n\mathrm{}}{}w^{}.$$
(2.22)
The proof of Theorem 2.11 follows the same lines as the proof of Theorem 2.9, with the difference that in this case one needs to generalize the first moment formula (2.6) (i) in the sense that $`_{\overline{D}}K_x^{w,(n)}(\mathrm{d}y)h(y)=h(x)`$ for any $`m`$-harmonic function $`h`$, i.e., $`h๐(\overline{D})`$ satisfying $`_{ij}m_{ij}\frac{^2}{x_ix_j}h(x)=0`$ for $`xD`$. The kernel $`K_x^{w,(n)}`$ now converges to the $`m`$-harmonic measure on $`D`$ with mean $`x`$, and this implies (2.22).
Again, in dimensions $`d2`$, the problem of finding a โreasonableโ class $`๐ฒ`$ satisfying the assumptions of Theorem 2.11 is so far unresolved. The problem with verifying conditions (i)โ(iv) from Definition 1.1 in an explicit set-up is that (i) and (ii) usually require some smoothness of $`w`$, while (iv) requires that one can prove the same smoothness for $`F_cw`$, which is difficult.
The proofs of Theorems 2.9 and 2.11 are based on the same principle. For any diffusion matrix $`w`$, let $`H_w`$ denote the class of $`w`$-harmonic functions, i.e., functions $`h๐(\overline{D})`$ satisfying $`_{ij}w_{ij}(x)\frac{^2}{x_ix_j}h(x)=0`$ on $`D`$. If $`w`$ belongs to one of the renormalization classes in Theorems 2.9 and 2.11, then $`H_w`$ has the property that $`T_{x,t}^ch(H_w)H_w`$ for all $`c>0`$, $`x\overline{D}`$, and $`t0`$, where $`T_{x,t}^ch(y):=h(x+(yx)e^{ct})`$ is the semigroup with generator $`_{i=1}^dc(x_iy_i)\frac{}{y_i}`$, i.e., the operator in (1.1) without the diffusion part. In this case we say that $`w`$ has invariant harmonics; see \[Swa00\]. As a consequence, one can prove that the iterated kernels satisfy $`_{\overline{D}}K_x^{w,(n)}(\mathrm{d}y)h(y)=h(x)`$ for all $`hH_w`$ and $`x\overline{D}`$. If $`s_n\mathrm{}`$, then this implies that $`K_x^{w,(n)}`$ converges to the unique $`H_w`$-harmonic measure on $`_wD`$ with mean $`x`$. Diffusion matrices from $`๐ฒ_{\mathrm{cat}}`$ do not in general have invariant harmonics. Therefore, to prove Theorem 1.4, we need new techniques.
Note that in the renormalization classes from Theorems 2.9 and 2.11, the unique attraction point $`w^{}`$ does not depend on $`\gamma ^{}`$. Therefore, by Lemma 2.8, these renormalization classes contain a unique fixed shape, which is given by $`\{\lambda w^{}:\lambda >0\}`$.
## 3 Connection with branching theory
From now on, we focuss on the renormalization class $`๐ฒ_{\mathrm{cat}}`$. We will show that for this renormalization class, the rescaled renormalization transformations $`\overline{F}_\gamma `$ from (2.9) can be expressed in terms of the log-Laplace operators of a discrete time branching process on $`[0,1]`$. This will allow us to use techniques from the theory of spatial branching processes to verify Conjecture 2.7 for the renormalization class $`๐ฒ_{\mathrm{cat}}`$ in the case $`\gamma ^{}<\mathrm{}`$.
### 3.1 Poisson-cluster branching processes
We first need some concepts and facts from branching theory. Finite measure-valued branching processes (on $``$) in discrete time have been introduced by Jiลina \[Jir64\]. We need to consider only a special class. Let $`E`$ be a separable, locally compact, and metrizable space. We call a continuous map $`๐ฌ`$ from $`E`$ into $`_1((E))`$ a continuous cluster mechanism. By definition, an $`(E)`$-valued random variable $`๐ณ`$ is a Poisson cluster measure on $`E`$ with locally finite intensity measure $`\mu `$ and continuous cluster mechanism $`๐ฌ`$, if its log-Laplace transform satisfies
$$\mathrm{log}E\left[\text{e}^{๐ณ,f}\right]=_E\mu (\mathrm{d}x)\left(1_{(E)}๐ฌ(x,\mathrm{d}\chi )\text{e}^{\chi ,f}\right)(fB_+(E)).$$
(3.1)
For given $`\mu `$ and $`๐ฌ`$, such a Poisson cluster measure exists, and is unique in distribution, provided that the right-hand side of (3.1) is finite for $`f=1`$. It may be constructed as $`๐ณ=_i\chi _{x_i}`$, where $`_i\delta _{x_i}`$ is a (possibly infinite) Poisson point measure with intensity $`\mu `$, and given $`x_1,x_2,\mathrm{}`$, the $`\chi _{x_1},\chi _{x_2},\mathrm{}`$ are independent random variables with laws $`๐ฌ(x_1,),๐ฌ(x_2,),\mathrm{}`$, respectively.
Now fix a finite sequence of functions $`q_k๐_+(E)`$ and continuous cluster mechanisms $`๐ฌ_k`$ ($`k=1,\mathrm{},n`$), define
$$๐ฐ_kf(x):=q_k(x)\left(1_{(E)}๐ฌ_k(x,\mathrm{d}\chi )\text{e}^{\chi ,f}\right)(xE,fB_+(E),k=1,\mathrm{},n),$$
(3.2)
and assume that
$$\underset{xE}{sup}๐ฐ_k1(x)<\mathrm{}(k=1,\mathrm{},n).$$
(3.3)
Then $`๐ฐ_k`$ maps $`B_+(E)`$ into $`B_+(E)`$ for each $`k`$, and for each $`(E)`$-valued initial state $`๐ณ_0`$, there exists a (time-inhomogeneous) Markov chain $`(๐ณ_0,\mathrm{},๐ณ_n)`$ in $`(E)`$, such that $`๐ณ_k`$, given $`๐ณ_{k1}`$, is a Poisson cluster measure with intensity $`q_k๐ณ_{k1}`$ and cluster mechanism $`๐ฌ_k`$. It is not hard to see that
$$E^\mu \left[\text{e}^{๐ณ_n,f}\right]=\text{e}^{\mu ,๐ฐ_1\mathrm{}๐ฐ_nf}(\mu (E),fB_+(E)).$$
(3.4)
We call $`๐ณ=(๐ณ_0,\mathrm{},๐ณ_n)`$ the Poisson-cluster branching process on $`E`$ with weight functions $`q_1,\mathrm{},q_n`$ and cluster mechanisms $`๐ฌ_1,\mathrm{},๐ฌ_n`$. The operator $`๐ฐ_k`$ is called the log-Laplace operator of the transition law from $`๐ณ_{k1}`$ to $`๐ณ_k`$. Note that we can write (3.4) in the suggestive form
$$P^\mu \left[\mathrm{Pois}(f๐ณ_n)=0\right]=P\left[\mathrm{Pois}\left((๐ฐ_1\mathrm{}๐ฐ_nf)\mu \right)=0\right].$$
(3.5)
Here, if $`\mu `$ is an $`(E)`$-valued random variable, then $`\mathrm{Pois}(\mu )`$ denotes an $`๐ฉ(E)`$-valued random variable such that conditioned on $`\mu `$, $`\mathrm{Pois}(\mu )`$ is a Poisson point measure with intensity $`\mu `$.
### 3.2 The renormalization branching process
We will now construct a Poisson-cluster branching process on $`[0,1]`$ of a special kind, and show that the rescaled renormalization transformations on $`๐ฒ_{\mathrm{cat}}`$ can be expressed in terms of the log-Laplace operators of this branching process.
By Lemma 5.4 below, for each $`\gamma >0`$ and $`x[0,1]`$, the SDE
$$\mathrm{d}๐ฒ(t)=\frac{1}{\gamma }(x๐ฒ(t))\mathrm{d}t+\sqrt{2๐ฒ(t)(1๐ฒ(t))}\mathrm{d}B(t),$$
(3.6)
has a unique (in law) stationary solution. We denote this solution by $`(๐ฒ_x^\gamma (t))_t`$. Let $`\tau _\gamma `$ be an independent exponentially distributed random variable with mean $`\gamma `$, and set
$$๐ต_x^\gamma :=_0^{\tau _\gamma }\delta _{๐ฒ_x^\gamma (t/2)}dt(\gamma >0,x[0,1]).$$
(3.7)
Define constants $`q_\gamma `$ and continuous (by Corollary 5.10 below) cluster mechanisms $`๐ฌ_\gamma `$ by
$$q_\gamma :=\frac{1}{\gamma }+1\text{and}๐ฌ_\gamma (x,):=(๐ต_x^\gamma )(\gamma >0,x[0,1]),$$
(3.8)
and let $`๐ฐ_\gamma `$ denote the log-Laplace operator with (constant) weight function $`q_\gamma `$ and cluster mechanism $`๐ฌ_\gamma `$, i.e.,
$$๐ฐ_\gamma f(x):=q_\gamma \left(1_{([0,1])}๐ฌ_\gamma (x,\mathrm{d}\chi )\text{e}^{\chi ,f}\right)(x[0,1],fB_+[0,1],\gamma >0).$$
(3.9)
We now establish the connection between renormalization transformations on $`๐ฒ_{\mathrm{cat}}`$ and log-Laplace operators.
###### Proposition 3.1
(Identification of the renormalization transformation) Let $`\overline{F}_\gamma `$ be the rescaled renormalization transformation on $`๐ฒ_{\mathrm{cat}}`$ defined in (2.9). Then
$$\overline{F}_\gamma w^{1,p}=w^{1,๐ฐ_\gamma p}(p,\gamma >0).$$
(3.10)
Fix a diffusion matrix $`w^{\alpha ,p}๐ฒ_{\mathrm{cat}}`$ and migration constants $`(c_k)_{k0}`$. Define constants $`\overline{s}_n`$ and $`\gamma _n`$ as in (2.8) and (2.11), respectively, where $`\beta :=1/\alpha `$. Then Proposition 3.1 and formula (2.10) show that
$$\overline{s}_nF^{(n)}w^{\alpha ,p}=w^{1,๐ฐ_{\gamma _{n1}}\mathrm{}๐ฐ_{\gamma _0}\left({\scriptscriptstyle \frac{p}{\alpha }}\right)}.$$
(3.11)
Here $`๐ฐ_{\gamma _{n1}},\mathrm{},๐ฐ_{\gamma _0}`$ are the log-Laplace operators of the Poisson-cluster branching process $`๐ณ=(๐ณ_n,\mathrm{},๐ณ_0)`$ with weight functions $`q_{\gamma _{n1}},\mathrm{},q_{\gamma _0}`$ and cluster mechanisms $`๐ฌ_{\gamma _{n1}},\mathrm{},๐ฌ_{\gamma _0}`$. We call $`๐ณ`$ (started at some time $`n`$ in an initial law $`(๐ณ_n)`$) the renormalization branching process. By formulas (3.4) and (3.11), the study of the limiting behavior of rescaled iterated renormalization transformations on $`๐ฒ_{\mathrm{cat}}`$ reduces to the study of the renormalization branching process $`๐ณ`$ in the limit $`n\mathrm{}`$.
### 3.3 Convergence to a time-homogeneous process
Let $`๐ณ=(๐ณ_n,\mathrm{},๐ณ_0)`$ be the renormalization branching process introduced in the last section. If the constants $`(\gamma _k)_{k0}`$ satisfy $`_n\gamma _n=\mathrm{}`$ and $`\gamma _n\gamma ^{}`$ for some $`\gamma ^{}[0,\mathrm{})`$, then $`๐ณ`$ is almost time-homogeneous for large $`n`$. More precisely, we will prove the following convergence result.
###### Theorem 3.2
(Convergence to a time-homogenous branching process) Assume that $`(๐ณ_n)\underset{n\mathrm{}}{}\mu `$ for some probability law $`\mu `$ on $`([0,1])`$.
(a) If $`0<\gamma ^{}<\mathrm{}`$, then
$$(๐ณ_n,๐ณ_{n+1},\mathrm{})\underset{n\mathrm{}}{}(๐ด_0^\gamma ^{},๐ด_1^\gamma ^{},\mathrm{}),$$
(3.12)
where $`๐ด^\gamma ^{}`$ is the time-homogenous branching process with log-Laplace operator $`๐ฐ_\gamma ^{}`$ in each step and initial law $`(๐ด_0^\gamma ^{})=\mu `$.
(b) If $`\gamma ^{}=0`$, then
$$\left(\left(๐ณ_{k_n(t)}\right)_{t0}\right)\underset{n\mathrm{}}{}\left(\left(๐ด_t^0\right)_{t0}\right),$$
(3.13)
where $``$ denotes weak convergence of laws on path space, $`k_n(t):=\mathrm{min}\{k:0kn`$, $`_{l=k}^{n1}\gamma _lt\}`$, and $`๐ด^0`$ is the super-Wright-Fisher diffusion with activity and growth parameter both identically $`1`$ and initial law $`(๐ด_0^0)=\mu `$.
The super-Wright-Fisher diffusion was studied in \[FS03\]. By definition, $`๐ด^0`$ is the time-homogeneous Markov process in $`[0,1]`$ with continuous sample paths, whose Laplace functionals are given by
$$E^\mu \left[\text{e}^{๐ด_t^0,f}\right]=\text{e}^{\mu ,๐ฐ_t^0f}(\mu [0,1],fB_+[0,1],t0).$$
(3.14)
Here $`๐ฐ_t^0f=u_t`$ is the unique mild solution of the semilinear Cauchy equation
$$\{\begin{array}{ccc}\hfill \frac{}{t}u_t(x)& =& \frac{1}{2}x(1x)\frac{^2}{x^2}u_t(x)+u_t(x)(1u_t(x))(t0,x[0,1]),\hfill \\ \hfill u_0& =& f.\hfill \end{array}$$
(3.15)
For a further study of the renormalization branching process $`๐ณ`$ and its limiting processes $`๐ด^\gamma ^{}`$ ($`\gamma ^{}0`$) we will use the technique of embedded particle systems, which we explain in the next section.
### 3.4 Weighted and Poissonized branching processes
In this section, we explain how from a Poisson-cluster branching process it is possible to construct other branching processes by weighting and Poissonization. We first need to introduce spatial branching particle systems in some generality.
Let $`E`$ again be separable, locally compact, and metrizable. For $`\nu ๐ฉ(E)`$ and $`fB_{[0,1]}(E)`$, we adopt the notation
$$f^0:=1\text{and}f^\nu :=\underset{i=1}{\overset{m}{}}f(x_i)\text{when}\nu =\underset{i=1}{\overset{m}{}}\delta _{x_i}(m1).$$
(3.16)
We call a continuous map $`xQ(x,)`$ from $`E`$ into $`_1(๐ฉ(E))`$ a continuous offspring mechanism.
Fix continuous offspring mechanisms $`Q_k`$ ($`1kn`$), and let $`(X_0,\mathrm{},X_n)`$ be a Markov chain in $`๐ฉ(E)`$ such that, given that $`X_{k1}=_{i=1}^m\delta _{x_i}`$, the next step of the chain $`X_k`$ is a sum of independent random variables with laws $`Q_k(x_i,)`$ ($`i=1,\mathrm{},m`$). Then
$$E^\nu \left[(1f)^{X_n}\right]=(1U_1\mathrm{}U_nf)^\nu (\nu ๐ฉ(E),fB_{[0,1]}(E)),$$
(3.17)
where $`U_k:B_{[0,1]}(E)B_{[0,1]}(E)`$ is defined as
$$U_kf(x):=1_{๐ฉ(E)}Q^k(x,\mathrm{d}\nu )(1f)^\nu (1kn,xE,fB_{[0,1]}(E)).$$
(3.18)
We call $`U_k`$ the generating operator of the transition law from $`X_{k1}`$ to $`X_k`$, and we call $`X=(X_0,\mathrm{},X_n)`$ the branching particle system on $`E`$ with generating operators $`U_1,\mathrm{},U_n`$. It is often useful to write (3.17) in the suggestive form
$$P^\nu \left[\mathrm{Thin}_f(X_n)=0\right]=P\left[\mathrm{Thin}_{U_1\mathrm{}U_nf}(\nu )=0\right](\nu ๐ฉ(E),fB_{[0,1]}(E)).$$
(3.19)
Here, if $`\nu `$ is an $`๐ฉ(E)`$-valued random variable and $`fB_{[0,1]}(E)`$, then $`\mathrm{Thin}_f(\nu )`$ denotes an $`๐ฉ(E)`$-valued random variable such that conditioned on $`\nu `$, $`\mathrm{Thin}_f(\nu )`$ is obtained from $`\nu `$ by independently throwing away particles from $`\nu `$, where a particle at $`x`$ is kept with probability $`f(x)`$. One has the elementary relations
$$\mathrm{Thin}_f(\mathrm{Thin}_g(\nu ))\stackrel{๐}{=}\mathrm{Thin}_{fg}(\nu )\text{and}\mathrm{Thin}_f(\mathrm{Pois}(\mu ))\stackrel{๐}{=}\mathrm{Pois}(f\mu ),$$
(3.20)
where $`\stackrel{๐}{=}`$ denotes equality in distribution.
We are now ready to describe weighted and Poissonized branching processes. Let $`๐ณ=(๐ณ_0,\mathrm{},๐ณ_n)`$ be a Poisson-cluster branching process on $`E`$, with continuous weight functions $`q_1,\mathrm{},q_n`$, continuous cluster mechanisms $`๐ฌ_1,\mathrm{},๐ฌ_n`$, and log-Laplace operators $`๐ฐ_1,\mathrm{},๐ฐ_n`$ given by (3.2) and satisfying (3.3). Let $`๐ต_x^k`$ denote an $`(E)`$-valued random variable with law $`๐ฌ_k(x,)`$. Let $`h๐_+(E)`$ be bounded, $`h0`$, and put $`E^h:=\{xE:h(x)>0\}`$. For $`fB_+(E^h)`$, define $`hfB_+(E)`$ by $`hf(x):=h(x)f(x)`$ if $`xE^h`$ and $`hf(x):=0`$ otherwise.
###### Proposition 3.3
(Weighting of Poisson-cluster branching processes) Assume that there exists a constant $`K<\mathrm{}`$ such that $`๐ฐ_khKh`$ for all $`k=1,\mathrm{},n`$. Then there exists a Poisson-cluster branching process $`๐ณ^h=(๐ณ_0^h,\mathrm{},๐ณ_n^h)`$ on $`E^h`$ with weight functions $`(q_1^h,\mathrm{},q_n^h)`$ given by $`q_k^h:=q_k/h`$, continuous cluster mechanisms $`๐ฌ_1^h,\mathrm{},๐ฌ_n^h`$ given by
$$๐ฌ_k^h(x,):=(h๐ต_x^k)(xE^h),$$
(3.21)
and log-Laplace operators $`๐ฐ_1^h,\mathrm{},๐ฐ_n^h`$ satisfying
$$h๐ฐ_k^hf:=๐ฐ_k(hf)(fB_+(E^h)).$$
(3.22)
The processes $`๐ณ`$ and $`X^h`$ are related by
$$(๐ณ_0^h)=(h๐ณ_0)\text{implies}(๐ณ_k^h)=(h๐ณ_k)(0kn).$$
(3.23)
###### Proposition 3.4
(Poissonization of Poisson-cluster branching processes) Assume that $`๐ฐ_khh`$ for all $`k=1,\mathrm{},n`$. Then there exists a branching particle system $`X^h=(X_0^h,\mathrm{},X_n^h)`$ on $`E^h`$ with continuous offspring mechanisms $`Q_1^h,\mathrm{},Q_n^h`$ given by
$$Q_k^h(x,):=\frac{q_k(x)}{h(x)}P\left[\mathrm{Pois}(h๐ต_x^k)\right]+\left(1\frac{q_k(x)}{h(x)}\right)\delta _0()(xE^h),$$
(3.24)
and generating operators $`U_1^h,\mathrm{},U_n^h`$ satisfying
$$hU_k^hf:=๐ฐ_k(hf)(fB_{[0,1]}(E^h)).$$
(3.25)
The processes $`๐ณ`$ and $`X^h`$ are related by
$$(X_0^h)=(\mathrm{Pois}(h๐ณ_0))\text{implies}(X_k^h)=(\mathrm{Pois}(h๐ณ_k))(0kn).$$
(3.26)
Here, the right-hand side of (3.24) is always a probability measure, despite that it may happen that $`q_k(x)/h(x)>1`$. The (straightforward) proofs of Propositions 3.3 and 3.4 can be found in Section 7.1 below. If (3.23) holds then we say that $`๐ณ^h`$ is obtained from $`๐ณ`$ by weighting with density $`h`$. If (3.26) holds then we say that $`X^h`$ is obtained from $`๐ณ`$ by Poissonization with density $`h`$. Proposition 3.4 says that a Poisson-cluster branching process $`๐ณ`$ contains, in a way, certain โembeddedโ branching particle systems $`X^h`$. Poissonization relations for superprocesses and embedded particle systems have enjoyed considerable attention, see \[FS04\] and references therein.
A function $`hB_+(E)`$ such that $`๐ฐ_khh`$ is called $`๐ฐ_k`$-superharmonic. If the reverse inequality holds we say that $`h`$ is $`๐ฐ_k`$-subharmonic. If $`๐ฐ_kh=h`$ then $`h`$ is called $`๐ฐ_k`$-harmonic.
### 3.5 Extinction versus unbounded growth for embedded particle systems
In this section we explain how embedded particle systems can be used to prove Theorem 1.4. Throughout this section $`(\gamma _k)_{k0}`$ are positive constants such that $`_n\gamma _n=\mathrm{}`$ and $`\gamma _n\gamma ^{}`$ for some $`\gamma ^{}[0,\mathrm{})`$, and $`๐ณ=(๐ณ_n,\mathrm{},๐ณ_0)`$ is the renormalization branching process on $`[0,1]`$ defined in Section 3.2. We write
$$๐ฐ^{(n)}:=๐ฐ_{\gamma _{n1}}\mathrm{}๐ฐ_{\gamma _0}.$$
(3.27)
In view of formula (3.11), in order to prove Theorem 1.4, we need the following result.
###### Proposition 3.5
(Limits of iterated log-Laplace operators) Uniformly on $`[0,1]`$,
$$\begin{array}{ccccc}\hfill (\mathrm{i})& \hfill \underset{n\mathrm{}}{lim}๐ฐ^{(n)}p& =& 1\hfill & (p_{1,1}),\hfill \\ \hfill (\mathrm{ii})& \hfill \underset{n\mathrm{}}{lim}๐ฐ^{(n)}p& =& 0\hfill & (p_{0,0}),\hfill \\ \hfill (\mathrm{iii})& \hfill \underset{n\mathrm{}}{lim}๐ฐ^{(n)}p& =& p_{0,1,\gamma ^{}}^{}\hfill & (p_{0,1}),\hfill \end{array}$$
(3.28)
where $`p_{0,1,\gamma ^{}}^{}:[0,1][0,1]`$ is a function depending on $`\gamma ^{}`$ but not on $`p_{0,1}`$.
In our proof of Proposition 3.5, we will use embedded particle systems $`X^h=(X_n^h,\mathrm{},X_0^h)`$ obtained from $`๐ณ`$ by Poissonization with certain $`h`$ taken from the classes $`_{1,1}`$, $`_{0,0}`$, and $`_{0,1}`$.
###### Lemma 3.6
(Embedded particle system with $`h_{1,1}`$) The constant function $`h_{1,1}(x):=1`$ is $`๐ฐ_\gamma `$-harmonic for each $`\gamma >0`$. The corresponding embedded particle system $`X^{h_{1,1}}`$ on $`[0,1]`$ satisfies
$$P^{n,\delta _x}\left[|X_0^{h_{1,1}}|\right]\underset{n\mathrm{}}{}\delta _{\mathrm{}}$$
(3.29)
uniformly<sup>2</sup><sup>2</sup>2Since $`_1[0,\mathrm{}]`$ is compact in the topology of weak convergence, there is a unique uniform structure compatible with the topology, and therefore it makes sense to talk about uniform convergence of $`_1[0,\mathrm{}]`$-valued functions (in this case, $`xP^{n,\delta _x}\left[|X_0^{h_{1,1}}|\right]`$). for all $`x[0,1]`$.
In (3.29) and similar formulas below, $``$ denotes weak convergence of probability measures on $`[0,\mathrm{}]`$. Thus, (3.29) says that for processes started with one particle on the position $`x`$ at times $`n`$, the number of particles at time zero converges to infinity as $`n\mathrm{}`$.
###### Lemma 3.7
(Embedded particle system with $`h_{0,0}`$) The function $`h_{0,0}(x):=x(1x)`$ $`(x[0,1])`$ is $`๐ฐ_\gamma `$-superharmonic for each $`\gamma >0`$. The corresponding embedded particle system $`X^{h_{0,0}}`$ on $`(0,1)`$ is critical and satisfies
$$P^{n,\delta _x}\left[|X_0^{h_{0,0}}|\right]\underset{n\mathrm{}}{}\delta _0$$
(3.30)
locally uniformly for all $`x(0,1)`$.
Here, a branching particle system $`X`$ is called critical if each particle produces on average one offspring (in each time step and independent of its position). Formula (3.30) says that the embedded particle system $`X^{h_{0,0}}`$ gets extinct during the time interval $`\{n,\mathrm{},0\}`$ with probability tending to one as $`n\mathrm{}`$. We can summarize Lemmas 3.6 and 3.7 by saying that the embedded particle system associated with $`h_{1,1}`$ grows unboundedly while the embedded particle system associated with $`h_{0,0}`$ becomes extinct as $`n\mathrm{}`$.
We will also consider an embedded particle systems $`X^{h_{0,1}}`$ for a certain $`h_{0,1}`$ taken from $`_{0,1}`$. It turns out that this system either gets extinct or grows unboundedly, each with a positive probability. In order to determine these probabilities, we need to consider embedded particle systems for the time-homogeneous processes $`๐ด^\gamma ^{}`$ ($`\gamma ^{}[0,\mathrm{})`$) from (3.12) and (3.13). If $`h_{0,1}`$ is $`๐ฐ_\gamma ^{}`$-superharmonic for some $`\gamma ^{}>0`$, then Poissonizing the process $`๐ด^\gamma ^{}`$ with $`h`$ yields a branching particle system on $`(0,1]`$ which we denote by $`Y^{\gamma ^{},h}=(Y_0^{\gamma ^{},h},Y_1^{\gamma ^{},h},\mathrm{})`$. Likewise, if $`h_{0,1}`$ is twice continuously differentiable and satisfies
$$\frac{1}{2}x(1x)\frac{^2}{x^2}h(x)h(x)(1h(x))0,$$
(3.31)
then Poissonizing the super-Wright-Fisher diffusion $`๐ด^0`$ with $`h`$ yields a continuous-time branching particle system on $`(0,1]`$, which we denote by $`Y^{0,h}=(Y_t^{0,h})_{t0}`$. For example, for $`m4`$, the function $`h(x):=1(1x)^m`$ satisfies (3.31).
###### Lemma 3.8
(Embedded particle system with $`h_{0,1}`$) The function $`h_{0,1}(x):=1(1x)^7`$ is $`๐ฐ_\gamma `$-superharmonic for each $`\gamma >0`$. The corresponding embedded particle system $`X^{h_{0,1}}`$ on $`(0,1]`$ satisfies
$$P^{n,\delta _x}\left[|X_0^{h_{0,1}}|\right]\underset{n\mathrm{}}{}\rho _\gamma ^{}(x)\delta _{\mathrm{}}+(1\rho _\gamma ^{}(x))\delta _0,$$
(3.32)
locally uniformly for all $`x(0,1]`$, where
$$\rho _\gamma ^{}(x):=\{\begin{array}{cc}P^{\delta _x}[Y_k^{\gamma ^{},h_{0,1}}0k0]\hfill & (0<\gamma ^{}<\mathrm{}),\hfill \\ P^{\delta _x}[Y_t^{0,h_{0,1}}0t0]\hfill & (\gamma ^{}=0).\hfill \end{array}$$
(3.33)
We now explain how Lemmas 3.63.8 imply Proposition 3.5. In doing so, it will be more convenient to work with weighted branching processes than with Poissonized branching processes. A little argument (which can be found in Lemma 7.12 below) shows that Lemmas 3.63.8 are equivalent to the next proposition.
###### Proposition 3.9
(Extinction versus unbounded growth) Let $`h_{1,1}`$, $`h_{0,0}`$, and $`h_{0,1}`$ be as in Lemmas 3.63.8. For $`\gamma ^{}[0,\mathrm{})`$, put $`p_{1,1,\gamma ^{}}^{}(x):=1`$, $`p_{0,0,\gamma ^{}}^{}(x):=0`$ $`(x[0,1])`$, and
$$p_{0,1,\gamma ^{}}^{}(0):=0\text{and}p_{0,1,\gamma ^{}}^{}(x):=h_{0,1}(x)\rho _\gamma ^{}(x)(x(0,1]),$$
(3.34)
with $`\rho _\gamma ^{}`$ as in (3.33). Then, for $`(l,r)=(1,1),(0,0)`$, and $`(0,1)`$,
$$P^{n,\delta _x}\left[๐ณ_0,h_{l,r}\right]\underset{n\mathrm{}}{}\text{e}^{p_{l,r,\gamma ^{}}^{}\left(x\right)}\delta _0+\left(1\text{e}^{p_{l,r,\gamma ^{}}^{}\left(x\right)}\right)\delta _{\mathrm{}},$$
(3.35)
uniformly for all $`x[0,1]`$.
Formula (3.35) says that the weighted branching process $`๐ณ^{h_{l,r}}`$ exhibits a form of โextinction versus unbounded growthโ. More precisely, for large $`n`$ the total mass of $`h_{l,r}๐ณ_0`$ is close to $`0`$ or $`\mathrm{}`$ with high probability.
Proof of Proposition 3.5 By (3.4),
$$๐ฐ^{(n)}p(x)=\mathrm{log}E^{n,\delta _x}\left[\text{e}^{๐ณ_0,p}\right](pB_+[0,1],x[0,1]).$$
(3.36)
We first prove formula (3.28) (ii). For $`(l,r)=(0,0)`$, formula (3.35) says that
$$P^{n,\delta _x}[๐ณ_0,h_{0,0}]\underset{n\mathrm{}}{}\delta _0$$
(3.37)
uniformly for all $`x[0,1]`$. If $`p_{0,0}`$, then we can find $`r>0`$ such that $`prh_{0,0}`$. Therefore, (3.37) implies that for any $`p_{0,0}`$,
$$P^{n,\delta _x}[๐ณ_0,p]\underset{n\mathrm{}}{}\delta _0.$$
(3.38)
By (3.36) it follows that
$$๐ฐ^{(n)}p(x)=\mathrm{log}E^{n,\delta _x}\left[\text{e}^{๐ณ_0,p}\right]\underset{n\mathrm{}}{}0,$$
(3.39)
where the limits in (3.38) and (3.39) are uniform in $`x[0,1]`$. This proves formula (3.28) (ii). To prove formula (3.28) (iii), note that for any $`p_{0,1}`$ we can choose $`0<r_{}<r_+`$ such that $`r_{}h_{0,1}p+h_{0,0}r_+h_{0,1}`$. Therefore, (3.35) implies that
$$P^{n,\delta _x}[๐ณ_0,p+๐ณ_0,h_{0,0}]\underset{n\mathrm{}}{}\text{e}^{p_{0,1,\gamma ^{}}^{}\left(x\right)}\delta _0+\left(1\text{e}^{p_{0,1,\gamma ^{}}^{}\left(x\right)}\right)\delta _{\mathrm{}}.$$
(3.40)
Using moreover (3.37), we see that
$$P^{n,\delta _x}[๐ณ_0,p]\underset{n\mathrm{}}{}\text{e}^{p_{0,1,\gamma ^{}}^{}\left(x\right)}\delta _0+\left(1\text{e}^{p_{0,1,\gamma ^{}}^{}\left(x\right)}\right)\delta _{\mathrm{}}.$$
(3.41)
By (3.36), it follows that
$$๐ฐ^{(n)}p(x)=\mathrm{log}E^{n,\delta _x}\left[\text{e}^{๐ณ_0,p}\right]\underset{n\mathrm{}}{}p_{0,1,\gamma ^{}}^{}(x)$$
(3.42)
where all limits are uniform in $`x[0,1]`$. This proves (3.28) (iii). The proof of (3.28) (i) is similar but easier.
## 4 Discussion, open problems
### 4.1 Discussion
Consider a $`([0,1]^2)^^2`$-valued process $`๐ฑ=(๐ฑ_\xi )_{\xi ^2}=(๐ฑ_\xi ^1,๐ฑ_\xi ^2)_{\xi ^2}`$, solving a system of SDEโs of the form
$$\begin{array}{ccc}\hfill \mathrm{d}๐ฑ_\xi ^1(t)& =& \underset{\eta :|\eta \xi |=1}{}\left(๐ฑ_\eta ^1(t)๐ฑ_\xi ^1(t)\right)\mathrm{d}t+\sqrt{2\alpha ๐ฑ_\xi ^1(t)(1๐ฑ_\xi ^1(t))}\mathrm{d}B_\xi ^1(t),\hfill \\ \hfill \mathrm{d}๐ฑ_\xi ^2(t)& =& \underset{\eta :|\eta \xi |=1}{}\left(๐ฑ_\eta ^2(t)๐ฑ_\xi ^2(t)\right)\mathrm{d}t+\sqrt{2p(๐ฑ_\xi ^1(t))๐ฑ_\xi ^2(t)(1๐ฑ_\xi ^2(t))}\mathrm{d}B_\xi ^2(t),\hfill \end{array}$$
(4.1)
where $`\alpha >0`$ is a constant, $`p`$ is a nonnegative function on $`[0,1]`$ satisfying $`p(0)=0`$ and $`p(1)>0`$, and $`(B_\xi ^i)_{\xi ^2}^{i=1,2}`$ is a collection of independent Brownian motions. We call $`๐ฑ`$ a system of linearly interacting catalytic Wright-Fisher diffusions with catalyzation function $`p`$. It is expected that $`๐ฑ`$ clusters, i.e., $`๐ฑ(t)`$ converges in distribution as $`t\mathrm{}`$ to a limit $`(๐ฑ_\xi (\mathrm{}))_{\xi ^2}`$ such that $`๐ฑ_\xi (\mathrm{})=๐ฑ_0(\mathrm{})`$ for all $`\xi ^2`$ and $`๐ฑ_0(\mathrm{})`$ takes values in the effective boundary associated with the diffusion matrix $`w^{\alpha ,p}`$ (see (2.3)). Heuristic arguments, based on renormalization, yield a formula for the clustering distribution $`(๐ฑ_0(\mathrm{}))`$ in terms of the diffusion matrix $`w^{}`$ which is the unique solution of the asymptotic fixed point equation (2.16) (ii) in the renormalization class $`๐ฒ_{\mathrm{cat}}^{0,1}`$; see Conjecture A.3 in Appendix A.2 below.
The present paper is inspired by the work of Greven, Klenke and Wakolbinger \[GKW01\]. They study a model that is closely related to (4.1), but where $`๐ฑ^1`$ is replaced by a voter model. They show that their model clusters and determine its clustering distribution $`(๐ฑ_0(\mathrm{}))`$, which turns out to coincide with the mentioned prediction for (4.1) based on renormalization theory. In fact, they believe their results to hold for the model in (4.1) too, but they could not prove this due to certain technical difficulties that a $`[0,1]`$-valued catalyst would create, compared to the simpler $`\{0,1\}`$-valued voter model.
The work in \[GKW01\] not only provides the main motivation for the present paper, but also inspired some of our techniques for proving Theorem 1.4. This concerns in particular the proof of Proposition 3.1, which makes the connection between renormalization transformations and a branching process. We hope that conversely, our techniques may shed some light on the problems left open by \[GKW01\], in particular, the question whether their results stay true if the voter model catalyst is replaced by a Wright-Fisher catalyst. It seems plausible that their results may not hold for the model in (4.1) if the catalyzing function $`p`$ grows too fast at $`0`$. On the other hand, our proofs suggest that $`p`$ with a finite slope at $`0`$ should be OK. (In particular, while deriving formula (3.40), we use that $`p`$ can be bounded from above by $`r_+h_{0,1}`$ for some $`r_+>0`$, which requires that $`p`$ has a finite slope at $`0`$.)
Our results are also interesting in the wider program of studying renormalization classes in the sense of Definition 1.1. We conjecture that the class $`๐ฒ_{\mathrm{cat}}^{0,1}`$, unlike all renormalization classes studied previously, contains no fixed shapes (see the discussion following Lemma 2.8). In fact, we expect this to be the usual situation. In this sense, the renormalization classes studied so far were all of a special type.
### 4.2 Open problems
The general program of studying renormalization classes in the sense of Definition 1.1 contains a wealth of open problems. In our proofs, we make heavy use of the single-way nature of the catalyzation in (1.7), in particular, the fact that $`๐ฒ^1`$ is an autonomous process which allows one to condition on $`๐ฒ^1`$ and consider $`๐ฒ^2`$ as a process in a random environment created by $`๐ฒ^1`$. As soon as one leaves the single-way catalytic regime one runs into several difficulties, both technically (it is hard to prove that a given class of matrices is a renormalization class in the sense of Definition 1.1) and conceptually (it is not clear when solutions to the asymptotic fixed shape equation (2.16) (ii) are unique). Therefore, it seems at present hard to verify the complete picture for renormalization classes on the unit square that arises from the numerical simulations described in Section 2.2 and Figures 2 and 3, unless one or more essential new ideas are added.
In this context, the study of the nonlinear partial differential equation (2.18) and its fixed points seems to be a challenging problem. This may be a hard problem from an analytic point of view, since the equation is degenerate and not in divergence form. For the renormalization class $`๐ฒ_{\mathrm{cat}}`$, the quasilinear equation (2.18) reduces to the semilinear equation (3.15), which is analytically easier to treat and moreover has a probabilistic interpretation in terms of a superprocess. For a study of the semilinear equation (3.15) we refer to \[FS03\]. We do not know whether solutions to equation (2.18) can in general be represented in terms of a stochastic process of some sort.
Even for the renormalization class $`๐ฒ_{\mathrm{cat}}`$, several interesting problems are left open. One of the most urgent ones is to prove that the functions $`p_{0,1,\gamma ^{}}^{}`$ are not constant in $`\gamma ^{}`$, and therefore, by Lemma 2.8 (c), $`๐ฒ_{\mathrm{cat}}^{0,1}`$ contains no fixed shapes. Moreover, we have not investigated the iterated renormalization transformations in the regime $`\gamma ^{}=\mathrm{}`$. Also, we believe that the convergence in (3.28) (ii) does not hold if the condition that $`p`$ is Lipschitz is dropped, in particular, if $`p`$ has an infinite slope at $`0`$ or an infinite negative slope at $`1`$. For $`p_{0,0}`$, it seems plausible that a properly rescaled version of the iterates $`๐ฐ^{(n)}p`$ converges to a universal limit, but we have not investigated this either. Finally, we have not investigated the convergence of the iterated kernels $`K^{w,(n)}`$ from (2.4) (in particular, we have not verified Conjecture A.2) for the renormalization class $`๐ฒ_{\mathrm{cat}}`$.
Our methods, combined with those in \[BCGdH95\], can probably be extended to study the action of iterated renormalization transformations on diffusion matrices of the following more general form (compared to (1.4)):
$$w(x)=\left(\begin{array}{cc}g(x_1)& 0\\ 0& p(x_1)x_2(1x_2)\end{array}\right)(x=[0,1]^2),$$
(4.2)
where $`g:[0,1]`$ is Lipschitz, $`g(0)=g(1)=0`$, $`g>0`$ on $`(0,1)`$, and $`p`$ as before. This would, however, require a lot of extra technical work and probably not generate much new insight. The numerical simulations mentioned in Section 2.2 suggest that many diffusion matrices of an even more general form than (4.2) also converge under renormalization to the limit points $`w^{}`$ from Theorem 1.4, but we donโt know how to prove this.
Part II
Outline of Part II In Section 5, we verify that $`๐ฒ_{\mathrm{cat}}`$ is a renormalization class, we prove Proposition 3.1, which connects the renormalization transformations $`F_c`$ to the log-Laplace operators $`๐ฐ_\gamma `$, and we collect a number of technical properties of the operators $`๐ฐ_\gamma `$ that will be needed later on. In Section 6 we prove Theorem 3.2 about the convergence of the renormalization branching process to a time-homogeneous limit. In Section 7, we prove the statements from Section 3.5 about extinction versus unbounded growth of embedded particle systems, with the exception of Lemma 3.7, which is proved in Section 8. In Section 9, finally, we combine the results derived by that point to prove our main theorem.
## 5 The renormalization class $`๐ฒ_{\mathrm{cat}}`$
In this section we prove Theorem 1.4 (a) and Proposition 3.1, as well as Lemmas 2.12.8 from Section 2. The section is organized according to the techniques used. Section 5.1 collects some facts that hold for general renormalization classes on compact sets. In Section 5.2 we use the SDE (1.7) to couple catalytic Wright-Fisher diffusions. In Section 5.3 we apply the moment duality for the Wright-Fisher diffusion to the catalyst and to the reactant conditioned on the catalyst. In Section 5.4 we prove that monotone concave catalyzing functions form a preserved class under renormalization.
### 5.1 Renormalization classes on compact sets
In this section, we prove the lemmas stated in Section 2. Recall that $`D^d`$ is open, bounded, and convex, and that $`๐ฒ`$ is a prerenormalization class on $`\overline{D}`$, equipped with the topology of uniform convergence.
Proof of Lemma 2.1 To see that $`(x,c,w)\nu _x^{c,w}`$ is continuous, let $`(x_n,c_n,w_n)`$ be a sequence converging in $`\overline{D}\times (0,\mathrm{})\times ๐ฒ`$ to a limit $`(x,c,w)`$. By the compactness of $`\overline{D}`$, the sequence $`(\nu _{x_n}^{c_n,w_n})_{n0}`$ is tight, and each limit point $`\nu ^{}`$ satisfies
$$\nu ^{},A_x^{c,w}f=0(f๐^{(2)}(D)).$$
(5.1)
Therefore, by \[EK86, Theorem 4.9.17\], $`\nu ^{}`$ is an invariant law for the martingale problem associated with $`A_x^{c,w}`$. Since we are assuming uniqueness of the invariant law, $`\nu ^{}=\nu _x^{c,w}`$ and therefore $`\nu _{x_n}^{c_n,w_n}\nu _x^{c,w}`$. The continuity of $`F_cw(x)`$ is a simple consequence of the continuity of $`\nu _x^{c,w}`$.
Proof of Lemma 2.2 Formula (2.1) (i) follows from the fact that rescaling the time in solutions $`(๐ฒ_t)_{t0}`$ to the martingale problem for $`A_x^{c,w}`$ by a factor $`\lambda `$ has no influence on the invariant law. Formula (2.1) (ii) is a direct consequence of formula (2.1) (i).
Proof of Lemma 2.3 This follows by inserting the functions $`f(x)=x_i`$ and $`f(x)=x_ix_j`$ into the equilibrium equation (5.1).
Proof of Lemma 2.4 If $`x_wD`$, then $`๐ฒ_t:=x`$ ($`t0`$) is a stationary solution to the martingale problem for $`A_x^{c,w}`$, and therefore $`\nu _x^{c,w}=\delta _x`$ and $`F_cw(x)=w(x)=0`$. On the other hand, if $`x_wD`$, then $`๐ฒ_t:=x`$ ($`t0`$) is not a stationary solution to the martingale problem for $`A_x^{c,w}`$ and therefore $`_{\overline{D}}\nu _x^{c,w}(\mathrm{d}y)|yx|^2>0`$. Let $`\mathrm{tr}(w(y)):=_iw_{ii}(y)`$ denote the trace of $`w(y)`$. By (2.2) (ii), $`\frac{1}{c}\mathrm{tr}(F_cw)(x)=\frac{1}{c}_{\overline{D}}\nu _x^{c,w}(\mathrm{d}y)\mathrm{tr}(w(y))=_{\overline{D}}\nu _x^{c,w}(\mathrm{d}y)|yx|^2>0`$ and therefore $`F_cw(x)0`$.
From now on assume that $`๐ฒ`$ is a renormalization class. Note that
$$K^{w,(n)}=\nu ^{c_{n1},F^{(n1)}w}\mathrm{}\nu ^{c_0,w}(n1),$$
(5.2)
where we denote the composition of two probability kernels $`K,L`$ on $`\overline{D}`$ by
$$(KL)_x(\mathrm{d}z):=_{\overline{D}}K_x(\mathrm{d}y)L_y(\mathrm{d}z).$$
(5.3)
Proof of Lemma 2.5 This is a direct consequence of Lemmas 2.1 and 2.3. In particular, the relations (2.6) follow by iterating the relations (2.2).
Proof of Lemma 2.6 Recall that $`\mathrm{tr}(w(y))`$ denotes the trace of $`w(y)`$. Formulas (2.5) and (2.6) (ii) show that
$$_{\overline{D}}K_x^{w,(n)}(\mathrm{d}y)|yx|^2=s_n_{\overline{D}}K_x^{w,(n)}(\mathrm{d}y)\mathrm{tr}(w(y)).$$
(5.4)
Since $`\overline{D}`$ is compact, the left-hand side of this equation is bounded uniformly in $`x\overline{D}`$ and $`n1`$, and therefore, since we are assuming $`s_n\mathrm{}`$,
$$\underset{n\mathrm{}}{lim}\underset{xD}{sup}_{\overline{D}}K_x^{w,(n)}(\mathrm{d}y)\mathrm{tr}(w(y))=0.$$
(5.5)
Since $`w`$ is symmetric and nonnegative definite, $`\mathrm{tr}(w(y))`$ is nonnegative, and zero if and only if $`y_wD`$. If $`f๐(\overline{D})`$ satisfies $`f=0`$ on $`_wD`$, then, for every $`\epsilon >0`$, the sets $`C_m:=\{x\overline{D}:|f(x)|\epsilon +m\mathrm{tr}(w(x))\}`$ are compact with $`C_m\mathrm{}`$ as $`m\mathrm{}`$, so there exists an $`m`$ (depending on $`\epsilon `$) such that $`|f|<\epsilon +m\mathrm{tr}(w)`$. Therefore,
$$\begin{array}{c}\underset{n\mathrm{}}{lim\; sup}\underset{x\overline{D}}{sup}\left|_{\overline{D}}K_x^{w,(n)}(\mathrm{d}y)f(y)\right|\underset{n\mathrm{}}{lim\; sup}\underset{x\overline{D}}{sup}_{\overline{D}}K_x^{w,(n)}(\mathrm{d}y)|f(y)|\hfill \\ \epsilon +m\underset{n\mathrm{}}{lim\; sup}\underset{x\overline{D}}{sup}_{\overline{D}}K_x^{w,(n)}(\mathrm{d}y)\mathrm{tr}(w(y))=\epsilon .\hfill \end{array}$$
(5.6)
Since $`\epsilon >0`$ is arbitrary, (2.7) follows.
Proof of Lemma 2.8 By (2.10), (2.12), and (2.13), $`w_\gamma ^{}^{}=lim_n\mathrm{}(\overline{F}_\gamma ^{})^nw`$ for each $`w๐ฒ`$. By Lemma 2.1 (b), $`\overline{F}_\gamma ^{}:๐ฒ๐ฒ`$ is continuous, so $`w_\gamma ^{}^{}`$ is the unique fixed point of $`\overline{F}_\gamma ^{}`$. This proves part (a).
Now let $`0w๐ฒ`$ and assume that $`\widehat{๐ฒ}=\{\lambda w:\lambda >0\}`$ is a fixed shape. Then $`\widehat{๐ฒ}s_nF^{(n)}w\underset{n\mathrm{}}{}w_\gamma ^{}^{}`$ whenever $`s_n\mathrm{}`$ and $`s_{n+1}/s_n1+\gamma ^{}`$ for some $`0<\gamma ^{}<\mathrm{}`$, which shows that $`\widehat{๐ฒ}=\{\lambda w_\gamma ^{}^{}:\lambda >0\}`$. Thus, $`๐ฒ`$ can contain at most one fixed shape, and if it does, then the $`w_\gamma ^{}^{}`$ for different values of $`\gamma ^{}`$ must be constant multiples of each other. This proves part (c) and the uniqueness statement in part (b).
To complete the proof of part (b), note that if $`w^{}=w_\gamma ^{}^{}`$ does not depend on $`\gamma ^{}`$, then $`w^{}๐ฒ`$ solves (2.16) (i) for all $`0<\gamma ^{}<\mathrm{}`$, hence $`F_cw^{}=(1+\frac{1}{c})^1w^{}`$ for all $`c>0`$, and therefore, by scaling (Lemma 2.2), $`F_c(\lambda w^{})=\lambda F_{c/\lambda }(w^{})=\lambda (1+\frac{\lambda }{c})^1w^{}=(\frac{1}{\lambda }+\frac{1}{c})^1w^{}`$.
### 5.2 Coupling of catalytic Wright-Fisher diffusions
In this section we verify condition (i) of Definition 1.1 for the class $`๐ฒ_{\mathrm{cat}}`$, and we prepare for the verification of conditions (ii)โ(iv) in Section 5.3. In fact, we will show that the larger class $`\overline{๐ฒ}_{\mathrm{cat}}:=\{w^{\alpha ,p}:\alpha >0,p๐_+[0,1]\}`$ is also a renormalization class, and the equivalents of Theorem 1.4 (a) and Proposition 3.1 remain true for this larger class. (We do not know, however, if the convergence statements in Theorem 1.4 (b) also hold in this larger class; see the discussion in Section 4.2.)
For each $`c0`$, $`w\overline{๐ฒ}_{\mathrm{cat}}`$ and $`x[0,1]^2`$, the operator $`A_x^{c,w}`$ is a densely defined linear operator on $`๐([0,1]^2)`$ that maps the identity function into zero and, as one easily verifies, satisfies the positive maximum principle. Since $`[0,1]^2`$ is compact, the existence of a solution to the martingale problem for $`A_x^{c,w}`$, for each $`[0,1]^2`$-valued initial condition, now follows from general theory (see \[RW87\], Theorem 5.23.5, or \[EK86, Theorem 4.5.4 and Remark 4.5.5\]).
We are therefore left with the task of verifying uniqueness of solutions to the martingale problem for $`A_x^{c,w}`$. By \[EK86, Problem 4.19, Corollary 5.3.4, and Theorem 5.3.6\], it suffices to show that solutions to (1.7) are pathwise unique.
###### Lemma 5.1 (Monotone coupling of Wright-Fisher diffusions)
Assume that $`0x\stackrel{~}{x}1`$, $`c0`$ and that $`(P_t)_{t0}`$ is a progressively measurable, nonnegative process such that $`sup_{t0,\omega \mathrm{\Omega }}P_t(\omega )<\mathrm{}`$. Let $`๐ฒ,\stackrel{~}{๐ฒ}`$ be $`[0,1]`$-valued solutions to the SDEโs
$$\begin{array}{ccc}\hfill \mathrm{d}๐ฒ_t& =& c(x๐ฒ_t)\mathrm{d}t+\sqrt{2P_t๐ฒ_t(1๐ฒ_t)}\mathrm{d}B_t,\hfill \\ \hfill \mathrm{d}\stackrel{~}{๐ฒ}_t& =& c(\stackrel{~}{x}\stackrel{~}{๐ฒ}_t)\mathrm{d}t+\sqrt{2P_t\stackrel{~}{๐ฒ}_t(1\stackrel{~}{๐ฒ}_t)}\mathrm{d}B_t,\hfill \end{array}$$
(5.7)
where in both equations $`B`$ is the same Brownian motion. If $`๐ฒ_0\stackrel{~}{๐ฒ}_0`$ a.s., then
$$๐ฒ_t\stackrel{~}{๐ฒ}_tt0\text{a.s.}$$
(5.8)
Proof This is an easy adaptation of a technique due to Yamada and Watanabe \[YW71\]. Since $`_{0+}\frac{\mathrm{d}x}{x}=\mathrm{}`$, it is possible to choose $`\rho _n๐[0,\mathrm{})`$ such that $`_0^{\mathrm{}}\rho _n(x)dx=1`$ and
$$0\rho _n(x)\frac{1}{nx}1_{(0,1]}(x)(x0).$$
(5.9)
Define $`\varphi _n๐^{(2)}()`$ by
$$\varphi _n(x):=_0^{x0}dy_0^ydz\rho _n(z).$$
(5.10)
One easily verifies that $`\varphi _n(x)`$, $`x\varphi _n^{}(x)`$, and $`x\varphi _n^{\prime \prime }(x)`$ are nonnegative and converge, as $`n\mathrm{}`$, to $`x0`$, $`x0`$, and $`0`$, respectively. By Itรดโs formula:
$$\begin{array}{cccc}\hfill E[\varphi _n(๐ฒ_t\stackrel{~}{๐ฒ}_t)]& =& E[\varphi _n(๐ฒ_0\stackrel{~}{๐ฒ}_0)]\hfill & \hfill (\mathrm{i})\\ & & +c(x\stackrel{~}{x})_0^tE[\varphi _n^{}(๐ฒ_s\stackrel{~}{๐ฒ}_s)]dsc_0^tE[(๐ฒ_s\stackrel{~}{๐ฒ}_s)\varphi _n^{}(๐ฒ_s\stackrel{~}{๐ฒ}_s)]ds\hfill & \hfill (\mathrm{ii})\\ & & +_0^tE\left[P_s\left(\sqrt{๐ฒ_s(1๐ฒ_s)}\sqrt{\stackrel{~}{๐ฒ}_s(1\stackrel{~}{๐ฒ}_s)}\right)^2\varphi _n^{\prime \prime }(๐ฒ_s\stackrel{~}{๐ฒ}_s)\right]ds.\hfill & \hfill (\mathrm{iii})\end{array}$$
(5.11)
Here the terms in (ii) are nonpositive, and hence, letting $`n\mathrm{}`$ and using the elementary estimate
$$|\sqrt{y(1y)}\sqrt{\stackrel{~}{y}(1\stackrel{~}{y})}||y\stackrel{~}{y}|^{\frac{1}{2}}(y,\stackrel{~}{y}[0,1]),$$
(5.12)
the properties of $`\varphi _n`$, and the fact that the process $`P`$ is uniformly bounded, we find that
$$E[0(๐ฒ_t\stackrel{~}{๐ฒ}_t)]E[0(๐ฒ_0\stackrel{~}{๐ฒ}_0)]=0,$$
(5.13)
by our assumption that $`๐ฒ_0\stackrel{~}{๐ฒ}_0`$. This shows that $`๐ฒ_t\stackrel{~}{๐ฒ}_t`$ a.s. for each fixed $`t0`$, and by the continuity of sample paths the statement holds for all $`t0`$ almost surely.
###### Corollary 5.2 (Pathwise uniqueness)
For all $`c0`$, $`\alpha >0`$, $`p๐_+[0,1]`$ and $`x[0,1]`$, solutions to the SDE (1.7) are pathwise unique.
Proof Let $`(๐ฒ^1,๐ฒ^2)`$ and $`(\stackrel{~}{๐ฒ}^1,\stackrel{~}{๐ฒ}^2)`$ be solutions to (1.7) relative to the same pair $`(B^1,B^2)`$ of Brownian motions, with $`(๐ฒ_0^1,๐ฒ_0^2)=(\stackrel{~}{๐ฒ}_0^1,\stackrel{~}{๐ฒ}_0^2)`$. Applying Lemma 5.1, with inequality in both directions, we see that $`๐ฒ^1=\stackrel{~}{๐ฒ}^1`$ a.s. Applying Lemma 5.1 two more times, this time using that $`๐ฒ^1=\stackrel{~}{๐ฒ}^1`$ a.s., we see that also $`๐ฒ^2=\stackrel{~}{๐ฒ}^2`$ a.s.
###### Corollary 5.3 (Exponential coupling)
Assume that $`x[0,1]`$, $`c0`$, and $`\alpha >0`$. Let $`๐ฒ,\stackrel{~}{๐ฒ}`$ be solutions to the SDE
$$\mathrm{d}๐ฒ_t=c(x๐ฒ_t)\mathrm{d}t+\sqrt{2\alpha ๐ฒ_t(1๐ฒ_t)}\mathrm{d}B_t,$$
(5.14)
relative to the same Brownian motion $`B`$. Then
$$E\left[|\stackrel{~}{๐ฒ}_t๐ฒ_t|\right]=e^{ct}E\left[|\stackrel{~}{๐ฒ}_0๐ฒ_0|\right].$$
(5.15)
Proof If $`๐ฒ_0=y`$ and $`\stackrel{~}{๐ฒ}_0=\stackrel{~}{y}`$ are deterministic and $`y\stackrel{~}{y}`$, then by Lemma 5.1 and a simple moment calculation
$$E\left[|\stackrel{~}{๐ฒ}_t๐ฒ_t|\right]=E[\stackrel{~}{๐ฒ}_t๐ฒ_t]=e^{ct}|\stackrel{~}{y}y|.$$
(5.16)
The same argument applies when $`y\stackrel{~}{y}`$. The general case where $`๐ฒ_0`$ and $`\stackrel{~}{๐ฒ}_0`$ are random follows by conditioning on $`(๐ฒ_0,\stackrel{~}{๐ฒ}_0)`$.
###### Corollary 5.4 (Ergodicity)
The Markov process defined by the SDE (3.6) has a unique invariant law $`\mathrm{\Gamma }_x^\gamma `$ and is ergodic, i.e, solutions to (3.6) started in an arbitrary initial law $`(๐ฒ_0)`$ satisfy $`(๐ฒ_t)\underset{t\mathrm{}}{}\mathrm{\Gamma }_x^\gamma `$.
Proof Since our process is a Feller diffusion on a compactum, the existence of an invariant law follows from a simple time averaging argument. Now start one solution $`\stackrel{~}{๐ฒ}`$ of (3.6) in this invariant law and let $`๐ฒ`$ be any other solution, relative to the same Brownian motion. Corollary 5.3 then gives ergodicity and, in particular, uniqueness of the invariant law.
###### Remark 5.5
(Density of invariant law) It is well-known (see, for example \[Ewe04, formula (5.70)\]) that $`\mathrm{\Gamma }_x^\gamma `$ is a $`\beta (\alpha _1,\alpha _2)`$-distribution, where $`\alpha _1:=x/\gamma `$ and $`\alpha _2:=(1x)/\gamma `$, i.e., $`\mathrm{\Gamma }_x^\gamma =\delta _x`$ $`(x\{0,1\})`$ and
$$\mathrm{\Gamma }_x^\gamma (\mathrm{d}y)=\frac{\mathrm{\Gamma }(\alpha _1+\alpha _2)}{\mathrm{\Gamma }(\alpha _1)\mathrm{\Gamma }(\alpha _2)}y^{\alpha _11}(1y)^{\alpha _21}\mathrm{d}y(x(0,1)).$$
(5.17)
$`\mathrm{}`$
We conclude this section with a lemma that prepares for the verification of condition (iv) in Definition 1.1 for the class $`๐ฒ_{\mathrm{cat}}`$.
###### Lemma 5.6
(Monotone coupling of stationary Wright-Fisher diffusions) Assume that $`c>0`$, $`\alpha >0`$ and $`0x\stackrel{~}{x}1`$. Then the pair of equations
$$\begin{array}{ccc}\hfill \mathrm{d}๐ฒ_t& =& c(x๐ฒ_t)\mathrm{d}t+\sqrt{2\alpha ๐ฒ_t(1๐ฒ_t)}\mathrm{d}B_t,\hfill \\ \hfill \mathrm{d}\stackrel{~}{๐ฒ}_t& =& c(\stackrel{~}{x}\stackrel{~}{๐ฒ}_t)\mathrm{d}t+\sqrt{2\alpha \stackrel{~}{๐ฒ}_t(1\stackrel{~}{๐ฒ}_t)}\mathrm{d}B_t\hfill \end{array}$$
(5.18)
has a unique stationary solution $`(๐ฒ_t,\stackrel{~}{๐ฒ}_t)_t`$. This stationary solution satisfies
$$๐ฒ_t\stackrel{~}{๐ฒ}_tt\text{a.s.}$$
(5.19)
Proof Let $`(๐ฒ_t,\stackrel{~}{๐ฒ}_t)_{t0}`$ be a solution of (5.18) and let $`(๐ฒ_t^{},\stackrel{~}{๐ฒ}_t^{})_{t0}`$ be another one, relative to the same Brownian motion $`B`$. Then, by Lemma 5.3, $`E[|๐ฒ_t๐ฒ_t^{}|]0`$ and also $`E[|\stackrel{~}{๐ฒ}_t\stackrel{~}{๐ฒ}_t^{}|]0`$ as $`t\mathrm{}`$. Hence we may argue as in the proof of Corollary 5.4 that (5.18) has a unique invariant law and is ergodic. Now start a solution of (5.18) in an initial condition such that $`๐ฒ_0\stackrel{~}{๐ฒ}_0`$. By ergodicity, the law of this solution converges as $`t\mathrm{}`$ to the invariant law of (5.18) and using Lemma 5.1 we see that this invariant law is concentrated on $`\{(y,\stackrel{~}{y})[0,1]^2:y\stackrel{~}{y}\}`$. Now consider, on the whole real time axis, the stationary solution to (5.18) with this invariant law. Applying Lemma 5.1 once more, we see that (5.19) holds.
### 5.3 Duality for catalytic Wright-Fisher diffusions
In this section we prove Theorem 1.4 (a) and Proposition 3.1. Moreover, we will show that their statements remain true if the renormalization class $`๐ฒ_{\mathrm{cat}}`$ is replaced by the larger class $`\overline{๐ฒ}_{\mathrm{cat}}:=\{w^{\alpha ,p}:\alpha >0,p๐_+[0,1]\}`$. We begin by recalling the usual moment duality for Wright-Fisher diffusions.
For $`\gamma >0`$ and $`x[0,1]`$, let $`๐ฒ`$ be a solution to the SDE
$$\mathrm{d}๐ฒ(t)=\frac{1}{\gamma }(x๐ฒ(t))\mathrm{d}t+\sqrt{2๐ฒ(t)(1๐ฒ(t))}\mathrm{d}B(t),$$
(5.20)
i.e., $`๐ฒ`$ is a Wright-Fisher diffusion with a linear drift towards $`x`$. It is well-known that $`๐ฒ`$ has a moment dual. To be precise, let $`(\varphi ,\psi )`$ be a Markov process in $`^2=\{0,1,\mathrm{}\}^2`$ that jumps as:
$$\begin{array}{cccc}\hfill (\varphi _t,\psi _t)& & (\varphi _t1,\psi _t)\hfill & \text{with rate }\varphi _t(\varphi _t1)\hfill \\ \hfill (\varphi _t,\psi _t)& & (\varphi _t1,\psi _t+1)\hfill & \text{with rate }\frac{1}{\gamma }\varphi _t.\hfill \end{array}$$
(5.21)
Then one has the following duality relation (see for example Lemma 2.3 in \[Shi80\] or Proposition 1.5 in \[GKW01\])
$$E^y\left[๐ฒ_t^nx^m\right]=E^{(n,m)}\left[y^{\varphi _t}x^{\psi _t}\right](y[0,1],(n,m)^2),$$
(5.22)
where $`0^0:=1`$. The duality in (5.22) has the following heuristic explanation. Consider a population containing a fixed, large number of organisms, that come in two genetic types, say I and II. Each pair of organisms in the population is resampled with rate $`2`$. This means that one organism of the pair (chosen at random) dies, while the other organism produces one child of its own genetic type. Moreover, each organism is replaced with rate $`\frac{1}{\gamma }`$ by an organism chosen from an infinite reservoir where the frequency of type I has the fixed value $`x`$. In the limit that the number of organisms in the population is large, the relative frequency $`๐ฒ_t`$ of type I organisms follows the SDE (5.20). Now $`E[๐ฒ_t^n]`$ is the probability that $`n`$ organisms sampled from the population at time $`t`$ are all of type I. In order to find this probability, we follow the ancestors of these organisms back in time. Viewed backwards in time, these ancestors live for a while in the population, until, with rate $`\frac{1}{\gamma }`$, they jump to the infinite reservoir. Moreover, due to resampling, each pair of ancestors coalesces with rate $`2`$ to one common ancestor. Denoting the number of ancestors that lived at time $`ts`$ in the population and in the reservoir by $`\varphi _s`$ and $`\psi _s`$, respectively, we see that the probability that all ancestors are of type I is $`E^y[๐ฒ_t^n]=E^{(n,0)}[y^{\varphi _t}x^{\psi _t}]`$. This gives a heuristic explanation of (5.22).
Since eventually all ancestors of the process $`(\varphi ,\psi )`$ end up in the reservoir, we have $`(\varphi _t,\psi _t)(0,\psi _{\mathrm{}})`$ as $`t\mathrm{}`$ a.s. for some $``$-valued random variable $`\psi _{\mathrm{}}`$. Taking the limit $`t\mathrm{}`$ in (5.22), we see that the moments of the invariant law $`\mathrm{\Gamma }_x^\gamma `$ from Corollary 5.4 are given by:
$$\mathrm{\Gamma }_x^\gamma (\mathrm{d}y)y^n=E^{(n,0)}[x^\psi _{\mathrm{}}](n0).$$
(5.23)
It is not hard to obtain an inductive formula for the moments of $`\mathrm{\Gamma }_x^\gamma `$, which can then be solved to yield the formula
$$\mathrm{\Gamma }_x^\gamma (\mathrm{d}y)y^n=\underset{k=0}{\overset{n1}{}}\frac{x+k\gamma }{1+k\gamma }(n1).$$
(5.24)
In particular, it follows that
$$\mathrm{\Gamma }_x^\gamma (\mathrm{d}y)y(1y)=\frac{1}{1+\gamma }x(1x).$$
(5.25)
This is the important fixed shape property of the Wright-Fisher diffusion (see formula (2.17)).
We now consider catalytic Wright-Fisher diffusions $`(๐ฒ^1,๐ฒ^2)`$ as in (1.7) with $`p๐_+[0,1]`$ and apply duality to the catalyst $`๐ฒ^2`$ conditioned on the reactant $`๐ฒ^1`$. Let $`(๐ฒ_t^1,๐ฒ_t^2)_t`$ be a stationary solution to the SDE (1.7) with $`c=1/\gamma `$. Let $`(\stackrel{~}{\varphi },\stackrel{~}{\psi })`$ be a $`^2`$-valued process, defined on the same probability space as $`(๐ฒ^1,๐ฒ^2)`$, such that conditioned on the past path $`(๐ฒ_t^1)_{t0}`$, the process $`(\stackrel{~}{\varphi },\stackrel{~}{\psi })`$ is a (time-inhomogeneous) Markov process that jumps as:
$$\begin{array}{cccc}\hfill (\stackrel{~}{\varphi }_t,\stackrel{~}{\psi }_t)& & (\stackrel{~}{\varphi }_t1,\stackrel{~}{\psi }_t)\hfill & \text{with rate }p(๐ฒ_t^1)\stackrel{~}{\varphi }_t(\stackrel{~}{\varphi }_t1),\hfill \\ \hfill (\stackrel{~}{\varphi }_t,\stackrel{~}{\psi }_t)& & (\stackrel{~}{\varphi }_t1,\stackrel{~}{\psi }_t+1)\hfill & \text{with rate }\frac{1}{\gamma }\stackrel{~}{\varphi }_t.\hfill \end{array}$$
(5.26)
Then, in analogy with (5.22),
$$E[(๐ฒ_0^2)^nx_2^m|(๐ฒ_t^1)_{t0}]=E^{(n,m)}[(๐ฒ_t^2)^{\stackrel{~}{\varphi }_t}x_2^{\stackrel{~}{\psi }_t}|(๐ฒ_t^1)_{t0}]((n,m)^2,t0).$$
(5.27)
We may interpret (5.26) by saying that pairs of ancestors in a finite population coalesce with time-dependent rate $`2p(๐ฒ_t^1)`$ and ancestors jump to an infinite reservoir with constant rate $`\frac{1}{\gamma }`$. Again, eventualy all ancestors end up in the reservoir, and therefore $`(\stackrel{~}{\varphi }_t,\stackrel{~}{\psi }_t)(0,\stackrel{~}{\psi }_{\mathrm{}})`$ as $`t\mathrm{}`$ a.s. for some $``$-valued random variable $`\stackrel{~}{\psi }_{\mathrm{}}`$. Taking the limit $`t\mathrm{}`$ in (5.27) we find that
$$E[(๐ฒ_0^2)^nx_2^m|(๐ฒ_t^1)_{t0}]=E^{(n,m)}[x_2^{\stackrel{~}{\psi }_{\mathrm{}}}|(๐ฒ_t^1)_{t0}]((n,m)^2,t0).$$
(5.28)
###### Lemma 5.7
(Uniqueness of invariant law) For each $`c>0`$, $`w\overline{๐ฒ}_{\mathrm{cat}}`$, and $`x[0,1]^2`$, there exists a unique invariant law $`\nu _x^{c,w}`$ for the martingale problem for $`A_x^{c,w}`$.
Proof Our process being a Feller diffusion on a compactum, the existence of an invariant law follows from time averaging. We need to show uniqueness. If $`(๐ฒ^1,๐ฒ^2)=๐ฒ_t^1,๐ฒ_t^2)_t`$ is a stationary solution, then $`๐ฒ^1`$ is an autonomous process, and $`(๐ฒ_0^1)=\mathrm{\Gamma }_x^{1/c}`$, the unique invariant law from Corollary 5.4. Therefore, $`((๐ฒ_t^1)_t)`$ is determined uniquely by the requirement that $`(๐ฒ^1,๐ฒ^2)`$ be stationary. By (5.28), the conditional distribution of $`๐ฒ_0^2`$ given $`(๐ฒ_t^1)_{t0}`$ is determined uniquely, and therefore the joint distribution of $`๐ฒ_0^2`$ and $`(๐ฒ_t^1)_{t0}`$ is determined uniquely. In particular, $`(๐ฒ_0^1,๐ฒ_0^2)=\nu _x^{c,w}`$ is determined uniquely.
###### Remark 5.8
(Reversibility) It seems that the invariant law $`\nu _x^{c,w}`$ from Lemma 5.7 is reversible. In many cases (densities of) reversible invariant measures can be obtained in closed form by solving the equations of detailed balance. This is the case, for example, for the one-dimensional Wright-Fisher diffusion. We have not attempted this for the catalytic Wright-Fisher diffusion. $`\mathrm{}`$
The next proposition implies Proposition 3.1 and prepares for the proof of Theorem 1.4 (a).
###### Proposition 5.9
(Extended renormalization class) The set $`\overline{๐ฒ}_{\mathrm{cat}}`$ is a renormalization class on $`[0,1]^2`$, and
$$\overline{F}_\gamma w^{1,p}=w^{1,๐ฐ_\gamma p}(p๐_+[0,1],\gamma >0).$$
(5.29)
Proof To see that $`\overline{๐ฒ}_{\mathrm{cat}}`$ is a renormalization class we need to check conditions (i)โ(iv) from Definition 1.1. By Lemma 5.2, the martingale problem for $`A_x^{c,w}`$ is well-posed for all $`c0`$, $`w๐ฒ_{\mathrm{cat}}`$ and $`x[0,1]^2`$. By Lemma 5.7, the corresponding Feller process on $`[0,1]^2`$ has a unique invariant law $`\nu _x^{c,w}`$. This shows that conditions (i) and (ii) from Definition 1.1 are satisfied. Note that by the compactness of $`[0,1]^2`$, any continuous function on $`[0,1]^2`$ is bounded, so condition (iii) is automatically satisfied. Hence $`๐ฒ`$ is a prerenormalization class. As a consequence, for any $`p๐_+[0,1]`$, $`\overline{F}_\gamma w^{1,p}`$ is well-defined by (1.2) and (2.9). We will now first prove (5.29) and then show that $`\overline{๐ฒ}_{\mathrm{cat}}`$ is a renormalization class.
Fix $`\gamma >0`$, $`p๐_+[0,1]`$, and $`x[0,1]^2`$. Let $`(๐ฒ_t^1,๐ฒ_t^2)_t`$ be a stationary solution to the SDE (1.7) with $`\alpha =1`$ and $`c=1/\gamma `$. Then
$$\overline{F}_\gamma w_{ij}^{1,p}(x)=(1+\gamma )E[w_{ij}^{1,p}(๐ฒ_0^1,๐ฒ_0^2)](i,j=1,2).$$
(5.30)
Since $`w_{ij}^{1,p}=0`$ if $`ij`$, it is clear that $`\overline{F}_\gamma w_{ij}^{1,p}(x)=0`$ if $`ij`$. Since $`(๐ฒ_0^1)=\mathrm{\Gamma }_x^\gamma `$ it follows from (5.25) that $`\overline{F}_\gamma w_{11}^{1,p}(x)=x_1(1x_1)`$. We are left with the task of showing that
$$\overline{F}_\gamma w_{22}^{1,p}(x)=๐ฐ_\gamma p(x_1)x_2(1x_2).$$
(5.31)
Here, by (2.2) (ii),
$$\begin{array}{ccc}\hfill \overline{F}_\gamma w_{22}^{1,p}(x)& =& (1+\gamma )E[p(๐ฒ_0^1)๐ฒ_0^2(1๐ฒ_0^2)]\hfill \\ & =& (\frac{1}{\gamma }+1)E[(๐ฒ_0^2x_2)^2].\hfill \end{array}$$
(5.32)
By (5.28), using the fact that $`E[๐ฒ_0^2]=x_2`$ (which follows from (5.27) or more elementary from (2.6) (i)), we find that
$$E[(๐ฒ_0^2x_2)^2]=E[(๐ฒ_0^2)^2](x_2)^2=E^{(2,0)}[x_2^{\stackrel{~}{\psi }_{\mathrm{}}}](x_2)^2=P^{(2,0)}[\stackrel{~}{\psi }_{\mathrm{}}=1]x_2(1x_2)(t0).$$
(5.33)
Note that $`P^{(2,0)}[\stackrel{~}{\psi }_{\mathrm{}}=1]`$ is the probability that the two ancestors coalesce before one of them leaves the population. The probability of noncoalescence is given by
$$P^{(2,0)}[\stackrel{~}{\psi }_{\mathrm{}}=2]=E\left[\text{e}^{_0^{{\scriptscriptstyle \frac{1}{2}}\tau _\gamma }2p\left(y_t^1\right)dt}\right],$$
(5.34)
where $`\tau _\gamma `$ is an exponentially distributed random variable with mean $`\gamma `$. Combining this with (5.32) and (5.33) we find that
$$\begin{array}{ccc}\hfill \overline{F}_\gamma w_{22}^{1,p}(x)& =& (\frac{1}{\gamma }+1)E\left[1\text{e}^{_0^{\tau _\gamma }p\left(y_{t/2}^1\right)dt}\right]x_2(1x_2)\hfill \\ & =& q_\gamma E\left[1\text{e}^{๐ต_x^\gamma ,p}\right]x_2(1x_2)\hfill \\ & =& ๐ฐ_\gamma p(x_1)x_2(1x_2),\hfill \end{array}$$
(5.35)
where we have used the definition of $`๐ฐ_\gamma `$.
We still have to show that $`\overline{๐ฒ}_{\mathrm{cat}}`$ satisfies condition (iv) from Definition 1.1. For any $`\alpha >0`$ and $`p๐_+[0,1]`$, by scaling (Lemma 2.2) and (5.29),
$$F_cw^{\alpha ,p}=\alpha F_{\frac{c}{\alpha }}w^{1,{\scriptscriptstyle \frac{p}{\alpha }}}=\alpha (1+\frac{\alpha }{c})^1\overline{F}_{\frac{c}{\alpha }}w^{1,{\scriptscriptstyle \frac{p}{\alpha }}}=w^{\left({\scriptscriptstyle \frac{1}{\alpha }}+{\scriptscriptstyle \frac{1}{c}}\right)^1,\left({\scriptscriptstyle \frac{1}{\alpha }}+{\scriptscriptstyle \frac{1}{c}}\right)^1๐ฐ_{{\scriptscriptstyle \frac{c}{\alpha }}}\left({\scriptscriptstyle \frac{p}{\alpha }}\right)}.$$
(5.36)
By Lemma 2.1, this diffusion matrix is continuous, which implies that $`๐ฐ_{\frac{c}{\alpha }}(\frac{p}{\alpha })`$ is continuous.
Our proof of Propostion 5.9 has a corollary.
###### Corollary 5.10
(Continuity in parameters) The map $`(x,\gamma )๐ฌ_\gamma (x,)`$ from $`[0,1]\times (0,\mathrm{})`$ to $`_1([0,1])`$ and the map $`(x,\gamma ,p)๐ฐ_\gamma p(x)`$ from $`[0,1]\times (0,\mathrm{})\times ๐_+[0,1]`$ to $``$ are continuous.
Proof By Lemma 2.1, the diffusion matrix in (5.36) is continuous in $`x,\gamma `$, and $`p`$, which implies the continuity of $`๐ฐ_\gamma p(x)`$. It follows that the map $`(x,\gamma )๐ฌ_\gamma (x,\mathrm{d}\chi )\text{e}^{\chi ,f}`$ is continuous for all $`f๐_+[0,1]`$, so by \[Kal76, Theorem 4.2\], $`(x,\gamma )๐ฌ_\gamma (x,)`$ is continuous.
Proof of Theorem 1.4 (a) We need to show that $`๐ฒ_{\mathrm{cat}}`$ is a renormalization class and that $`F_c`$ maps the subclasses $`๐ฒ_{\mathrm{cat}}^{l,r}`$ into themselves. It has already been explained in Section 2 that the latter fact is a consequence of Lemma 2.4. Since in Proposition 5.9 it has been shown that $`\overline{๐ฒ}_{\mathrm{cat}}`$ is a renormalization class, we are left with the task to show that $`F_c`$ maps $`๐ฒ_{\mathrm{cat}}`$ into itself. By (5.29) and scaling, it suffices to show that $`๐ฐ_\gamma `$ maps $``$ into itself.
Fix $`0x\stackrel{~}{x}1`$. By Lemma 5.6, we can couple the processes $`๐ฒ_x^\gamma `$ and $`๐ฒ_{\stackrel{~}{x}}^\gamma `$ from (3.6) such that
$$๐ฒ_x^\gamma (t)๐ฒ_{\stackrel{~}{x}}^\gamma (t)t0\text{a.s.}$$
(5.37)
Since the function $`z1e^z`$ on $`[0,\mathrm{})`$ is Lipschitz continuous with Lipschitz constant $`1`$,
$$\begin{array}{c}\left|๐ฐ_\gamma p(\stackrel{~}{x})๐ฐ_\gamma p(x)\right|\hfill \\ =\left|(\frac{1}{\gamma }+1)E\left[1\text{e}^{_0^{\tau _\gamma }p\left(๐ฒ_{\stackrel{~}{x}}^\gamma \left(t/2\right)\right)dt}\right](\frac{1}{\gamma }+1)E\left[1\text{e}^{_0^{\tau _\gamma }p\left(๐ฒ_x^\gamma \left(t/2\right)\right)dt}\right]\right|\hfill \\ (\frac{1}{\gamma }+1)E\left[_0^{\tau _\gamma }\left|p(๐ฒ_{\stackrel{~}{x}}^\gamma (t/2))p(๐ฒ_x^\gamma (t/2))\right|dt\right]\hfill \\ (\frac{1}{\gamma }+1)LE\left[_0^{\tau _\gamma }\left|๐ฒ_{\stackrel{~}{x}}^\gamma (t/2)๐ฒ_x^\gamma (t/2)\right|dt\right]\hfill \\ =(\frac{1}{\gamma }+1)L\gamma (\stackrel{~}{x}x)=L(1+\gamma )|\stackrel{~}{x}x|,\hfill \end{array}$$
(5.38)
where $`L`$ is the Lipschitz constant of $`p`$ and we have used the same exponentially distributed $`\tau _\gamma `$ for $`๐ฒ_x^\gamma `$ and $`๐ฒ_{\stackrel{~}{x}}^\gamma `$.
### 5.4 Monotone and concave catalyzing functions
In this section we prove that the log-Laplace operators $`๐ฐ_\gamma `$ from (3.9) map monotone functions into monotone functions, and monotone concave functions into monotone concave functions. We do not know if in general $`๐ฐ_\gamma `$ maps concave functions into concave functions.
###### Proposition 5.11
(Preservation of monotonicity and concavity) Let $`\gamma >0`$. Then:
(a) If $`f๐_+[0,1]`$ is nondecreasing, then $`๐ฐ_\gamma f`$ is nondecreasing.
(b) If $`f๐_+[0,1]`$ is nondecreasing and concave, then $`๐ฐ_\gamma f`$ is nondecreasing and concave.
Proof Our proof of Proposition 5.11 is in part based on ideas from \[BCGdH97, Appendix A\]. The proof is quite long and will depend on several lemmas. We remark that part (a) can be proved in a more elementary way using Lemma 5.6.
We recall some facts from Hille-Yosida theory. A linear operator $`A`$ on a Banach space $`V`$ is closable and its closure $`\overline{A}`$ generates a strongly continuous contraction semigroup $`(S_t)_{t0}`$ if and only if
$$\begin{array}{cc}\hfill (\mathrm{i})& ๐(A)\text{ is dense},\hfill \\ \hfill (\mathrm{ii})& A\text{ is dissipative},\hfill \\ \hfill (\mathrm{iii})& (1\alpha A)\text{ is dense for some, and hence for all }\alpha >0.\hfill \end{array}$$
(5.39)
Here, for any linear operator $`B`$ on $`V`$, $`๐(B)`$ and $`(B)`$ denote the domain and range of $`B`$, respectively. For each $`\alpha >0`$, the operator $`(1\alpha \overline{A}):๐(\overline{A})V`$ is a bijection and its inverse $`(1\alpha \overline{A})^1:V๐(\overline{A})`$ is a bounded linear operator, given by
$$(1\alpha \overline{A})^1u=_0^{\mathrm{}}S_tu\alpha ^1e^{t/\alpha }dt(uV,\alpha >0).$$
(5.40)
If $`E`$ is a compact metrizable space and $`๐(E)`$ is the Banach space of continuous real functions on $`E`$, equipped with the supremumnorm, then a linear operator $`A`$ on $`๐(E)`$ is closable and its closure $`\overline{A}`$ generates a Feller semigroup if and only if (see \[EK86, Theorem 4.2.2 and remarks on page 166\])
$$\begin{array}{cc}\hfill (\mathrm{i})& 1๐(\overline{A})\text{ and }\overline{A}1=0,\hfill \\ \hfill (\mathrm{ii})& ๐(A)\text{ is dense},\hfill \\ \hfill (\mathrm{iii})& A\text{ satisfies the positive maximum principle},\hfill \\ \hfill (\mathrm{iv})& (1\alpha A)\text{ is dense for some, and hence for all }\alpha >0.\hfill \end{array}$$
(5.41)
If $`\overline{A}`$ generates a Feller semigroup and $`g๐(E)`$, then the operator $`\overline{A}+g`$ (with domain $`๐(\overline{A}+g):=๐(\overline{A})`$) generates a strongly continuous semigroup $`(S_t^g)_{t0}`$ on $`๐(E)`$. If $`g0`$ then $`(S_t^g)_{t0}`$ is contractive. If $`(\xi _t)_{t0}`$ is the Feller process with generator $`\overline{A}`$, then one has the Feynman-Kac representation
$$S_t^gu(x)=E^x[u(\xi (t))\text{e}^{_0^tg\left(\xi \left(s\right)\right)ds}](t0,xE,g,u๐(E)).$$
(5.42)
Let $`๐^{(n)}([0,1]^2)`$ denote the space of continuous real functions on $`[0,1]^2`$ whose partial derivatives up to $`n`$-th order exist and are continuous on $`[0,1]^2`$ (including the boundary), and put $`๐^{(\mathrm{})}([0,1]^2):=_n๐^{(n)}([0,1]^2)`$. Define a linear operator $`B`$ on $`๐([0,1]^2)`$ with domain $`๐(B):=๐^{(\mathrm{})}([0,1]^2)`$ by
$$Bu(x,y):=y(1y)\frac{^2}{y^2}u(x,y)+\frac{1}{\gamma }(xy)\frac{}{y}u(x,y).$$
(5.43)
Below, we will prove:
###### Lemma 5.12
(Feller semigroup) The closure in $`๐([0,1]^2)`$ of the operator $`B`$ generates a Feller semigroup on $`๐([0,1]^2)`$.
Write
$$\begin{array}{ccc}๐_+& :=& \{u๐([0,1]^2):u0\},\hfill \\ ๐_{1+}& :=& \{u๐^{(1)}([0,1]^2):\frac{}{y}u,\frac{}{x}u0\},\hfill \\ ๐_{2+}& :=& \{u๐^{(2)}([0,1]^2):\frac{^2}{y^2}u,\frac{^2}{xy}u,\frac{^2}{x^2}u0\}.\hfill \end{array}$$
(5.44)
Let $`\overline{๐ฎ}`$ denote the closure of a set $`๐ฎ๐([0,1]^2)`$. We need the following lemma.
###### Lemma 5.13
(Preserved classes) Let $`g๐([0,1]^2)`$ and let $`(S_t^g)_{t0}`$ be the strongly continuous semigroup with generator $`\overline{B}+g`$. Then, for each $`t0`$:
(a) If $`g\overline{๐_{1+}}`$, then $`S_t^g`$ maps $`\overline{๐_+๐_{1+}}`$ into itself.
(b) If $`g\overline{๐_{1+}๐_{2+}}`$, then $`S_t^g`$ maps $`\overline{๐_+๐_{1+}๐_{2+}}`$ into itself.
To see why Lemma 5.13 implies Proposition 5.11, let $`(๐ฑ(t),๐ฒ(t))_{t0}`$ denote the Feller process in $`[0,1]^2`$ generated by $`\overline{B}`$. It is easy to see that $`๐ฑ(t)=๐ฑ(0)`$ a.s. for all $`t0`$. For fixed $`๐ฑ(0)=x`$, the process $`(๐ฒ(t))_{t0}`$ is the diffusion given by the SDE (5.20). Therefore, by Feynman-Kac, for each $`g๐([0,1]^2)`$,
$$E^y\left[\text{e}^{_0^tg(x,๐ฒ\left(s\right))ds}\right]=S_t^g1(x,y),$$
(5.45)
where $`1`$ denotes the constant function $`1๐([0,1]^2)`$. By (3.9),
$$๐ฐ_\gamma f(x)=(\frac{1}{\gamma }+1)\left(1\mathrm{\Gamma }_x^\gamma (\mathrm{d}y)E^y\left[\text{e}^{_0^{\tau _\gamma }f\left(๐ฒ_x\left(s\right)\right)ds}\right]\right)(f๐_+[0,1]),$$
(5.46)
where $`\mathrm{\Gamma }_x^\gamma `$ is the invariant law of $`(๐ฒ(t))_{t0}`$ from Corollary 5.4 and $`\tau _\gamma `$ is an exponential time with mean $`\gamma `$, independent of $`(๐ฒ(t))_{t0}`$. Setting $`g(x,y):=f(y)`$ in (5.45), using the ergodicity of $`(๐ฒ(t))_{t0}`$ (see Corollary 5.4), we find that for each $`z[0,1]`$ and $`t0`$,
$$\begin{array}{ccc}\hfill \mathrm{\Gamma }_x^\gamma (\mathrm{d}y)E^y\left[\text{e}^{_0^tf\left(๐ฒ\left(s\right)\right)ds}\right]& =& \underset{r\mathrm{}}{lim}P^z[๐ฒ(r)\mathrm{d}y]E^y\left[\text{e}^{_0^tg(x,๐ฒ\left(s\right))ds}\right]\hfill \\ & =& \underset{r\mathrm{}}{lim}S_r^0S_t^g1(x,z).\hfill \end{array}$$
(5.47)
It follows from Lemma 5.13 that for each fixed $`r,t`$, and $`z`$, the function $`xS_r^0S_t^g1(x,z)`$ is nondecreasing if $`f`$ is nonincreasing, and nondecreasing and convex if $`f`$ is nonincreasing and concave. Therefore, taking the expectation over the randomness of $`\tau _\gamma `$, the claims follow from (5.46) and (5.47).
We still need to prove Lemmas 5.12 and 5.13.
Proof of Lemma 5.12 It is easy to see that the operator $`B`$ from (5.43) is densely defined, satisfies the positive maximum principle, and maps the constant function $`1`$ into $`0`$. Therefore, by Hille-Yosida (5.41), we must show that the range $`(1\alpha B)`$ is dense in $`๐([0,1]^2)`$ for some, and hence for all $`\alpha >0`$. Let $`๐ซ_n`$ denote the space of polynomials on $`[0,1]^2`$ of $`n`$-th and lower order, i.e., the space of functions $`f:[0,1]^2`$ of the form
$$f(x,y)=\underset{k,l0}{}a_{kl}x^ky^l\text{ with }a_{k,l}=0\text{ for }k+l>n\text{.}$$
(5.48)
Set $`๐ซ_{\mathrm{}}:=_n๐ซ_n`$. It is easy to see that $`B`$ maps the space $`๐ซ_n`$ into itself, for each $`n0`$. Since each $`๐ซ_n`$ is finite-dimensional, a simple argument (see \[EK86, Proposition 1.3.5\]) shows that the image of $`๐ซ_{\mathrm{}}`$ under $`1\alpha B`$ is dense in $`๐([0,1]^2)`$ for all but countably many, and hence for all $`\alpha >0`$.
As a first step towards proving Lemma 5.13, we prove:
###### Lemma 5.14
(Smooth solutions to Laplace equation) Let $`\alpha >0`$, $`g๐^{(2)}([0,1])`$, $`g0`$, $`v๐([0,1]^2)`$, and assume that $`u๐^{(\mathrm{})}([0,1]^2)`$ solves the Laplace equation
$$(1\alpha (B+g))u=v.$$
(5.49)
(a) If $`g๐_{1+}`$, then $`v๐_+๐_{1+}`$ implies $`u๐_+๐_{1+}`$.
(b) If $`g๐_{1+}๐_{2+}`$, then $`v๐_+๐_{1+}๐_{2+}`$ implies $`u๐_+๐_{1+}๐_{2+}`$.
Proof Let $`u^y:=\frac{}{y}u`$, $`u^{xy}:=\frac{^2}{xy}u`$, etc. denote the partial derivatives of $`u`$ and similarly for $`v`$ and $`g`$, whenever they exist. Set $`c:=\frac{1}{\gamma }`$. Define linear operators $`B^{}`$ and $`B^{\prime \prime }`$ on $`๐([0,1]^2)`$ with domains $`๐(B^{})=๐(B^{\prime \prime }):=๐^{(\mathrm{})}([0,1]^2)`$ by
$$\begin{array}{ccc}\hfill B^{}& :=& y(1y)\frac{^2}{y^2}+\left(c(xy)+2(\frac{1}{2}y)\right)\frac{}{y},\hfill \\ \hfill B^{\prime \prime }& :=& y(1y)\frac{^2}{y^2}+\left(c(xy)+4(\frac{1}{2}y)\right)\frac{}{y}.\hfill \end{array}$$
(5.50)
Then
$$\begin{array}{cccccc}\hfill \frac{}{y}Bu& =& (B^{}c)u^y,\hfill & \hfill \frac{}{y}B^{}u& =& (B^{\prime \prime }c2)u^y,\hfill \\ \hfill \frac{}{x}Bu& =& Bu^x+cu^y,\hfill & \hfill \frac{}{x}B^{}u& =& B^{}u^x+cu^y.\hfill \end{array}$$
(5.51)
Therefore, it is easy to see that
$$\begin{array}{cccc}\hfill (\mathrm{i})& \hfill (1\alpha (B^{}c+g))u^y& =& v^y+\alpha g^yu,\hfill \\ \hfill (\mathrm{ii})& \hfill (1\alpha (B+g))u^x& =& v^x+\alpha (cu^y+g^xu),\hfill \\ \hfill (\mathrm{iii})& \hfill (1\alpha (B^{\prime \prime }2c2+g))u^{yy}& =& v^{yy}+\alpha (2g^yu^y+g^{yy}u),\hfill \\ \hfill (\mathrm{iv})& \hfill (1\alpha (B^{}c+g))u^{xy}& =& v^{xy}+\alpha (cu^{yy}+g^yu^x+g^{xy}u+g^xu^y),\hfill \\ \hfill (\mathrm{v})& \hfill (1\alpha (B+g))u^{xx}& =& v^{xx}+\alpha (2cu^{xy}+2g^xu^x+g^{xx}u),\hfill \end{array}$$
(5.52)
where in (i) and (ii) we assume that $`v๐^{(1)}([0,1]^2)`$ and in (iii)โ(v) we assume that $`v๐^{(2)}([0,1]^2)`$. By Lemma 5.12, the closure of the operator $`B`$ generates a Feller processes in $`[0,1]^2`$. Exactly the same proof shows that $`B^{}`$ and $`B^{\prime \prime }`$ also generate Feller processes on $`[0,1]^2`$. Therefore, by Feynman-Kac, $`u`$ is nonnegative if $`v`$ is nonnegative and $`u^y,\mathrm{},u^{xx}`$ are nonnegative if the right-hand sides of the equations (i)โ(v) are well-defined and nonnegative. (Instead of using Feynman-Kac, this follows more elementarily from the fact that $`B,B^{}`$, and $`B^{\prime \prime }`$ satisfy the positive maximum principle.) In particular, if $`g^y,g^x0`$ and $`v๐^{(1)}([0,1]^2)`$, $`v,v^y,v^x0`$, then it follows that $`u,u^y,u^x0`$. If moreover $`g^{yy},g^{xy},g^{xx}0`$ and $`v๐^{(2)}([0,1]^2)`$, $`v^{yy},v^{xy},v^{yy}0`$, then also $`u^{yy},u^{xy},u^{yy}0`$.
In order to prove Lemma 5.13, based on Lemma 5.14, we will show that the Laplace equation (5.49) has smooth solutions $`u`$ for sufficiently many functions $`v`$. Here โsuffiently manyโ will mean dense in the topology of uniform convergence of functions and their derivatives up to second order. To this aim, we make $`๐^{(2)}([0,1]^2)`$ into a Banach space by equipping it with the norm
$$u_{(2)}:=u+u^y+u^x+u^{yy}+2u^{xy}+u^{xx}.$$
(5.53)
Here, to reduce notation, we denote the supremumnorm by $`f:=f_{\mathrm{}}`$. Note the factor 2 in the second term from the right in (5.53), which is crucial for the next key lemma.
###### Lemma 5.15
(Semigroup on twice diffferentiable functions) The closure in $`๐^{(2)}([0,1]^2)`$ of the operator $`B`$ generates a strongly continuous contraction semigroup on $`๐^{(2)}([0,1]^2)`$.
Proof We must check the conditions (i)โ(iii) from (5.39). It is well-known (see for example \[EK86, Proposition 7.1 from the appendix\]) that the space $`๐ซ_{\mathrm{}}`$ of polynomials is dense in $`๐^{(2)}([0,1]^2)`$. Therefore $`๐(B)=๐^{(\mathrm{})}([0,1]^2)`$ is dense, and copying the proof of Lemma 5.12 we see that $`(1\alpha B)`$ is dense for all but countably many $`\alpha `$. To complete the proof, we must show that $`B`$ is dissipative, i.e., that
$$(1\epsilon B)u_{(2)}u_{(2)}(\epsilon >0,u๐^{(\mathrm{})}([0,1]^2)).$$
(5.54)
Using (5.51), we calculate
$$\begin{array}{ccc}\hfill \frac{}{y}(1\epsilon B)u& =& (1\epsilon (B^{}c))u^y,\hfill \\ \hfill \frac{}{x}(1\epsilon B)u& =& (1\epsilon B)u^x\epsilon cu^y,\hfill \\ \hfill \frac{^2}{y^2}(1\epsilon B)u& =& (1\epsilon (B^{\prime \prime }2c2))u^{yy},\hfill \\ \hfill \frac{^2}{xy}(1\epsilon B)u& =& (1\epsilon (B^{}c))u^{xy}\epsilon cu^{yy},\hfill \\ \hfill \frac{^2}{x^2}(1\epsilon B)u& =& (1\epsilon B)u^{xx}2\epsilon cu^{xy}.\hfill \end{array}$$
(5.55)
Using the disipativity of $`B,B^{}`$, and $`B^{\prime \prime }`$ with respect to the supremumnorm (which follows from the positive maximum principle) we see that $`(1\epsilon (B^{}c))u^y=(1+\epsilon c)(1\frac{\epsilon }{1+\epsilon c}B)u^y(1+\epsilon c)u^y`$ etc. We conclude therefore from (5.55) that
$$\begin{array}{ccc}\hfill (1\epsilon B)u_{(2)}& & (1\epsilon B)u+(1\epsilon (B^{}c))u^y+(1\epsilon B)u^x\epsilon cu^y\hfill \\ & & +(1\epsilon (B^{\prime \prime }2c2))u^{yy}+2(1\epsilon (B^{}c))u^{xy}2\epsilon cu^{yy}\hfill \\ & & +(1\epsilon B)u^{xx}2\epsilon cu^{xy}\hfill \\ & & u+(1+\epsilon c)u^y+u^x\epsilon cu^y\hfill \\ & & +(1+\epsilon (2c+2))u^{yy}+2(1+\epsilon c)u^{xy}2\epsilon cu^{yy}\hfill \\ & & +u^{xx}2\epsilon cu^{xy}u_{(2)}\hfill \end{array}$$
(5.56)
for each $`\epsilon >0`$, which shows that $`B`$ is dissipative with respect to the norm $`_{(2)}`$.
Proof of Lemma 5.13 Let $`g๐^{(2)}([0,1]^2)`$. Then $`ugu`$ is a bounded operator on both $`๐([0,1]^2)`$ and $`๐^{(2)}([0,1]^2)`$, so we can choose a $`\lambda >0`$ such that
$$gu\lambda u\text{and}gu_{(2)}\lambda u_{(2)}$$
(5.57)
for all $`u`$ in $`๐([0,1]^2)`$ and $`๐^{(2)}([0,1]^2)`$, respectively. Put $`\stackrel{~}{g}:=g\lambda `$. By Lemma 5.12, $`\overline{B}+\stackrel{~}{g}`$ generates a strongly continuous contraction semigroup $`(S_t^{\stackrel{~}{g}})_{t0}=(e^{\lambda t}S_t^g)_{t0}`$ on $`๐([0,1]^2)`$. Note that $`(1\alpha (B+\stackrel{~}{g}))`$ is the space of all $`v๐([0,1]^2)`$ for which the Laplace equation $`(1\alpha (B+\stackrel{~}{g}))u=v`$ has a solution $`u๐^{(\mathrm{})}([0,1]^2)`$. Therefore, by Lemma 5.14, for each $`\alpha >0`$:
$$\begin{array}{cc}\hfill (\mathrm{i})& \text{If }g๐_{1+}\text{, then }(1\alpha (\overline{B}+\stackrel{~}{g}))^1\text{ maps }(1\alpha (B+\stackrel{~}{g}))๐_+๐_{1+}\text{ into }๐_+๐_{1+}\text{.}\hfill \\ \hfill (\mathrm{ii})& \text{If }g๐_{1+}๐_{2+}\text{, then }(1\alpha (\overline{B}+\stackrel{~}{g}))^1\text{ maps }(1\alpha (B+\stackrel{~}{g}))๐_+๐_{1+}๐_{2+}\hfill \\ & \text{into }๐_+๐_{1+}๐_{2+}\text{.}\hfill \end{array}$$
(5.58)
By Lemma 5.15, the restriction of the semigroup $`(S_t^{\stackrel{~}{g}})_{t0}`$ to $`๐^{(2)}([0,1]^2)`$ is strongly continuous and contractive in the norm $`_{(2)}`$. Therefore, by Hille-Yosida (5.39), $`(1\alpha (B+\stackrel{~}{g}))`$ is dense in $`๐^{(2)}([0,1]^2)`$ for each $`\alpha >0`$. It follows that $`(1\alpha (B+\stackrel{~}{g}))๐_+๐_{1+}`$ is dense in $`๐_+๐_{1+}`$ and likewise $`(1\alpha (B+\stackrel{~}{g}))๐_+๐_{1+}๐_{2+}`$ is dense in $`๐_+๐_{1+}๐_{2+}`$, both in the norm $`_{(2)}`$. Note that we need density in the norm $`_{(2)}`$ here: if we would only know that $`(1\alpha (B+\stackrel{~}{g}))`$ is a dense subset of $`๐([0,1]^2)`$ in the norm $``$, then $`(1\alpha (B+\stackrel{~}{g}))๐_+๐_{1+}`$ might be empty. By approximation in the norm $`_{(2)}`$ it follows from (5.58) that:
$$\begin{array}{cc}\hfill (\mathrm{i})& \text{If }g๐_{1+}\text{, then }(1\alpha (\overline{B}+\stackrel{~}{g}))^1\text{ maps }๐_+๐_{1+}\text{ into itself.}\hfill \\ \hfill (\mathrm{ii})& \text{If }g๐_{1+}๐_{2+}\text{, then }(1\alpha (\overline{B}+\stackrel{~}{g}))^1\text{ maps }๐_+๐_{1+}๐_{2+}\text{ into itself.}\hfill \end{array}$$
(5.59)
Using also continuity in the norm $``$ we find that:
$$\begin{array}{cc}\hfill (\mathrm{i})& \text{If }g๐_{1+}\text{, then }(1\alpha (\overline{B}+\stackrel{~}{g}))^1\text{ maps }\overline{๐_+๐_{1+}}\text{ into itself.}\hfill \\ \hfill (\mathrm{ii})& \text{If }g๐_{1+}๐_{2+}\text{, then }(1\alpha (\overline{B}+\stackrel{~}{g}))^1\text{ maps }\overline{๐_+๐_{1+}๐_{2+}}\text{ into itself.}\hfill \end{array}$$
(5.60)
For $`\epsilon >0`$ let
$$G_\epsilon :=\epsilon ^1\left((1\epsilon (\overline{B}+\stackrel{~}{g}))^11\right)$$
(5.61)
be the Yosida approximation to $`\overline{B}+\stackrel{~}{g}`$. Then
$$e^{G_\epsilon t}=e^{\epsilon ^1t}\underset{n=0}{\overset{\mathrm{}}{}}\frac{t^n}{n!}(1\epsilon (\overline{B}+\stackrel{~}{g}))^n(t0),$$
(5.62)
and therefore, by (5.60), for each $`t0`$:
$$\begin{array}{cc}\hfill (\mathrm{i})& \text{If }g๐_{1+}\text{, then }e^{G_\epsilon t}\text{ maps }\overline{๐_+๐_{1+}}\text{ into itself.}\hfill \\ \hfill (\mathrm{ii})& \text{If }g๐_{1+}๐_{2+}\text{, then }e^{G_\epsilon t}\text{ maps }\overline{๐_+๐_{1+}๐_{2+}}\text{ into itself.}\hfill \end{array}$$
(5.63)
Finally
$$e^{\lambda t}S_t^gu=S_t^{\stackrel{~}{g}}u=\underset{\epsilon 0}{lim}\text{e}^{G_\epsilon t}u(t0,u๐([0,1]^2)),$$
(5.64)
so (5.63) implies that for each $`t0`$:
$$\begin{array}{cc}\hfill (\mathrm{i})& \text{If }g๐_{1+}\text{, then }S_t^g\text{ maps }\overline{๐_+๐_{1+}}\text{ into itself.}\hfill \\ \hfill (\mathrm{ii})& \text{If }g๐_{1+}๐_{2+}\text{, then }S_t^g\text{ maps }\overline{๐_+๐_{1+}๐_{2+}}\text{ into itself.}\hfill \end{array}$$
(5.65)
Using the continuity of $`S_t^g`$ in $`g`$ (which follows from Feynman-Kac (5.42)) we arrive at the statements in Lemma 5.13.
## 6 Convergence to a time-homogeneous process
### 6.1 Convergence of certain Markov chains
Section 6 is devoted to the proof of Theorem 3.2. In the present subsection, we start by formulating a theorem about the convergence of certain Markov chains to continuous-time processes. In Section 6.2 we specialize to Poisson-cluster branching processes and superprocesses. In Section 6.3, finally, we carry out the necessary calculations for the specific processes from Theorem 3.2.
Let $`E`$ be a compact metrizable space. Let $`A:๐(A)๐(E)`$ be an operator defined on a domain $`๐(A)๐(E)`$. We say that a process $`๐ฒ=(๐ฒ_t)_{t0}`$ solves the martingale problem for $`A`$ if $`๐ฒ`$ has sample paths in $`๐_E[0,\mathrm{})`$ and for each $`f๐(A)`$, the process $`(M_t^f)_{t0}`$ given by
$$M_t^f:=f(๐ฒ_t)_0^tAf(๐ฒ_s)ds(t0)$$
(6.1)
is a martingale with respect to the filtration generated by $`๐ฒ`$. We say that existence (uniqueness) holds for the martingale problem for $`A`$ if for each probability measure $`\mu `$ on $`E`$ there is at least one (at most one (in law)) solution $`๐ฒ`$ to the martingale problem for $`A`$ with initial law $`(๐ฒ_0)=\mu `$. If both existence and uniqueness hold we say that the martingale problem is well-posed. For each $`n0`$, let $`X^{(n)}=(X_0^{(n)},\mathrm{},X_{m(n)}^{(n)})`$ (with $`1m(n)<\mathrm{}`$) be a (time-inhomogeneous) Markov process in $`E`$ with $`k`$-th step transition probabilities
$$P_k(x,\mathrm{d}y)=P\left[X_k^{(n)}\mathrm{d}y|X_{k1}^{(n)}=x\right](1km(n)).$$
(6.2)
We assume that the $`P_k`$ are continuous probability kernels on $`E`$. Let $`(\epsilon _k^{(n)})_{1km(n)}`$ be positive constants. Set
$$A_k^{(n)}f(x):=(\epsilon _k^{(n)})^1(_EP_k(x,\mathrm{d}y)f(y)f(x))(1km(n),f๐(E)).$$
(6.3)
Define $`t_0^{(n)}:=0`$ and
$$t_k^{(n)}:=\underset{l=1}{\overset{k}{}}\epsilon _l^{(n)}(1km(n)),$$
(6.4)
and put
$$k^{(n)}(t):=\mathrm{max}\{k:\mathrm{\hspace{0.33em}0}km(n),t_k^{(n)}t\}(t0).$$
(6.5)
Define processes $`๐ฒ^{(n)}=(๐ฒ_t^{(n)})_{t0}`$ with sample paths in $`๐_E[0,\mathrm{})`$ by
$$๐ฒ_t^{(n)}:=X_{k^{(n)}(t)}^{(n)}(t0).$$
(6.6)
By definition, a space $`๐`$ of real functions is called an algebra if $`๐`$ is a linear space and $`f,g๐`$ implies $`fg๐`$.
###### Theorem 6.1
(Convergence of Markov chains) Assume that $`(X_0^{(n)})\mu `$ as $`n\mathrm{}`$ for some probability law $`\mu `$ on $`E`$. Suppose that there exists at most one (in law) solution to the martingale problem for $`A`$ with initial law $`\mu `$. Assume that the linear span of $`๐(A)`$ contains an algebra that separates points. Assume that
$$(\mathrm{i})\underset{n\mathrm{}}{lim}\underset{k=1}{\overset{m(n)}{}}\epsilon _k^{(n)}=\mathrm{},(\mathrm{ii})\underset{n\mathrm{}}{lim}\underset{k:t_k^{(n)}T}{sup}\epsilon _k^{(n)}=0,$$
(6.7)
and
$$\underset{n\mathrm{}}{lim}\underset{k:t_k^{(n)}T}{sup}A_k^{(n)}fAf_{\mathrm{}}=0(f๐(A))$$
(6.8)
for each $`T>0`$. Then there exists a unique solution $`๐ฒ`$ to the martingale problem for $`A`$ with initial law $`\mu `$ and moreover $`(๐ฒ^{(n)})(๐ฒ)`$, where $``$ denotes weak convergence of probability measures on $`๐_E[0,\mathrm{})`$.
Proof We apply \[EK86, Corollary 4.8.15\]. Fix $`f๐(A)`$. We start by observing that
$$f(X_k^{(n)})\underset{i=1}{\overset{k}{}}\epsilon _i^{(n)}A_i^{(n)}f(X_{i1}^{(n)})(0km(n))$$
(6.9)
is a martingale with respect to the filtration generated by $`X^{(n)}`$ and therefore,
$$f(๐ฒ_t^{(n)})\underset{i=1}{\overset{k^{(n)}(t)}{}}\epsilon _i^{(n)}A_i^{(n)}f(๐ฒ_{t_{i1}^{(n)}}^{(n)})(t0)$$
(6.10)
is a martingale with respect to the filtration generated by $`๐ฒ^{(n)}`$. Put
$$t^{(n)}:=t_{k^{(n)}(t)}^{(n)}(t0)$$
(6.11)
and set
$$\varphi _t^{(n)}:=A_{k^{(n)}(t)+1}^{(n)}f(๐ฒ_{t^{(n)}}^{(n)})1_{\{t<t_{m(n)}^{(n)}\}}(t0)$$
(6.12)
and
$$\xi _t^{(n)}:=f(๐ฒ_t^{(n)})+_{t^{(n)}}^t\varphi _s^{(n)}ds(t0).$$
(6.13)
Then we can rewrite the martingale in (6.10) as
$$\xi _t^{(n)}_0^t\varphi _s^{(n)}ds.$$
(6.14)
By \[EK86, Corollary 4.8.15\] and the compactness of the state space, it suffices to check the following conditions on $`\varphi ^{(n)}`$ and $`\xi ^{(n)}`$:
$$\begin{array}{cc}\hfill (\mathrm{i})& \underset{nN}{sup}\underset{tT}{sup}E\left[|\xi _t^{(n)}|\right]<\mathrm{},\hfill \\ \hfill (\mathrm{ii})& \underset{nN}{sup}\underset{tT}{sup}E\left[|\varphi _t^{(n)}|\right]<\mathrm{},\hfill \\ \hfill (\mathrm{iii})& \underset{n\mathrm{}}{lim}E\left[\left(\xi _T^{(n)}f(๐ฒ_T^{(n)})\right)\underset{i=1}{\overset{r}{}}h_i(๐ฒ_{s_i}^{(n)})\right]=0,\hfill \\ \hfill (\mathrm{iv})& \underset{n\mathrm{}}{lim}E\left[\left(\varphi _T^{(n)}Af(๐ฒ_T^{(n)})\right)\underset{i=1}{\overset{r}{}}h_i(๐ฒ_{s_i}^{(n)})\right]=0,\hfill \\ \hfill (\mathrm{v})& \underset{n\mathrm{}}{lim}E\left[\underset{t[0,T]}{sup}\left|\xi _t^{(n)}f(๐ฒ_t^{(n)})\right|\right]=0,\hfill \\ \hfill (\mathrm{vi})& \underset{nN}{sup}E\left[\varphi ^{(n)}_{p,T}\right]<\mathrm{}\text{for some }p(1,\mathrm{}],\hfill \end{array}$$
(6.15)
for some $`N0`$ and for each $`T>0`$, $`r1`$, $`0s_1<\mathrm{}<s_rT`$, and $`h_1,\mathrm{},h_r๐(E)`$. Here $``$ is separating, i.e., $`hd\mu =hd\nu `$ for all $`h`$ implies $`\mu =\nu `$ whenever $`\mu ,\nu `$ are probability measures on $`E`$. In (vi):
$$g_{p,T}:=\left(_0^T|g(t)|^pdt\right)^{1/p}(1p<\mathrm{})$$
(6.16)
and $`g_{\mathrm{},T}`$ denotes the essential supremum of $`g`$ over $`[0,T]`$.
The conditions (6.15) (i)โ(vi) are implied by the stronger conditions
$$\begin{array}{cc}\hfill (\mathrm{i})& \underset{n\mathrm{}}{lim}\underset{0tT}{sup}\xi _t^{(n)}f(๐ฒ_t^{(n)})_{\mathrm{}}=0,\hfill \\ \hfill (\mathrm{ii})& \underset{n\mathrm{}}{lim}\underset{0tT}{sup}\varphi _t^{(n)}Af(๐ฒ_t^{(n)})_{\mathrm{}}=0,\hfill \end{array}$$
(6.17)
where we denote the essential supremumnorm of a real-valued random variable $`X`$ by $`X_{\mathrm{}}:=inf\{K0:|X|K\text{a.s.}\}`$. Condition (6.17) (ii) is implied by (6.7) (i) and (6.8). To see that also (6.17) (i) holds, set
$$M_n:=\underset{0tT}{sup}\varphi _t^{(n)}_{\mathrm{}},$$
(6.18)
and estimate
$$\underset{0tT}{sup}\xi _t^{(n)}f(๐ฒ_t^{(n)})_{\mathrm{}}M_nsup\{\epsilon _k^{(n)}:\mathrm{\hspace{0.17em}1}km(n),t_k^{(n)}T\}.$$
(6.19)
Condition (6.17) (ii) implies that $`lim\; sup_nM_n<\mathrm{}`$ and therefore the right-hand side of (6.19) tends to zero by assumption (6.7) (ii).
### 6.2 Convergence of certain branching processes
In this section we apply Theorem 6.1 to certain branching processes and superprocesses.
Throughout this section, $`E`$ is a compact metrizable space and $`A:๐(A)๐(E)`$ is a linear operator on $`๐(E)`$ such that the closure $`\overline{A}`$ of $`A`$ generates a Feller process $`\xi =(\xi _t)_{t0}`$ in $`E`$ with Feller semigroup $`(P_t)_{t0}`$ given by $`P_tf(x):=E^x[f(\xi _t)]`$ ($`t0,f๐(E)`$).
Let $`\alpha ๐_+(E)`$ and $`\beta ,f๐(E)`$. By definition, a function $`tu_t`$ from $`[0,\mathrm{})`$ into $`๐(E)`$ is a classical solution to the semilinear Cauchy problem
$$\{\begin{array}{ccc}\hfill \frac{}{t}u_t& =& \overline{A}u_t+\beta u_t\alpha u_t^2(t0),\hfill \\ \hfill u_0& =& f\hfill \end{array}$$
(6.20)
if $`tu_t`$ is continuously differentiable (in $`๐(E)`$), $`u_t๐(\overline{A})`$ for all $`t0`$, and (6.20) holds. We say that $`u`$ is a mild solution to (6.20) if $`tu_t`$ is continuous and
$$u_t=P_tf+_0^tP_{ts}(\beta u_s\alpha u_s^2)ds(t0).$$
(6.21)
###### Lemma 6.2
(Mild and classical solutions) Equation (6.20) has a unique $`๐_+(E)`$-valued mild solution $`u`$ for each $`f๐_+(E)`$, and $`f>0`$ implies that $`u_t>0`$ for all $`t0`$. If moreover $`f๐(\overline{A})`$ then $`u`$ is a classical solution. For each $`t0`$, $`u_t`$ depends continuously on $`f๐_+(E)`$.
Proof It follows from \[Paz83, Theorems 6.1.2, 6.1.4, and 6.1.5\] that for each $`f๐(E)`$, (6.20) has a unique solution $`(u_t)_{0t<T}`$ up to an explosion time $`T`$, and that this is a classical solution if $`f๐(\overline{A})`$. Moreover, $`u_t`$ depends continuously on $`f`$. Using comparison arguments based on the fact that $`\overline{A}`$ satisfies the positive maximum principle (which follows from Hille-Yosida (5.41)) one easily proves the other statements; compare \[FS04, Lemmas 23 and 24\].
We denote the (mild or classical) solution of (6.20) by $`๐ฐ_tf:=u_t`$; then $`๐ฐ_t:๐_+(E)๐_+(E)`$ are continuous operators and $`๐ฐ=(๐ฐ_t)_{t0}`$ is a (nonlinear) semigroup on $`๐_+(E)`$.
Since $`E`$ is compact, the spaces $`\{\mu (E):\mu (E)M\}`$ are compact for each $`M0`$. In particular, $`(E)`$ is locally compact. We denote its one-point compactification by $`(E)_{\mathrm{}}=(E)\{\mathrm{}\}`$. We define functions $`F_f๐((E)_{\mathrm{}})`$ by $`F_f(\mathrm{}):=0`$ and
$$F_f(\mu ):=\text{e}^{\mu ,f}(f๐_+(E),f>0,\mu (E)).$$
(6.22)
We introduce an operator $`๐ข`$ with domain
$$๐(๐ข):=\{F_f:f๐(A),f>0\},$$
(6.23)
given by $`๐ขF_f(\mathrm{}):=0`$ and
$$๐ขF_f(\mu ):=\mu ,Af+\beta f\alpha f^2\text{e}^{\mu ,f}(\mu (E)).$$
(6.24)
Note that $`๐ขF_f๐((E)_{\mathrm{}})`$ for all $`F_f๐(๐ข)`$.
###### Proposition 6.3
($`(\overline{A},\alpha ,\beta )`$-superprocesses) The martingale problem for the operator $`๐ข`$ is well-posed. The solutions to this martingale problem define a Feller process $`๐ด=(๐ด_t)_{t0}`$ in $`(E)_{\mathrm{}}`$ with continuous sample paths, called the $`(\overline{A},\alpha ,\beta )`$-superprocess. If $`๐ด_0=\mathrm{}`$ then $`๐ด_t=\mathrm{}`$ for all $`t0`$. If $`๐ด_0=\mu (E)`$ then
$$E^\mu \left[\text{e}^{๐ด_t,f}\right]=\text{e}^{\mu ,๐ฐ_tf}(f๐_+(E)).$$
(6.25)
Proof Results of this type are well-known, see for example \[EK86, Theorem 9.4.3\], \[Fit88\], and \[ER91, Thรฉorรจme 7\]. Since, however, it is not completely straightforward to derive the proposition above from these references, we give a concise autonomous proof of most of our statements. Only for the continuity of sample paths we refer the reader to \[Fit88, Corollary (4.7)\] or \[ER91, Corollaire 9\].
We are going to extend $`๐ข`$ to an operator $`\widehat{๐ข}`$ that is linear and satisfies the conditions of the Hille-Yosida Theorem (5.41). For any $`\gamma ๐_+(E)`$ and $`\mu (E)`$, let $`\mathrm{Clust}_\gamma (\mu )`$ denote a random measure such that on $`\{\gamma =0\}`$, $`\mathrm{Clust}_\gamma (\mu )`$ is equal to $`\mu `$, and on $`\{\gamma >0\}`$, $`\mathrm{Clust}_\gamma (\mu )`$ is a Poisson cluster measure with intensity $`\frac{1}{\gamma }\mu `$ and cluster mechanism $`๐ฌ(x,)=(\tau _{\gamma (x)}\delta _x)`$, where $`\tau _{\gamma (x)}`$ is exponentially distributed with mean $`\gamma (x)`$. It is not hard to see that
$$E\left[\text{e}^{\mathrm{Clust}_\gamma \left(\mu \right),f}\right]=\text{e}^{\mu ,๐ฑ_\gamma f}(f๐(E),f>0),$$
(6.26)
where $`๐ฑ_\gamma f(x):=(\frac{1}{f(x)}+\gamma (x))^1`$. Note that since $`๐ฑ_\gamma 1`$ is bounded, the previously mentioned Poisson cluster measure mentioned above is well-defined. By definition, we put $`\mathrm{Clust}_\gamma (\mathrm{}):=\mathrm{}`$.
Define a linear operator $`๐ข_\alpha `$ on $`๐((E))_{\mathrm{}})`$ by
$$๐ข_\alpha F(\mu ):=\underset{\epsilon 0}{lim}\epsilon ^1\left(E[F(\mathrm{Clust}_{\epsilon \alpha }(\mu ))]F(\mu )\right)$$
(6.27)
with as domain $`๐(๐ข_\alpha )`$ the space of all $`F๐((E)_{\mathrm{}})`$ for which the limit exists. Define a linear operator $`๐ข_\beta `$ by
$$๐ข_\beta F(\mu ):=\underset{\epsilon 0}{lim}\epsilon ^1\left(F((1+\epsilon \beta )\mu )F(\mu )\right)$$
(6.28)
with domain $`๐(๐ข_\beta ):=๐((E))_{\mathrm{}})`$. Define $`P_t^{}:(E)_{\mathrm{}}(E)_{\mathrm{}}`$ by $`P_t^{}\mu ,f:=\mu ,P_tf`$ $`(t0,f๐(E),\mu (E))`$ and $`P_t^{}\mathrm{}:=\mathrm{}`$ ($`t0`$). Finally, let $`๐ข_{\overline{A}}`$ be the linear operator on $`๐((E))_{\mathrm{}})`$ defined by
$$๐ข_{\overline{A}}F(\mu ):=\underset{\epsilon 0}{lim}\epsilon ^1\left(F(P_\epsilon ^{}\mu )F(\mu )\right),$$
(6.29)
with as domain $`๐(๐ข_{\overline{A}})`$ the space of all $`F`$ for which the limit exists. Define an operator $`\widehat{๐ข}`$ by
$$\widehat{๐ข}:=๐ข_\alpha +๐ข_\beta +๐ข_{\overline{A}},$$
(6.30)
with domain $`๐(\widehat{๐ข}):=๐(๐ข_\alpha )๐(๐ข_{\overline{A}})`$. If $`f๐(\overline{A})`$, $`f>0`$, and $`F_f`$ is as in (6.22), then it is not hard to see that $`\widehat{๐ข}F_f(\mathrm{})=0`$ and
$$\widehat{๐ข}F_f(\mu ):=\mu ,\overline{A}f+\beta f\alpha f^2\text{e}^{\mu ,f}(\mu (E)).$$
(6.31)
In particular, $`\widehat{๐ข}`$ extends the operator $`๐ข`$ from (6.24). Since $`๐(\overline{A})`$ is dense in $`๐(E)`$, it is easy to see that $`\{F_f:f๐(\overline{A}),f>0\}`$ is dense in $`๐((E)_{\mathrm{}})`$. Hence $`๐(\widehat{๐ข})`$ is dense. Using (6.27)โ(6.29) it is not hard to show that $`\widehat{๐ข}`$ satisfies the positive maximum principle. Moreover, by Lemma 6.2, for $`f๐(\overline{A})`$ with $`f>0`$, the function $`tF_{๐ฐ_tf}`$ from $`[0,\mathrm{})`$ into $`๐((E)_{\mathrm{}})`$ is continuously differentiable, satisfies $`F_{๐ฐ_tf}๐(\widehat{๐ข})`$ for all $`t0`$, and
$$\frac{}{t}F_{๐ฐ_tf}=\widehat{๐ข}F_{๐ฐ_tf}(t0).$$
(6.32)
From this it is not hard to see that $`\widehat{๐ข}`$ also satisfies condition (5.41) (ii), so the closure of $`\widehat{๐ข}`$ generates a Feller semigroup $`(S_t)_{t0}`$ on $`๐((E)_{\mathrm{}})`$. It is easy to see that $`S_tF_f=F_{๐ฐ_tf}`$ $`(t0)`$. By \[EK86, Theorem 4.2.7\], this semigroup corresponds to a Feller process $`๐ด`$ with cadlag sample paths in $`(E)_{\mathrm{}}`$. This means that $`E^\mu [F_f(๐ด_t)]=F_{๐ฐ_tf}(\mu )`$ for all $`f๐(\overline{A})`$ with $`f>0`$. If $`\mu =\mathrm{}`$ this shows that $`๐ด_t=\mathrm{}`$ for all $`t0`$. If $`\mu (E)`$ we obtain (6.25) for $`f๐(\overline{A})`$, $`f>0`$; the general case follows by approximation.
Now let $`(q_\epsilon )_{\epsilon >0}`$ be continuous weight functions and let $`(๐ฌ_\epsilon )_{\epsilon >0}`$ be continuous cluster mechanisms on $`E`$. Assume that
$$Z_\epsilon (x):=๐ฌ_\epsilon (x,\mathrm{d}\chi )\chi ,1<\mathrm{}(xE)$$
(6.33)
and define probability kernels $`K_\epsilon `$ on $`E`$ by
$$K_\epsilon (x,\mathrm{d}y)f(y):=\frac{1}{Z_\epsilon (x)}๐ฌ_\epsilon (x,\mathrm{d}\chi )\chi ,f(fB(E)).$$
(6.34)
For each $`n0`$, let $`(\epsilon _k^{(n)})_{1km(n)}`$ (with $`1m(n)<\mathrm{}`$) be positive constants. Let $`๐ณ^{(n)}=(๐ณ_0^{(n)},\mathrm{},๐ณ_{m(n)}^{(n)})`$ be a Poisson-cluster branching process with weight functions $`q_{\epsilon _1^{(n)}},\mathrm{},q_{\epsilon _{m(n)}^{(n)}}`$ and cluster mechanisms $`๐ฌ_{\epsilon _1^{(n)}},\mathrm{},๐ฌ_{\epsilon _{m(n)}^{(n)}}`$. Define $`t_k^{(n)}`$ and $`k^{(n)}(t)`$ as in (6.4)โ(6.5). Define processes $`๐ด^{(n)}`$ by
$$๐ด_t^{(n)}:=๐ณ_{k^{(n)}(t)}^{(n)}(t0).$$
(6.35)
###### Theorem 6.4
(Convergence of Poisson-cluster branching processes) Assume that $`(๐ณ_0^{(n)})\rho `$ as $`n\mathrm{}`$ for some probability law $`\rho `$ on $`(E)`$. Suppose that the constants $`\epsilon _k^{(n)}`$ fulfill (6.7). Assume that
$$\begin{array}{cccc}\hfill (\mathrm{i})& \hfill q_\epsilon (x)๐ฌ_\epsilon (x,\mathrm{d}\chi )\chi ,1& =& 1+\epsilon \beta (x)+o(\epsilon ),\hfill \\ \hfill (\mathrm{ii})& \hfill q_\epsilon (x)๐ฌ_\epsilon (x,\mathrm{d}\chi )\chi ,1^2& =& \epsilon \mathrm{\hspace{0.17em}2}\alpha (x)+o(\epsilon ),\hfill \\ \hfill (\mathrm{iii})& \hfill q_\epsilon (x)๐ฌ_\epsilon (x,\mathrm{d}\chi )\chi ,1^21_{\{\chi ,1>\delta \}}& =& o(\epsilon )\hfill \end{array}$$
(6.36)
for each $`\delta >0`$, and
$$K_\epsilon (x,\mathrm{d}y)f(y)=f(x)+\epsilon Af(x)+o(\epsilon )$$
(6.37)
for each $`f๐(A)`$, uniformly in $`x`$ as $`\epsilon 0`$. Then $`(๐ด^{(n)})(๐ด)`$, where $`๐ด`$ is the $`(\overline{A},\alpha ,\beta )`$-superprocess with initial law $`\rho `$.
Here $``$ denotes weak convergence of probability measures on $`๐_{(E)}[0,\mathrm{})`$.
Proof We apply Theorem 6.1 to the operator $`๐ข`$, where we use the fact that if we view $`_1(๐_{(E)}[0,\mathrm{}))`$ as a subspace of $`_1(๐_{(E)_{\mathrm{}}}[0,\mathrm{}))`$ (note the compactification), equipped with the topology of weak convergence, then the induced topology on $`_1(๐_{(E)}[0,\mathrm{}))`$ is again the topology of weak convergence.
By Proposition 6.3, solutions to the martingale problem for $`๐ข`$ are unique. Since $`F_fF_g=F_{f+g}`$ and $`๐(A)`$ is a linear space, the linear span of the domain of $`๐ข`$ is an algebra. Using the fact that $`๐(A)`$ is dense in $`๐(E)`$ we see that this algebra separates points. Therefore, we are left with the task to check (6.8).
Define $`๐ฐ_\epsilon :๐_+(E)๐_+(E)`$ by
$$๐ฐ_\epsilon f(x):=q_\epsilon (x)๐ฌ_\epsilon (x,\mathrm{d}\chi )\left(1\text{e}^{\chi ,f}\right)(xE,f๐_+[0,1],f>0,\epsilon >0),$$
(6.38)
and define transition probabilities $`P_\epsilon (\mu ,\mathrm{d}\nu )`$ on $`(E)_{\mathrm{}}`$ by $`P_\epsilon (\mathrm{},):=\delta _{\mathrm{}}`$ and
$$P_\epsilon (\mu ,\mathrm{d}\nu )\text{e}^{\nu ,f}=\text{e}^{\mu ,๐ฐ_\epsilon f}.$$
(6.39)
We will show that
$$\underset{\epsilon 0}{lim}\epsilon ^1(๐ฐ_\epsilon ff)(Af+\beta f\alpha f^2)_{\mathrm{}}=0(f๐(A),f>0).$$
(6.40)
Together with (6.39) this implies that
$$P_\epsilon (\mu ,\mathrm{d}\nu )F_f(\nu )=F_f(\mu )+\epsilon ๐ขF_f(\mu )+o(\epsilon )(f๐(A),f>0),$$
(6.41)
uniformly in $`\mu (E)_{\mathrm{}}`$ as $`\epsilon 0`$. Therefore, the result follows from Theorem 6.1.
It remains to prove (6.40). Set $`g(z):=1z+\frac{1}{2}z^2e^z`$ $`(z0)`$ and write
$$๐ฐ_\epsilon f(x)=q_\epsilon (x)๐ฌ_\epsilon (x,\mathrm{d}\chi )\left(\chi ,f\frac{1}{2}\chi ,f^2+g(\chi ,f)\right).$$
(6.42)
Since
$$g(z)=_0^zdy_0^ydx_0^xdte^t(z0),$$
(6.43)
it is easy to see that $`g`$ is nondecreasing on $`[0,\mathrm{})`$ and (since $`0e^t1`$ and $`_0^xdte^t1`$)
$$0g(z)\frac{1}{2}z^2\frac{1}{6}z^3(z0).$$
(6.44)
Using these facts and (6.36) (ii) and (iii), we find that
$$\begin{array}{c}q_\epsilon (x)๐ฌ_\epsilon (x,\mathrm{d}\chi )g(\chi ,f)\hfill \\ f_{\mathrm{}}q_\epsilon (x)\left\{๐ฌ_\epsilon (x,\mathrm{d}\chi )g(\chi ,1)1_{\{\chi ,1\delta \}}+๐ฌ_\epsilon (x,\mathrm{d}\chi )g(\chi ,1)1_{\{\chi ,1>\delta \}}\right\}\hfill \\ f_{\mathrm{}}q_\epsilon (x)\left\{\frac{1}{6}\delta ๐ฌ_\epsilon (x,\mathrm{d}\chi )\chi ,1^21_{\{\chi ,1\delta \}}+\frac{1}{2}๐ฌ_\epsilon (x,\mathrm{d}\chi )\chi ,1^21_{\{\chi ,1>\delta \}}\right\}\hfill \\ =\frac{1}{6}\delta f_{\mathrm{}}\left(\epsilon \mathrm{\hspace{0.17em}2}\alpha (x)+o(\epsilon )\right)+o(\epsilon ).\hfill \end{array}$$
(6.45)
Since this holds for any $`\delta >0`$, we conclude that
$$q_\epsilon (x)๐ฌ_\epsilon (x,\mathrm{d}\chi )g(\chi ,f)=o(\epsilon )$$
(6.46)
uniformly in $`x`$ as $`\epsilon 0`$. By (6.36) (i) and (6.37),
$$\begin{array}{c}q_\epsilon (x)๐ฌ_\epsilon (x,\mathrm{d}\chi )\chi ,f=\left(q_\epsilon (x)๐ฌ_\epsilon (x,\mathrm{d}\chi )\chi ,1\right)\left(K_\epsilon (x,\mathrm{d}y)f(y)\right)\hfill \\ =\left(1+\epsilon \beta (x)+o(\epsilon )\right)\left(f(x)+\epsilon Af(x)+o(\epsilon )\right)\hfill \\ =f(x)+\epsilon \beta (x)f(x)+\epsilon Af(x)+o(\epsilon ).\hfill \end{array}$$
(6.47)
Finally, write
$$\begin{array}{c}q_\epsilon (x)๐ฌ_\epsilon (x,\mathrm{d}\chi )\chi ,f^2\hfill \\ =q_\epsilon (x)๐ฌ_\epsilon (x,\mathrm{d}\chi )\left(\chi ,f(x)^2+2\chi ,f(x)\chi ,ff(x)+\chi ,ff(x)^2\right).\hfill \end{array}$$
(6.48)
Then, by (6.36) (ii),
$$q_\epsilon (x)๐ฌ_\epsilon (x,\mathrm{d}\chi )\chi ,f(x)^2=f(x)^2\left(\epsilon \mathrm{\hspace{0.17em}2}\alpha (x)+o(\epsilon )\right).$$
(6.49)
We will prove that
$$q_\epsilon (x)๐ฌ_\epsilon (x,\mathrm{d}\chi )\chi ,ff(x)^2=o(\epsilon ).$$
(6.50)
Then, by Hรถlderโs inequality, (6.36) (ii), and (6.50),
$$\begin{array}{c}\left|q_\epsilon (x)๐ฌ_\epsilon (x,\mathrm{d}\chi )\chi ,ff(x)\chi ,f(x)\right|\hfill \\ \left(q_\epsilon (x)๐ฌ_\epsilon (x,\mathrm{d}\chi )\chi ,ff(x)^2\right)^{1/2}\left(q_\epsilon (x)๐ฌ_\epsilon (x,\mathrm{d}\chi )\chi ,f(x)^2\right)^{1/2}\hfill \\ \left(o(\epsilon )(2\alpha (x)\epsilon +o(\epsilon ))\right)^{1/2}=o(\epsilon ).\hfill \end{array}$$
(6.51)
Inserting (6.49), (6.50) and (6.51) into (6.48) we find that
$$q_\epsilon (x)๐ฌ_\epsilon (x,\mathrm{d}\chi )\chi ,f^2=\epsilon \mathrm{\hspace{0.17em}2}\alpha (x)f(x)^2+o(\epsilon ).$$
(6.52)
Inserting (6.46), (6.47) and (6.52) into (6.42), we arrive at (6.40). We still need to prove (6.50). To this aim, we estimate, using (6.47),
$$\begin{array}{c}q_\epsilon (x)๐ฌ_\epsilon (x,\mathrm{d}\chi )\chi ,ff(x)^21_{\{\chi ,1\delta \}}\hfill \\ \delta ff(x)_{\mathrm{}}q_\epsilon (x)๐ฌ_\epsilon (x,\mathrm{d}\chi )\chi ,ff(x)\hfill \\ =\delta ff(x)_{\mathrm{}}\left(\epsilon Af(x)+o(\epsilon )\right)\hfill \end{array}$$
(6.53)
and, using (6.36) (iii),
$$\begin{array}{c}q_\epsilon (x)๐ฌ_\epsilon (x,\mathrm{d}\chi )\chi ,ff(x)^21_{\{\chi ,1>\delta \}}\hfill \\ ff(x)_{\mathrm{}}q_\epsilon (x)๐ฌ_\epsilon (x,\mathrm{d}\chi )\chi ,1^21_{\{\chi ,1>\delta \}}=o(\epsilon ).\hfill \end{array}$$
(6.54)
It follows that
$$q_\epsilon (x)๐ฌ_\epsilon (x,\mathrm{d}\chi )\chi ,ff(x)^2\delta \epsilon ff(x)_{\mathrm{}}Af(x)+o(\epsilon )$$
(6.55)
for any $`\delta >0`$. This implies (6.50) and completes the proof of (6.40).
### 6.3 Application to the renormalization branching process
Proof of Theorem 3.2 (a) For any $`f_0,\mathrm{},f_k๐_+[0,1]`$ one has
$$\begin{array}{c}E\left[\text{e}^{๐ณ_n,f_0}\mathrm{}\text{e}^{๐ณ_{n+k},f_k}\right]\hfill \\ =E\left[\text{e}^{๐ณ_n,f_0}\mathrm{}\text{e}^{๐ณ_{n+k1},f_{k1}+๐ฐ_{\gamma _{nk}}f_k}\right]\hfill \\ =\mathrm{}=E\left[\text{e}^{๐ณ_n,g_k}\right],\hfill \end{array}$$
(6.56)
where we define inductively
$$g_0:=f_k\text{and}g_{m+1}:=f_{km1}+๐ฐ_{\gamma _{nk+m}}g_m.$$
(6.57)
By the compactness of $`[0,1]`$ and Corollary 5.10, the map $`(\gamma ,f)๐ฐ_\gamma f`$ from $`(0,\mathrm{})\times ๐_+[0,1]`$ to $`๐_+[0,1]`$ (equipped with the supremumnorm) is continuous. Using this fact and (6.56) we find that
$$E\left[\text{e}^{๐ณ_n,f_0}\mathrm{}\text{e}^{๐ณ_{n+k},f_k}\right]\underset{n\mathrm{}}{}E\left[\text{e}^{๐ด_n^\gamma ^{},f_0}\mathrm{}\text{e}^{๐ด_{n+k}^\gamma ^{},f_k}\right].$$
(6.58)
Since $`f_1,\mathrm{},f_k`$ are arbitrary, (3.12) follows.
Proof of Theorem 3.2 (b) We apply Theorem 6.4 to the weight functions $`q_\gamma `$ and cluster mechanisms $`๐ฌ_\gamma `$ from (3.8) and to $`A_{\mathrm{WF}}=x(1x)\frac{^2}{x^2}`$ with domain $`๐(A_{\mathrm{WF}})=๐^{(2)}[0,1]`$, and $`\alpha =\beta =1`$. It is well-known that $`\overline{A}_{\mathrm{WF}}`$ generates a Feller semigroup \[EK86, Theorem 8.2.8\]. We observe that
$$๐ฌ_\gamma (x,\mathrm{d}\chi )\chi ,f=E\left[2_0^{\tau _\gamma }f(๐ฒ_x^\gamma (t))\right]=2E[\tau _\gamma ]E\left[f(๐ฒ_x^\gamma (0))\right]=\gamma \mathrm{\Gamma }_x^\gamma (\mathrm{d}y)f(y),$$
(6.59)
where $`\mathrm{\Gamma }_x^\gamma `$ is the equilibrium law of the process $`๐ฒ_x^\gamma `$ from Corollary 5.4. It follows from (5.24) that
$$\begin{array}{cccc}\hfill (\mathrm{i})& \hfill \mathrm{\Gamma }_x^\gamma (\mathrm{d}y)(yx)& =& 0,\hfill \\ \hfill (\mathrm{ii})& \hfill \mathrm{\Gamma }_x^\gamma (\mathrm{d}y)(yx)^2& =& \frac{\gamma x(1x)}{1+\gamma },\hfill \\ \hfill (\mathrm{iii})& \hfill \mathrm{\Gamma }_x^\gamma (\mathrm{d}y)(yx)^4& =& O(\gamma ^2),\hfill \end{array}$$
(6.60)
uniformly in $`x`$ as $`\gamma 0`$. Therefore, for any $`\delta >0`$,
$$\begin{array}{cccc}\hfill (\mathrm{i})& \hfill \mathrm{\Gamma }_x^\gamma (\mathrm{d}y)(yx)& =& 0,\hfill \\ \hfill (\mathrm{ii})& \hfill \mathrm{\Gamma }_x^\gamma (\mathrm{d}y)(yx)^2& =& \gamma x(1x)+o(\gamma ),\hfill \\ \hfill (\mathrm{iii})& \hfill \mathrm{\Gamma }_x^\gamma (\mathrm{d}y)1_{\{|yx|>\delta \}}& =& o(\gamma ),\hfill \end{array}$$
(6.61)
uniformly in $`x`$ as $`\gamma 0`$. Consequently, a Taylor expansion of $`f`$ around $`x`$ yields
$$\mathrm{\Gamma }_x^\gamma (\mathrm{d}y)f(x)=f(x)+\gamma \frac{1}{2}x(1x)\frac{^2}{x^2}f(x)+o(\gamma )(f๐^{(2)}[0,1]),$$
(6.62)
uniformly in $`x`$ as $`\gamma 0`$. (For details, in particular the uniformity in $`x`$, see for example Proposition \[Swa99, B.1.1\].) This shows that condition (6.37) is satisfied. Moreover,
$$\begin{array}{c}๐ฌ_\gamma (x,\mathrm{d}\chi )\chi ,1=E[2\tau _\gamma ]=\gamma ,\hfill \\ ๐ฌ_\gamma (x,\mathrm{d}\chi )\chi ,1^2=E[(2\tau _\gamma )^2]=_0^{\mathrm{}}z^2\frac{1}{\gamma }e^{z/\gamma }dz=2\gamma ^2,\hfill \\ ๐ฌ_\gamma (x,\mathrm{d}\chi )\chi ,1^3=E[(2\tau _\gamma )^3]=_0^{\mathrm{}}z^3\frac{1}{\gamma }e^{z/\gamma }dz=6\gamma ^3,\hfill \end{array}$$
(6.63)
which, using the fact that $`q_\gamma =(\frac{1}{\gamma }+1)`$, gives
$$\begin{array}{c}q_\gamma ๐ฌ_\gamma (x,\mathrm{d}\chi )\chi ,1=1+\gamma ,\hfill \\ q_\gamma ๐ฌ_\gamma (x,\mathrm{d}\chi )\chi ,1^2=2\gamma +o(\gamma ),\hfill \\ q_\gamma ๐ฌ_\gamma (x,\mathrm{d}\chi )\chi ,1^3=o(\gamma ).\hfill \end{array}$$
(6.64)
This shows that (6.36) is fulfilled. In particular,
$$q_\gamma ๐ฌ_\gamma (x,\mathrm{d}\chi )\chi ,1^21_{\{\chi ,1>\delta \}}\delta ^1q_\gamma ๐ฌ_\gamma (x,\mathrm{d}\chi )\chi ,1^3=o(\gamma )$$
(6.65)
for all $`\delta >0`$.
## 7 Embedded particle systems
In this section we use embedded particle systems to prove Proposition 3.5. An essential ingredient in the proofs is Proposition 7.15 (a), which will be proved in the Section 8.
### 7.1 Weighting and Poissonization
Proof of Proposition 3.3 Obviously $`q_k^h๐_+(E^h)`$ for each $`k=1,\mathrm{},n`$. Since $`h๐_+(E)`$ and $`h`$ is bounded, it is easy to see that the map $`\mu h\mu `$ from $`(E)`$ into $`(E^h)`$ is continuous, and therefore the cluster mechanisms defined in (3.21) are continuous. Since
$$๐ฐ_k^hf(x)=\frac{q_k(x)}{h(x)}E\left[1\text{e}^{h๐ต_x,f}\right]=\frac{๐ฐ_k(hf)(x)}{h(x)}(xE^h,fB_+(E^h)),$$
(7.1)
formula (3.22) holds on $`E^h`$. To see that (3.22) holds on $`E\backslash E^h`$, note that by assumption $`๐ฐ_khKh`$ for some $`K<\mathrm{}`$, so if $`xE\backslash E^h`$, then $`๐ฐ_kh(x)=0`$. By monotonicity also $`๐ฐ_k(hf)(x)=0`$, while $`h๐ฐ_k^hf(x)=0`$ by definition. Since $`sup_{xE^h}๐ฐ_k^h1(x)=sup_{xE^h}\frac{๐ฐ_kh(x)}{h(x)}K<\mathrm{}`$, the log-Laplace operators $`๐ฐ_k^h`$ satisfy (3.3). If $`๐ณ`$ is started in an initial state $`๐ณ_0`$, then the Poisson-cluster branching process $`๐ณ^h`$ with log-Laplace operators $`๐ฐ_1^h,\mathrm{},๐ฐ_n^h`$ started in $`๐ณ_0^h=h๐ณ_0`$ satisfies
$$\begin{array}{ccc}\hfill E\left[\text{e}^{h๐ณ_k,f}\right]& =& E\left[\text{e}^{๐ณ_0,๐ฐ_1\mathrm{}๐ฐ_k\left(hf\right)}\right]\hfill \\ & =& E\left[\text{e}^{๐ณ_0,h๐ฐ_1^h\mathrm{}๐ฐ_k^h\left(f\right)}\right]=E\left[\text{e}^{๐ณ_k^h,f}\right](fB_+(E^h)),\hfill \end{array}$$
(7.2)
which proves (3.23).
Proof of Proposition 3.4 We start by noting that by (3.2),
$$๐ฐ_kf(x)=q(x)E\left[1\text{e}^{๐ต_x^k,f}\right]=q_k(x)P[\mathrm{Pois}(f๐ต_x^k)0](xE,fB_+(E)).$$
(7.3)
Into (3.24), we insert
$$\begin{array}{c}P\left[\mathrm{Pois}(h๐ต_x^k)\right]\hfill \\ =P[\mathrm{Pois}(h๐ต_x^k)|\mathrm{Pois}(h๐ต_x^k)0]P[\mathrm{Pois}(h๐ต_x^k)0]+\delta _0P[\mathrm{Pois}(h๐ต_x^k)=0].\hfill \end{array}$$
(7.4)
Here and in similar formulas below, if in a conditional probability the symbol $`\mathrm{Pois}()`$ occurs twice with the same argument, then it always refers to the same random variable (and not to independent Poisson point measures with the same intensity, for example). Using moreover (7.3) we can rewrite (3.24) as
$$Q_k^h(x,)=\frac{๐ฐ_kh(x)}{h(x)}P[\mathrm{Pois}(h๐ต_x^k)|\mathrm{Pois}(h๐ต_x^k)0]+\frac{h(x)๐ฐ_kh(x)}{h(x)}\delta _0().$$
(7.5)
In particular, since we are assuming that $`h`$ is $`๐ฐ_k`$-subharmonic, this shows that $`Q_k^h(x,)`$ is a probability measure. Let $`X^h`$ be the branching particle system with offspring mechanisms $`Q_1^h,\mathrm{},Q_k^h`$. Let $`Z_x^{h,k}`$ be random variables such that $`(Z_x^{h,k})=Q_k^h(x,)`$. Then, by (3.18), (3.24), (3.20), and (7.3),
$$\begin{array}{c}U_k^hf(x)=P[\mathrm{Thin}_f(Z_x^{h,k})0]=\frac{q_k(x)}{h(x)}P[\mathrm{Thin}_f(\mathrm{Pois}(h๐ต_x^k))0]\hfill \\ =\frac{q_k(x)}{h(x)}P[\mathrm{Pois}(hf๐ต_x^k)0]=\frac{1}{h(x)}๐ฐ_k(hf)(x)(xE^h).\hfill \end{array}$$
(7.6)
If $`xE\backslash E^h`$, then $`๐ฐ_k(hf)(x)๐ฐ_k(h)(x)h(x)=0=:h๐ฐ^h(f)(x)`$. This proves (3.25). To see that $`Q_k^h`$ is a continuous offspring mechanism, by \[Kal76, Theorem 4.2\] it suffices to show that $`xQ_k^h(x,\mathrm{d}\nu )\text{e}^{\nu ,g}`$ is continuous for all bounded $`g๐_+(E^h)`$. Indeed, setting $`f:=1e^g`$, one has $`Q_k^h(x,\mathrm{d}\nu )\text{e}^{\nu ,g}=Q_k^h(x,\mathrm{d}\nu )(1f)^\nu =1๐ฐ_k^hf(x)=1๐ฐ_k(hf)(x)/h(x)`$ which is continuous on $`E^h`$ by the continuity of $`q_k`$ and $`๐ฌ_k`$.
To see that also (3.26) holds, just note that by (3.19), (3.25), and (3.5),
$$\begin{array}{c}P^{(\mathrm{Pois}(h\mu ))}[\mathrm{Thin}_f(X_n^h)=0]=P[\mathrm{Thin}_{U_1^h\mathrm{}U_n^hf}(\mathrm{Pois}(h\mu ))=0]\hfill \\ =P[\mathrm{Pois}((hU_1^h\mathrm{}U_n^hf)\mu )=0]=P[\mathrm{Pois}((๐ฐ_1\mathrm{}๐ฐ_n(hf))\mu )=0]\hfill \\ =P^\mu [\mathrm{Pois}(hf๐ณ_n)=0]=P^\mu [\mathrm{Thin}_f(\mathrm{Pois}(h๐ณ_n))=0].\hfill \end{array}$$
(7.7)
Since this formula holds for all $`fB_{[0,1]}(E^h)`$, formula (3.26) follows.
###### Remark 7.1
(Boundedness of $`h`$) Propositions 3.3 and 3.4 generalize to the case that $`h`$ is unbounded, except that in this case the cluster mechanism in (3.21) and the offspring mechanism in (3.24) need in general not be continuous. Here, in order for (3.22) and (3.25) to be well-defined, one needs to extend the definition of $`๐ฐ_kf`$ to unbounded functions $`f`$, but this can always be done unambiguously \[FS03, Lemma 9\]. $`\mathrm{}`$
### 7.2 Sub- and superharmonic functions
This section contains a number of pivotal calculations involving the log-Laplace operators $`๐ฐ_\gamma `$ from (3.9). In particular, we will prove that the functions $`h_{1,1}`$, $`h_{0,0}`$, and $`h_{0,1}`$ from Lemmas 3.6, 3.7, and 3.8, respectively, are $`๐ฐ_\gamma `$-superharmonic.
We start with an observation that holds for general log-Laplace operators.
###### Lemma 7.2
(Constant multiples) Let $`๐ฐ`$ be a log-Laplace operator of the form (3.2) satisfying (3.3) and let $`fB_+(E)`$. Then $`๐ฐ(rf)r๐ฐf`$ for all $`r1`$, and $`๐ฐ(rf)r๐ฐf`$ for all $`0r1`$. In particular, if $`f`$ is $`๐ฐ`$-superharmonic then $`rf`$ is $`๐ฐ`$-superharmonic for each $`r1`$, and if $`f`$ is $`๐ฐ`$-subharmonic then $`rf`$ is $`๐ฐ`$-superharmonic for each $`0r1`$.
Proof If $`๐ณ`$ is a branching process and $`๐ฐ`$ is the log-Laplace operator of the transition law from $`๐ณ_0`$ to $`๐ณ_1`$ then, using Jensenโs inequality, for all $`r1`$,
$$\text{e}^{\mu ,๐ฐ\left(rf\right)}=E^\mu \left[\text{e}^{๐ณ_1,rf}\right]=E^\mu \left[\left(\text{e}^{๐ณ_1,f}\right)^r\right]\left(E^\mu \left[\text{e}^{๐ณ_1,f}\right]\right)^r=\text{e}^{\mu ,r๐ฐf}.$$
(7.8)
Since this holds for all $`\mu (E)`$, it follows that $`๐ฐ(rf)r๐ฐf`$. The proof of the statements for $`0r1`$ is the same but with the inequality signs reversed.
We next turn our attention to the functions $`h_{1,1}`$ and $`h_{0,0}`$.
###### Lemma 7.3
(The catalyzing function $`h_{1,1}`$) One has
$$๐ฐ_\gamma (rh_{1,1})(x)=\frac{1+\gamma }{\frac{1}{r}+\gamma }(\gamma ,r>0,x[0,1]).$$
(7.9)
In particular, $`h_{1,1}`$ is $`๐ฐ_\gamma `$-harmonic for each $`\gamma >0`$.
Proof Recall (3.7)โ(3.9). Let $`\sigma _{1/r}`$ be an exponentially distributed random variable with mean $`1/r`$, independent of $`\tau _\gamma `$. Then
$$๐ฐ_\gamma (rh_{1,1})(x)=(\frac{1}{\gamma }+1)E\left[1\text{e}^{_0^{\tau _\gamma }rdt}\right]=(\frac{1}{\gamma }+1)P[\sigma _{1/r}<\tau _\gamma ]=(\frac{1}{\gamma }+1)\frac{\gamma }{\frac{1}{r}+\gamma },$$
(7.10)
which yields (7.9).
###### Lemma 7.4
(The catalyzing function $`h_{0,0}`$) One has $`๐ฐ_\gamma (rh_{0,0})rh_{0,0}`$ for each $`\gamma ,r>0`$.
Proof Let $`\mathrm{\Gamma }_x^\gamma `$ be the invariant law from Corollary 5.4. Then, for any $`\gamma >0`$ and $`fB_+[0,1]`$,
$$\begin{array}{ccc}\hfill ๐ฐ_\gamma f(x)& =& (\frac{1}{\gamma }+1)E\left[1\text{e}^{๐ต_x^\gamma ,f}\right](\frac{1}{\gamma }+1)E[๐ต_x^\gamma ,f]\hfill \\ & =& (\frac{1}{\gamma }+1)E\left[_0^{\tau _\gamma }f(๐ฒ_x^\gamma (t/2))dt\right]=(1+\gamma )\mathrm{\Gamma }_x^\gamma ,f(x[0,1]),\hfill \end{array}$$
(7.11)
where we have used that $`\tau _\gamma `$ is independent of $`๐ฒ_x^\gamma `$ and has mean $`\gamma `$. In particular, setting $`f=rh_{0,0}`$ and using (5.25) we find that $`๐ฐ_\gamma (rh_{0,0})rh_{0,0}`$.
The aim of the remainder of this section is to derive various bounds on $`๐ฐ_\gamma f`$ for $`f_{0,1}`$. We start with a formula for $`๐ฐ_\gamma f`$ that holds for general $`[0,1]`$-valued functions $`f`$.
###### Lemma 7.5
(Action of $`๐ฐ_\gamma `$ on $`[0,1]`$-valued functions) Let $`๐ฒ_x^\gamma `$ be the stationary solution to (3.6) and let $`\tau _{\gamma /2}`$ be an independent exponentially distributed random variable with mean $`\gamma /2`$. Let $`(\beta _i)_{i1}`$ be independent exponentially distributed random variables with mean $`\frac{1}{2}`$, independent of $`๐ฒ_x^\gamma `$ and $`\tau _{\gamma /2}`$, and let $`\sigma _k:=_{i=1}^k\beta _i`$ $`(k0)`$. Then
$$1๐ฐ_\gamma f(x)=E\left[\underset{k0:\sigma _k<\tau _\gamma }{}\left(1f(๐ฒ_x^\gamma (\sigma _k))\right)\right](\gamma >0,fB_{[0,1]}[0,1],x[0,1]).$$
(7.12)
Proof By Lemma 7.3, the constant function $`h_{1,1}(x):=1`$ satisfies $`๐ฐ_\gamma h_{1,1}=h_{1,1}`$ for all $`\gamma >0`$. Therefore, by Proposition 3.4, Poissonizing the Poisson-cluster branching process $`๐ณ`$ with the density $`h_{1,1}`$ yields a branching particle system $`X^{h_{1,1}}=(X_n^{h_{1,1}},\mathrm{},X_0^{h_{1,1}})`$ with generating operators $`U_{\gamma _{n1}}^{h_{1,1}},\mathrm{},U_{\gamma _0}^{h_{1,1}}`$, where
$$U_\gamma ^{h_{1,1}}f=๐ฐ_\gamma f(fB_{[0,1]}[0,1],\gamma >0).$$
(7.13)
By (3.18) and (7.5),
$$U_\gamma ^{h_{1,1}}f(x)=1E\left[(1f)^{\mathrm{Pois}\left(๐ต_x^\gamma \right)}|\mathrm{Pois}(๐ต_x^\gamma )0\right](fB_{[0,1]}[0,1],x[0,1],\gamma >0).$$
(7.14)
Therefore, (7.12) will follow provided that
$$P[\mathrm{Pois}(๐ต_x^\gamma )|\mathrm{Pois}(๐ต_x^\gamma )0]=\left(\underset{k0:\sigma _k<\tau _{\gamma /2}}{}\delta _{๐ฒ_x^\gamma (\sigma _k)}\right).$$
(7.15)
Indeed, it is not hard to see that
$$\mathrm{Pois}(๐ต_x^\gamma )\stackrel{๐}{=}\underset{k>0:\sigma _k<\tau _{\gamma /2}}{}\delta _{๐ฒ_x^\gamma (\sigma _k)}.$$
(7.16)
This follows from the facts that $`๐ต_x^\gamma =2_0^{\tau _{\gamma /2}}\delta _{๐ฒ_x^\gamma (s)}ds`$ and
$$\underset{k>0:\sigma _k<\tau _{\gamma /2}}{}\delta _{\sigma _k}\stackrel{๐}{=}\mathrm{Pois}(\mathrm{2\hspace{0.17em}1}_{(\tau _{\gamma /2},0]}).$$
(7.17)
Conditioning $`\mathrm{Pois}(\mathrm{2\hspace{0.17em}1}_{(\tau _{\gamma /2},0]})`$ on being nonzero means conditioning on $`\tau _{\gamma /2}>\sigma _1`$. Since $`\tau _{\gamma /2}\sigma _1`$, conditioned on being nonnegative, is exponentially distributed with mean $`\gamma /2`$, using the stationarity of $`๐ฒ_x^\gamma `$, we arrive at (7.15).
The next lemma generalizes the duality (5.22) to mixed moments of the Wright-Fisher diffusion $`๐ฒ`$ at multiple times. We can interpret the left-hand side of (7.18) as the probability that $`m_1,\mathrm{},m_n`$ organisms sampled from the population at times $`t_1,\mathrm{},t_n`$ are all of the genetic type I.
###### Lemma 7.6
(Sampling at multiple times) Fix $`0t_1<\mathrm{}<t_n=t`$ and nonnegative integers $`m_1,\mathrm{},m_n`$. Let $`๐ฒ`$ be the diffusion in (5.20). Then
$$E^y\left[\underset{k=1}{\overset{n}{}}๐ฒ_{t_k}^{m_k}\right]=E\left[y^{\varphi _t}x^{\psi _t}\right],$$
(7.18)
where $`(\varphi _s,\psi _s)_{s[0,t]}`$ is a Markov process in $`^2`$ started in $`(\varphi _0,\psi _0)=(m_n,0)`$, that jumps deterministically as
$$(\varphi _s,\psi _s)(\varphi _s+m_k,\psi _s)\text{at time}tt_k(k<n),$$
(7.19)
and between these deterministic times jumps with rates as in (5.21).
Proof Induction, with repeated application of (5.22).
For any $`m1`$, we put
$$h_m(x):=1(1x)^m(x[0,1]).$$
(7.20)
The next lemma shows that we have particular good control on the action of $`๐ฐ_\gamma `$ on the functions $`h_m`$.
###### Lemma 7.7
(Action of $`๐ฐ_\gamma `$ on the functions $`h_m`$) Let $`m1`$ and let $`\tau _\gamma `$ be an exponentially distributed random variable with mean $`\gamma `$. Conditional on $`\tau _\gamma `$, let $`(\varphi _t^{},\psi _t^{})_{t0}`$ be a Markov process in $`^2`$, started in $`(\varphi _0^{},\psi _0^{})=(m,0)`$ that jumps at time $`t`$ as:
$$\begin{array}{cccc}\hfill (\varphi _t^{},\psi _t^{})& & (\varphi _t^{}1,\psi _t^{})\hfill & \text{with rate}\varphi _t^{}(\varphi _t^{}1),\hfill \\ \hfill (\varphi _t^{},\psi _t^{})& & (\varphi _t^{}1,\psi _t^{}+1)\hfill & \text{with rate}\frac{1}{\gamma }\varphi _t^{},\hfill \\ \hfill (\varphi _t^{},\psi _t^{})& & (\varphi _t^{}+m,\psi _t^{})\hfill & \text{with rate}1_{\{\tau _{\gamma /2}<t\}}.\hfill \end{array}$$
(7.21)
Then the limit $`lim_t\mathrm{}\psi _t^{}=:\psi _{\mathrm{}}^{}`$ exists a.s., and
$$๐ฐ_\gamma h_m(x)=E^{(m,0)}\left[1(1x)^\psi _{\mathrm{}}^{}\right](m1,x[0,1]).$$
(7.22)
Proof Let $`๐ฒ_x^\gamma `$, $`\tau _{\gamma /2}`$, and $`(\sigma _k)_{k0}`$ be as in Lemma 7.5. Then, by (7.12),
$$๐ฐ_\gamma h_m(x)=1E\left[\underset{k0:\sigma _k<\tau _{\gamma /2}}{}\left(1๐ฒ_x^\gamma (\sigma _k)\right)^m\right].$$
(7.23)
Let $`(\varphi ^{},\psi ^{})=(\varphi _t^{},\psi _t^{})_{t0}`$ be a $`^2`$-valued process started in $`(\varphi _0^{},\psi _0^{})=(m,0)`$ such that conditioned on $`\tau _\gamma `$ and $`(\sigma _k)_{k0}`$, $`(\varphi ^{},\psi ^{})`$ is a Markov process that jumps deterministically as
$$(\varphi _t^{},\psi _t^{})(\varphi _t^{}+m,\psi _s^{})\text{at time}\sigma _k(k1:\sigma _k<\tau _{\gamma /2})$$
(7.24)
and between these times jumps with rates as in (5.21). Then $`(\varphi _t^{},\psi _t^{})(0,\psi _{\mathrm{}}^{})`$ as $`t\mathrm{}`$ a.s. for some $``$-valued random variable $`\psi _{\mathrm{}}^{}`$, and (7.22) follows from Lemma 7.6, using the symmetry $`y1y`$. Since $`\sigma _{k+1}\sigma _k`$ are independent exponentially distributed random variables with mean one, $`(\varphi ^{},\psi ^{})`$ is the Markov process with jump rates as in (7.21).
The next result is a simple application of Lemma 7.7.
###### Lemma 7.8
(The catalyzing function $`h_1`$) The function $`h_1(x):=x`$ $`(x[0,1])`$ is $`๐ฐ_\gamma `$-subharmonic for each $`\gamma >0`$.
Proof Since $`\psi _{\mathrm{}}^{}1`$ a.s., one has $`1(1x)^\psi _{\mathrm{}}^{}x`$ a.s. $`(x[0,1])`$ in (7.22). In particular, setting $`m=1`$ yields $`๐ฐ_\gamma h_1h_1`$.
We now set out to prove that $`h_7`$, which is the function $`h_{0,1}`$ from Lemma 3.8, is $`๐ฐ_\gamma `$-superharmonic. In order to do so, we will derive upper bounds on the expectation of $`\psi _{\mathrm{}}^{}`$. We derive two estimates: one that is good for small $`\gamma `$ and one that is good for large $`\gamma `$.
In order to avoid tedious formal arguments, it will be convenient to recall the interpretation of the process $`(\varphi ^{},\psi ^{})`$ and Lemma 7.6. Recall from the discussion following (5.22) that $`(๐ฒ_x^\gamma (t))_t`$ describes the equilibrium frequency of genetic type $`I`$ as a function of time in a population that is in genetic exchange with an infinite reservoir. From this population we sample at times $`\sigma _k`$ ($`k0`$, $`\sigma _k<\tau _{\gamma /2}`$) each time $`m`$ individuals, and ask for the probability that they are not all of the genetic type II. In order to find this probability, we follow the ancestors of the sampled individuals back in time. Then $`\varphi _t^{}`$ and $`\psi _t^{}`$ are the number of ancestors that lived at time $`t`$ in the population and the reservoir, respectively, and $`E[1(1x)^\psi _{\mathrm{}}^{}]`$ is the probability that at least one ancestor is of type I.
###### Lemma 7.9
(Bound for small $`\gamma `$) For each $`\gamma (0,\mathrm{})`$ and $`m1`$,
$$\frac{1}{m}E^{(m,0)}[\psi _{\mathrm{}}^{}]\frac{1}{m}\underset{i=0}{\overset{m1}{}}\frac{1+\gamma }{1+i\gamma }=:\chi _m(\gamma ).$$
(7.25)
The function $`\chi _m`$ is concave and satisfies $`\chi _m(0)=1`$ for each $`m1`$.
Proof Note that
$$E\left[\left|\{k0:\sigma _k<\tau _{\gamma /2}\}\right|\right]=1+\gamma .$$
(7.26)
We can estimate $`(\varphi ^{},\psi ^{})`$ from above by a process where ancestors from individuals sampled at different times cannot coalesce. Therefore,
$$E^{(m,0)}[\psi _{\mathrm{}}^{}](1+\gamma )E^{(m,0)}[\psi _{\mathrm{}}],$$
(7.27)
where $`(\varphi ,\psi )`$ is the Markov process in (5.21). Note that if $`(\varphi ,\psi )`$ is in the state $`(m+1,0)`$, then the next jump is to $`(m,1)`$ with probability
$$\frac{\frac{1}{\gamma }(m+1)}{\frac{1}{\gamma }(m+1)+m(m+1)}=\frac{1}{1+m\gamma }$$
(7.28)
and to $`(m,0)`$ with one minus this probability. Therefore,
$$\begin{array}{ccc}\hfill E^{(m+1,0)}[\psi _{\mathrm{}}]& =& \frac{1}{1+m\gamma }E^{(m,1)}[\psi _{\mathrm{}}]+\left(1\frac{1}{1+m\gamma }\right)E^{(m,0)}[\psi _{\mathrm{}}]\hfill \\ & =& \frac{1}{1+m\gamma }\left(E^{(m,0)}[\psi _{\mathrm{}}]+1\right)+\left(1\frac{1}{1+m\gamma }\right)E^{(m,0)}[\psi _{\mathrm{}}]\hfill \\ & =& E^{(m,0)}[\psi _{\mathrm{}}]+\frac{1}{1+m\gamma }.\hfill \end{array}$$
(7.29)
By induction, it follows that
$$E^{(m,0)}[\psi _{\mathrm{}}]=\underset{i=0}{\overset{m1}{}}\frac{1}{1+i\gamma }.$$
(7.30)
Inserting this into (7.27) we arrive at (7.25). Finally, since
$$\frac{^2}{\gamma ^2}\frac{1+\gamma }{1+i\gamma }=\frac{2i(i1)}{(1+i\gamma )^3}0(i0,\gamma 0),$$
(7.31)
the function $`\chi _m`$ is convex.
###### Lemma 7.10
(Bound for large $`\gamma `$) For each $`\gamma (0,\mathrm{})`$ and $`m1`$,
$$E^{(m,0)}[\psi _{\mathrm{}}^{}](\frac{1}{\gamma }+1)\underset{k=1}{\overset{m}{}}\frac{1}{k}+\frac{3}{2}.$$
(7.32)
Proof We start by observing that $`\frac{}{t}E[\psi _t^{}]=\frac{1}{\gamma }E[\varphi _t^{}]`$, and therefore
$$E[\psi _{\mathrm{}}^{}]=\frac{1}{\gamma }_0^{\mathrm{}}E[\varphi _t^{}]dt.$$
(7.33)
Unlike in the proof of the last lemma, this time we cannot fully ignore the coalescence of ancestors sampled at different times. In order to deal with this we use a trick: at time zero we introduce an extra ancestor that can only jump to the reservoir when $`t\tau _\gamma `$ and there are no other ancestors left in the population. We further assume that all other ancestors do not jump to the reservoir on their own. Let $`\xi _t`$ be one as long as this extra ancestor is in the population and zero otherwise, and let $`\varphi _t^{\prime \prime }`$ be the number of other ancestors in the population according to these new rules. Then we have at a Markov process $`(\xi ,\varphi ^{\prime \prime })`$ started in $`(\xi _0,\varphi _0^{\prime \prime })=(1,m)`$ that jumps as:
$$\begin{array}{cccc}\hfill (\xi _t,\varphi _t^{\prime \prime })& & (\xi _t,\varphi _t^{\prime \prime }1)\hfill & \text{with rate}(\varphi _t^{\prime \prime }+1)\varphi _t^{\prime \prime },\hfill \\ \hfill (\xi _t,\varphi _t^{\prime \prime })& & (\xi _t,\varphi _t^{\prime \prime }+m)\hfill & \text{with rate}1_{\{\tau _{\gamma /2}<t\}},\hfill \\ \hfill (\xi _t,\varphi _t^{\prime \prime })& & (\xi _t1,\varphi _t^{\prime \prime })\hfill & \text{with rate}\frac{1}{\gamma }1_{\{\tau _{\gamma /2}t\}}1_{\{\varphi _t^{\prime \prime }=0\}}.\hfill \end{array}$$
(7.34)
It is not hard to show that $`(\xi ,\varphi ^{\prime \prime })`$ and $`\varphi ^{}`$ can be coupled such that $`\xi _t+\varphi _t^{\prime \prime }\varphi _t^{}`$ for all $`t0`$. We now simplify even further and ignore all coalescence between ancestors belonging to the process $`\varphi ^{\prime \prime }`$ that are introduced at different times. Let $`\varphi _t^{(k)}`$ be the number of ancestors in the population that were introduced at the time $`\sigma _k`$ $`(k0)`$. Thus, for $`t<\sigma _k`$ one has $`\varphi _t^{(k)}=0`$, for $`t=\sigma _k`$ one has $`\varphi _t^{(k)}=m`$, while for $`t>\sigma _k`$, the process $`\varphi _t^{(k)}`$ jumps from $`n`$ to $`n1`$ with rate $`(n+1)n`$. Then it is not hard to see that, for an appropriate coupling, $`\varphi _t^{\prime \prime }_{k0:\sigma _k<\tau _{\gamma /2}}\varphi _t^{(k)}`$ for all $`t0`$. We let $`\xi ^{}`$ be a process such that $`\xi _0^{}=1`$ and $`\xi _t^{}`$ jumps to zero with rate
$$\frac{1}{\gamma }1_{\{\tau _{\gamma /2}t\}}\underset{k0:\sigma _k<\tau _{\gamma /2}}{}1_{\{\varphi _t^{(k)}=0\}}.$$
(7.35)
Then for an appropriate coupling $`\xi _t^{}\xi _t`$ $`(t0)`$. Thus, we can estimate
$$_0^{\mathrm{}}E[\varphi _t^{}]dt_0^{\mathrm{}}E[\xi _t^{}]dt+_0^{\mathrm{}}E\left[\underset{k0:\sigma _k<\tau _{\gamma /2}}{}\varphi _t^{(k)}\right]dt.$$
(7.36)
Set $`\rho :=inf\{t\tau _{\gamma /2}:\varphi _t^{(k)}=0k0\text{ with }\sigma _k<\tau _{\gamma /2}\}`$ and $`\pi :=inf\{t0:\xi _t^{}=0\}`$. Then
$$_0^{\mathrm{}}E[\xi _t^{}]dt=E[\tau _{\gamma /2}]+E[\rho \tau _{\gamma /2}]+E[\pi \rho ]=\frac{3}{2}\gamma +E[\rho \tau _{\gamma /2}].$$
(7.37)
Since
$$\begin{array}{ccc}\hfill E[\rho \tau _{\gamma /2}]& & _0^{\mathrm{}}E\left[1_{\{_{k0:\sigma _k<\tau _{\gamma /2}}\varphi _t^{(k)}0\}}\right]dt\hfill \\ & & _0^{\mathrm{}}E\left[\underset{k0:\sigma _k<\tau _{\gamma /2}}{}1_{\{\varphi _t^{(k)}0\}}\right]dt,\hfill \end{array}$$
(7.38)
using moreover (7.36) and (7.37), we can estimate
$$_0^{\mathrm{}}E[\varphi _t^{}]dt\frac{3}{2}\gamma +_0^{\mathrm{}}E\left[\underset{k0:\sigma _k<\tau _{\gamma /2}}{}(\varphi _t^{(k)}+1_{\{\varphi _t^{(k)}0\}})\right]dt.$$
(7.39)
Since $`E\left[\left|\{k0:\sigma _k<\tau _{\gamma /2}\}\right|\right]=1+\gamma `$, we obtain
$$_0^{\mathrm{}}E[\varphi _t^{}]dt\frac{3}{2}\gamma +(1+\gamma )_0^{\mathrm{}}E[\varphi _t^{(0)}+1_{\{\varphi _t^{(0)}0\}}]dt.$$
(7.40)
Since $`\varphi _t^{(0)}`$ jumps from $`n`$ to $`n1`$ with rate $`(n+1)n`$, the expected total time that $`\varphi _t^{(0)}=n`$ equals $`1/((n+1)n)`$, and therefore
$$_0^{\mathrm{}}E[\varphi _t^{(0)}+1_{\{\varphi _t^{(0)}0\}}]dt=\underset{n=1}{\overset{m}{}}\frac{1}{(n+1)n}(n+1_{\{n0\}})=\underset{n=1}{\overset{m}{}}\frac{1}{n}.$$
(7.41)
Inserting this into (7.40), using (7.33), we arrive at (7.32).
###### Lemma 7.11
(The catalyzing function $`h_{0,1}`$) One has $`๐ฐ_\gamma (h_{0,1})h_{0,1}`$ for each $`\gamma >0`$. Moreover, for each $`r>1`$ and $`\gamma >0`$,
$$\underset{x(0,1]}{sup}\frac{๐ฐ_\gamma (rh_{0,1})(x)}{rh_{0,1}(x)}<1.$$
(7.42)
Proof Recall that $`h_{0,1}(x)=h_7(x)=1(1x)^7`$ $`(x[0,1])`$. We will show that
$$E^{(7,0)}[\psi _{\mathrm{}}^{}]<7$$
(7.43)
for each $`\gamma (0,\mathrm{})`$. The function $`\chi _m(\gamma )`$ from Lemma 7.9 satisfies
$$\chi _m(1)=\frac{1}{m}\underset{n=1}{\overset{m}{}}\frac{2}{n}<1(m5).$$
(7.44)
Since $`\chi _m(\gamma )`$ is concave in $`\gamma `$ and satisfies $`\chi _m(0)=1`$, it follows that $`\chi _m(\gamma )<1`$ for all $`0<\gamma 1`$ and $`m5`$. By Lemma 7.10, for all $`\gamma 1`$,
$$E^{(m,0)}[\psi _{\mathrm{}}^{}]2\underset{k=1}{\overset{m}{}}\frac{1}{k}+\frac{3}{2}<m(m7).$$
(7.45)
Therefore, if $`m7`$, then $`m^{}:=E^{(m,0)}[\psi _{\mathrm{}}^{}]<m`$. It follows by (7.22) and Jensenโs inequality applied to the concave function $`z1(1x)^z`$ that
$$๐ฐ_\gamma h_m(x)1(1x)^{E^{(m,0)}[\psi _{\mathrm{}}^{}]}=1(1x)^m^{}h_m(x)(x[0,1],\gamma >0).$$
(7.46)
This shows that $`h_m`$ is $`๐ฐ_\gamma `$-superharmonic for each $`\gamma >0`$. By Lemma 7.2, for each $`r>1`$,
$$\frac{๐ฐ_\gamma (rh_m)(x)}{rh_m(x)}\frac{r๐ฐ_\gamma (h_m)(x)}{rh_m(x)}\frac{1(1x)^m^{}}{1(1x)^m}(x(0,1]).$$
(7.47)
By Lemma 7.3 and the monotonicity of $`๐ฐ_\gamma `$,
$$\frac{๐ฐ_\gamma (rh_m)(x)}{rh_m(x)}\frac{๐ฐ_\gamma (r)(x)}{rh_m(x)}\frac{1+\gamma }{1+r\gamma }\frac{1}{(1(1x)^m)}(x(0,1]).$$
(7.48)
Since the right-hand side of (7.47) is smaller than $`1`$ for $`x(0,1)`$ and tends to $`m^{}/m<1`$ as $`x0`$, since the right-hand side of (7.48) is smaller than $`1`$ for $`x`$ in an open neighborhood of $`1`$, and since both bounds are continuous, (7.42) follows.
### 7.3 Extinction versus unbounded growth
In this section we show that Lemmas 3.63.8 are equivalent to Proposition 3.9. (This follows from the equivalence of conditions (i) and (ii) in Lemma 7.12 below.) We moreover prove Lemmas 3.6 and 3.8 and prepare for the proof of Lemma 3.7. We start with some general facts about log-Laplace operators and branching processes.
For the next lemma, let $`E`$ be a separable, locally compact, metrizable space. For $`n0`$, let $`q_n๐_+(E)`$ be continuous weight functions, let $`๐ฌ_n`$ be continuous cluster mechanisms on $`E`$, and assume that the associated log-Laplace operators $`๐ฐ_n`$ defined in (3.2) satisfy (3.3). Assume that $`0h๐_+(E)`$ is bounded and $`๐ฐ_n`$-superharmonic for all $`n`$, let $`E^h:=\{xE:h(x)>0\}`$, and define generating operators $`U_n^h:B_{[0,1]}(E^h)B_{[0,1]}(E)`$ as in (3.25). For each $`n0`$, let $`(๐ณ_0^{(n)},๐ณ_1^{(n)})`$ be a one-step Poisson cluster branching process with log-Laplace operator $`๐ฐ_n`$, and let $`(X_0^{(n),h},X_1^{(n),h})`$ be the one-step branching particle system with generating operator $`U_n^h`$. (In a typical application of this lemma, the operators $`๐ฐ_n`$ will be iterates of other log-Laplace operators, and $`๐ณ_0^{(n)},๐ณ_1^{(n)}`$ will be the initial and final state, respectively, of a Poisson cluster branching process with many time steps.)
###### Lemma 7.12
(Extinction versus unbounded growth) Assume that $`\rho ๐_{[0,1]}(E^h)`$ and put
$$p(x):=\{\begin{array}{cc}h(x)\rho (x)\hfill & \text{if}xE^h,\hfill \\ 0\hfill & \text{if}xE\backslash E^h.\hfill \end{array}$$
(7.49)
Then the following statements are equivalent:
$$\begin{array}{cc}\hfill (\mathrm{i})& P^{\delta _x}\left[|X_1^{(n),h}|\right]\underset{n\mathrm{}}{}\rho (x)\delta _{\mathrm{}}+(1\rho (x))\delta _0\hfill \\ & \text{locally uniformly for }xE^h,\hfill \\ \hfill (\mathrm{ii})& P^{\delta _x}\left[๐ณ_1^{(n)},h\right]\underset{n\mathrm{}}{}\text{e}^{p\left(x\right)}\delta _0+\left(1\text{e}^{p\left(x\right)}\right)\delta _{\mathrm{}}\hfill \\ & \text{locally uniformly for }xE,\hfill \\ \hfill (\mathrm{iii})& ๐ฐ_n(\lambda h)(x)\underset{n\mathrm{}}{}p(x)\hfill \\ & \text{locally uniformly for }xE\lambda >0,\hfill \\ \hfill (\mathrm{iv})& 0<\lambda _1<\lambda _2<\mathrm{}:๐ฐ_n(\lambda _ih)(x)\underset{n\mathrm{}}{}p(x)\hfill \\ & \text{locally uniformly for }xE(i=1,2).\hfill \end{array}$$
Proof of Lemma 7.12 It is not hard to see that (i) is equivalent to
$$P^{\delta _x}[\mathrm{Thin}_\lambda (X_1^{(n),h})0]\underset{n\mathrm{}}{}\rho (x)$$
(7.50)
locally uniformly for $`xE^h`$, for all $`0<\lambda 1`$. It follows from (3.19) and (3.25) that $`h(x)P^{\delta _x}[\mathrm{Thin}_\lambda (X_1^{(n),h})0]=hU^h(\lambda )(x)=๐ฐ(\lambda h)(x)`$ $`(xE)`$, so (i) is equivalent to
$$\begin{array}{c}(\mathrm{i})^{}๐ฐ_n(\lambda h)(x)\underset{n\mathrm{}}{}p(x)\hfill \\ \text{locally uniformly for }xE0<\lambda 1.\hfill \end{array}$$
By (3.4), condition (ii) implies that
$$\text{e}^{๐ฐ_n\left(\lambda h\right)\left(x\right)}=E^{\delta _x}\left[\text{e}^{\lambda ๐ณ_1,h}\right]\underset{n\mathrm{}}{}\text{e}^{p\left(x\right)}$$
(7.51)
locally uniformly for $`xE`$ for all $`\lambda >0`$, and therefore (ii) implies (iii). Obviously (iii)$`(\mathrm{i})^{}`$(iv) so we are done if we show that (iv)$``$(ii). Indeed, (iv) implies that
$$E^{\delta _x}\left[\text{e}^{\lambda _1๐ณ_1^{\left(n\right)},h}\text{e}^{\lambda _2๐ณ_1^{\left(n\right)},h}\right]\underset{n\mathrm{}}{}0$$
(7.52)
locally uniformly for $`xE`$, which shows that
$$P^{\delta _x}\left[c<๐ณ_1^{(n)},h<C\right]\underset{n\mathrm{}}{}0$$
(7.53)
for all $`0<c<C<\mathrm{}`$. Using (iv) once more we arive at (ii).
Our next lemma gives sufficient conditions for the $`n`$-th iterates of a single log-Laplace operator $`๐ฐ`$ to satisfy the equivalent conditions of Lemma 7.12. Let $`E`$ (again) be separable, locally compact, and metrizable. Let $`q๐_+(E)`$ be a weight function, $`๐ฌ`$ a continuous cluster mechanism on $`E`$, and assume that the associated log-Laplace operator $`๐ฐ`$ defined in (3.2) satisfies (3.3). Let $`๐ณ=(๐ณ_0,๐ณ_1,\mathrm{})`$ be the Poisson-cluster branching process with log-Laplace operator $`๐ฐ`$ in each step, let $`0h๐_+(E)`$ be bounded and $`๐ฐ`$-superharmonic, and let $`X^h=(X_0^h,X_1^h,\mathrm{})`$ denote the branching particle system on $`E^h`$ obtained from $`๐ณ`$ by Poissonization with a $`๐ฐ`$-superharmonic function $`h`$, in the sense of Proposition 3.4.
###### Lemma 7.13
(Sufficient condition for extinction versus unbounded growth) Assume that
$$\underset{xE^h}{sup}\frac{๐ฐh(x)}{h(x)}<1.$$
(7.54)
Then the process $`X^h`$ started in any initial law $`(X_0^h)_1(E^h)`$ satisfies
$$\underset{k\mathrm{}}{lim}|X_k^h|=\mathrm{}\text{or}k0\text{ s.t. }X_k^h=0\text{a.s.}$$
(7.55)
Moreover, if the function $`\rho :E^h[0,1]`$ defined by
$$\rho (x):=P^{\delta _x}[X_n^h0n0](xE^h)$$
(7.56)
satisfies $`inf_{xE^h}\rho (x)>0`$, then $`\rho `$ is continuous.
Proof of Lemma 7.13 Let $`๐`$ denote the tail event $`๐=\{X_n^h0n0\}`$ and let $`(_k)_{k0}`$ be the filtration generated by $`X^h`$. Then, by the Markov property and continuity of the conditional expectation with respect to increasing limits of $`\sigma `$-fields (see Complement 10(b) from \[Loe63, Section 29\] or \[Loe78, Section 32\])
$$P[X_n^h0n0|X_k]=P(๐|_k)\underset{k\mathrm{}}{}1_๐\text{a.s.}$$
(7.57)
In particular, this implies that a.s. on the event $`๐`$ one must have $`P[X_{k+1}^h=0|X_k^h]0`$ a.s. By (3.19) and (3.25), $`P^{\delta _x}[X_1^h0]=U^h1(x)=(๐ฐh(x))/h(x)`$, which is uniformly bounded away from one by (7.54). Therefore, $`P[X_{k+1}^h=0|X_k^h]0`$ a.s. on $`๐`$ is only possible if the number of particles tends to infinity.
The continuity of $`\rho `$ can be proved by a straightforward adaptation of the proof of \[FS04, Proposition 5 (d)\] to the present setting with discrete time and noncompact space $`E`$. An essential ingredient in the proof, apart from (7.54), is the fact that the map $`\nu P^\nu [X_n^h]`$ from $`๐ฉ(E)`$ to $`_1(๐ฉ(E))`$ is continuous, which follows from the continuity of $`Q^h`$.
We now turn our attention more specifically to the renormalization branching process $`๐ณ`$. In the remainder of this section, $`(\gamma _k)_{k0}`$ is a sequence of positive constants such that $`_n\gamma _n=\mathrm{}`$ and $`\gamma _n\gamma ^{}`$ for some $`\gamma ^{}[0,\mathrm{})`$, and $`๐ณ=(๐ณ_n,\mathrm{},๐ณ_0)`$ is the Poisson cluster branching process on $`[0,1]`$ defined in Section 3.2. We put $`๐ฐ^{(n)}:=๐ฐ_{\gamma _{n1}}\mathrm{}๐ฐ_{\gamma _0}`$. If $`0h๐[0,1]`$ is $`๐ฐ_{\gamma _k}`$-superharmonic for all $`k0`$, then $`๐ณ^h`$ and $`X^h`$ denote the branching process and the branching particle system on $`\{x[0,1]:h(x)>0\}`$ obtained from $`๐ณ`$ by weighting and Poissonizing with $`h`$ in the sense of Propositions 3.3 and 3.4, respectively.
Proof of Lemma 3.6 By induction, it follows from Lemma 7.3 that
$$๐ฐ^{(n)}(\lambda h_{1,1})=\frac{_{k=0}^{n1}(1+\gamma _k)}{_{k=0}^{n1}(1+\gamma _k)1+\frac{1}{\lambda }}(\lambda >0).$$
(7.58)
It is not hard to see (compare the footnote at (2.12)) that
$$\underset{k=0}{\overset{\mathrm{}}{}}(1+\gamma _k)=\mathrm{}\text{if and only if}\underset{k=0}{\overset{\mathrm{}}{}}\gamma _k=\mathrm{}.$$
(7.59)
Therefore, since we are assuming that $`_n\gamma _n=\mathrm{}`$,
$$๐ฐ^{(n)}(\lambda h_{1,1})\underset{n\mathrm{}}{}h_{1,1},$$
(7.60)
uniformly on $`[0,1]`$ for all $`\lambda >0`$. The result now follows from Lemma 7.12 (with $`h=h_{1,1}`$ and $`\rho (x)=1`$ $`(x[0,1])`$).
###### Remark 7.14
(Conditions on $`(\gamma _n)_{n0}`$) Our proof of Lemma 3.6 does not use that $`\gamma _n\gamma ^{}`$ for some $`\gamma ^{}[0,\mathrm{})`$. On the other hand, the proof shows that $`_n\gamma _n=\mathrm{}`$ is a necessary condition for (3.29). $`\mathrm{}`$
We do not know if the assumption that $`\gamma _n\gamma ^{}`$ for some $`\gamma ^{}[0,\mathrm{})`$ is needed in Lemma 3.7. We guess that it can be dropped, but it will greatly simplify proofs to have it around.
We will show that in order to prove Lemmas 3.7 and 3.8, it suffices to prove their analogues for embedded particle systems in the time-homogeneous processes $`๐ด^\gamma ^{}`$ ($`\gamma ^{}[0,\mathrm{})`$). More precisely, we will derive Lemmas 3.7 and 3.8 from the following two results. Below, $`(๐ฐ_t^0)_{t0}`$ is the log-Laplace semigroup of the super-Wright-Fisher diffusion $`๐ด^0`$, defined in (3.15). The functions $`p_{0,1,\gamma ^{}}^{}`$ ($`\gamma ^{}[0,\mathrm{})`$) are defined in (3.34).
###### Proposition 7.15
(Time-homogeneous embedded particle system with $`h_{0,0}`$)
(a) For any $`\gamma ^{}>0`$, one has $`(๐ฐ_\gamma ^{})^nh_{0,0}\underset{n\mathrm{}}{}0`$ uniformly on $`[0,1]`$.
(b) One has $`๐ฐ_t^0h_{0,0}\underset{t\mathrm{}}{}0`$ uniformly on $`[0,1]`$.
###### Proposition 7.16
(Time-homogeneous embedded particle system with $`h_{0,1}`$)
(a) For any $`\gamma ^{}>0`$, one has $`(๐ฐ_\gamma ^{})^n(\lambda h_{0,1})\underset{n\mathrm{}}{}p_{0,1,\gamma ^{}}^{}`$ uniformly on $`[0,1]`$, for all $`\lambda >0`$.
(b) One has $`๐ฐ_t^0(\lambda h_{0,1})\underset{t\mathrm{}}{}p_{0,1,0}^{}`$ uniformly on $`[0,1]`$, for all $`\lambda >0`$.
Proposition 7.15 (a) will be proved in Section 8.2.
Proof of Proposition 7.16 (a) By formula (7.42) from Lemma 7.11, for each $`r>1`$ the function $`rh_{0,1}`$ satisfies condition (7.54) from Lemma 7.13. Set $`\rho (x):=P^{\delta _x}[Y_n^{\gamma ^{},rh_{0,1}}0n]`$. Then, by (3.19) and (3.25),
$$\begin{array}{ccc}\hfill \rho (x)& =& \underset{n\mathrm{}}{lim}P^{\delta _x}[Y_n^{\gamma ^{},rh_{0,1}}0]=\underset{n\mathrm{}}{lim}(U_\gamma ^{}^{rh_{0,1}})^n1(x)\hfill \\ & =& \underset{n\mathrm{}}{lim}\frac{(๐ฐ_\gamma ^{})^n(rh_{0,1})(x)}{rh_{0,1}(x)}\frac{h_1(x)}{rh_{0,1}(x)}(x(0,1]),\hfill \end{array}$$
(7.61)
where $`h_1(x)=x`$ $`(x[0,1])`$ is the $`๐ฐ_\gamma ^{}`$-subharmonic function from Lemma 7.8. It follows that $`inf_{x(0,1]}\rho (x)>0`$ and therefore, by Lemma 7.13, $`\rho `$ is continuous in $`x`$.
By Lemma 7.13, we see that the Poissonized particle system $`X^{rh_{0,1}}`$ exhibits extinction versus unbounded growth in the sense of Lemma 7.12, which implies the statement in Proposition 7.16 (a).
Proof of Propositions 7.15 (b) and 7.16 (b) These statements follow from results in \[FS03\]. Indeed, \[FS03, Proposition 2\] implies that for any $`fB_+[0,1]`$ and $`x[0,1]`$,
$$\begin{array}{cccc}\hfill ๐ฐ_t^0f(x)& \underset{t\mathrm{}}{}& 0\hfill & \text{if }f(0)=f(1)=0,\hfill \\ \hfill ๐ฐ_t^0f(x)& \underset{t\mathrm{}}{}& p_{0,1,\gamma ^{}}^{}(x)\hfill & \text{if }f(0)=0,f(1)>0.\hfill \end{array}$$
(7.62)
To see that the convergence in (7.62) is in fact uniform in $`x[0,1]`$ we use the fact that each function $`fB_+[0,1]`$ with $`f(0)=f(1)=0`$ can be bounded as $`fr1_{(0,1)}`$ for some $`r1`$, and that each function $`fB_+[0,1]`$ with $`f(0)=0`$ and $`f(1)>0`$ can be bounded as $`\epsilon 1_{\{1\}}fr1_{(0,1]}`$ for some $`0<\epsilon 1`$ and $`r1`$. Therefore, by the monotonity of $`๐ฐ_t^0`$, it suffices to show that $`๐ฐ_t^0(r1_{(0,1)})`$, $`๐ฐ_t^0(r1_{(0,1]})`$, and $`๐ฐ_t^0(\epsilon 1_{\{1\}})`$ converge uniformly on $`[0,1]`$. By \[FS03, Lemma 15\], these functions are continuous for each $`t>0`$, and since moreover the limit functions are continuous, it suffices to show that the convergence is monotone. Thus, we claim that
$$\begin{array}{cccc}\hfill ๐ฐ_t^0(r1_{(0,1)})& & 0\hfill & (r1),\hfill \\ \hfill ๐ฐ_t^0(r1_{(0,1]})& & p_{0,1,\gamma ^{}}^{}\hfill & (r1),\hfill \\ \hfill ๐ฐ_t^0(\epsilon 1_{\{1\}})& & p_{0,1,\gamma ^{}}^{}\hfill & (0<\epsilon 1).\hfill \end{array}$$
(7.63)
By (an obvious analogue of) Lemma 7.2, it suffices to show that $`1_{(0,1)}`$ and $`1_{(0,1]}`$ are $`๐ฐ_t^0`$-superharmonic, while $`1_{\{1\}}`$ is $`๐ฐ_t^0`$-subharmonic for each $`t0`$. Let $`(๐ด_t^{0,h_{1,1}})_{t0}`$ be the branching particle system obtained from $`(๐ด_t^0)_{t0}`$ by Poissonization with the constant function $`h_{1,1}:=1`$. Then $`๐ด^{0,h_{1,1}}`$ is a system of binary splitting Wright-Fisher diffusions, which was also studied in \[FS03\]. One has (compare (3.19))
$$๐ฐ_t^01_{(0,1)}(x)=P[\mathrm{Thin}_{๐ฐ_t^01_{(0,1)}}(\delta _x)0]=P^{\delta _x}[\mathrm{Thin}_{1_{(0,1)}}(Y_t^{0,h_{1,1}})0]=P^{\delta _x}[Y_t^{0,h_{1,1}}((0,1))>0].$$
(7.64)
Likewise,
$$๐ฐ_t^01_{(0,1]}(x)=P^{\delta _x}[Y_t^{0,h_{1,1}}((0,1])>0]\text{and}๐ฐ_t^01_{\{1\}}(x)=P^{\delta _x}[Y_t^{0,h_{1,1}}(\{1\})>0].$$
(7.65)
Using the fact that the points $`0,1`$ are traps for the Wright-Fisher diffusion and that in a binary splitting Wright-Fisher diffusion, particles never die, it is easy to see that $`P^{\delta _x}[Y_t^{0,h_{1,1}}((0,1))>0]`$ and $`P^{\delta _x}[Y_t^{0,h_{1,1}}((0,1])>0]`$ are nonincreasing in $`t`$, while $`P^{\delta _x}[Y_t^{0,h_{1,1}}(\{1\})>0]`$ is nondecreasing in $`t`$.
We now show that Propositions 7.15 and 7.16 imply Lemmas 3.7 and 3.8, respectively.
Proof of Lemma 3.7 We start with the proof that the embedded particle system $`X^{h_{0,0}}`$ is critical. For any $`fB_+[0,1]`$ and $`k1`$, we have, by Poissonization (Proposition 3.4) and the definition of $`๐ณ`$,
$$\begin{array}{c}h_{0,0}(x)E^{k,\delta _x}[X_{k+1}^{h_{0,0}},f]=E^{k,(\mathrm{Pois}(h_{0,0}\delta _x))}[X_{k+1}^{h_{0,0}},f]=E^{k,\delta _x}[\mathrm{Pois}(h_{0,0}๐ณ_{k+1}),f]\hfill \\ =E^{k,\delta _x}[๐ณ_{k+1},h_{0,0}f]=(\frac{1}{\gamma }+1)E[๐ต_x^\gamma ,h_{0,0}f]=(\frac{1}{\gamma }+1)\mathrm{\Gamma }_x^{\gamma _{k1}},h_{0,0}f,\hfill \end{array}$$
(7.66)
where $`\mathrm{\Gamma }_x^\gamma `$ is the invariant law of $`๐ฒ_x^\gamma `$ from Corollary 5.4. In particular, setting $`f=1`$ gives $`h_{0,0}(x)E^{k,\delta _x}[|X_{k+1}^{h_{0,0}}|]=h_{0,0}(x)`$ by (5.25).
To prove (3.30), by Lemma 7.12 it suffices to show that
$$๐ฐ^{(n)}(\lambda h_{0,0})\underset{n\mathrm{}}{}0$$
(7.67)
uniformly on $`[0,1]`$ for all $`0<\lambda 1`$. We first treat the case $`\gamma ^{}>0`$. Then, by Theorem 3.2 (a), for each fixed $`l1`$ and $`f๐_+[0,1]`$,
$$๐ฐ_{\gamma _{n1}}\mathrm{}๐ฐ_{\gamma _{nl}}f\underset{n\mathrm{}}{}(๐ฐ_\gamma ^{})^lf$$
(7.68)
uniformly on $`[0,1]`$. Therefore, by a diagonal argument, we can find $`l(n)\mathrm{}`$ such that
$$(๐ฐ_\gamma ^{})^{l(n)}h_{0,0}๐ฐ_{\gamma _{n1}}\mathrm{}๐ฐ_{\gamma _{nl(n)}}h_{0,0}_{\mathrm{}}\underset{n\mathrm{}}{}0.$$
(7.69)
Using the fact that the function $`h_{0,0}`$ is $`๐ฐ_\gamma `$-superharmonic for each $`\gamma >0`$ and the monotonicity of the operators $`๐ฐ_\gamma `$, we derive from Proposition 7.15 (a) that
$$๐ฐ^{(n)}(\lambda h_{0,0})๐ฐ_{\gamma _{n1}}\mathrm{}๐ฐ_{\gamma _{nl(n)}}h_{0,0}\underset{n\mathrm{}}{}0$$
(7.70)
uniformly on $`[0,1]`$ for all $`0<\lambda 1`$. This proves (7.67) in the case $`\gamma ^{}>0`$.
The proof in the case $`\gamma ^{}=0`$ is similar. In this case, by Theorem 3.2 (b), for each fixed $`t>0`$ and $`f๐_+[0,1]`$,
$$๐ฐ_{\gamma _{n1}}\mathrm{}๐ฐ_{\gamma _{k_n(t)}}f(x_n)\underset{n\mathrm{}}{}๐ฐ_t^0f(x)x_nx[0,1],$$
(7.71)
which shows that $`๐ฐ_{\gamma _{n1}}\mathrm{}๐ฐ_{\gamma _{k_n(t)}}f`$ converges to $`๐ฐ_t^0f`$ uniformly on $`[0,1]`$. By a diagonal argument, we can find $`t(n)\mathrm{}`$ such that
$$๐ฐ_t^0(h_{0,0})๐ฐ_{\gamma _{n1}}\mathrm{}๐ฐ_{\gamma _{k_n(t(n))}}(h_{0,0})_{\mathrm{}}\underset{n\mathrm{}}{}0,$$
(7.72)
and the proof proceeds in the same way as before.
Proof of Lemma 3.8 By Lemma 7.12 and the monotonicity of the operators $`๐ฐ_\gamma `$ it suffices to show that
$$\begin{array}{cc}\hfill (\mathrm{i})& \underset{n\mathrm{}}{lim\; sup}๐ฐ^{(n)}(h_{0,1})p_{0,1,\gamma ^{}}^{},\hfill \\ \hfill (\mathrm{ii})& \underset{n\mathrm{}}{lim\; inf}๐ฐ^{(n)}(\frac{1}{2}h_{0,1})p_{0,1,\gamma ^{}}^{},\hfill \end{array}$$
(7.73)
uniformly on $`[0,1]`$. We first consider the case $`\gamma ^{}>0`$. By (7.68) and a diagonal argument, we can find $`l(n)\mathrm{}`$ such that
$$(๐ฐ_\gamma ^{})^{l(n)}h_{0,1}๐ฐ_{\gamma _{n1}}\mathrm{}๐ฐ_{\gamma _{nl(n)}}h_{0,1}_{\mathrm{}}\underset{n\mathrm{}}{}0.$$
(7.74)
Therefore, by Proposition 7.16 (a), the fact that $`h_{0,1}`$ is $`๐ฐ_{\gamma _k}`$-superharmonic for each $`k0`$, and the monotonicity of the operators $`๐ฐ_\gamma `$, we find that
$$๐ฐ^{(n)}h_{0,1}๐ฐ_{\gamma _{n1}}\mathrm{}๐ฐ_{\gamma _{nl(n)}}h_{0,1}\underset{n\mathrm{}}{}p_{0,1,\gamma ^{}}^{},$$
(7.75)
uniformly on $`[0,1]`$. This proves (7.73) (i). To prove also (7.73) (ii) we use the $`๐ฐ_\gamma `$-subharmonic (for each $`\gamma >0`$) function $`h_1`$ from Lemma 7.8. By Lemma 7.2 also $`\frac{1}{2}h_1`$ is $`๐ฐ_\gamma `$-subharmonic. By bounding $`\frac{1}{2}h_1`$ from above and below with multiples of $`h_{0,1}`$ it is easy to derive from Proposition 7.16 (a) that
$$(๐ฐ_\gamma ^{})^n(\frac{1}{2}h_1)\underset{n\mathrm{}}{}p_{0,1,\gamma ^{}}^{}$$
(7.76)
uniformly on $`[0,1]`$. Arguing as before, we can find $`l(n)\mathrm{}`$ such that
$$(๐ฐ_\gamma ^{})^{l(n)}(\frac{1}{2}h_1)๐ฐ_{\gamma _{n1}}\mathrm{}๐ฐ_{\gamma _{nl(n)}}(\frac{1}{2}h_1)_{\mathrm{}}\underset{n\mathrm{}}{}0.$$
(7.77)
Therefore, by (7.76) and the facts that $`\frac{1}{2}h_1`$ is $`๐ฐ_{\gamma _k}`$-subharmonic for each $`k0`$ and $`\frac{1}{2}h_1\frac{1}{2}h_{0,1}`$,
$$๐ฐ^{(n)}(\frac{1}{2}h_{0,1})๐ฐ_{\gamma _{n1}}\mathrm{}๐ฐ_{\gamma _{nl(n)}}(\frac{1}{2}h_1)\underset{n\mathrm{}}{}p_{0,1,\gamma ^{}}^{},$$
(7.78)
uniformly on $`[0,1]`$, which proves (7.73) (ii). The proof of (7.73) in case $`\gamma ^{}=0`$ is completely analogous.
## 8 Extinction on the interior
### 8.1 Basic facts
In this section we prove Proposition 7.15 (a). To simplify notation, throughout this section $`h`$ denotes the function $`h_{0,0}`$. We fix $`0<\gamma ^{}<\mathrm{}`$, we let $`Y^h:=Y^{\gamma ^{},h}`$ denote the branching particle system on $`(0,1)`$ obtained from $`๐ด^\gamma ^{}=(๐ด_0^\gamma ^{},๐ด_1^\gamma ^{},\mathrm{})`$ by Poissonization with $`h`$ in the sense of Proposition 3.4, and we denote its log-Laplace operator by $`U_\gamma ^{}^h`$. We will prove that
$$\rho (x):=P^{\delta _x}\left[Y_n^h0n0\right]=0(x(0,1)).$$
(8.1)
Since for each $`n`$ fixed, $`x\rho _n(x):=P^{\delta _x}[Y_n^h0]`$ is a continuous function that decreases to $`\rho (x)`$, (8.1) implies that $`\rho _n(x)0`$ locally uniformly on $`(0,1)`$, which, by an obvious analogon of Lemma 7.12, yields Proposition 7.15 (a).
As a first step, we prove:
###### Lemma 8.1
(Continuous survival probability) One has either $`\rho (x)=0`$ for all $`x(0,1)`$ or there exists a continuous function $`\stackrel{~}{\rho }:(0,1)[0,1]`$ such that $`\rho (x)\stackrel{~}{\rho }(x)>0`$ for all $`x(0,1)`$.
Proof Put $`p(x):=h(x)\rho (x)`$. We will show that either $`p=0`$ on $`(0,1)`$ or there exists a continuous function $`\stackrel{~}{p}:(0,1)(0,1]`$ such that $`p\stackrel{~}{p}`$ on $`(0,1)`$. Indeed,
$$\begin{array}{c}p(x)=h(x)P^{\delta _x}\left[Y_n^h0n0\right]=\underset{n\mathrm{}}{lim}h(x)P^{\delta _x}\left[Y_n^h0\right]\hfill \\ =h(x)\underset{n\mathrm{}}{lim}(U_\gamma ^{}^h)^n1(x)=\underset{n\mathrm{}}{lim}(๐ฐ_\gamma ^{})^nh(x)(x(0,1)),\hfill \end{array}$$
(8.2)
where we have used (3.19) and (3.25) in the last two steps. Using the continuity of $`๐ฐ_\gamma ^{}`$ with respect to decreasing sequences, it follows that
$$๐ฐ_\gamma ^{}p=p.$$
(8.3)
We claim that for any $`fB_{[0,1]}[0,1]`$, one has the bounds
$$\mathrm{\Gamma }_x^\gamma ,f๐ฐ_\gamma f(x)(1+\gamma )\mathrm{\Gamma }_x^\gamma ,f(\gamma >0,x[0,1]).$$
(8.4)
Indeed, by Lemma 7.5, $`๐ฐ_\gamma f(x)1E[(1f(๐ฒ_x^\gamma (0)))]=\mathrm{\Gamma }_x^\gamma ,f`$, while the upper bound in (8.4) follows from (7.11).
By Remark 5.5, $`(0,1)x\mathrm{\Gamma }_x^\gamma ,f`$ is continuous for all $`fB_{[0,1]}[0,1]`$. Moreover, $`\mathrm{\Gamma }_x^\gamma ,f=0`$ for some $`x(0,1)`$ if and only if $`f=0`$ almost everywhere with respect to Lebesgue measure.
Applying these facts to $`f=p`$ and $`\gamma =\gamma ^{}`$, using (8.3), we see that there are two possibilities. Either $`p=0`$ a.s. with respect to Lebesgue measure, and in this case $`p=0`$ by the upper bound in (8.4), or $`p`$ is not almost everywhere zero with respect to Lebesgue measure, and in this case the function $`x\stackrel{~}{p}(x):=\mathrm{\Gamma }_x^\gamma ,f`$ is continuous, positive on $`(0,1)`$, and estimates $`p`$ from below by the lower bound in (8.4).
### 8.2 A representation for the Campbell law
(Local) extinction properties of critical branching processes are usually studied using Palm laws. Our proof of formula (8.1) is no exception, except that we will use the closely related Campbell laws. Loosely speaking, Palm laws describe a population that is size-biased at a given position, plus โtypicalโ particle sampled from that position, while Campbell laws describe a population that is size-biased as a whole, plus a โtypicalโ particle sampled from a random position.
Let $`๐ซ`$ be a probability law on $`๐ฉ(0,1)`$ with $`_{๐ฉ(0,1)}๐ซ(\mathrm{d}\nu )|\nu |=1`$. Then the size-biased law $`๐ซ_{\mathrm{size}}`$ associated with $`๐ซ`$ is the probability law on $`๐ฉ(0,1)`$ defined by
$$๐ซ_{\mathrm{size}}():=_{๐ฉ(0,1)}๐ซ(\mathrm{d}\nu )|\nu |1_{\left\{\nu \right\}}.$$
(8.5)
The Campbell law associated with $`๐ซ`$ is the probability law on $`(0,1)\times ๐ฉ(0,1)`$ defined by
$$๐ซ_{\mathrm{Camp}}(A\times B):=_{๐ฉ(0,1)}๐ซ(\mathrm{d}\nu )\nu (A)1_{\left\{\nu B\right\}}$$
(8.6)
for all Borel-measurable $`A(0,1)`$ and $`B๐ฉ(0,1)`$. If $`(v,V)`$ is a $`(0,1)\times ๐ฉ(0,1)`$-valued random variable with law $`๐ซ_{\mathrm{Camp}}`$, then $`(V)=๐ซ_{\mathrm{size}}`$, and $`v`$ is the position of a โtypicalโ particle chosen from $`V`$.
Let
$$๐ซ^{x,n}():=P^{\delta _x}[Y_n^h]$$
(8.7)
denote the law of $`Y^h`$ at time $`n`$, started at time $`0`$ with one particle at position $`x(0,1)`$. Note that by criticality, $`_{๐ฉ(0,1)}๐ซ^{x,n}(\mathrm{d}\nu )|\nu |=1`$. Using again criticality, it is easy to see that in order to prove the extinction formula (8.1), it suffices to show that
$$\underset{n\mathrm{}}{lim}๐ซ_{\mathrm{size}}^{x,n}\left(\{1,\mathrm{},N\}\right)=0(x(0,1),N1).$$
(8.8)
In order to prove (8.8), we will write down an expression for $`๐ซ_{\mathrm{Camp}}^{x,n}`$. Let $`Q^h`$ denote the offspring mechanism of $`Y^h`$, and, for fixed $`x(0,1)`$, let $`Q_{\mathrm{Camp}}^h(x,)`$ denote the Campbell law associated with $`Q^h(x,)`$. The next proposition is a time-inhomogeneous version of Kallenbergโs famous backward tree technique; see \[Lie81, Satz 8.2\].
###### Proposition 8.2
(Representation of Campbell law) Let $`(๐ฏ_k,V_k)_{k0}`$ be the Markov process in $`(0,1)\times ๐ฉ(0,1)`$ with transition laws
$$P[(๐ฏ_{k+1},V_{k+1})|(๐ฏ_k,V_k)=(x,\nu )]=Q_{\mathrm{Camp}}^h(x,)((x,\nu )(0,1)\times ๐ฉ(0,1)),$$
(8.9)
started in $`(๐ฏ_0,V_0)=(\delta _x,0)`$. Let $`(Y^{h,(k)})^{k1}`$ be branching particle systems with offspring mechanism $`Q^h`$, conditionally independent given $`(๐ฏ_k,V_k)_{k0}`$, started in $`Y_0^{h,(k)}=V_k\delta _{๐ฏ_k}`$. Then
$$๐ซ_{\mathrm{Camp}}^{x,n}=(๐ฏ_n,\delta _{๐ฏ_n}+\underset{k=1}{\overset{n}{}}Y_{nk}^{h,(k)}).$$
(8.10)
Formula (8.10) says that the Campbell law at time $`n`$ arises in such a way, that an โimmortalโ particle at positions $`๐ฏ_0,\mathrm{},๐ฏ_n`$ sheds off offspring $`V_1\delta _{๐ฏ_1},\mathrm{},V_n\delta _{๐ฏ_n}`$, distributed according to the size-biased law with one โtypicalโ particle taken out, and this offspring then evolve under the usual forward dynamics till time $`n`$. Note that the position of the immortal particle $`(๐ฏ_k)_{k0}`$ is an autonomous Markov chain.
We need a bit of explicit control on $`Q_{\mathrm{Camp}}^h`$.
###### Lemma 8.3
(Campbell law) One has
$$Q_{\mathrm{Camp}}^h(x,A\times B)=\frac{\frac{1}{\gamma ^{}}+1}{h(x)}P[\mathrm{Pois}(h๐ต_x^\gamma ^{})\mathrm{d}\chi ]\chi (A)1_{\{\chi A\}},$$
(8.11)
where the random measures $`๐ต_x^\gamma ^{}`$ are defined in (3.7).
Proof By the definition of the Campbell law (8.6), and (3.24),
$$\begin{array}{ccc}\hfill Q_{\mathrm{Camp}}^h(x,A\times B)& =& Q^h(x,\mathrm{d}\chi )\chi (A)1_{\{\chi B\}}\hfill \\ & =& \frac{\frac{1}{\gamma ^{}}+1}{h(x)}P[\mathrm{Pois}(h๐ต_x^\gamma ^{})\mathrm{d}\chi ]\chi (A)1_{\{\chi B\}}+\left(1\frac{\frac{1}{\gamma ^{}}+1}{h(x)}\right)0.\hfill \end{array}$$
(8.12)
Recall that by (3.7),
$$๐ต_x^\gamma ^{}:=_0^{\tau _\gamma _{}}\delta _{๐ฒ_x^\gamma ^{}(t/2)}dt,$$
(8.13)
where $`(๐ฒ_x^\gamma ^{}(t))_t`$ is a stationary solution to the SDE (3.6) with $`\gamma =\gamma ^{}`$. By Lemma 8.3, the transition law of the Markov chain $`(๐ฏ_k)_{k0}`$ from Proposition 8.2 is given by
$$P[๐ฏ_{k+1}\mathrm{d}y|๐ฏ_k=x]=\frac{\frac{1}{\gamma ^{}}+1}{h(x)}E[\mathrm{Pois}(h๐ต_x^\gamma ^{})(\mathrm{d}y)]=\frac{1+\gamma ^{}}{h(x)}h(y)\mathrm{\Gamma }_x^\gamma ^{}(\mathrm{d}y),$$
(8.14)
where $`\mathrm{\Gamma }_x^\gamma ^{}`$ is the invariant law of $`๐ฒ_x^\gamma ^{}`$ from Corollary 5.4. In the next section we will prove the following lemma.
###### Lemma 8.4
(Immortal particle stays in interior) The Markov chain $`(๐ฏ_k)_{k0}`$ started in any $`๐ฏ_0=x(0,1)`$ satisfies
$$(๐ฏ_k)_{k0}\text{ has a cluster point in }(0,1)\text{a.s.}$$
(8.15)
We now show that Lemma 8.4, together with our previous results, implies Proposition 7.15 (a).
Proof of Proposition 7.15 (a) We need to prove (8.1). By our previous analysis, it suffices to prove (8.8) under the assumption that $`\rho 0`$. By Proposition 8.2,
$$๐ซ_{\mathrm{size}}^{x,n}=\left(\delta _{๐ฏ_n}+\underset{k=1}{\overset{n}{}}Y_{nk}^{h,(k)}\right).$$
(8.16)
Conditioned on $`(๐ฏ_k,V_k)_{k0}`$, the $`(Y_{nk}^{h,(k)})_{k=1,\mathrm{},n}`$ are independent random variables with
$$P\left[Y_{nk}^{h,(k)}0\right]P\left[Y_m^{h,(k)}0m0\right]=P[\mathrm{Thin}_\rho (V_k\delta _{๐ฏ_k})0].$$
(8.17)
Therefore, (8.8) will follow by Borel-Cantelli provided that we can show that
$$\underset{k=1}{\overset{\mathrm{}}{}}P[\mathrm{Thin}_\rho (V_k\delta _{๐ฏ_k})0|๐ฏ_{k1}]=\mathrm{}\text{a.s.}$$
(8.18)
Define $`f(x):=P[\mathrm{Thin}_\rho (V_k\delta _{๐ฏ_k})0|๐ฏ_{k1}=x]`$ $`(x(0,1))`$. We need to show that $`_{k=1}^{\mathrm{}}f(x)=\mathrm{}`$ a.s. Using Lemma 8.1 and Lemma 8.3 we can estimate
$$f(x)P[\mathrm{Thin}_{\stackrel{~}{\rho }}(V_k\delta _{๐ฏ_k})0|๐ฏ_{k1}=x]=_{๐ฉ(0,1)}Q_{\mathrm{Camp}}^h(x,\mathrm{d}y,\mathrm{d}\nu )\{1(1\stackrel{~}{\rho })^{\nu \delta _y}\}>0$$
(8.19)
for all $`x(0,1)`$. Since $`๐ฌ_\gamma ^{}`$, defined in (3.8), is a continuous cluster mechanism, also $`Q_{\mathrm{Camp}}^h(x,)`$ is continuous as a function of $`x`$, hence the bound in (8.19) is locally uniform on $`(0,1)`$, hence Lemma 8.4 implies that there is an $`\epsilon >0`$ such that
$$P[\mathrm{Thin}_\rho (V_k\delta _{๐ฏ_k})0|๐ฏ_{k1}]\epsilon $$
(8.20)
at infinitely many times $`k1`$, which in turn implies (8.18).
### 8.3 The immortal particle
Proof of Lemma 8.4 Let $`K(x,\mathrm{d}y)`$ denote the transition kernel (on $`(0,1)`$) of the Markov chain $`(๐ฏ_k)_{k0}`$, i.e., by (8.14),
$$K(x,\mathrm{d}y)=(1+\gamma ^{})\frac{y(1y)}{x(1x)}\mathrm{\Gamma }_x^\gamma ^{}(\mathrm{d}y).$$
(8.21)
It follows from (5.24) that
$$K(x,\mathrm{d}y)y(1y)=\frac{x(1x)+\gamma ^{}(1+\gamma ^{})}{(1+2\gamma ^{})(1+3\gamma ^{})}.$$
(8.22)
Set
$$g(x):=K(x,\mathrm{d}y)y(1y)x(1x)(x(0,1)).$$
(8.23)
Then
$$M_n:=๐ฏ_n(1๐ฏ_n)\underset{k=0}{\overset{n1}{}}g(๐ฏ_k)(n0)$$
(8.24)
defines a martingale $`(M_n)_{n0}`$. Since $`g>0`$ in an open neighborhood of $`\{0,1\}`$,
$$P[(๐ฏ_k)_{k0}\text{ has no cluster point in }(0,1)]P[\underset{n\mathrm{}}{lim}M_n=\mathrm{}]=0,$$
(8.25)
where in the last equality we have used that $`(M_n)_{n0}`$ is a martingale.
## 9 Proof of the main result
Proof of Theorem 1.4 Part (a) has been proved in Section 5.3. It follows from (2.12), (2.13), (3.10), and (3.11) that part (b) is equivalent to the following statement. Assuming that
$$(\mathrm{i})\underset{n=1}{\overset{\mathrm{}}{}}\gamma _n=\mathrm{}\text{and}(\mathrm{ii})\gamma _n\underset{n\mathrm{}}{}\gamma ^{}$$
(9.1)
for some $`\gamma ^{}[0,\mathrm{})`$, one has, uniformly on $`[0,1]`$,
$$๐ฐ_{\gamma _{n1}}\mathrm{}๐ฐ_{\gamma _0}(p)\underset{n\mathrm{}}{}p_{l,r,\gamma ^{}}^{},$$
(9.2)
where $`p_{l,r,\gamma ^{}}^{}`$ is the unique solution in $`_{l,r}`$ of
$$\begin{array}{ccccc}\hfill (\mathrm{i})& \hfill ๐ฐ_\gamma ^{}p^{}& =& p^{}\hfill & \text{if }0<\gamma ^{}<\mathrm{},\hfill \\ \hfill (\mathrm{ii})& \hfill \frac{1}{2}x(1x)\frac{^2}{x^2}p^{}(x)p^{}(x)(1p^{}(x))& =& 0(x[0,1])\hfill & \text{if }\gamma ^{}=0.\hfill \end{array}$$
(9.3)
It follows from Proposition 3.5 that the left-hand side of (9.2) converges uniformly to a limit $`p_{l,r,\gamma ^{}}^{}`$ which is given by (3.34). We must show $`1^{}`$ that $`p_{l,r,\gamma ^{}}^{}_{l,r}`$ and $`2^{}`$ that $`p_{l,r,\gamma ^{}}^{}`$ is the unique solution in this class to (9.3). We first treat the case $`\gamma ^{}>0`$.
$`1^{}`$ Since $`p_{0,0,\gamma ^{}}^{}0`$ and $`p_{1,1,\gamma ^{}}^{}1`$, it is obvious that $`p_{0,0,\gamma ^{}}^{}_{0,0}`$ and $`p_{1,1,\gamma ^{}}^{}_{1,1}`$. Therefore, by symmetry, it suffices to show that $`p_{0,1,\gamma ^{}}^{}_{0,1}`$. By Lemmas 7.8 and 7.11, $`xp1(1x)^7`$ implies $`x๐ฐ_{\gamma _k}p1(1x)^7`$ for each $`k`$. Iterating this relation, using (9.2), we find that
$$xp_{0,1,\gamma ^{}}^{}(x)1(1x)^7.$$
(9.4)
By Proposition 5.11, the left-hand side of (9.2) is nondecreasing and concave in $`x`$ if $`p`$ is, so taking the limit we find that $`p_{0,1,\gamma ^{}}^{}`$ is nondecreasing and concave. Combining this with (9.4) we conclude that $`p_{0,1,\gamma ^{}}^{}`$ is Lipschitz continuous. Moreover $`p_{0,1,\gamma ^{}}^{}(0)=0`$ and $`p_{0,1,\gamma ^{}}^{}(1)=1`$ so $`p_{0,1,\gamma ^{}}^{}_{0,1}`$.
$`2^{}`$ Taking the limit $`n\mathrm{}`$ in $`(๐ฐ_\gamma ^{})^np=๐ฐ_\gamma ^{}(๐ฐ_\gamma ^{})^{n1}p`$, using the continuity of $`๐ฐ_\gamma ^{}`$ (Corollary 5.10) and (9.2), we find that $`๐ฐ_\gamma ^{}p_{l,r,\gamma ^{}}^{}=p_{l,r,\gamma ^{}}^{}`$. It follows from (9.2) that $`p_{l,r,\gamma ^{}}^{}`$ is the only solution in $`_{l,r}`$ to this equation.
For $`\gamma ^{}=0`$, it has been shown in \[FS03, Proposition 3\] that $`p_{l,r,0}^{}`$ is the unique solution in $`_{l,r}`$ to (9.3) (ii). In particular, it has been shown there that $`p_{0,1,0}^{}`$ is twice continuously differentiable on $`[0,1]`$ (including the boundary). This proves parts (b) and (c) of the theorem.
## Appendix A Appendix: Infinite systems of linearly interacting diffusions
### A.1 Hierarchically interacting diffusions
For any $`N2`$, the hierarchical group with freedom $`N`$ is the set $`\mathrm{\Omega }_N`$ of all sequences $`\xi =(\xi _1,\xi _2,\mathrm{})`$, with coordinates $`\xi _k`$ in the finite set $`\{0,\mathrm{},N1\}`$, which are different from $`0`$ only finitely often, equipped with componentwise addition modulo $`N`$. Setting
$$\xi :=\mathrm{min}\{n0:\xi _k=0k>n\}(\xi \mathrm{\Omega }_N),$$
(A.1)
$`\xi \eta `$ is said to be the hierarchical distance between two sites $`\xi `$ and $`\eta `$ in $`\mathrm{\Omega }_N`$.
Let $`D^d`$ be open and convex, and let $`๐ฒ`$ be a renormalization class on $`\overline{D}`$. Let $`\sigma `$ be a continuous root of a diffusion matrix $`w๐ฒ`$ as in Remark 1.2. Consider a collection $`๐ฑ=(๐ฑ_\xi )_{\xi \mathrm{\Omega }_N}`$ of $`\overline{D}`$-valued processes, solving a system of SDEโs of the form
$$\mathrm{d}๐ฑ_\xi (t)=\underset{k=0}{\overset{\mathrm{}}{}}\frac{c_k}{N^k}\left(๐ฑ_\xi ^{k+1}(t)๐ฑ_\xi (t)\right)\mathrm{d}t+\sqrt{2}\sigma (๐ฑ_\xi (t))\mathrm{d}B_\xi (t)(t0,\xi \mathrm{\Omega }_N),$$
(A.2)
where $`(B_\xi )_{\xi \mathrm{\Omega }_N}`$ is a collection of independent standard Brownian motions, with initial condition
$$๐ฑ_\xi (0)=\theta D(\xi \mathrm{\Omega }_N).$$
(A.3)
Here the $`(c_k)_{k0}`$ are positive constants satisfying $`_kc_k/N^k<\mathrm{}`$, and $`๐ฑ_\xi ^k(t)`$ denotes the $`k`$-block average around $`\xi `$:
$$๐ฑ_\xi ^k(t):=\frac{1}{N^k}\underset{\eta :\xi \eta k}{}๐ฑ_\eta (t)(k0).$$
(A.4)
(Note that $`|\{\eta :\xi \eta k\}|=N^k`$.) Under suitable additional assumptions on $`\sigma `$, one can show that (A.2) has a unique (weak or strong) solution (see \[DG93a, DG96, Swa00\]). We call $`๐ฑ`$ a system of hierarchically interacting $`\overline{D}`$-valued diffusions with migration constants $`(c_k)_{k0}`$ and local diffusion rate $`w_{ij}=_k\sigma _{ik}\sigma _{jk}`$. Such systems are used to model gene frequencies or population sizes in population biology \[SF83\].
The long-time behavior of the system in (A.2) depends crucially on the recurrence versus transience of the continuous-time random walk on $`\mathrm{\Omega }_N`$ which jumps from a point $`\xi `$ to a point $`\eta \xi `$ with rate
$$a(\eta \xi ):=\underset{k=\xi \eta }{\overset{\mathrm{}}{}}\frac{c_{k1}}{N^{2k1}}.$$
(A.5)
This random walk is recurrent if and only if
$$\underset{k=0}{\overset{\mathrm{}}{}}\frac{1}{d_k}=\mathrm{},\text{where}d_k:=\underset{n=0}{\overset{\mathrm{}}{}}\frac{c_{k+n}}{N^n}$$
(A.6)
(see \[DG93a, Kle96\]; a similar problem is treated in \[DE68\]). Assuming that the law of $`๐ฑ(t)`$ converges weakly as $`t\mathrm{}`$ to the law of some $`\overline{D}^{\mathrm{\Omega }_N}`$-valued random variable $`๐ฑ(\mathrm{})`$, one expects that in the recurrent case $`๐ฑ(\mathrm{})`$ must have the following properties:
$$\begin{array}{ccc}\hfill (\mathrm{i})& ๐ฑ_\xi (\mathrm{})=๐ฑ_\eta (\mathrm{})\hfill & \text{a.s.}\xi ,\eta \mathrm{\Omega }_N,\hfill \\ \hfill (\mathrm{ii})& ๐ฑ_\xi (\mathrm{})_wD\hfill & \text{a.s.}\xi \mathrm{\Omega }_N.\hfill \end{array}$$
(A.7)
Here $`_wD`$ is the effective boundary of $`D`$, defined in (2.3). If $`๐ฑ(t)`$ converges in law to a limit $`๐ฑ(\mathrm{})`$ satisfying (A.7), then we say that $`๐ฑ`$ clusters. In the transient case, it is believed that solutions of (A.2) do not cluster. (For compact $`\overline{D}`$ these facts were proved in \[Swa00\].)
An important tool in the study of solutions to (A.2) is the so-called interaction chain. This is the chain $`(๐ฑ_0^0(t),๐ฑ_0^1(t),\mathrm{})`$ of block-averages around the origin. Heuristic arguments suggest that in the local mean field limit $`N\mathrm{}`$, the interaction chain converges to a certain well-defined Markov chain.
###### Conjecture A.1
Fix $`w๐ฒ`$, $`\theta D`$, and positive numbers $`(c_k)_{k0}`$ such that for $`N`$ large enough, $`_kc_k/N^k<\mathrm{}`$. For all $`N`$ large enough, let $`๐ฑ^N`$ be a solution to (A.2)โ(A.3), and assume that $`t_N`$ are constants such that, for some $`n1`$, $`lim_N\mathrm{}N^nt_N=T[0,\mathrm{})`$. Then
$$(๐ฑ_0^{N,n}(t_N),\mathrm{},๐ฑ_0^{N,0}(t_N))\underset{N\mathrm{}}{}(I_n^w,\mathrm{},I_0^w),$$
(A.8)
where $`(I_n^w,\mathrm{},I_0^w)`$ is a Markov chain with transition laws
$$P[I_k^w\mathrm{d}y|I_{k1}^w=x]=\nu _x^{c_k,F^{(k)}w}(\mathrm{d}y)(x\overline{D},0kn1)$$
(A.9)
and initial state
$$I_n^w=๐ฒ_T,\text{where}\mathrm{d}๐ฒ_t=c_n(\theta ๐ฒ_t)\mathrm{d}t+\sqrt{2}\sigma ^{(n)}(๐ฒ_t)\mathrm{d}B_t,๐ฒ_0=\theta ,$$
(A.10)
and $`\sigma ^{(n)}`$ is a root of the diffusion matrix $`F^{(n)}w`$.
Rigorous versions of conjecture A.1 have been proved for renormalization classes on $`\overline{D}=[0,1]`$ and $`\overline{D}=[0,\mathrm{})`$ in \[DG93a, DG93b\].
Note that the iterated kernels $`K^{w,(n)}`$ defined in (2.4) are the transition probabilities from time $`n`$ to time $`0`$ of the interaction chain in the mean-field limit:
$$K_x^{w,(n)}(\mathrm{d}y)=P[I_0^w\mathrm{d}y|I_n^w=x](x\overline{D},n0).$$
(A.11)
Lemma 2.6 expresses the fact that the system $`๐ฑ^N`$ clusters in the local mean-field limit $`N\mathrm{}`$. The condition $`s_n\mathrm{}`$ in Lemma 2.6 means that $`_{k0}\frac{1}{c_k}=\mathrm{}`$, which, in a sense, is the $`N\mathrm{}`$ limit of condition (A.6).
### A.2 The clustering distribution of linearly interacting diffusions
Let $`D^d`$ be open, bounded, and convex, and let $`๐ฒ`$ be a renormalization class on $`\overline{D}`$. Fix migration constants $`(c_k)_{k0}`$ and assume that $`s_n\mathrm{}`$ and $`s_{n+1}/s_n1+\gamma ^{}`$ for some $`\gamma ^{}[0,\mathrm{}]`$. Recall the definition of the iterated probability kernels $`K^{w,(n)}`$ in (2.4). Recall Conjecture 2.7. Assuming that the rescaled renormalized diffusion matrices $`s_nF^{(n)}w`$ converge to a limit $`w^{}`$, we can make a guess about the limit of the iterated probability kernels $`K^{w,(n)}`$.
###### Conjecture A.2
(Limits of iterated probability kernels) Assume that $`s_nF^{(n)}ww^{}`$ as $`n\mathrm{}`$. Then, for any $`w๐ฒ`$,
$$K^{w,(n)}\underset{n\mathrm{}}{}K^{},$$
(A.12)
where $`K^{}`$ has the following description:
1. If $`0<\gamma ^{}<\mathrm{}`$, then
$$K_x^{}=\underset{n\mathrm{}}{lim}P^x[I_n^\gamma ^{}],$$
(A.13)
where $`(I_n^\gamma ^{})_{n0}`$ is the Markov chain with transition law $`P[I_{n+1}^\gamma ^{}|I_n^\gamma ^{}=x]=\nu ^{1/\gamma ^{},w^{}}`$.
2. If $`\gamma ^{}=0`$, then
$$K_x^{}=\underset{t\mathrm{}}{lim}P^x[I_t^0],$$
(A.14)
where $`(I_s^0)_{s0}`$ is the diffusion process with generator $`_{i,j=1}^dw_{ij}^{}(y)\frac{^2}{y_iy_j}`$.
3. If $`\gamma ^{}=\mathrm{}`$, then
$$K_x^{}=\underset{\gamma \mathrm{}}{lim}\nu _x^{1/\gamma ,w^{}}.$$
(A.15)
For each $`N2`$, let $`๐ฑ^N=(๐ฑ_\xi ^N)_{\xi \mathrm{\Omega }_N}`$ be a system of hierarchically interacting diffusions as in (A.2) and (A.3). If $`\gamma ^{}=0`$, then because of Conjectures A.1 and A.2, we expect<sup>3</sup><sup>3</sup>3For $`\gamma ^{}>0`$, the situation is more complex. In this case at the right-hand side of (A.16) we expect the law $`_{\overline{D}}P^\theta [๐ฒ_T\mathrm{d}x]K_x^{}`$, where $`๐ฒ`$ solves the SDE $`\mathrm{d}๐ฒ_t=\frac{1}{\gamma ^{}}(\theta ๐ฒ_t)\mathrm{d}t+\sqrt{2}\sigma ^{}(๐ฒ_t)\mathrm{d}B_t`$ and $`\sigma ^{}`$ is a root of the diffusion matrix $`w^{}`$. Note that in this case the right-hand side of (A.16) depends on $`T`$. that
$$\underset{n\mathrm{}}{lim}\underset{N\mathrm{}}{lim}(๐ฑ_0^N(N^nT))=K_\theta ^{}(T>0),$$
(A.16)
where $`K^{}`$ is the kernel in (A.14).
In particular, consider the case that the migration constants $`(c_k)_{k0}`$ are of the form $`c_k=r^k`$ for some $`r>0`$. In this case, $`s_{n+1}/s_n\frac{1}{r}1`$, and $`s_n\mathrm{}`$ if and only if $`r1`$. One can check (see (A.6)) that for fixed $`N2`$, the random walk with the kernel $`a`$ in (A.5) is recurrent if and only if $`r1`$. The critical case $`r=1`$ corresponds to a critically recurrent random walk. For a precise definition of critical recurrence, see \[Kle96, formula (1.15)\]. For $`r=1`$, we expect that the double limit in (A.16) can be replaced by a single limit. More precisely, for each fixed $`N2`$, we expect that
$$\underset{t\mathrm{}}{lim}(๐ฑ_0^N(t))=K_\theta ^{}.$$
(A.17)
In this case, we call $`K_\theta ^{}`$ the clustering distribution of $`๐ฑ^N`$. The clustering distribution of linearly interacting isotropic diffusions was studied in \[Swa00\]. We expect (A.17) to hold, even more generally, for all systems of linearly interacting diffusions with a critically recurrent migration mechanism. In particular, we expect (A.17) to hold for symmetric nearest-neighbor interaction on $`^d`$ in the critical dimension $`d=2`$. If one is ready to make this enormous leap of faith, then combining Conjectures 2.7 and A.2, one arrives at the following conjecture.
###### Conjecture A.3
(Critical clustering) Let $`D^d`$ be open, bounded, and convex, and let $`๐ฒ`$ be a renormalization class on $`\overline{D}`$. Assume that the asymptotic fixed point equation (2.16) (ii) has a unique solution $`w^{}`$ in $`๐ฒ`$. Let $`\sigma `$ be a continuous root of a diffusion matrix $`w๐ฒ`$. Let $`๐ฑ=(๐ฑ_\xi )_{\xi ^2}`$ be a $`\overline{D}^^2`$-valued process, solving the system of SDEโs
$$\mathrm{d}๐ฑ_\xi (t)=\underset{\eta :|\eta \xi |=1}{}\left(๐ฑ_\eta (t)๐ฑ_\xi (t)\right)\mathrm{d}t+\sigma (๐ฑ_\xi (t))\mathrm{d}B_\xi (t),$$
(A.18)
with initial condition $`๐ฑ_\xi (0)=\theta \overline{D}`$ $`(\xi ^2)`$. Then
$$๐ฑ_\xi (t)\underset{t\mathrm{}}{}I_{\mathrm{}}^\theta (\xi ^2),$$
(A.19)
where $`(I_s^\theta )_{s0}`$ is the diffusion with generator $`_{i,j}w_{ij}^{}(y)\frac{^2}{y_iy_j}`$ and initial condition $`I_0^\theta =\theta `$.
### Acknowledgements
We thank Anton Wakolbinger and Martin Mรถhle for pointing out reference \[Ewe04\] and the fact that the distribution in (5.17) is a $`\beta `$-distribution.
|
warning/0506/astro-ph0506054.html
|
ar5iv
|
text
|
# VLA 3.5 cm continuum sources in the Serpens cloud core
## 1 Introduction
The Serpens cloud core (size $`6^{}`$ in diameter) is one of the more active, nearby star-forming regions which has been the subject of many observational studies during the last $``$ 15 years. The core is populated by a dense and extremely young cluster whose members are found in many different evolutionary stages. Near-IR surveys have identified more than 150 near-IR sources embedded in the cloud core (Eiroa & Casali, 1992; Sogawa et al. 1997; Giovannetti et al. 1998; Kaas 1999). Many of these sources have been classified as Class II young stellar objects (YSOs) - i.e. embedded T Tauri stars - by means of mid-IR ISO observations (Kaas et al. 2004); these observations also reveal a large number of flat-spectrum sources and Class I protostars. Submillimeter, millimeter, and far-IR observations show that the cloud core is also populated by many Class 0 protostars and protostellar condensations (Casali et al. 1993; Hurt & Barsony 1996; Testi & Sargent 1998). Line profiles of different molecules and transitions suggest infall motions of the gas associated with some of the Class 0 protostars (e.g Hurt et al. 1996; Gregersen et al., 1997; Mardones et al. 1997; Wolf-Chase et al. 1998; Hogerheidijde et al. 1999; Williams & Myers 2000; Narayanan et al. 2002). X-ray sources have been detected and identified with some of the YSOs (Preibisch 1998, 2003, 2004). In addition, an embedded young brown dwarf has been identified in the Serpens cloud (Lodieu et al. 2002). Further signspots related to star formation activity are found in the Serpens core: HH objects (e.g Davis et al. 1999; Ziener & Eislรถffel 1999), H<sub>2</sub> emission (e.g. Eiroa et al. 1997; Herbst et al. 1997; Hodapp 1999), and molecular outflows (White et al. 1995; Wolf-Chase et al; 1998; Williams & Myers 2000). Thus, the Serpens core represents an excellent laboratory for studying physical processes associated with star formation and their mutual inter-relations.
Very little has been done, however, in the radio continuum. Such observations are relevant because, among other aspects, they precisely reveal the positions of YSO objects, trace stellar winds at positions very close to the driving sources, and provide insights into thermal and synchroton emission in YSOs, stellar coronal activity of YSOs and magnetic fields. Radio continuum observations of Serpens were reported by Rodrรญguez et al. (1980) and Snell & Bally (1986). Rodrรญguez et al. (1989) and Curiel et al. (1993) studied a somehow unique triple radio-source with thermal and non-thermal emission components proposed to be a radio jet precursor of Herbig-Haro objects. This radio source is associated with the Class 0/Class I protostar smm 1/FIRS 1, the most luminous object embedded in the cloud core (Harvey et al. 1984; Casali et al. 1993). Smith et al. (1999) has analysed some VLA data of the radio emission coming from the SVS 4 region (Strom et al. 1976; Eiroa & Casali 1989). However, a detailed analysis of the radio emission in the Serpens cloud core is still lacking.
In this work we present VLA 3.5 cm continuum observations carried out by us toward the Serpens cloud. Our aim is to make a first approach to the Serpens cluster behaviour at radio continuum wavelengths, filling at least partly the existent gap at these frequencies. The layout of the paper is as follows. In Section 2 we present details of the VLA observations. Section 3 presents the VLA sources detected in the data; in that section we suggest associations of the radio sources with Serpens YSOs identified in other wavelengths regimes. Section 4 presents a discussion on the plausible nature of the radio sources and their clustering in the overall frame of the Serpens YSO cluster. Section 5 presents our conclusions. Finally, a short description of the individual radio sources and their counterparts together with their corresponding positions are given in the appendix (Section A).
## 2 Observations
Observations at 3.5 cm continuum where made with the Very Large Array (VLA) of the National Radio Astronomy Observatory (NRAO)<sup>1</sup><sup>1</sup>1NRAO is a facility of the National Science Foundation operated under cooperative agreement by Associated Universities Inc. in its C/D and D arrays during 1993 October 24th and 1994 February 6th, respectively. The phase center of the interferometer was at the position of the infrared source SVS 20 (Eiroa & Casali, 1992) and at the position of the Serpens thermal radio jet (Curiel et al. 1993) for the C/D and D configurations, respectively (Table 1). The observations were made in both circular polarizations with an effective bandwidth of 100 MHz. The absolute amplitude calibrator was 1328+307, with an adopted flux density of 5.27 Jy, while 1749+096 was used as phase calibrator. The data were edited, calibrated and imaged following standard procedures using the NRAO AIPS package. In order to measure the flux density of the individual radio sources, We produced cleaned maps setting the ROBUST parameter of the AIPS task IMAGR to 0 to optimize the tradeoff between angular resolution and sensitivity. The resulting synthesized beam sizes and rms sensitivities of the maps are given in Table 1.
## 3 Results
We have detected 22 VLA 3.5 cm continuum sources towards the Serpens core. Fig 1 presents 3.5 cm continuum contour maps of the SVS 20 and Serpens triple radio source fields. Table 2 gives the J2000.0 equatorial coordinates and peak and total flux densities. These flux densities have been corrected by the primary beam response of the VLA. All the continuum sources are point-like for our synthesized beam sizes, with the exception of VLA 7, which is the well known Serpens triple radio source (Curiel et al. 1993), and VLA 17, which is associated with the near-IR source EC 95 (see also Smith et al. 1999). We also note that SVS 20 (phase center of the C/D configuration) and the Serpens thermal radio jet (phase center of the D configuration) are separated by 2.4 and therefore both sources are at the border but within the two overlapping primary beams of the VLA (5.4โ at 3.5 cm). However, we have preferred to manage the two sets of data (C/D and D configurations) separately, without merging them into a single data set, since most of the detected sources are not overlapping within these two primary beams (see Fig. 1); in this way, more accurate flux density measurements can be achieved. We have nevertheless checked the flux density of the compact source VLA 16 (SVS20) as measured with the D data set after primary beam correction, finding that it is fully consistent (within $`1\sigma ,6\%`$) with the flux density measured with the C/D configuration.
Following the formulation given by Anglada et al. (1998), we have estimated that the number of random background 3.5 cm continuum sources expected within a field size of 5.4 (primary beam size) with a detection threshold of $``$ 0.08 mJy ($``$ 5$`\sigma `$) is $``$ 1. This means that almost all the detections within the primary beam of the two sets of data (C/D and D configurations) are likely associated with star formation in the Serpens core and, in fact, this is in good agreement with our findings, as it is shown in the following. The last column of Table 2 indicates whether the radio sources have plausible counterparts in other wavelength ranges (see Section A, appendix for a short description of the radio continuum sources and their associated counterparts). 16 out of the 22 VLA sources (including in this number 4 sources outside the primary beam, see note in Table 2) are suggested to be associated with near-IR objects, ISO mid-IR sources, mm/sub-mm sources, and X-ray sources. All these sources are YSOs embedded in the Serpens core; their names together with the VLA names are given in Table 3. Given the uncertainties in the absolute positions and the radio beam sizes, as well as the fact that positions of the near-IR sources can differ $`23`$ arcsec in each equatorial coordinate when comparing different near-IR works (Eiroa & Casali 1992, Sogawa et al. 1997, Giovannetti et al. 1998, Kaas 1999, Pontoppidan et al. 2004), an association between a radio and a near-IR source is done when their positions coincide within a projected angular distance of 5 arcsec. This is a conservative estimate given the uncertainties between the radio and near-IR positional reference frames, as well as the near-IR random positional errors (estimated to be $`3`$ arcsec, considering the aforementioned differences among different authors); the relative radio position uncertainties are very low, of the order of 0.1 arcsec. Last column of Table 3 gives the usual updated SED classification of YSOs (e.g. Andrรฉ & Montmerle 1994, Greene et al. 1994) of the Serpens radio sources. Some objects are marked as belonging to two different evolutionary classes, i.e., Class 0/Class I. In these cases, the Class 0 classification is based on their submillimeter emission (Casali et al. 1993; Hurt & Barsony 1996; Wolf-Chase et al. 1998; Hogerheidje et al. 1999; Brown et al. 2000), while the Class I classification is taken from their ISO mid-IR emission (Kaas et al. 2004). Davis et al. (1999) suggested that the Serpens protostellar objects are borderline objects between Class 0 and Class I protostars. It is worth to note that the SED Class 0/Class I โambiguityโ has recently been noted in Cep E by Noriega-Crespo et al. (2004), who suggest that a reacommodation of the Class 0/Class I SED classification is needed given flux-limit sensitivities of current mid-IR observations (e.g. Spitzer). In general, the flat-spectrum and Class II classifications in Table 3 are taken from ISO fluxes of the objects (Kaas et al. 2004), although we make the following considerations. VLA 15 is associated with the near-IR source EC 88 and with ISO 312. We note that ISO 312 is associated with EC 88 and EC 89 by Kaas et al. (2004), because both near-IR sources are not resolved by the ISO beam. However, Pontoppidan et al. (2004) have recently separated the mid-IR ISO emission from both EC 88 and EC 89 undependently. Using the data of these later authors, we classify EC 88 as a Class I protostar, and EC 89 as a flat-spectrum source. A similar case is represented by VLA 17/VlA 17โ and the flat-spectrum source ISO 317 (Kaas et al. 2004), recently resolved by Pontoppidan et al. (2004) into two ISO sources. In this case, VLA 17 is associated with EC 95, which has a Class II SED (using the fluxes given by Pontoppidan et al. 2004), and VLA 17โ is associated with EC 92, with a Class I SED. Finally, VLA 16 is associated with the binary star SVS 20/EC 90, which is ISO 314 (Kaas et al. 2004); in this case, Haisch et al. (2002) classify both components of the binary SVS 20 as flat-spectrum sources on the basis of ground-based 10 $`\mu `$m observations.
Five out of the 16 radio sources associated with near/mid-infrared ones (VLA 5, VLA 7, VLA 9, VLA 10, and VLA 11) are found in the NW cloud clump, i.e., the FIRS 1/S 68N field located towards the NW of the Serpens reflection nebula (SRN), while the rest of the sources (VLA 12, VLA 13, VLA 14, VLA 15, VLA 16, VLA 17, VLA 17โ, VLA 18, VLA 19, VLA 19โ, and VLA 20) are located in the Southeastern (SE) cloud clump (South from SRN), extending from SVS 20 towards the SVS 4 complex.
With respect to the other detected radio sources, VLA 3, VLA 6, and VLA 8 have no identifications at other wavelength regimes and they most likely are background extragalactic objects. On the other hand, VLA 1 and VLA 2 have dubious associations. These two sources are located close to the radio source S 68/5 detected at 1.4 GHz by Snell & Bally (1986). It could be that S 68/5 was not resolved at 1.4 GHz into our two sources since it approximately lies in the middle of VLA 1 and 2. By adding the integrated flux densities of VLA 1 and VLA 2 and comparing them with the 1.4 GHz flux density of S 68/5, we obtain a negative spectral index suggesting that it could be a background extragalactic object. A faint 2MASS source, 18293641+0115093, only detected at $`J`$ (17.00 mag) and $`H`$ (15.74 mag) is located at a distance of $``$ 3<sup>โฒโฒ</sup> of either VLA 1 and 2. Finally, the weak radio source VLA 4 is at a projected distance of $``$ 6โ from the near-IR source EC 27 ($`K`$ = 13.7 mag), but we think they are unrelated. Both radio and near-IR sources are $``$ 25โ towards the Northwest of S 68N.
## 4 Discussion
### 4.1 Nature of the radio sources
Centimeter continuum emission has been detected towards a large number of low luminosity YSOs, usually powering outflows and most of them in the Class 0/Class I protostellar phase, but also few in the Class II phase (embedded T Tauri stars). That emission is very weak with flux density values in the range of tenths to a few mJy. The emission is usually compact, although when observed at subarcsecond resolution, the sources tend to present a jet-like structure (e.g., Reipurth et al. 2004). Measurements of spectral indices at cm wavelengths reveal that thermal free-free emission dominates at these frequencies, although there exists a non-thermal component in few cases, as it is the case of the Serpens radio jet - our source VLA 7 (Curiel et al. 1993). Theoretical approaches explaining the origin of the radio emission usually invoke winds (disk-winds or X-winds) from an embedded protostellar object since stellar radiation produces too little ionization to account for the continuum observations (Torrelles et al. 1985; Curiel et al. 1989; Shang et al. 2004, and references therein). On the other hand, radio counterparts of X-ray YSOs have also been explained in terms of non-thermal gyrosynchroton emission arising from coronally active stars; in fact, this is the suggestion made by Smith et al. (1999) for VLA 17 (EC 95), which is associated with a very strong X-ray source (Preibisch 1998, 2003). We note that the Serpens VLA sources follow the general properties just described, although we have no information on their radio continuum SED and sizes at the subarcsecond scale.
The Serpens radio sources are associated with YSOs in different evolutionary stages, i.e., there are Class 0, Class I, flat-spectrum and Class II objects, but the trend is to be associated with the youngest Class 0/Class I protostars (last column of Table 3). Some of the Class 0 sources might present observational evidences of infalling gas, as described in the appendix. This variety of evolutionary phases might be the cause why there is no correlation of the radio flux density neither with the observed K magnitudes, nor with mid-IR fluxes, nor with SED IR spectral indexes. Radio molecular observations reveal a widespread, very complex outflow activity in the Serpens cloud core, which makes it complicated to identify individual molecular outflows (McMullin et al. 1994; White et al. 1995, Davis et al. 1999). Nevertheless, some of the radio sources can be associated with relatively well defined individual outflows: VLA 7 is the Serpens radio jet; VLA 5, VLA 10, VLA 11, and VLA 13 are associated with molecular and/or H<sub>2</sub> outflows. All of them are Class 0/Class I objects. Further, VLA 9 is a Class I object associated with the cometary nebulosity EC 53; VLA 18 and VLA 20 also shows some indications of nebulosity with a cometary morphology (very faint in the case of VLA 18). This is suggestive of outflows in these sources. Thus, it is likely that most of the Serpens radio sources are YSO thermal radio jets as those found associated with many other protostellar objects (Reipurth et al. 2004), at least in the cases of VLA 5, VLA 7, VLA 10, VLA 11, and VLA 13.
On the other hand, VLA 9, VLA 16, VLA 17, and VLA 20 are associated with X-ray sources. The radio emission of the Class II source VLA 17 (EC 95) has been explained by Smith et al. (1999) as due to gyrosynchrotron emission in an extreme stellar corona. VLA 9 (EC 53) is a Class I object, but with very deep CO absorption bands in its K-spectrum and EXOR-type properties; thus a more evolved Class II nature can not be ruled out. VLA 16 is associated with one of the components of the flat-spectrum binary source SVS 20, most likely the Northern component, while VLA 20 is a Class II object. Interestingly, the X-ray/radio luminosity ratio of all Serpens radio sources with X-ray emission approximately follows the ratio of coronal active stars (Gรผdel & Benz 1993, Smith et al. 1999), which suggests that these Serpens radio sources could be examples of YSOs with that kind of activity.
### 4.2 Clustering
The Serpens VLA sources can be grouped into two groups. As mentioned above (ยง3), the first group (VLA 5, VLA 7, VLA 9, VLA 10, and VLA 11) are located towards the NW clump of the Serpens cloud core. The rest of the sources form a second group located in the SE clump of the Serpens core (Table 3 and Fig. 1). Both core clumps are well known from a variety of molecules, like NH<sub>3</sub>, C<sup>18</sup>0, NH<sup>+</sup> or CS (Ungerechts & Gรผsten 1984; White et al. 1995; McMullin et al. 2000; Olmo & Testi 2002). The radio sources in the NW condensation are all associated with very young Class 0/Class I protostars, while at least half of the radio sources in the SE clump are associated with more evolved flat-spectrum and Class II sources. This result is coherent with the picture of star forming activitiy emerging from results in other wavelengths. Testi et al. (2000) already noticed subclustering of YSOs within both clumps of the Serpens mm/sub-mm sources. Kaas et al (2004) find a similar distribution for the ISO Class I sources, while their Class II sources are more widespread distributed, although with a trend to be more concentrated in the SE condensation, which is also the case of the near-IR cluster (Eiroa & Casali 1992; Giovannetti et al. 1998; Kaas 1999). It is most likely that we observe radio emission in the NW clump only from Class 0/Class I sources due to the scarcity of more evolved stars in that part of the cloud. Thus, the data suggest that although stars are presently forming in both NW and SE cloud condensations, it began earlier or it was more efficient in the SE clump. An interesting case is that of the stellar group SVS 4. From the $``$ 10 stars forming the SVS 4 subcluster and sharing a common near-IR nebulosity, 5 of them are most likely associated with radio sources. Two of these sources, VLA 15 (EC 88) and VLA 17โ (EC 92) are classified as Class I objects; the rest are more evolved Class II stars.
## 5 Conclusions
The Serpens core is known to host a rich YSO cluster with more than 150 members, including a large fraction of embedded very young Class 0/Class I protostars. Our VLA 3.5 cm observations reveal a number of radio continuum sources associated with Serpens YSOs and provide a first insight into the radio continuum properties of the young cluster. Most of these radio sources, 10 out of 16, are associated with Class 0/Class I objects, one is a flat-spectrum source, and only 4 out of 16 are associated with more evolved Class II (embedded Classical T Tauri stars) sources. The source VLA 18 (GCNM 122) has not been classified in terms of evolutionary classes. At first sight the ratio between Class 0/Class I and flat-spectrum/Class II sources detected in radio continuum seems to be in contrast with the YSO population found at infrared wavelengths, where Kaas et al. (2004) identified 20 Class I, 13 flat-spectrum and 43 Class II sources. This result plausibly reflects the fact that radio continuum sources tend to be preferentially associated with the earliest phases of star formation, rather than with the more evolved T Tauri phase. The data do not allow us to trace any firm conclusion on the nature of the radio emission. We tentatively suggest that the emission from Class 0/Class I sources is due to thermal radio jets, as suggested by Anglada et al. (1998) and Reipurth et al. (2004) for similar sources, while the radio emission from the Serpens Class II sources could have a stellar coronal origin, as suggested by Smith et al. (1999) for EC 95. A determination of radio spectral indices is needed in order to ellucidate between both or any other different alternatives, which requires simultaneous observations at different radio frequencies.
CE and JMT acknowledge partial finantial support from Spanish grants PNAYA 2001-1124 and PNAYA 2002-00376, respectively. SC is grateful to CONACyT, Mรฉxico and DGAPA, UNAM for their finantial support. The comments of an anonymous referee helped us to improve the content of the paper.
## Appendix A Appendix: comments on individual sources
### A.1 VLA5
VLA 5 is the Class 0 source S 68N/smm 9 and ISO 241 (Casali et al. 1993, McMullin et al. 1994, White et al. 1995, Davis et al. 1999, Kaas et al. 2004). H<sub>2</sub>O maser emission and compact molecular outflows in several lines, e.g., CO, CS are observed (Wolf-Chase et al. 1998, Williams & Myers 2000). Infalling gas in H<sub>2</sub>CO and CS lines has been suggested (Hurt et al. 1996, Mardones et al. 1997, Williams & Myers 2000). Hodapp (1999) noticed the presence of a very red, near-IR double star, smm 9-IR, associated with some nebulosity near the nominal position of the smm source (see the excellent Figs. 1 and 2 in Hodappโs paper); he interestingly pointed out the intriguing possibility of an association of the Class 0 source with one of the near-IR objects. Table 4 lists the positions of the objects associated with VLA 5. The position of smm 9-IR has been estimated by us using the $`K`$ image of Kaas (1999).
### A.2 VLA7
VLA 7 is the well studied triple radio-jet associated with the most prominent Class 0 object in Serpens, FIRS 1/smm 1 (Curiel et al. 1993, Harvey et al. 1984, Casali et al. 1993, Hurt & Barsony 1996). H<sub>2</sub>O maser emission, molecular outflows and a molecular envelope are also found (e.g. Dinger & Dickinson 1980, White et al. 1995, Curiel et al. 1996, Hogerheijde et al. 1999, Williams & Myers 2000). An H<sub>2</sub> near-IR jet with a point-like source at its apex, EC 41, is observed close to the position of the radio/smm source. Both objects likely are different YSOs (Eiroa & Casali 1989, Hodapp 1999). Kaas et al.(2004) have detected two mid-IR sources in this region: ISO 258a, a Class I source likely associated with EC 41, and ISO 258b, which could be associated with VLA 7/smm 1, although Kaas et al. (2004) find it unlikely. Table 5 lists the positions of the sources.
### A.3 VLA 9
VLA 9 is associated with the variable near-IR reflection nebulosity EC 53 and the Class I source smm 5/ISO 265 (Eiroa & Casali 1992, Casali et al. 1993, Hodapp et al. 1996, Horrobin et al. 1997, Sogawa et al. 1997, Kaas 1999, Davis et al. 1999, Kaas et al. 2004). There is no near-IR point-like source embedded in the nebulosity; Hodapp (1999) suggested an EXor-type object nature for the embedded stellar object in EC 53. The $`K`$-spectrum presents strong CO absorption bands (Casali & Eiroa, 1996). There is no prominent molecular line emission although marginal blueshifted CO emission close to smm 5 might be present (White et al. 1995, McMullin et al. 2000, Williams & Myers 2000, Narayanan et al. 2002). An X-ray source is associated with EC 53 (Preibisch 2003, 2004). VLA 9 likely denotes the position of the EXor-type star in EC 53 (Table 6).
### A.4 VLA 10
VLA 10 coincides with the Class 0 source smm 10/ISO 270 (Davis et al. 1999, Testi et al. 1998, Kaas et al. 2004) and with a very red, highly variable near-IR object (Hodapp 1999). The position of smm 10-IR has been estimated by us using the $`K`$ image of Kaas (1999). It might be a second example of a near-IR object associated with a Class 0 object (Table 7). A CS weak outflow is detected by Williams & Myers (2000).
### A.5 VLA 11
VLA 11 is associated with the near-IR source GCNM 53 and the Class I source ISO 276, just located towards the Northwest of SRN, Table 8 (Giovannetti et al. 1997, Kaas et al. 2004). The source is at the basis of a H<sub>2</sub> jet oriented in the North-West direction with a position angle of around 135 degrees (Herbst et al. 1997).
### A.6 VLA 12
VLA 12 is at a angular distance of $``$ 5<sup>โฒโฒ</sup> of the Class II source EC 79/ISO 304 (Table 9). EC 79 is star 2 of the SVS 4 complex. The spectral type of EC 79 is late K, M as suggested by NaI, CaI and CO absorptions present in its $`K`$ spectrum (Eiroa & Djupvik 2005, in preparation). We tentatively associate the VLA source with the Class II/T Tauri-like YSO.
### A.7 VLA 13
VLA 13 coincides with the Class 0 object smm 4 and maybe with ISO 308 (Table 10); this protostellar object shows evidences of infalling gas and molecular outflows (Casali et al. 1993, White et al. 1995, Hurt et al. 1996, Mardones et al. 1997, Gregersen et al. 1997, Davis et al. 1999, Hogerheidijde et al. 1999, Larson et al. 2000, Narayanan et al. 2002, Kaas et al. 2004). Two compact $`H_2`$ condensations (Eiroa et al. 1997, Herbst et al. 1997) are located close to smm 4 and likely related to it.
### A.8 VLA 14
VLA 14 is associated with the Class II source EC 84/ ISO 309, a member of the SVS 4 stellar subgroup (Table 11). NaI and CO absorptions are detected in the $`K`$ spectrum of this source (Eiroa & Djupvik 2005, in preparation).
### A.9 VLA 15
VLA 15 is most likely associated with the variable star EC 88 (Eiroa & Casali 1992, Kaas 1999). EC 89 is found very close to it and has faint Br<sub>ฮณ</sub> emission (Eiroa & Djupvik 2005, in preparation). Both belong to the SVS 4 stellar group. Kaas et al. (2004) associate the Class I object ISO 312 with both EC 88/EC 89 stars, but using the results of Pontoppidan et al. (2004) EC 88 (VLA 15) is a Class I object, while EC 89 is a flat-spectrum source (Table 12).
### A.10 VLA 16
VLA 16 is spatially associated with smm 6, ISO 314, an X-ray source and the near-IR binary SVS 20/EC 90, the strongest near-IR source embedded in the Serpens cloud core (Eiroa et al. 1987, Eiroa & Casali 1992, Casali et al. 1993, Davis et al. 1999, Preibisch 2003, Kaas et al. 2004); both components of the binary present flat-spectrum SEDs (Haisch et al. 2002). The source is detected at 8.4 GHz by Smith et al. (1999). The near-IR binary is surrounded by a ring of nebulosity with two spiral arms seen in opposite directions; $`H_2`$ emission is detected at the end of the SW arm. The ring could be a circumbinary disk (Eiroa et al. 1997a); alternatively, Huard et al. (1997) suggest that the ring could delineate evacuated, bipolar cavities. The brightest component of the binary, SVS 20S, presents a featureless $`K`$ spectrum, while SVS 20N has strong Br<sub>ฮณ</sub> and CO bands in emission (Herbst & Rayner 1994, Eiroa & Djupvik 2005, in preparation), indicating the presence of an accretion disk. It might be that VLA 16 is associated with SVS 20N (Table 13).
### A.11 VLA 17 and VLA17โ
VLA 17 is associated with EC 95, a young star with strong X-ray emission, while VLA 17โ is most likely related to the nearby star EC 92 (Table 14). EC 95 and EC 92 belong to the SVS 4 group and present NaI, CaI and CO absorptions in their $`K`$ spectra (Preibisch 1999, Eiroa & Djupvik 2005, in preparation). The flat-spectrum source ISO 317 corresponds to both stars (Kaas et al. 2004), but it is resolved by Pontoppidan et al (2004) into two distinct mid-IR sources: a Class II source (EC 95) and a Class I source (EC 92). Smith et al. (1999) suggest optically thin synchrotron emission as the origin of the VLA 17 (EC 95) radio continuum emission.
### A.12 VLA 18
VLA 18 is spatially associated with the weak near-IR source GCNM 122 (Table 15), which shows a very faint cometary-like nebulosity. The radio source was also detected at 8.4 GHz by Smith et al. (1999). GCNM 122 is variable: Giovannetti et al. (1998) gives a $`K`$ magnitude of 14.9, while we estimate 16.3 mag.
### A.13 VLA 19 and VLA 19โ
The pair of weak radio sources VLA 19 and VLA 19โ are likely associated with the Class I source ISO 331 (Table 16), but have no near-IR counterparts. VLA 19 was detected also at 8.4 GHz by Smith et al. (1999).
### A.14 VLA 20
VLA 20 is associated with the Class II object EC 117/ISO 338 and a X-ray source (Table 17). The source was detected at 8.4 GHz by Smith et al. (1999). There are NaI, CaI, and CO absorptions in the $`K`$ spectrum of the source (Eiroa & Kaas 2005, in preparation). A nearby 3 mm source ($``$ 4 arcsec southwards) has been detected by Testi & Sargent (1998).
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.